Evaluation Of Landslide Susceptibility And Software System Development Based On Various Machine Learning Methods

The evaluation of landslide susceptibility is usually considered as the first step in the evaluation of landslides by analyzing different combinations of geological conditions that affect landslides and evaluating the geographical spatial distribution of potential landslides.In the past,landslide susceptibility evaluation empirical models or semi-quantitative models based on the Arc GIS platform were subject to a certain degree of subjectivity.However,machine learning algorithms are widely used because they train models through samples and can obtain results objectively.However,the Arc GIS platform is not compatible with machine learning algorithms.Most of the researchers use Arc GIS to extract and process evaluation factors,and then export the factor data to use other software for model training.Such multiplatform switching makes the operation cumbersome and complicated,and the error rate is higher.At the same time,it is impossible to streamline and integrate this set of operations into a simple,fast,and automated susceptibility evaluation system that does not require a lot of manual operations.Based on the theories of machine learning and GIS spatial analysis,this paper studies the automated landslide susceptibility evaluation system.Mainly by summarizing and comparing the methods and results of the existing landslide susceptibility evaluation system,the evaluation factor data extraction,factor grading,feature data preprocessing and machine learning model prediction of prone areas are discussed,and the Python language is used to These technologies are programmed and integrated into a complete landslide-prone automated partitioning program system to solve the aforementioned multi-platform operation problems.The main results achieved are as follows:(1)Summarize a set of landslide automated susceptibility evaluation system with wide applicability and great procedural feasibility,and provide technical guidance for subsequent realization of landslide automated susceptibility evaluation.On the basis of the existing landslide susceptibility evaluation system,the grid unit is used as the basis for the evaluation unit,the improved fast clustering method is used to screen the non-landslide units,the evaluation index system is established by 7 evaluation factors,and the frequency is compared.Method for factor grading,four typical machine learning algorithms as the evaluation model,ROC curve and accuracy as the evaluation standard,and natural breakpoint method as the zoning method for the automatic landslide susceptibility evaluation system.(2)Using Python language programming to realize the evaluation factor data extraction,grading,preprocessing and output thematic maps and other technologies.Taking Lantian County as an example,the data extraction algorithms for the seven factors of elevation,slope,aspect,profile curvature,stratum lithology,rainfall,and vegetation coverage in the evaluation index system are studied,and the extracted factors are compared with the frequency ratio method.The data is classified and analyzed,and finally the data set is processed with missing value processing,coding processing and non-dimensional processing,and transformed into a data set that can be directly used for model training and prediction.(3)The frequency ratio method is used to classify the factors,and the correlation between the evaluation factors and the occurrence of landslides is analyzed.The statistical results show that: in the range of 500m-1000 m in elevation,the slope is in the range of 10°-35°,the rainfall is in the range of 650mm-800 mm,the NDVI is in the range of 0.4-0.6,the aspect is west and northwest,slope type It is concave,and the stratum lithology is that the frequency ratio of the landslide on the loose rock group and the soft and hard rock group is greater than 1,which is closely related to the occurrence of landslides.(4)Program calls and parameter optimization of logistic regression model,multi-layer perceptron model,support vector machine model,and XGboost model were implemented using Python language programming,and Lantian County was evaluated for landslide susceptibility.The classification accuracy,ROC curve and AUC value are used to evaluate the performance of the model.The calculation results show that the classification accuracy of the four models are 84.43%,82.81%,91.31%,and 88.15%;the four models under the ROC curve The area AUC values were 0.9024,0.9031,0.9287,0.9108,respectively.Combining the two results shows that the support vector machine model has the best generalization performance in the evaluation of landslide susceptibility in this study area.(5)The feasibility of the program system implemented this time is tested by using four models to predict the proportion of historical disaster points and the frequency ratio in the prone areas.The statistical results show that the proportions of landslides in the high-prone areas of the four models are 64.57%,54.33%,80.32%,and 69.05% respectively,indicating that the landslides are mainly concentrated in the high-prone areas,which is more in line with the actual situation;the landslide frequency ratios in the high-prone areas are respectively 3.92,1.78,3.01,3.76,all of which are far greater than 1,all show strong correlation.Combining the two shows that the program developed this time is more specific and feasible.(6)Designed the work flow of the two modules of the automated landslide susceptibility evaluation system,and developed the landslide automated susceptibility evaluation system.The system integrates evaluation factor extraction and preprocessing modules,evaluation model calculation and prediction result verification modules,and realizes factor selection,factor extraction,factor grading,data preprocessing,output thematic maps,model selection,model parameter configuration,calculation results and The display of susceptibility zoning results can easily and quickly evaluate landslide susceptibility.