A global-local experience of 2D acoustic FWI on a real data set

In this work, we illustrate an example of acoustic 2D FWI on a real data set extracted from a 3D volume. The FWI is carried out in two successive steps. The first step is based on a global optimization procedure that makes use of genetic algorithms. It is aimed at the estimation of a velocity model close enough to the global minimum such that the second step, based on a local optimization, quickly converges to the optimal solution. The local step also increases the resolution of the estimated velocity field. An L1-norm misfit function, computed on the envelope of the observed and synthetic data to avoid cycle-skipping effects, is used to increase the robustness of the solution. A low-velocity region at approximately 1.2 km depth is observed in the global inversion that is better detailed by the local inversion. The alignment of the CIGs at shallow depth after Kirchhoff pre-stack depth migration supports these results that are obtained with a very limited manpower effort.