| Subsurface rock properties are manifested in seismic records as variations in travel times, amplitudes and waveforms. It is commonly acknowledged that travel times are sensitive to the long wavelength part of the velocity, whereas the amplitudes are sensitive to the short wavelength part of the velocity. The inherent sensitivity of seismic velocity at different wavelengths suggests an approach that decomposes the waveform data into travel time and amplitude components. Therefore I propose a divide-and-conquer approach to the waveform inversion problem, where I first estimate the smoothly varying background velocity from the travel time and the rapidly changing perturbations from the amplitude, and then combine the perturbation with the background to obtain a starting model to be used for the final full waveform inversion. For estimating the background velocity, I use the flatness of events as the objective criterion, and simulated annealing as a search tool. Three different model parameterization schemes, namely, constant velocity blocks, splines and arctangent models are compared, with the arctangent having the most flexibility and least artifacts. Having obtained the background velocities, I then analyze the amplitude-versus-offset effects (AVO) to estimate the perturbations to the background, for which I use a linearized inversion method. The combination of the perturbation and background should be sufficiently close to the true model so that the inverse problem becomes quasi-linear. A full elastic waveform inversion is used to fine-tune the combined model to obtain P-wave, S-wave velocity and density, again using either a non-linear optimization method or an iterative linearized solution. Application of the inversion algorithm to synthetic data from an 84-layer model was able to predict the full reflectivity data and recover the true model parameters. Application to one seismic line in the Carolina Trough area found a continuous thin gas zone which produces strong Bottom Simulating Reflectors (BSRs). As the first step in extending the algorithm to weak transversely isotropic media, amplitude and travel times are examined. It appears that the anisotropy parameters are not constrained by the amplitudes within small incidence angles. A full waveform inversion is needed. |
