Modeling the potential distribution of shallow-seated landslides using the weights of evidence method and a logistic regression model: a case study of the Sabae Area, Japan

Abstract

A number of statistical methods are typically used to effectively predict potential landslide distributions. In this study two multivariate statistical analysis methods were used (weights of evidence and logistic regression) to predict the potential distribution of shallow-seated landslides in the Kamikawachi area of Sabae City, Fukui Prefecture, Japan. First, the dependent variable (shallow-seated landslides) was divided into presence and absence, and the independent variables (environmental factors such as slope and altitude) were categorized according to their characteristics. Then, using the weights of evidence (WE) method, the weights of pairs comprising presence (w +(i)) or absence (w −(i)), and the contrast values for each category of independent variable (evidence), were calculated. Using the method that integrated the weights of evidence method and a logistic regression model, score values were calculated for each category of independent variable. Based on these contrast values, three models were selected to sum the score values of every gird in the study area. According to a receiver operating characteristic curve analysis (ROC), model 2 yielded the best fit for predicting the potential distribution of shallow-seated landslide hazards, with 89% correctness and a 54.5% hit ratio when the occurrence probability (OP) of landslides was 70%. The model was tested using data from an area close to the study region, and showed 94% correctness and a hit ratio of 45.7% when the OP of landslides was 70%. Finally, the potential distribution of shallow-seated landslides, based on the OP, was mapped using a geographical information system.