Evolutionary polynomial regression to alert rainfall-triggered landslide reactivation

The derivation of an alert model for landslide risk management is a paramount problem for those sites which are affected by complex landslides involving strategic infrastructures as well as towns. This is a quite common scenario all over the world and then it is a primary problem for the management of geomorphological risk. Along the Adriatic Coast of south Italy, Petacciato landslide is peculiar, since it showed 11 reactivations between 1924 and 2009. It is a deep-seated landslide, and the history of its reactivations shows that even if generally related to quite abundant rainfall periods, there is no clear correlation between rainfall events and reactivations. For this reason, here, an analysis based on a data-driven evolutionary modeling technique is attempted, in order to identify an alert model based on cumulative rainfall heights. Modeling results are quite interesting and encouraging, since they are able to provide landslide forecasting whereas no false positive are ever returned. This work shows the results of this attempt as well as an analysis of the input to the modeling approach, in order to identify which are those cumulative rainfall heights which are physically sound with respect to the particular landslide.