Susceptibility assessment of regional landslides under different sampling strategies
Landslide susceptibility assessment is the first step in risk assessment and provides a data basis for early warning and prediction of geological disasters.In the process of susceptibility evaluation,the quality and quantity of the sample set are the key factors that determine the accuracy of landslide susceptibility modeling.In order to ex-plore how to obtain a high-quality landslide sample set,this paper takes the southern mountainous area of Shiyan City,Hubei Province as the study area,and selects evaluation factors such as slope and aspect that are related to the development characteristics of landslides in the study area.The paper establishes training and test data sets using four sampling strategies,i.e.,random sampling,landslide buffering sampling,information value con-strained sampling,and synthetic minority oversampling technique(SMOTE).Then,the paper select logistic re-gression model and support vector machine model for susceptibility evaluation,respectively,and compares and an-alyzes the accuracy of the four sampling strategies by using the ROC curves.The results show that the SMOTE sampling strategy has obtained the highest evaluation accuracy.Under this sampling strategy,the accuracy of the support vector machine model(0.930 4)is better than that of the logistic regression model(0.866 3),and more accurate evaluation results are obtained.This paper optimizes the landslide samples in the area through the mixed sampling technology,which provides a new idea for the selection of landslide susceptibility evaluation samples.
国家科技期刊平台
NSTL
万方数据
