Conclusions and Future Work

In this paper we discussed application of LSJIMP on shallow water-bottom data-sets. At first, results did not look promising, but they motivated us to probe into fine details of the method and to come up with improvements that would make it work better. Crosstalk models and regularization operators proposed under LSJIMP seem to break down for extremely shallow water-bottom data sets like SYN-1, as shown through equations (12) and (13). Discriminating between signal and crosstalk on the basis of moveout did not work very well for this case. We also proposed and tested a new scheme for generating crosstalk, which seems to be a slight improvement over the previous one.

As was evident in Figure , we were not able to suppress primary crosstalk effectively on multiple images. Derivative between images could suppress it but its convergence was really slow and our new proposed scheme for difference did not work. Possible reason for slow convergence might be presence of bad eigen value spectrum, that results in appearence of smooth components at the very end. Preconditioning might handle the problem of slow convergence.

There lies a lot of potential in designing appropriate weighting functions and crosstalk-modeling operators that can make the method of Least-Squares Joint Imaging more effective. We would also like to see how our method works if, instead of the NMO operator, we use a migration operator for imaging. This might yield better results.