Although I am enthusiastic about the new potential I see in using DAMF (data adaptive multidimensional filtering) for determining a pilot trace, DAMF is complicated and unfamiliar, so we first review the sequence of steps that follow the preparation of the pilot traces. Kjartansson's pilot trace is simply a moved out CDP stack. Kjartansson crosscorrelates pilots against individual traces and then scans each crosscorrelation for a maximum value, a process I will call picking. It is the picked times that are plotted in the midpoint-offset plane. The principle of superposition of seismic events of various stepouts applies before picking, not after, so it is not surprising that Kjartansson's attempted decomposition of his already noisy time picks yields an unconvincing map of time anomalies. I plan to attack this problem by simultaneously doing two things differently.
First, my DAMF pilot traces should be stable in the presence of velocity residual, varying reflector image, and shot statics but totally independent of the single effect of delay near and below the geophones. (This makes them a good base from which to measure those delays.) Second, I do not pick the crosscorrelation until after slant stacking (which transforms offset to depth). Thus, I pick to collapse a volume of coherency as a function of depth, midpoint, and slowness to get a plane of slowness as a function of depth and midpoint.
This plan raises an interesting question, ``What is the theoretical validity of the operation of backprojecting a crosscorrelation function?'' The validity is certainly limited by some commutivity issues. I plan to revisit this question after my codes are more complete.