GNGTS 2022 - Atti del 40° Convegno Nazionale

482 GNGTS 2022 Sessione 3.3 closely reproduces the velocity variations of the true model, even though there is a strong low velocity anomaly in the bottom right part. Analyzing the evolution of the negative log-likelihood ( Fig . 3 ) we can observe that after a small number of iterations we have reached the stationary regime, where the likelihood still oscillates but around the same values. We have decided to consider the first 500 iterations as the burn-in period and we have discarded the corresponding samples, so only the following 9500 samples for each chain were used to generate the mean and standard deviation models. Fig. 2 - Results of the inversion procedure, on the left part the initial model from which the inversion procedure started and the original model; on the right, the estimated mean model and the posterior standard deviation estimated by the GB-MCMC algorithm. Fig. 3 - On the left part, comparison between the observed data and the predicted data from the estimated mean model; on the right, the predicted data from the initial model from which the inversion procedure started and the evolution of the negative log-likelihood value during the GB-MCMC sampling. The vertical red line indicates the end of the burn-in.