GNGTS 2019 - Atti del 38° Convegno Nazionale

320 GNGTS 2019 S essione 2.1 statistical method proposed by Baiesi and Paczuski (2004) and modified by Zaliapin et al. (2008), known as the Nearest-Neighbor (NN) Clustering Analysis. With this choice, we avoid the arbitrary distinction among foreshocks, mainshocks and aftershocks. The choice of the NN method is also motivated by the fact that it provides a straightforward solution for identifying sequences of dependent events in a catalog, as explained in Zaliapin and Ben-Zion (2013). By grouping all the triggered events that share the same parent, and by progressively doing it back in time (if the parents themselves have been triggered by a previous ancestor), we can identify all the clusters through an agglomerative clustering approach. For the sake of example, Fig. 2 shows the clusters related to the three main seismic sequences occurred in central Italy in the last two decades: the 1997 Umbria Marche, the 2009 L’Aquila earthquake and the most recent Norcia-Amatrice sequences. We then compare the characteristics of the sequences of dependent events across the different crustal tectonic regions. Specifically, we look at the following cluster properties: (1) number of dependent events; (2) duration of the sequence; (3) spatial dispersion. To compare statistically the cluster properties across the three areas, we implement the two- sample Kolmogorov-Smirnov test. The goal is to check whether the observations are compatible with the null hypothesis that, for a given cluster property, the distributions of its values referred to different areas are samples from the same population. As for the background seismicity, we follow two complementary approaches for testing whether it is compatible with the null hypothesis of a stationary Poisson process: A) We test the null hypothesis of random and exponentially distributed inter-event times by means of the Runs and Lilliefors tests; B) We check whether the number of earthquakes in time windows of equal length conforms with the expectation of the Poisson distribution of equal seismicity rate by means of the conditional chi-square test (also named Poisson dispersion test). Results. We never find strong evidence against the null hypothesis of no difference among the distributions referred to the three tectonic crustal areas. In Fig. 3, we show the cumulative distribution function plots of the cluster properties for each catalog. The P -values returned by the two-sample Kolmogorov-Smirnov test are often way higher than 0. This is also true for the several additional tests that we have implemented to check the stability and robustness of our findings. We reach the same conclusion when analyzing subcatalogs that are highly homogeneous from the tectonic point of view, i.e when considering the effect of the deformation style alone. As for the background seismicity, we do not find significant evidence against the null hypothesis of uncorrelated inter-event times (Runs test), and the Lilliefors test does not show any significant departure from the Poisson hypothesis. However, the conditional chi-square test shows significant departures in Southern California and Italy. This result indicates that the background seismicity exhibits a higher time variability than what expected by a time homogeneous Poisson process. Fig. 3 - Cumulative distribution function plots of the cluster properties distinguished by catalog. From left to right: number of dependent events per sequence, duration, areal extent.