Investigating features of earthquake clustering: a comparative analysis for selected sequences In Italy

A number of studies aimed at seismic hazard assessment, as well as at the space-time analysis of earthquakes occurrence (e.g. Gentili et al., 2017), require preliminary declustering of the earthquake catalog. Moreover, the identification and statistical characterization of seismic clusters provide useful insights about the features of seismic energy release and their relation to physical properties of the crust within a given region. Earthquake clustering, in fact, is an fundamental aspect of seismicity, with typical features in space, time, and energy domains that provide key information about earthquake dynamics. In spite of the overall agreement about the existence of different types of clusters (sequences, swarms, bursts, etc.), there is no agreed formal definition of seismic clusters, nor a unique method to identify them. Most of the declustering algorithms available in literature are based on a deterministic space-time-window scheme or on a stochastic branching model (e.g. ETAS model by Ogata, 1998), which are generally suitable for large earthquakes, characterized by evident aftershock series clearly emerging from the background seismicity. Since various methods, relying on different physical/statistical assumptions, may lead to diverse classifications of earthquakes into main events and related events, we investigate the classification differences among different declustering techniques. Various techniques, including classical space-time windows methods (Gardner and Knopoff, 1974; Uhrhammer, 1986), are considered for this purpose. In particular, a statistical method for detection of earthquake clusters, based on "nearest-neighbor distances" of events in space-time-energy domain, is applied (Baiesi and Paczuski 2004; Zaliapin et al., 2008). The method allows for a robust data-driven identification of seismic clusters, and permits to disclose possible complex features in the internal structure of the identified clusters (Zaliapin and Ben-Zion, 2013). The application of the nearest-neighbor technique requires preliminary computation of the scaling parameters that characterize seismicity, specifically the b-value of the Gutenberg-Richter law (Gutenberg and Richter, 1954) and the fractal dimension of epicenters distribution (e.g. Grassberger, 1983). For this purpose we consider average robust estimates of the parameters the Unified Scaling Law for Earthquakes (USLE) in the study regions (Nekrasova et al., 2011, 2016). Accordingly, a formal selection and comparative analysis of earthquake clusters is carried out for the most relevant earthquakes in Northeastern Italy, as reported in the bulletins compiled at the National Institute of Oceanography and Experimental Geophysics since 1977 (Peresan and Gentili, 2016; Gentili et al., 2011). The analysis is then extended to consider earthquake sequences occurred in areas characterized by a different seismotectonic setting, such as the area struck by the recent Central Italy earthquakes. For this purpose we consider two databases of Italian seismicity: the historical catalog CPTI15 (Rovida et al., 2016) and the instrumental catalog, composed by the catalog of Lolli and Gasperini (2006) and updated since 2005 using the data from the Italian Seismological Instrumental and parametric Data-basE (http://iside. rm.ingv.it/iside/). The similarities and basic differences, between the clusters identified by the nearest-neighbor method and using other approaches, are investigated for the selected sequences, with special emphasis on seismicity of north-eastern Italy. Results from clusters identification turn out quite robust with respect to the time span and completeness level of the input catalog. Moreover, the study shows that the data-driven approach, based on the nearest-neighbor distances, c