GNGTS 2018 - 37° Convegno Nazionale

770 GNGTS 2018 S essione 3.3 MULTI-PARAMETER FULL-WAVEFORM INVERSION FOR COMPLEX SHAPED SHALLOW TARGETS: PRELIMINARY RESULTS AND CRITICAL ASPECTS D. Teodor 1 , C. Comina 1 , L.V. Socco 2 , R. Brossier 3 , F. Khosro Anjom 2 , P.T. Trinh 3 , J. Virieux 3 1 Dipartimento di Scienze della Terra, Università degli Studi di Torino, Italy 2 Dipartimento di Ingegneria dell’Ambiente, del Territorio e delle Infrastrutture, Politecnico di Torino, Italy 3 Institut des Sciences de la Terre (IsTerre) Univ. Grenoble Alpesly Introduction. Full-Waveform Inversion (FWI) is a non-linear data-fitting technique based on the full wavefield comparison between observed data and synthetic solution obtained by solving the wave equation (Tarantola, 2005). It provides quantitative seismic images of the subsurface physical parameters with a theoretical local half-wavelength resolution (Born and Wolf, 1970; Virieux and Operto, 2009). While the fitting is mainly driven by body waves for upper crust imaging (Sirgue et al. , 2010; Warner et al. , 2013), data for shallow targets are dominated by surface waves (SWs) which require a specific investigation for quantitative geotechnical imaging, especially in noisy environments. For shallow environments reconstruction, elastic propagation should be considered because shallow heterogeneities and topography impact strongly the wavefield. The elastic multi- parameter FWI is facing two main challenges: the first is related to the differential sensitivity of the algorithm with respect to each parameter class; the second is the increasing computational cost when considering needed visco-elastic propagation embedded into the optimization workflow: only linearized formulation based on Newton methods are currently considered. In this local optimization, an initial model is needed and it should be accurate enough to avoid local minima issues. Although techniques are developed to overcome cycle-skipping effects induced by a crude initial model for upper crust imaging (van Leeuwen and Herrmann, 2013; Métivier et al. , 2016,), near-surface applications are still very sensitive to the initial model design. The complex structure of the wavefield with many phases combined altogether makes the building of the initial model as an important crucial step for FWI. For shallow structures, the analysis of the dispersion curves (DCs) of SWs fundamental mode provides shear velocity structures. Recently, Socco et al. (2017) and Socco and Comina (2017) have proposed a workflow to extract as well compressive velocity structures using a wavelength-depth relationship sensitive to Poisson’s ratio. The mitigation of the strong lateral variations is performed by a clustering analysis (Khosro Anjom et al. , 2017). Are these S-wave and P-wave velocity models kinematically compatible for preventing cycle-skipping issues (Virieux and Operto, 2009)? This is the purpose of the present work while the influence of frequency window and offset range should play an important role when progressively building the image. Dataset, Methodology and Tools. Synthetic data are built as observed data on a conceptual model (Fig.1a) inspired by the CNR test site nearby Turin (Italy) and has already been presented in previous studies (Teodor et al. , 2017; Khosro Anjom et al. , 2017). For such propagation simulation, a spectral-element code (SEM46) is used where attenuation could be described by combining standard-linear-solid mechanisms and where anisotropy could be considered if needed (Trinh et al. , 2018). In this initial feasible synthetic investigation, only isotropic elastic propagation is considered, but more complex propagation would be expected when facing real data analysis. Avertical point source characterized by a Ricker wavelet with a central frequency of 16 Hz is used for 11 shots recorded by 72 stations. Following the previously mentioned procedure based on dispersion curve analysis, Teodor et al. (2018) built an initial P-wave velocity model (Vp) and S-wave velocity model (Vs) which are shown in figure 1c. The density is taken as constant with a value of 1800 kg/m 3 . Figures 1b and 1d show vertical traces for the first shot in the target model and in the initial model, respectively. One can appreciate some similar structures of waveforms, thanks to the DC analysis. Full-Waveform Inversion tests and results. We started the FWI of the synthetic dataset from the 1 – 40 Hz frequency range in the frame of a multi-scale approach. The density is