You are on page 1of 116

Common Mid Point Stacking

The Common Midpoint Stack


Definitions
Seismic lines are shot using the arrangement depicted in Figure 1 .
Figure 1
A source-point is recorded by a spread of receivers, which are usually velocity-sensitive
geophones on land, and pressure-sensitive hydrophones at sea. The source-spread system is
then rolled along by a distance equal to the source interval, and the next source-point is recorded
at the new receiver positions. At sea, the roll-along is eected by towing the source-spread
system behind the ship! on land, it is done electronically, and the planted geophones are not
moved.
"e see rom Figure 1 that trace # rom record A, trace $ rom record %, and trace & rom record '
sample the same subsurace point! when the relector is parallel to the surace, that subsurace
point lies directly beneath a surace point around which the three source-receiver pairs are
symmetrically disposed ( Figure # ).
Figure 2
This surace point is the common mid-point, so called because it is shared by several or many
ray-paths, each distinguished by its source-to-receiver oset. *ach raypath corresponds to a
seismic trace, but the several or many traces are derived rom dierent source points, and so
appear on dierent source records.
The operation o sorting all the traces having a common midpoint is called gathering, and the
suite o traces thus assembled is called a common-midpoint gather, or simply a cmp gather. The
purpose o the gathering is clear rom Figure # + because the raypaths impinge on the same
portion o the relector, the traces record substantially the same signal; because the raypaths are
otherwise dierent in space and in time, the traces record different ambient noise. Thus, addition
o all the traces having a common midpoint , ater appropriate corrections , enhances the
signal-to-noise ratio. The addition o these traces is called cmp stacking.
"hen the relector has dip, the traces sharing a common mid-point do not share a common depth
point ( Figure - ).
Figure 3
The area rom which the relections come in is thereby extended, in the updip direction. Stac.ing
these traces has a tendency to reduce the amplitude o the stac.ed trace, and thus the general
phase appearance o the relector. The magnitude o this alteration depends on the amount o dip
and the steps ta.en to correct or it. For most commonly encountered dips, however, we have
observed that the stac.ing process is very tolerant o the dips ( the solution to this problem, by
the way, is dip moveout optimization, or /01- see /eregows.i, 123& ).
%eore we can stac. the traces o the gather, we must ma.e the appropriate corrections we
spo.e o earlier. These corrections ensure that the signal is substantially the same on all the
traces. This means that+
the traces must sample substantially the same portion o the subsurace!
the signal must have the same time on all the traces! and
the signal must have the same shape on all the traces.
A Review of the Prestack Processes
A relection signal changes rom trace to trace along a common-midpoint gather or several
reasons. The obvious dierences are those o relection time! more subtle dierences occur in the
shape or character o the relection.
Time dierences may be static, aecting the entire trace by the same amount, or dynamic,
varying with relection time. 4n general, static time dierences arise rom raypath irregularities in
the near-surace! dynamic time dierences arise rom dierences in the raypath length ( Figure 1
and Figure # ).
Figure 2
Figure 1
The irst o the time corrections occurs early in the processing! this is the 5removal5 o the near-
surace eected by the datum corrections. "hen they require no processing analysis, they are
oten incorporated with the initial processes.
To ma.e these corrections, we irst establish a seismic datum, generally below the
inhomogeneous near-surace. This datum now serves as our surace o 6ero time, and all
subsequent timing is done rom this origin. "e see rom Figure - that the total correction has
contributions rom two parts o the ray-path+ rom source to datum on the way down, and rom
datum to receiver on the way up.
Figure 3
These are the source static and the receiver static, respectively.
/etermination o both the source static and the receiver static requires some .nowledge o the
velocity (or velocities) in the near-surace. (The thickness o the near-surace is, in this context, a
trivial problem! it is simply the elevation at the surace minus the elevation o the datum.) 4
seismic wor. is done with dynamite in drilled holes, the near-surace velocity is inerred rom the
uphole time ( Figure $ ). The depth o the hole divided by the direct arrival time to an up-hole
geophone is the average near-surace velocity. This velocity, in turn, is used to compute the
source and receiver statics.
Figure 4
4 the seismic source is a surace source such as an air gun or a vibrator, then the near-surace
velocity must be deduced rom previous wor. in the area, rom a special velocity survey, or rom
reraction arrivals. "e must be prepared to repeat the survey down the line i surace conditions
presage a change in near-surace velocity.
1nce we have calculated the source and receiver corrections, we have only to apply them! that
is, to subtract these times rom the seismic trace. For each record, we would apply the source
static to all the traces at once! we say the source static is a common-source correction. Then we
would apply the receiver statics as common-receiver corrections ( Figure 7 ).
Figure 5
The ob8ect o the datum corrections is to simulate the trace that would have been obtained i the
sources and receivers had all been on the seismic datum. To the extent this is achieved, the
remaining trace-to-trace time variations are such that the relection alignment is essentially a
hyperbola ( Figure & ). The increase in time with oset is, o course, the normal moveout.
Figure 6
Although we may regard the travel-time pattern as a hyperbola, deined as in Figure & , the real
pattern is not so simple. This is because the equation or travel time does not contain only two
terms, but rather an ininite number o them. The so-called higher-order terms can be ignored as
long as the eective spread length (ater the mute) is not much greater than the depth o the
relector.
For a given 6ero-oset time, and over a given range o source-to-receiver osets, the travel-time
pattern o Figure & is uniquely characteri6ed by a variable that happens to have dimensions o
length9time. :eophysicists were thereore quic. to call this variable a velocity! because the
normal-moveout correction according to this variable is that used or cmp stac.ing, it came to be
.nown as a stac.ing velocity! inally, the determination o this variable came to be called a
velocity analysis. This use o the word 5velocity5 is not a good one, but we seem to be stuc. with
it.
"hen we perorm a velocity analysis, we are searching or hyperbola that best its the travel-time
pattern o the relection. The goodness o the it may be determined visually or numerically. There
is some interpretation involved, o course, because the best it does not always correspond to a
legitimate primary relection. 4n Figure ; , or instance, a velocity o 1;<; m9s yields a good
correction at 1.<7 s, but it also does so at #.1< s and -.17 s.
Figure 7
'learly, the latter two are multiples, so we pic. or interpolate a higher velocity or those relection
times, even though the velocity analysis gives no indication that the higher velocity corresponds
to a primary event.
"e see, thereore, that the desired nmo correction aligns primary relections, but misaligns
multiple events, thereby attenuating the multiples when we come to stac. the traces.
=aving perormed the ield-statics and nmo corrections, we hope that the relection alignment is
perect. 4n the real world, o course, it is not, because our simple models do not perectly simulate
the real world. The thic.ness and velocity o the near-surace are highly variable! these variations
in turn may aect the nmo determination.
This imprecision leaves us with a series o small, unsystematic timing errors rom trace to trace
along the gather, which we reer to as residual static..
4n the standard determination o residual statics , the process is called automatic statics, or
autostatics the static is considered to comprise our terms+ the residual source static, the
residual receiver static, the residual normal moveout, and the eect o the structure (sometimes
called the dip term). 'onveniently, these our terms correspond to the our directions along which
we may gather data.
Autostatics programs vary rom one processing house to another, but they all have three
important steps in common+
First is an estimate o the trace-to-trace time dierences!
Then, we decompose those time dierences into the our components!
Finally, we apply the computed source and receiver statics to the traces.
Further, we oten constrain the autostatics solution by invo.ing the principle o surface
consistency. This stipulates that trace-to-trace misalignment can be regarded as a unction o
source and receiver locations at the surface, rather than o raypaths in the subsurace. 1 course
surace consistency is not always appropriate. "e remember also that time-variant corrections
may be required.
So, to the degree possible, we have aligned the relections on the cmp gather. There remains the
question o wave shape or character, because we do not wish to add relections that are out o
phase or too unli.e in requency content.
The reason we might have a problem is that the traces o a gather all come rom dierent source-
points. 1n land data shot with dynamite, source variables , charge si6e, depth, cavity eects,
etc. , can result in dierent wavelets at each shotpoint. At sea, the ailure o one air gun in the
array can also aect the source signature. *ven when we have a controlled source such as
vibrators on land, the pulse-shaping eects o the near-surace can modiy the downgoing
wavelets dierently at each source position. Finally, i we are wor.ing in the transition 6one, we
may very well have a combination o sources and receivers ( Figure 3 ).
Figure 8
To standardi6e all the pulse shapes, we generally require some sort o wavelet processing. The
irst ob8ective is to correct all the source signatures to a uniorm shape! a second ob8ective is to
select a stable and compact shape. "ith marine and land vibroseis data, we .now the
approximate source signatures! wavelet processing is thus deterministic, as we preer. "here the
source signature is un.nown, as with impulsive land data, we determine the outgoing pulse shape
statistically.
1nce we .now the source signature, we can calculate its amplitude and phase spectra. Acting on
these with the amplitude-requency and phase-requency responses o the required ilter, we
emerge with the desired wavelet. This ilter is then applied to all the traces, so that the relections
in a cmp gather now align according to both time and shape.
And so it remains to stac. the data, and we begin with an analysis o the conventional stac. and
its beneits.
The Mean-Amplitude Stack
A mean-amplitude stac. is a common-midpoint stac. in which each stac.ed trace is a simple
summation o the traces in the gather, out to the limits o the mutes. The resulting trace is a
statistical average o the constituent traces. 4n essence, we add all the sample values at a given
time, accounting or the ramp in the mute 6one, and divide by the number o traces! the
normalized trace has amplitudes comparable to the input.
'0> stac.ing has the beneit o improving the signal-to-noise ratio by a actor equal to the square
root o the number o traces in the stac.. "e should qualiy that now with a ew observations
about both signal and noise.
1. As much recent wor. has shown, the relection coeicient is not generally independent
o the oset ( Figure 1 ).
Figure 1
For some important geologic models, the relection coeicient exhibits a strong
dependence on oset or angles o incidence greater than about #<?. This behavior is the
basis or that ield o processing called amplitude-versus-offset (avo analysis.
For now, we may say that the change in relection coeicient with angle depends on the
change in >oisson@s ratio across the interace. 4n Figure # , or example, a thin briny layer
has a >oisson@s ratio about equal to that o the material above it! the relection coeicient
(and thereore the amplitude) is nearly constant at all osets.
Figure 2
4n Figure - , however, a gassy layer has a much lower >oisson@s ratio.
Figure 3
As the oset and the incident angle increase, so does the amplitude.
#. Amplitude need not change as only a unction o oset or it to aect the stac..
Figure $ reminds us that in the presence o dip, the relection points represented by each
trace do not converge at a point.
Figure 4
The resulting smear is bad enough, but i the physical properties change enough to
cause changes in the normal-incidence relection coeicient within the relection area,
then we may get spurious amplitudes over one or more traces.
4n general, whenever we see a signiicant amount o dip on the near-trace section or
brute stac., we preer to do a prestack partial migration, or dip-moveout processing, prior
to stac.ing.
-. Another simpliying assumption about the signal concerns the correction or the
decrease in amplitude with time. This decrease is due to spherical spreading, additional
spreading rom reraction, and various propagation eects. To compensate or the
spherical spreading, which is ully determined, we multiply each sample by its time t. To
compensate or the remaining amplitude losses, we multiply each sample by an
additional actor ekt, so that the total amplitude ad8ustment is tekt.
The exponential expansion is intended only as an approximate, temporary correction.
Thereore, we should understand now that the amplitude relations among the traces o a
gather are probably not correct! certainly, now that we have a better .nowledge o the
velocity behavior along the line, we can do better.
$. The improvement in the signal-to-noise ratio eected by stac.ing assumes that
each trace is a composite o relection signal and random noise. *ven i we restrict the
discussion to ambient noise , noise which is 8ust 5there5 in the absence o a shot , this
assumption applies only to certain types o noise. Thus, the noise generated by towing a
streamer may well be random, but common-mode intererence may not be.
Thereore, i we are to spea. o such improvements, we must say that we are relating the
signal to incoherent, uncorrelated noise. For most other ambient noises, the signal-to-
noise improvement generally is less than .
7. 'ountering the above is the understanding that, or certain recording geometries and
certain source-generated noise, is the minimum improvement attainable. The
common-midpoint gather acts as a very long array, long enough to suppress virtually all
wavelengths o ground roll or water waves ( Figure 7 ,
Figure 5
Figure & ,
Figure 6
Figure ; , and Figure 3 ).
Figure 7
To ensure continuity o the geophones across the gather, which is the primary criterion o
the stack-array, it is more eicient to set up the ield geometry in one o two ways+
Figure 8
At sea, the group length is equal to the group interval, and the source interval is hal the
group interval!
1n land, the group length, the group interval, and the source interval are all equal, and
the source is between groups o a split spread.
1ther geometries are possible but they require some degree o mixing in creating the
stac.-array.
The eectiveness o the stac.-array depends on the constancy o the source-generated
noise rom record to record, and thereore rom trace to trace in the gather. Still, the worst
case is that o random variations in the source-generated noise! then, the near-total
suppression o the noise reduces to statistical, or , suppression.
"e see, thereore, that even straight stac.ing implies assumptions that occasionally bear
examination. Aevertheless, the method is powerul, robust, and extremely useul.
Features of the Conventional Stack
=ere, we summari6e the beneits and limitations o the common-midpoint stac..
1bviously, a ma8or beneit o stac.ing is the improvement in the signal-to-noise ratio.
Stac.ing also allows us to display the suite o stac.ed traces in a orm that resembles a
6ero-oset, normal-incidence section, such as might be obtained through the method
ideali6ed in Figure 1 and Figure # .
Figure 1
The ield method o multiple coverage allows us to record the required traces more
eiciently, and gives us the signal-to-noise beneits as well.
Figure 2
"e understand, however, that in the stac.ed trace so obtained, the wavelet and the
relection coeicient have been averaged over a range o osets, and sometimes over an
enlarged portion o the relector.
For all its appearances, however, the stac.ed section should not be regarded as
equivalent to a geologic cross-section, i only or reasons o dip ( Figure - ).
Figure 3
Aevertheless, as a irst approximation to the subsurace structure, obtained without any
prior knowledge of that structure, the cmp stac. is remar.ably accurate. Aaturally, beore
we draw too many inerences about the structure, we must account or whatever vertical
exaggeration is present in the display.
%y virtue o the residual normal moveout, the stac. is also an excellent attenuator o
certain long-path multiples. The eect is greatest when the velocity increases rapidly with
time, and can be improved (at relection times beyond the end o the mute) by a
lengthening o the spread. Short-path multiples and water-bottom reverberation are best
treated by deconvolution methods.
The stac. also has economic beneits! when we stac. A-old data, we reduce by a
actor o A the number o traces that need subsequent processing. This is no small
consideration or marine lines that are routinely greater than $3-old.
Stac.ing is not very useul against noise bac.-scattered in the near-surace rom a
scatterer to the side o the spread. 4t can be useul i the scatterer is behind the spread (
Figure $ )! where it ails in the latter case, we may require the use o the requency-
wavenumber ilters.
Figure 4
Irreular !eometr"
=aving said that common depth point and common reflection point are not synonyms or common
midpoint, we must ac.nowledge that sometimes even the midpoints are not common. Surace
conditions sometimes prohibit a regular progression o source points, or even a straight line o
source points. 4n such cases, the gather may consist o traces whose midpoints lie within a hal
midpoint interval o the nominal midpoint location ( Figure 1 (a), (c)).
Figure 1
Alternatively, we may gather midpoints that lie within a circle of stack ( Figure 1 (c))! this may
include some overlap and thereore mild mixing o the stac.ed traces.
The ultimate case, o course, is the croo.ed line. *ven in level terrain, this .ind o geometry may
be dictated by cultural or even political concerns! in mountainous terrain, it is all but unavoidable (
Figure # ). 1bviously, a croo.ed line cannot be processed in the manner that a straight line
would. 4nstead, we must set up the processing geometry to account or the recording geometry.
Figure 2
The irst step is to determine the source-to-receiver osets. 4 the source and receiver positions
are .nown to good accuracy, this is easily done. 4t is then a simple matter to compute the
midpoint positions ( Figure - and Figure $ ).
Figure 3
From there,
Figure 4
the tas. is to sort all the scattered midpoints into bins ( Figure 7 )! each bin is essentially a gather,
and we wish to treat the constituent traces in much the same manner as we treat the traces o a
common-midpoint gather.
Figure 5
Thereore, we adapt the binning methods to the overall curvature o the line, pursuing what is
oten called a binning strategy. There are several considerations that dictate the si6e and shape
o each bin.
1. The line connecting the bin centers passes through the greatest density o midpoints,
and is oten computed using a locali6ed center-o-gravity scheme. This line is
subsequently regarded as the nominal physical location o the stac.ed section. The stack
line need not be straight, o course. 4t may consist o a series o segments, each with a
dierent a6imuth, as long as the stac. line is unbro.en.
#. 4 the bin centers satisy the above conditions, then each bin should have a uniorm
distribution o osets. A bin with only short osets is useless or multiple attenuation,
whereas a bin with only ar osets 8eopardi6es the velocity analysis. 1 course, i a bin is
near a sharp bend in the line, the continuity criterion above may render the desired oset
uniormity impossible. 4t is also useul to have a reasonably uniorm number o traces
within each bin.
-. "e also wish to minimi6e the distribution o source-to-receiver a6imuths within each
bin. This is a more subtle problem, and one that may be expensive to accommodate. 4
there is a signiicant amount o dip, traces having the same oset but dierent a6imuths
have dierent relection times or a given relection depth. This dierence in relection
time alters both the apparent structure and the velocity determination. 4n practice,
accommodation o these eects may require iterations through a --/ model! in general,
thereore, we preer to restrict the a6imuth distribution when we select traces or binning (
see Barner et al., 12;2 ).
$. 4t is permissible or bins to overlap, as indeed they must at bends in the stac. line. 4t is
also permissible or bins to vary in their cross-line dimension, or diameter, as long as the
variation rom one bin to the next is small. Their inline dimension, or bin interval,
however, should remain constant.
7. "e are not constrained to put every scattered midpoint into a bin, as long as the oset
and a6imuth conditions are met. 4n Figure & , or example, we can probably widen the
bins to include the traces to the northeast, in the manner shown.
Figure 6
"e do this in Figure ; , and then reali6e that little would be gained by widening them
urther, to include the midpoints south o the line.
Figure 7
The inal bin coniguration ( Figure 3 ) shows that bin diameter and orientation change along the
line, to conorm as closely as possible to both the original shot line and the spread o midpoints.
Figure 8
4n eect, we may regard the binning strategy as a search or a 5best-it5 to the shot line. (And, o
course, a similar strategy may be used or marine lines plagued by severe streamer eathering.)
Stackin# Alterative Forms
A$T%R&ATI'% F(RMS (F STAC)I&!
There are several alternatives to mean-amplitude common midpoint stac.ing. To place these in
perspective, we must see clearly that stac.ing accomplishes several ob8ectives.
The enhancement o signal relative to ambient noise. This is particularly useul at late
relection times.
The enhancement o signal relative to coherent noise. This is particularly useul over the
time range at which source-generated noise is received, and is strengthened i the ield
geometry satisies the stac.-array criteria.
The enhancement o primary signal relative to multiples. This is particularly useul or
multiples involving only relectors at medium depths, and or velocities increasing
mar.edly with depth.
The maintenance o subsurace continuity in those situations where it would otherwise be
impaired by missing source positions, missing receiver positions, local highly absorptive
weathering, and blind spots caused by reraction, aulting, and racturing.
The empirical conclusion rom many years o experience is that the ordinary mean-amplitude
stac. provides, at minimum cost, an excellent balance among these several ob8ectives. 4n some
situations, however, one o these ob8ectives is much more important than the others. Then, we
must be prepared to stri.e a different balance, so that we may strengthen one beneit at the
expense o the others.
Sometimes, thereore, we consider alternative types o stac., select one appropriate to the
problem, and as. whether its beneit 8ustiies its ris.s and its costs.
*eihtin +" Mutes
The simplest weights to apply to data are < and 1. "hen the weights are all 1, o course, we have
a mean-amplitude stac.. "hen some o the weights are <, we have muted those samples.
0ost muting is done in the early stages o processing. "e mute irst brea.s, noisy traces,
transients, and certain .inds o coherent noise, all on a record basis. 1ur rule or well-deined
large-amplitude noise is that it is almost always better to mute it than to rely on the stac. to
attenuate it.
1n the common-midpoint gather, the most obvious reason or muting is the pulse distortion
caused by normal-moveout correction ( Figure 1 ,
Figure 1
Figure # , and Figure - ).
Figure 2
"e oten call this distortion nmo stretch; in act, however, the nmo correction itsel is a stretch o
the time axis to ma.e all traces loo. li.e 6ero-oset traces.
Figure 3
"e remember the mechanism or the distortion. 4n Figure $ we have the 6ero-oset and ar-oset
traces o an uncorrected cmp gather.
Figure 4
The width o the pulse, which is centered at 1.<<< s, is $< ms (so its mean requency is #7 =6)!
or the variables shown, the pulse extends rom <.23< s to 1.<#< s on the near trace, and rom
1.17$ s to 1.12$ s on the ar trace. The nmo correction according to a velocity o #$$< m9s brings
the middle o the pulse to 1.<<< s, as it should ( Figure 7 ). The onset o the pulse, however, is
corrected to <.2;& s, and the end o the pulse to 1.<#- s. The nmo correction has stretched the
ar-oset pulse by ; ms, thereby lowering its 5requency5 to #1 =6.
Figure 5
As we can see rom Figures 1-7 , the distortion is worst at ar osets and early times. %ut, as we
see rom Figure 7 and the exercise, the amount o the distortion , and, by extension, the degree
o spectral change on the stac.ed trace , can be calculated in advance. The permissible
spectral change should be time-variant, to accommodate the decreasing expectation o good
bandwidth with time. :iven this speciication, the computer then may calculate an appropriate
mute , we may call it an 5automute5- on each trace in time. 4n general, this is probably suicient
or a irst estimate o the velocity distribution.
"hat the computer cannot do is decide a mute based on considerations other than nmo stretch.
4n particular, there may be troublesome source-generated noise, such as a head wave or a shear
reraction. 4n this case, the processor or the analyst has to ma.e a decision based on a visual
inspection o the gathers. And, o course, it is best i the mute is selected gather by gather, rather
than on representative gathers.
So the choice o the mute depends on the stac.ing velocities and on the source-generated noise,
and thereore varies along the line. 4t also depends, to a lesser degree, on the requencies. 4 we
wish to preserve the high requencies, then the maximum allowable distortion must be low ,
perhaps #<C or so , to guard against the inclusion o lower-requency traces in the stac..
1n the other hand, we cannot aord to mute too willingly or too severely i there is a severe
multiple problem in the area. 4n this case, a higher distortion allowance may be necessary to
include enough traces to give us a ull cycle o residual nmo, or best attenuation o the multiple.
Then, we might need to extend the distortion limit! many programs deault to 7<C.
The muting o marine data may depend on the water depth. 1ten, a shallow water bottom may
cause relected, reracted, and direct arrivals to appear on the near traces. =ere, the
simultaneous arrival o relections and coherent noise prevents us rom muting only the noise. 4n
such cases, we preer to ilter the traces in the f-k domain.
To assess the eects o the mute, and to see i the mute is hurting us in any way, it is helpul to
generate a series o stac. panels, such as those o Figure & ,
Figure 6
Figure ; ,
Figure 7
Figure 3 ,
Figure 8
Figure 2 ,
Figure 9
Figure 1< ,and Figure 11
Figure 11
.
Figure 10
=ere, a portion o the line has been stac.ed with dierent mutes, and thereore dierent olds o
coverage at a critical relection time. As we add more and more o the ar osets ( Figures &-3 ),
the shallow section beneits. *ventually, however, the mute ails to .ill the early waterborne
arrival, and the shallow section deteriorates ( Figures 2-11 )
*eihtin +" (ffset
To improve the stac. response o primaries relative to multiples, we may weight the traces by
oset. *ssentially, we put more emphasis on the ar osets, where the multiples are more out o
phase. The scheme may not be elegant, but it is easy and inexpensive, and thereore a good
alternative to straight stac.ing, especially when the problem is a particular multiple.
For best results, the weights are calculated by the program based on the velocities o the primary
and the multiple. The velocities are determined in the usual way the moveout dierence between
the primary and the problematic multiple is then used to calculate the weights. 0ore-advanced
schemes may also account or requency content, amplitude ratios between primary and multiple,
and perhaps the level o random noise.
Figure 1 , and Figure # describe how oset weighting wor.s.
Figure 1
4n Figure 1 ,
Figure 2
we see the nmo-corrected travel-time patterns corresponding to a primary and, 8ust below it, a
noninterering multiple. The primary is well aligned, and thereore yields a good stac. response,
but the misalignment o the multiple is not enough to cause it to stac. out. %y giving more weight
to the ar traces than to the near traces ( Figure # ), we ensure that each in-phase trace is added
to an out-o-phase trace. The multiple is minimi6ed, but the primary is unaected. "e see the
result on real data in Figure - .
Figure 3
"e should note that oset-weighted stac.ing is most eective in those cases where mean-
amplitude stac.ing would wor. airly well anyway. 4n general, this means that we need at least
one-hal cycle o residual normal moveout across the gather, and preerably one or more ull
cycles. :iven this level o rnmo, stac.ing without weights reduces the multiple amplitude
measurably! oset weighting simply ta.es it urther. Figure $ quantiies the improvement.
Figure 4
The curves show the approximate ration o multiple amplitude to primary amplitude, as a unction
o stac.ing velocity! the primary velocity is assumed to be -<7< m9s. "e see that or multiple
velocities below #7<< m9s, the oset-weighted stac. improves the enhancement o the primary,
relative to the multiples, by about 1< d%.
Aow we must as. what we lose or the beneit o increased multiple suppression. The irst and
ma8or loss is that we no longer have suppress ion o the ambient noise. The second loss is
that the same degree o nmo stretch on the ar traces produces more spectral change on the
stac.ed trace.
"e can minimi6e these problems by using a time-variant weighting. 4n the shallow section, where
higher requencies are more critical and multiple suppression is more eective, the weighting
approaches that o a mean-amplitude stac.. The weighting can then increase over the 6one
where multiple suppression is less eective. /eeper still, the weighting is removed as the
relection signal alls toward the ambient noise.
( 4n the literature, we sometimes see the term 5optimum-weight stac.5 when the meaning is
oset-weighted stac.. 4n general, however, an optimum-weight stac. is any technique that
involves a trace-weighting scheme based on some optimi6ation criterion. The term is thereore
ambiguous, and we are probably better o or not using it. )
Diversit" Stackin
A ma8or use o this method is in compensating or large noise bursts in unsummed vibroseis data.
4n eect, we weight each trace in inverse proportion to the amount o noise in it+ the greater the
noise, the lower the weight. The method proceeds as ollows+
1. 'alculate the average power within a speciic time window or each trace. "e do this
by calculating the energy and dividing by the time.
#. 'ompute a scaling actor such that the product o actor and average power is a
constant.
-. Ta.ing the scaling actor to apply to the center o its window, interpolate linearly
between actors, thereby emerging with a gain trace.
$. 'ross-multiply the data trace and the gain trace to yield a scaled trace.
7. Add the scaled traces, add the gain traces, and divide the irst sum by the second.
The process assumes a constant signal amplitude rom trace to trace! by so doing, it ascribes the
trace-to-trace dierences in average power entirely to the dierences in noise power (a good
assumption i the noise amplitude is much higher than the signal amplitude, which was our initial
condition anyway). Furthermore, it does so while preserving the original amplitude variations o
the signal in the input records.
Although the diversity stac., or inverse-power-weighted stack, is most common as a vibroseis
summing technique, it has uses in the stac.ing step which ma.e its inclusion here logical.
principally, these uses are in the stac.ing o marine data contaminated by noise , intererence,
really , rom other seismic vessels.
4n general, we preer to shoot marine data when other crews are inactive. This is because marine
wor. is essentially a continuous operation, meaning that the recorders are wor.ing almost
constantly. 4n the past, to prevent the recording o noise rom other crews, companies would
speciy a maximum intererence level, beyond which the crew must stop recording. 'rews
generally cooperated with each other on a 5time-sharing5 basis, whereby they too. turns
recording.
4n many cases, however, the crew-intererence speciication was too conservative, so that much
time was lost waiting or the interering crew to inish. "ith diversity stac.ing, noise that loo.s
virulent, even hopeless, on a ield record can be suppressed to reasonable, even imperceptible,
levels on the inal stac..
Figure 1 shows ive consecutive shot records showing the sort o noise we get rom interering
crews! here, the pea. amplitudes are at least three times greater than many old contract
speciications would allow.
Figure 1
Figure # , however, shows what these noise patterns loo. li.e on common-midpoint gathers.
Figure 2
1bviously, even a mean-amplitude stac. should go a long way toward attenuating the worst o
the noise.
1ne o the reasons the crew noise is so disorgani6ed on the gathers (as opposed to its
appearance on the record) is that the source o the intererence is not synchroni6ed with the
source o the survey ship. The noise arrives at dierent times on ad8acent records, so that
gathering urther misaligns it. 0ore important, however, is that the asynchronous noise
guarantees that the noise does not appear on all the traces of the gather. Thus, i A is the number
o traces in the gather, 0 is the number o traces on which the noise appears, and the noise is
suiciently incoherent, then the suppression is by a actor o . This becomes i A and
0 are equal.
"e see 8ust how eective mean-amplitude stac.ing is in Figure - ,
Figure 3
which is a stac. o the line rom which the records and gathers o Figure 1 , and Figure # are
ta.en. For comparison, Figure $ shows the same line without any interering crew noise.
Figure 4
*arly in the section, we see that alternative stac.ing is unnecessary! it is only in the deeper part
o the section that the intererence is apparent.
To combat the deeper noise, we resort to the diversity stac., which requires only that the signal
amplitude be approximately uniorm rom trace to trace. Figure 7 ,
Figure 5
Figure & ,
Figure 6
and Figure ; ,
Figure 7
compares the results o the mean-amplitude and diversity stac.s to the case o no intererence,
and Figure 3 quantiies the improvement or three levels o signal-to-ambient noise.
Figure 8
=ere, the poststac. signal-to-burst-noise ratio is plotted as a unction o the prestac. ratio or a
&<-old stac.. The lower curve represents the improvement due to the mean-amplitude stac.. The
upper curves represent the improvement due to the diversity stac. or three signal-to-noise ratios.
As always, we must as. what we lose or this beneit. First, we accept that the beneit exists only
or noise that temporarily exceeds the signal. Second, we have sub8ected the data to a nonlinear
process, so that we no longer have complete conidence in the relations between inal relection
amplitudes, nor in the eect o subsequent iltering (or inverse-iltering) operations. Third, the
stac.-array criteria may no longer be satisied.
Trimmed-Mean Stackin
The trimmed-mean stac. is an alternative to the diversity stac. in areas o high intererence rom
other vessels. This is essentially a mean-amplitude stac. perormed ater traces having
anomalously high amplitudes have been discarded. The amount o trimming is speciied by the
trimming variable which ranges rom < to <.7! this is simply the portion o traces at either end o
the amplitude range. Thus, these stac.s are sometimes called -trimmed-mean stac.s. ("hen
D <, o course, the trimmed-mean stac. reduces to the conventional mean-amplitude stac..)
Figure 1 ,
Figure 1
Figure # ,
Figure 2
Figure - ,
Figure 3
Figure $ ,
Figure 4
and Figure 7 show the eect o trimming $<C ( D <.#) o the amplitudes on a section
contaminated by severe intererence.
Figure 5
As with the diversity stac., the relative incoherence o the noise on the gather helps the trimming
do its 8ob.
4n deciding on the level o trimming, two considerations are important. 1ne is that we preer the
re8ected samples to be distributed symmetrically about the mean. This proviso permits us to be
reasonably conident that we do not change the estimate o the mean or a normal distribution o
amplitudes.
"e should also avoid discarding too many traces, lest we compromise too severely the beneits
o the subsequent stac.ing. A trimming o $<C ( D <.#) on a line that is nominally &<-old causes
a #-d% increase in the level o the ambient noise, relative to ull old. To the degree possible, we
should observe this #-d% limit! this may require us to ensure that the seismic source and the
noise source are not synchroni6ed.
Median Stack
Along with the diversity stac. and the trimmed-mean stac., another stac. with the property o
excluding anomalously high amplitudes is the median stack. 4n the series <.2, 1.-, 1.<, <.2, 1.1,
;.<, or example, the mean-amplitude stac. would have an output value o #.<, which is obviously
wrong. The median, however, is the middle value o the distribution or, in the case o an even
number o amplitudes, the average o the two middle values. 4n our example, the median value,
1.<7, ma.es much more sense than the mean! a trimmed-mean value o 1.<3 would also be
sensible. (And, o course, the median stac. is the same as an -trimmed-mean stac. with D
<.7.)
4n constructing a median stac., it is important to exclude rom consideration any 6ero values that
result rom muting. Further, the median is inappropriate i there are high amplitudes , say, rom
a multiple , on more than about a third o the traces.
The median stac. combines very low cost with quite remar.able eicacy against large noise, and
it does not require (as does the diversity stac.) that the noise come in bursts. 0edian stac.ing is
standard in the processing Eertical Seismic >roiling ( ES> )
The loss that we ace, in using both the trimmed-mean and median stac.s, is that they do less
well against multiples, and again are nonlinear! a median stac. is sub8ect to abrupt changes in
the amplitude values o consecutive samples o the output. Although the resulting appearance o
high-requency noise may be removed by iltering, there remains the uncertainty o relative
amplitudes which is always introduced by nonlinearity.
Coherence Stack
There are several variations to the coherence stac., but they are all designed to enhance the
amplitudes o coherent signal while suppressing the amplitudes o incoherent noise.
4n essence, we perorm on nmo-corrected gathers the same sort o coherence scan we do in a
velocity analysis! indeed, semblance is a common measure o the coherence (basically output
energy divided by input energy). 4 the semblance window contains a primary event, the
semblance is high, and this high value is assigned to the center o the window. 4 there is no
primary in the window, the semblance is low, and we assign a low value at the center. 4n this
manner, we emerge with a coherence trace, interpolating linearly between window centers. The
coherence stac. is then ormed by scaling the stac.ed trace by the semblance.
The coherence stac. (which appears under many trade names) is very eective in improving the
ratio o properly corrected primaries to both multiples and noise. 4ts wea.ness springs directly
rom its strength+ it is intolerant o minor errors in the stac.ing velocity. Thus, each poor velocity
analysis, or inappropriate velocity interpolation, leads to wea. 6ones in the aected relections.
The result, which tends to loo. geologically unnatural, is oten unacceptable to the interpreter.
A variation to the coherence stac. is the correlation stac.. =ere, we cross-correlate the stac.ed
trace with each o its input traces. 4 the correlation coeicient alls below a certain threshold, we
scale down that input trace (or .ill it entirely i the coeicient is much below the threshold). "e
may reine this by ma.ing the process iterative! we restac. without the oending trace, and
perorm another set o cross-correlations.
Velocity Filtering
'%$(CIT, FI$T%RI&!
4n ield wor. one o the ma8or unctions o the spread is to allow us to recogni6e dierent types o
seismic waves by their characteristic velocities across the spread. 4n particular, we can separate
dierent arrivals on the basis o these characteristic velocities by using two-dimensional ilters.
Two-dimensional ilters are useul in seismic processing because o their ability to re8ect events
having a particular velocity across the spread, or across the section. Such ilters are thereore
called velocity filters; because they are invariably applied in the requency domain ( as well as in
the slant-stac. domain ), they are also .nown as fre!uency-wavenumber (f-k filters.
The Two-Dimensional Fourier Transform
The seismic ield record, which represents signal amplitudes as a unction o time and o oset, is
oten reerred to as a time-domain representation. To get to its fre!uency-domain counterpart, we
remember the concept o the one-dimensional Fourier transorm+ we correlate the trace with a
sinusoid, which gives a measure o the amplitude o that requency in the trace. /oing this or all
requencies, we get the spectral density and phase o the trace. Filtering the trace is then a
matter o scaling the spectrum as a unction o requency, using a scale actor o one or
requencies we wish to pass, 6ero or those we wish to re8ect, and a suitable ramp in between.
Feassembly o the spectral components yields the iltered trace.
Seismic data also have a spatial requency. "e see this in Figure 1 ! at the let is a sinusoidal
waveorm in time, and at the right is the waveorm as it is recorded at a spread o our receivers.
Figure 1
The arriving wave ront is represented by the alignment o the open circles. 4 the receiver spacing
is small enough to prevent aliasing (as it is here), then the 5trace5 connecting the receiver
amplitudes may be regarded as a waveorm in space.
Gust as the temporal requency o a waveorm is the reciprocal o its temporal period, so is the
spatial requency o a waveorm the reciprocal o its spatial period. Spatial period is simply the
apparent wavelength , so spatial requency is simply . "e call this quantity the
wavenumber, and it has units o inverse length! oten, we speciy the wavenumber in terms o
some unit length, such as so many cycles per receiver interval, or per .ilometer, or per mile.
4n Figure 1 , the waves are coming up to the surace rom the let. Further, because they impinge
on the receivers one at a time, they are seen as having a hori6ontal velocity. This is the apparent
velocity Eap, because it is not a true measure o the speed o the waveront. "e quantiy it as the
dierence in receiver coordinates divided by the dierence in arrival times, or "#t.
"e remember that the apparent velocity is equal to the true velocity divided by the cosine o the
angle o wave propagation. *quivalently, it is equal to the true velocity divided by the sine o the
angle o dip. Thus, we may say that the waves in Figure 1 impinge on the receivers one at a time
by reason o dip, and that the dip may be quantiied as the dierence in arrival times divided by
the dierence in receiver coordinates, or "#t.
4n Figure # , the waves are coming up rom directly below, rom a lat relector.
Figure 2
%ecause all the receivers sample the same phase at the same time, the apparent wavelength and
the apparent velocity are ininite, and the wavenumber is thereore 6ero. Finally, in Figure - , the
waves are coming up rom the right.
Figure 3
The apparent wavelength is the same as it is in Figure 1 , so the wavenumber and the dip have
the same magnitude! however, they must have the opposite sign. 4n this situation, we may also
deine a negative apparent velocity, meaning that the wave is traveling right to let.
"avenumbers and dips, unli.e requencies, are both positive and negative. The sign convention
is this+ A wave front that is down to the right on a section, or down from near offset to far offset on
a record, has a positive wavenumber, and the reflector from which it comes has a positive dip. A
wave front that is down to the left on a section, or down from far offset to near offset on a record,
has a negative wavenumber, and the reflector from which it comes has a negative dip.
4t is possible or a waveront with a positive wavenumber to appear as though it has a negative
wavenumber. "e see this in Figure $ , where the waveront can be interpreted (wrongly) as
coming up rom the right.
Figure 4
1 course, it is also interpreted as having a negative apparent velocity and a negative dip.
Figure $ is an example o spatial aliasing, and it occurs because the receivers are too widely
spaced to prevent the ambiguity. To avoid spatial aliasing. we require the receiver spacing to be
smaller than Eap9#f$ , where f$ is the Ayquist requency. The quantity E9f$ is the inverse o the
Ayquist wavenumber k$.
Feturning now to the waveorms o Figure 1 , reproduced in gray in Figure 7 , we see the eect o
the same apparent velocity (or the same amount o dip) on a waveorm o higher requency.
Figure 5
1bviously, the higher requency results in a higher wavenumber. Furthermore, when we graph
requency and wavenumber or the two waveorms ( Figure & ), we see that the relation is simple
and direct+ %f two events have the same apparent velocity, their respective fre!uency-
wavenumber pairs lie in a radial line in the f-k plot.
Figure 6
A urther result apparent in Figure & is that the slope o the radial line is simply f#k; because . D
l9 a, the slope a is the velocity o the wavetrain. Thus, dierent velocities have dierent
slopes on the f-k plot, even i the events cross or overlap partially on the time-distance plot.
"e can extend the results o Figure & by adding the requency-wavenumber pairs rom Figure # ,
and Figure - . This we do in Figure ; , which represents a generali6ed f-k plot. The vertical axis
represents, again, the temporal requency, and ranges rom 6ero to the Ayquist requency.
Figure 7
The hori6ontal axis represents the wavenumber, and ranges between the negative and positive
Ayquist wavenumbers.
The Two-Dimensional Filter
The concept o a #-/ ilter is much the same as that o a 1-/ ilter. This time, however, we
correlate an n-trace record with an n-trace sinusoid having a certain velocity, such as Figure 1 .
Figure 1
The irst correlation gives us a value or one -. pair. Heeping the requency the same and
changing the velocity ( Figure # ), we correlate again to get a value or a second -. pair.
Figure 2
=aving done this or the ull range o velocities to the limit o aliasing ( Figure - ), we then go to
the next requency component, and begin again.
Figure 3
"hen we have reached the alias requency, we have the complete -. density. Then, the velocity
ilter is a matter o scaling the components as a unction o requency and wavenumber.
Feassembly o the components yields the velocity-iltered record.
4n practice, this sort o sample-by-sample, trace-by-trace correlation would be slow and
expensive. So we do it in the -. domain, harmonic by harmonic (so we can incorporate requency
iltering at the same time) and wavenumber by wavenumber. %ut the principle is the same+ we
use other traces, and an alignment across them, to optimi6e a stepout and subtract it.
As an example, we see in Figure $ a 5relection5 (event 1), crossed by 5ground roll5 (event #).
Figure 4
%ecause their spectra overlap, requency iltering would be ineective. 4n the -. plot, however,
there is no overlap ( Figure 7 ).
Figure 5
So, we designate the re8ect band by a wedge emanating rom the origin. As with one-dimensional
iltering, we apply a taper between the re8ect 6one and the pass 6one, steep enough to prevent
the attenuation o -. components we need, but gentle enough to prevent ringing. The re8ection is
best i the event is straight, and i the event amplitudes on all traces are equal. (The latter
condition is essentially a requirement that noise amplitudes are equal on all traces.) The iltered
record is shown in Figure & ! we see that the ground roll has been much reduced.
Figure 6
Feduced, but not eliminated. "e see why in Figure ; , which extends the -. plot o Figure 7 .
Figure 7
The trace spacing o the ground roll, which has a low velocity on the record, is too large to
prevent aliasing o the higher requencies. This aliasing shows up in the negative-wavenumber
side o Figure ; , as though event # were actually a series o events traveling rom the ar traces
to the near traces! the 5velocity5 varies with requency. %ecause the re8ect region in the -. plot
must emanate rom the origin, a substantial portion o the -. representation o event # is
untouched. The high-requency remnant o Figure & is characteristic o uniltered aliases.
"e prevent spatial aliasing in the ield by using a iner group interval. 4n the processing, we
simulate the iner interval by trace interpolation. Typically, the procedure begins with trace-to-
trace correlations within a sliding time gate to determine the dominant coherence direction. Then,
a simple halway-point interpolation between the time samples along that direction gives the trace
value in the center o the gate ( this trace interpolation procedure is also eective in the slant
stac. domain).
(1n real data, the interpolation scheme must be sensitive to the presence o incoherent noise!
otherwise, we run the ris. o converting that noise into spurious, laterally coherent signal.
"e also note that trace interpolation cannot replace proper sampling! we cannot expect it to
provide geologic resolution greater than that actually recorded in the ield.)
Figure 3 ,
Figure 8
Figure 2 ,
Figure 9
and Figure 1< show the eect o the combination o trace interpolation ( Figure 3 )
Figure 10
and f-k iltering ( Figure 2 ) on the record o Figure $ . "ith the interpolated traces removed, so
that we restore the original trace spacing ( Figure 1< ), we see that the reduction o the ground
roll is much greater.
Prestack Filterin
The irst time we perorm an f-k analysis is usually in the ield, where we use it to discern the
dierent types o arrival on a noise spread. 4n traditional practice, we then design a receiver
pattern, determine its #-/ response, and apply that response to the f-k spectrum o the data. 4n so
doing, we can see how the receiver array would aect the noise record.
As we might guess, prestac. f-k iltering is usually reserved or reraction, ground roll, and
streamer noise that source and receiver arrays have been unsuccessul in eliminating. %eore we
resort to it, however, we should ma.e sure that stac.ing , in general and in the stac.-array , is
also unable to remove the problem. >restac. iltering is an expensive process, and we must
weigh its cost against its beneit.
1ne inal use o prestac. f-k iltering is or the removal o multiples (but, again, only where nothing
else is suicient). 4n a sense, this is a time-variant iltering that requires good .nowledge o
multiple velocities.
The irst step is to correct the cmp gathers according to the velocity o the multiple system, so that
the primaries are overcorrected ( Figure 1 ).
Figure 1
The f-k plot o this gather then has the multiples aligned on the requency axis, and the primaries
entirely on one side o it ( Figure # ).
Figure 2
To the extent that elimination o the multiples can be eected, the resulting gather has primaries
only ( Figure - ).
Figure 3
The eectiveness o the process can be 8udged by comparing velocity analyses beore and ater
the process ( Figure $ ).
Figure 4
1n real data, o course, there may be many multiple systems, each with its own 5velocity.5 So the
choice o one multiple-velocity distribution is equivalent to the choice o one multiple-generating
system. This means that f-k iltering o multiples gives improved results only in highly speciic
cases. 4n general then, we use f-k iltering to remove multiples only i there is one multiple
system. *ven then, the eect is greatest or intermediate or deep hori6ons, where the velocity
contrast with the primaries is greatest. 4 there are several multiple systems, f-k iltering is not an
eective option.
Poststack Filterin
The apparent velocity o a (hyperbolic) relection varies over a wide range, rom slightly greater
than its true velocity to ininity. The low end o this range, which occurs at early times and long
osets, may overlap the velocity o some coherent noise. "orse, that coherent noise can be in a
different part o the record. Thereore, prestac. f-k iltering oten needs a wide pass-band, and
may result in indierent noise attenuation.
4t may also be that the coherent noise train is not visible on the ield record. 4n Figure 1 (a), or
example, the section is plagued by noise that stac.s well.
Figure 1
Any f-k iltering we wish to apply must be poststac.. 4n the present context, what we hope to
accomplish is more appropriately called dip filtering.
The principle is the same, o course, except that alignment o data on an f-k plot has signiicance
as a dip rather than as a velocity. Thus, a hori6ontal event on a seismic section shows up as a
vertical alignment in f-k space. 0ost events o seismic interest lie within a small wedge straddling
the requency axis! most coherent-noise trains, on the other hand, exhibit a much steeper dip.
Thus, the section o Figure 1 (a) is an ideal candidate or dip iltering, because the signal and the
noise should be easily separable on an f-k plot. Ater dip iltering, the relections are much clearer
( Figure 1 (b)).
1n the stac.ed section, o course, both positive and negative dips are equally li.ely. Thus, it
ma.es sense to plot both positive and negative wavenumbers. As we might expect, however, a
reversal o dip may be construed as spatial aliasing. 1ur rule here is airly simple+ i an alignment
does not pass through the origin (that is, D . D <), then the event is an alias. Almost all
legitimate events on an -. plot pass through the origin. (The exception is a dispersed event.)
"hat we need to do is study the stac.ed section closely, and determine the highest dip o
interest. 4 the noise is so severe that we cannot determine this rom the section, then we use the
highest design dip set up or the ield wor.. To this, we add a small saety actor, and use this
inal dip as the lowest design dip or the iltering. The highest design dip or the iltering can be
whatever we wish (provided, again, that we prevent wraparound by cutting o the ilter at the
spatial Ayquist).
"e do run the ris., i the noise is high, o removing poorly deined primaries having steep dip.
4nteractive processing can help us evaluate this problem quic.ly.
The Choice of 'aria+les
The processing required beore we do an f-k analysis and iltering depends on the reason or the
analysis. 4 we are studying a noise spread or determination o optimum source or receiver
arrays, then we probably want to do no processing that aects statics or moveout. As regards
moveout, the ground roll and reraction that we wish to analy6e do not display any hyperbolic
curvature. Further, i we do ma.e some nmo correction, then we alter the apparent velocities o
the ground roll and the reraction. Inless all the stepouts are still within the re8ect 6one, the nmo
correction could render the f-k analysis useless.
Statics are a dierent matter. "e have said that coherence on an f-k plot is a unction o apparent
velocity, or apparent dip. 4t is reasonable to expect, thereore, that severe statics will hamper the
analysis. 1n the other hand, we recogni6e that ground roll, traveling as it does along the surace,
is a wea. unction o the topographic irregularity. Furthermore, static corrections that would
optimi6e relection data may not necessarily do the same or reraction data. 1n balance,
thereore, it is usually best to have neither nmo nor statics corrected beore f-k analysis o ield
data. The rule o thumb is that, to attenuate an unwanted wave train, we wish to .eep it straight.
The same is not true, o course, or any poststac. analysis.
The Fourier transorm o a short time waveorm may produce ambiguities in the requency
domain. The longer the time waveorm, the better the resolution o the component requencies. A
similar result arises when we consider Fourier transorms to yield wavenumber densities+ the
greater the number o traces considered, the better resolved the wavenumber distribution. Thus,
i we have a seismic record in which all the events have comparable moveouts, or apparent
velocities, then the input to the f-k analysis should be the entire record (or, at least, as much o it
as contains signal).
4n reality, a record usually has a wide range o moveouts. "hat happens on the f-k plot then is
that the ma8or lobe broadens, encompassing the larger range o wavenumbers and requencies.
4n general, this is not a problem as long as the noise is well deined! ater all, the f-k analysis here
is to be used as a re&ect ilter. And we reali6e that because there are so many more time samples
than there are space samples, the temporal requencies are always better resolved than the
spatial requencies.
1nce we have the f-k plot, the tas. is to design the ilter, which may be a wedge emanating rom
the origin ( it may also be a polygon ). The width o the ilter depends on the range o velocities
we wish to ilter. As with requency ilters, we also need to speciy slopes, because a sharp cuto
o the wedge causes a ringy impulse response.
4ndeed, Figure 1 , and Figure # show that a poor choice o slopes puts us at ris. o 5creating5
events.
Figure 1
The slopes o Figure -
Figure 3
, and Figure $ do not have this problem.
Figure 2
(And slopes need not be straight, but may have curved edges.) 4n some circumstances, however,
we are orced to steep slopes or narrow pass-bands, in which case the consequent ringing
imparts a 5wormy5 appearance to the section.
Figure 4
Then, the standard processing remedy is to add bac. some o the uniltered section to the iltered
section, in ratios o 1+1 up to 1+-.
Finally, we must understand that f-k iltering has penalties attached to its use. "e have said that
noise re8ection by f-k iltering wor.s best i the traces have the same noise amplitude. 4n general,
this requires amplitude manipulation that also aects the signal. "e may equali6e the traces ater
the iltering, but this sacriices our conidence in the stac.ed amplitudes, and prevents any avo
analysis we might wish to do.
As an alternative however, some programs store the gain in the header. This gain can then be
removed ater -. iltering, which thus prevents this compromising o amplitudes.
Final Filtering
-andpass Filterin
%y the time we get to stac., we .now the usable signal bandwidth o the data. 4n general, then,
our next step is to design a bandpass filter with an output9input response o one over that
bandwidth, and a decreasing response on both sides o it ( Figure 1 ). 4n the requency domain,
we multiply the amplitude spectrum o the seismic trace by the response o the ilter. 4n the time
domain, we convolve the input time series with the ilter operator.
Figure 1
4n Figure 1 , we see that the nominal cutoff fre!uencies f' and f( do not actually have ull
response. Father, they represent -- d% points, a remnant o the bandpass ilter@s electrical-
engineering heritage! engineers regard the hal-power (-- d%) point as the edge o the pass-band.
For our purposes, this is acceptable, because a --d% loss o the cuto requencies is not li.ely to
be noticeable on the iltered trace.
A bandpass ilter is also characteri6ed by its slopes, which tell us how much a requency outside
the pass-band is reduced by the ilter. 1ne way to deine the slopes is to speciy amplitude values
or our dierent requencies! the inner two are the cuto requencies, and the outer two are the
null requencies. 4n a design sense, the null requencies have a response o 6ero, but with inite-
length ilters, this may not be reali6ed.
Another way to speciy a slope is as a taper, that is, so many d% per octave. Thus, i a ilter is
deined as 1< to &< =6 with slopes o 1# d%9oct and -& d%9oct, we would expect a 7-=6
component to have one-ourth the amplitude o a 1<-=6 component (or, equivalently, 17 d% below
pea. response).
The choice o slopes represents a compromise. The ininite slope o a boxcar ilter cannot be
achieved with operators o inite length, so the 5ideal5 bandpass ilter would probably have an
amplitude spectrum li.e that o Figure # (a).
Figure 2
Inortunately, the impulse response o this ilter 5rings5 ( Figure # (b))! the ringing is a
superposition o the cuto requencies.
4n Figure - , we see the eect o using a gentler slope.
Figure 3
The time-domain operator is shorter (that is, it has ewer non6ero components), which permits
eicient and economic time-domain iltering.
4n both Figure # and Figure - , the slopes are not linear tapers. Father, they are scaled by the
irst quarter-cycle o a cos# unction, which is general practice. This permits the amplitude
spectrum to have rounded corners at the cuto and null requencies. 4n general, however, even a
linear taper is eective at reducing the ringing.
Although the ilter o Figure - has a short, nonringy operator, we see rom its response that it is
slow to re8ect requencies above the cuto. To establish the correct operator length, thereore, we
remember that the eective length o the operator is inversely proportional to the requency
bandwidth. For most applications, an operator length between #<< and $<< ms is appropriate.
Time-'ariant Filterin
%ecause the amplitude spectrum o a seismic trace is li.ely to get narrower with time, most inal
ilters are time-variant. 4n turn, time-variance oten necessitates space-variance, so that important
mar.er hori6ons retain the same ilter across the section.
4n principle, a time-variant ilter starts out as two or more time-invariant ilters, and proceeds in the
ollowing manner ( Figures 1-& ).
1. "e decide which ilters to apply to the various trace segments. >erhaps we choose 1<
to 3< =6 down to 1.3 s, 1< to &7 =6 rom #.# to -.< s, and 17 to 7< =6 rom -.7 s to the
end o the trace.
#. "e ilter the entire trace three times, once by each ilter selected ( Figure 1 ,
Figure 1
Figure # , and Figure - ).
Figure 3
Figure 2
-. "e scale each iltered trace to have ull amplitude within the speciied times, and
decreasing amplitude beyond them.
Thus, or the irst trace o Figure $ ,
Figure 4
Figure 7 ,
Figure 5
and Figure & , we .eep ull amplitude or the irst 1.3 s, steadily decrease the amplitude
rom 1.3 to #.# s, then 6ero the trace beyond #.# s.
Figure 6
Similarly, the second trace is 6eroed to 1.3 s, ramped up to #.# s, ull amplitude to -.< s,
ramped down to -.7 s, then 6eroed beyond -.7 s. Finally, the third trace is 6eroed to -.<
s, ramped up to -.7 s, and ull amplitude beyond -.7 s.
$. Then we add the three traces ( Figure ; ).
4n practice, economic considerations do not permit us to ilter the entire trace three times.
Figure 7
4nstead, we would apply the irst ilter to a segment whose length is #.# s plus hal the length o
the operator (to minimi6e edge eects). "e would truncate the trace in a similar ashion or the
other ilters, and then scale and add as above.
4n the parlance, the trace o Figure ; is said to have a time-variant ilter. This is quite dierent
rom a time-varying ilter, in which the ilter response changes with each sample o the trace.
>erhaps we should call the irst a time-gated ilter, or simply a gated ilter, to describe the process
better and to avoid conusion.
%etween the iltered time gates, the ramp should be restricted to 5noneventul5 areas, especially
when stratigraphic interpretation is attempted, and should be at least as long as the operator. The
problem is that the summed amplitude response is not smooth, but rather has a discontinuity. 4n
Figure 3 ,
Figure 8
curve A is the high-cut portion o the irst ilter rom Figure 1 , and Figure # ! its slope is assumed
to be -& d%9oct. 'urve % is the counterpart rom the second ilter. 'urve ' represents the
normali6ed summation a quarter o the way down the ramp (that is, at 1.2 s).
1bviously, such a spectrum must have a ringy pulse shape. The situation improves as we
approach #.# s ( Figure 2 ), but the basic message remains+ we cannot trust a wiggle that is in the
ramp between two ilters.
Figure 9
A nonlinear ramp may help, but the problem will persist to some degree.
Filter Panels
Final ilters are selected on the basis o ilter panels ( Figure 1 ,
Figure 1
Figure # ,
Figure 2
Figure - ,
Figure 3
Figure $ ,
Figure 4
Figure 7 ,
Figure 5
Figure & ,
Figure 6
Figure ; ,
Figure 7
and Figure 3 ), output or a portion o the line that we regard as representative, or several portions
i we encounter dierent types o signal-to-noise problem (or geology) along the line.
Figure 8
1ur goal is a ilter with the widest bandwidth compatible with an acceptable level o noise. Such a
ilter is by nature time-variant, because the signal spectrum is changing and the relative
importance o the noise is rising. Ater we set up the ilter panels, thereore, we ma.e the
bandwidth decision as a unction o time, and then speciy the ilter.
There are several approaches to interpreting ilter panels. 4n Figures 1-3, we .eep the low cut
constant at 1< =6, and increase the high cut rom #< to 3< =6. This approach shows what limit to
the bandwidth is the maximum or acceptable noise. To determine the limit on the low end, we
use Figure 2 ,
Figure 9
Figure 1< ,
Figure 10
Figure 11 ,
Figure 11
Figure 1# , and Figure 1- , which .eeps the high cut constant at &< =6, and decreases the low
cut rom #< to 7 =6.
Figure 12
For this approach, some .ind o amplitude equali6ation is required. 4n both igures, the uniltered
stac. is shown or comparison.
Figure 13
Another approach, .nown as Time Eariant Spectral "hitening ( Jilma6, 123; ), uses narrow ilter
panels- usually a 1<-=6 bandwidth ( Figure 1$ ) - and has a second and separate unction, which
requires both equali6ed and unequali6ed panels.
Figure 14
4 an equali6ed high-requency panel shows a relection event clearly, then, o course, this band
will have been included in the analysis o Figures 1-3 and 2-1-. %ut i this event is lower in
amplitude than a lower-requency panel (o the same bandwidth in hert6), then we may conclude
that the deconvolution schemes have not done enough 4n this case, the data should be sub8ected
to a spectrum-whitening program next, and the ilter analysis repeated.
4t is oten tempting to concentrate our eorts on the high requencies, meaning that the choice o
the low-cut ilter oten gets less attention than it deserves. %ut we .now that good temporal
resolution requires good bandwidth, not 8ust good highs.
"e should also consider our reasons or a particular choice o low-cut requency. 4 we are led to
a high setting because it reduces hori6ontal 5banding5 across the section, we iner one o two
things+ the low-requency noise has not been attenuated in the ield (by geophone arrays or by
the stac.-array), or we are loo.ing at residual multiples. The lesson is that we should not be doing
with a inal ilter what should be done by proper ield design.
Finally, we note that the choice o ilters determines the inal appearance o the section.
Thereore, it is advisable to include the interpreter in the selection process. An understanding o
the exploration ob8ective is important to the selection.
Display
Final Amplitude Manipulation
=ow we treat the inal amplitude depends, as usual, on the exploration ob8ectives. /o we want a
well-balanced section, or eye appeal, or a true-amplitude display, rom which we can iner
dierences in acoustic impedanceK 4s the overall structural picture paramount, or do we hope to
learn something rom the brea.-up o a relectionK
4nterpretersL needs are twoold. First, they need to .now that the displayed variations o relection
strength along a particular relector are genuine, whether that relector is shallow or deep.
Second, they need to be able to assess the ratio o relection strength between shallow and deep
relections. These needs may be less important or a reconnaissance survey, but they are crucial
or any detailed wor..
*arly in the processing, we ad8usted the amplitudes on the record by a total actor o tekt. The
time actor was a ully determined compensation or the losses caused by spherical spreading!
the exponential actor was an empirically determined compensation or the unspeciied 5other5
losses. Aow we wish to remove the exponential correction, and replace it with one based on our
improved .nowledge o those losses.
First, we remember that, in a hori6ontally stratiied earth, the downgoing (and upcoming) seismic
rays are not spherical, but rather are curved by reraction. For a near-normal-incidence path,
thereore, an additional decay o the pressure amplitude can be attributed to the velocity
distribution. The total decay, which is very large, is called geometrical divergence; it is by ar the
dominant agent o the decay o seismic traces with time. "e characteri6e the pressure amplitude
as being inversely proportional to Er#t, where Er is the rms velocity.
1bviously, we are luc.y in this regard! the rms velocity is well approximated by the stac.ing
velocity. At this stage, thereore, we have a airly detailed picture o the velocity ield appropriate
to the seismic line. 4t is a simple matter to replace ekt by Er#9Eo#, where Eo is the velocity at the
time reerence (usually, but not necessarily, the irst sample). "e then need only a scaling actor,
or proportionality constant, to bring the amplitudes within the limits o the plotter and the eye.
%ut o course, this is all too simple, so there must be a complication or two. First is that the
expression Er#t is strictly applicable only to marine wor., where hydrophones measure the
pressure amplitude. 1n land, geophones measure particle velocity, which is 2<? out o phase with
pressure near the source. Fortunately, by the time o the earliest relection o interest, much o
that phase dierence has been removed, ma.ing the Er#t correction a suicient approximation or
land wor..
Another complication is that we cannot compensate correctly both primaries and multiples. 4
multiples do remain, with their smaller rms velocities, they are overcompensated by the
divergence correction, and appear larger than they should. "e must also .eep in mind that the
divergence actor is strictly applicable to the near traces, and presumes the /ix condition o lat,
parallel layering. At arther osets and in more complex layering, the simple relation between rms
velocity and geometrical divergence brea.s down.
Finally, we recall that we must terminate the divergence correction, lest we increase the noise
amplitudes.
And so we have made an amplitude ad8ustment that is suicient or most cases. "here it is not,
we require a inal exponential expansion, speciied as so many d% per second. A typical
expansion would be - d%9s. "e note, however, that it is li.ely to be much less i the
deconvolution has included an inverse-M operation . 4n that case, o course, the inverse-M should
also have been terminated when the stac.ed signal alls below the noise.
4 we do use a supplemental expansion, we must account or it whenever we compute the ratio o
relection strength between shallow and deep relectors. "e should also account or it when we
assess the varying strength o a single relector, i that relector is aulted or has suicient dip that
its relection time changes signiicantly. 'learly, then, we use the exponential expansion only
when we must.
"e may then wish to perorm a trace equali6ation or reconnaissance lines, but never or detailed
wor.. "here we do so, the goal is the balancing o amplitudes, so a time- and space-variant
automatic gain control (agc) is preerred! a typical window length is <.7 s. 4n that case, it is still a
good idea to provide a comparison section without equali6ation ( Figure 1 , and Figure #
Figure 2
).
Figure 1
Displa" Scales
The choice o scales depends on the type o survey , reconnaissance, semidetail, or detail. For
the irst, high requencies are less critical, so there is no problem with speciying a timing scale o
ive centimeters per second (about # in9s). 4 we need to loo. at the traces in greater detail, then
we double the time scale to 1< cm9s ($ in9s). And i the highest-requency interpretation is being
done, some interpreters choose a timing scale o #< cm9s (3 in9s). 4n some companies,
particularly in the Inited States, the standard timing scales are #.7, 7, and 1< in9s, respectively,
or the three types o survey. 4n portions o western 'anada, the scales are -.;7, ;.7, and 17 in9s
( Ta+le ./, below )
1nce we have selected the timing scale, we concentrate on the trace spacing, the hori6ontal
scale. Again, or a given vertical scale, we have several choices o hori6ontal scale, depending on
our requirements ( Ta+le ./, below ). The irst one that comes to mind is a scale that matches the
map on which the interpreter will be drawing the map. For recon wor., this is 1+7< <<<, so that #
cm on the section equals 1 .m on the ground. For prospect delineation, we .eep the proportions
the same, so that the scale is 1+#7 <<<, or $ cm to 1 .m. 4n both cases, the eect or typical
velocities is a vertical exaggeration o about #+1. (Some companies, again usually in the Inited
States, preer to have their ull-scale sections at a scale o 1+#$ <<<, to match the topographic
sheets produced by the IS:S. This hori6ontal scale combines with the vertical scale o 7 in9s to
yield the same approximate #+1 vertical exaggeration.)


%0ploration
o+1ective
2ori3ontal
scale
'ertical
scale
'ertical
e0aeration
Fecon
1+7<,<<<
# cm D 1 .m
7 cm9s #+1
Semidetail
1+#7,<<<
$ cm D 1 .m
1< cm9s #+1
Fine detail
1+#7,<<<
$ cm D 1.m
#< cm9s $+1
Ta+le ./ )ommon display scales. *hese represent common practice, but many variations are
possible.
To convert a scale o 1+7< <<< or 1+#7 <<< to a trace spacing, we remember that in most cases,
the trace interval, which is the same as the common-midpoint interval, is equal to hal the group
interval used in the ield.
There is also merit in producing a section at 5natural5 scales, so that the proportions o the
eatures are approximately correct. "e can do this by either halving the time scale or doubling
the trace spacing. And some interpreters ind it very helpul to have a 5squashed5 display, made
at perhaps 19# the vertical scale and 197 the hori6ontal scale. For a small price, we have a section
on which the geologic continuity is made most apparent! this is surely more cost-eective than
many techniques or enhancing the signal-to-noise ratio by processing. Further, such a section
brings a ma8or increase to the clarity o the aults! indeed, sometimes it alone demonstrates the
existence o shallow-angle aults.
"hatever our choice o scales, we do everyone a service by providing a scale bar on the label or
at the top o the section ( Figure 1 ).
Figure 1
Displa" Modes
"hile wiggles-only is used or some detail wor., most interpreters preer wiggles with variable
area or variable area only. 1thers preer variable density, and still others preer rectified variable
area ( Figure 1 ).
Figure 1
1ne consideration in the choice o display amplitude is that, as a matter o principle, we would
preer to display the entire waveorm. Ater all, there is no reason to believe that subtleties o a
waveorm will not appear at or near a pea. or a trough. %ut there is no doubt that relections
appear clearer i they are represented by blac. and white bands across the section! this requires
that the traces overlap. 4n practice, thereore, we see. a middle course, considering the ollowing
actors+
A very deensible rule is to set the amplitude level (sometimes called the plot gain such
that the strongest relections in the 6one o interest overlap by about 1<C. This is
generally sound in good record areas!
4n poorer record areas, and in particular where the problem is statics, it helps to
increase the amplitudes. 'ompanies wor.ing habitually in such areas may thereore have
a standard practice involving much more than 1<C!
The very act that the data are poor means that little interpretation o trace subtleties is
possible. The emphasis thereore swings away rom the preservation o the waveorm,
toward the accentuation o the alignment!
"hatever compromise is adopted, there may be dierent ways o expressing it. Figure
# illustrates a situation where the machine calculates the average absolute amplitude on
each trace (or, i we are using agc, in each window o each trace), and sets this average
equal to an excursion o hal the trace spacing.
Figure 2
Felections stronger than this average thus produce some overlap. Eariations on this
norm, or reasons o poor data or emphasis on trace subtleties, may then be expressed
as plus or minus so many d% o plot gain!
4n addition to the variable o amplitude, there is the variable o bias ( Figure - ). For
variable area, the bias moves the 6ero-amplitude level away rom the center o the trace
width. For va-wiggle, the bias blac.s in the pea.s to a base line that is no longer
coincident with 6ero amplitude!
Figure 3
4 the purpose o increasing the amplitude and the bias is to maximi6e the impression o
continuity, we preer va-wiggle, because the signal clipped by reason o large amplitudes
is still visible in the wiggle. 1n the other hand, i we wish to remove the accentuation o
the pea. with respect to the trough, we may choose va only!
4 we have chosen va-wiggle, and the priority is not alignment, but rather relection
character, then we should reduce the amplitude and the bias. "e do so because, as we
have said, subtleties o a waveorm are 8ust as li.ely to occur on pea.s as on troughs,
and they must be seen!
4n the extreme case, the exploration problem practically lies in the character o the
relections. Then, we use 6ero or slightly negative bias, and allow only the slightest
clipping. 0any interpreters then preer variable-area only!
For squashed displays, which allow us to see relection continuity through 6ones o poor
signal-to-noise ratio, variable density is mandatory. This retains some dynamic range in a
trace that is very narrow. Again, we wish to .eep the amplitudes and bias relatively low!
4 it were not or the diiculties o reproduction, variable density would probably be the
preerred mode or all displays. At the wor. station, these diiculties disappear, and some
degree o hori6ontal squash is oten required or long lines. This is when variable density
proves its value! the results are both revealing and pleasing!
4 a section is later to be reduced photographically, variable-area only is preerred! the
wiggles may not survive the reduction!
0icrostac.s are best plotted with variable density, again because we are trying to
accommodate the ull dynamic range in a very narrow trace. 4 the gather alignment is
more important than the relection character, however, a strongly biased va-only plot is
also permissible! the trace is then eectively on-o!
Finally, we note that many interpreters preer to have two wor.ing sections, one at
normal polarity, the other at reverse polarity. An interesting variation to this is the double
variable-area display ( Figure $ , and Figure 7 ).
Figure 4
This is produced by blac.ing in half of each pea. and each trough.
Figure 5
At the wor. station, double (rectiied) va in red and blue is very useul or ault
interpretation.
Polarit" Conventions
Several actors enter any discussion o polarity conventions+
the sense o the source, that is, whether it is a compression or a rareaction!
the sign o the relection coeicient, that is, whether it is positive or negative!
the manner o connection o the receiver, through the ampliier, to the digiti6er!
the signs o the numbers recorded on tape!
the direction o delection on the monitor record! and
the direction o delection on the section.
The 12;7 S*: polarity convention, or an impulsive source emitting a compression, requires the
hydrophone connections to be such that a compression in the water (that is, a notional positive
relection) is recorded on tape as a negative number, displayed on the monitor record as a
downswing, and displayed on the section as a swing to the let (that is, a 5white5).
The particle velocity measured by a geophone is in phase, in the ar ield, with the excess
pressure measured by a hydrophone. Thus, the counterpart convention or land wor. requires the
geophone connections to be such that an upward motion o the geophone case (that is, the
surace o the ground) yields the same negative number, the same downswing, and the same
5white5.
The convention that a positive pressure or an upward motion should be recorded as a negative
number sometimes is considered odd, but it is universally accepted. Any convention will do,
because we still retain ull choice o the display polarity.
The convention that a positive pressure or an upward motion should be displayed as a
downswing is rooted in history. "e all e"pect the irst brea.s to brea. downward, and it seems
pointless at this late date even to consider a change.
=owever, the convention that a positive pressure or an upward motion should be displayed on the
section as a 5white5 is questionable. True, it ollows the convention or a monitor record. %ut given
a compressional source and modern wide-band processing to 6ero phase, it implies that a
positive relection is represented by a 5white,5 and a negative relection by a 5blac..5 This has
three signiicant disadvantages+
%ecause a 5blac.5 is ar more visible than a 5white,5 it means that positive relections
are less clear than negative relections. Although the small and less-important relectors
in the earth are airly well balanced between positive and negative, the important
relectors (particularly, or seismic-stratigraphic purposes, the erosional unconormities)
tend to be positive. 4t is these which should be given the beneit o the 5blac.5 display!
A very important case o this occurs when there is a gas-liquid relection . *ven i this
relection is wea., we need to see it, and one o its recognition criteria is that it may be
assumed to be positive ( Figure 1 , and Figure #
Figure 2
)!
Figure 1
The positive-to-the-let convention is the opposite o that used or acoustic and density
logs, and or pseudolog traces computed rom the seismic traces.
From time to time, the possibility has been raised o a new and revised S*: polarity convention.
That is why, in this course, we have reerred to the original S*: convention as the '+,- S*:
convention. Any need or a revised convention, however, has been set aside by the current
practice in the industry, which is to call or two displays, one 5S*: normal5 and the other 5S*:
reversed.5 Then, the interpreter may choose to wor. either one , the 5normal5 i he has to tie
other surveys displayed in this way, and the 5reversed5 otherwise.
"hatever the interpreter may choose to do, it is essential that the observer be sure to chec. and
record the polarity o the ield equipment, and that the processor be sure to control and label the
polarity o the display.
Plot Annotation
"e annotate the section on both the top and the side. Top labeling is as ollows+
The direction o plotting conorms to the way maps are drawn+ east is to the right. The
exception to this rule is a line that runs due north, in which case north is to the right.
Sometimes, the interpreter may as. that the plotting direction be reversed, to allow better
comparison between a line running due north and one running a ew degrees east o
south ( Figure 1 ). Bine direction must be annotated on each section, probably on the top
right, and certainly on the label!
Figure 1
A inal section must always have line intersections annotated across the top, at the
appropriate intersecting trace. %ecause the section will eventually be olded along the
intersection, an intersection tic. mar. at the bottom ensures straight olding. 4
intersecting lines have been processed at the same time, then we must chec. the line
ties or both position and relection character. Any misties must be investigated
immediately! no section may be transmitted to the client or the interpreter unless all
misties have been investigated, and those associated with the ield wor. or the
processing have been resolved!
Also across the top o the section are proiles o elevation, loating datum, residual or
total statics, water depth, old o coverage (sometimes on the bottom), and tabulations o
the stac.ing velocities used!
The trace annotation must ma.e it clear whether the numbering is by ground station or
by common midpoint!
A line showing the hori6ontal extent o the cmp gather is also useul, in giving the
interpreter a eeling or the blurring that occurs in the stac.!
Also useul are annotations o ground conditions, rivers, bends in the line, and no-
permit 6ones.
The side label gives the interpreter inormation about the ield wor., the processing, and the
display. This critical inormation is the only guide to the nature o the data, and can alert the user
to its probable limitation+
1bviously, the label must agree with the observer@s report on all matters pertaining to
the ield wor.. A diagram o the spread or streamer geometry is a useul addition!
The label must also agree with the processing sequence. 4n this regard, it is always
easier to read the label i the processing steps are displayed in sequence. The
5generali6ed5 sequence made speciic by numbers, as used in the past, is cumbersome!
All inormation relative to pulse shape (that is, o phase manipulations as well as
requency manipulations) should be included!
A location map is generally desirable! the line appropriate to the section should be
highlighted!
Finally, the processor and the supervisor sign and date the section.
And so we have a inal stac. that we are pleased and proud to send to the interpreter. 1ur
shipment includes one ilm plot and two or three prints, which are rolled with the printed side out,
or each scale requested. "e also include our test plots+ decon and ilter tests and velocity
analyses, along with the seismic data on tape or wor.station interpretation. Some interpreters
may as. or plots o the corrected gathers. 4t is now up to the interpreter to ind that prospect.