Velocity model building by geometrical spreading focusing

A reliable and fast workflow for smooth velocity model building is of great interest for subsequent depth migration. Also full-waveform inversion benefits from geologically reasonable initial velocity models. At the same time, seismic data enhancement by wavefront attributes is accepted. These attributes are extracted from seimic data by common-reflection-surface (CRS) stack and locally characterize the wavefronts. We propose a novel tomography based on them. It is valid for reflection, diffraction and passive seismic data inversion. The idea of the new approach is minimizing geometrical spreading of diffracted and/or fictitious NIP waves. During the inversion the wavefront attributes remain fixed at common midpoints and serve as initial conditions for kinematic and dynamic ray tracing. The inverse problem turns out to be naturally parametrized by the velocity model as the only unknown. It significantly decreases the tomographic matrix dimension and improves the data/unknowns ratio if compared to conventional wavefront attributes tomography. The reduction of Fréchet derivatives is also an attractive feature if a generalization to 3D anisotropic media is considered. We provide both Fréchet derivatives and adjoint-state method formulation for the gradient of the new objective function. The algorithm combined with the L-BFGS-B quasi-Newton solver was tested on a salt body synthetic dataset. It converges to a consistent smooth velocity image.