Study On The Influence Of Sample Selection On Geological Disaster Susceptibility Assessment

Affected by the local geological environment and human engineering activities,Liulin County of Shanxi Province suffers from frequent geological disasters,which threaten the local social development and people’s life and property safety.The high accuracy of geological disaster susceptibility evaluation is very important for geological disaster prevention and control.The reasonable selection of samples is very important to improve the accuracy of susceptibility evaluation.Based on the investigation and collation of geological environment data and geological disaster data in Liulin County,the natural geography,geological structure and human engineering activities in the study area are analyzed.Based on GIS technology,the two selection conditions of positive and negative sample proportion and influence distance are analyzed and studied.Negative samples are selected according to "distance","independence" and "randomness".The random forest algorithm was used to evaluate the susceptibility of geological disasters to models with different sample proportions and influence distances.The main achievements are as follows:According to the analysis of 32 landslides,71 landslides,1 debris flow and 47 unstable slopes,it is found that the geological disasters caused by human engineering activities are significantly higher than the natural factors.Geological disasters are mainly distributed in the central and northern Liulin County along both sides of roads and rivers.Based on the previous research experience and the mechanism of geological disaster in Liulin County,16 evaluation factors were preliminarily selected.Through correlation and factor analysis,the coefficient of variation of four variables,namely slope,surface roughness,surface cutting depth and elevation,is highly correlated,and aspect,distance from fault and population density have little influence on the occurrence of geological disasters.Finally,nine factors,including elevation,rainfall and normalized vegetation index(NDVI),were selected to construct an evaluation system of geological disaster susceptibility in the study area.Random forest algorithm was used to construct the susceptibility model,and error analysis,confusion matrix and ROC curve were used to test each model.The high and very high susceptibility areas in the study area are mainly distributed in Mucun Town,Liulin Town,the middle of Lijiawan Town and the west of Chengjiazhuang,which are the key areas for disaster prevention and reduction.Distance from roads,distance from water system,NDVI and rainfall are the main factors affecting the occurrence of geological disasters in the study area.The main reason for the difference in the evaluation results of susceptibility under different sampling conditions is the change in the "representativeness" of randomly selected negative samples.The results show that different sample proportion and influence distance have great influence on the prediction accuracy.The sample ratio is 1:10,and the model with influence distance of 1000 m is optimal.This paper has 27 figures,22 tables and 72 references.