and
, notice that each crossline midpoint gather is occupied
by one, and only one receiver line. Therefore, we can conceptualize a 3-D CMP
gather as a 2-D CMP gather with nonzero crossline offset, and thus remove the
crossline offset axis from five-dimensional CMP-sorted data, saving
considerable memory and computational waste. This approach is similar to
Biondi's 1997 combination of azimuth moveout (AMO)
Biondi et al. (1998) and common-azimuth wave equation depth migration
Biondi and Palacharla (1996) for the prestack imaging of primaries in data with
narrow-azimuth geometries. Unfortunately, the AMO transformation is not
generally valid for multiples.
HEMNO is well-suited to image multiples with the data geometry described above, primarily because HEMNO images pegleg multiples with a vertical time shift. Rather than correlating wavefields across possibly-undersampled axes like migration, HEMNO uses a measurement of the data's zero-offset time dip to account for structure-induced moveout variations. Because the crossline offset axis is removed, the computational cost increase of applying HEMNO to 3-D data versus 2-D data is only proportional to the number of crossline midpoints.
My particular LSJIMP implementation, presented in Section
, uses HEMNO, in conjunction with three amplitude
normalization operators to produce a ``true amplitude'' image of pegleg
multiples. From Figure
, recall that Snell Resampling
moves multiple energy across offset to make the multiple's AVO response
comparible with its primary. For this reason, the use of Snell Resampling in
the crossline direction runs contrary to the stated assumption that we store
only one crossline offset bin per 3-D CMP gather. Therefore, in this thesis, I
do not apply crossline Snell Resampling for narrow-azimuth data. In practice,
little useful angular information is anyway obtained in the crossline direction,
since in most cases the data will have a maximum crossline offset of merely a
few hundred meters.