Conclusions

On the CGG Green Canyon 3-D dataset, LSJIMP again demonstrated an excellent ability to separate primaries and multiples. The data subset shown in the paper came from a sedimentary minibasin, which boasts a simple velocity profile, high signal-to-noise ratio, and fairly mild (but non-trivial) dips. The full dataset released by CGG contains much data recorded over salt. In preliminary tests on one salt region, NMO and HEMNO proved unable to correctly image primaries or multiples. The salt bodies often exhibit crossline dips of over 30 degrees, which, when combined with the data's inherent sparsity, severely test even the most advanced imaging techniques.

The data subset is particularly well-suited for Radon demultiple, with its large velocity gradient and gentle geology. I tested least-squares hyperbolic Radon demultiple (LSHRTD) and found that LSJIMP compares quite favorably, both in terms of computational efficiency, multiple separation, and amplitude preservation.

A quantitative study of prestack reflection amplitudes confirmed what was suspected; LSJIMP's ability to remove multiples and random noise, as well as its ability to use multiples and other constraints to interpolate missing traces, improves prestack amplitude analysis.