Fast velocity model evaluation with synthesized wavefields

Introduction

Because of the high velocity contrast between salt and sediments, accurate interpretation of subsurface salt structures is vital to obtaining a high-quality seismic image. A variety of tools have been developed to aid interpreters in this task, including several different approaches to automatic seismic image segmentation (Lomask et al., 2007; Halpert, 2011). Methods such as these can provide several possible salt body interpretations, whether through uncertainty analysis of the boundary (Lomask, 2007) or interpreter guidance of segmentation results (Halpert et al., 2011). Because of computational constraints, it is often not feasible to run full migrations to test each of the possible models. Therefore, methods to quickly and efficiently test the imaging effects of each model are of great interest. Several approaches to the problem of fast re-migration of seismic data have been proposed. Many of these approaches rely on a fast variant of beam migration (Hill, 1990), one form of target-oriented imaging. Beam-based methods (Wang et al., 2008) operate in the post-stack domain, and often rely on qualitative judgments of image quality to evaluate different velocity models. Pre-stack imaging methods using reverse-time migration (RTM) have also been proposed (Wang et al., 2011), and would allow for analysis of valuable velocity information in the form of subsurface-offset or angle gathers. However, at present this approach is likely only feasible when testing a small number of limited-size models. A further option, and the one explored here, is to synthesize new datasets from an initial image, and use these datasets to test velocity models. First, a new source wavefield can be derived from the initial image using modified prestack exploding reflector modeling (Guerra, 2010). This allows information from the inevitable velocity errors in the initial model to be preserved and, hopefully, corrected. Second, a form of Born modeling (Stolt and Benson, 1986) can be used to synthesize a new dataset from the initial image (Tang and Biondi, 2010; Tang, 2011). The new dataset can be designed with an arbitrary acquisition geometry, creating opportunities for target-oriented imaging. Furthermore, since a generalized source function is used, a useful image with prestack velocity information can be obtained by migrating only a single shot. Crosstalk artifacts can be avoided by targeting only small, isolated locations from the initial image. This approach also allows for a simple quantitative measure of image quality or focusing. In the following sections, this method will be described and demonstrated using data from the 2D Sigsbee synthetic model. The accuracy and computational efficiency of the results show great promise for this method working in conjuction with automatic segmentation as part of a full 3D interactive interpretation toolbox.

Fast velocity model evaluation with synthesized wavefields