next up previous [pdf]

Next: Linear least-squares inversion Up: Reconciling processing and inversion: Previous: Reconciling processing and inversion:

Introduction

Migration with the full wave equation is, in principle, capable of creating an image of the subsurface by reversing the propagation of all wave modes (Wapenaar et al., 1987). It is a computationally demanding process, which is why the wave equation is approximated by the one-way wave equation. Migration with the one-way approximation is formulated such that, ideally, the only seismic events present in the input data are primary reflections. When this is not the case, because the pre-processing was unable to completely eliminate noise (multiple reflections, for instance), the final image presents migrated events that do not correspond to the geology. In other words, these events are not modeled by the physical theory on which the migration method is based. They are frequently called migration noise or migration artifacts.

As is widely known, inversion problems are very sensitive to noise. In seismic, several authors have shown that coherent noise degrades the performance of the velocity inversion and yields erroneous results (Li and Symes, 2007; Chauris and Noble, 2001; Shin and Min, 2006). Therefore it is usually necessary to include some kind of a priori information about the model as an additional regularization equation in the inversion or to perform noise suppression prior to inversion.

Noise also affects least-squares inversion of migrated images into reflectivity. In offshore data, multiples are the main coherent noise. Valenciano (2008) addresses the problem of migrated multiple reflections in inversion by applying a pre-processing step to attenuate them. Here, using the Sigsbee2b dataset (Paffenholz et al., 2002), we characterize the multiples in the subsurface-offset domain, detail the pre-processing step and show the impact it has on inversion.


next up previous [pdf]

Next: Linear least-squares inversion Up: Reconciling processing and inversion: Previous: Reconciling processing and inversion:

2009-04-13