Time reversal and plane-wave decomposition in seismic interferometry, inversion and imaging
Tao, Yi, active 2012
MetadataShow full item record
This thesis concerns the study of time reversal and plane-wave decomposition in various geophysical applications. Time reversal is a key step in seismic interferometry, reverse time migration and full waveform inversion. The plane-wave transform, also known as the tau-p transform or slant-stack, can separate waves based on their ray parameters or their emergence angles at the surface. I propose a new approach to retrieve virtual full-wave seismic responses from crosscorrelating recorded seismic data in the plane-wave domain. Unlike a traditional approach where the correlogram is obtained from crosscorrelating recorded data, which contains the full range of ray parameters, this method directly chooses common ray parameters to cancel overlapping ray paths. Thus, it can sometime avoid spurious arrivals when the acquisition requirement of seismic interferometry is not strictly met. I demonstrate the method with synthetic examples and an ocean bottom seismometer data example. I show a multi-scale application of plane-wave based full waveform inversion (FWI) with the aid of frequency domain forward modeling. FWI uses the two-way wave-equation to produce high-resolution velocity models for seismic imaging. This technique is implemented by an adjoint-state approach, which viii involves a time-reversal propagation of the residual wavefield at receivers, similar to seismic interferometry. With a plane-wave transformed gather, we can decompose the data by ray parameters and iteratively update the velocity model with selected ray parameters. This encoding approach can significantly reduce the number of shots and receivers required in gradient and Hessian calculations. Borrowing the idea of minimizing different data residual norms in FWI, I study the effect of different scaling methods to the receiver wavefield in the reverse time migration. I show that this type of scaling is able to significantly suppress outliers compared to conventional algorithms. I also show that scaling by its absolute norm generally produces better results than other approaches. I propose a robust stochastic time-lapse seismic inversion strategy with an application of monitoring Cranfield CO2 injection site. This workflow involves two steps. The first step is the baseline inversion using a hybrid starting model that combines a fractal prior and the low-frequency prior from well log data. The second step is to use a double-difference inversion scheme to focus on the local areas where time-lapse changes have occurred. Synthetic data and field data show the effectiveness of this method.