Feature Processing Method In Rainfall- Induced Landslide Susceptibility Assessment

Landslides are one of the most widespread geological hazards throughout the world, which trend to be increasingly frequent in mountainous areas due to the accelerating urbanization. Landslides seriously threat the safety of people’s life and property. Landslides are the coupling action result of many factors. Of various landslide disasters, the rainfall-induced landslides have the widest spatial distribution and the highest frequency of occurrence. Effective assessment and prediction of rainfall-induced landslides susceptibility is a worldwide problem due to the suddenness and uncertainty of landslides. To reduce the casualties and property losses caused by landslides, rainfall-induced landslide susceptibility assessment has become an important and meaningful issue in the risk mitigation research for mitigation and prevention.According to the domestic and foreign research results, characteristics of rainfall-induced landslides were analyzed. On the basis of analyzing the influence of various features to landslides, a certainty factor feature weighted method was proposed to reflect different impact of features on landslides.4 feature sets and 3 training sets were contructed and applied to 3 models. A total of 36 rainfall-induced landslide susceptibility assessment models were built for comparison and analysis. Detail contents are as follows:(1) The spatial and temporal distribution characteristics of rainfall-induced landsldes were analyzed and summarized. Based on the impact of rainfall on landslides,3 types of rainfall triggering landslides were proposed:rainstorm, continuous precipitation and antecedent precipitation.(2) Different impact of various features on landslides were studied. Given 3 types of rainfall triggering landslides, a trinity comprehensive rainfall representation method was proposed, expressing daily rainfall, continuous raining days, antecedent effective rainfall, which improved the calculation of the rainy decay factor by considering the topography and vegetation impact.4 feature sets were contructed based on feature selection.(3) To highlight the different impact of various features on landslides, a certainty factor feature weighted and feature quantification method was proposed. According to the sensitivity of different feature intervals/categories on landslides, features were quantified by certainty factor. Landslide certainty factor ratio was proposed to weight features.3 training sets were contructed by overlay statistics sample construction method, including non-feature weighted training set, Pearson relative coefficient feature weighted training set and certainty factor feature weighted training set 3 training sets combined with 4 feature sets were applied to LR, ANN and SVM models respectively for comparison analysis.Wencheng County is located in southeast coast of Zhejiang Province, China, which is steeply mountainous and affected by typhoon seriously. Rainfall-induced landslides occur every year. Therefore, Wencheng County was chosen as the study area in this paper. The experimental results showed that the occurrence mechanisom of rainfall-induced landslides was very complex and the effects of various features on landslides differed markedly, so it is essential for feature selection, feature quantification and feature weighted. The result of feature selection was closely related with feature processing method and construction model. The certainty factor feature weighted and feature quantification method proposed in this papar could improve the landslide susceptibility assessement result, which is an important supplement to rainfall-induced landslides susceptibility assessment work.