Abstract:

Landslide is one of the three major natural hazards in China. It is, therefore, very important to study how to perform landslide risk assessment and zoning. The related studies may provide policymakers with theoretic basis in formulating strategies and policies for control of landslides. On the basis of the field surveys and relevant researches already done, 11 factors, such as terrain, lithology, vegetation, land use, precipitation, fault, human activities, etc. have beer used as evaluation indexes in performing landslide risk assessment for Huadu District with the aid of the logistic regression-weighted support vector machine model. Landslide risks in the region were sorted into 5 grades, and in the end the model in fitting accuracy with ROC curve has been verified. Results show that the risk was quite high in a large portion of Timian Town, and a certain portion of Huadong and Chini towns, which is spatially consistent with distribution of the 65 landslide disaster inventory data; The regions, very low, low, moderate, high and very high in risk accounted for 28.19%, 31.31%, 25.54%,1.73% and 3.24%, respectively, of the district in area. Verification of fitting accuracy with ROC curve shows that the model of logistic regression-weighted SVM may be used effectively to assess landslide risks in this region, with fairly high assessment accuracy, grading ability and objectivity.

Key words: landslide hazard, risk assessment, logistic regression, support vector machine, Huadu District