Shallow landslide susceptibility assessment in a semiarid environment - A Quaternary catchment of KwaZulu-Natal, South Africa

This paper presents an evaluation of landslide susceptibility maps based on the Frequency Ratio (FR) Landslide Susceptibility Index (LSI), logistic regression (LR), and artificial neural network (ANN) methods, applied within a Quaternary catchment in the KwaZulu-Natal region, South Africa. Landslide occurrence in this particular catchment is generally induced by heavy rainfall events where shallow landslips occur mostly on steep slopes thereby threatening informal housing at the base of slopes. The scope of this study is to compare the performance of landslide susceptibility models and rate the importance of landslide causal factors. Thirteen landslide causal factors, including altitude, slope angle, aspect, slope total curvature, slope plan curvature, slope profile curvature, proximity to stream network, Topographic Wetness Index (TWI), lithology, land cover, proximity to faults, proximity to dolerite sills and dykes, and proximity to the road network were considered in the study. Furthermore each causal factor was rated according to the above mentioned methods, and it was found that the most influential landslide causal factors are the, altitude, land cover, and lithology. The resulting susceptibility maps were classified into 5 classes ranging from Very low to Very high susceptibility. Model performance was tested using an independent validation set comprising 10% of all mapped landslides. For verification of the models' performance, receiver operating curves (ROCS) were calculated and the areas under curve (AUC) for the success rate curves were 0.736, 0.748, 0.827, and 0.809 for FR, LSI, LR and ANN respectively. The findings revealed that models showed promising results for shallow landslide susceptibility modelling since they all give accuracy values higher than 73%. The LR and ANN methods however proved to be superior in representing landslide susceptibility throughout the study area. (C) 2015 Elsevier B.V. All rights reserved.