GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.3 GNGTS 2024 fnal ensemble. As a prior mean model we used a gradient model, with velocity values increasing from top to botom and ranging from the minimum to the maximum value of the original model (Fig.2-c). We observe that the model obtained with the EB-FWI (Fig.2-d) contains all the main features visible in the DCT-compressed version of the original porton of the Marmousi model. The main diferences are placed on the botom and in lateral portons of the model, where the algorithm is sometmes not able to correct high or low wrong velocity values. Anyway, this happens in the less illuminated parts of the model, characterized by higher values of the standard deviaton (Fig.2-e). A comparison of observed and predicted data is shown in Fig.3, along with their diference. The represented shot is the third of the fve simulated, and its positon corresponds to the center of the horizontal extension of the model. The represented seismograms show a good ft between observed and predicted data. Fig.3-b shows the shot gather computed on the gradient model, used as the mean of the prior distributon. Considering that this is the seismogram associated with the mean of the startng ensemble and comparing the observed data (Fig.3-a) with the seismogram corresponding to the mean of the last ensemble (Fig.3-c), we clearly see that the algorithm appears able to properly reproduce the main events in the data. Fig. 2 – a) True model, porton of the synthetc Marmousi model; b) true model afer the DCT compression; c) mean model of the prior distributon; d) fnal result of the EB-FWI; e) standard deviaton associated to the inversion result.