SEISMIC DATA ENHANCEMENT BASICS
Below are the individual processing steps, described in the order in which they are applied. Some are applied only on marine data, some to both land and marine data, as noted in the text. Difficult data has been purposely chosen for the examples so that the results may be judged normal as opposed to near-perfect conditions.
The purpose of predictive deconvolution is to remove reverberations or ringing effects from seismic records caused by the source bubble and water bottom multiple reflections two issues that do not affect land seismic data processing. The procedure in predictive deconvolution is to design a least squares filter whose output is the inverse of the reverberating train of the input. Then by delaying this output and convolving it with the wavelet complex, we should be left with the primary reflection only.
If we let:
[fi] ni = 0 = filter coefficients
[xi] = input trace
[ri] ni = 0 = autocorrelation coefficients of the input (primary)
[bi] ni = 0 = cross
correlation coefficients of the desired output and actual
f0 r0 + f1 r1 + fn rn = b0
f0 r1 + f1 r0 + fn rn -1 = b1
f0 r2 + f1 r1 + fn rn -2 = b2
f0 rn + f1 rn-1 + fn r0 = bn
With the automatic marine
option the program computes its own gate start times and prediction
distances from the parameters geophone distance, water velocity,
water bottom depth, gun depth and streamer depth. The program will
attempt to calculate parameters for the number of gates requested
(maximum of five gates). If this is not possible, because of
insufficient room to grade operators then the program will decrease
the number of gates until only the water bottom multiples present
User defined gates may optionally be used by supplying gate start time, prediction distances and correlation window sizes for all gates which are to be used for all traces on the record.
To perform predictive convolution on a seismic record by means of successive iterations, the program automatically computes a prediction time from the autocorrelation of the input trace and uses this time to design the first prediction operator. After this operator has been applied, a new autocorrelation is calculated and a second operator is designed and so on. In this way, an operator for each reverberating wave train may be computed and applied separately. The computation of this normalized autocorrelation is independent of the unnormalized autocorrelation used in the predictive operator calculation. As an option, the program will output the prediction time for each iteration.
With this program, the number of iterations is usually set equal to the number of different reverberation periods which are present on the data. The following section of this paper displays examples and autocorrelations before and after successive iterations.
Analysis of the predominate reverberant periods and the frequency
content of the data is an important early step in the definition of
the processing parameters and sequence.
The bubble period (from the autocorrelation of the non-deconvolved
record) is approximately 110 ms long.
The water bottom multiple is 920 ms long.
Predictive deconvolution with a short (36 ms) prediction distance
and 1664 ms operator length effectively eliminates the bubble (see
second record from right hand side of displays).
Predictive deconvolution with a longer (850 ms) prediction distance,
164 ms operator, effectively eliminates the water bottom multiple
(see third record from the right hand side of examples).
5. Therefore, if iterative predictive deconvolution is used, this process should be run twice as shown by the fourth record from the right hand side of the examples. This effectively eliminates both types of reverberation events. For economy, the bubble can be eliminated before stack and the water bottom multiple can be eliminated after stack.
Bandpass filter parameters are then chosen from the harmonic analysis after deconvolution. The chosen time variant filter is shown in colour on both harmonics. This analysis is performed periodically along the project lines, so that some control over changing water depth effect and geologic conditions is obtained. Since the predictive deconvolution programs self-design operators from water depth and spread geometry when the marine option is used, or the autocorrelation amplitudes of the reverberating events when the iterative option is used, it is usually only necessary to test predictive deconvolution in a few selected locations.
After analysis and testing for deconvolution and filter parameters, the proof of any system is in the results of production processing. A deep water example is illustrated below (Fig 2). The section shown is highly faulted, but without too much throw on an individual fault. More recent sediments overlie the faulted blocks, and the velocity analysis (Fig 3) indicates interval velocities of 12,000 to 14,000 feet per second immediately below water bottom. (See interpretive overlay on deconvolved section). A wedge of low velocity material (7600 ft/sec) exists on the right hand side, followed by sections of approximately 10-12,000 ft/sec., 15-16,000 ft/sec., 10-11,000 ft/sec. and terminating in a rather indefinite series of interval velocities of about 17,000 ft/sec.
This series of interval velocities typifies one of the most serious seismic exploration problems, namely a very hard water bottom, with the attendant reduction in energy penetration and the strong multiple to primary energy ratio. Predictive deconvolution is one of the most effective tools for elimination of such multiples; at approximately 2.0 seconds the multiple has been virtually eliminated by this method.
Note also that the multiple attenuation is not significantly aided by differential normal moveout, since the data was shot with a 12 trace cable (200 ft. group interval, 855 ft. offset), the data is only stacked 6 fold, and normal moveout is small due to the high average velocity of the section.
Other multiples, in particular peg-leg multiples generated at the 14,000-11,000 ft/sec. interface (the reflection at 1.5 seconds, right hand edge of example), are also present. These events are evident on all the velocity analyses and the average velocities suggest that they cannot be primary reflections. These also interfere with legitimate primaries and are extremely difficult to remove from the section. Other methods, to be described in later sections of this paper may be effective.
Because of the rapid decrease in energy penetration, scaling of the final section becomes another matter of concern. Consider, for example, a section with virtually no primary energy. If predictive deconvolution was used to eliminate all bubbles and water bottom multiples, then the resulting section would be very low in amplitude, and only the water bottom event would be visible. Unfortunately, the relative amplitudes are seldom retained after deconvolution and some form of scaling is normally used to obtain a fairly constant RMS amplitude over the entire trace. If primary energy does not appreciably contribute to the RMS calculation, the amplitude of the residual bubble and multiple after decon will be raised to nearly its original level. It then appears as if the deconvolution was ineffective and has in fact made the record noisier. This is far from the case, as seen in the example below (Fig 4). The two sections show a portion of the previous example before and after scaling, scaled in such a way that the water bottom reflection amplitudes are identical in both cases. The decon has obviously attenuated the reverberations, but the data dependent scaling has enhanced the 2nd and 3rd bounce multiples at the expense of the primary reflections. Thus, it is important to appreciate the effects of time variant scaling on the result of a de-reverberated section.
Another approach to the bubble problem is to analyze the signature of the source impulse and its associated reverberatory train. The signature can be found on the near trace prior to the water bottom arrival or on an auxiliary channel especially used to record the signature. In the example shown below the signature trace has been extracted from the near trace and displayed to show the reverberatory content.
Auto correlate the record and pick manually, or automatically, the
bubble pulse periodicities and their amplitudes taking the zero lag
value as unity. This yields a time series of spikes at those times
with those amplitudes.
Using the time series obtained from (a) (with suitably modified
amplitudes if necessary), convolve the wavelet found in (b) with the
time series to obtain a synthetic signature.
(d) Design a deconvolution operator using least squares techniques from this synthetic signature and filter the entire sequence of records.
The approach obviously has to cope with a number of assumptions but can be used successfully in shallow water cases where a direct signature may not be available.
For cases in which interference can be predicted in time and space by its velocity characteristics, a coherent noise attenuation program (CONA) is used, which can reduce the amplitude of water bottom multiples and interbed multiples to approximately the level of the random noise. The results of the process are show below. The first illustration (Fig 7) shows a 12 trace record before and after the CONA process. The compete removal of the water bottom multiple events is evident without any degradation of primary events. No spectrum whitening is involved, so phase and frequency characteristics of the data are unchanged.
A stacked section with and without the CONA process is shown below. The reflections which cross the first water bottom are clearly visible after application of CONA.
The specialized programs described here for marine seismic data processing, and where applicable on land seismic, provide versatile and consistent removal of source bubble, water bottom, single-bounce, and peg-leg multiple interference effects, even under adverse data quality conditions. Most of the deconvolution operators are developed automatically from analysis of the actual data set. The coherent noise attenuation process allows for manual inputs if shallow water depth or other conditions prevent the use of more automatic techniques.
DIGITAL PROCESSING OF SPARKER SEISMIC DATA
Experience with sparker surveys, combined digital corrections and
display of airborne magnetic and gravity data, showed great promise
in the 1970s and 1980s, and may still be useful in remote, hostile,
or ecologically sensitive environments.
· Optical presentation and scale uniformity: The client has their choice of display made and scale. Time scale changes, sometimes present on shipboard monitor sections, are eliminated.
Predictive deconvolution: Water bottom ringing, which makes
interpretation difficult, is attenuated.
Trace compositing: Signal to noise ratio of deep reflections may be
enhanced by compositing adjacent groups of traces.
Time variant filtering: Reflection quality may be enhanced by the
use of digital filters.
· Time variant scaling: Deep data and low amplitude zones may be brought out by the use of data dependent scaling programs.
Page Views ---- Since 15 Apr 2023
Copyright 2023 by Accessible Petrophysics Ltd.
CPH Logo, "CPH", "CPH Gold Member", "CPH Platinum Member", "Crain's Rules", "Meta/Log", "Computer-Ready-Math", "Petro/Fusion Scripts" are Trademarks of the Author