Fig. 3 shows the input data (top panel), the zero-traces inserted migration (middle panel) and the SMM (bottom panel). The irregularity introduced by the mask is mild (the sixth trace has been eliminated in two regions). Inserting zero traces and imaging on a regular grid introduces noise farther away from the missing traces as depth increases. Overall, the image is full of incoherent high-frequency noise. Instead, the SMM result introduces no such problems. Even if the jump in sampling rate is large at the place of the missing traces (the step becomes instead of dx), the reflections off the side of the grid are minor. This makes me believe that for jumps in the grid step of the order of half of dx and under, the artifacts would be negligible.
Fig. 4 shows the effect of severe irregularities. Both images are severely affected. The S/N ratio of the zero-traces inserted surrogate irregular migration result is lowered to the limit of interpretablity, and any mistake while interpolating the traces (if we interpolate instead of inserting zero traces) would probably have severe effects as well. The SMM result is plagued by strong spurious reflections, but we notice that these are: a) localized, and b) highly coherent and dipping in the opposite direction from the true local dips. These are prone to filtering with a reasonable apriori assumption about the direction of the dip, and can be obtained as well from unmigrated data. The SMM image overall is crisp and interpretable. The fact that the incoherent noise has not been increased is particularly important because input datasets already suffer from a S/N problem, such as those in crustal seismology.