Full-waveform Inversion in the Data and Image Spaces

Full-waveform inversion of seismic data requires the knowledge of an initial model sufficiently close to the true model to prevent cycle-skipping. A common approach to solve this problem is to rely on velocity models estimated using ray tomography in combination with depth migration. This usually yields a smooth velocity model with kinematic properties similar enough to the true model to prevent the cycle-skipping problem. We propose an alternative approach based on wave theory only. By performing model fitting in the image space based on differential semblance a low resolution velocity model with good kinematic properties can be obtained. Cycle-skipping can be avoided by using this model as a starting model for full-waveform inversion in the data space. We show synthetic examples where an object function based on differential semblance is used to estimate an initial model and a least-squares object function is then minimized to refine the initial model.