Global 3D acoustic FWI using sparse model parameterization

Estimating a starting velocity model for Full Waveform Inversion can be challenging. It requires several passes of migration velocity analysis to obtain a model accurate enough to prevent cycle skipping. We present an alternative approach to estimate a 3D starting model using a global optimization method called Very Fast Simulated Annealing. To constraint he optimization problem with a large number of unknowns, we parameterize the 3D model with surfaces and velocities surrounding them and solve for the optimal parameters. The final estimated model from VFSA serves as a starting model for FWI. We demonstrate the effectiveness by comparing FWI results along a few 2D lines from the 3D model. The proposed method is largely automated and reduces numerous man hours required to build a starting model. We apply our proposed method to one toy model and one complex synthetic model. In both cases, we were able to obtain acceptable results. Use of VFSA with a sparse parameterization method makes our 3D global inversion a practical tool.