Landslide Hazard Identification Based On SBAS-InSAR And Machine Learning

Lanzhou is located in the middle of the transition from the Qinghai Tibet Plateau to the Loess Plateau.The precipitation in this region is concentrated in summer,and less in autumn,winter and spring.The characteristics of precipitation concentration leads to landslides in the rainy season in Lanzhou,which is located in the Loess Plateau.In addition,with the expansion of urban construction in Lanzhou to the slopes on both sides of the river valley and loess tableland,a series of strong human activities,such as slope cutting,mountain cutting,irrigation in surrounding mountainous areas,farmland irrigation and mining of mineral resources,have increased the frequency of landslides in Lanzhou.Therefore,it is necessary to investigate the hidden dangers of landslides in Lanzhou.Taking the main urban area of Lanzhou as the research object,including Chengguan district,Anning district,Xigu district and Qilihe district,this paper combines SBAS-InSAR and machine learning to build threshold model to accurately identify landslide hazards in Lanzhou.Firstly,SBAS-InSAR technology is used to obtain the land surface deformation rate and time series cumulative deformation variables of Lanzhou city from 2015 to 2020 based on sentinel-1a data.The landslides in Lanzhou city are cataloged by earth optical image;then the MLP model is used to evaluate the sensitivity of landslides in Lanzhou city;finally,combined with the deformation results and landslide sensitivity evaluation results,the threshold model is constructed to accurately identify the location of landslide hazards in Lanzhou City,and the accuracy of the threshold model is verified by theoretical accuracy and field investigation.This study will provide data support and thinking for urban planning and disaster prevention and mitigation in Lanzhou city.The main conclusions are as follows(1)Based on the SBAS-InSAR technology,the surface deformation characteristics of the main urban area of Lanzhou are obtained.The main urban area of Lanzhou is stable,but the surface deformation of some areas is relatively large,mainly concentrated in the mountainous areas around Lanzhou city,land creation and Qingshi areas in Chengguan district,shajingyi in Anning district,Liuquan town in Xigu district and Yingmenqian village in Qilihe district.Through statistical analysis,it is found that the large deformation area is located in the area with low elevation,low slope,close to roads,railways,faults and rivers,and weak rock strata.(2)Based on the principles of systematisms,representativeness,hierarchy,operability and multicollinearity analysis method,the influencing factors of landslide are selected,such as terrain factors(slope,elevation,aspect),geological factors(lithology,distance from fault),hydrological factors(distance from river),human engineering factors(distance from road,distance from railway).A total of 152 landslides were interpreted in the main urban area of Lanzhou by optical image,including 138 unstable slopes and 14 old/ancient landslides,with a total area of 7.189 km~2.Then the sensitivity of landslides in Lanzhou was predicted by the interpreted landslides and MLP model.The ROC curve surface product was 0.839,and the evaluation effect was ideal.The area of extremely low sensitive area,low sensitive area,medium sensitive area,highly sensitive area and extremely highly sensitive area is 11.76 km~2,589.85 km~2,112.25 km~2,227.87 km~2 and 123.87 km~2respectively.The extremely highly sensitive area and highly sensitive area are distributed in the north and south mountains of the main urban area of Lanzhou.(3)Combined with the results of SBAS-InSAR technology and MLP model to build threshold model,the location of landslides in Lanzhou is identified.Hidden danger of landslides in Lanzhou are mainly distributed in the northern mountainous areas of Chengguan district,Qilihe district and Xigu District.The coincidence rate of hidden danger of landslide and landslides is 92.76%.After smooth and screening treatment,there are 168 hidden danger of landslides.The distribution characteristics of hidden danger of landslides,large deformation area and landslide distributed in low altitude,sunny slope,medium slope,soft lithology and close to artificial facilities and rivers.The results show that the theoretical accuracy of the threshold model is high,and the field investigation also verifies the accuracy of the threshold model.