The CASC93 dataset Nabelek and Zandt (1993) is a passive seismic experiment where arrivals from distant earthquakes were recorded by an irregular 2D array of seismometers in central Oregon.
As we can see in Figure 2, the interpolated result with the PEF estimated from the shifted data looks better than the original multiscale result. In this case 7 different scales were used for (b), while for (c) 10 additional versions of the data were used. The difference is most pronounced in the area with coarser sampling, since the different grid positions would return more unique information than in the densely-sampled area.
Next, a 3D synthetic case is examined. The ``quarter-dome'' model Claerbout (1993) has been sub-sampled randomly so that only 10 percent of traces remain. It was then interpolated with both a standard multiscale PEF as well as a multiscale PEF estimated with different grid positions. The results are shown in Figure 3. The number of versions of the data varies even more in this case, where 5 different scales were used in the multiscale estimation, while 39 additional versions of the data were used when grid position was varied. The results are improved, although nearly 8 times as many copies of data were used. However most of those versions of the data were at the coarser scales.