Seismic imaging and velocity model building with the linearized eikonal equation and upwind finite-differences

Ray theory plays an important role in seismic imaging and velocity model building. Although rays are the high-frequency asymptotic solutions of the wave equation and therefore do not usually capture all details of the wave physics, they provide a convenient and effective tool for a wide range of geophysical applications. Especially, ray theory gives rise to traveltimes. Even though wave-based methods for imaging and model building had attracted significant attentions in recent years, traveltime-based methods are still indispensable and should be further developed for improved accuracy and efficiency. Moreover, there are possibilities for new ray theoretical methods that might address the difficulties faced by conventional traveltime-based approaches. My thesis consists of mainly four parts. In the first part, starting from the linearized eikonal equation, I derive and implement a set of linear operators by upwind finite differences. These operators are not only consistent with fast-marching eikonal solver that I use for traveltime computation but also computationally efficient. They are fundamental elements in the numerical implementations of my other works. Next, I investigate feasibility of using the double-square-root eikonal equation for near surface first-break traveltime tomography. Compared with traditional eikonal-based approach, where the gradient in its adjoint-state tomography neglects information along the shot dimension, my method handles all shots together. I show that the double-square-root eikonal equation can be solved efficiently by a causal discretization scheme. The associated adjoint-state tomography is then realized by linearization and upwind finite-differences. My implementation does not need adjoint state as an intermediate parameter for the gradient and therefore the overall cost for one linearization update is relatively inexpensive. Numerical examples demonstrate stable and fast convergence of the proposed method. Then, I develop a strategy for compressing traveltime tables in Kirchhoff depth migration. The method is based on differentiating the eikonal equation in the source position, which can be easily implemented along with the fast-marching method. The resulting eikonal-based traveltime source-derivative relies on solving a version of the linearized eikonal equation, which is carried out by the upwind finite-differences operator. The source-derivative enables an accurate Hermite interpolation. I also show how the method can be straightforwardly integrated in anti-aliasing and Kirchhoff redatuming. Finally, I revisit the classical problem of time-to-depth conversion. In the presence of lateral velocity variations, the conversion requires recovering geometrical spreading of the image rays. I recast the governing ill-posed problem in an optimization framework and solve it iteratively. Several upwind finite-differences linear operators are combined to implement the algorithm. The major advantage of my optimization-based time-to-depth conversion is its numerical stability. Synthetic and field data examples demonstrate practical applicability of the new approach.