Probabilistic approaches for assessing rainfall thresholds triggering shallow landslides. The study case of peri-Vesuvian area (Southern Italy)

Abstract

Ash-fall pyroclastic soil deposits covering steep carbonate slopes in the peri-Vesuvian area (southern Italy) are periodically involved in shallow landslides (about 700 events were recorded during the last three centuries, as reported by CASCINI et alii, 2008), triggered by intense and/or prolonged rainfall events, which evolve as catastrophic debris flows. In the last decades, many studies have been focused on estimating reliable relationships among the triggering of shallow landslides and the amount and duration of rainfall events, as well as the role played by antecedent soil hydrological conditions. Results of these studies have been expected to give information on temporal hazard to landslide onset to be used for setting a reliable early warning system. In this paper we present probabilistic approaches to assess rainfall thresholds triggering shallow landslides by classical empirical methods and to manage the uncertainties related to biases of data. At this scope, rainfall events related to the occurrence of debris flows along slopes of the Sarno and Lattari Mountains, known from chronicles of the last century, were analyzed by means of the empirical models of Intensity-Duration (I-D) (CAINE, 1980) and rainfall recorded in the day of the landslide occurrence (P) vs the antecedent cumulated rainfall (Pa) (CROZIER & EYLES, 1980). In order to limit and to assess uncertainties related to biases of rainfall data, a comparison with the regional probability model of high intensity rainfall, carried out in the framework of the VAPI Project (ROSSI & VILLANI, 1994) has been carried out. Moreover, rainfall data were processed by a bivariate logistic regression model resulting in the assessment of probability to landslide triggering, given an assumed rainfall event. The I-D empirical rainfall thresholds obtained by Caine model (1980) were compared to rainfall thresholds estimated by deterministic approaches (DE VITA et alii, 2013; NAPOLITANO et alii, 2016) showing a good match.