TIME: 3:30 PM

LOCATION: GMCS 314

SPEAKER: Miguel Dumett, San Diego State University, Computational Science Research Center

ABSTRACT: Given an Earth model and a macro velocity model, a subsurface reflectivity image is acquired through a migration process. Typical migration algorithms exhibit structures that are not well defined and/or with some artifacts due to a series of data sampling characteristics. To remediate this problem, a functional that boosts sparsity is used and a convenient equivalent formulation is solved. The approach is tested with synthetic data for a collection of signal sources and receivers. Then the inverse problem of obtaining a subsurface reflector description is solved using data migrated and migrated back by a discrete version of the Kirchhoff migration. The method is robust to noise and to relatively imprecise macro velocities. Moreover, it requires a small fraction of the data required by traditional migration algorithms to produce the same level of accuracy although for an equivalent amount of data it is computationally more intensive. As an alternative it is shown that a high-order modified mimetic difference scheme can be used to discretize the Eikonal equation that is needed to solve for computing the travel times between sources and receivers. A quasi-Newton approach suggests that the solution of the system of nonlinear equations can be reduced to solving numerically a few linear systems. The extension of this mimetic approach to more general geometries is under development.

HOST: Jose Castillo