Landslide Hazard Identification Based On InSAR And Machine Learning In Zhouqu County

As one of the serious geological disasters in China,landslides are mainly influenced by tectonic activities and other environments,which are mostly found in the southwest region of China.The complex geological structure of the Bailong River basin,coupled with concentrated rainfall and human activities,has resulted in frequent landslide disasters in the region,which pose significant risks to the ecological environment,critical infrastructure,and the safety of individuals and their property.As a typical area with frequent landslide hazards,Zhouqu County is of great practical significance for the study of landslide risks in the area to promote local economic and social development.Synthetic aperture radar interferometry(InSAR)technology can monitor surface deformation in a wide range and with high accuracy,but it is easily affected by topography and other factors;Machine learning model to identify landslides can effectively supplement the data deficiency of incoherent areas when InSAR technology monitors large displacements The combination of the two factors is conducive to comprehensive and timely detection of landslide hazards,facilitating the development of making targeted protection measures.Therefore,taking Zhouqu County as the research object,SBAS-InSAR technology is used to extract and analyze the surface deformation information and circling the landslide hazard areas and unstable slopes.Using optical remote sensing images as a foundation,various machine learning models are employed to detect landslide-prone locations in Zhouqu County.The precision of each model’s extraction results is then assessed and analyzed to determine the most suitable machine learning model for landslide detection in the region.The landslide identification results are analyzed with the overlay of optical images according to the marked landslide potential hazard areas and a typical area is selected for example demonstration.The research results have implications for the exploration,monitoring,and early warning of other potential landslides in the region,and play an active role in regional disaster prevention and mitigation efforts.The main work and conclusions of the paper are as follows:(1)Enclosing landslide hazard areas based on SBAS-InSAR technology.The surface deformation rates of Zhouqu County from January 2020 to May 2021 were inferred using SBAS-InSAR technology and Sentinel-1A lift-track data.The results showed that the lifttrack deformation rates ranged from-144.68 mm/a to 111.48 mm/a,and the lift-track deformation rates ranged from-114.60 mm/a to 141.18 mm/a.The surface deformation rates obtained from different track The surface deformation rates obtained from different track data are highly consistent.The accuracy of the monitoring results of the deformation of the lifting track is evaluated by internal inspection,in which the accuracy is higher along the Qinling Mountains,which is basically consistent with the spatial distribution of the effective deformation information,indicating that the experimental results have high reliability.Set10mm/a～10mm/a as the threshold value to segment the deformation results and circle the landslide hazard area and unstable slope.By studying the correlation between deformation results and precipitation data,it is concluded that rainfall is one of the main causes of landslides in the region.(2)Landslide hazard identification analysis based on machine learning models.Based on the existing landslide data set,the logistic regression(LG)model,random forest(RF)model and multilayer perceptron(MLP-NN)neural network model were used to identify landslide hazards,and the recognition accuracy of each model was quantitatively evaluated.The results show that the MLP-NN model has the highest theoretical accuracy.Comparing the model identification results with the superimposed surface deformation rate information,the superimposed agreement of LG,RF and MLP-NN models are 74.72%,80.00%and 90.50%.respectively,so the multilayer perceptron neural network model is more suitable for landslide hazard identification in Zhouqu County.(3)Combined SBAS-InSAR and machine learning for landslide hazard identification.Based on SBAS-InSAR technology and multilayer perceptron neural network model to identify landslide hidden danger areas,superposition analysis is performed to identify active landslide hidden danger points,and then compared with optical remote sensing images for analysis.The results show that most of the identified landslide potential sites are located in areas such as historical landslides or bare ground.As a result,three typical historical landslide areas,namely Jiangdingya,Yahuokou and Suoertou,were studied empirically,and precipitation data and landslide time-series deformation curves were used to reveal the landslide deformation mechanism,and it was found that the landslide deformation in the area was positively correlated with precipitation with a certain lag.