You are on page 1of 18

AN ANALYSIS OF STACKING, RMS, AVERAGE,

AND INTERVAL VELOCITIES OVER A HORIZONTALLY


LAYERED GROUND *

BY

M. AL-CHALABI **

ABSTRACT
AL-CHALABI, M., 1974, An Analysis of Stacking, rms, Average, and Interval Velocities
of a Horizontally Layered Ground, Geophysical Prospecting 22, 458-475.
A correct derivation of rms, average and interval velocities from one another and from
common depth point stacking velocities requires a clear understanding of the relationships
between these velocities. We relate the average velocity to the rms velocity through a
“heterogeneity factor” which is a quantity that gives a measure of the degree of velocity
heterogeneity in the ground. The interval velocity is a quantity which varies according to
the method of its derivation. The difference between rms and stacking velocities depends
on the heterogeneity factor and on the length of the spread. Unless allowed for, this dif-
ference can reverse the advantages of long spreads and cause large errors in interval
velocity determinations. It may be removed through a number of techniques. The ac-’
curacy of stacking velocities in the presence of random “noise” is independent of the
heterogeneity factor. Relevant expressions can be broken down into simple formulae
which give the accuracy quickly and with good precision.

INTRODUCTION

Much of our information about the velocity distribution in the ground is


derived from stacking velocities obtained from common depth point stacks.
These stacking velocities are used as bases for estimating the rms velocities
and are often treated as being synonymous with them. The rms velocities
may be used to estimate interval and average velocities. A correct derivation
of one velocity from another requires a clear understanding of the relationships
between these velocities. In the present work, we deal quantitatively with
these relationships and look into the accuracy of stacking and rms velocities.
We pay a particular attention to the role played by the velocity heterogeneity
of the ground in determining these relationships. To keep the treatment as
simple as possible we restrict the investigation to a horizontally layered ground
and deal with basic principles only.
* Paper presented at the Thirty-fifth Meeting of the European Association of Explorat-
ion Geophysicists in Brighton, June rg73.
** Exploration and Production Research Division, BP Research Centre, Sunbury-on-
Thames, Middlesex, England.
STACKING, RMS, AVERAGE, AND INTERVAL VELOCITIES 459

THEREL~TIONBETWEENAVERAGEAND RMS VELOCITIES


On a horizontally stratified ground, the average velocity to the nth interface
is given by
I fi
T/a= - zd Vk.tk: (I!
To n-1

where vk is the velocity of the kth layer, tk is the two way traveltime within
the Kth layer and To is the two way normal incidence time defined by

To = i tk = 2 i h&k,
k-1 k=l

hk being the thickness of the kth layer.

The corresponding rms velocity is given by


%
V rms = + i 0; t, (2)
( 0 n-1 )

In appendix I we show that

vz, = !$ 2 hk -f hj @kv-~’
Crna - (3)
k=1 ilk+1

where D is the depth to the nth interface. The quantity

gives a measure of the degree of velocity heterogeneity in the ground. We shall


call this quantity the heterogeneity factor. It is a positive quantity, being equal
to zero only when all of the layers have the same velocity (a homogeneous
ground). Its value is independent of the order of layering.

From equation (3) we get


Crna - vZ,
vz, =g (5)
i.e.,
Vrma/Va = (I + &!I” (6)

Equations (3-6) illustrate quantitatively the observation that the rms velocity
equals the average velocity when the ground is homogeneous and progressively
exceeds it as the ground becomes more heterogeneous.
460 M. AL-CHALABI

THE INTERVAL VELOCITY


It is well known from work with continuous velocity logs and synthetic
seismograms that the velocity of an interval does not remain constant over
any significant thickness of rock. A truer picture is the squiggly form of a
velocity log. Let us then look into the meaning of the quantity which we are
calling interval velocity.
The interval velocity zle can be calculated from
v,” = (Vi Tb - vi Tt) / (T&)- TJ (7)
where Vt and Irb are the rms velocities at the top and bottom of the interval
and it and rb are the corresponding normal incidence traveltimes (Dix 1955).
Consider an interval consisting of n segments (or layers) each one having a
uniform velocity ze!(figure I). The interval velocity W of this interval is then
given by
w2 = (q&+1 TTb+1- vf ~1)/(%*1- 4 by equation (7)

SURFACE
___--------A

TRAVELTIME
----
RMS ------ m
VELOCITY SURFACE
r v1 =I
T
v, =2
2 5
E
e
E. w------v-
s v. ------ ---
-I J 3
---------
f
z Yl
I z n+t
. Yl+r
---------
--------
Fig. I. A sketch of an interval consisting of n segments.

At the jth segment, equation (7) gives


VZj+l Tj+l - v; Tj = 20; (Tj+l - Tj).

Therefore,

w2 =
i $ bj+l - g/b,+1 - 4 (8)
i-1
which is the formula for the rms velocity of the interval under consideration.
Therefore, we conclude that an interval velocity calculated from two rms velocities
is the urns velocity of the interval.
STACKING,RMS,AVERAGE,AND INTERVAL VELOCITIES 461

The interval velocity may also be calculated from‘the two average velocities
at the top and bottom of the interval. In this case, a procedure similar to that
used in deriving equation (8) gives

w = zi q (y+1- q)/(Q&+1- 7.n) (9)


I-1

which is the formula for the average velocity of the interval. Therefore,
an interval velocity calmdated from two average velocities is the average velocity
of the interval.
Hence, the interval velocity is a quantity that depends on the method by
which it is calculated. Suppose we had rms velocities obtained from a CDP
stack and average velocities obtained from measurements, at a well nearby.
In this case, an interval velocity calculated using the rms velocities will be
larger than the velocity of the same interval calculated. using the average
velocities. The difference between the two results depends on the heterogeneity
of the interval in accordance with the difference between rms and average
velocities discussed in the previous section.

THE STACKING VELOCITY AS AN ESTIMATE OF THE RMS VELOCITY

The rms velocity is usually estimated from the stacking velocity. In the
absence of errors, the quantity
B = V, - Vrms (10)
represents the bias in this estimate. It can be shown (Al-Chalabi 1973) that the
bias is a non-negative quantity which increases as the quantity
m
x F/c (v, h) Pzk
k-1

increases. Fk(v, h) is a complicated non-negative function of the velocities


and thicknesses of the layers, increasing in magnitude as the heterogeneity
factor g increases. p(= sin B&r) is the ray parameter, 02 being the angle of
incidence at the ith layer.
Over the same ground, the ray parameter increases with increasing offset
and decreases with increasing depth. Therefore, there is always a rapid in-
crease in the bias B as the maximum offset increases. However, for the same
spread geometry, increasing depth does not necessarily reduce the bias; when
there is a large increase in g with depth, the consequent increase in Fk(v, 12)
can swamp the decrease in p and cause B to increase. The fact that the dis-
crepancy between the stacking velocity and the true rms velocity could in-
crease with depth has generally been overlooked in the geophysical literature.
These considerations are illustrated in figures z and 3. We based our data
462 M. AL-CHALABI

INTERVAL VELOCITY

1.5 2~10003m’sec) 4 El
o- *
INTERFACE
1 1615
2 1776
l-

z 1693

=2-
z 4 1960

i 5 2091
6 2164

3- 7 2403

6 2504

Q L 93 ~
2704
4-
z CONVENTIONAL STACK (EQUATION II)

7 J-TERM STACK
SPREAD LENQTH 3656 m p?EEEq

Fig. 2. A simplified model from the North Sea showing the bias in the estimate of VT,,.

INTERVAL VELOCITY
(1000 mpc) GIAS(m/sec)
2 3 4 5 100 200
I
I - INTERFACE
1 3962

2934

2670
2676

5 3017

6 3002
3041

3434

3566
i 3603

-CONVENTIONAL STACK (EQUATION ii)

0 J-TERM STACK
pLEFiq
SPREAD LENGTH 3656m

Fig. 3. A simplified model from Alaska showing the bias in the estimate of Arms.
STACKING, RMS, AVERAGE, AND INTERVAL VELOCITIES 463

on a 3658 m (12 ooo ft) spread length, which is somewhat longer than is
normally used in practice. The stacking velocity corresponding to each inter-
face was calculated from

where m is the number of stacked traces and X5 and Ti are respectively the
offset and total traveltime corresponding to the ith trace. This value of Vi
corresponds to the slope of a least squares fit through the traveltimes in the
Tz-X2 plot. For our purpose, this is an adequate simulation of the process
of obtaining stacking velocities in practice. The bias value is indicated by a
black triangle above each line.
Model I (figure 2) is simplified from an actual model from the North Sea.
There are only small variations in interval velocity in the top four layers. Conse-
quently, the heterogeneity factor is low and the bias decreases steadily with
depth as p gets smaller. The increase in interval velocity below the fourth
interface and the larger increase below the sixth interface produce large
increases in the heterogeneity factor. Thus, at a depth of 3000 m, the bias
is much larger than at shallower horizons, say at 1500 m. This example stresses
the possibility of significant increases in the bias with depth despite the
decrease in the spread length/depth ratio.
Model 2 (figure 3) is simplified from Alaskan well data. The direct waves
travel with a high velocity through the fast permafrost layer allowing the
use of the full record from the second interface downwards. Hence, a large
bias exists at shallow levels caused by a large spread length/depth ratio. The
large bias is further emphasized by the high heterogeneity factor arising from
the sudden decrease in velocity below the permafrost layer. In this example,
the bias decreases steadily with depth, except for an insignificant rise at the
8th interface.
In the previous two examples, the stacking velocity was calculated from
exact traveltimes. These times correspond to the onset times of the reflection
wavelets on the traces of a CDP gather. In practice, random variations in
the quality of the reflection data generally cause random time shifts in the
wavelets. These variations arise from several factors such as signal-to-noise
ratio, imperfect static corrections, irregular spread geometry, etc. We shall
refer to the net effect of these factors as random “noise”. For our purpose,
it was convenient to simulate this noise by random time errors (jitter) super-
imposed on the exact reflection times of the horizon being considered. Bodoky
and Szeidovitz (1972) make a statistical investigation of these time errors.
They accept the hypothesis that these errors are normally distributed.
Figure 4 shows a histogram of stacking velocities corresponding to the
464 M. AL-CHALABI

seventh interface of model I (figure 2). This histogram represents a population


of 200 stacking velocities. In calculating each stacking velocity, a different
time jitter from the same random number distribution (of 2.3 ms standard
deviation) was superimposed on the exact reflection times for interface 7.

TRUE RMS ZERO ERROR


VELOCITY VELOCITY

!----
7005

2403

Fig. 4. A histogram
2481

of stacking
STACKINO VELOCITY
2457 2493

velocities
2499 2505

for interface
m/m

7 of model I.

The stacking velocity was then calculated by equation (II). Figure 4 shows
that the peak of the histogram is located at the zero error velocity, i.e. the
stacking velocity which would have been obtained had the data been jitter
free as in the data of figure 2. This histogram is typical of a large number of
histograms corresponding to other tested models. In these histograms, X2-tests
show that the stacking velocities are normally distributed at the 5% level
of significance, the mean falling close to the zero error velocity. When the time
errors are large, some skewness in the distribution should be expected. This
is because a negative time error increases the velocity by a greater amount
than an equal positive error would decrease it.
Thus, in our model, the difference between a given stacking velocity and
the true rms velocity is made up of the sum of a random error caused by
random noise and a systematic component which is the bias. Methods of
estimating the accuracy of stacking velocities in the presence of random noise
are given in the last section. Methods for removing the bias are given in appen-
dix 2. Semi-systematic factors such as a laterally variable weathering velocity
(see for example Schneider 1971, figure 26) are not covered by the present work.

ERRORS CAUSED BY THE BIAS


In many practical applications the stacking velocity can be safely treated
as the rms velocity. However, when the bias is large, serious errors could arise
unless allowance is made for it. The determination of interval velocity is one
example. Stacking velocities are generally used for I’8 and I’b in equation (7)
STACKING, RMS, AVERAGE, AND INTERVAL VELOCITIES 465

to calculate the velocity of a given interval. When the bias values in Vt and Vb
are of comparable magnitudes they tend to cancel out, leaving the interval
velocity relatively unaffected. In many cases, however, a large bias exists
on one interface of the interval only. In such cases, the bias in Vt will not
cancel that in Vb and the calculated interval velocity will contain a large error.
For example, if no correction is made for the bias in calculating the velocity
of the interval between the sixth and seventh interface of model I (figure 2)
the interval velocity will be overestimated by 14%. In the same way, the
velocity of the interval between the 4th and 5th interface of model z (figure 3)
will be underestimated by 9%. These errors are too large for most purposes.
They indicate the need to take into account any significant bias when calcula-
ting the interval velocity. If the bias on one interface is negligible, the error
in interval velocity is roughly of the order of B,D/h where B, is the larger
bias and h and D are respectively the thickness and average depth of the
interval. This result can be verified from equation (7).
The use of long spreads is another example where the improved accuracy
of the stacking velocity (gained by using a long spread) could be swamped
by the increased bias. An example is shown in figure 5. The data correspond
to interface 7 of model I. Random noise is simulated by a random jitter of

Fig. 5. A plot showing the increase in net error with increasing spread length when the
bias is not allowed for.
466 M. AL-CHALABI

5 ms standard deviation superimposed on the true traveltimes. The stacking


velocity is calculated by equation (II). The probable error plot refers to the
magnitude of 0.675 (I~where c9 is the standard error in the stacking velocity cal-
culated by equation (12) below. The net error is the total magnitude of the proba-
ble error and the bias. As the spread gets longer the rate of increase of the bias
increases while the rate of decrease of the probable error decreases. Therefore,
after a certain optimum value of spread length (2,000 m in this case) the net
error starts to increase. This example is representative of a large number of
practical cases. It shows that large,bias values should be corrected. Otherwise,
increasing the spread length becomes an expensive way of accumulating more
errors in what is treated as the rms velocity.

THE ACCURACY OF STACKING VELOCITIES

We now consider the accuracy of stacking velocities in the presence of


random effects which we are collectively calling random noise. We disregard
the bias and treat V, and V,,, as equivalent. We again simulate the random
noise by a random time jitter superimposed on the traveltime data.

Equations fey Estimating the Accuracy


Tests have shown that for all practical purposes the accuracy of stacking
velocities is independent of the heterogeneity factor, i.e. of the type of velocity
variations in the ground. Figure 6 summa&es this fact. The coefficient of
variation is plotted against the heterogeneity factor for a large number of
models. The stacking velocity corresponds to the bottom interface in each
model. The models were generated in such a way that the reflection times of
the bottom interface always produced the same NM0 in the outermost trace.

Fig. 6. A plot of the coefficient of variation of stacking velocity versus the heterogeneity
factor showing the scattering of points about a horizontal line.
STACKING,RMS,AVERAGE,AND INTERVALVELOCITIES 467

The variance C( of each stacking velocity was obtained numerically. It corres-


ponded to the variance of a population of zoo stacking velocities. These velocities
were generated in a way similar to that used for the velocities of figure 4. The
standard deviation of the time jitter was 5.8 ms. The points in figure 6 plot
about a horizontal line showing no correlation with the heterogeneity factor.
The horizontal line corresponds to the theoretical value of the coefficient of
variation which is calculated by equation (12) below.
These results suggest that the accuracy of stacking velocities can be estimated
from the traveltime data, without taking the actual velocity distribution into
account. If the stacking velocity is obtained through equation (II), or through
an algorithm equivalent to it, then the stacking velocity variance is given by

(14
where

CS~is the variance of the time jitter, m is the number of stacked traces, and Xi
and Tz are respectively the offset and traveltime corresponding to the ith
trace. Replacing Tt by ATi + TO, where ATt is the NM0 of the ith trace and
TO is the two-way normal incidence traveltime, equation (12) reduces to

03)

A Method for a Quick Estimation of the Accuracy of Stacking Velocities

Suppose that we replaced the ratio of the two sums in equation (13) by
the quantity K’lATz where K’ is a constant and AT, is an effective NMO, the
exact expression for which is to be determined by experiment. In this case an
approximately linear relationship will exist between this quantity and c(.

Thus, we would write (13) as


rs; cu K n; Vi/AT; 04)

where K = 0.25 K’. The form of equation (14) is quite useful. A large number
of experiments was conducted to establish whether the use of this empirical
relation was justified. The experiments consisted of searching for an expression
for AT, which would produce consistent values of K over a velocity range of
468 M. AL-CHALABI

1524-7620 m/set (5000-25 ooo ft/sec), a time range of 1.0-5.0 set and a spread
length of up to 3658 m (12 000 ft). Among the expressions tested were

( i ATf/m)‘/“, ( g AT, X,1 z X,), ( g AT: Xi/ g Xi)%, and AT,,,


d-1 s-1 i-1 4-l *=1
the latter being the NM0 corresponding to the largest offset. In all of these
expressions it was assumed that the offsets of the traces in the CDP gather
increased at equal increments. The expression
AT, = AT,,,
produced the most consistent values of K. Assuming a r&fold cover, a value
of K was determined in the above ranges on a least squares basis. This value was
0.196. Accordingly, we may write (14) as

cu2.4 6; V:/(NAT;,,) (IS)


where N is the multiplicity of cover. The standard error of the stacking velocity
is
~8 cu I.54 bt v,/ (l/F AT,,,) (16)
The discrepancy between the exact value of bg of equation (12) and that
calculated by equation (16) decreases as time and velocity increase and as
the spread length decreases. Over ranges of practical interest, the discrepancy is
usually less than 5%. In specific problems where the ranges Vs, N, TO, and
X are narrower than those quoted above, a different value for K may be
determined, if required, to obtain c9 with even greater precision. It was
therefore concluded that the use of equations (15) and (16) was justified for
gathers in which the offsets of successive traces increase at a fairly regular
interval and the offset of the innermost trace is small relative to the total
spread length. When the offset of the innermost trace is not small (say greater
than six times the increment between the offsets of successive traces) the
discrepancy could become quite large (15% or more). In this case, and in cases
where the offsets of successive traces do not increase at regular intervals, the
accuracy estimate may be obtained from equations (12) or (13).
From equation (16), the coefficient of variation is

100 G/VS cu 154 cft/(vc ATmaz) (17)


Equations (15-17) can be used to estimate the accuracy of stacking velocities
directly. A slide-rule is adequate for the purpose. Equation (17) was used to
construct a chart of the coefficient of variation as a function of AT,,, for
various values of N (figure 7). This chart simplifies further the estimation
STACKING, RMS, AVERAGE, AND INTERVAL VELOCITIES 469

of the accuracy of stacking velocities. As an example, suppose that we obtained


a velocity from a z x 12 stack (N = 24). Suppose also that the NM0 of the
outer traces was about 95 ms and that bt was estimated to be 5 ms. For AT,,,
= 95 and N = 24, we read 0.33 on the vertical axis. Therefore, the estimated
accuracy of the stacking velocity is 0.33 x 5 s 1.7 per cent.

A CHART FOR DETERMINING THE ACCURACY


OF_‘.‘, FROM THE NM0 OF MAXIMUM OFFSET

N = FOLD OF COVER
PLOTS ARE NORMALISED
FOR-+=lms

AT,,,,, (ms)
Fig. 7. Chart for determining the accuracy of V8 from the NM0 of maximum offset.

In a strict sense the above considerations apply only to stacking velocities


obtained by an algorithm totally equivalent to equation (II). However, the
principle of equation (II) is in many ways very similar to the- measure of
coherence used in other methods. Therefore, equation (12)-(17) should provide
a good estimate of the accuracy of stacking velocities obtained by other
methods. The value of ct is the quantity most prone to uncertainty. This value
is a function of the signal-to-noise ratio, of the dominant signal and noise fre-
quencies, of poor statics, etc. A fairly reasonable estimate of bt may be ob-
tained by inspecting the record of the corrected CDP gather. Methods of
estimating crt more rigorously are not central to the present work. A useful
discussion of the estimation of ot has been given by Bodoky and Szeidovitz
(1974.
470 M. AL-CHALABI

APPENDIX I

DERIVATION OF EQUATION (3)


From equations (I) and (2) we get

(3)

APPENDIX 2

METHODS OF REMOVING THE BIAS


We present below three methods for estimating the true rms velocity from
the stack data. Other methods (e.g. Brown 1969) might also be useful.

A. The Model Simulation Method


In this method, the stacking velocities corresponding to all major reflectors
are picked out. An approximate model of the ground is then constructed and
used as a basis for estimating the bias values. The procedure comprises the
following steps :
I) The stacking velocities are assumed to represent the true rms velocities
of the ground. The interval velocities are computed in equation (7). The thick-
ness of each layer is then worked out.
2) For each interface in the resulting model, reflection traveltimes corres-
ponding to all offsets used in the stack are computed by a ray-tracing method.
3) A new stacking velocity corresponding to each interface is computed
from these traveltimes. Equation (II) may be used for this purpose. An esti-
mate of the bias at each interface is obtained by subtracting the rms velocity
(the original stacking velocity) from the newly calculated stacking velocity.
4) The estimated bias values are subtracted from the original stacking
STACKING, RMS, AVERAGE, AND INTERVAL VELOCITIES 471

velocities. The resulting velocities will be usually fairly close to the true rms
velocities. <They are regarded as the rms velocities of another approximate
model. Steps 1-3 are repeated.
5) Steps 1-4 are repeated 3 or 4 times. This is usually sufficient for the
estimated rms velocities to converge to their correct values. Once the final
rms velocities have been obtained, the determination of the thickness and
velocity of each layer becomes a trivial matter.
In some cases, the estimated rms velocities oscillate at successive stages
without converging. In such cases, the last two steps can be replaced by:
4) The model defined by the original stacking velocities is regarded as a
first stage approximation. The set of stacking velocities obtained in step 3
are again treated as the rms velocities of the ground (without subtracting
the bias values). Steps 1-3 are repeated twice to generate a second stage and
a third stage approximation models. In general, the discrepancy between
the estimated and the correct bias continues to increase at successive stages.
5) For each interface, the bias may be regarded as a function of the stage
of approximation. This bias is then treated as ordinate, the stages being equally
spaced along the abscissa (figure BI). A quadratic fitted to the bias values at
the three stages of approximation may be extrapolated back to the zeroth
stage to obtain an estimate of the true bias. The quadratic extrapolation (or
some variations of it) works consistently well although the principle is arbitrary.
An example of the quadratic extrapolation is shown graphically in figure BI.
The estimated bias corresponds to interface 7 in model I.

MODEL 1
INTERFACE 7

/ QUADRATIC CURVE
#c-- EXTRAPOLATION

1 2 3
STABES

Fig. BI. An example of quadratic extrapolation in model simulation.


Geophysical Prospecting, \‘ol. zz 31
472 M. AL-CHALABI

The method of model simulation is simple and requires very limited computer
time. The bias estimates are relatively insensitive to small errors in the stacking
velocity (up to about 50 m/set). Such errors are carried through to the estimated
rms velocity. The main difficulty with this method lies in picking out the
major reflectors (interfaces) correctly.

B. The Method of Shifting Stack


The decrease in the bias with decreasing offset [decreasing ray parameter 9)
was discussed in the text. At zero offset (p = o) the bias vanishes. The principle
of the method is to estimate the velocity at zero offset by extrapolating back
from a set of stacking velocities. Each of these stacking velocities is obtained
from a different set of traces from the trace gather. Suppose we had data for
a z4-fold stack. We can stack the inner 12 traces (traces 1-12) and obtain a
stacking velocity, then stack the inner 14 traces (traces 1-14) and obtain another
stacking velocity, then stack the inner 16 traces and so on until all 24 traces
had been stacked. We should then end up with 7 different stacking velocities,
increasing as the maximum offset increases. The chosen traces could have also
been varied in a different manner, e.g. we could have stacked traces 1-14, then
3-16, then 5-18, etc. Let us define an effective offset as being a single quantity
that represents the offsets of all of the stacked traces. A convenient but
arbitrary form of the effective offset is

where Xj is the offset of the ith trace and m is the number of stacked traces.
We can plot the seven velocities versus their effective offset and extrapolate
back to the zero offset velocity. A low order polynomial fitted to the points
may be used in the extrapolation. In fact, on a V, versus Xz plot, the points
would fall almost on a straight line. The near-linear relationship between V8
and Xi is not an obvious one. It was deduced from a large number of numerical
experiments. It can also be justified analytically; The V, versus Xi plot is re-
commended for the purpose of this ,method. The I’,” - Xi relationship was also
found to be approximately linear, but to a slightly lesser extent than the V8
- Xi relationship. Figure B2 shows an example of I/‘, - Xz and V, - X, plots.
The example corresponds to interface 7 of model I.
In practice, the V, - Xi points will generally scatter about a straight
line because of the presence of noise in the stacked data. The rms velocity
may be estimated from the intercept of a straight line fitting the points on a
weighted least squares basis, the weights being proportional to the inverse
of the variance of each velocity.
STACKING, RMS, AVERAGE, AND INTERVAL VELOCITIES 473

x: (EFFECTIVE OFFSET)’

0 1’0 PO 30 40 50 (km)’
,

I.0 1’5 2.0 km


X, EFFECTIVE OFFSET

Fig. Bz. Estimation of the true YWZS velocity by a linear extrapolation from the V8
versus Xl plot.

An important feature of this method is that the estimate of V,,, at each


interface is independent of errors made at other interfaces. The principle of
the method can be extended to other problems in which the stacking velocity
varies systematically with offset such as problems involving dipping layers.
C. The Three-Term Series Truncation
Methods of obtaining stacking velocities are essentially based on a two-term
truncation of the series
T;=C,+C,X;+C,X,4+CqX;+... VW
where X6 and Tg are respectively the offset and true reflection time corres-
ponding to the ith trace, C,(j = I, 2, . . ., co) is the coefficient of the jth term
and C, = I/V& (T aner and Koehler 1969). The bias in the estimate of Vrms
arises as a result of this truncation. One possibility for reducing the bias would
be to base the stack on a larger number of terms. The three-term truncation
of equation (BI) is particularly useful because it usually gives a very good
approximation to the true traveltime (Al-Chalabi 1973). If a stack is carried
out according to this three-term truncation the stacking velocity obtained from
vs" = I/C,
should be very close to the true rms velocity. This approach was tested on a
number of models including those of figures z and 3. Stacking velocities
were obtained by using a least squares criterion similar to that used in deriving
equation (II). Results for models I and z are shown in figures z and 3. The bias
values corresponding to these three-term stacking velocities are indicated
by white triangles below each line. They demonstrate the drastic reduction
474 M. AL-CHALABI

El?0 ERROR

MODEL 1
INTERFACE 8

TRUNCATION
4 = 2.3ms

1
6300
STACKINB VELOCITY

I 2530 2560 2591

J-TERM TRUNCATION

8100 6200 6300 6400 fthec


STACKING VELOCITY
2469 2499 2530 2560 m/set

Fig. B3. A comparison between the histograms of two-term and three-term stacking
velocities. Despite the high variance of the three-term velocities there is here a clear
advantage in stacking according to three terms.

I MODEL 2

I
I
I INTERFACE 7

i
I
i I
10 20 30 40 mhec
RANGE WITHIN TRUE RMS VELOCITY
Fig. B4. Plots of the probability of being within a given range of the true YWLSvelocity
for a two-term and a three-term stack.
STACKING, RMS, AVERAGE, AND INTERVAL VELOCITIES 475

in the bias achieved by a three-term stack. However, numerical experiments


in the presence of simulated random noise indicate that the variance of a
velocity obtained from a three-term stack is generally about IO times greater
than the variance of the corresponding velocity obtained by the conventional
two-term stack. Figure B3 shows a comparison between the histograms of
stacking velocities produced by a two-term truncation and a three-term
truncation. The histograms correspond to the eighth interface in model I.
They were generated in the same way as the histogram of figure 4. The standard
error of the added time jitter was 2.3 ms. As in all of the tested models, the
three-term stacking velocities are centred about the true rms velocity but
their variance is much larger than in the case of the two-term velocities.
In this particular example, there is some advantage in stacking according to
three-terms despite the large variance; the three term stack produces a closer es-
timate of the true rms velocity than a two term stack. Usually, however, the im-
provement gained by stacking according to three terms is offset by the loss
of accuracy. An example is shown in figure B4. In this example, the bias and
the random errors are combined to give the probability of being within a certain
range of the true rms velocity. The plots correspond to the seventh interface of
model 2. The standard deviation of the added jitter is 1.2 ms. Although the jitter
is low, the probability of being within a close range of the true rms velocity
reverses in favour of a two-term stack at a range of only 14 m/set.
In conclusion, the method of three-term stack is not recommended for
general use. It is presented here mainly because it arises as an interesting
possibility when considering the time-distance relationship of equation (BI).

ACKNOWLEDGEMENTS
I am grateful to Dr Peter O’Brien for reading the manuscript and to him
and Dr Andrew Lucas for some interesting discussions. I thank the Chairman
and Board of Directors of the British Petroleum Company Ltd. for their
permission to publish this work.

REFERENCES
AL-CHALABI, M., 1973, Series approximation in velocity and traveltime computation.
Geophysical Prospecting 21, 783-795.
BODOKY, T., and SZEIDOVITZ, Zs., 1972, The effect of normal correction errors on the
stacking of common-depth point traces. Geophysical Transactions of the Hungarian
Geophysical Institute Roland Eotvijs, 20 (3-4). 47-57.
BROWN, R. J. S., 1969, Normal-moveout and velocity relations for flat and dipping beds
and for long offsets. Geophysics 34, rSo-195.
DIX, C. H., 1955, Seismic velocities from surface measurements. Geophysics 20, 68-86.
SCHNEIDER, W. A., 1971, Developments in seismic data processing and analysis (196%
1970). Geophysics 36, 1043-1073.
TANER, M. T., and KOEHLER, F., 1969, Velocity spectra-Digital computer derivation of
velocity functions. Geophysics 34, 859-881.

You might also like