In marine environments, three-dimensional reflection seismic data is normally acquired in a so-called ``wide tow'' streamer configuration, illustrated in Figure 1. Ironically, the crossline azimuth range of this data is less than with most land acquisition geometries, so geophysicists often call towed streamer data ``narrow azimuth'' data. Note that the crossline shot interval, , is chosen such that the outermost receiver line on one swath overlaps the innermost receiver line on the previous swath. Figure 2 illustrates that such an acquisition geometry produces a regularly sampled crossline CMP axis, if cable feathering is absent. In one sense, this geometry boasts some degree of optimality, as it produces a well-sampled 3-D image at a minimum cost.
Unfortunately, geometry shown in Figure 2 causes the 3-D extension of the SRME method of multiple prediction to fail spectacularly. SRME requires to be relatively small-in practice, roughly the same as commonly chosen crossline receiver line spacing parameters van Borstelen (2003). 3-D field datasets commonly have a crossline shot interval of up to ten times the crossline receiver line spacing. Workarounds for the 3-D sampling problem include: ignoring crossline structure and using a 2-D prediction, massive (270,000 CPU hours) shot interpolation Kleemeyer et al. (2003), sparse inversion of the crossline multiple contribution gathers Hokstad and Sollie (2003); van Dedem and Verschuur (2002), and novel acquisition geometries Paffenholz (2003). Currently, none of these methods combines proven accuracy with computational/cost efficiency.