2D multiscale non-linear velocity inversion

An efficient and robust non-linear inversion method for velocity optimizati on combining a global random search followed by a simplex technique is pres ented. The background velocity field is estimated at different spatial scal es by analysing image gathers after iterative prestack depth migrations. Fi rst, the global random search is used to determine the main features/trends of the velocity model (large-scale component). Then, the simplex technique improves the resolution of the velocity field by estimating smaller-scale features. A measure of the quality of the velocity model (objective functio n) is based on flattening offset events in depth-migrated image gathers. To help constrain the solution, the algorithm can incorporate a priori inform ation about the model and a smoothness condition. This 2D velocity estimati on offers the benefit of being semi-automatic (requiring minimal human inte rvention) as well as providing a global and objective solution (which is a useful approach to an interpretation-derived velocity-estimation technique) . The method is applied to a real data set where AVO analysis is carried ou t after prestack depth migration, as structural effects are non-negligible. It is demonstrated that the method can successfully estimate a laterally i nhomogeneous velocity model at a computational cost modest compared with an interpretation-based iterative prestack depth velocity-analysis technique.