| The feasibility of automatic background velocity estimation in 2D laterally varying media is evaluated in this research. To automate the process, background velocity estimation is cast as a non linear inverse problem. Cubic splines are used in the velocity model parameterization. Some selected shot gathers or some constant offset sections are used as input data for the inversion. A global optimization technique, called very fast simulated annealing (VFSA), is used in the search for the model with minimum error. A Kirchhoff summation scheme based on first arrival travel times is used to migrate/model the input data involved in the estimation process. Two cost functions are evaluated and compared. The first one is defined in the recorded data or (x,t) domain and is based on a reflection tomography criterion (RTC). The other cost function is defined in the migrated data or (x,z) domain and is based on a migration misfit criterion (MMC). Depth relaxation is used to improve the convergence and quality of the estimates while simultaneously reducing the computational cost.;The proposed schemes resulted in good velocity estimates when applied to synthetic and real data, showing the feasibility of their application for the current generation of computers. For the velocity estimation procedure in the recorded data domain travel times do not have to be picked, and in the migrated data domain the depth images do not have to be manually compared. Interpreter intervention is required only to reduce the ambiguity inherent in the velocity estimation problem. The application of the proposed velocity estimation schemes to the complex Marmousi velocity model showed some of the limitations of the proposed schemes for areas with complex structural features. Some suggestions are provided to overcome these limitations. Extension of the proposed schemes to 3D models is straightforward but probably feasible only for the fastest available parallel computers. |
