Evaluation of prediction capability of the MaxEnt and Frequency Ratio methods for landslide susceptibility in the Vernazza catchment (Cinque Terre, Italy)

Description

The research is focused on the Vernazza catchment, an area of 5,75 km2 belonging to the Vernazza municipality in the Cinque Terre National Park, Italy; here, landslide susceptibility maps are produced using two different statistical methods by analyzing several intrinsic factors controlling landslides. It is also intended to evaluate the maps to determine the comparison between the coverage of high susceptibility areas obtained through different methods. The first statistically based method is a presence–absence (Frequency Ratio) method, while the second one is a presence-only (MaxEnt) method; the acquisition and preparation of the predisposition factors are also described, as well as their sensitivity and hierarchy regarding the landslide susceptibility models. Furthermore, in order to understand the effective improvement brought by the performance of the models, a validation using the receiving operator characteristics (ROC) and the area under curve (AUC) techniques has been carried out. The role played by variables such as land use and FAS is well visible: the outputs generated through both methods show a uniform distribution of very high susceptibility values on the medium-lower right portion of the catchment, and also the "aspect" variable, in which the value of each cell in a dataset indicates the direction of the cell's slope faces, strongly influences the results since the west–south west-facing cells are considered as prone to generate landslides. Results obtained for assessing landslide susceptibility show good prediction rate curves for both the tested methodologies, with higher values for the frequency ratio susceptibility model. However, for the MaxEnt susceptibility models, these values are lower, though, without ever decreasing below 0.60. In both cases, future developments of the adopted methods could involve a further distinction of landslide type to evaluate the potential of model prediction specifically for each landslide category.

Additional details