An Automatic Workflow to Infer Focal Mechanisms of Microearthquakes: Application to Southern Italy

The focal mechanism of an earthquake is a key element to constrain the slipping fault and the stress field. Despite that, its calculation may be very challenging in the case of small-magnitude earthquakes because of the difficulty in determining the first motion polarities of signals hidden in the noise. In this study, we tested the convolutional first motion (CFM)-a convolutional neural network-to detect P-wave polarities on about 16 years of seismicity that occurred in Irpinia (Southern Italy). CFM found 175 earthquakes with 8 P polarities or more and a small-error location. By inverting for the focal mechanisms with the retrieved polarities, we determined a dominance of normal faulting mechanisms produced by an extensional NE-SW oriented stress field. Moreover, we found that the spatial heterogeneity of the mechanisms (measured by the Kagan angle) decreases for interevent distances lower than about 3 km. Furthermore, we demonstrated that this heterogeneity can be produced by a fault distribution whose orientations follow a Cauchy distribution with parameter $k$ = 0.3-0.4, which is an indication of the degree of fault misalignment. We also found that such a misalignment in the southern volume of the fault system is about 1.7 times higher than in the northern volume. Our automatic inference of focal mechanisms highlights structural complexity as a key factor controlling seismicity, demonstrating the potential of automated approaches to characterize fault systems capable of generating M7 earthquakes.