Semi-automated template matching and machine-learning--based analysis of the Castelsaraceno microearthquake sequence (High Agri Valley, Southern Apennines, Italy).

The accurate characterization of microearthquake sequences allows to better understand the physical processes involved in earthquake nucleation and to gain insights on the geometry of fault structures at depth. Standard workflows for the study of earthquake sequences involve manual detection and phase-picking steps, requiring a huge amount of work from expert seismologists, particularly for microseismic events. We show how the investigation of a low-magnitude sequence, occurred in August 2020 close to Castelsaraceno village (southern Apennines) benefited from the application of 4-step semi-automated template matching and machine-learning--based workflow. First, the phase-picking was automatically performed through a deep-learning algorithm on 202 microearthquakes detected between July and October 2020, then an automatic multi-step absolute and relative earthquake location procedure was applied. A total of 72 high-accuracy relocated events clustered in time (7--12 August) and in a narrow range of depths (10--12 km) were recognized as belonging to the sequence. The Ml 2.1 foreshock and the Ml 2.9 mainshock also identified a persistent asperity.