GNGTS 2018 - 37° Convegno Nazionale

GNGTS 2018 S essione 3.1 575 2017), we make use of a processing sequence that includes a band-pass filtering between 5 Hz and 15 Hz, trace-envelope computation, and trace-by-trace normalization. Finally, we choose the L 2 norm difference between the predicted and the observed data, as the misfit function. Initial model. Due to the non-linearity of the misfit function, the starting model in the FWI procedure plays an important role. To assure the convergence of a local optimization method, the initial model must be accurate enough to give a good match between the observed and predicted data. To attain this, we make use of the velocity model obtained from MVA. More details on the estimation of the model can be found in (Tognarelli et al. , 2010). Figure 2 shows the MVAvelocity model used to pre-stack depth migrate (PSDM) the data. The migration results are illustrated in Fig. 3a where 6 GIGs evenly spaced along the profile and displayed up to 2 km of depth can be observed. The flattening of the events on the CIGs, obtained after the PSDM, constitutes a good validation test for the reliability of the MVA velocity model. However, many gathers still present a residual move-out that can be reduced using the FWI based on a gradient line-search method. Inversion procedure and results. As the local optimization method, we use the steepest descent algorithm (Nocedal et al., 2006), where the descend direction corresponds to the negative direction of the gradient of the misfit function, and the step length is obtained by a line search that satisfies the Wolfe conditions (Wolfe, 1969). The gradient is computed using the adjoint method (Plessix, 2006). Special attention is given to include the processing sequence in the computation of both the misfit function and its gradient. The unknowns are the velocity values on the grid nodes situated below the sea floor, for a total of 981 x 77 unknowns. The velocities range of the unknowns is between 1400m/s and 4500m/s. We perform 200 iterations for the minimization procedure. Fig. 2a and 2b show the velocity model obtained by MVA and at the end of the FWI optimization procedure, respectively. The starting model is refined mainly in the upper part just below the sea-bed. Figure 3a and 3b show a comparison of 6 CIGs positioned along the seismic profile before and after the FWI inversion. A significant improvement of the horizontal alignment of the events at about 1 km of depth can be noted passing from Fig. 3a to Fig. 3b. Fig. 3 - CIGs obtained by pre-stack depth migrating the data using (a) the MVA velocity model and (b) the final ve- locity model obtained at the end of the local optimization procedure. Conclusion. In this work, we made an acoustic FWI experience on a portion of the CROP M12A marine seismic profile acquired in the framework of the Italian Deep Crust Project. The processing sequence applied to the data reduces the non-linearity of the misfit function and strengthens the reliability of the whole procedure against the cycle-skipping phenomenon. Starting from a velocity model obtained by the MVA, we estimated, by means of a gradient-