Full Waveform Inversion (FWI) is used to estimate subsurface parameters, such as the P-wave velocity, in an iterative process. Although having been studied for several years, practical application of the technique on real data has proven difficult. From the theory, the full update is the reverse-time migrated residuals multiplied by the full Hessian. We demonstrate for a 2D acoustic marine synthetic case that the methodology can be applied with reasonable results, using the PSPI migration with a deconvolution imaging condition and a line search to calculate the step length to invert P-wave velocity. We tested different mutes applied in the migrated residuals, resulting in higher resolution when the mute is not too narrow or wide. We present a study of the importance of the initial velocity model for the convergence we find that the best initial model gives the best inversion result. However, we obtained good responses for initial velocity models not close to the global minimum such as velocity models with smoothed flat layers based on simulated wells in the real velocity model. We showed that when using only one well to generate the initial model, our best result is when we do not sample the high velocity body. Our results suggest that we can have a higher resolution inversion if an initial model is generated by an interpolation of two or more wells.
View full article as PDF (4.68 Mb)