Conclusion

Multiple attenuation can be cast as a problem where events are separated according to their patterns (e.g., multidimensional spectra). A pattern is made up of both the kinematic and amplitude information that PEFs can approximate. The pattern-based technique is a two-step procedure. Firstly, non-stationary time/space domain PEFs are estimated for both primaries and multiples. These filters are estimated from a model of the multiples usually computed with SRMP and a model of the primaries computed with the Spitz approximation. The Spitz approximation assumes that the noise and signal are uncorrelated. Secondly, multiples are separated from the primaries in a least-squares sense according to their multidimensional spectra.

As illustrated with the Sigsbee2B dataset, this approach has the potential to separate primaries and multiples very well as long as an accurate model exists for the PEFs estimation. When no model of the primaries exist, the Spitz approximation, which convolves the data with the noise PEFs, leads to a very good attenuation of the multiples if high dimensions filters are used (i.e., 3-D vs. 2-D). Indeed, primaries and multiples are less likely to look similar.

Comparing adaptive and pattern-based subtraction indicates that the latter removes the multiples almost always better, except in areas where the primaries and multiples are correlated. An important property of the pattern-based approach is that it seems to cope well with modeling inadequacies. For instance, with the Gulf of Mexico dataset, although not properly reconstructed, short-offset multiples were nonetheless removed.

Multiple attenuation can be viewed as a two steps process where multiples are first predicted and then subtracted. Both steps are equally important but most of the efforts are usually concentrated on the prediction step and not the subtraction. Since in practice it remains impossible to get a perfect multiple model due to the limitations of the acquisition geometry and interpolation/extrapolation techniques, new subtraction methods are needed. The pattern-based method presented in this Chapter is a successful tool for removing coherent energy in seismic data. This technique offers a viable alternative to adaptive subtraction by being less sensitive to errors in the multiple model. In addition, compared to adaptive subtraction, the pattern-based technique alleviates the strong assumption that primaries have minimum energy.

Chapter illustrates the pattern-based method with a 3-D field data example where the acquisition geometry limits our ability to predict surface-related multiples accurately.