| Landslides brought about huge casualties and property losses,its research with great practical significance and research value has become an essential part of eco-environmental construction in China.According to investigation,the colluvial landslide is one of the main kind of landslides occurred in our country,whose outbreak frequency is high and abruptness is strong.Meanwhile,the colluvial landslide is also the most widely distributed landslide type in China,especially prone to happen in those areas with frequent human activities or huge environment disturbance,it seriously restricts the economic development of the affected areas.Therefore,it is particularly important to study the colluvial landslide.For these reasons,this thesis chooses Zhushan County,the typical colluvial landslide prone area,as the study area,the regional geological background information is fully collected,the geological disaster survey is completed through a combination method of remote sensing interpretation and field investigation,the basic characteristics and development regularities of the colluvial landslide in study area is summarized subsequently.After that,the stability states of individual colluvial landslides are assessed and the regional landslide susceptibility analysis are completed.Based on above researches,the thesis gets some conclusions as below:1.The geological background profile and landslide distribution overview of the study area are briefly introduced.To provide the reliable basis for constructing stability evaluation index system as well as determining the landslide susceptibility conditional factors,the development regularities of colluvial landslide in Zhushan County are summarized as per geological environment conditions and morphological characteristics respectively.2.Depend on the above development regularities,stability evaluation index system for colluvial landslide is established consists of five components,fourteen indicators:(1)features of slope topography: hillslope morphology,average slope angle,relative relief;(2)features of sliding bed: stratum lithology,dip angle of rock stratum,slope structure;(3)features of sliding zone: dip angle of sliding surface,cohesion,internal friction angle;(4)features of sliding mass: thickness,permeability;(5)external factors: precipitation,engineering activities,water action of reservoir.3.Based on the above index system,take the typical 177 colluvial landslides that have specific information as the statistical objects.Firstly,the comprehensive contribution rate is used to quantify the evaluation indexes of these landslides.Then,the Fast ICA is used for feature extraction to obtain the independent components.The Fast ICA is one of the most typical algorithm of independent component theory.4.On the basis of these independent components,an intelligent evaluation model of monomer colluvial landslide stability is established by using particle swarm optimization support vector machine(PSO-SVM).The stability state of 177 colluvial landslides are assessed,then fifty colluvial landslides are picked to verify the evaluation model,the result shows that the PSO-SVM stability evaluation model is effective.5.Depend on the development regularities of colluvial landslide,this thesis chooses ArcGIS as the basic tool,regular grid as the evaluation unit.By considering the following five aspects: geological features,topographical features,hydrological features,climatic and environmental features,external induced factors,the initial conditional factors database for landslide susceptibility analysis are created.This database consists of eighteen initial conditional factors: engineering rock group,distance to geological structures,slope structure,slope angle,altitude,slope aspect,curvature,profile curvature,plan curvature,terrain roughness index,distance to drainage,stream power index(SPI),catchment area,topographic/bedding plane intersection angles(TOBIA),erosion topography index(LS),topographic wetness index(TWI),distance to road,average annual rainfall.The relationship between each factor and the landslide occurred is detailed analyzed by using three relation parameters,namely,the percentage of each factor class,the percentage of landslides fall in each factor class,the information content of each factor class.6.The Genetic algorithm-optimized rough set theory is used to reduce the redundant information of 18 initial factors as well as to obtain the core factors.Meanwhile,the correlation of 18 initial factors is judged by the Pearson correlation coefficient,thereby the factors with high correlation with other factors and have lower weight are removed to avoid the repeated calculation of weights.Finally,eight representative conditional factors of landslide susceptibility are achieved,namely,distance to road,engineering rock group,topographic/bedding plane intersection angles(TOBIA),slope angle,distance to drainage,slope structure,altitude,stream power index(SPI).7.A comprehensive landslide susceptibility evaluating model(LSI model)is established by calculating the landslide susceptibility index(LSI).(1)Before performing the analysis by means of back-propagation neural network(BPNN)models,the preprocessing of samples for locating the non-landslide area is done.Firstly,the significance of sample pretreatment is elaborated.Secondly,the filter criteria are displayed for locating the non-landslide area.Thirdly,the K-means cluster method and the two-step cluster method is applied to do the sample pretreatment respectively.Results reveal that the K-means cluster analysis cannot qualify the filter criteria,however,the two-step cluster method can meet.After validation,this paper demonstrates that an accurate sampling strategy out performs random sampling,when training a landslide susceptibility classifier.(2)The LSI model is established,and the corresponding LSI map is obtained.First,estimate the weights of 8 conditional factors by means of the BPNN models whose input layer consists of 8 core conditional factors as well as output layer are the values represent whether landslide occurred(“1” represent landslide area,“0” represent the non-landslide area).Then calculate the ratings for the different classes of the conditional factors using a probabilistic method.Finally,the weighted sum of the ratings and the weights is used as the LSI value for each pixel,these values were classified into five LSI classes using the natural break method in ArcGIS software to obtain the LSI map of Zhushan County.8.For the comparison of the comprehensive landslide susceptibility evaluating model,two commonly used landslide susceptibility evaluation models are established by using SPSS and ArcGIS software,namely,the logistic regression model and the information content model.Moreover,the corresponding LSI maps are obtained.During creating the information content model,the analytic hierarchy process(AHP)is used to calculate the weight of each conditional factor.Subsequently,three verification methods are used to verify the accuracy and reliability of these models,namely,the receiver operating characteristic curve(ROC)and area under the curve,the seed cell area index(SCAI),the existing landslide data comparison.The conclusions are as follows:By means of the ROC verification,the AUC values sorted from the largest to smallest are: the LSI model,the logistic regression model,the AHP-information content model.Then,the evaluation models that show better performance in the SCAI test are: the AHP-information content model and the LSI model.In the end,verification of the three maps are also performed by comparing them with the existing landslide data,the result shows that the LSI model has the most significant partition disparity,which leads to the best practicability. |
PDF Full Text Request