Lithological and seasonal control on rainfall thresholds for the possible initiation of landslides in central Italy

We compiled a catalogue of 442 rainfall events with landslides in the Abruzzo, Marche, and Umbria regions, central Italy, between February 2002 and August 2010. For each event, we determined the duration D (in hours) and the cumulated (total) event rainfall E (in mm), using rainfall measurements obtained from a dense network of 150 rain gauges. Landslides were mapped as single points using Google Earth®, and were attributed a level of mapping accuracy P, in four classes (P1 < 1 km2, P10 < 10 km2, P100 < 100 km2, P300 < 300 km2). Using this information, we determined cumulated event rainfall–rainfall duration (ED) thresholds, which are entirely equivalent to rainfall intensity–duration (ID) thresholds, and avoid unnecessary conversions. To determine the ED thresholds, we modified a statistical method first proposed to determine ID thresholds. Adopting a bootstrapping statistical technique, we expanded the method to determine the uncertainty associated with the parameters that define the power law threshold model, and the sensitivity of the thresholds to the number of (D,E) data points in the empirical data set. We determined ED thresholds for 1% and 5% exceedance probability levels: (i) for the entire study area, and for the three individual regions in the study area (Abruzzo, Marche, and Umbria); (ii) for three main lithological domains in the study area (post-orogenic sediments, flysch deposits, carbonate rocks); and (iii) for two seasonal periods (May–September, October–April). Comparison of the ED thresholds indicates that the cumulated amount of rainfall necessary to trigger landslides in flysch deposits is larger than in post-orogenic sediments and carbonate rocks. However, the ED thresholds for post-orogenic sediments and carbonate rocks were statistically indistinguishable in the study area. Also, the seasonal thresholds were statistically different for rainfall durations shorter than 12 h and longer than 100 h, but were indistinguishable in the range 12 < D < 100 h. The results obtained are directly applicable to landslide forecasting based on empirical rainfall thresholds, and have implications for landslide hazard and risk assessment, as well as for landslide erosion and landscape evolution studies, both in the study area and in the neighboring regions.