The recent geophysical literature has shown the building of an accurate initial model is the more appropriate way to reduce the ill-posedness of the Full Waveform Inversion, providing the necessary convergence of the misfit function toward the global minimum. Optimized models are useful as initial guess for more sophisticated velocity inversion and migration methods. I developed an automatic P-wave velocity inversion methodology using pre-stack two-dimensional seismic data. The proposed inversion strategy is fully automatic, based on the semblance measurements and guided by the paraxial traveltime approximation, so-called Finite-Offset Common-Reflection-Surface. It is performed in two steps, at first using image rays and an a priori known initial velocity model we determine the reflector interfaces in depth from time migrated section. The generated depth macro-model is used as input at the second step, where the parametrization of the velocity model is made layer by layer. Each layer is separated from each other by smoothed interfaces. The inversion strategy is based on the scan of semblance measurements in each common-midpoint gather guided by the Finite-Offset Common-Reflection-Surface traveltime paraxial approximations. For beginning the inversion in the second step, the finite-offset common-midpoint central rays is built by ray tracing from the velocity macro-model obtained in the first step. By using the arithmetic mean of total semblance calculated from the whole common-midpoint gathers as objective function, layer after layer, a global optimization method called Very Fast Simulated Annealing algorithm is applied in order to obtain the convergence of the objective function toward the global maximum. By applying to synthetic and real data, I showed the robustness of the inversion algorithm for yielding an optimized P-wave velocity macro-model from pre-stack seismic data.
|
