GIS-based ensemble modelling of fuzzy system and bivariate statistics as a tool to improve the accuracy of landslide susceptibility mapping

The primary objective is to propose and verify an ensemble approach based on fuzzy system and bivariate statistics for landslide susceptibility assessment (LSA) at Azarshahr Chay Basin (Iran). In this regard, various integrations of fuzzy membership value (FMV), frequency ratio (FR), and information value (IV) with index of entropy (IOE) were investigated. Aerial photograph interpretations and substantial field checking were used to identify the landslide locations. Out of 75 identified landslides, 52 (approximate to 70%) locations were utilized for the training of the models, whereas the remaining 23 (approximate to 30%) cases were employed for the validation of the models. Fourteen landslide conditioning factors including altitude, slope aspect, slope degree, lithology, distance to fault, curvature, land use, distance to river, topographic position index (TPI), topographic wetness index (TWI), stream power index (SPI), normalized difference vegetation index (NDVI), distance to road, and rainfall were prepared and utilized during the analysis. The FMV_IOE, FR_IOE, and IV_IOE models were designed utilizing the dataset for training. Finally, to validate as well as to compare the model's predictive abilities, the statistical measures of receiver operating characteristic (ROC), including sensitivity, accuracy, and specificity, were employed. The accuracy of 92.7, 92.5, and 91.8% of the models such as FMV_IOE, FR_IOE, and IV_IOE ensembles, respectively, was by the area under the receiver operating characteristic (AUROC) values developed from the ROC curve. For the validation dataset, the FMV_IOE model had the maximum sensitivity, accuracy, and specificity values of 95.7, 91.3, and 87.0%, respectively. Thus, the ensemble of FMV_IOE was introduced as a promising and premier approach that could be used for LSA in the study area. Also, IOE results indicated that altitude, lithology, and slope degree were main drivers of landslide occurrence. The results of the present research can be employed as a platform for appropriate basined management practices in order to plan the highly susceptible zones to landslide and hence minimize the expected losses.