Attenuation of 2D specularly-reflected multiples in image space

I propose a new method to attenuate specularly-reflected multiples in the image space. The method is based on the difference in mapping between primaries and multiples in Subsurface Offset Domain Common Image Gathers (SODCIGs). I migrate the data with a velocity slower than that of the primaries but faster than that of the multiples. The primaries are therefore undermigrated whereas the multiples are overmigrated. For positive data offsets, the primaries are mapped to positive subsurface offsets and the multiples to negative subsurface offsets in SODCIGs. I then apply a tapered mute to eliminate the primaries and do adjoint migration on the multiples with the same velocity model to get an estimate of the multiples in data space. Similarly, a tapered mute is applied to eliminate the multiples and adjoint migration used to obtain an estimate of the primaries in data space. The estimate of the multiples is adaptively matched to the data with the estimate of the primaries used as a weight function to prevent matching the primaries. I illustrate the method with a 2D synthetic dataset and show that the primaries can be well recovered although some residual from the water bottom multiple remains. INTRODUCTION Specularly-reflected multiples, such as water-bottom multiples, peg-leg multiples and internal multiples contaminate the seismic data to varying degrees making it more difficult to process and interpret. Water-bottom multiples are relatively easy to eliminate since their moveout is completely predictable. Peg-leg multiples and internal multiples, on the other hand, are difficult to eliminate if the subsurface is complex. Surface-Related Multiple Elimination (SRME) (Berhout and Verschuur, 1997; Verschuur and Berkhout, 1997; Weglein et al., 1997; Dragoset and Jericevic, 1998; Dragoset, 1999) can be used to eliminate all multiples with at least one bounce at the water surface. SRME has been proven to be very effective when the data is finely and regularly sampled in the space coordinates and when the data aperture is sufficient to capture all surface bounces of the multiples. In many practical situations this is difficult to achieve and a large amount of data interpolation and extrapolation is required. Even when all the sampling conditions are met, SRME cannot be used to eliminate internal multiples unless the data is downward-continued to the multiple-generating layer, which in general is difficult to do for all layers. 37