A comparison of statistical and deterministic methods for shallow landslide susceptibility zoning in clayey soils

Abstract

Shallow landslides are widespread in different geological contexts and generally occur as multiple events over large areas. When these phenomena involve fine-grained soils, they may cause serious consequences—in terms of environmental and property damages—and thus their spatial forecasting becomes a relevant issue for land use planning and design purposes. The existing literature provides several methods for landslide susceptibility assessment, categorized in qualitative and quantitative methods. When dealing with analyses at large scale (1:5000), quantitative methods are generally preferred. In this paper, landslide susceptibility maps are produced for a study area prone to shallow landsliding, located in Catanzaro (southern Italy). To this aim, two quantitative methods are implemented: the statistical “information value method” and the deterministic “TRIGRS model.” The two approaches are compared by means of two indicators of the grade of correctness of the landslide susceptibility maps: the area under curve of the ROC curve, AUC, and the overestimation index, OI. The results of the analyses in terms of AUC values demonstrate the effectiveness and consistency of both methods in performing the susceptibility mapping of the study area. When the OI values are considered, the results provided by the deterministic model are slightly better than the ones resulting from the statistical analysis. This does not come as a surprise for the case study at hand and it can be ascribed to the availability, within the study area, of: a reliable database of soil properties, and an in-depth knowledge of the behaviour of the considered landslides.