Full Waveform Inversion for Kinematic Equivalence

We present an approach to 3D full waveform inversion that can handle noisy data deficient in refracted energy. We use time domain modeling and monochromatic inversion, and obtain excellent kinematic agreement of modeled data with observed data for a narrow azimuth towed streamer dataset from the northwest shelf of Australia. We describe how we apply regularization and preconditioners to obtain a robust solution with respect to an imperfect starting model and various sources of noise in the field data. Even though we employ very simple isotropic constant density visco-acoustic wave propagation, we obtain near kinematic equivalence of the data modeled using the inverted velocities with the field data. We discuss how this kinematic equivalence in the framework of robust norms and regularized nonlinear optimization may imply that non-uniqueness comes also from the choice of physics used for wave propagation.