A Pattern Recognition Approach to the Subsequent Event of Damaging Earthquakes in Italy

Abstract:

In this study, we investigated the occurrence of large aftershocks following the most significant earthquakes that occurred in Italy after 1980. We applied a pattern recognition approach using statistical features to forecast the occurrence of strong earthquakes after target events. A window based method detects clusters of seismicity in the analyzed catalogue and classifies them as “type A” if, given a main shock of magnitude M, the subsequent strongest earthquake in the cluster has magnitude ≥M-1, of type B otherwise.

We tested a set of A cluster precursors available in literature (Vorobieva et al., 1993, Vorobieva,1999) and, in addition, we proposed a set of original features to characterize clusters. In particular, we investigated the temporal evolution of the radiated energy, the spatio-temporal distribution of the immediate foreshocks, including a detailed analysis of the dimension and shape of the region of preparation of the strong earthquake, the spatio-temporal evolution of the aftershocks occurring within a few days and the probability to have a strong earthquake depending on the time elapsed after the mainshock. Furthermore, we examined the spatial distribution of the two types of clusters inside the Italian territory. In order to characterize the feature depending on the cluster type, we used decision trees as classifiers on single feature separately. The generalization capability of resulting trees was improved by a pruning based on leave-one-out method; the performances of the classification are tested by leave-one-out method too. The analysis was performed on different time-spans after the mainshock to simulate the increase of information available as time passes during the seismic clusters.