Sunteți pe pagina 1din 11

1

Chapter 4
Velocity analysis
Introduction
The objective of velocity analysis is to determine the seismic velocities of layers in
the subsurface.
Seismic velocities are used in many processing and interpretation stages such as:
Spherical divergence correction
NMO correction and stacking
Interval velocity determination
Migration
Time to depth conversion.
There are different types of seismic velocities such as: the NMO, stacking, RMS,
average, interval (Dix), phase, group, and migration velocities.
The velocities that can be derived reliably from T-X data are the NMO and stacking
velocities.
The T-X curve of a single constant-velocity horizontal layer is a perfect hyperbola
given by:
T
2
(X) = T
2
(0) + X
2
/V
2
, (4.1)
where T(X) is the 2-way traveltime at offset X, T(0) is the 2-way traveltime at zero
offset, and V is the layer velocity.
In a series of plane horizontal constant-velocity layers, the exact offset (X
N
) and two-
way traveltime (T
N
) to the N
th
layer are given by the following parametric equations:
2

N
i
i
i i
N
pV
H pV
X
1
2
) ( 1
2 ,
and

N
i
i i
i
N
pV V
H
T
1
2
) ( 1
2 ,
where V
i
, and H
i
are the interval velocity and thickness of the i
th
layer; and p = dT/dX
is the parameter of the ray received at X = X
N
with T = T
N
.
However, we can express the exact T
N
in terms of X
N
in a Taylors series expansion
about X
N
= 0:
T
N
2
(X) = C
0
+ C
1
X
N
2
+ C
2
X
N
4
+.... (4.2)
The series (4.2) is an exact representation of the exact T-X curve of the N-th reflector
under two conditions:
1. X = 0.
2. Infinite number of terms is used in the series.
Putting X > 0 in equation (4.2) will result in losing some accuracy in the
representation of series (4.2) of the exact T-X curve.
Therefore, we can only use this series as long as we keep close to X = 0. A practical
measure is X/Z 1.
If infinite number of terms are used in equation (4.2), then:
2
2
1
2
1
0
) 0 (
2
N
N
i
i
N
i i
i
T t
V
H
C
1
]
1


1
]
1



, (4.2a)
and
2
1
2
1
1
/ 1
N RMS
N
i
i i
N
i
i
V
t V
t
C
1
1
1
1
]
1

. (4.2b)
3
where t
i
is the interval zero-offset two-way traveltime of the i-th layer and V
RMSN
is
the RMS velocity of the N-th reflector.
Since we cannot use infinite number of terms practically, we have to truncate
equation (4.2). This will introduce an error, and C
1
will deviate from 1/V
RMS
2
.
In general, the less terms left in the series, the greater is the error.
Truncating equation (4.2) to two terms, we get the following hyperbolic relation
between T and X:
T
N
2
(X) T
N
2
(0) + X
2
/V
RMSN
2
, (4.3a)
Note the two conditions above when estimating V
RMSN
from equation (4.3a).
The RMS velocity can be defined in terms of the true T-X curve as the square root of
the reciprocal of the coefficient of the X
2
term in the series approximation of the
exact T
2
-X
2
curve of multiple layers:
1
1
C
V
RMS

. (4.3b)
It can also be shown that V
RMS
is equal to the square root of the reciprocal of the
slope of the tangent to the exact T
2
-X
2
curve at X = 0. That is:
0
2 2
] / [
1

X
RMS
dX dT
V . (4.3c)
The RMS velocity V
RMSN
to the N-th reflector can also be defined in terms of the
properties of subsurface layers as (see equation (4.2b)) :
2 / 1
1
1
2

,
_

N
i
i
N
i
i i
RMSN
t
t V
V
(4.4)
4
where V
i
is the interval velocity and t
i
is the interval zero-offset two-way traveltime
of the i-th layer, and N is the number of layers.
The stacking velocity V
s
is found from the T-X data by fitting a best-fit hyperbola to
the true (nonhyperbolic) T-X curve, which takes the form:
T
2
(X) = T
2
(0) + X
2
/V
s
2
, (4.5)
where T(X) is the 2-way traveltime at offset X and T(0) is the 2-way traveltime at
zero offset.
Figure.
Figure.
Note that when fitting a hyperbola to the true T-X curve, we always choose the
constant term to be equal to T
2
(0) because it is a known condition and must be
satisfied. Therefore, this is a constrained fit.
The normal moveout (NMO) is the time difference between traveltime at a given
offset and zero offset. At small offsets, the NMO is approximated by:
T
NMO
(X) X
2
/[2V
NMO
2
T(0)]. (4.6)
The NMO velocity V
NMO
is found from the T-X data by searching for the velocity
that will best NMO-correct a certain reflection (i.e., makes it perfectly horizontal).
Figure.
The stacking and NMO velocities cannot be related directly to the properties of
subsurface layers.
However, at small offsets (offset depth), the NMO and stacking velocities are
approximately equal to the RMS velocity.
5
Since V
RMS
is related to layers properties (equation 4.4), the equivalence of the RMS,
NMO, and stacking velocities at small offsets can be used to relate V
S
and V
NMO
to
layer properties directly through Dix formula defined next.
The interval (Dix) velocity (V
N
) of the N-th layer can be calculated from the RMS (or
stacking and NMO at small offsets) velocities as follows:
where V
RMSN-1
and V
RMSN
are the RMS velocities to the top and bottom of the N-th
layer, and T(0)
N-1
and T(0)
N
are the zero-offset traveltimes to the top and bottom of
the N-th layer.
The NMO and stacking velocities are generally not equal because they are calculated
using different unrelated methods. However, in practice, they are considered equal if
only small offsets are used.
See this Excel Sheet for the effects of truncation and offset on V
RMS
calculation.
Velocity determination in practice
(1) The T
2
-X
2
method
If we approximate the true T-X curve by a hyperbola of the form given by equation
(4.5), then a plot of T
2
versus X
2
will give a line whose slope and intercept are 1/V
s
2
and T
2
(0), respectively.
Hence, we can use this equation to find the stacking velocity V
s
from the slope of the
best-fit line to the true T-X curve. This is called the T
2
-X
2
method.
2 / 1
1
1
2
1
2
) 0 ( ) 0 (
) 0 ( ) 0 (
1
]
1


N N
N RMSN N RMSN
N
T T
T V T V
V
(4.7)
6
The T
2
-X
2
method is not practical to use for common seismic exploration datasets
because it needs picking of T(X) at every offset, which has the following problems:
1. It is time consuming if done by humans because of the huge datasets commonly
encountered in seismic exploration.
2. It is prone to errors if done by machine especially in noisy datasets.
Therefore, it is mainly used with small datasets of relatively high S/N ratio (e. g.,
experimental or synthetic data).
(2) Constant-velocity stacks (CVS)
This method attempts to find the NMO velocity to each reflector.
This method has two variations:
(a) The first variation consists of the following steps:
1. A selected CMP gather is repeatedly NMO-corrected using a range of
constant velocity values.
2. The NMO-corrected gathers are displayed beside each other in panels.
3. Following a certain event, as it is NMO-corrected using different velocities;
we choose the velocity that best flattens (i.e., makes it perfectly horizontal)
the event as V
NMO
of that event.
4. Proceeding in this way for the other events of interest in the CMP gather, we
can build up a velocity function that is appropriate for the NMO correction of
this CMP gather.
5. Choose another CMP gather and repeat steps 1-4.
6. Interpolate the NMO velocities linearly for CMPs that were not analyzed.
7
(b) The second variation consists of the following steps:
1. A small portion of the line (consisting of several adjacent CMP gathers) is
repeatedly NMO-corrected and stacked using a range of constant velocity
values.
2. The constant-velocity stacks are displayed beside each other in panels.
3. Following a certain event, as it is stacked using different velocities; we choose
the velocity that produces the most laterally continuous stack of the event as
V
NMO
of that event.
4. Proceeding in this way for the other events of interest in the panel, we can
build up a velocity function that is appropriate for this portion of the line.
5. Choose another portion with different (or overlapping) CMP gathers and
repeat steps 1-4.
6. Interpolate the NMO velocities linearly for portions that were not analyzed.
The velocities found using variation (b) are often called stacking velocities (V
s
)
because a stacked section is used.
Important parameters to consider when using the CVS method are the minimum,
maximum, and increment in the trial NMO velocities.
The CVS method is especially useful in areas with complex structures.
8
(3) The velocity spectrum
This method attempts to find the stacking velocity to each reflector.
It maps the T-X data of a single CMP gather onto the velocity-spectrum plane.
In the velocity-spectrum plane, the vertical axis is T
0
and the horizontal axis is V
s
.
The method consists of the following steps:
(1) Select a CMP gather that has a relatively high S/N ratio. The CMP gather should
be sorted in offset.
(2) Determine the minimum and maximum T
0
that you want to analyze.
(3) Determine the minimum, maximum, and increment of V
s
to be attempted.
(4) Determine the gate width, w, around the reference time T
0
. This is usually equal
to the dominant period of the data.
(5) Start with the minimum T
0
and V
s
.
(6) Compute
2 2 2
0
/ ) (
s
V X T X T + , where T
0
and V
s
are set to T
0
and V
s
of step (5)
and X is the offset of the traces in that CMP gather.
(7) The amplitudes in a gate of width w centered about T(X) calculated from step (6)
are selected from all the traces in the gather.
(8) The average of the amplitudes corresponding to the first time sample of the gate
on all traces in the gather is computed and squared.
(9) Step (8) is repeated for all the time samples in the gate w.
(10) The squared averages are added together to give the stacked energy (E
s
).
(11) Sum up the squared amplitudes of the first sample on every trace.
(12) Repeat step (11) for all the other samples in the gate.
(13) The sums of the squares are add together to give the un-stacked energy (E
u
).
9
(14) Calculate the semblance NE = E
s
/E
u
.
(15) Now you have one point on the velocity-spectrum plane, namely (minimum T
0
,
minimum V
s
, NE).
(16) While fixing T
0
, increment V
s
and repeat steps (6)-(14) until you reach the
maximum V
s
.
(17) Increment T
0
by w/2 and repeat steps (5)-(16) until you reach the maximum T
0
.
(18) For an event occurring at a given T
0
, the corresponding V
s
is the one that is
associated with the maximum semblance occurring at that T
0
.
(19) Select another CMP gather and repeat steps (5)-(17).
You should end up with a set of picks (T
0
, V
s
) for every selected CMP.
To find the (T
0
, V
s
) sets for the other, unprocessed CMPs, we linearly interpolate
them.
E
s
, E
u
, and NE are measures of coherency (similarity) of the signal along a hyperbolic
curve. Other measures of signal coherency are often used such as:
The stacked amplitude.
The normalized stacked amplitude.
The un-normalized crosscorrelation.
The normalized crosscorrelation.
The energy-normalized crosscorrelation.
The coherency measure is usually displayed as:
A contour plot.
A gated row plot.
The time increment is selected as half the dominant period to avoid:
10
extra costs of processing if we choose less than half, and
mixing of closelyspaced reflections if we choose more than half.
Important parameters to consider when using the velocity spectrum method are the
minimum, maximum, and increment in the trial stacking velocities.
The velocities used in this method are called often the semblance velocities.
The velocity spectrum method is more suited for noise-contaminated datasets.
Factors affecting velocity estimates
Velocity estimation from the velocity spectrum is limited in accuracy and resolution
for the following reasons:
Spread length:
Adequate resolution in the velocity spectrum can only be attained with
spreads that span both near and far offsets.
Using only near offsets degrades the coherency peaks at later times due to the
low NMO associated with deep reflections.
Using only far offsets degrade the peaks at shallow times due to the high
NMO associated with shallow reflections.
Stacking fold: Using very low folds significantly shifts the coherency peaks in the
spectrum due to the loss of hyperbolic character of reflections.
S/N ratio: The accuracy of the velocity spectrum is limited when the S/N ratio is
poor due to the many erroneous peaks generated by aligning random noise.
11
Muting: It decreases the semblance magnitude because of the zeroed (muted)
amplitudes. If muting is used, the semblance calculation has to compensate for
the effect of muting.
Time gate length: Smaller time gates increase the computational costs, while
coarse ones reduce the temporal resolution of the spectrum (i.e., distinguishing
velocities of closely-spaced reflections).
Velocity sampling: The velocity range should be chosen to span the expected
stacking velocities of primary reflections across the area (i.e., in time and space).
The velocity increment should be chosen fine enough to give the required
resolution of the spectrum.
Departures from hyperbolic moveout: The moveout can depart from a hyperbola
due to anisotropy or lateral heterogeneities in the overburden. Using three-term
series fitting might help in picking the stacking velocities.

S-ar putea să vă placă și