Summary

I have shown that RTM is equivalent to GDM, where the generalized migration image at a point ${\bf {x}}$ is obtained by taking the dot-product of the GDM operator with the data. GDM is a generalization of simple diffraction-stack migration which sums the data over the hyperbola-like curve associated with the primary reflections. The advantage of GDM over RTM is that moveout-based noise can be precisely filtered from the migration kernel, which can lead to precise suppression of migration artifacts and avoids unintentional filtering of data. In contrast, the standard RTM approach can only filter the backpropagated data, which is an inseparable mixture of contributions from both the migration kernel and data. I demonstrated the effective reduction of migration noise by dip filtering the decomposed migration kernel for both realistic synthetic data and field data. Some other useful applications of filtered GDM include mitigation of migration artifacts due to aliasing and multiples (Zhou et al., 2003; Zhou, 2004).

The biggest challenge in implementing GDM is the demanding memory and computation expense of the migration operators. Thus, full blown GDM should be restricted for now to smaller data sets, but with wavelet compression it can be practical for target-oriented migration and iterative least-squares GDM.