NESTORE v1.0: A MATLAB based code to forecast strong aftershocks applied to Italian seismicity

NESTOREv1.0 is a MATLAB-based algorithm that uses a machine learning approach to provide a probabilistic forecasting of clusters in which a strong mainshock is followed by at least one subsequent earthquake of comparable magnitude. This objective is critical to mitigate seismic risk during strong seismic sequences because building already damaged by the mainshock occurrence can collapse, increasing damages and death toll. In particular, the algorithm distinguishes clusters in which the magnitude difference between the mainshock and the strongest aftershock is less than or equal to 1 (type A) and the other cases (type B). NESTOREv1.0 identifies clusters and is trained to distinguish the two typologies by using cluster seismicity parameters (features) on a training dataset. It is then able to produce a forecasting of A-type cluster for retrospective analysis on a test database or for ongoing clusters. For the application of NESTOREv1.0 to the Italian data, we focused our analysis on two areas covering the northeast of Italy and most of the complementary part of Italy. For these two areas, we used the seismic data recorded by the INGV and OGS seismic networks over the last 40 years. In particular, we trained NESTOREv1.0 for the clusters that occurred in the first 30 years of the catalogues and evaluated its performance in the period 2010-2021. We found for both areas that the percentage of correct forecasting of the cluster type a few hours after the occurrence of the mainshock is over 85%.