GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.3 GNGTS 2024 Fig. 3 – a) Observed seismogram; b) seismogram computed from the mean model of the prior distributon; c) predicted seismogram; d) diference between observed and predicted seismograms. Conclusions We presented an ensemble-based approach to FWI using the ES-MDA algorithm. To reduce the computatonal efort required by such an approach, we compress both data and model space through a 2D DCT. We applied the algorithm to a porton of the synthetc Marmousi model in the acoustc approximaton. The results are satsfactory: the mean of the fnal ensemble contains all the main features of the original, DCT compressed, model showing the main diferences on the deepest porton and on the edges. Even in data space we observe a good ft between observed and predicted data. The algorithm appears able to deal with the cycle skipping issue mitgatng it: to this end, some tests, not shown here for the lack of space, have been performed. Future steps of this research are further tests on the algorithm, with the aim of approaching the applicaton to feld data. The code will also be improved to run in parallel, to considerably reduce the overall computatonal tme. Further investgatons will be carried out to obtain a more reliable estmaton of the uncertaintes.