GNGTS 2017 - 36° Convegno Nazionale

596 GNGTS 2017 S essione 3.1 on the fault zone width (about 7-8 m), the occurrence of wide low-velocity zones interpreted as colluvial wedges. We may correctly estimate the Holocene displacement (about 15 m) and the minimum amount of cumulative offset accrued by the fault. Such study expands the knowledge of the fault behaviour, which had been previously constrained only for the most recent time interval by paleoseismic studies. References Improta, L., Zollo, A., Herrero, A., Frattini, M., Virieux, J., & Dell���������� ��� ����� ������� ������� �� ������� ’Aversana, P., 2002. Seismic imaging of complex structures by non-linear traveltime inversion of dense wide-angle data: Application to a thrust belt. Geophys. J. Int., 151, 264–278, doi: 10.1046/j.1365-246X.2002.01768.x. Improta, L., Villani, F., Bruno, P.P., Castiello, A., De Rosa, D., Varriale, F., Punzo, M., Brunori, C.A., Civico, R., Pierdominici, S., Berlusconi, A., & Giacomuzzi G., 2012. High-resolution controlled-source seismic tomography across the MiddleAterno basin in the epicentral area of the 2009, Mw 6.3, L’Aquila earthquake (centralApennines, Italy). ����� �� ������� ������ ���� ����� ����� ���� �� �������� ���� �������������������� Ital. J. Geosci. (Boll. Soc. Geol. It.), 131, 3, 373-388, doi: 10.3301/IJG.2011.35. Villani, F., Tulliani, V., Sapia, V., Fierro, E., Civico, R., and Pantosti, D., 2015b. Shallow subsurface imaging of the Piano di Pezza active normal fault (central Italy) by high-resolution refraction and electrical resistivity tomography coupled with time-domain electromagnetic data. Geophys. J. Int., 203 (3), 1482-1494, doi: 10.1093/gji/ggv399. Villani, F., Improta, L., Pucci, S., Civico, R., Bruno, P.P., and Pantosti, D., 2017. Investigating the architecture of the Paganica Fault (2009 Mw 6.1 earthquake, central Italy) by integrating high-resolution multiscale refraction tomography and detailed geological mapping, Geophys. J. Int., 208(1), 403-423, doi:10.1093/gji/ggw407. Near-surface model prediction and refinement by full waveform surface wave inversion Z. Xing, F. Rappisi, A. Mazzotti Department of Earth Sciences, Geophysics, University of Pisa, Italy Introduction. Genetic algorithm full waveform inversion (GA-FWI) is able to fairly predict complex shear-wave velocity (Vs) models from surface waves, even in the case when very limited or null a-priori information is available (Xing andMazzotti, 2017a). Out of consideration for computing time reduction, a two-grid approach (Sajeva et al. , 2016; Aleardi and Mazzotti, 2017; Mazzotti et al. , 2017), one coarse grid for the inversion and one fine grid for the modeling, is recommended for the method. Thus, we generally obtain smooth velocity models whose wavelengths are dependent on the coarse grid spacing. In this paper, we show that these models are suitable starting models for FWI approaches with local optimization methods and that, in general, significant details of the depth model can be retrieved. Instead, we do not discuss the influences caused by different surface wave modeling strategies (Thorbecke and Draganov, 2011; Groos et al. , 2014; Xing and Mazzotti, 2016) and by assumptions in wave modeling (Xing and Mazzotti, 2017b), thus we focus on model prediction and refinement. Surface wave modeling. The synthetic “observed” data modeling and the forward modeling in FWI for the tests shown in the paper are the same 2D time domain finite difference modeling. A 4 th order approximation is applied to derivatives along space while a 2 nd order along time. Convolutional perfectlymatched layer is used as the absorbing boundary condition. To guarantee the reliability of the modeled surface waves, 20 points per wavelength instead of 5, which is generally recommended for body wave modeling, is adopted in the modeling process. Two-grid genetic algorithm FWI. FWI is a data-driven process. In our GA-FWI code, based on an L1 norm, in each generation, operations of selection, recombination, mutation and reinsertion are executed to generate new models with ever-decreasing data misfits. Disregarding the possible existence of ambiguity, which could be alleviated by a-priori information, the model correspondent to the minimum data misfit in the last generation is selected as the inversion result. Alternatively, the mean model in the last generation could be chosen.