Toward an automatic full-wave inversion: Synthetic study cases

Abstract

Geophysics 2 ABSTRACTFull Waveform Inversion (FWI) in seismic scenarios continues to be a complex procedure for subsurface imaging that might require extensive human interaction, in terms of model setup, constraints and data preconditioning. The underlying reason is the strong non-linearity of the problem that forces the addition of a priori knowledge (or bias) in order to obtain geologically sound results. In particular, when the use of long offset receiver is not possible or may not favor the reconstruction of the fine structure of the model, one needs to rely on reflection data. As a consequence, the inversion process is more prone to get stuck into local minima. It is then possible to take advantage of the cross-correlation error functional, less subject to starting models error, in order to output a suitable background model for inversion of reflection data. By combining these functionals, high-frequency data content with poor initial models can be successfully inverted. If we can find simple parameterizations for such functionals we can can reduce the amount of uncertainty and manual work related to tuning FWI. Thus FWI might become a semi-automatized imaging tool. I NTRODUCTIONFull Waveform Inversion (FWI) represents a seismic imaging method able to improve Earth structural models up to spatial resolutions beyond the limits of standard Travel Time Tomography (TTT), and more adequate for seismic imaging. TTT only inverts the time residuals of (mostly) P-wave phases picked on the recorded field traces, requiring human interaction.On the other hand, FWI processes the whole waveforms achieving a finer resolution. Nevertheless, given our surface to surface acquisition limitations, noise effects and initial models with poor Geophysics 3 low frequency content, convergence to the true model cannot be guaranteed. Among the strongest concerns when using FWI is the matching of synthetic and data phases when they are apart more than half a cycle in time, an effect known as cycle skipping (Luo and Schuster, 1991;Warner and Guasch, 2014; Metivier et al., 2016). Some functional have been developed over the last decades to cope with this issue: e.g the cross-correlation (CC) traveltime functional (Luo and Schuster, 1991), the adaptive FWI from Warner and Guasch (2014), or the optimal transport distance (Métivier et al., 2016). Although less sophisticated, the CC is able to provide good background models as reported by, e.g., Jimenez-Tejero et al. (2015), but lack in resolution. On

The RECIPE approach to challenges in deeply heterogeneous high performance systems

“…In this paper, we report the characterization of a geophysical exploration application which builds upon Full Waveform Inversion (FWI), a cutting edge technique that aims to acquire the physical properties of the subsoil from a set of seismic measurements [48,49].…”

On the Use of Probabilistic Worst-Case Execution Time Estimation for Parallel Applications in High Performance Systems

“…The case study considered in this paper is the Full Waveform Inversion (FWI), a cutting edge technique that aims to acquire the physical properties of the subsoil from a set of seismic measurements [18,19]. Starting from a guess (initial model) of the variables being inverted (e.g., sound transmission velocity), the stimulus introduced and the recorded signals, FWI performs several phases of iterative computations to reach the real value of the set of variables being inverted with an acceptable error threshold.…”

“…For instance, the result of a weather prediction or large simulation behavior beyond those values observed in applications with a continuous flow of (big) data. In particular, we use an application for geophysical exploration as a representative, which builds upon Full Waveform Inversion (FWI), a cutting edge technique that aims to acquire the physical properties of the subsoil from a set of seismic measurements [18,19]. Having reliable timing bounds for this type of application allows setting the speed at which specific areas can be monitored as well as the resolution used for the collected data.…”