QUICK DETERMINATION OF THE FAULT MECHANISM FROM INITIAL P-WAVE AMPLITUDE DISTRIBUTION

When an earthquake occurs, there are only few seconds between the rupture beginning and its devastating effect on population and buildings. Having a reliable image of the seismic source in real-time is crucial for producing realistic strong motion shake maps, which are the most valuable piece of information to provide to the end-users, for an efficient planning of emergency operations. The focal mechanism (together with the rupture extent) represents a key parameter for a correct prediction of the ground shaking at target sites, since the source radiation pattern and directivity modulate the amplitude and frequency content of the radiated seismic wave field as a function of the azimuth with respect to the propagating rupture on the fault plane. While different methodologies have been proposed for the real-time magnitude estimation in EEWS, the problem of the rapid determination of the focal mechanism still lacks of a unique and validated solution. Currently, the real-time automatic determination of the focal mechanism takes advantage of a grid of precomputed solutions and is generally available within minutes after the earthquake detection and location. Here we develop a straightforward and robust methodology for the real-time determination of the focal mechanism using the azimuthal distribution of initial P-wave amplitude and a-priori constraints based on the local tectonic information. In our methodology, as soon as a few seconds of P-wave signals are available at a set of recording stations, we measure the initial P-wave peak amplitudes as the maximum absolute amplitude values of displacement, velocity and acceleration (Pd, Pv and Pa, respectively). The three parameters are estimated on the vertical component of the ground motion in progressively expanding P-wave time windows, starting from the arrival of the P-wave and stopping at the expected arrival of the S-waves. To account for the path attenuation effect, both Pd, Pv and Pa are corrected using precomputed, empirical scaling relationships, and then normalized to their maximum value. An empirical combination of the three parameters is then used to compare the observed amplitude distribution (as a function of azimuth and take-off angle) to the theoretical amplitude variation (i.e., theoretical P-wave amplitude radiation pattern), for a set of potential fault geometries. The comparison, through a dedicated algorithm, provides a first-order identification of the best solution for the fault mechanism (in terms of strike, dip and rake angles). We account for the available tectonic information through a specific prior distribution for strike, dip and rake angles and used a probabilistic, Bayesian, evolutionary approach, where the solution at each time step is used as a prior information for later times. At each iteration, the convergence of the solution is evaluated by comparing the current solution with the most likely triplet of the previous step and the convergence of the solution is declared if the solution does not show significant variations for a given time window. We apply the methodology to a large dataset of earthquakes, with magnitude ranging between 4 and about 7, to understand the potential limitations of the proposed approach. We explore the optimal combination of the Pwave peak amplitudes parameters (Pd, Pv, Pa) and investigate the use of different prior distribution to better constrain the fault geometry and rapidly determine the source focal mechanism.