Spatial landslide susceptibility mapping using integrating an adaptive neuro-fuzzy inference system (ANFIS) with two multi-criteria decision-making approaches

• Published in: Theoretical and applied climatology Vol. 146; no. 1-2; pp. 489 - 509
• Main Authors: Paryani, Sina, Neshat, Aminreza, Pradhan, Biswajeet
• Format: Journal Article
• Language: English
• Published: Vienna Springer Vienna 01.10.2021; Springer; Springer Nature B.V
• Subjects: Accuracy; Adaptive systems; Aquatic Pollution; Artificial neural networks; Atmospheric Protection/Air Quality Control/Air Pollution; Atmospheric Sciences; Climate science; Climatology; Damage prevention; Decision making; Earth and Environmental Science; Earth Sciences; Economics; Fuzzy logic; Inference; Land use; Land use management; Landslides; Landslides & mudslides; Mapping; Methods; Model accuracy; Modelling; Multiple criterion; Original Paper; Sensitivity; Slope processes; Specificity; Susceptibility; Training; Waste Water Technology; Water Management; Water Pollution Control
• DOI: 10.1007/s00704-021-03695-w
• ISSN: 0177-798X

Landslide is a type of slope process causing a plethora of economic damage and loss of lives worldwide every year. This study aimed to analyze spatial landslide susceptibility mapping in the Khalkhal-Tarom Basin by integrating an adaptive neuro-fuzzy inference system (ANFIS) with two multi-criteria decision-making approaches, i.e., the best-worst method (BWM) and the stepwise weight assessment ratio analysis (SWARA) techniques. For this purpose, the first step was to prepare a landslide inventory map, which was then divided randomly into the ratio of 70/30% for model training and validation. Thirteen conditioning factors were selected based on the previous studies and available data. In the next step, the BWM and the SWARA methods were utilized to determine the relationships between the sub-criteria and landslides. Finally, landslide susceptibility maps were generated by implementing ANFIS-BWM and ANFIS-SWARA ensemble models, and then several quantitative indices such as positive predictive value, negative predictive value, sensitivity, specificity, accuracy, root-mean-square-error, and the ROC curve were employed to appraise the predictive accuracy of each model. The results indicated that the ANFIS-BWM ensemble model (AUC = 75%, RMSE = 0.443) has better performance than ANFIS-SWARA (AUC = 73.6%, RMSE = 0.477). At the same time, the ANFIS-BWM model had the maximum sensitivity, specificity, and accuracy with values of 87.1%, 54.3%, and 40.7%, respectively. As a result, the BWM method was more efficient in training the ANFIS. Evidently, the generated landslide susceptibility maps (LSMs) can be very efficient in managing land use and preventing the damage caused by the landslide phenomenon. Graphical abstract