Conclusions on the 2-D Data Results  

In general, the LSJIMP method demonstrated high-quality separation results on the 2-D Mississippi Canyon data example. Primary energy was preserved and nicely uncovered from strong, shallow pegleg multiples. The method lacks somewhat in its ability to adequately model salt-related reflections, kinematically or in terms of their amplitudes. For one, the rugosity of the top of salt reflection negatively affected reflection coefficient estimation, especially since my model of reflection coefficients assumes spatial contuinuity, and thus to imperfect separation of salt-related multiples from the data.

While in some cases HEMNO could accurately model the kinematics of these multiples (see Figures and ), in cases where the salt geometry varied too fast spatially, HEMNO's performance suffered. I conclude that for moderate, spatially ``smooth'' dips, HEMNO works well. Failures point to migration (especially prestack depth migration) techniques to tackle the salt problem.

In section I investigated what, if anything, the multiples add to the LSJIMP inversion. The improved separation results obtained after adding the multiples confirmed a central assertion about LSJIMP: the use of multiple reflections in a global inversion add a useful constraint to discriminate between signal and noise.

In section , I applied the LSJIMP nonlinear updating scheme outlined in section and found that in particular, poorly-estimated reflection coefficients can be improved by the updating scheme, which in turn leads to improved separation results. However, I found that the result of updating the crosstalk weights was negligible to the separation results.