GNGTS 2019 - Atti del 38° Convegno Nazionale

GNGTS 2019 S essione 1.3 189 Aquestion may arise about the validity of the velocity field shown in Fig. 2. To try answering that question, we have various means to assess the reliability of the estimated model. One is via the checking of the match between the synthetic seismograms computed on this model and the observed seismograms. Other ways are to verify the convergence rate in the GA + GB FWI, and to compare with the P-wave velocity profiles recorded in deep wells, if available. Figure 3 shows the observed seismograms (in black) and the predicted seismograms (in red) of one common shot gather of the CROP/18A profile. Overall, the matching is quite satisfactory: as also evidenced in the close-ups at different source-receiver distances, no cycle skip is evident. This fair level of correspondence is common to many seismograms along the profile, with the exception of those located at the ends of the line where the decreased illumination of the subsurface causes a loss of sensitivity in the inversion. Conclusions. The FWI approach we have discussed seems to be appropriate for applications to difficult seismic data, like those of the CROP/18A profile recorded in the geothermal area of southern Tuscany, which show a scarcity of reflections. The use of transmitted waves, and the sequence of GA FWI, driven by a non-linear stochastic optimization, and of GB FWI, with a linear iterative optimization, allow for the estimation of a velocity model whose general morphology is not biased by an a-priori model and, at the same time, shows enough details to be used as an interpretation asset. If the objective of the inversion is estimation of the geometry of the subsurface velocity structure when no a-priori model exists, or when different interpretations on its morphology occur, a neutral starting model for the inversion is an opportune option since any a-priori information could bias the result toward one model or the other. Therefore, genetic algorithms, which do not even start from a model but from a random population of models, seem quite suitable. We may then leave to linear optimization tools the task of improving the resolution. We are not expert in volcanic or geothermal exploration and thus we leave to the specialists the evaluation of whether this type of results may be useful for their endeavors in this or other cases. Acknowledgements. The authors wish to thank Enel Green Power for making the data available, for the permission to publish the results and for the many recent and past discussions on the geology of the area. Part of this work is developed in the framework of the Research Projects of the University of Pisa (PRA 2018-2019). References Aleardi M., Tognarelli A. and Mazzotti, A.; 2016: Characterisation of shallow marine sediments using high-resolution velocity analysis and genetic-algorithm-driven 1D elastic full-waveform inversion . Near Surface Geophysics, 14 (5), 449 – 460, DOI: 10.3997/1873-0604.2016030. Fig. 3 - Comparison between observed seismograms (black) and the predicted seismograms (blue) computed on the velocity field illustrated in Fig. 2.