Effect of using various weighting methods in a process of landslide susceptibility assessment

Abstract

This study discusses the evaluation of the effect of using different weighting approaches in the process of landslide susceptibility assessment. Weighting process is needed, especially for landslide susceptibility assessment using bivariate statistical analysis, and can radically affect the resulting susceptibility map. The bivariate analysis belongs to a set of quantitative methods. The initial point for the bivariate analysis is selection and processing of input factors in the form of parametric maps, factors that play a dominant role in slope stability. The parametric maps were subsequently evaluated related to landslide inventory map. Another essential part of the bivariate analysis is the determination of the weight of the given input factors. Herein, four methods were applied to determine the weights of each class within reclassified input factors, as well as the total weight of the individual input factors. As a study area, the district of Kysucke Nove Mesto in Slovakia was chosen. Four prognostic maps were the result using entropy index, AHP method (AHP—analytic hierarchy process) (the input factors weighted as a whole), frequency ratio and landslide index (the weights were calculated for each class of input factor). Final landslide susceptibility maps were verified trough ROC curves (ROC—receiver operating characteristics). The accuracy of maps was ascertained by the size area (AUC—area under curve) under ROC curve. The highest accuracy was obtained for maps using weights calculated from the landslide index (88.5%) and the frequency ratio (88.4%).