GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 1.3 GNGTS 2024 MCMTpy waveform inversion package: testng a new method for moment tensor estmaton T. Mancuso 1 , C. Totaro 1 , B. Orecchio 1 1 Department of Mathematcal and Computer Sciences, Physical and Earth Sciences (University of Messina, Italy) Earthquake focal mechanism inversion is a seriously non-linear problem. It is well known that an accurate estmaton of focal mechanisms is fundamental to obtain good constraints on regional stress feld, to assess seismic hazard, and to beter understand tectonic processes. The procedures commonly used to compute focal mechanism solutons are based on the polarity of P-wave frst moton which may be biased by several factors (e.g., an inadequate coverage of seismic statons). Waveform inversion methods have so far demonstrated to be much more capable to furnish stable and reliable solutons (Prest et al., 2013), however needing a more accurate estmate of the associated errors which tend to be underestmated using linearized techniques (Scolaro et al., 2018). Bayesian inference is increasingly being applied to solve these non-linear problems because it has the advantage of quantfying uncertaintes of parameters (Vasyura-Bathke et al., 2020). In this work we present the applicaton of a new Python package MCMTpy (Yin & Wang, 2022), which exploits the ‘Cut-And-Paste’ waveform inversion algorithm (CAP, Zhao & Helmberger, 1994; Zhu & Helmberger, 1996) and Bayesian inference, using Markov Chain to implement the source locaton and focal mechanism inversion in a unique workfow. The new approach can simultaneously invert for magnitude, focal mechanism, source locaton, source depth and origin tme also providing a way to quantfy uncertaintes by statstcal inference. The main functons included in MCMTpy are source parameters inversion (i) under double-couple assumpton with Markov-Chain Monte Carlo (MCMC) method, (ii) under double-couple assumpton with a grid-search method and (iii) for the full moment tensor soluton with MCMC method. To test the robustness and limitatons of the new package, we applied the MCMTpy to the 2016 M w 6.0 Amatrice earthquake and to a smaller event (i.e., M w 3.2) of the same sequence. We performed several tests by varying the startng soluton, number of iteratons, network geometry, and the type of computaton (e.g., MCMC, grid-search method) and we compared our results with moment tensor solutons from other catalogues (e.g., Time Domain Moment Tensor). Test results were analysed in this study in order to evaluate reliability of moment tensor solutons and estmated uncertaintes in diferent inversion setngs. References