Landslide susceptibility assessment based on K-means cluster information model in Wenchuan and two neighboring counties, China

The study of landslide susceptibility evaluation is of great significance to both zoning of geological disasters and disaster prevention strategies. Taking Wenchuan and two surrounding counties (Li County and Mao County), which are prone to landslides, as an example, K-means cluster information model for landslide susceptibility mapping is proposed. After seven impact factors, i.e., slope angle, elevation, aspect, distance from the structure, distance from the water system, formation lithology and the land usage, are selected, the secondary classification for factors is carried out. The former five impact factors mentioned above were classified separately by K-means cluster analysis according to 159 landslide samples. At the sametime, the traditional isometric classification was also presented to compare with the K-means clustering method. The latter two impact factors were classified qualitatively. According to the differences of the above secondary classification methods and whether the landslide sample considers the area factor, the information model is subdivided into four categories (model a: K-means clustering quantitative model, model b: isometric classification quantitative model, model c: K-means clustering area model, and model d: isometric classification area model). The information of each secondary index was calculated separately, and the information distribution of the study area was obtained through spatial overlay analysis of ArcGIS. Then, the landslide susceptibility of the study area was divided into five grades by natural breakpoint method. Taking the principle of increasing susceptibility and Area Under Curve (AUC) as the accuracy evaluation indicators, three results were obtained. First, the overall effect of K-means clustering models (model a and model c) is better than that of isometric classification models (model b and model d). Second, the area models (model c and model d) are generally better than the quantitative models (model a and model b) under the same classification method. Third, With the above two advantages, the evaluation accuracy of model c is significantly improved compared with model b, and its AUC value is increased from 80.46% to 87.25%.