A deep catalogue of 56k focal mechanisms for the 2016 Amatrice, Italy earthquake sequence

The 2016 Amatrice, Italy earthquake sequence occurred on a normal fault system in the Central Apennines, and contained over 1,300 M>=3 earthquakes. With ~140 permanent or temporary seismic stations directly above the seismic activity, the sequence has been exceptionally well recorded. Starting from a deep learning-based catalogue of earthquake hypocentres (~900,000 re-located events from ~15 million seismic phases; Tan et al., 2021), we use a convolutional neural network classifier to predict P-wave first motion polarities, from which we compile a deep catalogue of earthquake focal mechanisms. The catalogue consists of >56'000 focal mechanisms, about 8'000 of which have nodal plane uncertainties below 25 degrees.

In contrast to existing, conventional focal mechanism catalogues, the deep catalogue samples almost the entire study region, and almost the entire magnitude range (~M0-M5), although nodal plane uncertainties generally tend to increase with decreasing magnitude. We use the focal mechanism catalogue to study the kinematics of the Amatrice earthquake sequence, to test the hypothesis of a coseismic rotation of the static stress field by large and small events, and to analyse the complexity of the stress field before, during and after the earthquake sequence.