Early Identification Of Geological Hazards And Landslides Susceptibility Evaluation Of Karakorum Highway (Domestic Section)

Karakoram Highway connecting China and Pakistan is an important part of the construction of China-Pakistan Economic Corridor,and its geographical location is extremely important.The highway passes through the Hindu Kush,Himalayas and Karakoram Mountains.The geological conditions along the route are extremely complicated,the geological structure is active,and historical earthquakes are frequent.The frequent occurrence of geological hazards,mainly are unstable slopes,landslides,ground subsidence,glacial movements,and debris flows,has become a major threat to highway safe operation.It is of great significance to research the deformation characteristics of geological hazards along the highway and the susceptibility assessment of landslides.This research can deeply understand the relationship between the distribution of geological hazards of Karakoram highway and various environmental factors,and can also provide effective decision-making support for the government to carry out Karakoram highway reconstruction and expansion and the construction of China-Pakistan Economic Corridor.However,due to the large space span of Karakoram highway,the traditional ground monitoring methods are difficult to meet the needs of a comprehensive survey of geological hazards in this region.With the emergence of InSAR technology in recent years,large-area,high-precision and long-term serial surface deformation monitoring has become possible.It has been widely used in the early identification of geological hazards such as landslides,unstable slopes and ground subsidence.In this study,SBAS-InSAR method is used to monitor the surface deformation and early identification of geological hazards in the 10 km buffer zone of Karakoram highway.Based on TRMM precipitation data and regional environmental characteristics,the movement characteristics and development rules of typical geological hazards are analyzed.Comprehensively consider the geological environment background,terrain conditions,hydrological conditions,vegetation coverage and road factors that affect the occurrence of landslides(including un-landslides and landslides)in the study area.The logistic regression model and random forest model were used to evaluate the susceptibility of landslides.Using the validation data set and AUC value of ROC curve to verify and compare the evaluation accuracy of the two models.The optimal evaluation results were combined with the deformation rate results to optimize the landslides susceptibility levels,and the following conclusions were reached:(1)Using the SBAS-InSAR method to process the 28 scene Sentinel-1A radar image from ESA,and obtain the time series results of ground deformation in the study area from 2016 to 2017.The rate along line of sight was-82.46~142.48 mm/a,and the maximum deformation rate along slope was-474 mm /a based on the geometric attitude of the radar.(2)Based on the slope deformation rate results combined with field verification and remote sensing image interpretation,a total of 280 geological disaster-causing bodies were identified in the study area,including 73 landslides,200 unstable slopes,and 7 ground subsidence.6 glacier movements were developed near Gonggar Mountain and Muztag Mountain.In addition,23 active debris flow ditches were found through field investigation and remote sensing interpretation.They are concentrated in the Gazi Valley,the Tashkurgan Basin and Khunjerab Pass.Analyzing the relationship between typical geological hazards and precipitation,in the summer when precipitation is concentrated,surface deformation accelerates,which can easily cause geological hazards.(3)Analyzing the environmental factors related to the occurrence of landslides in the study area,and finally the height difference,slope,aspect,curvature,topographic wetness index,normalized differential vegetation index,average annual precipitation,lithology,distance from faults and roads distance are used as factors in evaluating the susceptibility of landslides.Through the analysis of logistic regression model and random forest model,it is found that different factors have different influence on the landslide hazard,among which slope and distance from the road have the greatest influence on the landslide hazard.(4)In order to compare the advantages and disadvantages of the statistical model and machine learning model in the evaluation of the landslides susceptibility of Karakoram highway,the landslides susceptibility of the study area was evaluated based on logistic regression and random forest model.Susceptibility levels are divided into five categories: very low,low,medium,high,and very high.Both types of evaluation results show that the high-prone areas are concentrated in the entire section of the Gazi Valley,both sides of the Tashkurgan Valley,and Khunjerab Pass.The high and very high susceptibility in the logistic regression model accounted for 9.63% of the total results,and 58.17% of the historical landslides in the validation data set fell within the high and very high susceptible ranges.The AUC value under the ROC curve was 0.868;the results of random forest are 8.29%,98.79% and 0.981 respectively.From the aspect of verification accuracy,the random forest evaluation results were selected to participate in the optimization of landslides susceptibility assessment.(5)By establishing an optimization matrix of the deformation rate values and the susceptibility level values,the landslides susceptibility evaluation levels generated by the random forest model are optimized.The results showed that the susceptibility level of 2608 cells increased,of which the susceptibility level of 2176 cells increased by 1,303 susceptibility levels of cells increased by 2,119 susceptibility levels of cells increased by 3,10 Cell susceptibility rating increased by 4.