August 4, 1997
Paul G. A. Garossino [GTR-2718]
If done carefully, flexbinning has the power to dramatically increase the character continuity of the stacked data yielding a first order improvement in its interpretability. Done poorly it becomes a dangerous process that can introduce a destructive spatial alias with potentially disastrous effects on the resulting interpretation.
This paper introduces the -flexbin option of the USP sorting program sr3d2. The 3D marine dataset acquired over the Flamboyant Field area of offshore Trinidad has been used as a data example. Severe cable feathering was experienced during the acquisition of this data due to exceptionally strong tidal currents. This coupled with an asymmetric [12.5 by 40 meter] bin dimension resulted in a dataset having both a severe bin-wise fold variation and an erratic bin-wise source - receiver offset distribution. We will both demonstrate the data character instability brought about by this distribution of data and the effectiveness of flexbinning at solving the problem.
Algorithm
The flexbin algorithm implemented in sr3d2 was crafted based
on the minimalist philosophy "Introduce as slight a spatial alias as
possible". As such, the program was built to flex only when necessary
and when necessary only by the minimum amount necessary to regularize the
data. In areas of the survey where a full bin-wise distribution of offsets
is present no flexing is undertaken. In areas where holes in the bin-wise
offset distribution occur a controlled amount of flexing is used. The initial
step is to determine the offset distribution for a the active bin by construction
of a histogram of the input offsets. The histogram bin dimensions are based
on the sr3d2 command line entries for -dmin, -dmax and -ddel.
Data Example: Static Binning
When statically binned, a radical striping of fold magnitude is observed (fig. 2) in the Trinidad Tringas dataset varying in intensity as a function of both the amount of cable feathering and the acquisition direction. LI to LI fold differences range from zero to over 160 with a maximum fold reached of 197. The distribution of offsets varies almost as radically as the fold and even high fold is no guarantee that a reasonable offset distribution will be obtained.
Raw Nominal Fold Distribution Figure 2
High Fold
In an area of the survey where the fold is excessively high [LI 64, fold 100 -197] an LI was extracted and stacked (fig. 3). Examination of data from within the mute zone [~3000 ms and up] reveals that the near offsets were missing from a good portion of this line. A selection of bins (every 200th - fig. 4) were extracted substantiating that the very near offsets are indeed poorly represented [is present at all] here. A surgical mute applied to this LI (fig. 5) reveals that mid to far range offsets are in fact present continuously along the line.
Low Fold
In areas where the fold is very low one would expect a poor sampling of offsets. LI 69 [fold 0 - 70] was extracted as well (fig. 6). This stack reveals that at least the near offsets are present over most of the LI. The prestack bin data extracted from every 200th bin along the line (fig. 7) confirms that the near offsets are present but also reveals a dearth of mid to far range traces. In fact there is very little present on the entire LI except the near offsets. The resulting stacked data is essentially a near range stack of the dataset. Subsequently, the amount of destructive interference due to multiples and noise is significantly less than in the case of LI64 above resulting in traces with both a higher overall amplitude and a significantly different character. A surgical mute applied to this LI (fig. 8) reveals that problems that occur when the offset distribution is skewed to the near side.
Amplitude/Character Striping
The amplitude and character differences brought about by the variable bin-wise mix of trace offsets has a very pronounced effect on the spatial continuity of the resulting stacked dataset. This is most effectively illustrated by extracting a crossline record (DI 400 - fig. 9) which displays a marked striping of both amplitude and character. This striping manifests itself on a time slice (2192 ms - fig. 10) as localized loud amplitudes. The presence of such signal attributes is very undesirable for besides making interpretation difficult, it makes amplitude extractions dubious and wreaks havoc with coherency and horizon picking algorithms.
Data Example: Flex Binning
Now consider the same data zones extracted from the volume formed using the -flexbin option of sr3d2.
High Fold
Notice (fig. 11) that the shallow section has been populated verifying the presence of near offset traces. The overall amplitude of the LI is now greater and the continuity of events has been improved. The prestack gathers are shown both before (fig. 12) and after (fig. 13) the partial bin stack application. This regularization reduces the tendency to overweight the near offsets. Minimal differences are seen in the surgical mute result (fig. 14) though the shallow section is demonstrably more continuous. The most significant differences are at depth, between 1800 and 2600 milliseconds, from DI 800 to the end of the line.
Low Fold
Here (fig. 15) flexbinning has had a very beneficial effect. The overall amplitude level is both lower and more comparable to surrounding LI's while event character has been stabilized. This is especially evident over the zone of interest from 1800 to 2600 milliseconds from DI 1 to the center of the section. Flex binning has included traces required to regularize the offset distribution. This is most evident over the first half of the extracted line as is evident when comparing the original prestack data (fig. 7) with the same bins after flexing (fig. 16). The application of a partial bin stack (fig. 17) produces further regularization so that no one offset range is allowed to dominate the stack. An event character comparison is now possible between LI69 and LI64 allowing the correlation of events that were previously seemingly unrelated. The surgical mute stack (fig. 18) is now usable. Previously (fig. 8) it was a curiosity at best.
Amplitude/Character Striping
The striping previously evident (fig. 9) has been dramatically reduced (fig. 19). Character continuity now exists across the whole survey. In addition the shallow section is significantly more continuous with the inclusion of near offsets where required. The improvement in the interpretability of the time slice data (fig. 20) is quite remarkable.
The one pitfall of the procedure is occasional subtle interpolation errors due to aliasing of structural dip across the 40 meter bin spacing in the crossline direction. These errors manifest themselves as distinct terminations associated with a local break in the dip of an event. It is important therefore that a brute stack of the dataset be used for comparison in order to readily identify such events. If the interpreter is not privy to the original stack it is possible that a nasty misinterpretation of the data may occur.
The nominal fold after the flexbin process (fig. 21) shows very little of the striping evident in the static binned case. The little that does remain is due to gaps in coverage too extensive to be overcome by simple flexbinning.
Conclusions
When bin-wise fold and offset distribution irregularities are present a simple flexbin procedure followed by a partial bin stack prior to stacking the data can go a long way toward providing a much more stable dataset with respect to amplitude and character continuity. The pitfall of the process is the amount of spatial aliasing introduced during the operation.
Recommendations
Flexbinning is recommended when bin-wise fold, offset or even azimuth
distribution is found to be irregular to the point of affecting the amplitude
and character continuity of the stacked data. If an offset sort or surgical
mute of such a dataset is desired then flexbinning is manditory. A partial
bin stack of the data to regularize the fold within each offset range prior
to stacking the data is recommended in all cases.
Processing Flows
The flows used to bin and stack the 3D data used in these examples were:
Brute Stack
sr3d2 -Dsr3d2_volume -Psr3d1_table -go |\
bdmute -Pmute_picks -Mdiston |\
Flexbin Stack
sr3d2 -Dsr3d2_volume -Psr3d1_table -go \
-flexbin -dmin200 -dmax4200 -ddel50 |\
bdmute -Pmute_picks -Mdiston |\
Surgical Mute [add the following]
bdmute -Psurgical_mute_picks -Mdistoff
USP Notes.