Comparison and combination of different models for optimal landslide susceptibility zonation

It is important to compare different methods and apply combined models for landslide susceptibility zonation on a regional scale for land-use planning and hazard mitigation. The purpose of this study is attempt to obtain an optimal landslide susceptibility zonation in a severely landslide affected region where the available data are very limited. Six single models (analytical hierarchy process (AHP), logistic regression (LR), fuzzy logic (FL), weight of evidence integrated logistic regression (WL), artificial neural network (ANN) and support vector machine (SVM)), were applied to obtain the single landslide susceptibility zonations along the middle reaches of the Bailong River from Zhouqu to Wudu, southern Gansu, China, then these single models were compared, after which the three single models that performed better (LR, ANN and SVM) were selected to prepare the combined zonations. Six conditional independent environmental factors were selected as the explanatory variables that contribute to landslide occurrence (elevation, slope, aspect, distance from fault, lithology and settlement density). The mapped landslides in this region were randomly partitioned into two sets: 80% of the landslides were used for the model training and the remaining 20% were used for validation of the models. Receiver operating characteristic and cost curves were plotted as means of evaluating the quality of the susceptibility zonations for the single and combined models. Results show that the single LR, ANN and SVM are models with superior prediction performance and are more suitable for constructing the combined models in this study. Compared with single models, the combined models provided an improved prediction capability and reduced uncertainties.