Accelerating seismic imaging and velocity model building with approximate extended Born inversion
Symes, William W.
Doctor of Philosophy
Solving the inverse problem in exploration seismology usually consists of two main components: seismic imaging and velocity model building. The main goal of this thesis is to improve the efficiency, as well as the accuracy, of current cutting-edge methods for seismic imaging and velocity model building. This thesis relies upon two basic ideas: scale separation of the earth model and the model extension concept. The former separates the earth model into long scale background model and short scale reflectivity model while the latter extends the reflectivity model by adding extra degrees of freedom. Seismic imaging aims to recover the reflectivity model by solving a linearized inverse problem. Various migration algorithms approximate the solution by computing the adjoint of the (extended) Born modeling operator. I show that an inexpensive modification of the adjoint operator can lead to an approximate inverse to the extended Born modeling operator. This operator is further used to accelerate iterative Least Squares Migration, both extended and nonextended variants. The velocity model building part starts from velocity analysis, which updates the long wavelength information by minimizing an objective function that measures the violation of a semblance condition. I show that the replacement of the migration operator with the approximate inverse operator can lead to a better velocity update. Subsequent Full Waveform Inversion process perfects the velocity model by providing fine details with local optimization algorithm. The approximate inverse operator preconditions the steepest descent method to approximate the convergence rate of the Gauss-Newton method while preserving essentially the same cost.
Seismic Imaging; Velocity Model Building; Seismic Inversion; Optimization