Predicting storm triggered debris flow events: application to the 2009 Ionian-Peloritan disaster(Sicily, Italy)

Debris flows are shallow landslides triggered by extreme meteorological events and propagating into thedrainage lines of a slope as a fluid. A debris flow susceptibility map depicts the spatial probability for futurephenomena to be triggered in a given area. Stochastic approaches are widely used in landslide studiesfor the assessment of the susceptibility. In fact, they allow to obtain a predictive function which relates aresponse variable (presence/absence of landslides) and a set of physical-environmental variables which areexpected to control the slope failure phenomena. Future landslide occurrences are typically predicted bystudying a past landslide inventory, under the basic assumption that “new landslides will occur under thesame conditions which drove the past ones”. The present research is aimed at testing the basic assumption,in case of extreme event triggered landslide scenarios. The study case is the debris flow event occurred inthe Messina area in 2009. In particular, by applying logistic regression, a model was calibrated by exploitingan inventory dated at 2007 and validated with respect the 2009 inventory (forward chrono-validation).Moreover, a model was calibrated with the 2009 and validated in predicting the 2007 landslides (backwardchrono-validation). Under the basic assumption, the two modelling procedures should achieve the sameresults. Cross-validation procedures have been applied to investigate precision, reliability and robustnessof the models, both in terms of predictive performance and inner structure of the model. The results ofthe research attest for high performance and good agreement between the two chrono-validated models.However, some differences arose, indicating possible limits in the basic assumption.