Applying Information Theory and GIS-based quantitative methods to produce landslide susceptibility maps in Nancheng County, China

The main objective of the present study was to produce a landslide susceptibility map by implementing a novel methodology that combines Information Theory and GIS-based methods for the Nancheng County, China, an area with numerous reported landslide events. Specifically, the information coefficient that is estimated from Shannon’s entropy index was used to determine the number of classes of each landslide-related variable that maximizes the information coefficient, while three methods, logistic regression, weight of evidence, and random forest algorithm, were implemented to produce the landslide susceptibility map. The comparison of the various models was based on the assessment of a database of 112 past landslide events, which were divided randomly into a training dataset (70 %) and a validation dataset (30 %). The identification of the areas affected was established by analyzing airborne imagery, extensive field investigation, and the examination of previous research studies, while the morphometric variables were derived using remote sensing technology. The geo-environmental conditions in those locations were analyzed regarding their susceptibility to slide. In particular, 11 variables were analyzed: lithology, altitude, slope, aspect, topographic wetness index, sediment transport index, profile curvature, plan curvature, distance to rivers, distance to faults, and distance to roads. The comparison and validation of the outcomes of each model were achieved using statistical evaluation measures, the receiving operating characteristic, and the area under the success and predictive rate curves. Each model gave similar outcomes; however, the random forest model had a slightly higher predictive performance in terms of area under the curve (0.9220) against the ones estimated for the weight of evidence (0.9090) and the logistic regression model (0.8940). The same pattern of performance was reported when the success power of the models was calculated. Random forest was slightly better than the other two models in terms of area under the curve (0.9350) in comparison with the weight of evidence (0.9255) and logistic regression (0.9097). The predictive performance was estimated by using the validation dataset, while the success power of the models was estimated by using the training dataset. From the visual inspection of the produced landslide susceptibility maps, the most susceptible areas are located at the west and east mountainous areas, while moderate to low susceptibility values characterize the central area.