Global Optimization Procedure to Estimate a Starting Velocity Model for Local Full Waveform Inversion

Finding an efficient procedure to solve a seismic inversion problem, such as Full Waveform Inversion, still remains an open question. This is mainly due to the non linearity of the misfit function, characterized by the presence of multiple local minima, that cause unsuccessful results of local optimization strategies when the starting model is not in the basin of attraction of the global minimum. Therefore, the use of a global optimization strategy in order to estimate a suitable starting velocity model for Full Waveform Inversion is a crucial point. In this work we propose a new method which is based on the application of the Adaptive Simulated Annealing algorithm using a coarse inversion grid, different from the domain modelling one, that allows the reduction of the number of unknowns. Numerical results show that the application of our strategy to an acoustic Full Waveform Inversion provides a good starting model for a local optimization procedure, lying in the basin of attraction of the global minimum.