The Mechanism Analysis And Susceptibility Mapping Of The Landslides Along The Maoding River At The Upstream Of The Jinsha River

Jinsha River is located in the upstream of the Yangtze River.There is a closely relationship between the development history of Jinsha River and the construction and safety of the local hydropower stations,highways,railways and ecological environment.Jinsha River has an effect on the production and ecological environment in the middle and lower reaches of the Yangtze River.Due to climatic and environmental changes brought by the uplift of plateau,geological disasters such as collapse,slide and flow occur frequently in the plateau and surrounding areas.The upper Yangtze River basin is one of the areas suffering the strongest effects of geological disasters,which has a great impact on human engineering and life.In this study,the Maoding River at the upstream of the Jinsha River is selected as study area.Due to the effects of rapid uplift and valley cutting,there are also widely distributed geological disasters in the study area.It is found that there are many landslides which may be the potential disasters for future human engineering activities.At the same time,the study area is located between the downstream of the proposed Xulong dam site and the upstream of the proposed Simuda dam site.In order to develop the water resources sustainably,it is necessary to study the mechanism and susceptibility of the landslides in this area.The main research contents and achievements of this paper are as follows:First,the mechanism analysis of the landslides.Based on the topography,geology and remote sensing data of the study area,with the help of 3S(i.e.GPS,RS,and GIS),field investigation,and interpretation of remote sensing,we describe the spatial distribution characteristics of the landslides on both sides of the river.It is found that there are large scale and intensive distributed landslides in the study area,and the number on right bank is obviously larger than that on left bank.There are two types of landslide genesis mechanism in the study area,which are flexural and tensile crack(67.9%)and creep tensile crack(32.1%).Based on the analysis of the different factors(lithology,bank slope structure,etc.)affecting the landslide genesis mechanism,it is concluded that lithology,discontinuity orientation,bank slope structure,and the stage of deformation have great influences on the landslide genesis mechanism.Second,the kinematic analysis of slope.Through remote sensing technology to determine the orientation of a slope,combined with the field investigation of the structural plane(such as bedding plane,random joint),the characteristics of the structural plane developed on the slope can be described.Then,based on the right hand rule,the structural plane can be represented by a steronet combined with the rose diagram and orientation uniquely.Then the structural planes are divided into four groups by density point method,and the properties of each group are described in detail.Based on the theory of kinematics,the possible failure modes of the slope are determined,and the safety angle of the slope is evaluated by using stereographic projection technic and probabilistic kinematics analysis.By judging the failure modes of the potential failure blocks,the safety angles of the upper and lower parts of Gonda landslide are obtained.Third,the evaluation of landslide influencing factors.In combination with the previous research basis and the actual situation of the research area,the evaluation system of landslide susceptibility is preliminarily constructed by selecting 11 evaluation indexes that may affect the occurrence of landslide,and a detailed classification is carried out for each influencing factor.The spatial database of each influencing factor is constructed by DEM,geological map,remote sensing image and other basic information sources,and the influence of the seismic factor on landslide modeling is eliminated by preliminary judgment of each influencing factor.Then,based on the statistical theory,with the help of Arcgis platform on data management,SPSS statistical analysis module,Excel and Matlab on information processing,through the relative area density method(frequency ratio method),we analyze the importance of each factor classification interval.Further,the correlation of each influencing factor of the landslides is studied.And the correlation among rainfall,vegetation,and elevation is found to be highly correlated by variance inflation factor and tolerance method(VIF&T)and Pearson's correlation coefficient method,which provided guidance for the subsequent dimensional reduction analysis of the model.Finally,based on the informatics theory,we have deeply screened the influencing factors with good prediction ability.Based on information gain method,information gain ratio method and 5-fold cross-validation technique on a group of train data,we sort the prediction ability of the influencing factors.After that,the results are verified with the basic data of the study area.Curvature and slope aspect are proved to have poor prediction ability and removed from the landslide susceptibility model.Thus,a total of eight factors,i.e.,rainfall,vegetation,elevation,lithology,distance to river,relief,distance from fault,and slope angle,are selected to use in the prediction model.This procesure can provide solid data foundation for the subsequent landslide susceptibility modeling.At last,the analysis of landslide susceptibility models and the evaluation of the prediction results.Based on the selected landslide influencing factors,two models were constructed to evaluate the landslide susceptibility in the study area.The first model is AHP-PCA-ICM.This model integrates people's subjective judgment in determining the weight through analytic hierarchy process(AHP),and then adjusts the weight through the objective principal component analysis(PCA).Finally,the susceptibility is evaluated by applying the information content model(ICM).The second model is PCA-PSO-SVM.The model is completed through objective intelligent algorithm based on machine learning.It firstly solves the collinearity problem between the different impact factors by principal component analysis(PCA).Then,particle swarm optimization(PSO)algorithm is used to find the optimal model parameters,and finally with the help of Arcgis,Matlab and SPSS software,we apply support vector regression machine model(SVM)to evaluate landslide susceptibility.The comparative analysis of the effectiveness of the two models by using Sridevi Jadi experimental probabilistic precision analysis method,ROC curve,and Kappa index shows that the PCA-PSO-SVM model has advantages over the prediction results of AHP-PCA-ICM model,and that the PCA-PSO-SVM model is very suitable for the study of landslide sensitivity in this region.