Large scale wavefield reconstruction via weighted matrix factorization and seismic survey design

Seismic data acquisition plays a crucial role in identifying potential oil and gas reservoirs during the early phases of exploration. However, obtaining finely sampled seismic data can be costly and physically impossible. Recent developments in Compressive Sensing have resulted in seismic data being increasingly collected at random along spatial coordinates. Although random sampling improves acquisition efficiency, it shifts the burden from seismic acquisition to data processing. Wavefield recovery is one of the required processes for reconstructing fully seismic data from coarsely subsampled data. Among the various techniques proposed for wavefield reconstruction, matrix completion methods are computationally efficient and straightforward to implement. These methods exploit the low-rank structure of fully seismic data. However, matrix completion performs well at low-to-mid frequencies and degrades at higher frequencies due to the failure of low-rank structure to accurately approximate higher frequencies. To address this issue, this thesis proposed a recursively weighted matrix completion method. Although effective, this method is computationally expensive, and a more efficient method for handling 2D seismic data was also proposed. Compared to 2D seismic data, 3D seismic data can detect reflections outside of the 2D plane but poses a computational challenge due to its large scale. To overcome this challenge, this thesis proposed a parallel weighted reconstruction method to improve the reconstruction of 3D seismic data. Land seismic data presents a greater challenge due to contamination by ground roll, which consists of surface waves with a high spatial frequency content and large amplitude. To address this issue, a practical workflow was proposed in this thesis to improve the recovery of land seismic data. Although matrix completion is an efficient technique for reconstructing fully seismic data, the optimal acquisition design is still being investigated. Recent studies have shown that the spectral gap can be used to predict and characterize the quality of wavefield reconstruction via matrix completion for a given subsampling mask. Based on these findings, a simulation-free seismic survey design for both 2D and 3D seismic data was proposed in this thesis to obtain an improved subsampling survey by minimizing the spectral gap ratio. Furthermore, this concept was extended to the design of a time-lapse seismic survey, which is essential for reservoir management and monitoring geological carbon storage but is difficult and expensive to acquire. To improve the reconstruction of the time-lapse wavefield, a joint recovery model was proposed that leverages the benefits of the non-replicated baseline and monitor subsampled seismic data. A time-lapse seismic survey design that incorporates the joint recovery model with spectral gap was proposed to generate sparse, non-replicated time-lapse acquisition geometries that favor wavefield recovery.

https://slim.gatech.edu/Publications/Public/Thesis/2023/zhang2023THlsw