Quantitative assessment of landslide susceptibility on the Loess Plateau in China

Abstract

ABSTRACT This study was undertaken to produce landslide susceptibility maps by the frequency ratio (FR) and weight-of-evidence (WOE) methods for the Qingshui River Basin, and compare three combinations of different controlling factors to get the best number for analysis. Since conditioning factors create suitable conditions for landslides, 11 such parameters were used for this study: slope angle, aspect, altitude, valley depth, lithology group, distance to water bodies, stream power index, topographic wetness index, longitudinal curvature, cross-sectional curvature, and relief. Performances of models with 6, 8, and 11 of these factors were evaluated using two models to obtain reliable landslide susceptibility maps, investigate the effect of different numbers of factors, and determine the most effective. The receiver operating characteristic (ROC) curve and the area under the ROC curve (AUC) were used to verify the accuracy of the landslide susceptibility assessment results. AUCs for the prediction rate curve of FR and WOE, with 6, 8, and 11 landslide variables, were 0.765, 0.731, 0.702 and 0.771, 0.728, 0.717, respectively. The results indicate that WOE model performed better than the FR model in the basin and that accuracy of evaluation decreases (rather than increases) with an increase in number of variables. Abbreviations: FR: frequency ratio; WOE: weight-of-evidence