Interval Velocity Model

Interval Velocity Model
0
1934
1
DEPTH X1000 M
2376
Stacking velocities are

used to derive interval
velocities for depth
conversion
5
DISTANCE

The stacking velocities supplied to the interpreter by
the processing contractor are usually far from ideal
for deriving interval velocities.
Why are they usually not ideal?
10.1
Exercise 10.1
A digital version of this exercise is available as Ex 10.1.xls.

Given this stacking velocity function (the nearest to the well A location) find: 1. the interval stacking velocities,
2. the thicknesses of the units
3. the average velocity to each of the interfaces.
Compare your results to the results of Exercise 2.1 at the end of the Definitions 1 page
2.10 (File Ex 2.1.xls)
What are the possible sources of error in your resulting depth model?
VIS =
tbase - ttop
SP 78355
Time
msec
0
50
290
489
840
900
945
1030
1330
1586
1740
2070
VS,base2 tbase - VS, top2 ttop
VS
m/s
1480
1480
1710
1810
2140
2190
2278
2430
2810
2985
3260
3570
10.2
1/2
n Optical stacking velocity

o Refraction analysis
p Stacking velocity analysis
q Dip Corrected stacking velocities
r Interval Stacking velocities
s Interval velocity analysis
t Migration velocities
u Depth migration velocity analysis
v Residual velocity analysis
w Tomographic velocity analysis
We need the best RMS interval velocity model we can get

out of the stacking velocity data.
We may need to dip correct our stacking velocities.
We may need a bias correction.
We may need to interpolate the RMS velocities to the
times of our picked horizons.
After deriving the interval RMS velocities we will almost
certainly need to filter and smooth.
10.3
Dip Correction
Stacking velocity values can be dip corrected without recourse to

reprocessing with DMO.
Levins Dip Correction for 2D data is VSapp VS / Cos where
is the apparent dip, giving: VS VSapp x dx / [dx2 + (dt. VSapp )2].
Distance
mic
g seis
Dippin ent
ev
dx
Time
dt
Cos = dx / [dx2 + (dt. VA )2]

dx / [dx2 + (dt. VSapp )2]
Dip Correction
VS VSapp x dx / [dx2 + (dt. VSapp )2].

Given VSapp = 2476 m/s and a time dip of 170 msec / 240 m
find VS (remember dT = 2.dt and you need the time in seconds).
VSapp = 2476 m/s
dT =170 msec
dx = 240 m
10.4
Dip Correction
Levins dip correction VSapp VS / Cos works well for a

single dipping reflector on a dip line.
It also works for the case of a dip line with parallel dips in
the subsurface.
Only use in areas of steep dip (>25) on legacy data
Dip Correction
In the case of multiple

arbitrary dips Levins
formulae fails to make an
adequate correction. In fact,
a dip correction cannot be
made directly and one has to
resort to modelling the
observed stacking velocity.
V2
NMO
1
Tocos2ro
k=1
ij
j
V2t
k
rj
tn
n
k-1
ro
cos2ij
k
j=0 cos2r
j
Vn
n
with
To = tk
1
Everett and Hubral (1974) offer a recursive procedure for modelling VS

whereas Shah and Levin (1974) did not.
10.5
Bias Correction
When: maximum offset is greater than 1500m or

there are large contrasts in formation velocities
then the stacking velocity will not be a good approximation to
the RMS velocity.
We need to correct for this bias before we compute the
interval velocity.
There are four ways we can do this; the first three are
numerically the same.
Bias Correction
Method 1
This approach is used in velocity modelling software and
returns the Interval RMS Velocities.
M. Al-Chalabi gives:
Bias = VS - VRMS = Xmax2(VRM44-VRMS4)/(8To2VRMS5)
where
VRM44 = ( VIm4tm)/to
for m = 1 to current layer.
VS - VRMS - Xmax2(VRM44-VRMS4)/(8To2VRMS5) = error

Using the Dix derived interval velocities to calculate the VRMS
and VRM4 terms the modulus of the error is minimised by
changing the values of the interval velocities. (This could be
done in Excel using the Solver tool.)
10.6
Bias Correction
Method 1
Note that the equation is based on the three term Taylor series
expansion and as such will be in error because of the
neglected terms. However, since the expression is an infinite
series, including additional terms does not measurably
increase the accuracy of the solution as demonstrated by alChalabi (1974) when he first introduced the concept.
The XMAX term in the equation is the maximum offset, this
varies with the reflection time according to the mute that was
used before velocity analysis, not the mute that was applied to
remove NMO stretch after the NMO correction.
Bias Correction
Method 1
The file Bias.xls illustrates this method on worksheet exercise 1.
10.7
Bias Correction
Method 2
As before
VS - VRMS - Xmax2(VRM44-VRMS4)/(8To2VRMS5) = error
Using the Dix derived interval velocities (VIS) to calculate the
VRMS and VRM4 terms the error is made as small as possible,
recursively, by changing the values of VIS layer by layer.
A second tool in Excel, Goal Seek, can be used to do this.
Applying it to the second layer of our model we obtain VI =
3990 m/s compared to the original model value of 4000 m/s.
Bias Correction
Method 2
RMS & Stacking Velocity

Velocity/1000 m /s
Interval Velocity m /sec

0
2000
4000
1.4
6000
2.2
2.6
0
500
500
1000
1000
1500
1500
2000
2000
Depth m
Depth m
1.8
2500
2500
3000
3000
3500
3500
4000
4000
4500
4500
5000
5000
RMS Velocity
Theoretical Stacking Velocity
Original Model
10.8
Bias Correction
Method 2
Bias
Depth Conversion Error

Error in m
Bias - Percent
0.2 0.4 0.6 0.8
-5
1.2 1.4 1.6
1000
1000
2000
2000
Depth m
Depth m
-0.2 0
3000
4000
10
15
20
25
30
35
3000
4000
5000
5000
6000
6000
Bias Correction
Bias Correction
Bias Correction
Method 2
The file Bias.xls illustrates this method on worksheet exercise 2.
10.9
Bias Correction
Method 3
The bias correction can be estimated by modelling, an
approach that was used before optimisation was available in
spreadsheets:
Assume
VS=VRMS
Calculate
VIrms
Build model
using VIrms
for VI
Compute VS
for new
model*
Modelled
bias is
estimate.
* To compute the stacking velocity for the model it is necessary to:

Model Tx2 using the T0, x2 and x4 terms of the moveout
equation.
Plot Tx2 against x2
Fit a straight trend line.
VS is the inverse of the slope of the trendline.
Bias Correction
Method 3
Assume
VS=VRMS
Calculate
VIrms
Build model
using VIrms
for VI
Compute VS
for new
model*
Modelled
bias is
estimate.
For our 2 layer model we had: T0

VS
1.5
2000
2.0
2699
assuming the above are RMS velocities, then
T0
VRMS
VIrms
1.5
2000
2000
2.0
2699
4140
10.10
Bias Correction
Method 3
Assume
VS=VRMS
Calculate
VIrms
Build model
using VIrms
for VI
Compute VS
for new
model*
Modelled
bias is
estimate.
* To compute the stacking velocity for

the new model it is necessary to:
Model Tx2 using the T0, x2 and x4
terms of the moveout equation.
Plot Tx2 against x2
Fit a straight trend line.
VS is the inverse of the slope of the
trendline.
New Model
To = 1.5 sec
VI = 2000 m/s
To = 2 sec
VIRMS = 4140 m/s
Bias Correction
Method 3
Assume
VS=VRMS
Build model
using VIrms
for VI
Calculate
VIrms
Bias Correction by Modelling
y = 0.1316x + 4.0051
5.5
Compute VS
for new
model*
Modelled
bias is
estimate.
Trendline slope = 1/VS2

Slope = 0.1316 x 10-6
So for new model: VS = 2756 m/sec
and VRMS = 2699 m/sec
T squared
4.5
4
0
Modelled offsets
X squared
10
Linear (Modelled offsets)
10.11
Bias for new model is 57 m/s

This is estimate of bias in
original data.
Bias Correction
Method 3
Assume
VS=VRMS
Calculate
VIrms
Build model
using VIrms
for VI
Compute VS
for new
model*
Modelled
bias is
estimate.
For our 2 layer model:

VS = 2699 m/sec and the bias is estimated to be 57 m/sec.
Estimate the interval velocity of the second layer after applying
a bias correction.
Use VI = [(V2RMS2t2 - V2RMS1t1)/(t2 - t1)]1/2
Bias Correction
Method 4
Another method is suggested by the fact that VS tends to VNMO
at small offsets.
This method has to be implemented during processing or
reprocessing and was popular with contractors in the
1980s for determining VRMS.
The stacking velocity is determined at a number of different
offsets, a cross plot generated, and extrapolated back to zero
offset.
10.12
Bias Correction
Method 4
Bias Correction
y = 6E-06x + 2646
Applying the technique to

the T - X values for our
model:
2700
2695
2690
2685
Stacking Velocity
For the second layer of our

model we obtain 2646
m/sec giving VI = 4000
m/s.
2680
2675
2670
2665
2660
2655
0
2000000
4000000
6000000
8000000
10000000
Offset squared
Interpolation
Unless the stacking velocities have been picked to be

consistent with the horizons that define the boundaries in the
macrovelocity model there will be a need to interpolate.
The way in which we interpolate stacking velocities affects
the interval velocities derived from the interpolated values.
10.13
Interpolation
RMS Velocity
1000
2000
0.2
500
0.4
1000
0.6
1500
0.8
2000
4000
2000
2500
1.2
3000
1.4
3500
1.6
4000
Linear interpolation VRMS = 2273 m/sec
Interpolation
RMS Velocity
1000
2000
Interval Velocity
3000
2000
0.2
500
0.4
1000
0.6
1500
Depth
One Way Time
Interpolation needs
to take account of
the interval velocity
model. It should
either leave the
interval velocity
model obtained
from the
observations
unaltered or use a
proiri interval
velocity information
from well control.
Interval Velocity
3000
Depth
One Way Time
Linear
interpolation is
usually used
(being the default
in software) but
this is not
necessarily
correct since it
modifies the
interval velocity
model given by
the observations.
0.8
2000
2500
1.2
3000
1.4
3500
1.6
4000
Constant interval velocity

interpolation VRMS = 2338 m/sec
10.14
4000
Exercise 10.2
You have stacking velocities picked from Horizon based Velocity Analysis for two horizons
and find that you need to interpolate the stacking velocities at a constant time value which
just happens to lie mid way between the two sets of picks. The software you are using
offers you a choice of interpolation methods to use: linear in V, linear in V2, linear in V2t,
linear in sqrt(V2t) and with a user supplied compaction factor. You dont have a value for K
and decide to interpolate between your picked horizons so that the interval velocity remains
unchanged.
Which method do you choose?
Confirm your answer by interpolating between
T secs
1.315
1.885
VS m/s
1997
2548
at a time of 1.6 secs.
If you had chosen the other methods what would be the error in your interpolated value for
VS?
10.15
Frequently we find that the RMS velocity profiles or maps

are very noisy and it is quite common to find these being
smoothed before being used to compute interval velocities.
This may lead to erroneous interval RMS velocities.
It is preferable to derive the interval RMS velocities and then
smooth the data.
Interval RMS Velocity
Having corrected as far as possible the stacking velocities so

that they better approximate the RMS velocities and having
interpolated to the horizons of interest we can now compute
the interval RMS velocities.
Whether we use stacking velocities or time migration RMS
velocities we use Dix inverse RMS equation to produce an
estimate of the interval RMS velocities known as interval
stacking velocities or interval RMS velocities.
VIS =
VS,base2 tbase - VS,top2 ttop

tbase - ttop
10.16
1/2
Using the three time-velocity pairs derived in the exercise at the

end of the last section (page 9.44) compute the two interval
RMS velocities.
VIS =
VS,base2 tbase - VS,top2 ttop

tbase - ttop
1/2
The class results are:

Upper interval
Lower interval
10.17
Mean = 5445
S. D. = 236
Mean = 5020
S. D. = 294
10.18
Sensitivity
The error in interval velocity is proportional to the error in
RMS velocity.
The error in interval velocity is proportional to the travel
time error.
The error in interval velocity is proportional to the ratio of
RMS to interval velocity.
Interval velocity error is proportional to the ratio of time
thickness to depth.
See Hajnal and Sereda, Maximum uncertainty of interval velocity estimation, Geophysics November 1981.
Rule
Rule of thumb:
Try to avoid applying Dix equation to intervals
thinner than 200msec TWT.
Field Statics
If field statics are applied to a datum other than the average
of the topography / shot receivers, the interval stacking
velocities will be distorted.
From our 2 layer model we had VS = 1978 and 2680 m/s
after a 40 msec static. This results in an interval stacking
velocity of 4153 m/s compared to the 4140 m/s without the
static.
10.19
Smoothing
No matter how much care we have taken we will almost

certainly have a noisy interval RMS velocity map so what
can we do to reduce the noise?
Smoothing
No matter how much care is

taken in picking and correcting
the stacking velocity values,
resulting interval RMS velocity
maps and profiles will always
be noisy and contain artefacts.
These data usually need to be
edited or smoothed before
being incorporated into a
macrovelocity model.
8 miles
Data courtesy of Amoco
10.20
Smoothing
High frequency noise
Artefacts caused
by a fault zone
We may not always be as
fortunate in having so little noise.
8 miles
Smoothing
Editing
The use of Dix equation on thin intervals is unstable.
Consequently geologically unreasonable interval velocities
often result. These can be edited, either removed or replaced
by geologically reasonable values.
10.21
Smoothing
Noise
Most commonly our first
interval RMS velocity
maps will look like this,
noisy.
The effect is caused by
noise in the data and
misties between different
velocity profiles.
It is advisable to filter out
the worst points before
smoothing.
Smoothing
Filtering
There are several high
frequency features in this
map that are removed by
filtering.
The filter used in this
example was a robust
filter set at one standard
deviation.
Data courtesy of Norsk Hydro
10.22
Smoothing
Filtering
This map shows the result
of the robust filtering.
Several of the high
frequency anomalies have
been removed.
Notice the fenceline
anomalies between the
green and black surveys
in the centre.
Network error adjustment
algorithms minimise such
mistie errors.
Data courtesy of Norsk Hydro
Smoothing
Smoothing
Almost always we find that the interval RMS velocity profiles
or maps are noisy even after editing and filtering and it is
quite common to smooth them.
Techniques which are used, depending on availability of
software, include:
spatial averaging (mean and median)
polynomial fitting,
locally weighted regression (LOESS),
kriging.
10.23
Smoothing
Spatial Averaging
Smoothing techniques are applied as spatial running
averages. Commonly mean and median filters are available.
In mean filtering the smoothed value is the arithmetic
average of the data points in the filter window. In mean
filtering weights may be applied to the data values - usually
inverse with distance from the centre of the filter - so that
values closer to the centre of the filter have more influence
on the result.
In median filtering the data are rank ordered and the middle
value taken.
Smoothing
Spatial Averaging
Circle of inclusion
Data Points
Output value
may be at the
centre of the
circle
+
Centroid - average
location where
average value
may be posted
Radius of inclusion chosen interpretively

to smooth noise but not
geologically meaningful
anomalies.
10.24
Smoothing
Spatial Averaging
The radius of inclusion may be determined by testing a
number of different radii centred on a well location.
The resulting smoothed velocities (with calibration if
necessary) are used to make a depth conversion which is
compared to the well data:-
ERROR
The plot of depth conversion error

against radius of inclusion
demonstrates a well behaved
minimum when the radius is about
equal to the maximum offset.
RADIUS
Smoothing
Spatial Averaging
Alternatives are to use a square, or
a rectangle, with dimensions of the
order of twice the maximum offset.
10.25
Smoothing
Spatial Averaging
C.I. = 500 ft/s
C.I. = 500 ft/s
Smoothing
Surface Fitting
Polynomial (trend surface) fitting can be used to smooth noisy
maps. Care is needed with smoothing techniques so that
geologically meaningful gradients (such as those seen across fault
zones as in the example below) are not smoothed out too much.
10.26
Smoothing
Smoothing
LOESS filtering (locally weighted regression) is available in
some software as a smoothing tool. A polynomial surface is
fitted to the data points in a window and the predicted value
at the centre of the window out-put. A hybrid approach
based on mean and median filtering and trend surface fitting.
Kriging can also be used for smoothing.
Various forms of gridding also incorporate smoothing

functionality, smoothing being some form of localized
averaging over say 9 or 25 grid nodes.
Smoothing
Normalization
Interval velocity is often proportional to the depth of burial
or approximately proportional to reflection time.
So that none of the variation due to this dependence is
smoothed out interval RMS velocities ought to be
normalized before smoothing when depth or time ranges
are large. The remaining variations will be due to:
the noise we wish to eliminate.
lithology (which generally varies slowly)
variation in bed thickness
pressure cells (overpressure)
tectonic inversion
10.27
Smoothing
Normalization
When Interval Velocity is a function of depth
VI = fn(z)
it may be normalized with respect to depth.
VNI = VIobs - fn(zobs) + fn(zN)

The depth may be burial depth, depth to mid-point, etc. The
normalization depth zN is arbitrarily chosen.
The next slide illustrates the process graphically.
Smoothing
Normalization
Interval Stacking Velocity (ft/sec)
16000
VNI = VIobs - fn(zobs) + fn(zN)

VNI
Variation
due to depth,
fn(z), this
may or may
not be linear.
14000
12000
10000
8000
2000
4000
6000
10.28
zN
8000
10000
Mid point
depth (ft)
Smoothing
Normalization
Velocity - time cross
plots before and
after normalisation /
smoothing / removal
of normalisation
process showing
how smoothing has
reduced the scatter.
Smoothing
Carter (1993) advises

that data points from
areas of tectonic uplift
need to be corrected for
the uplift before
normalization - and the
effect of the uplift
restored after smoothing.
This presupposes that
we know the amount of
uplift.
Interval Velocity (m/sec)
Normalization
Correction?
Or estimate of
uplift?
2000
1900
1800
1700
1600
0
100
200
300
400
500
One Way Time (msec)

10.29
Smoothing
Normalization
A Triassic interval
velocity map from
seismic data showing
high frequency
anomalies that are
related to the geology.
Smoothing this map
would attenuate these
variations which are
related to structure.
From Carter 1993 Leading Edge
Smoothing
Normalization
The depth map to the
top of the Triassic. Many
of the variations are
spatially similar to the
variations in the Triassic
interval velocity map.
This suggests that the
major variations in
interval velocity are
related to depth of burial
and should not be
smoothed.
10.30
Smoothing
Normalization
A log-log plot of interval
velocity against depth to
mid-point for the Triassic
shows a strong linear
relationship.
Smoothing
Normalization
The depth normalized
interval velocity map for
the Triassic. Variations
are due to: noise (high frequency),
probably lithology
(regional gradients)
possibly tectonics
(steep linear gradients).
10.31
Smoothing
Normalization
Theoretically, the approach is complicated for functions
based on instantaneous velocity, because of the the
expression for interval velocity.
From Vi = Vo + kz the depth is given by:-
zn = Vo(eK t - 1)/K + zn-1 eK t

and the interval velocity is then
VI = (zn - zn-1) / t.
The formulae are equally complicated from Evjens and
Fausts formulae.
Smoothing
The smoothed interval RMS

velocities may be used in
depth conversion, event
migration or map migration of
interpreted time horizons to
generate a first
From a Schlumberger GeoQuest brochure
10.32
Quality Control
Work Flow
Interval
RMS
Velocities
Dix
INTERPRETATION
Bias
correction
DIP
CORRECTED
STACKING
VELOCITIES
INTERPOLATION
DIP CORRECTION
STACKING
VELOCITIES
RMS
Velocities
HORIZON
CONSISTENT
STACKING
VELOCITIES
Quality Control
The First Macrovelocity Model

The general scheme for building the first model is as follows:
Interval
RMS
Velocities
INTERPRETATION
Depth
Conversion
(or Ray
Migration)
Dix
Compare
& Adjust
Model
RMS
VELOCITIES
STACKING
VELOCITIES
Forward
Model
10.33
INITIAL
MACROVELOCITY
MODEL
Quality Control
The First Macrovelocity Model

The original data have been corrected, edited,
smoothed and inverted in the process of deriving our
What techniques are you familiar with for ensuring that
the macrovelocity model obtained from the seismic data
is consistent with the data themselves?
Quality Control
Stacking Value Matching

This technique ensures a macrovelocity model consistent with
the available data. It is suitable when original seismic tapes are
not available. The time picks and model velocities are iteratively
adjusted so that stacking velocity hyperbolae and modelled
moveouts are in acceptable agreement.
INITIAL
MACROVELOCITY
MODEL
OBSERVED
STACKING
VELOCITIES
MODEL OF
GATHER
To
Model
(Offset
Raytrace)
Time Horizons
and Modelled
Stacking
Velocities
Stacking
Velocity
Hyperbola
Model
Moveout
OPTIMIZE MODEL
10.34
REFINED
MACROVELOCITY
MODEL
Quality Control
Travel Time Inversion - Tomography

This optimisation technique ensures that the macrovelocity
model built from smoothed interval RMS velocities is still
consistent with all of the original data.
This technique is often implemented under the name of
travel time inversion. The optimisation of the offset raytraced model is a form of tomography.
The initial macrovelocity model will only be as good as the
original stacking velocity values, if they are inaccurate then
the optimised model will reflect the inaccuracy - not remove
the inaccuracy.
Summary
Before stacking velocities can be used to determine interval

RMS velocities they may require a dip correction.
A bias correction may be required if the maximum offsets
exceed 1500m or if there are large velocity contrasts between
formations.
It may be necessary to interpolate the RMS velocities. Any
interpolation should seek to take account of the interval velocity
model.
10.35
Summary
Interval RMS velocity maps may be derived from stacking

velocities by Dix equation.
The resultant maps will require editing and smoothing.
If the range of values is large, or changes abruptly,
normalization may be desirable prior to smoothing.
A number of techniques are available for smoothing.
After smoothing the resultant depth model ought to be
optimised with respect to the picked reflections and their
associated stacking velocities.
10.36

Interval Velocity Model

Uploaded by

Copyright:

Available Formats

Interval Velocity Model

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Interval Velocity Model

Uploaded by

Copyright:

Available Formats

Interval Velocity Model

Stacking velocities are