Probabilistic clustering of rainfall condition for landslide triggering.

Here we present a new fully automatic method to identify the probability of landslide occurrence associated to rainfall conditions characterized by measures of intensity or cumulative rainfall and rainfall duration. The method splits the rainfall events of the past in two groups: a group of events causing landslides and its complementary, then estimate their probabilistic distributions. The proposed approach can be conveniently integrated in existing early-warning system to improve the accuracy of the estimation of the real landslide occurrence probability associated to rainfall events and its uncertainty.