Landslide susceptibility maps using logistic regression model for the Oria river catchment (Gipuzkoa province). Strategies for variables processing

Abstract

We present a study in the Oria River basin (882 km2) (Gipuzkoa), where we carried out a comparative analysis among different input data processing strategies for the study of the susceptibility of landslides using a logistic regression model. The three adopted strategies were: (i) no variable (continuous or categorical) was transformed; (ii) the continuous variables were transformed into categorical by dividing the distribution of the values in equal-range classes; and (iii) the categorical variables were transformed into continuous by calculating the landslide density for each class. Within these three strategies, we considered a landslide inventory obtained from fieldwork and used a common set of variables, selected from an objective statistical procedure. We determined the advantage of transforming categorical variables into continuous by comparing the results of a logistic regression model applied to the different strategies. However, the statistical procedure for the selection of variables was unable to detect the bias produced by one of the explanatory variables, which was highlighted only after generating the susceptibility maps. Therefore, we performed a fourth Logistic Regression model run, and its map, considering the results of the first three.