Vous êtes sur la page 1sur 17

GEOPHYSICS, VOL. 46, NO. 5 (MAY 1981);P. 717-733, 26 FIGS.

Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fundamentals of frequency domain migration

Joong H. Chun* and Chester A. Jacewitz*

Fourier transform is the fundamental technique of this method.


I ABSTRACT
Frequency domain migration is founded upon the wave
1 The advantagesof this method include fast computing time good
performance under low SIN ratio conditions, and excellent per-
equation, and so includes diffractions and other effects. formance for steep dip. Disadvantages include difficulties with
This paper seeks to motivate and illuminate frequency do- widely varying velocity functions. Others discussing aspects of
main migration using straightforward geometric techniques frequency domain migration include Bolondi et al (1978), Hood
and simple frequency domain observations. (1978). and Whittlesey and Quay (1977).
This article is meant as an overview for those working in the
I I
geophysical industry who wish to know more about migration. In
INTRODUCTION particular, the authorshope to provide new insights into the funda-
mental aspects of migration with emphasis upon frequency do-
Migration of seismic data is a processof mapping one time sec-
main migration. There will be much reliance upon the geometry
tion onto a second time section in which events are repositioned
behind the usual physical and mathematical treatments of migra-
underthe appropriatesurface location and at the correct time Thus,
tion. The intimate geometric relations between the migration of a
a migration output should be a time section of the geologic depth
dipping event in time and the counterpartmigration in the frequency
section. No current migration technique handles all the difficulties
domain will be explained. Certain general effects of migration
of noise, rapidly varying velocities, steepdips, and other problems
upon seismic data will be considered. Some specific parameters
perfectly. Techniques vary greatly in performance relative to these
are considered to determine how they affect migration.
problems.
Migration in the frequency domain has been known for some
Three of the major techniques of migration are diffraction,
time Stolts insights have made it a practical method. We intend
finite-difference, and frequency domain migration. Diffraction
to illuminate his results from the geometric viewpoint.
migration is also known as Kirchhoff integral migration when
This article covers the fundamentals of migration. Any imple-
high-order approximationsare used. The finite-difference approach
mentation has problems of its own. Frequency domain migration
is commonly known as time domain or wave equation migration.
has the usual problems of finitely sampled data. Working in the
Frequencydomain migration may also be referred to as frequency-
frequency domain has special problems. These can be handled
wavenumber migration or Fourier transform migration.
effectively, but it is not our purpose to cover such material.
The diffraction type migration process is often described as a
statistical approach. A particular datum might have originated
from many possible subsurfacelocations. All suchpossibleorigins
are treated as equally likely in this approach. The major ad-
vantage of diffraction migration is good performance with steep
dip. One disadvantage is poor performance under low signal-to-
A -x 0
noise (S/N) ratio conditions.
Finite-difference migration is a deterministic approach (Claer-
bout, 1976). The migration procedure is modeled by the wave
equation. This equation is then approximated by a simpler type
of equation appropriate for migration. This last equation is then
approximated by a finite-difference algorithm. An advantage of
the finite-difference method is its good performance with a low
S/N ratio. Disadvantagesof this method include a relatively long
computing time and difficulty in handling steep dip data.
Frequency domain migration is also based upon a deterministic
(a) (b)
approachvia the wave equation (Stolt, 1978). Instead of utilizing FIG. I. A 90.degree reflector model. (a) Earth section, (b) record
finite-difference approximations, the two-dimensional (2-D) section.

Presentedat the 31st AnnualMidwest SEG Meeting, March 8, 1978, in Midland. Manuscript received by the Editor May 19. 1978; revised manuscript
receivedSeptember12, 1980.
*SeismographServiceCorp., P.O. Box 1590,Tulsa, OK 74102.
0016-8033/81/0501-717$03.00. 0 1981 Societyof Exploration Geophysicists.All rights reserved.

717
718 Chun and Jacewitz

A -x
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

\
\
\ 1 1
z z
90" 80"
(4 (b)

FIG. 2. A dipping reflector model. (a) Earth section, (b) record


(a) (b)
section.
FIG. 3. The fan earth model. (a) Some representativeearth dips-
open fan. (b) Corresponding record section-folded fan.

MIGRATION IN THE DEPTH DOMAIN tance AC is mapped vertically on the x-z plane as in Figure 2b
We will begin the study of migration fundamentals with special to the segment AC. The travel distance is thus AC = AC. The
cases, then work toward a complete understandingof the geometry earth model in Figure 2a is superimposedon Figure 2b as a dotted
of migration. A step-by-step procedure for performing migration line. From Figure 2b. we easily see the following relationship:
of a seismic event will be discussed. This procedure is similar AC AC
to mapping used in the frequency domain to accomplish migration. sin Ly, = ~ = - = tan oh. (1)
OA OA
Consider the vertical earth model (o., = go-degree reflector)
of Figure la. Suppose a seismic source is at A with the signal The equation above is a familiar description of the relationship
recorded at the same point. Then the only energy that can be re- between the migrated dip (a,) and recorded dip angle (oh). Since
corded at A is from the horizontal path in the ray theory approach. C maps to C under migration, this process moves data updip.
Any nonhorizontally traveling wave will be reflected downward The useof reflections explains many of the observedphenomena.
and not reach A. Map the travel distance of the horizontally The concept of diffraction is required for deeper understandingof
traveling ray on the depth plane (x-z) in the z-direction as in migration. Diffractions are usually associatedwith discontinuities.
Figure lb. Since OA = AC, the dip angle ob of the reflection is Reflections may be considered as a superposition of diffractions.
equal to 45 degrees. Thus, for a 90-degree reflector the reflection We are using geometric ideas to motivate and explain the exact
takes place only at one point at the surface and the recorded re- solution. For this purpose, the term reflection will generally be
flections are mapped along a 45.degree line on the depth plane as adequate. Thus, given a reflector, the reflection processgives rise
A moves along the surface. Migration maps the 45.degree re- to a reflection event on the record section. Since sections will be
flection of Figure I b onto the 90.degree reflector of Figure la. considered before and after migration, the general process of
Next, consider the dipping earth model in Figure 2a. Again. mapping a reflector or diffractor to the record section will be
assumethat source and receiver are both at A. The wave from A described as the diffraction process. Migration proceeds from a
will be reflected at C and will be recorded at A. The travel dis- record section to the earth model. Diffraction proceeds in the di-

0 -x A B

I I
\
\ I I
\
\ \ I I

(W
FIG. 4. A bounded dipping reflector. (a) Record section, (b) construction for migration.
Frequency Domain Migration 719

0 -x A - A (3) With point 0 as origin. draw a circle of radius OB.


(4) From point D draw a horizontal lint toward the circle
10" 10"

I
and find the intersecting point between the circle and the

z
20

30

40
J
z
line (point E).
(5) Connect origin 0 and point E, to find the migrated dip
angle a,.
(6) Construct point D by making ED = ED. D is the
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

50
migrated point of D. The angle of line DD is F =
70
o,/2. By projecting line CD with this angle to line
6 OE, the migration process is completed.

Validation of this migration process is given in Appendix A.


Thus. migration requires a two-step procedure. The first is to
FIG. 5. A correspondencebetween the record section and the depth
tind the correct line which gives the migrated dip angle, and the
section. (a) Wedge of events before migration. (b) half-disc of
events after migration. second is to map reflection points to their appropriate migrated
points linearly. This linear mapping property is important for
understandingfrequency domain migration.
The seismic section is a bounded diffraction model (the lengths
rection opposite to migration. For example, given a point diffrac- in x and 2 are limited). Consider the bounded fan diffraction model
tor. the diffraction processgives rise to a diffraction on the record of Figure 5a. By migrating this section. we obtain the section of
section. Just as reflection may refer to the reflection process or Figure Sb. For this model, an event inside the triangle ABO is
reflection data, so too with diffraction. The context should make migrated within a semicircle whose diameter is equal to OA.
the distinction clear. To understandthe correspondencebetween Figures 5a and 5b.
Consider an earth model with varying dip angles. This fan the reader should review Figure 2b. In this figure, the raypath
model appears in Figure 3a. In this model. the velocity is con- AC is perpendicular to the reflector OC. Return to Figure 5.
stant for each layer between the reflectors or ribs of the fan. Each The triangles of Figure 5a correspondto a traveltime and diffracted
layer has a different density. A reflection will be received from line pair. In Figure Sb the distance from A to the endpoint of a
each boundary because of the density contrast. If the source and segment corresponds to the raypath as in Figure 2a. Thus. each
receiver are placed at the same point along the surface, the recorded angle in Figure 5b (A to endpoint to 0) must be a right angle. Recall
seismic section will be a diffracted seismogram of Figure 3b. For the geometric fact that the collection of all points representing an
this fan earth model, one can view the migration process as an apex of a right triangle formed with a fixed hypotenuse OA must
unfolding (extcndirtg) of a folded fan (diffraction model). The necessarily form a semicircle. This should help explain the rela-
diffraction processcan be viewed as a folding of a fully extended tionship between Figures 5a and 5b.
fan (the earth model). Note that the intensity of folding is such The basic concept of migrating a fan model in the depth domain
that the steeper the dip of the unfolded fan, the more densely it is useful for understandingthe relationship between the migration
is folded on the diffraction model and vice versa. procedure in the time domain and a similar procedure in the fre-
A simple mechanical method of migration is illustrated in quency domain.
Figures 4a and 4b which uses equation (I ). An understandingof
this method is helpful for understanding frequency domain MIGRATION IN THE FREQUENCY DOMAIN
migration. A reflection segment to be migrated appears in Figure We have discusseda special type of seismic section, which was
4a. To migrate the event. one may follow this procedure: described as a fan model. This meant that the unmigrated section
closely resembled a partially opened oriental fan. A real seismic
(I) Extend line segment CD toward the surface to inter- section does not resemble this, but it can be imagined as being a
section point 0. set of scattered diffraction fans. A curved event may be con-
(2) From point D draw a vertical line to the surface. The sidered as consisting of many very small StKIight-linCsegments.
point of intersection is B in this example. A most important aspect of the 2-D Fourier transform is that, how-

X(t) xctr

t
0 -t
-t
P-
0

FIG. 6. Review of Fourier transforms-time domain. frequency domain amplitude, and phase. (a) Spike at time 0, (b) spike shifted in time
and (c) boxcar.
720 Chun and Jacewitz
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

f(x.2) F&,.z)
(a) UN
>L

IF&, , k,)l
(c)
FIG. 7, A 3-D sketchofthree
boundedhorizontal reflectors. (a) Line spike model in the depth domain. (b) Fourier transform in the x-direction.
(c) Typical amplitude contours in the 2-D frequency domain.

ever scattered events are in the depth domain, the Fourier trans- the Fourier transform of a rectangularpulse (boxcar) is a sine func-
formed data will be well organized and become a nicely gathered tion in the frequency domain as in Figure 6c.
fan. Thus, by applying the migration process to the gathered fan Consider the seismic section f(_r, z) of Figure 7. Assume that
while in the frequency domain, all the scattered dipping events there are three horizontal reflections (line spike series) located at
in the depth domain are handledeasily. We will review a few funda- Z1. Zz, and Z,. Figure 7a is a three-dimensional (3-D) sketch of
mental properties of the Fourier transform in both the I- and 2-D this section, The Fourier transform of this 2-D section with respect
cases, then consider how the folded fan earth model can be un- to the variable x appears in Figure 7b. This is a bounded seismic
folded in the frequency domain. section. In this figure, as the range of .r tends to x, the sine func-
It is well known that an impulse (spike) at I = 0 in the time tions located at Z1, Zz, Z3 wil! be reduced to spikes. If we per-
domain transformsto a constantfor all frequencies in the frequency form the Fourier transform once again in the :-direction, we would
domain as in Figure 6a. When the spike is located at I # 0. the obtain the 2-D transform of ,f(.~, -_)of Figure 7c. Since F (k,, k,)
amplitude spectrum remains the same but the phase spectrum is a complex function, we have only presented the contoursof the
changes. Thus, information regarding the time of the spike is pre- amplitude spectrum IF(k,. k,)l in Figure 7c. Thus. we see
served in the phase spectrum as in Figure 6b. It is also known that that a horizontal event in the x-z plane is mapped along the k,-axis.

FIG. 8. A 3-D sketchofthree boundedvertical reflectors. (a) Line spike model in the depth domain. (b) Fourier transform in the depth direction.
and (c) typical amplitude contours in the 2-D frequency domam.
Frequency Domain Migration 721
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

FIG. 9. Bounded dipping reflection segment. (a) Reflection in the


depth domain, (b) frequency domain representation.

and the width of the mapped event is inversely proportional to the


length of the horiiontal event in the X-Z plane. For the width of
(W
an event, one might use the size of a band containing 90 percent
!+dddddd X
of the energy. By applying the same logic, if there are three vert-
cal events on the f(s, z) section of Figure 8a, they will be mapped
along the &-axis as in Figures 8b and 8c.
These examples illustrate that the 2-D transform maps normal to 2
the dip direction of the events in the X-Z plane. In other words,
horizontal events are mapped along the k,-axis and vertical events
are mapped along the k,-axis. The width of a mapped event is
inversely proportional to its extent in the X-Z plane.
t
Consider the dipping reflection in the X-Z plane of Figure 9a. It
will be mapped on the k, - k, plane along the normal direction Cc)
to the dip angle in Figure 9b. The width will be inversely propor-
tional to the extension of the event in the x-2 plane. FIG. IO. Migration mapping. (a) Depth domain mapping. (b) con-
Thus, any event in the X-Z plane is mapped normal to its dip struction of frequency domain mapping, and (c) preservation of
angle in the k, - k, plane. Therefore, the dip angle of both spatial frequency.
diffracted and migrated events in the .X-Zplane must be preserved
in the k, - k, plane. Also, all events of the same dip are gathered
together in the frequency domain along a single line.
When events on the X-Z plane are not line spikes but have some
waveform, the mapping of such events is more complex.
Figure IOb shows the Fourier transform of the diffracted and
migrated events of Figure IOa. As mentioned before, the dif- point on the Vb line (point B).
fracted event (OD) and migrated event (OD) of Figure IOa will (3) Draw a vertical line from point B and find the inter-
be mapped along the V, and V, normal vectors of Figure lob. In seating point with the circle (point C).
the frequency domain, if V, is given, the V, vector may be found (4) Draw a line that passesthrough the origin and point C.
by following a procedure similar to that of the depth domain. This
procedure is: This will give us a migrated dip line (V,).
Note that frequency components along line V, must be mapped
(I) Draw a circle whose radius is equal to OA (A is an along the line V, after migration. However, we do not know
arbitrary point on the k,-axis). The center is at 0. exactly where the point B is mapped on the line V,. If we follow
(2) Draw a horizontal line from A and find the intersecting the same procedure used for the depth domain migration, it
722 Chun and Jacewitz

would appear that point B on the line V, maps at point C on the


line V,. That procedure was illustrated by Figures 4a and 4b
earlier. However. in the frequency domain. point B is directly
mapped to C. In other words, this is a \~erticn/ mapping (pro-
jection) while it was a slanted projection with the angle of a,/2
in the depth domain (see Appendix B). The basic reason for this
vertical mapping is illustrated in Figure 10~. The dark lines repre-
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

sent lines of fixed dip before migration. The results of migration


KX are indicated by the dashed lines. The spacing between these lines
is unchangedby migration. Thus. for such a-seismic section with
(a) (w a single dip, the frequency in the .r-direction is the same before
and after migration. In terms of k, and k, as in Figure lob. this
FIG. I I, Migration mapping in the frequency domain. (a) A line means the component of k, cannot change under the migration
of constant K, frequency and its migrated mapping. (b) a grid of
mapping. This mapping must be vertical in the k, - k, plane.
curves of constant K,.
In Figure I la. the horizontal line intersects all the dipping
lines. If we project frequency componentson this line to the circle,
all the points on the line are properly migrated. Therefore. if we

(a) (w
(4 (b)
FIG. 12. Wedge of dips prior to migration. (a) A wedge of events
in the depth domain, (b) the frequency domain representation of FIG. 13. Wedge of dips after migration. (a) A semi-disc in the depth
the wedge. domain, (b) the frequency domain equivalent.

(4

(b) w
FIG. 14. The full fan earth model. (a) The full fan, (b) the bounded fan of the record section. and (c) the migrated correspondenceof(b).
Frequency Domain Migration 723
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

UN w
FIG. 15. The frequency domain representationof Figure 14. (a) Fourier transform of the full fan, (b) Fourier transform of the bounded fan
with representativepulse spreading, and (c) migration of(b). Note the dotted lines in (b) and (c) correspondunder the migration mapping. as
do points A. A. B, and B.

project frequency components on a series of horizontal lines to ened line as explained before. For simplicity. it is presented as
their corresponding circles vertically as in Figure 1lb, the migra- a series of dipping lines (Figure 12b).
tion process in the frequency domain is completed. A mathe- Figure I3 shows a Fourier transform pair of the migrated fan
matical expression of this process may be written as model obtained from the diffracted fan model of Figure 12. Note
that the frequency domain migration may be viewed as unfolding
k,
F(k,, k,) = F(k,. w 1. (2) (Figure 13b) of the folded fan (Figure 12b). Also, in the depth
4TF
x z domain the triangle ABO of Figure l2a is mapped within a semi-
circle of Figure l3a whose diameter is equal to OA. In the fre-
Function F is the 2-D Fourier transform of the original section.
quency domain, the triangle A B 0 of Figure 12b is mapped
The transform of the migrated section is given by F. Thus the
within an area of quarter of a circle of Figure l3b.
migrated section may be computed directly in terms of the ori inal
We thus conclude the following:
seismic section. The reason for the scaling factor k,/ * k, + k,
in the frequency domain migration is explained in Appendix B.
(I) The dip angles of reelectors, both diffracted and
The Fourier transform pair of the diffracted fan earth model of
migrated, in the depth domain are preserved in the fre-
Figure 5a is presented in Figure 12. Since the model is bounded
quency domain. However, they are normal to the depth
in both X- and z-directions, the real transform should be a broad-
domain dip angle.
(2) The point-to-point mapping from the diffracted event
to a migrated event is a slanted projection with the
-X
angle of a,/2 in the depth domain. It is a vertical pro-
jection in the frequency domain.
(3) For the case of a complete I go-degree fan earth model
case of Figure 14a, the diffracted fan model is a split
45.degree folded fan model as in Figure l4b. If we im-
pose boundaries ABCD as in Figure l4b and migrate
back, then the fans are confined within two semi-
circles as in Figure 14~. The Fourier transform of the
models in the depth domain are presented in Figure 1.5.
Figure 15a is the Fourier transform of Figure l4a, and
Figure l5b is that of Figure l4b. Imposing the bound-
aries shown in Figure 14b results in broadening of
frequency components of each dipping event as ex-
plained before. Figure 15~ is the Fourier transform of
FIG. 16. Synthetic data: record section Figure l4c
724 Chun and Jacewitz

An example of scattered fans is depicted in Figure 16. It is as -x


if the fan model of Figure 14b were scattered across the section.
Migrating the scattered fan in the depth domain, by extending
each reflection to the surface and using the geometric migration
construction. is cumbersome. Because the Fourier transform
gatherstogether events of the same dip, the transform of Figure 16
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

resembles that of the gathered fan model of Figure 15b.


The most important aspect of frequency domain migration is
that the scattered fan in the depth domain becomes a nicely
gathered fan in the frequency domain. The information regarding
the spatial position of a dipping reflection in the depth domain is
preserved in phase relations within the complex numbers in the
frequency domain. By applying the simple migration process to
the model of Figure I6 in the frequency domain, then performing
an inverse Fourier transform, we obtain the migrated scatteredfan
model of Figure 17. FIG. 17. Migrated section from Figure 16.

MIGRATION PARAMETERS -X

7
For real data. there are many unknowns. We must consider the
effects each of these parameters has upon migration. Then, there
will be a coherent picture available of the interaction among these
variables and their relative importance.
I
T
et
1
Z
To consider various migration parameters. it is necessary to
distinguish the dip angles on the time and depth sections. Our
notation will be et for the dip angle on the time section, BZfor the
correspondingdip angle in depth, 8, for the migrated dip angle in
depth, and 6, for the migrated dip angle in time (Figure 18).
Converting from time to depth via velocity yields tan 0, = (a) (W
(V tan 8,)/2. The time is the two-way time while the depth is
the one-way depth. hence the division by 2. Migration in the
depth domain yields the usual formula sin BZ = tan 8,. Reconver-
sion to time is performed via tan 6, = (2 tan 6,)/V.
operations combine to produce:
These 1
z
tan G, = (2 tan{sin~[V(tan ft,)/2]})/V, (3)

so that
G, = tan- (2 tan{sin~[V(tan 0,)/2]}/V). (4)

It is important to realize that the resultant migrated dip angle


w (d)
6, depends upon the velocity of conversion V, not merely the FIG. 18. Migration from time section to time section. (a) Input
original dip angle et. The velocity V does not cancel in this record section, (b) conversion to depth domain. (c) migration in
equation. An alternate formula from Appendix C is depth domain. and (d) reconversion to the time domain.

0 -x B -x

1 i
2
T

FIG. 19. Migration path as a function of velocity. (a) Components of migration in the depth domain, (b) migration of a point under different
velocities.
Frequency Domain Migration 725
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

UN
(4
FIG. 20. Freql :y changes under migration. (a) Frequency domain, (b) time domain-migrating a seismic pulse.

tan tit the time section is a function of velocity, one can derive (see
tan Cl, = (5) Appendix E)
JI - /:jzlantl,
tan tit
d, = V2T ~
From this formula, for any fixed tit, 6, will approach 8, for small 4
velocities. That is, small velocities produce little noticeable and
change in dip angle. As the velocity is increased, e, will become
90 degrees. Beyond this range, there is no real angle for migration. dt = T [I - \:I - ii tan 0,) (IO)
Next, consider the migration position as a function of the angle
on an apparent depth section tlZ and the apparent depth of a point
The first equation indicates that a reflection segment on a seismic
Z. The following fundamental formulas from Appendix D are
time section migrates a horizontal distance proportional to the
illustrated in Figure 19a. If d = (d,, d,) represents the com-
square of the migration velocity. The second equation is more
ponents of the displacement of a point P by migration, then
difficult to visualize. Instead, we may eliminate the variable
d = (dl = 2 z sin(8,/2) = m, (6) tan tit to study the migration path of a point (see Appendix F).
Let us describe the migration path of a point P on a dipping
d, = z sin e,, (7)
reflection segment of a time section as velocity changes. The
migrated point P is on the parabolic path connecting P and 0 and
d, = ~(1 - cos 6,). symmetric about the I-axis (Figure l9b). For small velocities,
(8)
P will be found near P, while for large velocities. ? will be
These relations are only of preliminary interest as we are interested found near 0.
in changesfrom a time section to a migrated output time section. The change of frequency as a function of velocity can be ex-
However, note that in a depth section, a point on a dipping re- plained more easily from the frequency domain viewpoint. The
flection is displaced by migration linearly and proportional to its 2-D frequency domain mapping given for migration in equation
depth before migration. We can also leave the velocity and dip (2) leaves the k, component unchanged. Velocity changes also
angle constant. but vary the depth. For a constant velocity, a uni- leave the k, component unchanged. Therefore, in migration of
form change in time is equivalent to a change of depth (with no a dipping reflection on a seismic time section, only the k, com-
change in 6,). Thus, in the time section. the corresponding dis- ponent will be affected by velocity variation. Using a notation
placements d, d,, and d, are linearly proportional to the time of F,. F, for input frequencies in k, and k,. one can apply the
shift. conversions and mapping to determine
We may return to the time domain again to see how a point
migrates as a function of velocity. Using the same fundamental (I 1)
equations, and the fact that the depth of a reflection segment on
726 Chun and Jacewitz
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

I -x

\
\
\ \

o\\,
w /
0

FIG. 21. Reflections with a single dip. (a) Before migration, (b) /// \\
\\
after migration. /
/

I
I
-0

for the output time frequency Fi. As is expected for a small (3 Cd)
velocity, F, = F,. As discussed, this reflects the fact that the
seismic time section is only slightly ahered under migration by a FIG. 22. Diffraction of a bounded reflector. (a) Depth domain
low velocity. For V large enough, F, = 0. The dip of the re- earth model, (b) frequency domain, (c) diffracted (inverse process
flection is now 45 degrees in the depth section, so that it migrates of_migration), and (d) back to depth domain (record section).
to a vertical event with F f = 0. For velocities above the velocity
which would make the dip of a reflection 45 degrees in the depth
section, the reflection will be eliminated from the resulting sec-
tion. This phenomenon will be described in detail later. It is not mapping. Start with a geologic section, then consider a record
merely conversion to depth that alters F,. Migration itself will section. Figure 22a is a representativegeologic section. This event
alter this frequency even if one works only in depth domain. In is a bounded dipping reflector. In the frequency domain, as seen
this case the mapping is earlier, the event will be modulated by a sine function. An example
is drawn with two points A and B labeled in Figure 22b.
(12) Consider a record section generated by the diffraction mapping
Figure 20a illustratesthe two key points of the frequency domain which is the inverse of migration mapping (take V = 2 for
mapping. First, the mapping takes lines of one apparent dip into simplicity). Note in Figure 22c that the entire area occupied by
lines of the true dip. Second, the mapping is vertical. That is, dips steeper than 45 degrees (suprrsreep) has been zeroed. This
the frequency in .Y is unchanged by the migration mapping. is because line OA, which has a 90.degree dip in Figure 22b.
Let us return to the time domain to understandthese two points. is mapped to a 45.degree line on Figure 22~. No dip greater than
Figure 20b shows an event with a single dip. The associated 45 degrees can exist on a record section. Also, note that no infor-
waveform is complex. Nonetheless, after migration, the event mation has been lost via the diffraction mapping. The representa-
still has a single dip. The preservation of spatial frequencies is tive sine function of Figure 22b has merely been split in Figure
illustrated in Figures 2la and 21b. The horizontal spacing of the 22~. No part of it has been lost. Figure 22d showsthe section which
dipping reflections is unchanged by the migration process. How- resultsfrom such a typical geologic model. These figuresillustrate
ever, spacings A, B, and C before migration must be the same that the diffraction patterns of Figure 22d must be related to the
as A, B, and C after migration, respectively. Thus. the spac- distortion of the sine function in Figure 22~.
ings A, B, and C must be greater. This shows that the frc- If one performs migration via the frequency domain approach
quency in the time direction must be reduced by migration. This with the input section of Figure 22d, one merely retracesthe steps
may be confirmed by reconsideringthe frequency domain mapping successivelyto Figure 22c, Figure 22b. and, finally, Figure 22a,
directly. the correct geologic model. Migration appearsto entail no loss of
We may follow the processing through a complete cycle to information. but there is a more subtle consideration.
investigate the loss of input data under the frequency domain Proceed in a different direction. Start with the bounded dipping
Frequency Domain Migration 727
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

KZ
\
\\
\
\

Kx KX

\\
/ \\
/ \\

-.\\\
08 (c)

/
r/ I
IA
KX

(d) (e)
FIG. 23. Bounded dipping reflection model without edge diffraction. (a) Depth domain model (record section), (b) frequency domain. (c)
truncation, (d) migration, and (e) back to depth domain after migration.

reflector on the section of Figure 23a. The frequency domain model because of the supersteepdips shown in Figure 23b.
illustration in Figure 23b is again modulated by a representative The cause of the supersteepdata which is not of geologic origin
sine function. The migration mapping will not use any data in the is worth investigating. One cause of supersteepevents is clearly
supersteepregion indicated by 0 in Figure 23~. These data are noise which can occur at any dip, even supersteep. A second
lost. Figure 23e illustrates the result when viewed in the geologic cause of supersteepevents is a velocity which is too high. Any
plane after the migration. The dashed curves of Figure 23e repre- event will become supersteepif the velocity used for processing
sent low-amplitude data caused by the zeroing and stretching of is taken high enough. Thus, a poor choice of velocity may cause
the sine function of Figure 23b. If we start with the geologic geologically valid data to be interpreted as supersteep, and hence
model of Figure 23e and retrace the steps, we never recover the it will be discarded.
original section of Figure 23a. This is as it should be. The record We have seen how migrated seismic sections change as one
section of Figure 23a cannot be generated from any geologic varies certain parameters. Some results which might appear
Chun and Jacewitz
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

,_ ^_

. .

--. .._.
--

__ . _---

_ ._. ._. I

FIG. 24. Synthetic sections. (a) Top, before migration, (b) bottom, after frequency domain migration
Frequency Domain Migration 729
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

(a) UN
FIG. 25. Complex synthetic section. (a) Before migration, (b) after frequency domain migration.

negative have been seen to be natural consequences of the quency domain easily handles events of any dip and location.
migration process. Migration with a nonconstantvelocity is a delicate and complicated
problem in itself. Still, even the simple approximations using
APPLICATIONS TO REAL AND SYNTHETIC DATA
smoothed stacking velocities did a reasonably adequatejob.
A few frequency domain migration examples are presented to
show that the scattered fans were migrated correctly.
CONCLUSIONS
The section in Figure 24 contains bounded dipping reflections.
Notice the separation between the terminators of corresponding Whenever one is dealing with problems of a mathematical
events in Figure 24a. After migration, the events have been nature, it is always prudent to select the natural coordinate
connected making the reflectors continuous. Furthermore. the system if possible. In this coordinate system, the statement of
events have moved updip and the dips themselves have increased. the problem and methods of solution should become exceptionally
This section is relatively easy to visualize as a scattered fan. It clear. We pursuethe idea that the coordinatesof the Fourier trans-
has been gathered in the frequency domain and migrated in the form domain are the natural coordinates for dealing with migra-
simple manner we have discussed. Note that the reflections do tion. In particular, no matter where they are located in space, all
not terminate sharply. Sharp terminations on the geologic section events of the same dip appear along lines normal to that dip when
generate diffractions on the record section. But these synthetic viewed in the frequency domain. Events of the same dip are
reflections were sharply terminated, and the necessary diffrac- migrated together. Thus, dip lines through the origin in the
tions were absent. Note also the lowering of frequency along the Fourier transform domain are part of a coordinate system natural
time axis after migration. This effect follows from the frequency to migration. In the diffracted transform domain, straight lines of
domain migration mapping being downward toward lower k,. constant frequency in depth are quite natural. These transform
This effect increaseswith increasing dip. into circles, which are curves of constant frequency along the
The section of Figure 25 is more complex. Before migration. normal.
Figure 25a is a bewildering collection of crossing arcs. After The use of events of a single dip greatly simplifies our under-
migration in Figure 25b, the reflectors are seen to be smooth standing of migration. Using the Fourier transform, every event
curves. This section contains few straight-line segments. In- on a seismic section may be considered as a sum of events of a
stead, the curved lines may be considered to be short line seg- single dip and the section a sum of all of such events. Although it
ments of rapidly changing dip. Of course, the principle of super- may be more difficult to visualize what is occurring in this more
position allows an easy treatment of even this complicated complicated case, one knows that the principle of superposition
section. applies. Thus, the entire migration process may be considered
A real seismic section of Figure 26 is the most complicated as the simple procedure just described applied to a vast sum of
section in our illustrations. Nonetheless, migration via the fre- simple events.
730
Chun and Jacewitz
ACKNOWLEDGMENTS
REFERENCES
The authors wish to thank the management of Seismograph Bolondi.G.. Rocca.F., and Savelli. S.. 1978, A frequencydomainap-
Service Corp. for their cooperation and encouragement during preachto two-dimensional migration:Geophys.Prosp..v. 26. p. 750-
the preparation of this paper. Also, we wish to thank the many 772.
Claerbout,J. F.. 1976. Fundamental5 of data processing:New York,
other colleagues who provided both help and data as valuable in- McGraw-Hill Book Co.. Inc.
put to our work. In particular, the editors and reviewers of GEO- Hood, P., 1978,Finite differenceand wavenumbermigration:Geophys.
PHYSlCS have been most helpful. Dr. Enders A. Robinson deserves Prosp.,v. 26, p. 773-789.
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

St&. R. H., 1978. Migrationby Fouriertramform.Geophysics.v. 43.


special thanks for his advice and encouragement. p. 23-48.
Whittlesey.J. R.. and Quay. R. G.. 1977, Wave equationmigration:

FIG. 26. Real seismic section. (a) Before migration. (b) after frequency domain migration,
Frequency Domain Migration 731

Presented at the 47th Annual InternationalSEC Meeting. September by the geometric procedure. Now apply the principle of superposi-
20. in Calgary.
tion. Since each section of dip c(b is a combination of lines of this
dip, such a section must migrate to a section of dip (Y,.
APPENDIX A One need consider only sectionswith a single dip. Under migra-
MIGRATION MAPPING IN THE DEPTH DOMAIN tion, each line is moved about its intersection with the surface.
This point of intersection remains fixed. In particular. this means
Appendix A servesto prove the validity of the constructionillus-
that the frequency in X, the spatial frequency, is unchanged by
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

trated in Figure 4b. Steps (l)-(4) are straightforward geometric


migration. Thus. if the mapping which implements migration
procedures. Because of construction of a circle in step (3).
takes the form
OB = OE. Let d be the distance from E to the .x-axis. By step (4).
(1 = BD. Thus, combining these facts. (K,. K,) 4 (K,. K,).

BD one must have K, = K,. The mapping leaves the spatial fre-
=os
tancub quency invariant.
A point (K,, K,) in the frequency domain corresponds to an
d event of dip 0 with tan 8 = -K,/K,, This is becausethe Fourier
OE transform maps events to their normals. Thus the relation

= sin a,. tan ab = sin (Y,

This justifies step (5). may be applied to the case of one point (I?,. K,) to give
Now one must show that e = a,/2 and that D is the migrated f?,/i?, = K,/cKz.
counterpart of D. It is somewhat easier to demonstrate these re-
sults in reverse. That is, supposeD is the migrated point, making Since I?, = K,, this may be rewritten
BD = BD. What are the angle E and the distance ED? Note &z f K; = K;
that BD is perpendicular to OE. Thus, BDE and BDE are both
right triangles. Since they have the same hypotenuse BE. and or
have one other side equal in length, the third sides must be equal. K, zz d/K; - c2X.
Thus ED = ED as required. Computing F is then easy. Since
This validates mapping points vertically in Figure lob. In addition.
ED is horizontal, it is parallel to OB. Thus the exterior angle at
the equivalence between equation (2) and the geometric construc-
E is cx,. The interior angle is then 180 degrees - a,. Since the
tion of Figure IOb has been demonstrated.
triangle is isosceles, the remaining angles are both F. The angles
We still must discussthe weighting factor in equation (2). This
of a triangle sum to 180 degrees. Thus, 180 degrees - (Y, +
is not included in the geometric construction, but only in the
F + F = 180 degiees or E = a,/2.
algebraic formulation. Recall that each section with a single dip.
no matter how complex, migrates to another section with a single
APPENDIX B dip. Under the mapping of Figure 4b. the points of triangle OBD
MIGRATION MAPPING IN THE FREQUENCY DOMAIN migrate to those of triangle OBD. But
Appendix B has two purposes. The first is to show the corrcct- Area (OBD) = (I /2)(BD)(OD)
ness of the geometric procedure described for migration in the = (I/2)(BD)(OB cm q,)
frequency domain. The second is to indicate the validity of the = cm a, Area (OBD).
algebraic equivalent of equation (2).
This means, roughly. that before migration the event occupies
To validate the procedure, the relation tan ah = sin (Y, must
more area so that it is overemphasized. Thus. after migration,
hold in the construction. Referring to Figure lob, by step (I),
the event must be multiplied by the factor of cos (Y, to restore the
OA = OC. Let d be the perpendicular distance from C to the
k,-axis. By steps (2) and (3), d = AB. Thus correct balance. In the frequency domain, the event maps to its
normal so that
AB

tan
(yb= OA
cos a, = K,/dK; + K;.

But this is precisely the factor which occurs in equation (2).


d
oc

= sin cy,. APPENDIX C


THE time SECTION ANGLE BEFORE AND AFTER MIGRATION
This shows the construction produces the correct migrated angle
cx, when starting with (Yb. Thus, when considered as a mapping Appendix C derives a formula relating Bt (the premigration dip
between lines through the origin, the construction takes lines of angle on the time section) with 6, (the postmigration angle). The
dip (Ybto lines of dip cx,. One must still show the construction is notation follows migration parameters. Begin with the basic
correct as a mapping between points in the frequency domain. migration relation for dip angles on the depth section:
There are many ways to indicate the validity of migration tan 8, = sin e,
mapping. Stolts important paper (1978) first established this via
sin i!l
the wave equation and a change of variables in the Fourier trans- = cos A, z
form. Since the formula holds, a more descriptive indication of cos Fi,
why it shouldhold basedupon a raypath approachwill be presented. I
First, the mapping must map sections of dip (Ye to sections of zz- tan 6,.
dip ~1,. Certainly, each line of dip (Ybis mapped to one of dip (Y, VI + tan6,
732 Chun and Jacewitz

Here we have used the relations displacement in the time section. Equations (9) and (IO) will be
proved here. Consider a fixed point at time T on a reflector of
sin 0
tan 0 = - dip fit on the time section. Note that d, is the same whether mea-
cos 0 suredon the time section or on the depth section. After converting
and to depth. the point is located at depth 2 = (V/2) 7 on a reflector
dip tll. From Appendix D.
cos 0 = l/u/r + tantl.
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

d, = ; sin HZ.
The first is the definition of the tangent. The second is easily
checked by using
= (T) tan Oz.
sin8 + cos20 = 1.

One has, after converting from depth to time = (F) (: tan 0,).

or

or d, = (VT tan 8,)/4


I Also. the displacement in ; is given from Appendix D by
V2
tan 8, = tan G,
l
1 I + 4 tan8,.
d, = -_(I - cos e,,.

Then solve for tan e,. The result is Using the relation between cosine and sine.

VT
tan G, = tan HI/ Jybz (i, = -[I - (I - sin8 * )12]
2

By the migration relation,

APPENDIX D
d, = :[I - iI - tan2tl,112].
DISPLACEMENT OF A POINT UNDER MIGRATION
Appendix D refers to Figure 19a. Here we derive the formulas Reconverting to time
for rl, and d,. Let h be the length from p to the -r-axis, and let
; = BP. Migration means in particular that BP = BP. The angle d, = T {I - [I - (V tanfl,)/4])
OBP is 90 degrees - e,. L

Thus. Thus. since


h = BP sint90 degrees - 6,).
and dt = 1.
V
d,.
h = zcos g,.
d, = T{l - ]I - (V tan2ti,l/4]12}.
Therefore.

d, = 1 - h,
d,=z-:cosA,. APPENDIX F
MIGRATION OF A POINT AS A FUNCTION OF VELOCITY
and
Appendix F refers to Figure l9b and the proof that a point
d, = :(I - cos G,). migrates along a parabola as the migration velocity changes. Con-
The relation for d, also follows from basic trigonometry: tinuing from Appendix E. one has

rl, = BP cos(90 degrees - e,) d, = VTtan


so that 4

for the .r displacement. and


d, = z sin G,

Finally, the triangle PBP is isosceles with vertex angle 6,.


Drop a perpendicular from B to PP. Then one sees that

d for the t displacement. The last equation may be rewritten as


- = z sin(e,/2).
12
2
T(l -Ttanflt) = T - d,.
or

d = 2: sin(8,/2) So, squaring the last expression.

T (I - T tann,) = T2 ~ 2Trl, + d;.


APPENDIX E
Thus,
MIGRATION OF A time SECTION AS A
FUNCTION OF VELOCITY V
4 T tanfl, = 2Trl, - c/f
Appendix E usesthe notation of Figure 19a. In addition, d, is the
Frequency Domain Migration 733

From the expression for d,. one has (d, - 7) = T - T tan H,c/,

V For the choice of (iI = T.


4 T tanHt = T tan Otd,
0 = T - T tan H,d,

Combining these last two expressions, But this may be rewritten as

T tan 8,d, = 2Td, - df. tan e1 = T/d,.


Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Rearranging this equation. Hence, the point (X. T) has migrated to (0. 0).
The form of the relation shows that rl, has a maximum when
d: - 2Tll, + T tan etd, = U
d, = T. Thus. the parabolic path is uniquely defined by the follow-
This is a parabolic relation between d, and dt. Clearly for V = 0. ing two prop&es: (I) The parabola passes through (X, T) and
d, = 0 and d, = 0. One may rewrite the relation as (0. 0). and (2) the parabola is tangent to the r-axis at (0. 0).