Velocity model evaluation through Born modeling and migration: a feasibility study

Introduction

Building an accurate seismic velocity model is essential for obtaining an acceptable image of the subsurface. When the subsurface is especially complex, for example in geological settings dominated by irregularly-shaped salt bodies, this task becomes particularly challenging. The large contrast between salt and sediment velocities magnifies the effects of inaccurate salt interpretation, resulting in a poor image. Unfortunately, velocity model-building is a time-consuming process that often requires several iterations. A typical salt-interpretation and model-building workflow involves iterative sediment- and salt-flood migrations to identify the top and base of the salt bodies (Mosher et al., 2007). In situations where the top or (especially) base salt interpretation is uncertain or ambiguous, several different salt scenarios may be geologically feasible. Therefore, a means of quickly testing the effects of several different possible velocity models would be extremely useful for judging and refining salt interpretations. Here, we investigate a Born modeling and migration scheme that allows for fast remigrations of data simulated from an initial image, that incorporates prestack velocity information from the initial image's subsurface offset gathers.

An interactive interpretation and imaging environment would be a valuable model-building tool, and several different approaches have been proposed. Wang et al. (2008) introduced a fast migration scheme built on Gaussian beam imaging (Hill, 1990), that can quickly test different salt scenarios. This method relies on seismic demigration and redatuming of wavefields to reduce the computational expense of migrating with several different velocity models. However, this method operates in the poststack image domain, neglecting velocity information that can be obtained from prestack images, and is limited by the approximations inherent to beam imaging. A similar approach has been proposed using reverse time migration (RTM) in a ``layer-stripping" manner (Wang et al., 2011), but this remains too computationally intensive to test more than a very small number of possible models. Chauris and Benjemaa (2010) have proposed another method using RTM, which aims to reduce computational expense by summing over time-delays in the subsurface rather than sources. However, at present this method has only been demonstrated in two dimensions, and it remains unclear if an extension to 3D is feasible. Finally, fast migrations may also be achieved through the use of simulated datasets derived from an initial image. Guerra (2010) synthesized wavefields using prestack exploding reflector modeling as a means for performing wave-equation tomography in the image space. However, the significant amount of preprocessing required, especially in 3D, makes this approach less appealing for interactive testing of several velocity models.

Born modeling (Stolt and Benson, 1986) is based on a single-scattering approximation of the wave equation. By taking advantage of this approximation, we can simulate a new dataset (Tang and Biondi, 2010; Tang, 2011) from an initial image, whose size and acquisition geometry can be selected independently from those of the original dataset. Furthermore, the simulated data can be migrated using generalized sources, drastically reducing the number of shots required. In the examples we show, only a single shot is required, allowing for migrations well within an interactive time frame. In order to improve the accuracy of this method, we use a generalized source function derived from subsurface offset gathers of the initial image. This allows for a more accurate and data-driven result than if a simple wavelet were used as the source function; in addition, including non-zero subsurface offset information into this source function incorporates important velocity information available from the initial image.

In the following sections, we review the Born modeling methodology and outline the procedure for obtaining the generalized source function described above. We then demonstrate the method using simple 2D synthetic models. Crosstalk artifacts arising from the modeling procedure limit these examples to isolated image points along a single reflector in the subsurface; however, these tests indicate that this method can effectively provide information about the accuracy of different velocity models on an image. Further enhancements and the inclusion of phase-encoding (Romero et al., 2000) strategies should widen the applicability of the method. Ultimately, we hope to combine this method with elements of an automated image segmentation scheme (Halpert et al., 2011) to create a powerful tool for interactive interpretation and imaging.

Velocity model evaluation through Born modeling and migration: a feasibility study