The influence of the inventory on the determination of the rainfall-induced shallow landslides susceptibility using generalized additive models

• Published in: Catena (Giessen) 2020-10, Vol.193, p.104630, Article 104630
• Main Authors: Bordoni, Massimiliano, Galanti, Yuri, Bartelletti, Carlotta, Persichillo, Maria Giuseppina, Barsanti, Michele, Giannecchini, Roberto, Avanzi, Giacomo D'Amato, Cevasco, Andrea, Brandolini, Pierluigi, Galve, Jorge Pedro, Meisina, Claudia
• Format: Article
• Language: English
• Subjects: Cohen’s Kappa coefficient; Data-driven methods; Landslide inventories; Landslide susceptibility; Receiving operator curve
• ISSN: 0341-8162; 1872-6887
• DOI: 10.1016/j.catena.2020.104630

[Display omitted] •Role of inventories on shallow landslides susceptibility through data-driven model.•Agreement of models of an area built with different inventories is generally fair.•Multi-temporal inventory is more effective to obtain a complete susceptibility.•Phenomena detected by each tool influence directly the susceptibility assessment.•Criteria adopted in a shallow landslide inventory should be provided. Data-driven methodologies are widely used for the reconstruction of shallow landslide susceptibility maps. The outputs of these models strictly depend on the type of landslide inventories used for the modelling. Thus, this work performed a deep analysis of the role of different inventories on the definition of shallow landslide susceptibility of a particular area, modeled through a data-driven technique (Generalized Additive Model). Distinct shallow landslide inventories referred to the same rainfall event or to different rainfall events can produce different maps of susceptibility distribution for the same area. Despite a similar predictive capability, the models built for the same area produced different susceptibility maps, as estimated by low values of the Cohen's Kappa coefficient (0.13–0.46). If inventories obtained using different landslide detection methods are available, the models of susceptibility will be affected by the position and shape of the phenomena that could be detected by each tool. An inventory created integrating the visual interpretation of high-resolution images with field surveys allows to obtain more accurate inventories, to derive a better representation of the proneness of the territory towards shallow landsliding. In the case of inventories related to the same rainfall event carried out with different criteria, the susceptibility distribution is influenced by the choices adopted as regards, in particular, the type and the mapping criteria of phenomena included in the inventory. For shallow landslide inventories triggered by different rainfall events in the same area, the classification of the territory in terms of susceptibility is strictly connected to each specific rainfall event and reflects the relative shallow slope stability of a region only under similar rainfall conditions. Shallow landslide susceptibility obtained with a multi-temporal inventory, grouping landslides occurred in few years during more events, provides a more accurate estimation of the predisposition to shallow landsliding in consequence of the