Detection and characterization of seismic sequences in the normal fault system of the Irpinia region, Southern Italy

Understanding mechanical processes occurring on faults and catching the preparation phase of large magnitude events require a detailed characterization of the microseismicity, which can be enhanced today using advanced techniques for earthquake detection. These techniques decrease the detection threshold of seismic networks and provide augmented catalogs, which enable improved statistical analysis associated with event occurrence and size. However, seismic events recorded at the level of the noise typically emerge only at a few stations, making earthquake characterization challenging. This issue is further complicated in areas where seismicity occurs deep in the crust, as happens in the normal fault system of the Irpinia region, Southern Italy, where earthquakes occur at depths between 8 and 15 km.

In this work we focus on the detection and characterization of seismic sequences occurring in the Irpinia region featuring low magnitude mainshocks (Ml∼3), using data from the Irpinia Near Fault Observatory.

Event detection for the sequences is performed through the integration of a machine learning based detector (EQTransformer, Mousavi et al., 2020) and a template matching technique (Chamberlain et al., 2018), with the former providing a wider set of templates for the similarity search. This strategy outperforms auto-similarity techniques based on fingerprints (FAST, Yoon et al., 2015) and template matching grounded in manual catalogs. On average, the final catalog of the analyzed sequences increases the manually revised network bulletin by a factor 7. We compared P- and S- arrival time estimates, grounded in the machine learning phase picking and cross-correlation for template matching, using manual identifications to assess the reliability of automatic picks; the mean residual between manual and automatic values is ~0 for both P- and S-waves, with a larger residuals standard deviation for the latter.
We apply a double-difference location technique using both catalog and cross-correlation differential travel times for locating the events, with the goal of resolving and highlighting fault structures where seismicity takes place. We finally track the spatio-temporal evolution of the seismicity, and apply a mechanical model based on static stress, to discriminate whether sequences in the area are mainly triggered by static stress change, dynamic stressing, or aseismic mechanisms such as fluid diffusion.