Optimized Regional Landslide Susceptibility Evaluation In Maoxian Based On SBAS-InSAR Technique

Maoxian is located in the northwest of Sichuan Province and belongs to the Aba Tibetan and Qiang Autonomous Prefecture.It is located in the area where the Qinghai-Tibet Plateau transitions to the Sichuan Basin and the terrain changes abruptly.With the development of fold and fault structures and the strong neotectonic activity,the area is seriously affected by the earthquake,which makes the area a typical geological disaster prone area,mainly including landslide and collapse.Among them,the landslide activity is extremely strong,which threatens the national highway 213 and villages along the river.In order to ensure the safety of people's lives and property,and realize the harmonious coexistence and development of people and environment,it is necessary to carry out sensitivity evaluation of landslides in the region.The evaluation results can also provide an important basis for the government's land use planning and disaster prevention and mitigation.Taking the landslide from Feihong township to Diexi town in Maoxian as the research object.By analyzing the development characteristics of landslide disaster,the influencing factors were selected and analyzed,then the spatial database and sensitivity evaluation system of the study area were established.Based on GIS technology,combined with deterministic coefficient method and three machine learning methods,the sensitivity assessment and analysis of landslide geological hazard in the study area were carried out.With the aid of SBAS-InSAR technology to optimize the results,the sensitivity of the landslide is figure and the combination of InSAR data.In this way,the rationality of the landslide sensitivity is improved.The main achievements are as follows:(1)There are 122 landslide disaster points in the study area,which are mainly distributed in townships such as Huilong Township,Yonghe Township,Baixi Township,Shidaguan Township,and Diexi Town.Among them,23 landslide sites are distributed in Huilong Township,21 in Yonghe Township,17 in Baixi Township,13 in Shidaguan Township,12 in Feihong Township,and 10 in Diexi Township.By projecting the landslides to Google,it is found that the landslides are mainly distributed along the Minjiang River in groups or chains.(2)11 indicators,including elevation,slope,slope direction,river,relief and curvature,geological factors(stratum and fault)and environmental factors(land type,vegetation type and terrain humidity index),were selected to construct the sensitivity evaluation system of landslide.Taking ArcGIS as the platform,combining SPSS statistical analysis module and EXCEL information processing,the importance of each influence factor classification interval is analyzed by the method of certainty coefficient.Through statistical analysis,the landslide was distributed in the regions with elevation value of 1500~2000m,undulations value of 0~100m,curvature value of-0.5~0.5,slope of 0°~20°and slope direction of north and east,which was prone to landslide.Landslide disasters are mainly distributed on both sides of Minjiang River and its tributaries,near the fault zone.The strata are mainly Devonian and Silurian.Landslides are more widely distributed in areas with large arable land,herbaceous vegetation and TWI values.(3)Three combination models of machine learning methods were mainly used for evaluation,which are CF-SVM,CF-RF and CF-ANN.Based on the coefficients determined by CF as the basic data,three intelligent machine learning algorithms were used to evaluate the sensitivity of the research area.Based on ArcGIS software,the evaluation results of the three models divide the research area into fIve areas by the method of natural discontinuity,which are respectively very low sensitive area,low sensitive area,medium sensitive area,high sensitive area and very sensitive area.The accuracy of the three models were evaluated and compared.Finally,the CF-SVM model was optimized.The areas of the very low,low,medium,high and extremely sensitive areas in the sensitivity map of the research area were 597.38 km~2?229.82km~2?82.39km~2?101.12km~2 and 152.50km~2 respectively.(4)SBAS-InSAR technology and multi-temporal Sentient-1A data were used to extract the surface deformation rate graph of the study area from 2017 to 2019.Combined with InSAR data,the ROC accuracy of the CF-SVM model was up to 0.894,which improved to a certain extent.The rationality of landslide sensitivity zoning map has also been verified and improved.And through on-site verification,5 new landslide disaster hidden danger points occurred,and the corresponding sensitivity level of 2landslide disaster hidden danger points has increased,and the evaluation results are more reasonable.The areas with very low,low,medium,high and very high sensitivity of the optimized evaluation results were 641.59 km~2?198.77 km~2?79.10 km~2?128.92km~2 and114.83 km~2,respectively.The optimized landslide sensitivity evaluation results are consistent with the spatial distribution characteristics of the landslide points in the study area,indicating that the optimized results of CF-SVM model and SBAS-InSAR are suitable for the sensitivity evaluation of the study area,which can provide a new method for the future regional landslide sensitivity evaluation.