Seismic imaging with extended image volumes and source estimation

Seismic imaging is an important tool for the exploration and production of oil & gas, carbon sequestration, and the mitigation of geohazards. Through the process of seismic migration, images of subsurface geological structures are created from data collected at the surface. These images reflect changes in the physical rock properties such as wave speed and density. While significant progress has been made in the development of 3D imaging technology for complex geological areas, several challenges remain, some of which are addressed in this thesis. The first main contribution of this thesis is in the area of creating so-called subsurface-offset gathers, which play an increasingly important role in seismic imaging because they provide a multitude of information ranging from the reflection mechanism itself to information of the dips of specific reflectors and the accuracy of the background velocity model. Unfortunately, the formation and manipulation of these gathers come with exceedingly high computational and storage costs because extended image volumes are quadratic in the image size. These high costs are avoided by using techniques from modern randomized linear algebra that allow for compression of extended image volumes into low-rank factorized form—i.e., the image volume is approximately written as an outer product of a tall and a wide matrix. It is demonstrated that this factorization provides access to different types of sub-surface offset gathers, including common-image (point) gathers, without the need to explicitly form this outer product. As a result, challenging steep dip imaging situations, where conventional horizontal offset gathers no longer focus, can be handled. Moreover, extended image volumes for one background velocity model can directly be mapped to those of another background velocity model. As a result, factorization costs are incurred only once when examining imaging scenarios for different background velocity models. The second main contribution of this thesis is on the development of computationally efficient sparsity-promoting imaging techniques and on-the-fly source estimation. In this work, an adaptive technique is proposed where the unknown time signature of the sources is estimated during imaging. Without accurate knowledge of these source signatures, seismic images can be wrongly positioned and can have the wrong amplitudes hampering subsequent geophysical and geological interpretations. With the presented technique, this problem is mitigated. Finally, a contribution is made to address the detrimental effects of surface-related multiples. If not handled correctly, these multiples give rise to unwanted artifacts in the image. A new technique is introduced to address this issue in realistic settings where there is a strong density contrast at the ocean bottom. As a result, the surface-related multiples are mapped to the reflectors. Because bounce points at the surface can be considered as sources, this mapping of the multiples rather than removal increases the subsurface illumination.