To achieve this goal, I use three multiple removal techniques. The first method consists of separating the multiples with a high-resolution hyperbolic Radon transform Kostov and Nichols (1995); Lumley et al. (1995). This method is particularly effective at separating multiples and primaries when both have different moveouts. The high-resolution provides high separability in the Radon domain thanks to a Cauchy regularization of the model space Sacchi and Ulrych (1995). This method is very robust and might be the only choice for separating primaries and multiples with, for instance, land data.
The second method is very similar to the Delft approach Verschuur et al. (1992) where the multiples are first predicted and then adaptively subtracted. This method can predict any type of multiple as long as enough data are recorded. In addition, no model of the subsurface is needed. The multiples are removed by estimating non-stationary matching filters in the time domain Rickett et al. (2001). The Delft approach works particularly well for marine data. Some recent examples with land data show also promises Verschuur and Kelamis (1997).
The last method separates the primaries and the multiples based and their pattern Brown and Clapp (2000); Manin and Spitz (1995). The patterns are approximated with non-stationary multidimensional prediction-error filters (PEFs). These filters are estimated from a noise and a signal model. The noise model is provided by the Delft approach and the signal model is estimated by convolving the data and the noise PEFs. This method has the ability to work in any dimension and is robust to modeling uncertainties Guitton (2003).
These three methods are tested on a deep water, 2-D Gulf of Mexico dataset provided by WesternGeco. One interesting geological feature of this dataset is a shallow salt body that creates shadow zones and strong multiples. There are also off-plane/3-D multiples that make the noise attenuation quite challenging. With this dataset, I show that the best multiple attenuation result is obtained with the pattern-based approach with 3-D filters.
In the next section, I describe each multiple attenuation technique emphasizing their strengths and weaknesses. Then, I apply each of these methods on the 2-D Gulf of Mexico dataset.