GNGTS 2017 - 36° Convegno Nazionale

552 GNGTS 2017 S essione 3.1 Elastic full-waveform inversion of single-component marine seismic data: preliminary results on the Marmousi-2 model R. Lo Bue, A. Tognarelli, M. Aleardi, A. Mazzotti Earth Sciences Department, University of Pisa, Italy Introduction. The theoretical concepts of full-waveform inversion (FWI) date back to the early 1980s (Tarantola, 1984), but due to lack of sufficient computer power, the application of FWI to seismic data did not take off until a few years ago. In particular, over the past several years, the industry has been making large strides toward using gradient-based full-waveform inversion (FWI) to build velocity models for use with pre-stack depth migration (Sajeva et al., 2016; Aleardi et al., 2016) after many synthetic model exercises, the attention has turned to the use of field data with the acoustic approximation of the two-way wave propagation (Morgan et al., 2013). These studies have shown that if the acquired data provide long offsets and low frequencies in the range of 2 to 3 Hz, gradient-based FWI can iteratively build a high-fidelity velocity model by means of consecutive use of data with increasing frequency bandwidth (the so-called multiscale approach; Bunks et al., 1995). In addition, although the acquired data are not acoustic (but more realistically viscoelastic and anisotropic) it has been demonstrated that the 3D FWI can bring significant uplift to the details in the acoustic velocity field and thus create superior migrated images. Recent computational improvements allowed for the simulation of 3D elastic wavefields and thus undertake the challenge of elastic full-waveform inversion (EFWI). Differently fromacoustic FWI that is primarily focused on inverting diving waves, EFWI has the ability to simultaneously invert reflected and transmitted energy using traveltime, amplitude, and phase information. In this context, EFWI can theoretically be an optimal tool to derive high-resolution and reliable elastic characterizations of the subsurface that are crucial in many geophysical applications, but particularly in reservoir characterization studies in which only primary P-P reflections and 1D convolutional forward models are routinely used (e.g. Aleardi and Ciabarri, 2017). Obviously, the non-linearity and the ill-conditioning of FWI increase as many wave phenomena (multiples or converted waves) or different model parameters ( Vp, Vs , density, viscoelastic and anisotropic parameters) are simultaneously inverted (Operto et al., 2013). For this reason, applications of EFWI are primarily focused on inverting multicomponent seismic data (Sears et al., 2010; Prieux et al., 2013; Vigh et al., 2014) that compared to conventional single-component data, bring in additional information about shear wave velocity. However, acquiring multicomponent seismic data is expensive especially in deep-water areas, where hardware limitations prevent multicomponent technology from being extended to water depths in excess of 1500 m. For this reason, in this work we assess the ability of EFWI of single- component data to provide accurate elastic subsurface models that could be used as input for reservoir characterization studies. In the following we show some preliminary results obtained on the Marmousi-2 model, which is an elastic model that reproduces a geological profile of north Quenguela in the Quanza Basin in Angola (Martin et al. , 2006). We primarily focus our attention on the evaluation of the accuracy and quality of the Vp , Vs and corresponding Vp / Vs models. The inversion strategy, uses starting models that nicely approximate the true elastic model and moves from low frequencies to high frequencies, using both early arrivals, diving waves and reflected events. The density was kept constant to maintain the inversion at a simple level, which allowed us to draw essential conclusions. The Marmousi-2 model and the inversion approach. The Marmousi-2 elastic model (Figs. 1a and 1b) has been developed from the Marmousi-1 acoustic model. Both models aim to reproduce the geology of the north Quenguela in the Quanza Basin in Angola. Marmousi-2 preserves the Marmousi-1 lithologies and geological structures but is deeper and laterally more extended. The sedimentary sequence is quite simple in the left and right edges of the model, but it is very complex in the centre where thrust structures, salt bodies and high-angle normal