Study On Landslide Susceptibility Assessment And Zoning In Zhejiang Province Based On GIS And Machine Learning

Zhejiang Province is one of the provinces with serious geological hazards and landslides in China.In recent years,various losses and a series of adverse effects caused by frequent landslides have attracted wide attention from all walks of life.Therefore,it has important application value and practical significance to carry out landslide hazard susceptibility evaluation and zoning research for the economic and social development of the study area.In this paper,Zhejiang Province is taken as the research area,aiming at the widespread landslide geological disasters(a total of 3248).Based on the combination of ArcGIS and historical landslide disaster data,the influencing factors of landslide formation in the area are analyzed,and select the landslide impact factors,and use ArcGIS software to classify the status,so as to establish the evaluation index system of landslide disaster susceptibility in the research area.Then,the establishment of landslide susceptibility evaluation model and the study of susceptibility zoning in the study area are carried out by the statistical analysis method based on data drive and machine learning.The following three main achievements have been achieved in this paper:(1)Research and Analysis on influencing factors and distribution characteristics of landslide disasters in Zhejiang Province,Finally,the susceptibility evaluation index system of landslide disasters in the study area is established based on 9 factors,including slope,slope direction,elevation,curvature,stratum,rainfall,distance to fault zone,distance to water system and road.(2)By selecting 9 influencing factors of landslide,based on the proportion of landslide under each index factor,area proportion of each grade and density of landslide points,the status classification and correlation statistical analysis of each index factor are carried out.(3)Evaluation model(information quantity model,deterministic coefficient model,stochastic forest model and I+RF model,CF+RF model combination model)based on combination of GIS and machine learning and data drive to evaluate the susceptibility of landslide disasters in research area.According to the comparative analysis of evaluation zones of each model,and the comparison of accuracy test under each model,I+LR model is selected as the optimum model for this study,and the results of zoning are in good agreement with the historical landslide disaster distribution.The high risk area and high risk area of landslide accounted for 21.86% of the study area,and the distribution proportion of landslide disasters was 84.65%.It is mainly distributed in mountainous and hilly areas in southern Zhejiang Province,northwest and Southeast mountainous areas.These areas are more susceptibleThe results show that the statistical analysis method based on machine learning and data driving is applicable to the susceptibility assessment of landslide disasters in large areas of provincial level with good scientificalness and high accuracy.At the same time,the susceptibility assessment compared with CF model,information quantity model and random forest model,The combination model of stochastic forest model based on information quantity and stochastic forest model based on deterministic coefficient increases the relevant information that takes into account the factors of each index,i.e.weighting factor,so the evaluation accuracy is higher.