Research On Collapse And Landslide Susceptibility Assessment Method Based On Machine Learnin

Collapse and landslide geological disasters seriously endanger people’s life and property safety,causing serious casualties and economic losses annually in the world.In Guangzhou’s Baiyun District,the serious slope weathering combined with the large population,numerous engineering activities and frequent extreme weather such as floods and typhoons during the flood season lead to the frequent occurrence of geological disasters,among which collapse and landslide are the most important.This has a negative impact on the construction and operation within the district.In recent years,government departments have actively promoted the work of geological disaster prevention and control,among which the zoning of collapse and landslide geological disaster-prone areas is the cornerstone for the progress of prevention and control work.There are many models currently used for research on the susceptibility of landslide and collapse geological disasters,and in recent years,machine learning models have been widely applied.Due to the influence of modeling data on the accuracy of machine learning model algorithms,this paper uses three machine learning models to analyze the susceptibility of landslide in Baiyun District and compares it with the traditional information model.The main contents and results of this paper are as follows:1.Collect information on natural geography,stratigraphic geological environment and human activities in Baiyun District,combine with the results of the field investigation of the collapse and landslide disasters of Baiyun District in the 13 th FiveYear Plan,development characteristics and spatial and temporal distribution rules of landslide hazards in Baiyun District.2.The principle of machine learning is explained by a machine learning linear regression algorithm,and illustrates the workflow of machine learning.The three models that will be used to study landslide susceptibility in this paper are classification and regression tree model,random forest model and gradient boosting decision tree model.3.Based on the detailed analysis of landslide hazard influencing factors,development characteristics and spatial and temporal distribution rules,12 landslide susceptibility factors in the study area were selected.The relationship between each factor and the distribution of hazard points is also presented in the form of a graph,and the factors are graded and the relationship between each level and the number and relative density of hazard points is analyzed using statistical methods.4.The distance from the landslide point,the evaluation result of the information quantity model and the high-definition remote sensing image map are considered in the selection of the non-landslide point,and the features are pre-processed to improve the accuracy of the machine learning model.The traditional information quantity model for landslide susceptibility evaluation is influenced by some factors,and the evaluation results are prone to be inconsistent with the actual situation.In this paper,random forest model,gradient boosting decision tree model,and classification and regression tree model are used to evaluate the susceptibility of avalanche landslide in Baiyun District,and the accuracy index and AUC value derived from confusion matrix are used as evaluation index to compare the comprehensive accuracy of the three models.5.The random forest model is used to evaluate the importance of influence factors by calculating the Gini index of each feature that can assess the feature importance.The random forest algorithm with the best comprehensive accuracy was selected to derive the results of the avalanche landslide susceptibility evaluation in Baiyun District,and the equal interval method was used to divide the Baiyun District avalanche landslide susceptibility zone into high susceptibility zone,relatively high susceptibility zone,medium susceptibility zone,relatively low susceptibility zone and low susceptibility zone,which can be directly used in the susceptibility zoning of Baiyun District avalanche landslide geological hazards.6.The evaluation results and the distribution of the issued disaster sites were overlaid and compared,and it was found that the evaluation results are more consistent with the objective situation and have good reference.