Figure 7

Figure 8

We would like to similarly match the estimated multiples to the original data but we cannot, because nothing prevents the pattern-matching algorithm from attempting to match the primaries as well. This is obviously undesirable because, as much as possible, we want to keep the primaries as they are in the original data. Here is where the estimate of the primaries will help. We can estimate filters to simultaneously match both the estimated primaries and the estimated multiples to the data as in Guitton's thesis 2005 thus preventing the primaries to be matched to the same dataset as the multiples. An easier approach is to use the estimated primaries to compute a mask such that where the primaries are present the mask is zero and therefore the primaries are not matched when attempting to match the multiples Claerbout and Fomel (2002) . This is the approach I used here although in practice the other approach is likely to be better. In order to compute the mask I first computed the envelope of the estimated primaries, and then chose a threshold amplitude above which all samples were set to zero and below which all samples were set to one. This mask was then smoothed with a triangular filter both in offset and in time. Figure shows the envelope of the estimated primaries and the mask.

Figure 9

The result of applying the weighted adaptive matching to the estimated multiples is shown in Figure . The water-bottom multiple has not been well recovered, in contrast to the other multiples. The primaries didn't leak much into the multiples, which is a very satisfactory result. Figure shows the residual, obtained by subtracting the matched estimated multiples from the data. Here we see that although the result is not perfect, most of the multiple energy has been attenuated except for the water-bottom multiple. In particular, the primaries have been well-recovered.

Figure 10

Figure 11

4/5/2006