• FWI
• Surface Waves
• Two-Grid FWI
• MASW
• preconditioned conjugate gradient

This thesis presents a global-local surface wave full-waveform inversion (FWI) to estimate a high-resolution elastic shear wave velocity model of the near-surface at the site of Grenoble (France) using a seismic data acquired in the framework of the interPACIFIC project. A preliminary analysis is carried out by means of the MASW approach and dispersion curve assessment, obtaining 1-D shear wave velocity profiles comparable with nearby boreholes measurements. However, the main assumption of a 1-D velocity sub-surface distribution is not consistent with the experimental data because different Vs profiles occur for the same dataset up to 30 m. Therefore, a method able to overcome the 1-D limitation is required to tackle the lateral variability of the subsurface. A 2-D Vs model is obtained by inverting the surface waves in a two-step full-waveform inversion (FWI) approach. In the first, we perform a global FWI which employs a genetic algorithm (GA) as the optimization tool and a finite difference code as the forward modeling engine. Making use of a large search range centered on a constant velocity field, GA are used to estimate a long-wavelength velocity model minimizing the L2 norm of the difference between the observed and simulated seismic waveforms at the receiver locations. The objective is to find a sub-surface model that reproduces the full waveform, including the travel-times and amplitudes of the observed seismic data. In the second step, the obtained best model is used as the starting point for a gradient-based FWI. The local optimization method of preconditioned conjugate gradient is applied. The global full‐waveform inversion makes use of a two‐grid scheme in which the sub-surface is described by means of a fine grid for the finite-difference forward modelling and of a coarse grid for the stochastic inversion. A bilinear interpolation brings the coarse-grid models into the fine-grid models with the aim of having the resulting synthetic seismograms matching the events of the observed seismograms (predominantly the surface waves events). A good match between observed and predicted data allows to use the estimated velocity field as the starting point for a local, gradient‐based full‐waveform inversion supposed to intercept the global minima of the misfit function, with errors smaller than half of the wavelet period to avoid cycle-skipping artifacts. No a-priori information are used to guide the inversion. The models of the starting population for the global inversion are randomly distributed within constant search ranges and to accelerate the convergence, two strategy are considered: offset marching and frequency marching. Gradual increase of the maximum offset is implemented at low frequency (up to 15 Hz) reproducing, step-by-step, deeper portions of the sub-soil. In the GA inversion we consider frequency lower than 30 Hz, while we extend the frequency up to 60 Hz in the local inversion, adopting a frequency marching scheme. The final velocity model shows an improved resolution in the shallow part of the sub-surface. The reliability of the final velocity model is assessed by the data misfit achieved and by the comparison with the available nearby boreholes. The main structural features of the sub-surface model are fairly predicted, even in the case of null a-priori information.