Estimating Starting Models for Full Waveform Inversion Using a Global Optimization Method

Full waveform inversion (FWI) is formulated as a data fitting least squares minimization problem, which iteratively updates an initial velocity model with the scaled gradient. However the local optimization approaches can converge to a local minimum if the starting model used is not close enough to the optimal model. Here we propose a two-step technique in which we a use a global optimization method like very fast simulated annealing (VFSA) to generate a starting model containing low frequency components. The starting model is sparsely parameterized with a set of interfaces and velocities. The sparse velocity model thus obtained from VFSA is used as a starting model in a conventional gradient based FWI to obtain our final model containing short wavelength features required for detailed data fitting. We apply the proposed method to two field datasets from offshore India where we are successfully able to obtain velocity models that match the well log data and are helpful in geological interpretation of the study areas. Thus we demonstrate that the use of VFSA is practical for field data analysis.