A comparison between empirical landslide predictive models applied to the Marche region (central Italy)

This research contributes to the landslide forecast debate, analyzing the relationship between rainfall and slope failures in the eastward section of the Esino river basin, located in the Marche region (central Italy). Post-orogenic quaternary sediments, prone to rainfall-induced shallow landslides, characterize this 550 km2 wide area. Based on the review of historical landslides, the study area was affected by 234 landslides over the period 1953 to 2012. In order to determine rainfall thresholds for the possible occurrence of landslides, an approach that consider 3 empirical (statistical) methods have been applied: (i) the cumulative event rainfall– duration (ED) method, (ii) the intensity – duration (ID) method, and (iii) the Bayesian probabilistic method (one-dimensional and two-dimensional). These methods were applied for the same landslides and rainfall databases and for the same zone, which is characterized by comparable hydrogeological properties. The ED and ID method allowed the definition of local rainfall thresholds which, even if showing general consistency with the trends of the literature thresholds, highlight the need of considering limited areas with similar lithological settings. Moreover, it was possible to associate different ranges of ED or ID with the corresponding number of landslides expected, from rainfall events that could trigger 1 landslide to rainfall events that could trigger more than 10 landslides. On the other hand, the Bayesian method allowed to consider in a probabilistic framework all the past rainfall events, whether they resulted in landslides or not, and thus to select the critical rainfall parameters for the area of investigation. Results of this analysis show that the most significant rainfall variables in explaining the initiation of landslides are: cumulative event rainfall (E), daily rainfall (R), and five days antecedent rainfall (A5) only if coupled with daily rainfall. Finally, all the considered models were tested during a recent rainfall event that affected the study area on 2-4 May 2014 and triggered several landslides. The encouraging results obtained in this research have shown the benefits of applying a diversified methodological approach to study a complex problem as the landslide hazard in the Marche region