Full-waveform inversion (FWI) based on an elastic approximation of the Earth's physical behavior is a powerful tool for high-resolution model building. The elastic wave equation based wavefield extrapolation relaxes some of the limitations we face using ray-based methods. Still, it also introduces more challenges to the inversion. The high computation cost is becoming more tolerable with the development of modern computers, while the cycle-skipping and crosstalk between different parameters are still hampering our efforts to apply waveform inversion in practice. To addresses these issues, I first improve the nonzero-lag crosscorrelation objective function by adding a normalization term and a smooth weighting function. Then, I utilize a reflection-friendly local-similarity measurement. The algorithm strives to maximize the local similarities of the predicted and observed data by stretching/squeezing the observed data. It compares two data sets locally, and thus, performs better than the global correlation in matching multiple arrivals. The near-surface of the Earth plays a vital role in supporting the modern infrastructure and in imaging the deep Earth. I propose a wave-equation based inversion algorithm that inverts for S-wave velocities using fundamental- and higher-modes Rayleigh waves without picking an explicit dispersion curve. The proposed method aims to maximize the similarity of the phase velocity spectrum (f-v) of the observed and predicted surface waves with all-Rayleigh wave modes (if they exist) included in the inversion. A particular application to ambient noise data recorded by a fiber-optic cable is also shown. Reservoir characterization is an essential component of oil and gas production, as well as prediction. However, full-waveform inversion can easily fail to characterize deep-buried reservoirs due to illumination limitations. I introduce the facies-constrained waveform inversion by utilizing other geophysical data such as well logs to effectively reduce the crosstalk in the multiparameter estimation, where a Bayesian inversion and a more advanced deep learning network are utilized to build the connection between different geophysical data. Synthetic and field data examples demonstrate the effectiveness of the algorithms and also reveal some of their limitations.
|Date made available
|KAUST Research Repository