Depth Migration Before and After LSJIMP  

My particular LSJIMP implementation would likely come early in a modern seismic processing flow; after stacking velocity analysis, but before depth migration and interval velocity model building. Therefore a scenario of considerable practical interest is to view LSJIMP as a multiple suppression algorithm, and to treat the estimated LSJIMP primaries (after inverse NMO) as demultipled data, which would then be depth migrated.

WesternGeco supplied a depth interval velocity model with the data. I migrated the raw data and the LSJIMP primary data using a 2-D Extended Split-Step prestack depth migration algorithm Stoffa et al. (1990) with three so-called ``reference velocities'' to handle lateral velocity variation. Image sampling in depth is 6.67 meters. The migration algorithm outputs gathers as a function of depth, midpoint, and subsurface offset. Using the method of Sava and Fomel (2000), the offset gathers are converted to Angle-domain Common-image gathers (ADCIGs) as a function of opening angle at the reflector.

Figure shows angle stacks, after a z^1.5 gain, of the Mississippi Canyon raw data, the data after LSJIMP, and the difference of the two. The Figure is similar in style to Figure . The removed multiples are simplest to view on the left-hand side of the section, where the geology is less complicated than under the salt body. In the sedimentary region, we notice, as before, that LSJIMP can cleanly separate primaries from many different multiple reflections. In the salt region, the results are somewhat muddied, since migration strongly defocuses multiples. We see that much multiple energy has been removed, though much remains. Subsalt primaries, already difficult to spot without any multiple energy, are uncovered better, especially around 3500 meters depth. The dominant dip is negative (toward the surface with increasing midpoint).

For reasons explained in more detail in section , some primary energy is seen on the difference panel, where we hope to see only multiple energy. The loss of primary energy, while well below the clip value anywhere, is strongest for the top of salt reflection. Much of the lost energy has a high spatial wavenumber, and likely arises from diffractions which my implementation of LSJIMP cannot model. Also, the large velocity contrast at the top of salt gives rise to strong head waves, which have a high apparent velocity. These events, which are not flat after NMO for primaries, are filtered out as noise by the LSJIMP regularization which differences across offset.

 

stackcomp-weimig.gulf
Figure 12 Angle stacks after prestack depth migration. Panel (a): Raw data. Panel (b): LSJIMP estimated primaries after inverse NMO. Panel (c): Difference.

Figures and illustrate, after z^1.5 gain, ADCIGs at midpoints 55 and 344 (of 750), respectively. Compare these Figures to Figures and . Figure is extracted from the sedimentary region of the data. Notice that LSJIMP has quite cleanly separated multiples from the primaries, and certainly improved our ability to interpret the angle gather for amplitude-versus-angle phenomena.

 

angcomp.weimig.gulf.55
Figure 13 Angle gather comparison at CMP 55 of 750. Panel (a): Raw data. Panel (b): LSJIMP estimated primaries after inverse NMO. Panel (c): Difference.

Figure , on the other hand, is extracted from the salt-bearing region of the data. Visually, it is far more difficult on the angle gather to distinguish primaries from multiples, although peglegs from shallow reflectors, between 3500 and 4300 meters depth, are recognizable and cleanly removed from the data, uncovering some hidden primaries. Notice that some downcurving reflections within the salt (1900 to 2800 meters depth), which may be internal multiples, are attenuated by LSJIMP, since they are not flat like true signal events. Furthermore, the events with negative dip below 4000 meters, which may be out-of-plane reflections or diffractions, are also attenuated somewhat.

 

angcomp.weimig.gulf.344
Figure 14 Angle gather comparison at CMP 344 of 750. Panel (a): Raw data. Panel (b): LSJIMP estimated primaries after inverse NMO. Panel (c): Difference.