Insights into Extensional Tectonics from a Large Deep Learning Focal Mechanism Catalog

ABSTRACT Earthquake sequences play out on geologic fault and fracture systems, which are usually underconstrained by data. Modern deep learning earthquake detection and characterization methods now allow us to compute high-sensitivity and high-resolution seismicity catalogs, with which we can image at least the seismogenic parts of fault and fracture systems with much more detail than had been possible previously. Here, we use a convolutional neural network classifier and the SKHASH algorithm to compute a catalog of 16,600 well-constrained focal mechanisms (FMs) for the exceptionally well-monitored 2016 Mw 6.0 Amatrice–Mw 5.9 Visso–Mw 6.5 Norcia earthquake sequence in Italy. The resulting catalog paints a detailed picture of earthquake faulting kinematics in a fragmented extensional tectonic system. We observe that normal-faulting mechanisms dominate the seismic activity only over the depth range of 2–9 km. At shallower depths, for which the overburden may be too low for normal faults to be elastically loaded, strike-slip faulting is more common. The much-debated basal shear zone—an extensive about 2 km wide near-horizontal layer with distributed seismicity at 8–10 km depth—is characterized by much higher FM variability than the shallower parts of the crust, where the main normal faults host the largest earthquakes. In the north and south of the study region, the basal shear zone seismicity is sharply divided by the main normal faults, with predominantly normal faulting in the hanging wall, and predominantly strike-slip faulting in the footwall. The FMs from this study provide insight into deformation processes at the intersection of this basal shear zone and the major normal faults, which is where both the Amatrice and Norcia events nucleated.