简介：由于一些原因，壳牌在加拿大富特希尔斯地区获取的地震成像质量非常糟糕。不断提升地震成像的质量对壳牌未来在该地区的勘探获取成功非常关键。

Improving seismic imaging in Alberta

Continuously improving the seismic image quality is key to Shell’s future exploration success in the Canadian Foothills. 

Schematic diagram of rays that have the same receiver and the same offset but a 180º difference in acquisition azimuth. Consequently, the recorded wavefields from the shots have experienced different near-surface effects.

Seismic data quality in the Canadian Foothills (Figure 1) is notoriously poor for a number of reasons. First, the rough topography, combined with limited access, often leads to sparse data acquisition that results in inadequate subsurface illumination. Second, the geology is complex with rapid changes in near-surface velocities caused by thrusting, modern and ancient fluvial systems, and buried karsting in the carbonates. These near-surface conditions give poor source and receiver coupling. The karsting creates considerable scattering and ground-roll effects of the seismic wavefield that mask the primary reflection energy. Further, thrust sheets carry thick low-velocity clastic sediments, often greater than .6 miles (1 km). The resulting boundary produces large ray-bending effects. When combined, these factors make seismic depth processing extremely challenging, and often it is difficult to produce an image that enables a consistent structural interpretation of a prospect.

A recent prestack depth migration (PSDM) project over a well-documented displacement transfer zone in the Brazeau Thrust, known locally as the “Swing-Back,” was able to provide an improved structural image of the Limestone gas field (Figure 2a). The imaging problems caused by the Swing-Back result from a dramatic change in the along-strike velocity field where more than 1.2 miles (2 km) of Paleozoic carbonates disappear across a near-vertical dip parallel fault and are replaced by Cretaceous clastics. To identify possible seismic processing issues, Shell reviewed the project and concluded that it was executed to the current best practices.

Nevertheless, the existing product failed to image the geology below the Swing-Back correctly; hence there was a business need to investigate new methodologies in the effort to improve the image. We focused on the practices of ignoring the azimuth information and applying static shifts to compensate for near-surface velocity variations, believing that this would give the largest improvement. The physical assumptions made in these two steps are likely not valid in mountainous areas with rapidly varying near-surface geology and will possibly introduce larger errors than higher-order effects like neglecting anisotropy in the velocity depth model.

Azimuthal data regularization

Compared to conventional marine acquisition, land 3-D acquisition geometries have the advantage of a wide-azimuth distribution, even though this information is typically disregarded during the PSDM workflow. In this study, we used the available azimuth information during the data regularization and the velocity model building. Usually, data regularization is carried out in the constant offset domain with a trace interpolation scheme (Figure 2b). In the case of wide-azimuth land data, traces with the same offset but quite different azimuth (Figure 3) are often used to construct missing traces. Because these traces may have largely different ray-paths, this approach frequently yields interpolated traces that are of unsatisfactory quality. To improve on the interpolation result, we needed to use input traces that are more comparable, which meant traces that have similar azimuth and offset. Using just three 60º azimuth ranges for trace interpolation resulted in a noticeable improvement in the PSDM quality, especially in the shallow section (Figure 2c). At the near-vertical Swing-Back, where the wavefield is particularly complex, this simple prestack interpolation is an improvement but is still sub-optimal and not able to correct image artifacts like migration swings. In the future, Shell hopes square migration methods will help to improve further on this result.

Near-surface velocity model

The limited near-offset sampling in the seismic acquisition and the seismic noise mean that conventional reflection seismic in the Canadian Foothills has a very poor velocity resolution from the surface down to 3.300 ft (1,000 m) depth. Hence, for most PSDM projects, the near-surface velocity model is populated with velocities, sometimes as high as 13,120 to 14,765 ft/s (4,000 to 4,500 m/s), derived for the same geological unit in deeper parts of the data. However, a vertical seismic profile (VSP) acquired in another area showed that velocities could be as low as 1,640 to 2,625 ft/s (500 to 800 m/s) in the weathering impacted layer down to around 60 ft (20 m) depth. Beyond the weathering layer, velocities increased rapidly to 4,920 to 6,562 ft/s (1,500 to 2,000m/s) at a depth of about 150 ft (50 m) where the geology is much more consolidated. Onwards to a depth of 1,970 ft (600 m), the velocity reached the level typically used as a background velocity. Using the information from the VSP velocity profile, it was decided to replace the constant 136,120 ft/s velocity in the shallow part of the model with a varying near-surface velocity model. This model was computed by Shell’s front-end contractor, CGGVeritas, via tomographic inversion of the first arrival picks. It was consistent with the conventional refraction statics solution apart from some smoothing. The resulting near-surface velocity field had large vertical and lateral variations (Figure 4). The range of velocities was similar to those observed by the VSP. 

By incorporating detailed near-surface velocities in the depth model, travel-time differences associated with the near-surface layer are accounted for in the ray-tracing during the PSDM. This allows for the large ray-bending and hence results in a more accurate positioning of the seismic energy. The inclusion of the near-surface model into the depth model also allowed us to compensate for the near-surface velocity effects in a dynamic way during the migration instead of just as static time shifts. Hence, the refraction statics applied for improved shot processing had to be removed prior to the PSDM. It should be noted that the migration also has to be carried out from the surface to minimize the effects of the rough topography on the image. The large image improvement obtained by this simple and fast procedure demonstrates that it can be important to include the strong gradient of the near-surface velocities into the depth model.

Azimuthal velocity model updating

During the testing of the azimuthal data regularization, we observed that the data quality differed depending on the azimuth. We capitalized on this observation by honoring the azimuth during the velocity-model updating via generating directional common-image gathers (CIGs) with 12 azimuth (0-360º) and 15 offset slots. A first analysis of the CIGs showed that the residual moveout (RMO) varied with azimuth, probably caused by illumination issues and, to a much lesser extent, by anisotropy. Furthermore, due to the non-reciprocal acquisition geometry and shot-receiver coupling problems, reciprocal azimuths actually had different RMO and data quality.

Consequently, we picked RMO for each azimuth range independently and used the resulting 12 RMO datasets with their according azimuth information in a joint isotropic travel-time inversion (TTI). In this way, the TTI posted the velocity update approximately back into the direction from where the data originated. Compared to a standard TTI just using a single azimuth, the resulting model was more stable (Figure 5). For this specific 3-D project, the velocity difference between azimuthal TTI and conventional offset TTI has been as large as 1,475 ft/s (450 m/s).

After migrating with the azimuthal TTI updated model, flatter CIGs containing more coherent signal were obtained (Figure 6b), and the according stack of the multi-azimuth updating shows an enhanced image.

Conclusions

The total image improvement obtained for the Swing-Back highlights the advantage of incorporating a near-surface velocity from tomographic inversion and using multi-azimuth information for the data regularization, the PSDM, and the velocity model building sequence. The new image was of sufficient quality to support Shell’s business decisions, but further refinements are necessary to drive additional exploration efforts. Least-squares migration and joint inversion of refraction and reflection data for the model building have great potential to provide additional seismic uplift, and Shell currently has both techniques under development in research. Moreover, after removing assumptions like static shifts that introduce errors of first order in the seismic processing, the data are now of sufficient quality to include higher-order effects, like anisotropy, into the depth model.