The analysis of the Castelsaraceno microearthquake sequence (southern Italy) through a semi-automated template matching and machine-learning approach reveals an anti-apenninic fault

The accurate characterization of microearthquake sequences allows seismologists to shed light on the physical processes involved in earthquake nucleation, the deformation processes underlying rupture activation and propagation, and to image faults geometry at depth. The current methodologies used for this purpose first need the event detection and the phase-picking - usually manual-based - and earthquake locations, which require plenty of work even by expert analysts particularly in the case of microearthquake signals, commonly noise contaminated. Thus, improving standard procedures through semi-automatic or fully-automatic workflows would be an essential step forward towards the more efficient analysis of seismic sequences.

Here we show the results of a semi-automated template matching and machine-learning based workflow applied for the characterization of the foreshock-mainshock-aftershock microearthquake sequence occurred close to Castelsaraceno village (High Agri Valley, Southern Apennines, Italy) in August 2020. The analyses were performed on seismic data mainly recorded by a local seismic network belonging to the High Agri Valley geophysical Observatory (HAVO) deployed in the study area and located at a maximum epicentral distance of ~20 km from the seismicity cluster.

The application of the semi-automated single-station template matching technique to the continuous data-streams of the two nearest stations of the HAVO network (from 28th July to 12th October 2020) allowed us to detect more than twice the number of microearthquakes previously identified by standard manual detections. The phase-picking was automatically performed through a deep-learning algorithm (Phasenet) on the 202 ultimate detected microearthquakes. Finally, an automatic multi-step absolute and relative earthquake location procedure was carried out.

A total of 76 events were identified as belonging to the Castelsaraceno sequence, which occurred in a short time span (7-12 August) and in a limited range of depths (10 -12 km). Both the Ml 2.1 foreshock doublet and the Ml 2.9 mainshock occurred on 7 August ruptured the same seismogenic patch, thus suggesting the presence of a persistent asperity. The integrated analysis of the aftershocks distribution, the focal mechanism of the mainshock, and the geological framework of the study area, allowed revealing the seismogenic fault, not currently mapped in literature: a NE-SW striking (225°), high-angle (55°) fault with a left-lateral transtensional (rake -30°) kinematic. We also hypothesize that the seismic sequence occurred at depth in a brittle layer of the crystalline basement confined between two regions with more ductile rheology; futhermore, the estimated b-value (0.73±0.04) indicates the occurrence of the sequence in a relatively low-heterogeneity material and suggests the unimportant effect of pore-fluid pressure in driving its evolution.