Comparative performance of new hybrid ANFIS models in landslide susceptibility mapping

Abstract

Many landslides occur in the Karun watershed in the Zagros Mountains. In the present study, we employed a novel comparative approach for spatial modeling of landslides given the high potential of landslides in the region. The aim of the study was to combine adaptive neuro-fuzzy inference system (ANFIS) with grey wolf optimizer (GWO) and particle swarm optimizer (PSO) algorithms using the outputs of qualitative stepwise weight assessment ratio analysis (SWARA) and quantitative certainty factor (CF) models. To this end, 264 landslide positions and twelve conditioning factors including slope, aspect, altitude, distance to faults, distance to rivers, distance to roads, land use, lithology, rainfall, plan and profile curvature and TWI were then extracted considering regional characteristics, literature review and available data. In the next step, the multi-criteria SWARA decision-making model and CF probability model were used to evaluate a correlation between landslide distribution and conditioning factors. Ultimately, landslide susceptibility maps were generated by ANFIS-GWO and ANFIS-PSO hybrid models and the accuracy of models was assessed by ROC curve. According to the results, the area under the curve (AUC) for the hybrid models ${\text{ANFIS - GWO}}_{{\text{SWARA}}}$, ${\text{ANFIS - PSO}}_{{\text{SWARA}}}$, ${\text{ANFIS - GWO}}_{{\text{CF}}}$ and ${\text{ANFIS - PSO}}_{{\text{CF}}}$ was 0.789, 0.838, 0.850 and 0.879, respectively. The hybrid models ${\text{ANFIS - PSO}}_{{\text{CF}}}$ and ${\text{ANFIS - GWO}}_{{\text{SWARA}}}$ showed the highest and lowest prediction rate, respectively. Moreover, CF outperformed the SWARA method in terms of evaluating correlation between conditioning factors and landslides. The map produced in this study can be used by regional authorities to manage landslide risk.

Graphic abstract