(C. Liner and J. Seales)
In and earlier post, spectral decomposition (SD) of the the Dickman 3D data using simple bandpass filtering. This has been done using both seismic unix and SMT filtering tools. Here we consider the sensitivity of the SD output on choices of the filter parameters.
A broadband time slice (847 ms) through the data shows a well-defined channel. Aside from a slight amplitude change to the SW, we see no evidence of overbank or secondary channel features.
Figure 2 is an SU plot created using a filter centered on 41 Hz generated by a shell script. The image shows a clear feature that trends away from the channel to the SW. The filter parameters are shown along with a dash-yellow outline of the feature under discussion.
Equivalent 41 Hz time slices at 847 ms (Figures 3 and 4) were created using the trapezoid filter in SMT. Apart from grayscale color bar differences between SU and SMT, the time slice features are consistent between SU and SMT. Both filters created in SMT were centered on 41 Hz, but Figure 3 uses a 2 Hz top while Figure 4 is more constrained with a 0.2 Hz top. Visually, these results are equivalent indicating the SD attribute computed in this manner is not unstable with respect to filter parameter choices.
To study detailed variation between these SMT results, subtractive difference volumes were generated using functionality in SMT (Figures 5 and 6). These show there are, indeed, slight differences between the output of the two SMT filters created. Both figures show the same result, but with opposite amplitudes due to changing the subtraction order. The difference plots show energy in horizontal bands representing acquisition footprint (source orientation), a large fault in the NW corner, a bright karst (sinkhole) feature near the center, and a network of curved lineations of unknown origin. In an absolute sense, the maximum amplitude difference between the 2Hztop and 0.2Hztop filters is on the order of 10% in the mentioned features, otherwise it is less that 2%.
If this kind of spectral decomposition is to be generally useful, it is important to quantify the sensitivity on filter parameters beyond the visual difference described above. To accomplish this, we set out a grid of 27 points in the 3D survey area about 2500 ft apart (Figure 7). At each location, the amplitude for 2Hztop and 0.2Hztop results were measured and cross plotted in Excel (Figure 8).
Finally, we notice that an 847 time slice through the SPICE attribute volume (Figure 9) seems to indicate the same channel feature seen on 41 Hz spectral decomposition results. We are planning further investigation of this tantalizing relationship.