The averaging algorithm is quite mindless in that it does not account for structure. To date this has not presented a problem as structural changes that occur over three adjacent shots are not that large.
Two methods are available for choosing which shots get replaced. The manual method involves supplying the program with an XSD header value at pick location file containing at least one pick from any shot that you wish to replace. Of course this means that you have to actually examine at least a portion of every shot on the line.
If you are working with a terrabyte of data you may wish to use the automatic option. Here a three record rolling filter is cascaded along the line comparing the central shot with the ones on either side.
The comparison algorithm forms three sets of averaged autocorrelations [A1, A2, A3], one for each record of the rolling buffer, using a simple three step process:
1. An autocorrelation is calculated over a user defined window for every tenth trace of the record.
2. For each record these autocorrelations are summed and the result normalized by the number of autocorrelations calculated within each record.
3. The summed autocorrelation is in turn normalized by it's zero lag value.
Three crosscorrelations are then computed [C12, C13, C23] as follows:
The amplitude of the zero lag samples from the above crosscorrelations are compared. When
then the central shot record is significantly different from the surrounding pair. A threshold value is then calculated
[ C13(0) - C12(0) + C13(0) - C23(0) ] / 2
which if greater than a user defined threshold value will result in the central record being flagged for replacement.
During the 1990 Trinidad Tringas 3D seismic acquisition a time-sharing arrangement was entered into by Shell and Amoco to assure controlled shooting over their adjacent blocks. The agreement remained in place until Shell finished it's main survey and began infill shooting. At that point both boats began shooting simultaneously resulting in recorded data containing seismic arrivals from both sources (fig. 1).
The affected data was stacked and examined onboard using an IQC system to determine if the other source was going to be a problem. A decision was made to continue the acquisition.
The data was subsequently interpreted, prospects mapped, recommended and drilled. During the post appraisal process it was discovered that the seismic and geologic interpretations could not be reconciled. A reprocessing project is currently underway in part to answer the question "Why not?".
There are several facts that allow use of the shave routine to solve the multiple source problem in this instance:
1. The other source is an airgun cycling at about the same time interval as our source. The shooting however did not initiate simultaneously so that the onset of energy drifts about 1900 milliseconds with each successive record (fig. 2)
.2. The other source was sufficiently far away that the amplitudes associated with it's arrivals are not significant above about 1600 milliseconds.
3. Over the zone of interest [1800 to 3000 milliseconds] the amplitude level of the extraneous arrivals is about equal to that of the energy arriving from Amoco's source.
4. Below 3000 milliseconds the amplitude of the waveforms arriving from Shell's source are significantly greater than those from Amoco's source.
5. The 1900 millisecond drift means that every 4th record has no energy recorded from Shell's source as they are cycling their compressors during that time.
6. The Shell source is mobile causing the orientation of the direct arrivals to vary considerably throughout the data volume.
When Shell's direct arrivals interfere with the zone of interest on the central record in the filter buffer they have yet to arrive on the subsequent record and are of insignificant amplitude at that time on the preceding record. This allows the application of shave within a window that excludes the previous and subsequent Shell shots.
The data was passed through five iterations of shave using the overlapping windows shown in table1.
Window Start (ms) | Window End (ms) |
|---|---|
Threshold testing was done for each window to provide a difference threshold suitable for detection of the Shell source. This was accomplished by executing shave using the chosen window on a test suite of data then examine the printout file. Documentation for the crosscorrelation amplitudes are reported to the printout file for every position of the three record buffer:
In the event that C13 is a maximum the calculated threshold is also displayed:
(13 - 12 + 13 - 23) / 2 = 0.1341739
Should the calculated threshold be greater than the user supplied value then an anomalous event is detected and the record flagged for replacement:
(13 - 12 + 13 - 23) / 2 = 0.1341739
Anomalous Energy Detected on RecNum 328
Through correlation of this information with the undesired arrivals in the input seismic data one can quickly decide if the auto-detect mode of the shave routine will be of use. If not then one may always pick the records to be operated on in XSD and supply an XSD header value at pick location file.
The following flow was used in the application of shave to the Tringas 3D survey:
shave -s2000 -e3400 -thresh 0.02 -Sshavestats_2000_3400 |
shave -s3200 -e4000 -thresh 0.03 -Sshavestats_3200_4000 |
shave -s3950 -e4800 -thresh 0.04 -Sshavestats_3950_4800 |
shave -s4750 -e5500 -thresh 0.05 -Sshavestats_4750_5500 |
shave -s5200 -e6000 -thresh 0.06 -Sshavestats_5200_6000 -mute -Odataout
The output data (fig. 3) is now largely free from the effects of Shell's source. The output shave statistics files documents the average correlation difference for every record along with the number of records that exceed the threshold. The minimum and maximum differences encountered are also listed:
autodetect threshold = 2.0000000E-02
Total Anomalous Detections = 151
Minimum Threshold Detected = 9.6857548E-05
Maximum Threshold Detected = 0.1642547
Covariance Anomalies Above Threshold
------------------------------------
The shave routine can remove systematic noise from marine shot sorted data with minimal loss of fold. It has been successfully applied to the 1990 Tringas marine 3D survey. Without shave the processed results would have exhibited a significantly poorer focused image of the subsurface as illuminated by Amoco's source.
Return to the USP Home Page.
