Sei sulla pagina 1di 17

Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/ GEOPHYSICS, VOL.

L. 44, NO. I I (NOVEMBER 1979); P. 1813-1829, 23 FIGS.

The effect of subsurface sampling on one-dimensional


synthetic seismograms

M. Schoenberger* and F. K. Levin*

In generating one-dimensional (I-D) synthetic seismograms from sonic and density logs. the actual sub-
surface is replaced by a sequence of beds so constructed that the time required to traverse each bed is the
same. Since Wuenschel (1960) devised the method, geophysicists have assumed that as long as the beds
were thin, i.e., subsurface time sampling was fine enough to avoid aliasing in the source waveform,
the synthetic seismograms would be independent of the sampling rate. We have found this assumption to
be unjustified. Even when the subsurface is sampled more than twice per period at the maximum source
frequency, the traces generated by a synthetic seismogram program may change as the sampling changes.
For the wells that were analyzed, the reflections were delayed by an increasing amount as the sampling
became finer. The delay appeared to approach a constant value; but for the most cyclic of the logs, even
a OS-msec sample interval apparently was not small enough. However, when instead of computing the
velocity for a sample interval by simply summing transit times we turned to a method suggestedby Sher-
wood (1962), the total reflection time delay became less dependent on the sampling rate. Attenuation over
the seismic band increased as the sampling became finer, regardless of the method used to obtain velocity
from transit times. Although the time delays found in this study are greatly reduced when the method of
computing velocity from sonic log transit times is modified from that of simply summing sample values, it
is not clear that this more complicated procedure is free of problems due to correction of sonic log inte-
grated times to check-shots times. Part of the difficulty in tying check-shot traveltimes to seismic reflection
times may originate in the phenomena discussedhere.

INTRODUCTION
face used to generate a synthetic seismogramconsists
In two previous papers (Schoenberger and Levin, of beds of equal time thickness (Wuenschel, 1960).
1974, 1978), we examined with one-dimensional Seismograms from real subsurfacesconsist of over-
(1-D) synthetic seismogramsthe apparentattenuation lapping primary and multiple reflections from thou-
due to intrabed multiple reflections. Intrabed multi- sands of interfaces, whereas synthetic seismograms
ples-reflections bouncing between nearby interfaces consist of primary and multiple reflections from
-delay and elongate reflections by transferring fewer interfaces at equal time intervals.
energy from the beginning of a seismic pulse to later An important parameter which must be selected to
times. Quantification of this filtering phenomenon genera&ea synthetic seismogram is the time thickness
was the purpose of our previous papers. However, in of each interval. The time thickness also determines
thosepapers, we also noted that 1-D synthetic seismo- the sample rate of the source waveform (c.f., Ap-
gram programs may distort the effect of multiple re- pendix A). Users of synthetic seismogramstypically
flections by requiring them to arrive precisely at time choose a sample interval AT such that the maximum
samples. Unlike a real subsurfacewith beds varying source frequency is less than the Nyquist frequency
in thickness from centimeters to meters, the subsur- (1/2AT). Although this avoids aliasing of the source

Presented at the 48th Annual International SEG Meeting November 1, 1978 in San Francisco. Manuscript received by
the Editor November 22, 1978; revised manuscript received May 3, 1979.
*Exxon Production Research Company, Box 2 189, Houston, TX 7700 1.
0016.8033/79/ 1IOI-1813$03.00. @ 1979 Society of Exploration Geophysicists. All rights reserved
1813
Schoenberger and Levin

time - SECONDS
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

0
0 2000 4000 6000 8000 10,000 12,OW
DEPTH - FEET

FIG. I. Velocity and density logs for well 2 (Schoenbergerand Levin, 1978). The valuesof velocity and density
were obtained by averaging within I-msec equal two-way traveltime intervals.

waveform, it does not necessarily avoid aliasing- between beds in the upper half and those in the lower
type problems due to use of equal time thicknessesin half has been removed.
the derivation of synthetic seismograms. A synthetic Investigating the effect of subsurface sampling
seismogram can be viewed as resulting from a pro- rates on I-D synthetic seismograms is difficult. Ex-
cessfor which a sonic and a density log are the input, cept for subsurfaces too simple to be instructive,
and reflections, both primary and multiple, are the theory is of little help and computer studies, i.e.,
output. If this processwere linear, it would be neces- empirical investigations, are required. With little the-
sary to sample such that the maximum frequency of oretical guidance to indicate that the sampling rate
the earths impulse responsewas less than the Nyquist might be important, geophysicists have been disin-
frequency. In fact, the process of generating a syn- clined to search for sampling phenomena. We were
thetic seismogram is nonlinear. If the input density so disinclined. However, our earlier work had left a
logs are scaled, i.e., each density is multiplied by worrisome problem: for the same subsurfacesampled
the same constant, the output is invariant: the time at 2-msec and 4-msec intervals, attenuation of sig-
scale and reflection coefficients are unchanged. (The nals at frequencies below 60 Hz differed by about
reflection coefficients are independent of velocity 25 percent (Schoenberger and Levin, 1978, Figure
scaling also.) Furthermore, if the log is divided into 14). Since attenuation constants were computed di-
an upper and lower half and a synthetic seismogram rectly from the synthetic seismograms, changing the
is formed by feeding the output signal from the upper sampling interval must have changedthe seismograms
half into the lower half, the resulting seismogram in a nontrivial manner.
differs from that formed from the undivided log. In Ideally, an earth model with unequal bed travel-
splitting the log, the possibility of multiple reflections times could be approximated arbitrarily closely by
Effect of Subsurface Sampling 1815

time - SECONDS
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

DEPTH - FEET

FIG. 2. Velocity and density logs for well 18 (Schoenberger and Levin, 1978). The values of velocity and
density were obtained by averaging within 1-msec equal two-way traveltime intervals.

forming a finely spaced equal-traveltime model. By EXPERIMENTALPROCEDURE


sampling the logs at successively smaller intervals, Figures 1 and 2 are the sonic and density logs for the
we would hope that at some sample interval AT, the wells used in this study. To conform with our pre-
filtering effects would stabilize; and that for sample vious notation (Schoenberger and Levin, 1978), we
intervals less than AT,, the synthetic seismogram have numbered the wells 2 and 18. The logs of well 2
would be invariant. This stabilized seismogram (Figure 1) show large variations over small depth
should be the correct 1-D synthetic seismogram- intervals, while the logs of well 18 (Figure 2) are
the seismogram we would get if we could form one smoother. Attenuation constantspreviously reported
for the continuous, unsampled subsurface. To de- for a 2-msec sample interval were 0.41 dB/Hz (0.21
termine the extent to which stabilization could be dB/wavelength) for well 2 and 0.15 dB/Hz (0.06
achieved with practical sampling intervals, we re- dB/wavelength) for well 18.
turned to three wells that are the typical examples in In our earlier work, we computed the earth trans-
Schoenberger and Levi, (1978). We selected from mission filter by inserting an isolated reflector below
the three wells the two with significant transmission the well and then generating a synthetic reflection
filtering and sampled the sonic and density logs at seismogram. For the sample intervals in the work dis-
intervals of OS-, l-, 2-, and 4-msec two-way travel- cussed here, we were unable to use this technique;
times. We emphasize these were two wells with our synthetic seismogram program allows too few
significant transmission filtering; we believe that multiple reflections to let us apply our original method
for many of the other wells in our previous paper the with 0.5msec and l.O-msec sample intervals. In-
phenomena we discuss would be insignificant. stead, we computed the complete down-traveling
1816 Schoenberger and Levin

Hms

/
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

0 lms

o-

+ io 40 00 00 D
time - MILLISECONDS time - MILLISECONDS

WC. 3. Two-way earth transmissionfilter impulse re- FIG. 4. Two-way earth transmission filter impulse
sponsesfor well 2, sampled at 4.0-, 2.0-, l.O-, and responsesfor well 18, sampled at 4.0.) 2.0-, 1.O-, and
0.5msec equal two-way traveltimes. Zero time corre- 0.5-msec equal two-way traveltimes. Zero time corre-
spondsto arrival of the first energy, and each impulse spondsto arrival of the first energy. and each impulse
responsehas been normalized to have the same peak responsehas been normalized to have the same peak
amplitude. amplitude.

wavetrain at a point below the well. This wavetrain are the impulse responsesof the two-way earth trans-
convolved with itself is equivalent, to within a scale mission filters. In each case, the time delay of the
factor, to the reflection from an isolated, deep inter- wavetrain increases as the sampling becomes finer;
face. Our input was a spike at the surface. in addition, the shape of the impulse response varies
slightly as the sample interval changes. Note that the
RESULTS responsesfor well 18 develop a high-frequency oscil-
The synthetic seismograms resulting from these lation for sampling intervals of I .O msec and 0.5
simulations are displayed in Figures 3 and 4 for wells msec. We believe this behavior is a consequenceof
2 and 18, respectively. Since a spike was input, these the relatively smooth impedance log for well 18 com-

SOURCE
WAVEFORM AMPLITUDE SPECTRUM

2b 40
time - MILLISECONDS FREQUENCY - Hz

FIG. 5. The waveform and amplitude spectrum of a source to be used as an input to the earth models charac-
terized by the logs of Figures 1 and 2.
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/ Effect of Subsurface Sampling 1817

I I
0 20 40 60 60
time - MkUSECONDS

FIG. 6. Transmission seismograms for well 2 for FIG. 7. Transmission seismograms for well 18 for
sample intervals of 4.0-, 2.0-, 1.O-, and 0.5-msec sample intervals of 4.0-, 2.0-, l.O-, and 0.5.msec
two-way traveltimes. Inputs were appropriately two-way traveltimes. Inputs were appropriately
sampled versions of the waveform of Figure 5. Each sampled versions of the waveform of Figure 5. Each
seismogram has been normalized to have the same seismogram has been normalized to have the same
peak value. peak value.

24..

RUST TROUGH

0;s 1.0 2.0 4.0


0- SAMPLE
0.5 1.0
INTERVAL
2.0 4.0
- MILLISECONDS
SAMPLE INTERVAL - MILLISECONDS

FIG. 8b. time delays of synthetic transmission seis-


FIG. 8a. Variation in time delay of synthetic trans- mograms (displayed in Figures 6 and 7) as a function
mission seismogramsfor well 2, using as criteria the of sample interval. The time delays are measured by
delay of the first peak, the first zero crossing,the first the delay of the first trough of the seismogramrelative
trough, and the second zero crossing. to that of the source waveform.
1818 Schoenberger and Levin

30

45
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

WELL 2
2 A8

B
35 --\
1
0
time -
T T+12
MILLISECONDS s 30
3
FIG. 9. Window function by which the transmission
seismogram impulse responsesof Figures 3 and 4 are 5 25

x
multiplied prior to spectral analysis.

20

pared to that of well 2, and we will discuss the sub-


ject in greater detail later in the paper and in Ap- .15
0.5 1.0 2.0 4.0
pendix A. SAMPLE INTERVAL - MILLISECONDS
Since we are concerned with the effects of trans-
mission filtering at frequencies in the seismic band, FIG. 10. Attenuation coefficients (dB/Hz) versus
sample interval for wells 2 and 18.
phenomena such as the high-frequency oscillation
of the seismograms from well 18 (Figure 4) are ir-
relevant. To examine filtering effects over the seis-
mic band, we generated synthetic seismogramsusing
as input appropriately sampled versions of the
wavelet displayed in Figure 5. This pulse is a For both wells 2 and 18, the time delay of the trans-
minimum-phase version of a 35Hz Ricker wavelet mitted waveforms first trough is plotted versus sam-
with half-power points at 20 and 50 Hz. The two- ple interval in Figure 8b. As the sampling becomes
way transmission seismograms for wells 2 and 18, finer, more transmission loss occurs; more multiples
shown in Figures 6 and 7, respectively, were ob- are generated; and more energy is shifted to later
tained by convolving the input wavelet of Figure 5 times. However, our motivation for sampling with
with impulse responsesof Figures 3 and 4. smaller intervals was to find a value below which
As is obvious from Figures 6 and 7, the transmitted the results were independentof sampling interval. By
pulses are delayed relative to the direct arrival. To this measure, the seismogram from well 18 appears
determine the magnitude of these delays, we com- to be stabilizing at a 9.0-msec delay for a sample
pared each transmitted pulse to the source pulse interval of 1.0 msec and at a lO.O-msec delay for a
used to generate it. Since pulse shaping occurred as sample interval of 0.5 msec. On the other hand, the
well as time delay, there was some arbitrariness in- seismogram from well 2 still exhibits increasing
volved in measuring time delay. In Figure 8a, the delay with decreasing sample interval even at 0.5
delays of characteristic parts of the waveform are msec, the smallest interval used. Overall, changing
plotted versussample interval. Note that later parts of the sample interval from 4 to 0.5 msec changes the
the waveforms are delayed more than the earlier time delay on the well 2 seismogram from 8 to 24.5
parts, or, alternatively, transmissions through the msec.
synthetic earth model elongates the waveform. Of If our synthetic seismogram program had been
more importance is the similarity of the various more flexible, we would have sampled the subsur-
curves. The similarity permits us to choose as our face more finely than at a 0.5-msec two-way travel-
time delay measure the time shift of the first trough; time interval. Presumably, the time delay of the trans-
this was chosen, since it is the pulses most promi- mitted wavelet associated with well 2 would even-
nent feature. tually have saturated at a value greater than 24.5
Effect of Subsurface Sampling 1919

msec for a sample interval less than 0.5 msec. Figure displayed in Figure 9. In Figure 9, T is 40 msec for
8 implies that a synthetic seismogramgeneratedfrom a wavetrain sampled at a 4-msec interval; the window
logs sampled at a 4-msec two-way traveltime inter- partially eliminates multiple reflections delayed by
val would yield time delays significantly smaller than 40 to 52 msec and completely eliminates multiple
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the correct delays. An even poorer seismogramwould reflections arriving after 52 msec. For smaller sample
result from a primary-reflection-only model. Since intervals, T was increased by the incremental delays
primary-reflection-only seismograms produce no obtained from Figure 8b in order that, to the eye at
time delay, the correct time delay would be under- least, the windowed wavetrains for different sample
estimated by more than 24.5 msec. In addition, time intervals differed mainly by their delay relative to one
delays become progressively larger as more earth sec- another. Smaller sample intervals corresponded to
tion is encountered. Often, tying seismic data to wells larger time delays and, therefore, required longer
with synthetic seismograms is difficult, usually re- windows.
quiring time-variable time shifts to align the two types The amplitude spectrum for each of these win-
of data. Perhaps underestimation of time delays due dowed wavetrains was calculated and the cumulative
to intrabed multiple reflections has been a factor in attenuation was obtained from the slope of a least-
this problem. squares line fitted to values of the log amplitude
In addition to creating a time delay, redistribution spectrum between 0 and 65 Hz. This procedure as-
of energy by the generation of multiple reflections sumes the attenuation is of the form Kernorwhere f is
converts a pulse reflected from a deep, isolated inter- frequency, a is an attenuation coefficient, and K is a
face into a reflection wavetrain. In fact, the change in constant. The values of the attenuation coefficient a,
apparent attenuation with changing sample interval expressed in dB/Hz, are displayed as a function of
caused by this mechanism motivated our investiga- sample interval in Figure 10. As before, the attenua-
tion. Splitting a reflection wavetrain into intrabed tion produced by well 2 logs far exceeds that of well
and interbed reflections is an arbitrary procedure, and 18 logs. Furthermore, the attenuation coefficient at
the time following the arrival of a primary reflection
at which the split is made varies from one investiga-
tion to another. In our previous papers (Schoenberger
and Levitt, 1974, 1978), we classified as intrabed
55
reflections the first 40 msec of the wavetrains. We
maintain that classificationhere. However, the change
in time delay with changing sample interval apparent 50
on Figures 6 and 7 forced us to separateshaping- and
time-delay effects of multiple reflections somewhat
.45
differently. We multiplied the wavetrains (transmis-
sion seismograms) by the time-window function P
p 40
I

E 38

s I WELL 2

[ 30

a I
I
25
.lG-
I
E
L -19 .mI-.
2 - - - SMOOTHEOWGS
-ORIGINAL WGS
a-
WELL 18
.15,-
0.5 1.0 2.0 4.0
2% SAMPLE INTERVAL - MILLISECONDS
o.ms O.GlG GmG 0.Y 0.06 0.10 G.M
GPKnAL FRE(1ENCY - Fr -
FIG. 12. Attenuation coefficients (dB/Hz) versus
FIG. 11. The amplitude responseof a zero-phase filter sample interval for wells 2 and 18, from both
used to smooth the well logs of Figures 1 and 2. smoothed (filtered) and original logs.
1820 Schoenberger and Levin

28 each well has nearly stabilized at the l.O-msec sam-


ple interval.
24
Using attenuationcoefficient and time delay as cri-
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

teria for characterizingtransmissionfiltering, we con-


8
clude that sampling the sonic and density logs of well
gm
18 at a 0.5msec interval was adequate to ensure an
0 accurate synthetic seismogram. For well 2, at a 0.5
I
msec sample interval, the attenuation coefficient had
316
I stabilized, but the time delay had not. Hence, at well
I 2, the synthetic seismogramgeneratedfrom the small-
l 12 est sample interval available (0.5 msec) is of ques-
9
x
tionable validity.

c5 LOG PREPARATION FOR MODELING


With a sufficiently small sample interval, present
4 methods of synthetic seismogram generation seem
----sMmwwGs i able to produce stable transmission filters. We pre-
-0NGNAl WCS
viously observed (Schoenberger and Levitt, 1974,
0
0.5 1.0 20 4.0 1978) that large values of attenuation and large time
SAMPLE INTERVAL - MILLISECONDS delays result from large and rapid variation in acous-
tic impedance. Based upon the two wells analyzed
FIG. 13. time delays (measured at the first trough)
of transmissionseismogramsfor wells 2 and 18, from here, this type of variation also appears to require
both smoothed (filtered) and original logs. finer sampling to achieve stable transmission filter-
ing. It seemsreasonableto interpret this phenomenon
as a type of aliasing; i.e., the high-frequency compo-
nents of the acoustic impedance log appear to be low
frequencies when the sampling is too coarse. Such a
philosophy suggests smoothing the logs (antialias

0 20
time -
40
MILLISECONDS
a
0
II
I
0
I
lo
,
4ms

40 (10
time- MILLISECONDS
A

FIG. 14. Transmissionfilter impulse responsesfor the FtG. 15. Transmissionfilter impulse responsesfor the
smoothed logs of well 2. Each response has been smoothed logs of well 18. Each response has been
normalized to have the same peak amplitude. Zero normalized to have the same peak amplitude. Zero
time correspondsto arrival of the first energy. time correspondsto arrival of the first energy.
Effect of Subsurface Sampling
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

i 2il io oil do
llME - MILUSECONOS

FIG. 16. Transmission seismograms at well 2 for FIG. 17. Transmission seismograms at well 18 for
smoothed logs. Inputs were appropriately sampled smoothed logs. Inputs were appropriately sampled
versions of the waveform of Figure 5. Each seismo- versions of the waveform of Figure 5. Each seis-
gram has been normalized to have the same peak mogram has been normalized to have the same peak
value. value.

filtering) prior to computing synthetic seismograms. both 0.5- and l.O-msec sampling. Based upon ob-
We would hope, by using such a procedure, to attain servations of attenuation and time delays, we must
the same stabilized transmission filter as with the conclude that smoothing the logs is an undesirable
original log, but with a coarser sampling. practice.
These conjecturesseem reasonablebut are not cor- We became painfully aware that smoothing the
rect. The logs were smoothed by applying a zero- logs can have unanticipated effects on synthetic
phasespatialfilter with the amplitude spectrumshown seismograms when we plotted the transmitted wave-
in Figure 11. Note that this is a very gentle low-pass trains resulting from a spike input (Figures 14 and
filter with the half-power point (-3 dB) at O.O4/ft, 15). Oscillations that are noticeable but small for the
correspondingto a frequency greater than 200 Hz for unsmoothed logs grew when the logs were filtered
a velocity greater than 5000 ft/sec. (The sonic and until, for a 0.5-msec sample interval, the underlying
density logs had been digitized every foot.) In Figure behavior was hidden. Again convolution with the
12, we have superimposedon the plots of Figure 10 35-Hz wavelet (Figure 5) removed the oscillation
the attenuation coefficient values from synthetic (Figures 16 and 17).
seismogramscomputedfrom filtered (smoothed) logs.
Filtering of the logs has little effect for well 18. How- A DIFFERENT METHOD OF COMPUTING
VELOCITIES FOR EQUAL-TIME INTERVALS
ever, for well 2-the well with large and rapid varia-
tions in transit times and density-values did not In all of our simulations, we used a method for
stabilize even for a OS-msec sample interval. Since obtaining an equal-time interval subsurface model
the attenuation coefficient from the original well 2 from sonic log transit times that preserves the time
logs stabilized at 0.5 msec, smoothing the logs had of the first arrival across layers; i.e., we simply
a deleterious effect. The resultant time delays (Figure summed the transit times until the sample interval
13) are more predictable. For both wells, time delays selected had been attained and computed velocity
are more stable for the smoothed than for the original by dividing the depth interval needed to attain that
logs, but the values are significantly smaller. At a time interval by the time interval (time-average
0.5-msec sample interval, smoothing decreased the method). An alternative, more complicated proce-
time delay by 1.5 msec at well 18 and by 4.5 msec at dure that accountsfor the multiple reflections within
well 2. Assuming that the correct time delay is larger the time interval has been given by Sherwood (1962)
than that for the original logs at a 0.5-msec sample and dErceville and Kunetz (1963); it is equivalent
interval, smoothing the logs increased the errors at to the long-wavelength approximation used to obtain
1822 Schoenberger and Levin

elastic constants for anisotropic solids (Backus,


1962). The method, which we shall call the stress-
strain method, is described in detail in Appendix B.
With this method, we generated another set of syn-
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

thetic seismogramsfor wells 2 and 18 with the four


sample intervals. time delays computed from these
seismograms are plotted, along with those from the
time average seismograms, in Figure 18. The time
delays obtained from the two techniquesappear to be
comparable for these two wells. However, it should
be noted that these time delays are measured relative
to the direct arrival (or first arrival of energy).
Sherwood (1962) and dErceville and Kunetz
(1963) observed that the traveltime across a layer
obtained from this stress-strain technique is greater -THE AVERAGE METHOO 18

than the computed traveltime (summation of sonic O-


0.5 1.0 2.0 4.0
log sampled times). Sherwood noted that the extra
SAMPLE INTERVAL - MILLISECONDS
traveltime is the delay we would find if we could
compute the effect of energy multiply reflected among FIG. 18. time delay of first trough as a function of
the interfaces of all the beds being combined into one sample interval for wells 2 and 18. The upper curve
equal-time sample interval. For example, if we chose in each set was computedfrom synthetic seismograms
generated with the stress-strainmethod of Sherwood
to sample at a 2-msec interval (two-way time) and the (1962) and dErceville and Kunetz (1963); the lower
interval velocities of one-ft thick beds for a region of curves were taken from Figure 8b.
the subsurface were about 8000 ft/sec, a 2-msec
sample interval would include eight beds. Among the
nine interfaces, numerous intrabed multiple reflec-
tions would be possible. The time-average method
ignores the time delay due to these invisible multiple
32 ,
reflections; the stress-strain technique accounts for
them. As the size of the sample interval is reduced,
the number of beds combined into one interval de-
creasesand, hence, the possible number of reflections
from concealed interfaces also decreases. Under the
same conditions, the total number of intervals of the
equal-time sampled subsurface increases; conse-
quently, the number of reflections among the inter-
faces also increases. It is not obvious that reduced
WELL 2
time delays due to fewer reflections within each inter-
val and the increasedtime delays resulting from more
reflections between a greater number of intervals
should compensate for one another but, while not
complete, the compensation is considerable. The
increasedtraveltime of the first arrival due to multiple WELL 18
reflections within each layer was computed for each
4-- - - - STRESS-STRAIN MET00
of the stress-strain seismograms. These times were - time AVERAGE HEW00
added to the delay shown in Figure 18 due to multi-
o- :
ples among the layers. The resultant total time delays 0.5 1.0 2.0 4.0
are displayed in Figure 19. Since the effects of multi- SAMPLE INTERVAL - MILLISECONDS
ple reflections among the interfaces and within indi-
FIG. 19. Total time delay of first trough as a function
vidual equal-time beds have opposite trends, they of sample interval for wells 2 and 18. The data from
tend to compensatefor one another, and the variation Figure 18 were added to the delay of the first arrival
in total time delay with sample interval, for the stress- to obtain the total delay.
Effect of Subsurface Sampling 1823

time delay on sample interval, it had little effect on


the dependence of attenuation on sample interval
(Figure 20). Regardless of how we handled transit
times, the attenuation increased as sample interval
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

decreased. It may be noted that the attenuation for a


1-msec sample interval exceeds that for the 0.5-msec
interval for the stress-straintechnique applied to well
2. However, also included in Figure 20 are error
bars which indicate the deviation from linearity of
the log amplitude spectra. These are intended to be
WELL 2
50 percent confidence intervals basedupon a Gaussian
distribution of values about a straight line (Bennett
and Franklin, 1954). Certainly, even within this large
confidence interval, the values for the I-msec and
0.5-msec intervals are not significantly different.
Furthermore, the 40-msec window chosen to separate
intrabed and interbed multiple reflections was arbi-
trary. Increasing this time by 4 msec has the effect of
making the attenuation values for all four cases de-
picted in Figure 20 monotonically increase with de-
creasing sample interval. One final point should be
made concerning the effect of windowing the wave-
train prior to attenuation estimation. Increasing the
0. 5 1.0 2.0 4.0
SAMPLE INTERVAL - MILLISECONDS window length by 4 msec, in addition to making the
curves of Figure 20 monotonic, increasesattenuation
FIG. 20. Attenuation coefficient as a function of in well 2 for the OS-msec sample interval by I3 per-
sample interval for wells 2 and 18. The coefficients
cent for the time-average method and by 8.3 percent
from synthetic seismogramsgeneratedwith the stress-
strain method of Sherwood (1962) and of dErceville for the stress-strainmethod. There was a lesser effect
and Kunetz (1963) are plotted along with those from for the 1.O-msec interval, and in none of these cases
Figure 10 (time-average method). were attenuation values for well 18 affected.
In view of the decreased sensitivity of time delay
for the stress-strain method, the reader reasonably
might ask why this method is not used universally in
strain method, is greatly reduced. For both wells, the the computation of I-D synthetic seismograms. In
delay still increasesas the sample interval decreases spite of its obvious attractiveness, the method has
from 4 to 0.5 msec. However, for well 2, the increase associateddifficulties that are discussed later.
of 6 msec over that range contrasts with the more-
than- 18-msec increase found for the time-average DISCUSSION
method; for well 18, the stress-strainmethod reduces Because of the complex nature of the phenomena
the change in delay from 9.5 msec (time-average we are considering, we are unable to offer quantitative
method) to 5.5 msec. As the sample intervals de- explanations for the behavior illustrated by Figures
crease to zero, we pass to the continuous case, and 3-17. Our remarks in this section of the paper are
the two methods for finding velocity become equiva- qualitative. That attenuation and time delays in the
lent. Hence, for a given well, the curves of Figure 19 transmission seismogram should depend on the size
must coincide for sufficiently fine sampling. It is and rapidity of changes in impedance is certainly
interesting that although for both wells the change reasonable, since intrabed multiple reflections depend
of delay with sample interval seems to be approach- on those changes (ODoherty and Anstey, 1971;
ing a constantvalue, even for a 0.5-msec interval the Schoenberger and Levin, 1974). Sampling logs at
delay for the time-average method is less than the ever smaller intervals should eventually result in a
delay for the stress-strainmethod. seismogram that is both stable and identical to the
Although use of the alternative stress-strainmethod seismogram we would obtain from the continuous
of finding velocity greatly reduced the dependenceof subsurface-if we knew how to form a synthetic
1824 Schoenberger end Levin

coefficient is the one of our previous work (Schoen-


berger and Levin, 1974): the transmission coef-
ficient is

fi (1 - R;),
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

i=l

where R, , Rz, , are the reflection coefficients


above a deep, isolated reflecting surface. Note that
this is a two-way transmissioncoefficient, i.e., each
interface of the subsurface is traversed once from
above and once from below. By using a two-way
transmission coefficient, we avoid the necessity for
normalization by the quotient of acousticimpedances
of the initial and final media. For the unfiltered (orig-
inal) logs, the transmission coefficient plotted in
Figure 21 decreases rapidly with decreasing sample
10-@(
,., 1 I I interval. For sample intervals that are fine compared
0.5 1.0 2.0 4.0
to the intervals over which subsurface velocity and
SAMPLE INTERVAL - MILLISECONDS
density change, reflection coefficients should de-
FIG. 2 I. Transmissioncoefficients associatedwith the crease with decreasing sample interval, and the
acousticimpedancemodels for wells 2 and 18. Values transmission coefficient should either not decrease
were computed for sample intervals of 0.5, 1.O, 2.0, or increase. In the limit of continuoussampling, the
and 4.0 msec, for both the original and smoothedlogs.
transmission coefficient would be unity (Foster,
1975).
If the reflection coefficients do not change in mag-
nitude but simply double in number as the sample
seismogram for such a subsurface. Sampling at a de- interval is halved, the transmissioncoefficient should
creased interval increases the number of reflection square. In Figure 21, the transmission coefficient
coefficients but, provided the log values vary at a rate does follow a square-law curve for the unfiltered
slow compared with the sampling rate, decreasesthe logs for sample intervals between I .O and 4.0 msec.
magnitudeof individual coefficients. We can say little We conclude that, on the average, reflection coeffi-
more. As we have seen, there does seem to be a limit- cient magnitudesdid not decrease until the 0.5msec
ing sampling rate for both attenuationand time delay. sample interval was reached. This observation co-
The behavior of synthetic seismogramsfrom con- incides with our earlier conclusions concerning the
tinuous subsurfaces was explored in an important increase in both attenuation and time delay as the
paper by Foster (1975). Foster proved that, for a con- sample interval decreasedfrom 4.0 to 1.O msec. Note
tinuous 1-D subsurface, two-way transmission ef- also that smoothing the logs increased the transmis-
fects of the type we are considering vanish. Since in sion coefficient and reduced its variation with sam-
Fosters formulation properties of the subsurfaceand ple interval. The reflection coefficients for smoothed
derivatives of those properties are continuous, a dis- well 2 logs decreased significantly when the sample
crete approximation to the subsurfaces would nec- interval changed from 1.0 to 0.5 msec; however,
essarily require sample intervals that were small under the same condition, the attenuation (Figure
compared to any interval over which subsurfaceprop- 12) increased. This behavior illustratesthe point that
erties changed. It is because our sample intervals multiple reflections are functions of both magni-
were not small in this sense that we encounteredthe tude and spacing of reflection coefficients. Although
phenomena discussedhere. smoothing the logs did reduce the extremely short-
If the logs were sampled sufficiently finely, we period log variations, it had a different (and unclear)
should eventuall_y approach the sondnuous case. effect on the longer period variations (Figures I6
That our logs did not vary slowly compared with con- and 17) which contributed the multiple reflections
ventional sample rates can be verified by examining being analyzed here.
the transmission coefficient as a function of sample Another way of looking at the comparison be-
interval (Figure 21). Our definition of transmission tween sampling and log variation is by way of a linear
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/ Effect of Subsurface Sampling 1825

0
0 250 500 750 1000 0250500 750 1000
FREQUENCY - Hz FREQUENCY - Hz

FIG. 22. Amplitude spectra of the 0.5msec sampled reflection coefficient logs associatedwith smoothed and
unsmoothed(original) logs from wells 2 and 18. Each spectrum has been smoothed with a rectangular 50-Hz
wide window.

system analogy; i.e., we can speak of spectra and ing (see Appendix A). We resorted to numerical ex-
a Nyquist sampling frequency. Since, as mentioned periments because we saw no theoretical way to
earlier, our system is nonlinear, this is not rigorously explain what we had found previously (Schoenberger
valid; but it does provide some insight. We said the and Levin, 1978). We still have no theory to guide
reflection coefficients decreasewith finer sampling if us. Whatever the source of the oscillations, they did
the log values vary at a rate slow compared with that not interfere with transmissionof lower frequencies,
sample interval. For wells 2 and 18, the logs values for convolution of the spike response with a 35-Hz
did not vary slowly. Figure 22 is a plot of the ampli- wavelet removed them.
tude spectra of the reflection coefficient sequences As was pointed out to us by one of the referees of
for the smoothed and unsmoothed (original) logs of this paper, a subsurface whose acoustic impedance
wells 2 and 18 sampled at OS-msec two-way time varies slowly with depth (or traveltime) transmits
intervals. Even at the Nyquist frequency of 1000Hz, high frequencies efficiently (ODoherty and Anstey,
the spectra of the unsmoothed logs have appreciable 1971). Hence, we might have expected synthetic
amplitude. We needed finer sampling to avoid alias- seismograms generated from smoothed logs to have
ing. As a practical matter, we could not sample at a more apparent high frequencies than those generated
faster rate, for the logs were digitized every foot; from unsmoothed logs. Even with that insight, it is
finer time sampling would have required interpolation not clear why high frequencies should manifest them-
procedures we did not have, nor, considering the selves, as shown in Figures 14 and 15, as persistent
resolving power of logging sondes, could justify. sinusoids or why the sinusoids should become
Filtering the logs reduced the high-frequency content stronger as the sample interval decreased.
relative to the lower-frequency content but did not The work we have described was based on the
result in negligible values at the Nyquist frequency. generation of 1-D synthetic seismograms from sub-
Oscillations corresponding to the sample interval surfaces made up of beds all with the same time thick-
are obvious in Figures 14 and 15. This phenomenon ness. As long as we are tied to computers, there is no
had been observed whenever the sample interval was other obvious, practical method of including the intra-
small relative to the log variations. Some insight can bed multiples responsiblefor the phenomenawe have
be gained into the causes of the oscillations by ex- discussed. Although it is easy to show equal-time
amining a very slowly varying subsurface-one with sampling gives seismograms that converge to those
acoustic impedance increasing(or decreasing) mono- from a continuous subsurface for simple systems
tonically with depth. In this case, first-order intrabed such as linearly increasing impedance over a depth
multiples are opposite in sign from primary reflec- interval many wavelengths long (Berryman et al,
tions; however, equations describing higher-order 1958), practical convergence to the correct solution
intrabed multiples are too complicated to be illuminat- is not so clear for realistic, rapidly changing systems.
1626 Schoenberger and Levin

In view of Fosters (1975) work, we must conclude with a geophone in the well. If the first arrival time
that if subsurfaceproperties really were continuous, of the check-shot could be picked accurately, this
synthetic seismograms from continuous subsurfaces procedure would be consistent with the time-average
and from sampled subsurfaces would be identical. method used to obtain the velocities for our model.
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

However, real subsurfacesinclude numerous beds of However, a check-shot time really comes from a pulse
many thicknessesthat are separatedby distinct inter- that includes multiple reflections and, hence, is de-
faces. Neither the continuity demanded by Fosters scribed by the stress-strainmethod of Sherwood and
theory nor the requirement of several samples within dErceville and Kunetz. If check-shot corrected sonic
each constant-property zone may be realizable in logs (which already account for multiples) were to be
practice. A completely independentmethod of manu- used in the stress-strainvelocity calculations, the de-
facturing synthetic seismograms would be useful lay due to multiples would be introduced a second
for establishing the validity of the equal-time pro- time The resultant synthetic seismograms would
cedures. Analog models could furnish the needed have reflections arrivingtoo late. It would not be im-
data (Levin, 1953; Sherwood, 1962; Woods, 1975). possible to change sonic-log transit times in such a
way that would be consistent with check-shot
times, but as far as we know, no one does. Further-
CONCLUSIONS more, prorating a correction among the individual
It is not obvious how the sonic and density logs transit times would be difficult. Thus, the procedure
from a well should be sampled in order that a syn- presently used to correct integrated transit time is in-
thetic seismogram generated from the logs matches consistent and needs examination.
a seismogram recorded on the surface at the well. An intuitive notion for decreasing sensitivity to
dErceville and Kunetz (1963) suggestedthe logs be sample interval involves smoothing the logs to pro-
blocked as a function of the length of the signal rather vide antialias filtering. For reasonsdiscussedearlier,
than as a function of the velocity (and density) varia- smoothing increased the high-frequency content of
tion of the logs. However, our investigation indicates the seismograms and resulted in poorer synthetic
that for logs sampled to provide an equal-traveltime seismograms than those from the unsmoothed log.
model, the usual criterion of avoiding aliasing of the time delays seemed to stabilize at larger sample inter-
source waveform is inappropriate, Intuitively we feel vals for a smoothedlog, but stabilization did not occur
the finest sampling that is practical provides the most at the same values as for the unsmoothed logs; in
accurate results; Fosters (1975) work seems to lead addition, attenuation stabilized at larger sample
to the same conclusion. For one of the wells we intervals for unsmoothed logs. Nonlinear problems
analyzed, neither attenuation nor time delay changed can causestrangephenomena. Indeed, low-pass filter-
significantly when we decreased the sample interval ing of the well logs resulted in increased high-
from 1.O to 0.5 msec. For that well, we conclude 1.O frequency content of the transmission response.
msec equal-traveltime (two-way) layers are suffi- Our analysis was confined to the first 40 msec of a
ciently fine. For the other well, time delay increased wavetrain that persistedfor much longer. The 40 msec
when the sample interval was changed from 1.0 to examined represented our arbitrary distinction be-
0.5 msec. We conclude that a 1.0 msec interval is tween what we chose to view as intrabed multiple
inadequate. For both of these wells, the procedure reflections and what we assignedto interbed multiple
of Sherwood (1962) and of dErceville and Kunetz reflections. It is possible that if we had included as
(1963) would have allowed coarser sampling for intrabed multiple reflections more of the wavetrain,
equivalent timing errors; unfortunately, with the the results we report would need minor modification.
coarser sampling, attenuation effects would still be What is the significance of our investigation? For
underestimated. geophysicistsusing 1-D synthetic seismograms, there
Since we know the more complex sampling is a clear warning: what you compute can depend on
scheme of Sherwood and of dErceville and Kunetz the interval at which you sample the sonic and density
reduces sensitivity to the size of the sample interval, logs, and, if your logs vary rapidly over short inter-
why dont we proposethat scheme be used routinely? vals, will depend on the sample interval. For one of
The problem is the following: sonic-log transit times the wells we analyzed, changing the sample interval
as recorded in the field are not absolute. Sonic logs from 4 to 0.5 msec increased attenuation by 35 per-
are corrected by forcing the integrated sonic-log cent and the time delay by 16.5 msec. Furthermore,
transit times to match traveltimes from a source at with the primary-reflections-only synthetic seismo-
or near the surface, i.e., check-shot times, as recorded grams which are commonly used, there is neither
Effect of Subsurface Sampling 1827

attenuationnor time delay. These seismogramsdiffer Benyman, L. H., Goupillaud,P. L., and Waters,K. H.,
appreciably from finely sampled complete seismo- 1958, Reflectionsfrom multiple transition layers, I:
Geophysics, v. 23, p. 223-243.
grams. dErceville, I., and Kunetz, G., 1%3, Sur linfluence dun
The contribution of this paper may not be the trans- empilement de couches minces en sismique: Geophys.
Prosp., v. 11, p. 115-121.
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

mission phenomenaper se; we should not be surprised Foster, M., 1975, Transmission effects in the continuous
if we have uncovered computationally interesting, one-dimensional seismic model: Geophys. J. Roy. Astr.
practically unimportant behavior. Of significance Sot., v. 42, p. 519-527.
Levin, F. K., 1953, An analogue computer for studying
may be our questioning of the validity of 1-D seismo- seismic reflection complexes: Proc., Geophys. Sot. of
grams of the type introduced by Wuenschel (1958, Tulsa, v. 1, p. 70-71.
ODoherty, R. F., and Anstey, N. A., 1971, Reflections on
1960) and Seriff (1958) more than twenty years ago.
amplitudes: Geophys. Prosp., v. 19, p. 430-458.
They may not be as good a representation of field Schoenberger, M., and Levin, F. K., 1974, Apparent at-
data as we would like. The computations assume a tenuation due to in&abed multiples: Geophysics, v. 39,
p. 278-291. U
plane wave normally incident on plane, horizontal - 1978, Apparent attenuation due to intrabed multi-
interfaces separating beds of nonattenuating mate- ples, II: Geophysics, v. 43, p. 730-737.
rials. None of these assumptionsis realized in prac- Seriff, A. J., 1958, Multiple reflections in many-layered
systems: Geophysics, v. 23, p. 1073.
tice. We have no feeling for how deviations from the Sherwood, J. W. C., 1%2, The Seismoline, an analog
assumed system affect phenomena we have illus- computer of theoretical seismograms:Geophysics, v. 27,
trated p. 19-34.
Woods, J. P., 1975, A seismic model using sound waves
in air: Geophysics, v. 40, p. 593-607.
REFERENCES Wuenschel, P. C., 1958, Seismogram synthesiswith multi-
Backus,G. E., 1%2, Long-wave elastic anisotropy:J. ples and transmission coefficients: Geophysics, v. 23,
Geoph.Res., v. 67, p. 4427-4440. p. 1073.
Bennett,C. A., andFranklin,N. L., 1954,Statisticalanaly- - 1960, Seismogram synthesis including multiples
sis in chemistry and the chemical industry: New York, and transmission coefficients: Geophysics, v. 25, p.
JohnWiley and Sons,p. 228. 106-129.

APPENDIX A
COMPUTATION OF SYNTHETIC SEISMOGRAMS

The major advantageof Wuenschels (1958, 1960)


technique for computing 1-D synthetic seismograms nMo= -Pm2 R, Ri-1. (A- 2)
was equal-time sampling of logs measured in depth, i=l

a procedure that resulted in multiple reflections being 2nd-order intrabed response:


automatically included in the output. (The same tech-
nique was developed about the same time by Seriff. nM = P, (nM, + nMz),
Seriffs results are available only as an abstract of a
talk presented at the 1958 Annual International nM, = 2 RiRf_l$ RjRj-1, (A- 3)
SEG meeting.) To see just how remarkable an inno- i=l j=l

vation equal-time sampling was, let us look at ex- and


pressionsfor a few intrabed multiple-reflection com-
putations. To keep the expressionsas simple as pos- nM, = -2 RIRi-, (1 - RF-,).
sible, we will not consider the intrabed multiple 1=2
contributionsto a reflection but the multiple-reflection
contributionsto the impulse responseat the bottom of 3rd-order intrabed response:
an n-layered system. nM = P, (nM, + nMz + nM3),
Primary response:
nM, = - 2 RiRi_, 2 RJR*_, * (A-4)
i=l J=l
P, = fi (1 - Ri). (A- 1)
i=O

lst-order intrabed response:


2 RkRk-1,
k=j
1828 Schoenberger and Levin

GENERATION
OFSYNTHETIC
SEISMOGRAM
nM, = 2 R,R,_, (1 - RF-,) .
i=2

$ RkRk-I + 2 R,R,-,
1 ,
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

k=i-1 k=l
SAMPLE DEPTH-TIME
CONVERSION
INTERVAL
AT
A

nM, = -f: RiRi_,(l - Rf-.J(l - Rf_!_,). I Zi VSi AT


i=3
REFLECTION
COEFFICIENT
The steps involved in the construction of a syn-
thetic seismogramare illustrated by the flow chart of
Figure A- 1. The startingpoint or input to this system
is its set of velocity and density logs. Velocity logs
(recorded as transit time per foot) and density logs are TRANSMISSION
RESPONSE
digitized uniformly at 1- or 2-ft depth intervals. Next,
(step A) transit times are summed until integral multi-
ples of the selected sampling rate AT are achieved;
then the depth interval required for AT transit time is
divided by AT to obtain an average velocity. In turn, SOURCE
WAVEFORM
this is multiplied by the average density to obtain the
average impedance for that AT layer.
SYNTHETIC
TRANSMITTED
Step B, obtaining reflection coefficients at inter-
faces, is a straightforward application of the formula
nonlinear in Z,, the acoustic impedance of layer i, FIG. A- 1. Generation of time-average synthetic seis-
mogram.
zi+l - zi
Ri = (A- 5)
zi+l + Zi

The transmissionresponsecalculation (step C), as (1960). The final stage (step D) of synthetic seismo-
discussed in the text, is also a nonlinear operation. gram generation is convolution of the impulse re-
Details of its constructionare contained in Wuenschel sponse from step C with a source waveform.

APPENDIX B
COMPARISON OF TIME-AVERAGE AND STRESS-STRAIN METHODS
FOR COMPUTING INTERVAL VELOCITIES
Sherwood (1962) and dErceville and Kunetz The velocity V, of the equal-transit time layer in-
(1%3) proposed a method for computing an equal- volves both the velocity Vi and density pi of the thin
traveltime model which is different from that which layers; i.e.,
we used. The same procedure is inherent in Backus
1 PC
(1%2) discussion of anisotropic media. Instead of -=_ p.
(B-2)
preserving the first-arrival traveltimes when combin- v: d i=l PiVf

ing many thin layers into one equal-transit time layer,


they preserve the stress-strainrelationship. This pro- (dErceville and Kunetz assumed pi = p for all i.)
ceditre yields a density pc for the layer of thickness Sherwood noted that
d which is simply the averagedensity of the layer; i.e.,
The traveltime acrossthe equivalent homogeneous
bed is greater than the traveltime of the first arrival
pc=1 n (B-1) transmitted across the continuously varying bed.
d zIpidi This is due to the fact that in estimating the prop-
Effect of Subsurface Sampling 1829

erties of the equivalent homogeneous bed, we and


have essentially made allowance for the multiple
reflections that delay energy transmission across v;_ 1+2a+aZ=1+2a.
--
the inhomogeneousbed. V::
Downloaded 12/28/15 to 132.239.1.230. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

dErceville and Kunetz made the same point. That Therefore,


the traveltime is indeed greater follows directly from
Schwartzs inequality. The extent of this difference is
not obvious. If there is no density variation (pi = p for
(B-5)
all i), then
n
n x 7:
v:,_
-- i=l
G-3)
dErceville and Kunetz gave an equation equivalent
to equation (B-5) but expressed in reflection coeffi-
v: (2 Ti)2
cients and ratio of transit times. An extreme subsur-
i=l face would be one in which half of the traveltimes
where VP is the velocity of the tktarrival and 7i is were T and the other half were 27. In this case, the
the traveltime across the ith thin layer. Using fractional error would be
algebraic manipulation,

c-1
n2T2

1 2
=--. 1
a
-zz-
i=l j=* i=l VC 18
and
Even in this extreme case, the error is less than 6
l/2 i 5 (Ti - Tj) percent.
i=l j=1
1+ 03 -4) For well 2, which has both velocity and density
2 Ti variations, sampling at a 4-msec interval using equa-
( i=l > tion (B-2) resulted in a 12-msec increase in travel-
time through the well. Since the overall traveltime
for the original well is 2.0 set, this was a 0.6 percent
If the error in using VP instead of V, is (Y,
average increase.The traveltime for well 18 increased
v, = V,(l + a), by only 8 msec, for an average increaseof 0.4 percent.

Potrebbero piacerti anche