You are on page 1of 192

4 Migration

• Introduction • Exploding Reflectors • Migration Strategies • Migration Algorithms • Migration Parameters • As-
pects of Input data • Migration Velocities • Migration Principles • Kirchhoff Migration • Diffraction Summation
• Amplitude and Phase Factors • Kirchhoff Summation • Finite-Difference Migration • Downward Continuation •
Differencing Schemes • Rational Approximations for Implicit Schemes • Reverse Time Migration • Frequency-Space
Implicit Schemes • Frequency-Space Explicit Schemes • Frequency-Wavenumber Migration • Phase-Shift Migra-
tion • Stolt Migration • Summary of Domains of Migration Algorithms • Kirchhoff Migration in Practice •
Aperture Width • Maximum Dip to Migrate • Velocity Errors • Finite-Difference Migration in Practice •
Depth Step Size • Velocity Errors • Cascaded Migration • Reverse Time Migration • Frequency-Space Migra-
tion in Practice • Steep-Dip Implicit Methods • Depth Step Size • Velocity Errors • Steep-Dip Explicit Methods
• Dip Limits of Extrapolation Filters • Velocity Errors • Frequency-Wavenumber Migration in Practice •
Maximum Dip to Migrate • Depth Step Size • Velocity Errors • Stolt Stretch Factor • Wraparound • Residual Mi-
gration • Further Aspects of Migration in Practice • Migration and Spatial Aliasing • Migration and Random
Noise • Migration and Line Length • Migration from Topography • Exercises • Appendix D: Mathematical
Foundation of Migration • Wavefield Extrapolation and Migration • Stationary Phase Approximations • The
Parabolic Approximation • Frequency-Space Implicit Schemes • Stable Explicit Extrapolation • Optimum Depth
Step • Frequency-Wavenumber Migration • Residual Migration • References

4.0 INTRODUCTION

Migration moves dipping reflections to their true sub- which is at or near the salt dome flank. In contrast, re-
surface positions and collapses diffractions, thus increas- flections associated with the gently dipping strata have
ing spatial resolution and yielding a seismic image of moved little after migration.
the subsurface. Figure 4.0-1 shows a CMP-stacked sec- Figure 4.0-2 is an example with a different type
tion before and after migration. The stacked section in- of structural feature. The stack contains a zone of near-
dicates the presence of a salt dome flanked by gently horizontal reflections down to 1 s. After migration, these
dipping strata. Figure 4.0-1 also shows a sketch of two events are virtually unchanged. Note the prominent un-
prominent features — the diffraction hyperbola D that conformity that represents an ancient erosional surface
originates at the tip of the salt dome, and the reflection just below 1 s. On the stacked section, the unconfor-
B off the flank of the salt dome. After migration, note mity appears complex, while on the migrated section,
that the diffraction has collapsed to its apex P and the it becomes interpretable. The bowties on the stacked
dipping event has moved to a subsurface location A, section are untied and turned into synclines on the mi-
464 Seismic Data Analysis

FIG. 4.0-1. A CMP stack (a) before, (b) after migration; (c) sketch of a prominent diffraction D and a dipping event before
(B) and after (A) migration. Migration moves the dipping event B to its assumed true subsurface position A and collapses
the diffraction D to its apex P. The dotted line indicates the boundary of a salt dome.

grated section. The deeper event in the neighborhood


of 3 s is the multiple associated with the unconformity
above. When treated as a primary and migrated with
the primary velocity, it is overmigrated.
Figure 4.0-3a shows a stacked section that con-
tains fault-plane reflections conflicting with the shal-
low gently-dipping reflections. Note the accurate posi-
tioning of the fault planes and delineation of the fault
blocks on the migrated section in Figure 4.0-3b. From
the three examples shown in Figures 4.0-1, 4.0-2, and
4.0-3, note that migration moves dipping events in the
updip direction and collapses diffractions, thus enabling
us to delineate faults while retaining horizontal events
in their original positions.
The goal of migration is to make the stacked sec-
tion appear similar to the geologic cross-section in depth
along a seismic traverse. The migrated section, however,
commonly is displayed in time. One reason for this is
that velocity estimation based on seismic and other data
always is limited in accuracy. Therefore, depth conver-
sion is not completely accurate. Another reason is that
interpreters prefer to evaluate the validity of migrated
sections by comparing them to the unmigrated data.
Therefore, it is preferable to have both sections dis-
played in time. The migration process that produces
a migrated time section is called time migration. Time
migration, the main theme of Chapter 4, is appropriate
as long as lateral velocity variations are mild to moder-
ate.
FIG. 4.0-2. A CMP stack (a) before, (b) after migration. When the lateral velocity gradients are significant,
Migration unties the bowties on the stacked section and time migration does not produce the true subsurface im-
turns them into synclines (Taner and Koehler, 1977). age. Instead, we need to use depth migration, the output
Migration 465

FIG. 4.0-3. A CMP stack (a) before, (b) after migration. Migration collapses subtle diffractions associated with the growth
faults, moves the fault-plane reflections to the fault positions, and thus makes detailed structural interpretation easier.
466 Seismic Data Analysis

FIG. 4.0-4. A CMP stack (a) before, (b) after time migration. Time migration is adequate for accurate imaging of the
top-salt boundary, whereas depth migration is imperative for accurate imaging of the base-salt boundary (B).
Migration 467

of which is a depth section. Consider the data from an


area with intense salt tectonics in Figure 4.0-4. Time
migration has produced an acceptable image of the re-
gion above the salt. However, note the crossing of events
that is a manifestation of overmigration of the reflection
associated with the base-salt boundary (denoted by B
in Figure 4.0-4b). The improper migration is the result
of inadequate treatment by the time migration of the
effects of severe raypath bendings at the top-salt bound-
ary caused by the strong velocity contrast between the
salt layer and the overlying rocks.
Complex structures associated with salt diapirism,
overthrust tectonics and irregular water-bottom topog-
raphy usually are three dimensional (3-D) in character.
A stacked section really is the seismic response of a 3-
D subsurface on a two-dimensional (2-D) plane of pro-
file. Therefore, 2-D migration is not completely valid
for 3-D data from areas with complex 3-D structures.
Figure 4.0-5a is an inline stacked section from a land
3-D survey. Figure 4.0-5b is a 2-D migration of this
section, while Figure 4.0-5c is the same section after
3-D migration of the entire 3-D survey data. In particu-
lar, note the significant difference in the imaging of the
top of salt T and base of salt B. In 2-D migration, we
assume that the stacked section does not contain any
energy that comes from outside the plane of recording
(sideswipe). Three-dimensional imaging of the subsur-
face is discussed in Section 7.3.

Exploding Reflectors

When a stacked section is migrated, we use the migra-


tion theory applicable to data recorded with a coin-
cident source and receiver (zero-offset). To develop a
conceptual framework for discussing migration of zero-
offset data, we now examine two types of recording
schemes.
A zero-offset section is recorded by moving a sin-
gle source and a single receiver along the line with no
separation between them (Figure 4.0-6). The recorded
energy follows raypaths that are normal incidence to
reflecting interfaces. This recording geometry obviously
is not realizable in practice.
Now consider an alternative geometry (Figure 4.0-
6) that will produce the same seismic section. Imag-
ine exploding sources that are located along the reflect-
FIG. 4.0-5. A 2-D CMP stack (a) represents a 2-D cross-
ing interfaces (Loewenthal et al., 1976). Also, consider
section of a 3-D wavefield. Thus, it can contain energy from
one receiver located on the surface at each CMP lo- outside the plane of the 2-D line traverse. A 2-D migra-
cation along the line. The sources explode in unison tion (b) is inadequate when this kind of energy is present
and send out waves that propagate upward. The waves on the 2-D CMP-stacked section. (c) Clear imaging of the
are recorded by the receivers at the surface. The earth salt structure requires both 3-D data collection and 3-D mi-
model described by this experiment is referred to as the gration (Section 7.3). (Data courtesy Nederlandse Aardolie
exploding reflectors model. Maatschappij B.V.)
468 Seismic Data Analysis

FIG. 4.0-6. Geometry of zero-offset recording (left), and hypothetical simulation of the zero-offset experiment using exploding
reflectors (right) (Claerbout, 1985).

The seismic section that results from the exploding


reflectors model is largely equivalent to the zero-offset
section, with one important distinction. The zero-offset
section is recorded as two-way traveltime (from source
to reflection point to receiver), while the exploding re-
flectors model is recorded as one-way traveltime (from
the reflection point at which the source is located to
the receiver). To make the sections compatible, we can
imagine that the velocity of propagation is half the true
medium velocity for the exploding reflectors model.
The equivalence between the zero-offset section and
the exploding reflectors model is not quite exact, par-
ticularly in the presence of strong lateral velocity vari-
ations (Kjartansson and Rocca, 1979).
These concepts now are applied to the velocity-
depth model in Figure 4.0-7. Consider source-receiver
pairs placed along the earth’s surface at every tenth
midpoint. In this case, a zero-offset section is mod-
eled. At midpoint 130, five different arrivals are asso-
ciated with rays that are normal incidence to the first
interface. Alternatively, imagine receivers placed along
the earth’s surface at every tenth midpoint and sources
placed along the interface where the rays emerge at the
right angle to the interface (equivalent to the normal-
incidence rays of the zero-offset section). In the latter
case, the velocities indicated in Figure 4.0-7 must be FIG. 4.0-7. A velocity-depth model (top) and the zero-
halved to match the time axis with that associated with offset traveltime response (bottom) of the water-bottom re-
the zero-offset section. flector. Shown also are the normal-incidence rays used to
The interface can be sampled more densely by plac- compute the zero-offset traveltime trajectory. Note the five
ing receivers and sources at closer spacing (Figure 4.0- arrivals A, B, C, D, and E at CMP 130 — all from the water
8a). The next deeper interface can be modeled; that is, bottom.
Migration 469

FIG. 4.0-8. Exploding-reflector modeling of zero-offset traveltimes associated with (a) a water bottom, (b) a flat, and (c)
a dipping reflector. (d) The superposition of the normal-incidence traveltime responses in (a), (b), and (c). Shown on the
velocity-depth models in the left-hand column are the normal-incidence rays used to compute the traveltime trajectories.
The time sections shown on the right-hand column are equivalent to a zero-offset traveltime section with the vertical axis in
two-way time.
470 Seismic Data Analysis

the traveltime trajectory can be computed by placing (b) a migration algorithm compatible with the strat-
sources along this interface and leaving the receivers egy,
where they were on the surface (Figure 4.0-8b). Finally, (c) appropriate parameters for the algorithm,
the same experiment can be repeated for the third inter- (d) issues concerning the input data, and
face (Figure 4.0-8c). To derive the composite response (e) migration velocities.
from the velocity-depth model in Figure 4.0-8d (the left-
hand column), individual responses shown in Figures Migration strategies include:
4.0-8a, 4.0-8b and 4.0-8c from each interface are su-
perimposed. The result is shown in Figure 4.0-8d (the
right-hand column). We can imagine that sources were (a) 2-D versus 3-D migration,
placed at all three interfaces and turned on simultane- (b) post- versus prestack migration, and
ously. Such an experiment would cause the rays emerg- (c) time versus depth migration.
ing from the three interfaces to be recorded at receivers
placed on the surface, along the line. The spectrum of migration strategies extend from 2-D
Actually, Figure 4.0-8d (the right-hand column) poststack time migration to 3-D prestack depth migra-
represents the modeled zero-offset traveltime section. tion. Depending on the nature of the subsurface geol-
Seismic wavefields, however, are represented not only ogy, any other in-between combination can be selected.
by wave traveltimes but also by wave amplitudes. Figure In practice, 2-D/3-D poststack time migration is used
4.0-9a shows the modeled zero-offset wavefield section most often for a good reason — it is the least sensitive
based on the same velocity-depth model in Figure 4.0- to velocity errors, and it often yields results acceptable
8d (the left-hand column). The shallow complex inter- for a reliable interpretation. Table 4-1 is an overview of
face (horizon 1 in Figure 4.0-8a) caused the complicated different migration strategies applied to different types
response of the two simple interfaces (horizons 2 and 3) of seismic data (2-D, 3-D, stacked, and unstacked).
in this zero-offset traveltime section. Choice of an appropriate migration strategy re-
How valid is the assumption that a stacked sec- quires input from the interpreter as to the structural
tion is equivalent to a zero-offset section? The conven- geology and stratigraphy in an area. Dipping events on a
tional CMP recording geometry provides the wavefield stacked section call for time migration. Conflicting dips
at nonzero offsets. During processing, we collapse the with different stacking velocities is one case in which a
offset axis by stacking the data onto the midpoint-time conventional stacked section differs from a zero-offset
plane at zero offset. For CMP stacking, we normally as- section. Thus, strictly speaking, poststack migration
sume hyperbolic moveout. Figure 4.0-10 shows selected
CMP gathers modeled from the velocity-depth model
in Figure 4.0-8d (the left-hand column). Because of the
presence of strong lateral velocity variations, the hy- Table 4-1. Migration strategies.
perbolic assumption may not be appropriate for some
reflections on some CMP gathers (Figure 4.0-10a); how- Case Migration
ever, it may be valid for others (Figure 4.0-10b). We ob-
tain a stacked section (Figure 4.0-9b) that resembles the dipping events time migration
zero-offset section (Figure 4.0-9a) to the extent that the
hyperbolic moveout assumption is valid. The assump- conflicting dips with prestack migration
tion that a conventional stacked section is equivalent to different stacking velocities
a zero-offset section also is violated to varying degrees
in the presence of strong multiples and conflicting dips 3-D behavior of 3-D migration
with different stacking velocities (Chapter 5). While mi- fault planes and salt flanks
gration of unstacked data is discussed in Chapter 5, our
main focus in this chapter is on migration after stack. Case Migration

strong lateral velocity depth migration


Migration Strategies variations associated with
complex overburden structures
In practice, migration of seismic data requires decision
complex nonhyperbolic moveout prestack migration
making with regards to:
3-D structures 3-D migration
(a) an appropriate migration strategy,
Migration 471

Therefore, a case of complex overburden structures calls


for depth migration before stacking the data.
Furthermore, complex overburden structures, en-
countered in areas with salt tectonics, overthrust tec-
tonics and irregular water-bottom topographies can
often exhibit 3-D characteristics. Thus, imaging such
structures may require 3-D prestack depth migration.
Field surveys are designed such that line orienta-
tions are, as much as possible, along the dominant strike
and dip directions, so as to minimize 3-D effects. Un-
der these circumstances, the 2-D assumption for migra-
tion can be acceptable. However, if the subsurface has
a truly 3-D geometry, without a dominant dip or strike
direction, then it is imperative to do 3-D migration of
3-D data. In such cases, 2-D migration (whether post-
stack or prestack, time or depth) can lead to potential
problems in interpretation.
A practical alternative to 2-D prestack depth mi-
gration can be a prestack layer replacement to cor-
FIG. 4.0-9. (a) The zero-offset wavefield section equivalent rect for the complex nonhyperbolic moveout followed
to the zero-offset traveltime section in Figure 4.0-8d (the by time migration after stack. This, however, is appli-
right-hand column); (b) the CMP stack generated from the cable to situations involving a single overburden layer,
CMP gathers as in Figure 4.0-10. (Modeling by Deregowski such as irregular water-bottom topography for it to be
and Barley, 1981.) reasonably practical.

which assumes that the stacked section is equivalent to Migration Algorithms


a zero-offset section is not valid to handle the case of
conflicting dips. Instead, one needs to do prestack time The one-way-in-depth scalar wave equation is the ba-
migration. sis for common migration algorithms. These algorithms
Conflicting dips often are associated with salt do not explicitly model multiple reflections, converted
flanks and fault planes, which can have 3-D characteris- waves, surface waves, or noise. Any such energy present
tics. This then requires 3-D prestack time migration. In in data input to migration is treated as primary re-
Section 5.3, we shall discuss a practical alternative to flections. Migration algorithms can be classified under
2-D prestack time migration strategies. The alternative three main categories:
sequence includes the application of normal-moveout
(NMO) correction using velocities appropriate for flat
(a) those that are based on the integral solution to the
events followed by 2-D dip moveout correction (DMO)
scalar wave equation,
to correct for the dip and source-receiver azimuth ef-
(b) those that are based on the finite-difference solu-
fects on stacking velocities. As a result, conflicting dips
tions, and
are preserved during stacking, and thus, imaging can be
(c) those that are based on frequency-wavenumber im-
deferred until after stacking using 2-D poststack time
plementations.
migration strategies. This series of processing steps is
largely equivalent to 2-D prestack time migration and
results often are comparable. The same workflow also is Whatever the algorithm, it should desirably:
applicable to 3-D prestack time migration (Section 7.4).
Accurate imaging of targets beneath complex (a) handle steep dips with sufficient accuracy,
structures with strong lateral velocity variations re- (b) handle lateral and vertical velocity variations, and
quires depth migration. Aside from the problem of con- (c) be implemented, efficiently.
flicting dips with different stacking velocities, strong lat-
eral velocity variations associated with complex over- Figure 4.0-11 is a migrated CMP stacked section
burden structures usually cause conventional stacking with a major unconformity. The undermigration — in-
based on the hyperbolic moveout assumption to fail. complete imaging of the unconformity, is not because of
472 Seismic Data Analysis

FIG. 4.0-10. Selected CMP gathers modeled from the velocity-depth model in Figure 4.0-8d (the left-hand column). (a)
Gathers from the complex part of the velocity-depth model, (b) gathers from the simpler part of the velocity-depth model.
CMP locations are indicated in Figure 4.0-8d. The CMP stack is shown in Figure 4.0-9b. (Modeling by Deregowski and Barley,
1981.)

erroneously too low velocities. Although we should al- obliquity factor (angle-dependency of amplitudes), and
ways be aware of velocity errors when migrating seismic the phase shift inherent in Huygens’ secondary sources
data, the undermigration in Figure 4.0-11 is the result (Section 4.1).
of using a dip-limited algorithm. By using a steep-dip Another migration technique (Claerbout and Do-
algorithm, we can achieve a more accurate imaging of herty, 1972) is based on the idea that a stacked section
the unconformity (Figure I-9). can be modeled as an upcoming zero-offset wavefield
The three principle migration techniques are dis- generated by exploding reflectors. Using the explod-
cussed in this chapter in their historical order of devel- ing reflectors model, migration can be conceptualized
opment as outlined below. The first migration technique as consisting of wavefield extrapolation in the form of
developed was the semicircle superposition method that downward continuation followed by imaging. To under-
was used before the age of computers. Then came stand imaging, consider the shape of a wavefield at ob-
the diffraction-summation technique, which is based on servation time t = 0 generated by an exploding reflec-
summing the seismic amplitudes along a diffraction hy- tor. Since no time has elapsed and, thus, no propagation
perbola whose curvature is governed by the medium ve- has occurred, the wavefront shape must be the same as
locity. The Kirchhoff summation technique introduced the reflector shape that generated the wavefront. The
later (Schneider, 1978), but actually in use earlier, basi- fact that the wavefront shape at t = 0 corresponds to
cally is the same as the diffraction summation technique the reflector shape is called the imaging principle. To
with added amplitude and phase corrections applied to define the reflector geometry from a wavefield recorded
the data before summation. These corrections make the at the surface, we only need to extrapolate the wave-
summation consistent with the wave equation in that field back in depth then monitor the energy arriving at
they account for spherical spreading (Section 1.4), the t = 0. The reflector shape at any particular extrapola-
Migration 473

FIG. 4.0-11. A CMP stack (a) before, and (b) after migration. Note the undermigration of the unconformity event (U)
caused by the use of a dip-limited algorithm.

tion depth directly corresponds to the wavefront shape and Benson (1986) combine theory with practice in the
at t = 0. field of migration with an emphasis on the frequency-
Downward continuation of wavefields can be im- wavenumber methods.
plemented conveniently using finite-difference solutions Another frequency-wavenumber migration is the
to the scalar wave equation. Migration methods based phase-shift method (Gazdag, 1978). This method is
on such implementations are called finite-difference mi- based on the equivalence of downward continuation to a
gration. Many different differencing schemes applied to phase shift in the frequency-wavenumber domain. The
the differential operators in the scalar wave equation imaging principle is invoked by summing over the fre-
exist both in time-space and frequency-space domains. quency components of the extrapolated wavefield at
Claerbout (1985) provides a comprehensive theoretical each depth step to obtain the image at t = 0.
foundation of finite-difference migration and its practi- A reason for the wide range of migration algorithms
cal aspects. used in the industry today is that none of the algorithms
After the developments on Kirchhoff summation fully meets the important criteria of handling all dips
and finite-difference migrations, Stolt (1978) introduced and velocity variations while still being cost-effective.
migration by Fourier transform. This method involves Migration algorithms based on the integral solu-
a coordinate transformation from frequency (the trans- tion to the scalar wave equation, commonly known as
form variable associated with the input time axis) to Kirchhoff migration, can handle all dips up to 90 de-
vertical wavenumber axis (the transform variable asso- grees, but they can be cumbersome in handling lateral
ciated with the output depth axis), while keeping the velocity variations.
horizontal wavenumber unchanged. The Stolt method Finite-difference algorithms can handle all types of
is based on a constant-velocity assumption. However, velocity variations, but they have different degrees of
Stolt modified his method by introducing stretching dip approximations. Furthermore, differencing schemes,
in the time direction to handle the types of velocity if carelessly designed, can severely degrade the intended
variations for which time migration is acceptable. Stolt dip approximation.
474 Seismic Data Analysis

FIG. 4.0-12. A CMP stack (a) before, and (b) after migration. Lack of any event to the right of the dotted line on the
migrated section is a result of the finite line length.

Finally, frequency-wavenumber algorithms have is the critical parameter in Kirchhoff migration. A small
limited ability in handling velocity variations, partic- aperture causes removal of steep dips; it generates spu-
ularly in the lateral direction. As a result of limi- rious horizontal events and organizes the random noise
tations of the three main categories of migration al- uncorrelated from trace to trace.
gorithms — integral, finite-difference, and frequency- Depth step size in downward continuation is the
wavenumber methods, migration software has expanded critical parameter in finite-difference methods. An opti-
further to additional extensions and combinations of the mum depth step size is the largest depth step with the
basic algorithms. Residual migration — phase-shift or minimum tolerable phase errors. It depends on tempo-
constant-velocity Stolt migration followed by the appli- ral and spatial samplings, dip, velocity, and frequency.
cation of a dip-limited algorithm is one example. It also depends on the type of differencing scheme used
in the algorithm.
Finally, the stretch factor is the critical parameter
Migration Parameters
for Stolt migration. A constant-velocity medium implies
After deciding on the migration strategy and the ap- a stretch factor of 1. In general, the larger the vertical
propriate algorithm, the analyst then needs to decide velocity gradient, the smaller the stretch factor needs
on the migration parameters. Migration aperture width to be.
Migration 475

Aspects of Input Data major unconformity represented by the strong, near-


horizontal reflection. Events A dipping up to the left
When migrating seismic data, one needs to be con- and B dipping up to the right cross over one another
cerned with various aspects of the input data set itself: on sections that correspond to 90 percent and 95 per-
cent of stacking velocities, indicating undermigration.
(a) line length or areal extent, The same events are split away from one another in
(b) signal-to-noise ratio, and opposite directions on sections that correspond to 105
(c) spatial aliasing. percent and 110 percent of stacking velocities. The most
overall acceptable image is seen on the section that cor-
The line length must be sufficient to allow a steeply responds to the 100 percent of stacking velocities.
dipping event to migrate to its true subsurface location. Accuracy in event positioning after migration actu-
It is a fatal error to record short profiles in areas with ally depends on the combined effects of the performance
complex geology. Also, for 3-D migration, the surface of the migration algorithm used and the velocity errors.
areal extent of a 3-D survey is almost always larger than For example, the inherently undermigrating character
the target subsurface areal extent. of a 45-degree finite-difference algorithm can be, for an
Random noise at late times on a stacked section, event with a specific dip, coincidentally counterbalanced
when migrated, potentially can be hazardous for shal- by the overmigration effect of erroneously too high ve-
lower data. One may have to compromise on migration locities. In the presence of large vertical velocity gradi-
aperture for deep data to prevent this problem to occur. ents, a two-pass 3-D migration can also cause overmi-
Trace spacing must be small enough to prevent gration of steep dips (in the form of lateral translation)
spatial aliasing of steep dips at high frequencies. Al- even with the correct migration velocities.
though this is not an issue for modern prestack data, a Figure 4.0-14a shows a portion of a migrated
coarse shot-receiver spacing can degrade the fidelity of stacked section. Although this section does not contain
prestack migration severely. Old data and 3-D marine steep dips, accurate imaging of the faults along the low-
data in the crossline direction often are trace interpo- relief structures can be important to the interpreter.
lated prior to poststack migration.
Note the slight undermigration, which may be caused
Figure 4.0-12 is a CMP stacked section before and
by any of the following: (a) error in migration veloci-
after migration. From an interpretation viewpoint, the
ties, (b) a dip-limited algorithm that failed to focus the
reliable part of this migrated section is confined to the
upper central part. Lack of any reflection energy to the diffraction energy adequately, or (c) a possible 3-D be-
right of the dotted line does not mean that there is a havior of the geometry of the reflector. The section in
structural discontinuity there. It only means that the Figure 4.0-14a has been migrated with a dip-limited al-
reflections associated with the imbricate structure have gorithm. Using a proper algorithm, with the same veloc-
been migrated in the updip direction from right to left. ities, we get the migrated section in Figure 4.0-14b. The
As a result, a zone with no reflectors to the right of the resulting section shows slight overmigration, which can
dotted line is left behind because of the truncation of be attributed to errors in migration velocities. Lowering
the wavefield represented by the right-hand edge of the the velocities gives the improved, but not completely ac-
stacked section. The deeper part is useless, because the curate, image in Figure 4.0-14c. Perhaps, the remaining
noise dominates the section. issues in imaging may be attributed to 3-D effects.
As demonstrated by the example in Figure 4.0-14,
migration results generally are self-evident — under-
Migration Velocities and overmigration often can be recognized on a mi-
grated section. Problems in imaging often are traced to
accuracy in migration velocities. I consider migration
Horizontal displacement during migration is propor-
tional to migration velocity squared (equation 4-1). velocities as the weak link between the seismic section
Since velocities generally increase with depth, errors in and the geologic cross-section.
migration are usually larger for deep events than shal- In the next section, basic principles of migration
low events. Also, the steeper the dip, the more accurate are presented and the Kirchhoff summation, finite-
the migration velocities need to be, since displacement difference, frequency-space, and frequency-wavenumber
is proportional to dip. algorithms are reviewed. Practical aspects of the migra-
Figure 4.0-13 shows a portion of a CMP-stacked tion algorithms are expounded in Sections 4.2 through
section after 3-D poststack time migration using a per- 4.5. Specifically, key parameters for each category of
cent range of stacking velocities. Note the subtle under- the migration algorithms are analyzed using appropri-
and overmigration effects on dipping events below the ate synthetic and field data examples. Further aspects of
476 Seismic Data Analysis

FIG. 4.0-14. (a) A portion of a migrated CMP stack; note


the subtle undermigration at fault locations A and B caused
by the use of a dip-limited algorithm; (b) same data set but
migrated with an algorithm with no dip limitation; note
the subtle overmigration most likely due to erroneously too
high velocities; (c) same data set migrated with the same
algorithm as in (b) but with velocities adjusted to prevent
overmigration.

FIG. 4.0-13. A portion of a CMP-stacked section after 3-


D poststack time migration using, from top to bottom, 90, 4.1 MIGRATION PRINCIPLES
95, 100, 105 and 110 percent of stacking velocities. Note the
subtle under- and overmigration effects on dipping events
below the major unconformity represented by the strong, Consider the dipping reflector CD of the simple ge-
near-horizontal reflection. Note the effect of velocities used ologic section in Figure 4.1-1a. We want to obtain a
in migration on the positioning of the event A dipping up zero-offset section along the profile Ox. As we move the
to the left and event B dipping up to the right. source-receiver pair (s, g) along Ox, the first normal-
incidence arrival from the dipping reflector is recorded
at location A. In this discussion, we assume a normal-
migration in practice, including spatial aliasing, migra- ized constant-velocity medium v = 1 so that time and
tion response to random noise, line length, and irregular depth coordinates become interchangeable. The reflec-
topography are discussed in Section 4.6. The problem tion arrival at location A is indicated by point C on the
of conflicting dips with different stacking velocities that zero-offset time section in Figure 4.1-1b. As we move
requires dip-moveout (DMO) correction and prestack from location A to the right, normal-incidence arrivals
time migration, and the accompanying topic on migra- are recorded from the dipping reflector CD. The last
tion velocity analysis are deferred until Chapter 5. The arrival is recorded at location B, which is indicated by
problem of imaging beneath complex structures that re- point D in Figure 4.1-1b. In this experiment, diffrac-
quires earth imaging and modeling in depth is discussed tions off the edges of reflector CD are excluded to sim-
in Chapters 8 and 9, respectively. plify the discussion.
Migration 477

(c) Migration moves reflectors in the updip direction.

The example in Figure 4.0-1 demonstrates the


above observations. In particular, the dipping event (B)
has moved in the updip direction, become shorter, and
steepened after migration (A).
As mentioned in the previous section, conventional
migration output is displayed in time, as is the input
stacked section. To distinguish the two time axes, we
will denote the time axis on the stacked section as t
— event time in the unmigrated position, and the time
axis on the migrated section as τ — event time in the
migrated position.
We shall now examine the horizontal and verti-
cal displacements as seen on the migrated time section.
From Figure 4.1-2, consider a reflector segment CD. As-
sume that CD migrates to C D and that point E on
C D migrates to point E on CD. The horizontal and
vertical (time) displacements — dx and dt , and the dip
∆τ /∆x, all measured on the migrated time section (Fig-
ure 4.1-2), can be expressed in terms of medium velocity
v, traveltime t, and apparent dip ∆t/∆x as measured
on the unmigrated time section (Figure 4.1-2). Chun
and Jacewitz (1981) derived the following formulas:
v 2 t ∆t
dx = , (4 − 1)
4 ∆x

2
v∆t
dt = t 1 − 1− , (4 − 2)
FIG. 4.1-1. Migration principles: The reflection segment 2∆x
C D in the time section as in (b), when migrated, is moved
updip, steepened, shortened, and mapped onto its true sub- ∆τ ∆t 1
surface location CD as in (a). (Adapted from Chun and = . (4 − 3)
∆x ∆x 2
Jacewitz, 1981.) v∆t
1−
2∆x
Compare the geologic section in Figure 4.1-1a,
which is in depth, with the zero-offset seismic section To gain a quantitative insight into these expres-
in Figure 4.1-1b, which is in time. The true subsurface sions, we consider a numerical example. For a realis-
position of reflector CD is superimposed onto the time tic velocity function that increases with depth, consider
section for comparison. Clearly the true geologic posi- five reflecting segments at various depths. For simplic-
tion of reflector CD is not the same as the reflection ity, assume that quantity ∆t/∆x is the same for all (10
event position C D . ms per 25-m trace spacing). From equations (4-1), (4-
From this simple geometric construction, note that 2), and (4-3), compute the horizontal and vertical dis-
the reflection in the time section C D must be migrated placements dx and dt and the dips (in ms/trace) after
to its true subsurface position CD in the depth section. migration. The results are summarized in Table 4-2.
The following observations can be made from the geo- Refer to Table 4-2 and equations (4-1), (4-2), and
metric description of migration in Figure 4.1-1: (4.3) and make the following observations:

(a) The dip angle of the reflector in the geologic section (a) The time dip ∆τ /∆x on the migrated section is
is greater than in the time section; thus, migration always greater than the time dip ∆t/∆x on the
steepens reflectors. unmigrated section.
(b) The length of the reflector, as seen in the geologic (b) The horizontal displacement dx increases with
section, is shorter than in the time section; thus, event time t in the unmigrated position. At 4 s,
migration shortens reflectors. the horizontal displacement is more than 6 km.
478 Seismic Data Analysis

FIG. 4.1-2. Quantitative analysis of the migration process. Dipping event AB on the unmigrated section (left) is moved to
A B on the migrated section (right). The event after migration also is superimposed on the unmigrated section to compare the
position of the event before and after migration. Point C on dipping reflector AB is moved to C after migration. The amount
of horizontal displacement dx , vertical displacement dt , and the dip ∆τ /∆x after migration is calculated from equations (4-1),
(4-2) and (4-3).

Table 4-2. Horizontal and vertical displacements as a In Figure 4.1-1a, assume that the zero-offset sec-
result of migration of a series of dipping reflections with tion was recorded only between surface locations A and
the same apparent dip (10 ms/trace) as measured on B. The time section would include the event C D , but
the unmigrated stacked section at various depths, and when migrated, the event would migrate out of the sec-
changes in dip angle as measured on the migrated sec- tion, resulting in a blank migrated section (Figure 4.1-
tion in time. 1b). Therefore, the data on a stacked section are not
necessarily confined to the subsurface below the seismic
t v dx dt ∆t/∆x ∆τ /∆x
line. The converse is even more important; the structure
(s) (m/s) (m) (s) (ms/trace) (ms/trace)
below the seismic line may not be recorded on the seis-
mic section. Suppose that the data were recorded only
1 2500 625 0.134 10 11.5
between surface locations O and A. This time, the re-
2 3000 1800 0.400 10 12.5
sulting time section would be blank. So, we should not
3 3500 3675 0.858 10 14.0
record between A and B, and neither should we record
4 4000 6400 1.600 10 16.7
between O and A. Instead, we should record between O
5 4500 10125 2.820 10 23.0
and B in order to record the reflector of interest prop-
erly and also to migrate it properly.
In areas with a structural dip, line length must be
(c) The horizontal displacement dx is a function of the chosen by considering the horizontal displacements of
velocity squared. If there is a 20 percent error in dipping events from the structures causing the events.
the velocity used in migration, then the event is This is an important consideration, especially in 3-D
misplaced by an error of 44 percent. seismic work. The areal surface coverage of a survey
(d) The vertical displacement dt also increases with usually is larger than the areal subsurface coverage of
time and velocity. interest.
(e) The steeper the event dip, the more the horizontal To achieve a complete image of a dipping reflec-
and vertical displacements after migration. tor, also, the recording time must be long enough. For
Migration 479

FIG. 4.1-3. (a) A zero-offset section modeled from the dipping reflectors shown in (b). The medium velocity is constant 3500
m/s and the trace spacing is 25 m. The true dip angles of the reflectors vary from 0 to 45 degrees at 5-degree increment.
Migration of the zero-offset section (a) yields the model of the dipping reflectors (b).

FIG. 4.1-4. A portion of a CMP stack (a) before and (b) after migration. Note the group of events with a range of dips that
fan out from a fault plane. Migration has moved them in the updip direction, made them shorter and steeper.
480 Seismic Data Analysis

FIG. 4.1-6. (a) A velocity-depth model consisting of a syn-


clinal reflector; (b) selected normal-incidence arrivals on the
zero-offset section. Trace the bowtie in the time section.

example, if only OE seconds were recorded (Figure 4.1-


1), then the recorded segment C D would yield only
part of the complete image CD. An excellent example of
recording deeper in time and with longer line length for
steeper dips is shown in Figure 4.0-1. Proper imaging of
the salt dome boundary required that data be recorded
for more than 6 s.
The migration concepts described above are
demonstrated further by the dipping events model in
Figure 4.1-3. The edge diffractions are included here.
The dipping reflectors on the zero-offset section are
steepened, shortened, and moved in the updip direc-
tion as a result of migration. A field data example of a
series of dipping events on a stacked section before and
FIG. 4.1-5. Curved reflecting interfaces (synclines and an- after migration is shown in Figure 4.1-4. Note that the
ticlines) (a) before and (b) after migration. See text for de- steeper the dip, the more the event moves after migra-
tails. (Modeling courtesy Union Oil Company.) tion.
So far, only linear reflectors were considered. We
now consider a more realistic geologic situation that
involves curved reflecting interfaces. Figure 4.1-5 shows
Migration 481

FIG. 4.1-7. A portion of a CMP stack (a) before and (b) after migration. Anticlines appear bigger, while synclines appear
smaller than their actual sizes on the unmigrated section (a).

three synclines and a small anticlinal feature. The syn- time section. Complete the procedure by tracing the
clines appear as bowties on the zero-offset section. By traveltime trajectory in Figure 4.1-6b.
using the principles deduced from the geometry of Fig- Two field data examples containing synclinal and
ure 4.1-1, note that as a result of migration, segment anticlinal structures are shown in Figures 4.1-7 and
A of the bow tie moves in the updip direction to the 4.1-8. In Figure 4.1-7, note that the synclinal feature
left. Similarly, segment B moves to the right, while flat- broadens and the anticlinal feature narrows as a result
topped segment C does not move much at all. Conse- of migration. In Figure 4.1-8, the bow ties associated
quently, after migration the flanks of bow ties associated with two small synclinal basins A and B grow larger in
with synclines are opened up. On the other hand, the depth. After migration, the bowties are untied and the
small anticline seems to be broader on the zero-offset synclines are delineated.
section than it is on the migrated section. Again note
that segment D moves updip to the right, while segment
Kirchhoff Migration
E moves updip to the left as a result of migration. Thus,
synclines broaden and anticlines compress as a result of
Claerbout (1985) uses the harbor example in Figure 4.1-
migration. Migration velocities also affect the apparent 9 to describe the physical principles of migration. As-
size of the structure; higher velocities mean more mi- sume that a storm barrier exists at some distance z3
gration and, hence, smaller anticlinal structure. from the beach and that there is a gap in the barrier.
Why does a syncline look like a bowtie on the Imagine a calm afternoon breeze that comes from the
stacked section? The answer is in Figure 4.1-6, where ocean as a plane incident wave. The wavefront is par-
a symmetric syncline is modeled. Given the subsurface allel to the storm barrier. As we walk along the beach
model in Figure 4.1-6a, the normal-incidence rays can line, we see a wavefront different from a plane wave.
be computed to derive the zero-offset traveltime section The gap on the storm barrier has acted as a secondary
in Figure 4.1-6b. Only five CMP locations are shown source and generated the semicircular wavefront that is
for clarity. At locations 2 and 4, there are two distinct propagating toward the beach.
arrivals, while at location 3, there are three distinct ar- If we did not know about the storm barrier and the
rivals. By filling in the intermediate raypaths, the bow gap, we might want to lay out a receiver cable along the
tie character of the syncline can be constructed on the beach to record the approaching waves. This experiment
482 Seismic Data Analysis

FIG. 4.1-8. A portion of a CMP stack (a) before and (b) after migration. Migration unties the bowties and turns them into
synclines below A and B.

is illustrated in Figure 4.1-10 with the recorded time (the points on the reflecting interface) get closer to each
section. Physicists call the gap on the barrier a point other, superposition of the hyperbolas produces the re-
aperture. It is somewhat similar to a point source, since sponse of the actual reflecting interface (Figure 4.1-13).
both generate circular wavefronts. However, the ampli- In terms of the harbor example, this is like assuming
tudes on the wavefront that propagate outward from that the barrier is wiped out by a storm so that the
a point source are isotropic, while those from a point primary incident plane wave reaches the beach with-
aperture are angle-dependent. The point aperture on out modification. The diffraction hyperbolas, which are
the barrier acts as a Huygens’ secondary source. caused by sharp discontinuities at both ends of the re-
From the beach experiment, we find that Huygens’ flector in Figure 4.1-13, remain. These hyperbolas are
secondary source responds to a plane incident wave and equivalent to diffractions seen at fault boundaries on
generates a semicircular wavefront in the x − z plane. stacked sections.
The response in the x − t plane is the diffraction hyper- In summary, we find that reflectors in the subsur-
bola shown in Figure 4.1-11. face can be visualized as being made up of many points
Imagine that the subsurface consists of points along that act as Huygens’ secondary sources. We also find
each reflecting horizon that behave much as the gap that the zero-offset section consists of a superposition
on the storm barrier. From Figure 4.1-12, these points of the many hyperbolic traveltime responses. Moreover,
act as Huygens’ secondary sources and produce hyper- when there are discontinuities (faults) along the reflec-
bolic traveltime trajectories. Moreover, as the sources tor, diffraction hyperbolas often stand out.
Migration 483

FIG. 4.1-11. A point that represents a Huygens’ secondary


source (a) produces a diffraction hyperbola on the zero-offset
FIG. 4.1-9. The gap in the barrier acts as Huygens’ sec- time section (b). The vertical axis in this section is two-way
ondary source, causing the circular wavefronts that approach time, while the vertical axis in the time section in Figure
the beach line. (Adapted from Claerbout, 1985.) 4.1-10 is one-way time.

FIG. 4.1-12. Superposition of the zero-offset responses (b)


of a discrete number of Huygens’ secondary sources as in
(a).

FIG. 4.1-10. Waves recorded along the beach generated by


Huygens’ secondary source (the gap in the barrier in Figure FIG. 4.1-13. Superposition of the zero-offset responses (b)
4.1-9) have a hyperbolic traveltime trajectory. of a continuum of Huygens’ secondary sources as in (a).
484 Seismic Data Analysis

Diffraction Summation

Huygens’ secondary source signature is a semicircle in


the x − z plane and a hyperbola in the x − t plane.
This characterization of point sources in the subsurface
leads to two practical migration schemes. Figure 4.1-
14a shows a zero-offset section that consists of a single
arrival at a single trace. This event migrates to a semi-
circle (Figure 4.1-14b). From Figure 4.1-14, note that
the zero-offset section recorded over a constant-velocity
earth model consisting of a semicircular reflecting inter-
face contains a single blip of energy at a single trace as
in Figure 4.1-14a. Since this recorded section consists of
an impulse, the migrated section in Figure 4.1-14b can
be called the migration impulse response. An alternate
scheme for migration results from the observation that
a zero-offset section consisting of a single diffraction hy-
perbola migrates to a single point (Figure 4.1-15b).
The first method of migration is based on the su-
perposition of semicircles, while the second method is
FIG. 4.1-14. Principles of migration based on semicircle
based on the summation of amplitudes along hyperbolic
superposition. (a) Zero-offset section (trace interval, 25 m;
paths. The first method was used before the age of dig- constant velocity, 2500 m/s), (b) migration.
ital computers. The second method, which is known as
the diffraction summation method, was the first com-
puter implementation of migration.
The migration scheme based on the semicircle su-
perposition consists of mapping the amplitude at a sam-
ple in the input x − t plane of the unmigrated time sec-
tion onto a semicircle in the output x − z plane. The
migrated section is formed as a result of the superposi-
tion of the many semicircles.
The migration scheme based on diffraction sum-
mation consists of searching the input data in the x − t
plane for energy that would have resulted if a diffract-
ing source (Huygens’ secondary source) were located at
a particular point in the output x−z plane. This search
is carried out by summing the amplitudes in the x − t
plane along the diffraction curve that corresponds to
Huygens’ secondary source at each point in the x − z
plane. The result of this summation then is mapped
onto the corresponding point in the x − z plane. As
noted early in this section, within the context of time
migration, however, the summation result actually is
mapped onto the x − τ plane, where τ is the event time
in the migrated position.
The curvature of the hyperbolic trajectory for am-
plitude summation is governed by the velocity function.
The equation for this trajectory can be derived from the
FIG. 4.1-15. Principles of migration based on diffraction
geometry of Figure 4.1-15. A formal derivation also is summation. (a) Zero-offset section (trace interval, 25 m; con-
provided in Section D.2. Assuming a horizontally lay- stant velocity, 2500 m/s), (b) migration. The amplitude at
ered velocity-depth model, the velocity function used input trace location B along the flank of the traveltime hy-
to compute the traveltime trajectory is the rms veloc- perbola is mapped onto output trace location A at the apex
ity at the apex of the hyperbola at time τ (Section 3.1). of the hyperbola by equation (4-4).
Migration 485

From the triangle COA in Figure 4.1-15a, we note that and frequency characteristic. Otherwise, there would be
2 no amplitude cancelation when they are close to one an-
4x
t2 = τ 2 + 2
. (4 − 4) other. The waveform that results from the summation
vrms
must be restored in both phase and amplitude.
Having computed the input time t, the amplitude In summary, we must consider the following three
at input location B is placed on the output section at factors before diffraction summation:
location A, corresponding to the output time τ at the
apex of the hyperbola. (a) The obliquity factor or the directivity factor, which
From Section 3.1, reflection traveltimes in a layered describes the angle dependence of amplitudes and
earth approximate small-spread hyperbolas. This may is given by the cosine of the angle between the di-
seem to impose a serious restriction on the aperture rection of propagation and the vertical axis z (Fig-
width — the lateral extent of the diffraction hyperbola, ure 4.1-15).
in the summation process. However, the small-spread (b) The spherical spreading factor, which is propor-
approximation is valid even at large distances from the tional to 1/vr for 2-D wave propagation, and
apex, and the errors associated with it are insignificant (1/vr) for 3-D wave propagation.
at late times. In practice, this approximation is not usu- (c) The wavelet shaping factor, which is designed with
ally an issue. a 45-degree constant phase spectrum and an am-
plitude spectrum proportional to the square root of
Amplitude and Phase Factors the frequency for 2-D migration. For 3-D migration,
the phase shift is 90 degrees and the amplitude is
Now consider several factors associated with the am- proportional to frequency.
plitude and phase behavior of the waveform along the
diffraction hyperbola. From Figure 4.1-9, given the al-
ternative of standing at location A or B, we intuitively Kirchhoff Summation
think that it is safer to stand at location B. This is The diffraction summation that incorporates the obliq-
because the wave amplitude at location A, which is on uity, spherical spreading and wavelet shaping factors
the z-axis, is stronger than the wave amplitude at loca- is called the Kirchhoff summation, and the migration
tion B, which is at an oblique angle from the z-axis. As method based on this summation is called the Kirchhoff
mentioned earlier, this is one difference between a point migration. To perform this method, multiply the input
source with uniform amplitude response at all angles data by the obliquity and spherical spreading factors.
and the point aperture that produces a wavefront with Then apply the filter with the above specifications and
angle-dependent amplitudes. This angle dependence of sum along the hyperbolic path that is defined by equa-
amplitudes, which is described by the obliquity factor, tion (4-4). Place the result on the migrated section at
should be considered before summation. To correct for time τ corresponding to the apex of the hyperbola. In
the obliquity factor, the amplitude at location B in Fig- practice, the order of the filter application, specified by
ure 4.1-15a is scaled by the cosine of the angle between factor (c), and summation can be interchanged without
BC and CA before it is placed at output location A. sacrificing accuracy because the summation is a linear
Another factor is the spherical divergence of wave process and the filter is independent of time and space.
amplitudes. Again, from Figure 4.1-9, given the alterna- The velocity used in equation (4-4) is taken as the
tive of standing at location B or C, we prefer to stand at rms velocity, which can be allowed to vary laterally.
location C. The reason for this is that the wave ampli- However, lateral variation in velocity distorts the hy-
tude along the wavefront at location C, which is farther perbolic nature of the diffraction pattern and somehow
from the point aperture source, is weaker than the wave must be considered. The value for the rms velocity typi-
amplitude at location B. Wave energy decays as (1/r2 ), cally is that of the output time sample; that is, the apex
where r is the distance from the source to the wavefront, time τ of the hyperbola.
while amplitudes decay as (1/r). Thus, amplitudes must What was determined from a physical point of view
be scaled by factor (1/r) before summation for wave in the preceding discussion can be rigorously described
propagation in three dimensions. by the integral solution to the scalar wave equation.
Finally, there is a third factor that involves the Schneider (1978), Berryhill (1979) and Berkhout (1980)
inherent property of Huygens’ secondary source wave- are excellent references for the mathematical treatment
form. This factor is difficult to explain from a physical of the Kirchhoff migration method. The integral solu-
viewpoint. Nevertheless, it is obvious from Figure 4.1- tion of the scalar wave equation yields three terms; the
13 that Huygens’ secondary sources must respond as a far-field term which is proportional to (1/r), and two
wavelet along the hyperbolic paths with a unique phase other terms which are proportional to (1/r2 ). Hence,
486 Seismic Data Analysis

it is the far-field term that makes the most contribu- wavefield recorded along the beach (Figure 4.1-16a). As-
tion to the summation that is used in practical imple- sume that the barrier is 1250 m from the beach. Now
mentation of Kirchhoff migration. The output image move the recording cable into the water, 250 m from the
Pout (x0 , z = vτ /2, t = 0) at a subsurface location (x0 , z) beach. Start recording at the instant the plane wave hits
using only the far-field term is computed from the 2-D the barrier. The recorded section is shown in Figure 4.1-
zero-offset wavefield Pin (x, z = 0, t), which is measured 16b. Move the cable 500 m from the beach and record
at the surface (z = 0), by the following summation over the section in Figure 4.1-16c, followed by a recording
a spatial aperture 750 m from the beach to obtain the section in Figure
∆x cos θ 4.1-16d. Finally, 1000 m from the beach, record the sec-
Pout = √ ρ(t) ∗ Pin , (4 − 5) tion shown in Figure 4.1-16e.
2π x
vrms r
Note that each recording yields a hyperbola in
where vrms is the rms velocity at the output point which the apex moves closer to zero time. The actual
(x0 , z) and r = (x − x0 )2 + z 2 , which is the distance extent of the recording cable is denoted by the solid line
between the input (x, z = 0) and the output (x0 , z) on top of each frame. Had we recorded at the barrier
points. The asterisk denotes convolution of the rho fil- (1250 m from the beach), the apex of the hyperbola
ter ρ(t) with the input wavefield Pin . would be positioned at t = 0.
The rho filter ρ(t) corresponds to the time deriva- In Kirchhoff migration, the diffraction hyperbola
tive of the measured wavefield, which yields the 90- is collapsed by summing the amplitudes, then placing
degree phase shift and adjustment of the amplitude them at the apex. The alternative approach implied
spectrum by the ramp function ω of frequency (Ta- by the result of the experiment shown in Figure 4.1-
ble A-1 of Appendix A). For 2-D migration, the half- 16 is to use the hyperbola recorded a distance away
derivative of the wavefield is used. This is equivalent from the beach to construct the hyperbola that would
to the 45-degree phase shift and adjustment of the am- be recorded at another distance closer to the source of
plitude
√ spectrum by a function of frequency defined as the diffraction hyperbola. The process is stopped when
ω. Since the rho filter is independent of the spatial the hyperbola collapses to its apex. In the harbor ex-
variables, it actually can be applied to the output of the periment, this collapse occurs when the receiver cable
summation in equation (4-5). Finally, the far-field term
coincides with the barrier, or, equivalently, when t = 0.
in equation (4-5) is proportional to the cosine of the an-
As stated in the introductory section, this is called the
gle of propagation (the directivity term or the obliquity
imaging principle.
factor) and is inversely proportional to vr (the spherical
spreading term) in three dimensions.
√ In two dimensions,
the spherical spreading term is vr.
Equation (4-5) can be used to compute the wave- Downward Continuation
field at any depth z. The ouput image Pout is computed
at (x0 , z = vτ /2, t = 0) using the input wavefield Pin The harbor experiment described above can be simu-
at (x, z = 0, t − r/v). To obtain the migrated section at lated in the computer. Pretend that moving the receiver
an output time τ , equation (4-5) must be evaluated at cable from the beach into the water closer to the barrier
z = vτ /2 and the imaging principle must be invoked by is like moving the receiver cable from the surface down
mapping amplitudes of the resulting wavefield at t = 0 into the earth closer to the reflectors. Think of the gap
onto the migrated section at output time τ . The com- on the barrier as equivalent to a point diffractor on a re-
plete migrated section is obtained by performing the flecting interface causing the diffraction hyperbola (Fig-
summation in equation (4-5) and setting t = 0 for each ure 4.1-17a). Start with the wavefield recorded at the
output location. The range of the summation is called surface and move the receivers down to depth levels at
the migration aperture. finite intervals. Downward continuation of the upcom-
ing wavefield at the surface, therefore, can be considered
equivalent to lowering the receivers into the earth.
Finite-Difference Migration The computer-simulated wavefields at these differ-
ent depths are shown in Figure 4.1-17. By applying the
To describe the physical basis of finite-difference migra- imaging principle at each depth, the entire wavefield is
tion, recall the harbor example of Figure 4.1-9. Instead imaged. The final output from this process is the mi-
of taking the section recorded along the beach, which grated section. The last section (panel f ) at 1250 m
contains the diffraction hyperbola, then collapsing it to has only one arrival at t = 0. The recording cable is
get the migrated section in Figure 4.1-15, consider the on the storm barrier and the arrival from the gap oc-
following alternative procedure. Again, start with the curs at t = 0. As the cable moved into the ocean and
Migration 487

FIG. 4.1-16. Moving the receiver cable in the harbor experiment (Figure 4.1-9) from the beach into the water at discrete
intervals parallel to the beach line. Numbers on top indicate the distance of the receiver cable from the beach line.

FIG. 4.1-17. Computer simulation of the experiment illustrated in Figure 4.1-16. Here, we downward continue the receivers
at discrete depth intervals. The numbers on top indicate the distance of the receiver cable from the surface, z = 0.

recorded closer to the barrier, the recorded diffraction at a time that is equivalent to the distance from the sur-
hyperbola arrived earlier, and became shorter and more face to the diffractor as shown in Figure 4.1-18b. This
compressed. It collapsed to a point when the receivers is called time retardation.
coincided with the storm barrier over which the source Reconsider the results from the computer simula-
point forms a gap. tion of the harbor experiment in Figure 4.1-17. Sup-
There is one important difference between the pose we stopped recording at a depth of 1000 m before
physical experiment in Figure 4.1-16 and the computer-
simulated downward-continuation experiment in Figure
4.1-17. The receiver cable is the same at each step in
Figure 4.1-16, whereas the effective cable length gets
shorter and shorter toward the source (the gap in the
barrier) in Figure 4.1-17. This is because we started by
recording the wavefield at the surface (Figure 4.1-16a)
with a finite cable length. The recorded information is
confined to within the two raypaths depicted on the
section in Figure 4.1-17a. As the cable moves closer to
the source, the effective receiver cable containing the
information is confined to smaller and smaller lengths.
Although receivers are lowered vertically, energy moves
down along raypaths it originally took on the way up.
To relate these recordings at different depths (Fig- FIG. 4.1-18. (a) Superposition of the time sections in Fig-
ure 4.1-17), we superimpose them as shown in Figure ure 4.1-17; (b) removing the translational effect by retar-
4.1-18a. Moreover, the recordings can be shifted so that dation to place the energy at the apex of the hyperbola
the apexes of the hyperbolas coincide and are positioned obtained initially along the beach line.
488 Seismic Data Analysis

the barrier. The original hyperbola in Figure 4.1-17a Table 4-3. Simple extrapolation of a data vector in
was partially collapsed at this depth (Figure 4.1-17e). time.
Therefore, downward continuing to a depth short of the
true depth of the source causes undermigration. Diffrac- Operator Data Time Step
tions and dipping events also are undermigrated if in- -1.1 100 0
correctly low velocities are used for migration.
1.0 x 1
Assume that the recording continues and passes
beyond barrier position z3 (Figure 4.1-9). We infer that
the focused energy on the section at this depth (Figure 100 0
4.1-17f) would propagate through the focal point and -1.1 110 1
turn into hyperbolas that are the mirror images of those 1.0 x 2
in Figures 4.1-17a through 4.1-17e. We have downward
continued more than necessary. This yields overmigra- 100 0
tion, which also is caused by incorrectly high velocities. 110 1
From these observations, note that downward continu- -1.1 121 2
ing to a wrong depth is like downward continuing with 1.0 x 3
the wrong velocity (Doherty and Claerbout, 1974).
Another important issue to consider is how often
the extrapolated wavefield should be computed. When
going from one frame to another in Figure 4.1-17, what Equation (4-6a) is generalized as
should the depth step size be? This is discussed in detail
later in Section 4.3. (1.0) × P (t + 1) + (−1.1) × P (t) = 0, (4 − 7a)

which is rewritten in the form

Differencing Schemes P (t + 1) − P (t) = (0.1) × P (t), (4 − 7b)

where t is the time variable and P is the quantity being


Finite-difference migration algorithms are based on dif-
ferential solutions to the scalar wave equation that are extrapolated. Instead of defining the time interval as
used to downward continue the input wavefield recorded one unit, we can define it as an arbitrary increment
at the surface. A simple numerical example illustrates of time ∆t. Also, assume that the inflation rate is a.
the finite-difference method of solving differential equa- Equation (4-7b) then takes the more general form
tions (Claerbout, 1985). Assume that you have $100
P (t + ∆t) − P (t) = a P (t). (4 − 8a)
today. Given an annual inflation rate of 10 percent, for
the same buying power next year, you need $110. A Alternatively, we could use the average of the
computer algorithm can determine the face value of the present and future values on the right side of this equa-
present $100 in future years. Table 4-3 shows the results tion:
of extrapolation from one year to the next.
a
Given the present value, 100, find future values in P (t + ∆t) − P (t) = P (t + ∆t) + P (t) . (4 − 8b)
the data column. The following equation solves for the 2
unknown x: Equations (4-8a) and (4-8b) now can be put into
(1.0) × x + (−1.1) × (100) = 0, (4 − 6a) the form of equation (4-6a) as

which yields x = 110. We used a two-point operator and P (t + ∆t) + −1 − a P (t) = 0, (4 − 9a)
aligned it with the data column as indicated in Table
and
4-3. Similarly, we have
a a
(1.0) × x + (−1.1) × (110) = 0, (4 − 6b) 1− P (t + ∆t) + −1 − P (t) = 0. (4 − 9b)
2 2
which yields x = 121. By using the new value for x, we
obtain By using either equation (4-9a) or equation (4-9b), we
compute the future values of P (t) from a given initial
(1.0) × x + (−1.1) × (121) = 0, (4 − 6c) value as shown in Table 4-4.
which yields x = 133, and so on. By moving the oper- The operator in which the coefficient of the future
ator down in the time direction as shown in Table 4-3, value P (t + ∆t) is unity is called the explicit operator.
we extrapolate the data column into the future. Stability of the finite-difference solution — the problem
Migration 489

Table 4-4. Application of two-point implicit and ex- various aspects of the finite-difference migration meth-
plicit operators to extrapolate data P from t to t + ∆t. ods.

Explicit Implicit Data


Operator Operator Column
Rational Approximations for Implicit Schemes
−1 − a −1 − a/2 P (t)
The scalar wave equation is a two-way wave equation in
1 1 − a/2 P (t + ∆t) depth that describes propogation of both upcoming and
downgoing waves. If we consider the resulting wavefield
from the exploding reflectors model as the upcoming
waves, then we are really interested in a one-way wave
of wave amplitudes growing from one extrapolation step equation to downward continue the upcoming waves. In
to another, can be an issue with this type of operator fact, we normally use some rational approximation to
(Section D.6) An implicit operator produces stable re-
the one-way wave equation in finite-difference imple-
sults because of averaging on the right side of equation
mentations.
(4-9b), known as the Crank-Nicolson scheme. For the
To get the actual differential equation to be used
differential equations used in finite-difference migration
in downward extrapolation of the upcoming waves, and
algorithms, such as the parabolic equation described in
therefore to perform a finite-difference migration, the
Section D.3, scalar a becomes a matrix coefficient. Im-
general strategy is as follows:
plicit schemes require inversion of this matrix. However,
no inversion is needed with explicit schemes, since fu-
ture values can be written explicitly in terms of only (a) Start with the two-way scalar wave equation:
past values. ∂2P ∂2P 1 ∂2P
Equation (4-9a) is rewritten by redefining scalar a + − = 0, (4 − 12)
∂x2 ∂z 2 v 2 (x, z) ∂t2
as a ∆t to obtain
where x and z are the space variables, t is the time
P (t + ∆t) − P (t)
= a P (t). (4 − 10) variable, v is the velocity of wave propagation, and
∆t P (x, z, t) is the pressure wavefield.
The left side of equation (4-10) is the discrete repre- (b) Assume constant velocity and perform 3-D Fourier
sentation of the continuous derivative of P with respect transform of the pressure wavefield. This is equiv-
to time, dP/dt. Therefore, equation (4-10) is the finite- alent to substituting the plane-wave solution
difference equation that corresponds to the differential exp(ikx x + ikz z − iωt) to equation (4-12). The sub-
equation stitution yields the dispersion relation between the
dP transform variables
= a P (t). (4 − 11)
dt ω2
We have derived the differential equation that de- kz = ∓ − kx2 , (4 − 13a)
v2
scribes the inflation of money (equation 4-11). Now con-
where kx and kz are the wavenumbers in the x
sider the analysis in reverse order. We start with the dif-
and z directions, and ω is the angular temporal
ferential equation (4-11), and write the corresponding
frequency.
difference equation (4-10), which is the equation that
(c) We are interested in upcoming waves, hence we
is solved in the computer. This equation is written in
only need one of the two solutions. We also want to
either the explicit (equation 4-9a) or implicit (equation
invoke the exploding reflector model by replacing v
4-9b) form to extrapolate the present value of P to the
future. by v/2 to obtain the following paraxial dispersion
This example illustrates how finite-difference relation
schemes can solve differential equations in the com- 2ω vkx
2
puter. The scalar wave equation can be treated in a kz = 1− , (4 − 13b)
v 2ω
similar, but more complicated manner. Complications
arise because it is a partial differential equation that where the horizontal wavenumber kx has been nor-
contains the second derivatives of the wavefield with re- malized with respect to 2ω/v.
spect to depth, time, and spatial axes. Setting up the (d) Make a rational approximation to the square-root
computer algorithm is more involved and is not dis- expression in equation (4-13b) so as to derive a
cussed here. Claerbout (1976, 1985) provides details of differential equation (Sections D.3 and D.4). This
490 Seismic Data Analysis

approximation imposes a dip limitation to the dif- lapsing diffraction energy to the apex of the trav-
ferential equation. One approximation to the dis- eltime curve only. Hence, it is referred to as the
persion relation given by equation (4-13b) is ob- diffraction term. When lateral velocity variations
tained by Taylor expansion of the square root and are significant, the diffraction curve is somewhat
retaining the first two terms in the series (Section like a skewed hyperbola with its apex shifted later-
D.3) ally away from the diffraction source. This lateral
2 shift is accounted for by the thin-lens term given
2ω 1 vkx by equation (4-16b) (Section D.3). If the lateral
kz = 1− . (4 − 14a)
v 2 2ω velocity variations are significant, then the thin-
This dispersion equation is known as the 15-degree lens term is not negligible. Migration algorithms
approximation and is the basis for the first finite- that implement both the diffraction and thin-lens
difference time migration algorithm developed by terms represented by equations (4-16a,b) generally
Claerbout and Doherty (1972). Albeit no longer in are two-step schemes that alternately solve these
use, we shall review the 15-degree finite-difference two terms. To propagate one depth step, first apply
algorithm for its historical significance. the diffraction term on wavefield Q. The thin-lens
(e) Operate on the pressure wavefield P (kx , kz , ω) with term then is applied to the output from the diffrac-
the approximate form of the dispersion relation tion calculation. A migration method that includes
given by equation (4-14a), and inverse Fourier the effects of the thin-lens term is called depth mi-
transform in the z direction to get the differential gration, since the output section is in depth. Depth
form of the approximate one-way wave equation. migration is warranted if there are strong lateral
variations in velocity; in this case, the coefficient
2 of the thin-lens term cannot be negligible. If we
∂ 2ω 1 vkx
P (kx , z, ω) = −i 1− P (kx , z, ω). assume that velocity varies only in the vertical di-
∂z v 2 2ω rection, then v̄(z) = v(x, z). This makes the thin-
(4 − 14b) lens term of equation (4-16b) vanish, and we are
(f) Recall from Figure 4.1-17 that, after each left with the diffraction term of equation (4-16a).
downward-continuation step, we retard the wave- A migration method that implements the diffrac-
field by translating it in time so that after migra- tion term (equation 4-16a), only, is known as time
tion, events appear in their correct depth locations. migration, the output of which is in time τ of equa-
The time retardation is done by applying a linear tion (4-15b). When recast in terms of the τ vari-
phase shift to the pressure wavefield P able, equation (4-16a) takes the form
Q = P exp(−iωτ ), (4 − 15a) ∂2Q v2 ∂ 2 Q
= . (4 − 17)
where the retarded time is ∂τ ∂t 8 ∂x2
z This is the parabolic equation for time migration.
dz
τ= , (4 − 15b) (g) Finally, write down the difference forms of the dif-
0 v̄(z)
ferential operators either in implicit form to be used
and Q is the retarded wavefield. The velocity v̄(z) is in finite-difference solution of the parabolic equa-
the horizontally averaged v(x, z). Substitute equa- tion (4-17) for migration.
tion (4-15a) into (4-14b) to obtain the differen-
tial equation associated with the 15-degree finite-
Boundary and initial conditions are needed to solve
difference algorithm in two parts
the differential equations. The initial condition for mi-
∂2Q v ∂2Q gration is the recorded wavefield at the surface z = 0.
= , (4 − 16a)
∂z∂t 4 ∂x2 Also, in migration we assume that the wavefield is zero
and after a maximum observation time, typically the end
time of the recorded trace. Then there are the side
∂Q 1 1 ∂Q
=2 − , (4 − 16b) boundaries, beyond which assumptions must be made
∂z v̄(z) v(x, z) ∂t about the form of the wavefield.
where Q is the retarded wavefield. Derivation In the (x, z, t) coordinates, the seismic section is
of equations (4-16a,b) is based on the assump- represented by the x − t plane, while the migrated sec-
tion that velocity varies vertically. Nevertheless, tion (earth) is represented by the x − z plane. Finite-
in practice, the velocity function in equations (4- difference migration, as discussed here, extrapolates the
16a,b) can be varied laterally, provided the varia- x − t plane in finite increments of z and outputs the
tion is smooth. Equation (4-16a) accounts for col- wavefield at t = 0 at each step (Figure 4.1-19).
Migration 491

FIG. 4.1-20. Two algorithmic schemes to downward con-


tinue wavefields in the computer: (a) z-outer, and (b) t-
outer. The midpoint axis is perpendicular to the plane of
the paper. In both schemes, the CMP-stacked data are rep-
FIG. 4.1-19. The seismic section represented by the x − t
resented by the s-column at z = 0 (Claerbout, 1976).
plane at the surface z = 0 is downward continued to obtain
the time sections at discrete depth levels. The direction of
extrapolation is indicated by the thick arrow. The migrated
section is represented by the x − z plane at t = 0. in Figure 4.1-20, there are two samples from each depth
step collected into the migrated section.

There are two ways to downward continue the


wavefield recorded at the surface in the computer (Fig-
Reverse Time Migration
ure 4.1-20). Starting with the wavefield at the surface
z = 0 represented by the vectors in x — s1 , s2 , s3 , . . .,
which are perpendicular to the page, we can compute Another migration method, known as reverse time mi-
the wavefield at different depth levels using the order of gration (Baysal et al., 1983), extrapolates an initially
the computation shown in Figure 4.1-20a. Assume zero zero x − z plane backward in time, bringing in the seis-
value for the bottom of the extrapolated wavefield at mic data P (x, z = 0, t) as a boundary condition z = 0
each depth step. So, for example, using s7 , s8 , and 0, at each time step to compute snapshots of the x − z
compute the wavefield at position 1. Then use s6 , s7 , plane at different times. At time t = 0, this x − z plane
and the wavefield already computed at position 1, com- contains the migration result P (x, z, t = 0) (Figure 4.1-
pute that at position 2, and so on. Notice that, in this 21).
scheme, we compute the wavefield at all times for one The algorithmic structure for the reverse time mi-
depth step, then compute the wavefield at all times for gration is illustrated schematically in Figure 4.1-22.
the next depth step, followed by the next depth step, Start with the x − t section at the surface, z = 0. Also,
and so on. Hence, this is called the z-outer computa- consider an x − z frame at tmax . This frame is blank ex-
tional scheme. cept for the first row which is equal to the bottom row
The alternate scheme involves a different order of of the x−t section at tmax . Extrapolate this snapshot at
computation (Figure 4.1-20b). First, compute the wave- t = tmax to t = tmax −∆t by using the phase-shift oper-
field at one time for all depths, then using those al- ator exp(iω∆t). This yields a new snapshot of the x − z
ready computed values, compute the wavefield at the frame at t = tmax − ∆t. The first row of numbers in this
next shallower time for all depths, and so on. Hence frame is identical to the row in the x − t plane — the
this is called the t-outer computational scheme. original unmigrated section, at t = tmax − ∆t. Hence,
In both schemes, the output migrated section is ob- replace the first row in the snapshot at t = tmax − ∆t
tained by collecting the diagonal elements. Depending with the row of the x − t section at t = tmax − ∆t
on the depth step size, which can be conveniently de- and continue the extrapolation back in time. The last
fined as the number of time samples, one collects one or snapshot is at t = 0 that represents the final migrated
more samples at each depth level. In the example shown section.
492 Seismic Data Analysis

FIG. 4.1-21. Reverse time migration: Start with an all-zero


x − z plane at the bottom of the data cube and extrapolate
backward in time toward t = 0 to compute snapshots of the
x − z plane at different times. These snapshots of the sub-
surface are indicated by the horizontal planes; the direction
of extrapolation — reverse in time, is indicated by the thick FIG. 4.1-22. An algorithmic description of reverse time
arrow. At each time level, include the boundary value (x- migration.
slice at z = 0, indicated by the dotted lines) into the x − z
plane from the seismic section. The migrated section is the
x − z plane at t = 0 (the top horizontal plane).
where Q(x, z, ω) is the retarded wavefield in the
frequency-space domain.
Frequency-Space Implicit Schemes When recast for time migration, equation (4-19a)
becomes (Section D.4):
As discussed in Section 4.3, in practice the 15-degree 1 ∂3Q ∂2Q 8ω ∂Q
finite-difference migration can handle dips up to 35 de- i − +i 2 = 0, (4 − 19b)
2ω ∂τ ∂x2 ∂x2 v ∂τ
grees with sufficient accuracy. A steep-dip approxima-
where τ is the time variable associated with the mi-
tion to equation (4-13b) is achieved by continued frac-
tions expansion (Section D.4) as grated data.
Note that dropping the first term in equation (4-
2ω v 2 kx2 1 19a) and inverse Fourier transforming in time yields the
kz = 1− . (4 − 18)
v 8ω 2 v 2 kx2 15-degree diffraction equation (4-16a). Similarly, drop-
1− ping the first term in equation (4-19b) yileds the 15-
16ω 2
degree equation (4-17) for time migration.
This dispersion equation is known as the 45-degree ap-
proximation and is the basis of the most common imple- As for the 15-degree equation, the thin-lens equa-
mentation of steep-dip implicit finite-difference schemes tion (4-16b) also applies for the 45-degree equation.
(Kjartansson, 1979). When implemented in the frequency-space domain, the
Refer to the steps described earlier and replace the thin-lens term is represented by the phase-shift oper-
Taylor expansion given by equation (4-14a) with the ator of equation (4-15a). Again, the final step in the
continued fractions expansion given by equation (4-18). procedure is to write down the difference forms of the
Follow the subsequent steps to derive the correspond- differential operators in implicit form to be used in
ing differential equation associated with the 45-degree finite-difference solution of the 45-degree equation (4-
diffraction term (Section D.4): 19) for migration. Kjartansson (1979) provides an im-
plicit scheme in which the extrapolation is in z. Never-
v ∂3Q ∂2Q 4ω ∂Q theless, as for the 15-degree equation (4-17), it is trivial
i − +i = 0, (4 − 19a) to adapt his scheme for time migration with the extrap-
4ω ∂z∂x2 ∂x2 v ∂z
Migration 493

model. Operate on the pressure wavefield P and in-


verse Fourier transform in z to obtain the differential
equation

P (kx , z, ω) = −ikz P (kx , z, ω), (4 − 20)
∂z
whose solution can be used to extrapolate the wavefield
at the surface down in depth
P (kx , z, ω) = P (kx , 0, ω) exp(−ikz z). (4 − 21)
For a discrete depth step ∆z, equation (4-21) takes
the form
P (z + ∆z) = P (z) exp(−ikz ∆z), (4 − 22a)
where, for convenience, the variables kx and ω have been
omitted from P .
When designing extrapolation operators, whatever
the differencing scheme, the objective must be to en-
FIG. 4.1-23. An algorithmic description of frequency-space sure that the phase and amplitude of the actual op-
migration. erator closely resembles those of the desired operator
exp(−ikz ∆z).
olation in τ of equation (4-15b). The phase-shift op- Discretize the one-way wave equation (4-20) and
erator of equation (4-15a) is velocity-dependent when apply differencing approximation using an explicit
implemented for depth migration, and it is velocity- scheme such as
independent when implemented for time migration.
The 45-degree approximation given by equation (4- P (z + ∆z) = P (z) 1 − ikz ∆z , (4 − 22b)
19b) actually is fairly accurate in practice up to 60 de- and an implicit scheme (Section D.6):
grees. As described in Section D.4, the basic 45-degree
equation (4-19b) also can be adapted to obtain extrapo- 1 − ikz ∆z/2
P (z + ∆z) = P (z) . (4 − 22c)
lation schemes for imaging steeper dips up to 90 degrees. 1 + ikz ∆z/2
Nevertheless, a penalty is paid for steep-dip accuracy in The explicit extrapolation operator (1 − kz ∆z) of
terms of dispersive noise incurred by implicit schemes
equation (4-22b) actually is the first two terms of the
(Section 4.3).
Taylor expansion of the exact operator exp(−ikz ∆z) of
Steep-dip finite-difference algorithms may be more
equation (4-22a). Table 4-5 provides the amplitude and
conveniently implemented in the frequency-space do-
phase of the exact, explicit and implicit operators used
main than in the time-space domain. A general frame-
in equations (4-22a,b,c) for wavefield exptrapolation in
work for implementing such algorithms involve a loop
depth.
over the depth step z, and a loop over the frequency ω
A desired property of an extrapolation operator is
(Figure 4.1-23). For each depth step:
that it must be stable — its amplitude should be less or
equal to unity. The implicit operator defined by equa-
(a) Apply the shift term (equation 4-15a). tion (4-22c) is stable, while the explicit operator defined
(b) Apply the diffraction term (equation 4-19) by per- by equation (4-22b) causes amplitudes of the extrapo-
forming implicit extrapolation of each of the fre- lated wavefield grow with depth (Section D.6). In fact,
quency components of the wavefield.
the larger the depth step ∆z, the more unstable are
(c) Sum over the frequencies to invoke the imaging
the results of extrapolation. Another desired property
principle which is equivalent to setting t = 0.
of an extrapolation operator is that it should yield the
(d) Repeat the computation for all the depth steps to
least phase error. The inherently stable nature of im-
complete the imaging.
plicit schemes has been the compelling reason for their
use in practice. Recent developments in the design of
stable explicit schemes, however, now have made them
Frequency-Space Explicit Schemes widely accepted (Holberg, 1988; Hale, 1991; Soubaras,
1992).
Start with the paraxial dispersion relation given by In principle, the exact extrapolation operator in
equation (4-13b) adapted to the exploding reflectors equation (4-22a) can be inverse Fourier transformed to
494 Seismic Data Analysis

Table 4-5. Amplitude and phase of the exact, explicit such that, when Fourier transformed to the frequency-
and implicit extrapolation operators used in equations wavenumber domain, the difference between the actual
(4-22a,b,c). See Section D.6 for details. transform H(kx ) and the desired transform D(kx ) of
equation (4-23) is minimum, subject to the stability
Operator Amplitude Phase constraint that the amplitude of H(kx ) is never greater
Exact than unity within the propagation region kx ≤ (2ω/v).
exp(−ikz ∆z) 1 kz ∆z Details of the method of modified Taylor expansion
based on this design criterion by Hale (1991) are de-
Explicit scribed in Section D.5.
1 − ikz ∆z 1 + (kz ∆z)2 tan−1 (kz ∆z) As for the implicit schemes (Figure 4.1-23), a mi-
gration algorithm based on an explicit extrapolation fil-
Implicit ter design involves a loop over the depth step z and a
1 − ikz ∆z/2 kz ∆z loop over frequency ω. For each depth step:
1 tan−1
1 + ikz ∆z/2 1 − (kz ∆z/2)2
(a) Convolve the explicit extrapolation filter h(x) with
each of the frequency components of the wavefield.
frequency-space (ω − x) domain and applied to P (z) in (b) Sum over the frequencies to invoke the imaging
an explicit manner. Each output sample of P (z + ∆z) principle which is equivalent to setting t = 0.
at some x location for a frequency ω and velocity v is (c) Repeat the computation for all the depth steps to
computed independently by convolving an explicit fil- complete the imaging.
ter operator of a specified length centered at the out-
put location x with the input data array P (z) in the x The length of the filter coefficients h(x) determines
direction. In contrast, implicit schemes require solving the dip accuracy of the explicit operator. The larger
a set of linear equations to obtain the output samples the number of filter coefficients 2N + 1, the steeper the
of P (z + ∆z) — computationally more intensive than
dip accuracy. In practice, extrapolation filter lengths 7,
convolution. Efficiency is an advantage of the explicit
11, and 25 are often associated with 30-, 50-, and 70-
schemes over the implicit schemes.
degree dip accuracies. Phase error of the extrapolation
Another attractive property of stable explicit
schemes is their extension to 3-D extrapolation that pre- operator at steep dips may be reduced by increasing the
serves circular symmetry — a feature that is relatively number of coefficients. Also, lateral velocity variations
more difficult to attain with implicit schemes (Section can be accommodated by varying the velocity at each
7.3). x location of the filter coefficient.
Whether an explicit filter is computed by inverse
transforming the exact filter exp(−ikz ∆z) of equation
(4-22a) to the frequency-space domain or by Taylor ex- Frequency-Wavenumber Migration
pansion as in equation (4-22b), the problem is that nei-
ther approach yields a stable filter operator. A stable ex-
Frequency-wavenumber (f −k) migration is not as easily
plicit extrapolation filter in the frequency-space domain
can be designed using a constrained least-squares tech- explained as the Kirchhoff or finite-difference migration
nique (Holberg, 1988), or by a modified Taylor series from a physical point of view. Chun and Jacewitz (1981)
expansion of the exact extrapolation filter exp(−ikz ∆z) provide practical insight into the principles of f − k
(Hale, 1991). Another method of explicit operator de- migration.
sign based on on an alternative stability criterion is pre- In Section 1.2, we learned that dipping events in
sented by Soubaras (1992). the t − x domain map onto the f − k domain along ra-
By substituting for kz from equation (4-13b), the dial lines. The steeper the dip, the closer the radial line
exact extrapolation filter exp(−ikz ∆z) of equation (4- is to the wavenumber axis. Figure 4.1-24 shows dipping
22a) is expressed in the frequency-wavenumber domain events before and after migration in the t − x and f − k
as domains. The Nyquist wavenumber is 20 cycles/km and
2 the bandwidth is given by the corner frequencies 6, 12
2ω vkx
D(kx ) = exp −i 1− ∆z . (4 − 23) - 36, 48 Hz for the passband region of the spectrum.
v 2ω (See Figure 1.1-26 for the definition of corner frequen-
The objective is to find, for a specific frequency ω cies.) The red is associated with the flat part of the
and velocity v, a symmetric explicit filter with com- passband region and the blue is associated with the ta-
plex coefficients h(x) in the frequency-space domain per zone. Note that migration rotates the radial lines
Migration 495
496 Seismic Data Analysis

of the f − k plane and is superimposed on the energy


associated with the dipping reflections themselves.
Migration of a dipping event in the f − k domain is
sketched in Figure 4.1-25. Note that this figure is the f −
k equivalent of Figure 4.1-1. In both figures, we assume
velocity equal to 1. The vertical axis in Figure 4.1-25
represents the temporal frequency ω for the event in its
unmigrated position B, and the vertical wavenumber kz
for the event in its migrated position B .
Migration in the frequency-wavenumber domain in-
volves mapping the lines of constant frequency AB in
the ω − kx plane to circles AB in the kz − kx plane.
Therefore, migration maps point B vertically onto point
B . Note that in this process, the horizontal wavenum-
ber kx does not change as a result of mapping. When
this mapping is completed, the dipping event OB is
mapped along OB after migration; thus, the dip angle
θ̄ after migration is greater than the dip angle θ before
migration. For comparison, these two radial lines are
shown on the same plane kz − kx .
We now examine the diffraction hyperbola and its
collapse to the apex after migration in the f −k domain.
A diffraction hyperbola is represented by an inverted
triangular area in the frequency domain as shown in
Figure 4.1-26. The Nyquist wavenumber is 40 cycles/km
and the bandwidth is given by the corner frequencies 6,
12 - 36, 48 Hz for the passband region of the spectrum.
As for the dipping events model in Figure 4.1-24, the red
is associated with the flat part of the passband region
and the blue is associated with the taper zone. The
two edges in the right and left quadrant of the f − k
FIG. 4.1-25. Migration in the f − k domain. (Migration in
the t − x domain is illustrated in Figure 4.1-1.) (a) A dip- plane correspond to the asymptotes of the flanks of the
ping reflector is represented by a radial line OB in the f − k diffraction hyperbola, the base of the inverted triangle
plane. (b) After migration, the radial line OB maps onto an- corresponds to the high-frequency end of the passband,
other radial line OB , while B maps onto B . The horizontal and the tip of the triangle in the proximity of the origin
wavenumber is invariant under migration. For comparison, of the f − k plane corresponds to the low-frequency end
the f − k response of the dipping event before migration (a) of the passband. Migration turns the triangular area
has been superimposed on the f −k response after migration into a circular shape as shown in Figure 4.1-26.
(b). (Adapted from Chun and Jacewitz, 1981.) The f − k analysis of the diffraction hyperbola
shown in Figure 4.1-26 is based on the representation
of the hyperbola as a series of discrete dipping seg-
in the 2-D amplitude spectrum outward and away from ments. Figure 4.1-27 depicts a diffraction hyperbola in
the frequency axis. The steepest event represented by the t − x and f − k domains. We imagine that the hy-
radial line A maps onto radial line B after migration. perbola is made up of a series of dipping segments, such
The feather-like energy especially prominent in the left as A, B, C, D and E. The zero-dip segment A is at the
quadrant of the f − k spectrum is associated with the apex, while the steepest dip segment E is along the
flanks of the diffraction hyperbolas in the t − x domain. asymptotes. In the f − k domain, the zero-dip segment
The energy associated with the left flanks which are A maps along the frequency axis, while the dipping seg-
dipping opposite to the dipping reflections maps onto ments B, C and D map along the radial lines, increas-
the left quadrant of the f − k plane. And the energy ingly further away from the frequency axis. Finally, the
associated with the right flanks of the diffraction hy- asymptotic tail E maps along the radial line that rep-
perbolas that are dipping in the same direction as that resents the boundary between the propagation and the
of the dipping reflections maps onto the right quadrant evanescent region. The evanescent region corresponds
Migration 497
498 Seismic Data Analysis

FIG. 4.1-27. A hyperbola on the t − x plane maps onto an inverted triangular area on the f − k plane. (See text for details.)

to the energy that is located at or greater than 90 de- (a) Just as any other migration algorithm, start with
grees from the vertical. The opposite side of the hyper- the two-way scalar wave equation (4-12).
bola maps onto the second quadrant (negative kx ) in (b) Assume constant velocity and perform 3-D Fourier
the f − k domain. In the continuous case, a diffraction transform and obtain the dispersion relation be-
hyperbola is represented by a series of continuous radial tween the transform variables (equation 4-13a).
lines that constitute an inverted triangular area in the (c) Then, adapt the dispersion relation to the explod-
f − k domain (Figure 4.1-26). ing reflectors model by halving the velocity for the
A curious fact emerges from the f − k spectrum of upcoming waves (equation 4.13b).
the migrated section in Figure 4.1-26. We expect migra- (d) Operate on the pressure wavefield P and inverse
tion to collapse the diffraction hyperbola to a point at transform in z to obtain the differential equation
the apex. The spectrum of this migrated section really (4-20).
should be more like that in Figure 4.1-28 — a rectangle. (e) Obtain the solution given by equation (4-21).
Why is there a difference between this spectrum and the
spectrum after migration in Figure 4.1-26? The discrete form of this solution given by equation
If you start with a point and model it, you get the (4-22) is the basis for phase-shift migration in which
diffraction hyperbola in Figure 4.1-28. However, in re- velocity can be varied at each depth step in the vertical
ality we deal with a diffraction hyperbola as shown in direction.
Figure 4.1-26. The hyperbolas do not look different in The phase-shift method involves the following
the t − x domain, but note the difference in their f − k steps:
spectra. The f − k spectrum of the real-life diffraction,
which is always subjected to bandpass filtering (Figure
(a) Start with the stacked section — an approximation
4.1-26), is missing the energy above the 48-Hz line that
to the zero-offset section P (x, z = 0, t), and per-
is present in the f − k spectrum of the modeled diffrac-
form 2-D Fourier transform to get P (kx , z = 0, ω).
tion curve (Figure 4.1-28). These missing high frequen-
(b) By using equation (4-22), for each frequency ω, ex-
cies cause the difference between the spectra after mi-
trapolate the transformed wavefield P (kx , z, ω) at
gration.
depth z with a phase-shift operator exp (−ikz ∆z)
to get the wavefield P (kx , z + ∆z, ω) at depth
z + ∆z. At each z step, a new extrapolation op-
Phase-Shift Migration erator with the velocity defined for that z value is
computed.
Theory of the frequency-wavenumber (f − k) migration (c) As for any other migration, invoke the imaging
techniques is left to Section D.7. For now, we briefly principle t = 0 at each extrapolation step to ob-
review the f − k migration algorithms as follows: tain the migrated section P (kx , z, t = 0) in the
Migration 499

4.1-28. (a) A diffraction hyperbola modeled from an impulse in the t − x domain, (b) the f − k spectrum of (a), (c) migration
of (a), (d) the f − k spectrum of (c); (e) bandpass-filtered version of (a), (f) the f − k spectrum of (e), (g) migration of (e),
and (h) the f − k spectrum of (g).
500 Seismic Data Analysis

transform domain. The imaging condition t = 0


is met by summing over all frequency components
of the extrapolated wavefield at each depth step.
This is easily shown from the integral representing
the inverse Fourier transform of the extrapolated
wavefield (equation D-84).
(d) Repeat steps (b) and (c) for downward continua-
tion and imaging, respectively, for all depth steps to
get the migrated section in the transform domain
P (kx , z, t = 0).
(e) Final step involves inverse transforming in the x
direction to get the migrated section P (x, z, t = 0).

Figure 4.1-29 shows a flowchart of the phase-shift


method.
The phase-shift method (Gazdag, 1978) can only
handle vertically varying velocities. A way to accommo-
date lateral velocity variations judged to be acceptable
for time migration is to first stretch the CMP-stacked
section in the time direction so as to make it correspond
to a velocity field v̄(z) that only varies vertically. This
velocity field is obtained by averaging the original veloc-
ity field associated with the unstretched CMP-stacked FIG. 4.1-29. Flowchart for Gazdag’s phase-shift method of
section in the x direction. Following the stretching op- migration.
eration, the stacked section is migrated using the veloc-
ity function v̄(z) in the standard phase-shift migration
scheme. Finally, the migrated section is unstretched.
Gazdag and Squazzero (1984) extended the phase-
shift method to handle lateral velocity variations. To
achieve this, first the input wavefield is extrapolated by
the phase-shift method using a multiple number of lat-
erally constant velocity functions and a series of refer-
ence wavefields are created. The imaged wavefield then
is computed by interpolation from the reference wave-
fields. This migration method is known as phase-shift-
plus-interpolation. An alternative extension of phase-
shift migration to handle lateral velocity variations is
presented by Kosloff and Kessler (1987).

Stolt Migration

If the medium velocity is constant, migration can


be expressed as a direct mapping (Stolt, 1978) from
temporal frequency ω to vertical wavenumber kz FIG. 4.1-30. Flowchart for Stolt’s constant-velocity migra-
(Figure 4.1-25). Figure 4.1-30 is a flowchart of the tion method in the f − k domain.
Stolt algorithm; the mathematical details are left
to Section D.7. The equation for Stolt mapping is

where P (kx , z = 0, ω) is the zero-offset section and


v kz P (kx , kz , t = 0) is the migrated section in the frequency-
P (kx , kz , t = 0) =
2 kx + kz2
2
wavenumber domain.
v Note that Stolt migration involves, first, mapping
P kx , 0, ω = ky2 + kz2 ,
2 from ω to kz for a specific kx by using the dispersion
(4 − 24a) relation of equation (4-13a) recast as
Migration 501

v finite-difference and Kirchhoff migrations. In practice,


ω= kx2 + kz2 . (4 − 24b)
2 mapping in the f − k domain really is from ω − kx to
ωτ − kx rather than to kz − kx , where ωτ is the Fourier
The output of mapping is then scaled by the quantity dual of τ and is simply kz of equation (4-13b) scaled by
S v/2 (Section D.3):

v kz 2
S= . (4 − 24c) vkx
2 kx2
+ kz2 ωτ = ω 1− . (4 − 25)

Stolt’s algorithm for constant velocity thus involves


the following steps: One important concept must be pointed out from
equation (4-25). Note that for a constant kx , ωτ < ω;
(a) Start with the input wavefield P (x, z = 0, t) ap- thus, migration shifts the bandwidth to lower frequen-
proximated by the CMP stack, and apply 2-D cies. This is analogous to the conclusion derived in rela-
Fourier transform to get P (kx , z = 0, ω). tion to the NMO correction, since the latter also causes
(b) Map the wavefield from ω to kz using the dispersion data stretching to lower frequencies (Section 3.1). The
relation given by equation (4-24b). implication from equation (4-25) is demonstrated by
(c) Apply the scaling factor S of equation (4-24c) as the dipping events model in Figure 4.1-24. While the
part of the mapping procedure (Section D.7). bandwidth of the zero-dip event is retained after mi-
(d) Invoke the imaging principle by setting t = 0 and gration, the bandwidth of the event with steepest dip
obtain P (kx , kz , t = 0). has shifted from approximately 40 Hz to 36 Hz at the
(e) Finally, apply 2-D inverse transform to get the mi- high-frequency end of the spectrum. In fact, the shift in
grated section P (x, z, t = 0). bandwidth is dip-dependent; events with different dips
which have the same bandwidth before migration will
It may be questionable as to whether the constant- have different bandwidths after migration.
velocity Stolt method has value on its own as a practical
migration algorithm. Nevertheless, Stolt’s method can
be used efficiently to perform a constant-velocity mi-
Summary of Domains of Migration Algorithms
gration as the first step in a residual migration scheme
(Section 4.5). Additionally, the method constitutes an
essential procedural step for migration velocity analysis Migration algorithms described in this section are based
as described in Section 5.4. on the assumption that the input stacked section rep-
Stolt extended his method to handle velocity vari- resents a zero-offset acoustic wavefield. As such, these
ations (Section D.7). For the variable-velocity case, algorithms are all based on the scalar wave equation
Stolt’s extension consists of (4-12). Table 4-6 provides a list of the migration al-
gorithms described in this section with the associated
(a) modifying the input wavefield to make it appear as design and application domains. While there exist sev-
if it were the response of a constant-velocity earth, eral other migration algorithms, those listed in Table
(b) applying the constant-velocity algorithm outlined 4-6 are the most widely used in practice.
in Figure 4.1-30, and Also included in Table 4-6 are the types of velocity
(c) reversing the original modification of the input fields and dips each migration algorithm can accommo-
wavefield. date. It is clear that each algorithm is limited by either
the type of velocity variations or dip ranges. Therefore,
This modification essentially is a type of stretching each has an appropriate usage in practice depending
of the time axis (Section D.7) to make the reflection on field data and velocity characteristics (Section 4.0).
times approximately equivalent to those recorded for a Note that the lateral velocity variations implied by the
constant-velocity earth. The nature of stretching is de- velocity fields in Table 4-6 are mild to moderate and
scribed by the stretch factor W . The constant-velocity are within the bounds of time migration. Additionally,
case is equivalent to W = 1. the choice of migration algorithm depends on whether
Note that the phase-shift and Stolt migration out- your objective is imaging or migration velocity analysis.
puts normally are displayed in two-way vertical zero- While imaging is the subject of this chapter, migration
offset time τ = 2z / v, as are the outputs from the velocity analysis is discussed in Chapter 5.
502 Seismic Data Analysis

Table 4-6. Domains of migration algorithms. the phase-shift method was a compromise; it handles
dips of up to 90 degrees and velocities that can only
Algorithm Domain Dips and vary vertically.
Velocities Before a migration algorithm is used on field data,
Kirchhoff t−x up to 90 deg its impulse response must be tested. A band-limited im-
Summation time-space rms v(x, τ ) pulse response is generated by using an input that con-
tains an isolated wavelet on one trace only To also limit
the spatial bandwidth, this trace is replicated on either
Finite-Difference t−x up to 35 deg
side with the wavelet amplitude halved. The ideal mi-
15-deg Implicit time-space int v(x, τ )
gration algorithm should produce an impulse response
that has the shape of a semicircle. Kirchhoff migration
Finite-Difference ω−x up to 65 deg
produces the section shown in Figure 4.2-1d. The im-
45-deg Implicit frequency-space int v(x, τ )
pulse response indicates that Kirchhoff migration can
accurately handle dips up to 90 degrees. The dip on
Finite-Difference ω−x up to 80 deg
a migration impulse response is measured as the angle
70-deg Explicit frequency-space int v(x, τ )
θ between the vertical and a specified radial direction.
Note that migration can be limited to smaller dips (Fig-
Phase-Shift ω − kx up to 90 deg ure 4.2-1).
freq.-wavenumber int v(τ )

Stolt Method ω − kx up to 90 deg


with Stretch freq.-wavenumber rms v(x, τ ) Aperture Width

From the previous section, we know that Kirchhoff


migration involves a summation of amplitudes along
diffraction hyperbolas. Given the rms velocity at a par-
ticular time sample of a particular input trace, a hyper-
bolic traveltime trajectory associated with a fictitious
diffractor is overlaid on the input section with its apex
4.2 KIRCHHOFF MIGRATION
at that time sample. In theory, a diffraction hyperbola
IN PRACTICE extends to infinite time and distance. In practice, we
have to deal with a truncated hyperbolic summation
In this and the following three sections, the parameters path. The spatial extent that the actual summation
that affect performance of Kirchhoff summation, finite- path spans, called the migration aperture, is measured
difference, and f −k migration methods are discussed. In in terms of the number of traces the hyperbolic path
Kirchhoff migration, the important parameters are the spans.
aperture width used in summation and the maximum
dip to migrate. In finite-difference and phase-shift mi-
grations, the depth step size needs to be selected prop-
erly. The stretch factor is important in Stolt migration.
The responses of these methods to velocity errors also
are examined. All practical aspects are discussed using
synthetic models of two zero-offset sections — a model
of dipping-events and a model of a diffraction hyper-
bola. Real data examples also are used to evaluate the
choice of optimum parameters.
In Sections 4.2, 4.3, 4.4, and 4.5, migration re-
sults of different algorithms using various parameters
are compared with a desired migration. In all cases,
this desired migration was obtained using the phase-
shift method with appropriate parameters and veloci-
ties. This does not imply that the phase-shift method FIG. 4.2-1. Migration can be confined to a range of dips
always provides a desirable output; it only means that present on a seismic section. The impulse response for the
the data examples in this section were chosen so that dip-limited migration operator is a truncated semicircle. Dip
the phase-shift algorithm is appropriate. The choice of angle θ is measured from the vertical axis.
Migration 503

FIG. 4.2-2. Summation paths for Kirchhoff migration in a medium with (a) low velocity (2000 m/s), (b) high velocity (4000
m/s), and (c) vertically varying velocity. Migration aperture is small for low velocities and large for high velocities.

The curvature of the diffraction hyperbola is gov- nx = dx /∆x, where ∆x is the CMP interval. Therefore,
erned by the velocity function. Figure 4.2-2a shows the aperture width that is required is 2nx +1. Figure 4.2-
a number of low-velocity diffraction hyperbolas, while 4 also shows Kirchhoff migrations of the dipping events
Figure 4.2-2b shows a number of high-velocity hyper- using four different aperture widths. Small-aperture mi-
bolas. A low-velocity hyperbola has a narrower aper- gration eliminates steeply dipping events on the output
ture when compared to a high-velocity hyperbola. This section. Increasing the aperture width allows proper mi-
agrees with our intuition — high velocity means more gration of the steeply dipping events. From this we see
migration. In practice, we deal with a velocity function that using too small an aperture width causes a dip fil-
that at least varies with depth. The diffraction hyperbo- tering action during migration, because a small aperture
las can have different curvatures depending on the ve- excludes the steeper flanks of the diffraction hyperbola
locity value at a given time sample (Figure 4.2-2c). Be- from the summation.
cause of the vertical variation in velocity, aperture width For any given event position in time t before mi-
generally is time variant. For the usual case in which gration, the optimal value for the aperture width is de-
velocity increases with depth, migration of the shallow fined by twice the maximum horizontal displacement in
part of the section requires a narrow aperture, while migration for the steepest dip of interest in the input
migration of the deep portion requires a wider aperture section. In this case, the horizontal displacement asso-
(Figure 4.2-2c). This implies that, given the same dip, ciated with the 45-degree dipping event is computed by
deep events migrate farther than shallow events. substituting the values for v = 3500 m/s, ∆x = 25 m,
Figure 4.2-3 shows a zero-offset diffraction hyper- ∆t / ∆x = 12 ms/trace, where t = 2 s in equation (4-1).
bola (8 ms/trace dip along the asymptotes) and migra- The value for the horizontal displacement is 118 traces,
tions using four different aperture widths. The smaller giving an aperture width of 237 traces. Typically, we
the aperture, the less capable the migration is in col- consider somewhat larger values to allow for velocity
lapsing the diffraction hyperbola. In this case, use of an errors.
aperture width that is equal to the width of the input A good way to determine aperture width is to gen-
section (half aperture, 256 traces) yields the best result. erate diffraction hyperbolas as shown in Figure 4.2-2c
Figure 4.2-4 shows a synthetic zero-offset section using the regionally averaged, vertically varying veloc-
that consists of a number of dipping events ranging from ity. Clearly, the larger the aperture width, the more
0 to 45 degrees in increments of 5 degrees. Aperture traces are used in the summation. For the dipping
width is related closely to the horizontal displacement events in Figure 4.2-4, the optimal value of the half-
dx that takes place in migration as defined by equa- aperture width is 150 traces; increasing the width to
tion (4-1). The number of traces an event migrates is 300 traces resulted in no further improvement.
504 Seismic Data Analysis

FIG. 4.2-3. Tests for aperture width in Kirchhoff migration: (a) a zero-offset section that contains a diffraction hyperbola with
2500-m/s velocity, (b) desired migration using the phase-shift method; Kirchhoff migration using (c) 35-trace, (d) 70-trace,
(e) 150-trace, and (f) 256-trace half-aperture width.
Migration 505

FIG. 4.2-4. Tests for aperture width in Kirchhoff migration: (a) a zero-offset section that contains dipping events with
3500-m/s velocity, (b) desired migration using the phase-shift method; Kirchhoff migration using (c) 35-trace, (d) 70-trace,
(e) 150-trace, and (f) 256-trace half-aperture width.
506 Seismic Data Analysis

FIG. 4.2-5. Tests for aperture width in Kirchhoff migration: Insufficient aperture width causes removal of steeply dipping
events.
Migration 507

FIG. 4.2-6. Tests for aperture width in Kirchhoff migration: Insufficient aperture width causes spurious horizontal events in
deep, noisy part of a stacked section.
508 Seismic Data Analysis

FIG. 4.2-7. Tests for aperture width in Kirchhoff migration: Input to migration is a section (a) that contains band-limited
random noise uncorrelated from trace to trace. Note the spurious horizontal events in the deeper part of the section after
migration using small aperture (60 traces); these gradually disappear at increasingly larger apertures.

A test of aperture width on the CMP stacked data random noise and the velocity function used in migra-
example is shown in Figure 4.2-5. The small-aperture tion increases in time (Figure 4.2-7a). We see two inter-
migration causes smearing in the deeper part of the sec- esting phenomena on the migrated sections using three
tion. This smearing effect destroys the dipping events different apertures. First, in all three cases there is more
and produces spurious horizontally dominant events. smearing of noise in the deeper part of the data, where
Smearing is reduced gradually with increasing aperture. the velocities generally are higher than in the shallower
Figure 4.2-6 is the deeper portion of a stacked sec- part. Second, there is relatively more smearing in the
tion with migrations using different aperture sizes. The small-aperture migration compared with others at a
smearing effect is much more noticeable at small aper- given time in the section. Moreover, this smearing is
ture. The main difference between the stacked sections characterized by horizontally dominant spurious events,
in Figures 4.2-5 and 4.2-6 is that the latter, being deeper especially in the deeper part of the section. Note that
in time, contains a large amount of noise. This smearing even with a large aperture, some smearing still is present
phenomenon was not noticed in the noise-free synthetic in the deepest part of the section in Figure 4.2-7d. As
model in Figure 4.2-4. We now see that choice of aper- indicated in Figure 4.2-2b, because summation stops at
ture width is more critical than we originally thought. the bottom of the section, the effective aperture (CD)
In particular, a small aperture changes the noise char- at late times is much smaller than that used in other
acteristics of the section. parts of the section (AB). Remember that summation
Why do we see horizontally dominant smearing using very small aperture includes only the apex por-
with small-aperture migration? To answer this, we must tion of the diffraction hyperbola, where dips are nearly
do a simple experiment with a section that contains only flat. Therefore, the small aperture with a dip filtering
Migration 509

action passes flat or nearly flat events — those hori- imum dips. The smaller the maximum allowable dip,
zontal wavenumber components that are zero or nearly the smaller the aperture. This combination of maximum
zero. aperture width and maximum dip limit determines the
In conclusion, the following assessments are made actual effective aperture width used in migration. In
concerning the choice of aperture width. particular, diffraction hyperbolas along which summa-
tion is done are truncated beyond the specified maxi-
(a) Excessively small aperture width causes destruc- mum dip limit.
tion of steeply dipping events and rapidly varying A field data example of testing the maximum dip
amplitude changes. parameter is shown in Figure 4.2-9. Some steep dips are
(b) Excessively small aperture width organizes random lost on the section that corresponds to the 2 ms/trace
noise, especially in the deeper part of the section, maximum allowable dip. The 8 ms/trace dip appears
as horizontally dominant spurious events. to be optimum. The maximum dip parameter must be
(c) Excessively large aperture means more computer chosen carefully so that the steep dips of interest in the
time. More importantly, large apertures can de- input section are preserved. Finally, dip value can be
grade the migration quality in poor signal-to-noise changed spatially and in time; however, practical im-
ratio conditions. Use of large aperture will cause plementation can be cumbersome.
random noise at late times to creep into the good
shallow data. Aperture width always is a compro-
mise with noise.
(d) Sometimes it is better to use a smaller aperture Velocity Errors
than would theoretically be required to avoid the
adverse effect of noise on the migrated event. Noise We now examine the response of Kirchhoff migration
considerations may even require a time-dependent to velocity errors. Figure 4.2-10 shows the diffraction
aperture width. hyperbola and migrations using the 2500 m/s medium
(e) It is recommended that the aperture width be kept velocity, and 5, 10, and 20 percent lower velocities. With
constant in migrating all lines from a particular increasingly lower velocities, the diffraction hyperbola is
survey so that an overall uniformity in amplitude collapsed less and less taking the shape of a frown — it
characteristics on the migrated sections is main- is undermigrated.
tained. Figure 4.2-11 shows the same diffraction hyperbola
In practice, a regional velocity function and the steepest and migrations using the 2500 m/s medium velocity,
dip in a survey area are used to compute the optimal and 5, 10, and 20 percent higher velocities. With in-
aperture width that can be used over the entire set of creasingly higher velocities, the the diffraction hyper-
data from the area (equation 4-1). bola is inverted more and more taking the shape of a
smile — it is overmigrated.
The under-and overmigration effects resulting from
Maximum Dip to Migrate the use of erroneously low or high velocities on the dip-
ping events model are seen in Figures 4.2-12 and 4.2-
During migration, we can specify the maximum dip 13, respectively. Label the correct position of the event
we want migrated in the section. This may be useful with the steepest dip from the desired migration on the
when we need to suppress the steeply dipping coher- results of migrations with different velocities and note
ent noise. Figure 4.2-8 shows migrations of the dipping the event mispositioning caused by erroneously low and
events with four different maximum allowable dips. For high velocities. Compare with the desired migration and
a 4 ms/trace dip limit, events with dips greater than also note that the steeper the dip the more the under-
this value are suppressed. Similarly, for an 8 ms/trace and overmigration effect. Sensitivity of migration to ve-
dip value, events with dips greater than this value are locity errors can be measured quantitatively via equa-
suppressed. When the dip value is 12 ms/trace, no sup- tions 4-1 and 4-2.
pression occurs, since all events in the input section have From the migrated sections in Figure 4.2-14, note
dips less than this value. Limiting the dip parameter is that the bow tie becomes increasingly less resolved at
a way to reduce computational cost, since it is related lower velocities; this indicates undermigration.
to aperture width (equation 4-1), which determines the Figure 4.2-15 shows a CMP-stacked section and
cost. the desired migration. The steep left flank of the salt
From Figure 4.2-1, note that the Kirchhoff migra- dome has been imaged with acceptable accuracy. The
tion impulse response can be limited to various max- accuracy of the imaging of the slightly overturned right
(text continues on p. 520)
510 Seismic Data Analysis

FIG. 4.2-8. Tests for maximum dip to migrate in Kirchhoff migration: (a) a zero-offset section that contains a diffraction
hyperbola with 2500-m/s velocity, (b) desired migration; Kirchhoff migration using (c) 4-ms/trace, (d) 8-ms/trace, (e) 12-
ms/trace, and (f) 24-ms/trace maximum dip.
Migration 511

FIG. 4.2-9. Tests for maximum dip to migrate in Kirchhoff migration: A low value for maximum dip to migrate can be
hazardous. All dips of interest must be preserved during migration.
512 Seismic Data Analysis

FIG. 4.2-10. Tests for velocity errors in Kirchhoff migration: (a) a zero-offset section that contains a diffraction hyperbola
with 2500-m/s velocity, (b) desired migration; Kirchhoff migration using (c) the medium velocity of 2500 m/s, (d) 5 percent
lower, (e) 10 percent lower, and (f) 20 percent lower velocity.
Migration 513

FIG. 4.2-11. Tests for velocity errors in Kirchhoff migration: (a) a zero-offset section that contains a diffraction hyperbola
with 2500-m/s velocity, (b) desired migration; Kirchhoff migration using (c) the medium velocity of 2500 m/s, (d) 5 percent
higher, (e) 10 percent higher, and (f) 20 percent higher velocity.
514 Seismic Data Analysis

FIG. 4.2-12. Tests for velocity errors in Kirchhoff migration: (a) a zero-offset section that contains dipping events with
3500-m/s velocity, (b) desired migration; Kirchhoff migration using (c) the medium velocity of 3500 m/s, (d) 5 percent lower,
(e) 10 percent lower, and (f) 20 percent lower velocity.
Migration 515

FIG. 4.2-13. Tests for velocity errors in Kirchhoff migration: (a) a zero-offset section that contains dipping events with
3500-m/s velocity, (b) desired migration; Kirchhoff migration using (c) the medium velocity of 3500 m/s, (d) 5 percent higher,
(e) 10 percent higher, and (f) 20 percent higher velocity.
516 Seismic Data Analysis

FIG. 4.2-14. Tests for velocity errors in Kirchhoff migration: Undermigration manifested as inadequate handling of the
bowtie is caused by the use of velocities lower than those considered to be medium velocities.
Migration 517

FIG. 4.2-15. (a) A CMP-stacked section, and (b) migration using the phase-shift method. Event A is the water-bottom
multiple, and Event B is the peg-leg multiple associated with the top-salt boundary. These multiples are respectively denoted
by Events C and D on the migrated section.
518 Seismic Data Analysis

FIG. 4.2-16. Tests for velocity errors in Kirchhoff migration using, from top to bottom, 100, 95, 90, and 80 percent of rms
velocities. The input stacked section is shown in Figure 4.2-15a and the desired migration using the phase-shift method is
shown in Figure 4.2-15b.
Migration 519

FIG. 4.2-17. Tests for velocity errors in Kirchhoff migration using, from top to bottom, 100, 105, 110, and 120 percent of
rms velocities. The input stacked section is shown in Figure 4.2-15a and the desired migration using the phase-shift method
is shown in Figure 4.2-15b.
520 Seismic Data Analysis

flank can only be inferred by the lateral positioning of due to the bandlimited nature of seismic data. Steeper
the gently dipping reflections in the vicinity of the salt dips, in principle, can be migrated by a cascaded appli-
flank. cation of the 15-degree algorithm (Larner and Beasley,
Figure 4.2-16 shows results of Kirchhoff migration 1990).
of the stacked section in Figure 4.2-15 using velocities Finite-difference migration of stacked data cur-
lower than what may be optimum for imaging. While rently is performed using steep-dip algorithms based
the undermigration of the left flank of the salt dome on the continued fractions expansion to the scalar wave
is not so evident, the steeply dipping reflection off the equation. This approximation provides a theoretical dip
right flank intersects the gently dipping reflections as- accuracy up to 45 degrees. The basic 45-degree scheme
sociated with the surrounding strata, thus providing a can be improved to handle steeper dips up to 80 de-
clue for undermigration. grees with reasonable accuracy (Section D.4). The 45-
Figure 4.2-17 shows results of Kirchhoff migration degree finite-difference algorithm commonly is imple-
of the stacked section in Figure 4.2-15 using velocities mented using an implicit scheme in the frequency-space
higher than what may be optimum for imaging. While domain.
overmigration effects may be marginal on the section First, as we did for the Kirchhoff migration, we ex-
with a small velocity error (105 percent of optimum ve- amine the impulse response of the 15-degree implicit
locities), migration with higher velocity errors (110 and scheme. The shape of the impulse response of a desired
120 percent of optimum velocities) shows signs of over- migration algorithm with no dip limitation is a semicir-
migration in the form of crossing events along the left cle. The shape of the impulse response of the 15-degree
flank of the salt diapir. Under- and overmigration effects equation is, in theory, an ellipse (Claerbout, 1985) as
caused by large velocity errors often are detectable; nev- seen in Figure 4.3-1. The nature of the dispersive noise
ertheless, small velocity errors can cause subtle effects pattern inside the ellipse is discussed in the next section
making it difficult to judge whether there is under- or on depth step size. Isolated noise spikes in field data can
overmigration. Uncertainties in migration velocities in- introduce such noise patterns on migrated sections.
evitably cause uncertainties in the interpretation made The parts of the responses above the small circles in
from migrated sections or volumes of data. For instance, Figure 4.3-1 correspond to the evanescent energy, while
the shape of the salt diapir inferred from the results the parts below the circles correspond to the propagat-
shown in Figures 4.2-16 and 4.2-17 varies significantly ing energy (Claerbout, 1985). The parts below the cir-
depending on the percent velocity errors. cles are the useful part of the response. The evanescent

4.3 FINITE-DIFFERENCE MIGRATION


IN PRACTICE

As described in Section 4.1, finite-difference migration


is implemented using implicit and explicit schemes. In
this section, we shall include in our discussion the 15-
degree finite-difference algorithm because of its histori-
cal significance. Nevertheless, we shall primarily discuss
practical aspects of the steep-dip implicit and explicit
schemes in the frequency-space domain. Specifically, we
shall deal with the impulse responses, depth step size
and response to velocity errors in implicit and explicit
schemes.
The first finite-difference migration algorithm that
was introduced to the seismic industry was based on
the parabolic approximation to the scalar wave equation FIG. 4.3-1. Desired impulse response of a 90-degree mi-
gration algorithm is a semicircle (top), while the impulse
(Claerbout and Doherty, 1972). The algorithm was im-
response of the 15-degree migration algorithm is an ellipse
plemented in the time-space domain and designed using (bottom). The propagation zone is defined by the portion of
an implicit scheme. The parabolic approximation theo- the ellipse below the small circles, and the evanescent zone
retically limits the algorithm to handling dips up to 15 is defined by the portions above the small circles. For com-
degrees (Section D.3). Nevertheless, in practice, it can parison, the desired response has been superimposed on the
handle dips up to 35 degrees with sufficient accuracy impulse response of the 15-degree equation.
Migration 521

FIG. 4.3-2. (a) CMP stack, (b) desired migration by phase-shift method, (c) 15-degree finite-difference migration. The
finite-difference migration based on the parabolic equation has the inherent property of undermigrating the steep flank of the
diffraction and the steeply dipping event. See Figure 4.3-3 for a sketch of the migration results.

energy travels horizontally and is characterized by imag- imposing the desired semicircular response and measur-
inary wavenumbers kz , which occur when the quantity ing the angle between the indicated lines. Note from the
in the square root in equation (4-13b) becomes nega- measured angle in Figure 4.3-1 that the 15-degree im-
tive. This means that the evanscent region corresponds plicit scheme can be used to migrate dips up to approx-
to horizontal wavenumbers kx > 2ω/v. For negative kz , imately 35 degrees with sufficient accuracy. This is pri-
the exact extrapolator exp(−ikz z) is no longer a wave marily because errors associated with finite-difference
propagator; instead, it causes waves to decay rapidly in approximations used in particular implementations of
depth. Thus, evanescent energy is not expected to be the 15-degree equation usually are adjusted to cancel
present in recorded wavefields. The impulse response of some of the theoretical error associated with the 15-
the 15-degree finite-difference algorithm, however, sug- degree differential equation.
gests propagation in the region of evanescence. This is The dip-limited nature of the parabolic equation
not desirable; the parts of the elliptical wavefront above causes undermigration of steeply dipping events and
the small circles should be removed. Use of a depth step steep flanks of diffractions. This is demonstrated by the
size that is greater than the input sampling rate tends field data example in Figure 4.3-2. The two prominent
features, diffraction D and dipping event B, are located
to suppress the response in the evanescent region. Ex-
as shown in Figure 4.3-3 before and after migration.
cessively large depth steps, however, cause truncation of
the wavefront further into the propagating zone below
the small circles in Figure 4.3-1.
The impulse response (Figure 4.3-1) is used to Depth Step Size
estimate the maximum dip that the implicit finite-
difference algorithm can handle without serious ampli- Finite-difference migration involves downward continu-
tude distortions or phase errors. This is done by super- ation of the wavefield at the surface, such as a stacked
522 Seismic Data Analysis

Undermigration of the diffraction energy along the


steep flanks of the hyperbola is caused by the parabolic
approximation to the scalar wave equation. The dis-
persive noise that accompanies the undermigrated en-
ergy is an effect of approximating differential operators
with difference operators. The accuracy of this approxi-
mation decreases at large frequencies and wavenumbers
(Claerbout, 1985). Thus, the dispersive noise becomes
less with smaller trace spacing and sampling in depth
and time. For example, the difference operator of equa-
tion (4-10) becomes an increasingly better approxima-
tion to the differential operator of equation (4-11) as ∆t
is made smaller. To emphasize more strongly the pres-
ence of the dispersive noise, migrated sections in Fig-
ures 4.3-4 and 4.3-5 have been displayed with the same
display gain level as the input section. The dispersion
normally is much less pronounced on field data.
Figure 4.3-6 shows the dipping events model and
the 15-degree implicit finite-difference migration results
using four different depth step sizes. We can make the
following inferences:

(a) Increasing depth step size causes more and more


undermigration at increasingly steep dips.
(b) The waveform along reflectors is dispersed at steep
dips and large depth steps.
FIG. 4.3-3. A sketch of the diffraction D and steeply dip-
ping event before (B) and after (A) desired migration from (c) Kinks occur along reflectors at discrete intervals
the sections in Figure 4.3-2. The diffraction and the dipping that correspond to the depth step size. Kinks are
event after finite-difference migration using the parabolic more pronounced at increasingly steeper dips.
equation are denoted by F D − D and F D − B, respectively.
The first inference results from the parabolic ap-
section, and invoking the imaging principle so as to cre- proximation, the second from differencing approxima-
ate an image of the subsurface at t = 0. The downward tions, and the third from gradual undermigration to-
continuation takes place in the computer at discrete ward the base of each depth step. The kinks are good
depth intervals (Section 4.1). Depth step size governs for diagnostics; their presence indicates that the depth
performance of finite-difference migration. Inappropri-
step size that is used is too coarse for the dips present in
ate specification of this parameter can cause artifacts in
the data. In that case, smaller depth step size should be
the migrated section. We want to choose an optimum
used; then the kinks disappear altogether (Figure 4.3-
depth step size that is large for computational savings,
7). Nevertheless, kinks that characterize undermigra-
yet yields a tolerable error in positioning the events af-
ter migration. tion can be eliminated by a local adjustment of migra-
Figure 4.3-4 shows the constant-velocity diffraction tion velocities or interpolation between the wavefields
hyperbola and the 15-degree implicit finite-difference at the adjacent depth steps.
migrations using four different depth steps. Large depth It is apparent from Figure 4.3-6 that migration
steps cause severe undermigration as well as kinks along with 20-ms depth step, which corresponds to one-half of
the flank of the diffraction curve (especially apparent in the typical dominant period of recorded seismic waves,
the 60- and 80-ms cases). At smaller depth steps, such has the least dispersion with the least undermigration
as the 20- and 40-ms cases, more energy collapses to the — optimum accuracy in event positioning. Further de-
apex, but the 15-degree scheme fails to achieve a com- creasing depth step size does not improve migration sig-
plete focusing of the energy at the apex of the hyper- nificantly (Figure 4.3-7). The 15-degree implicit scheme
bola. Note also the dispersive noise that trails the unfo- causes precursive dispersion at large depth steps greater
cused energy. The dip-limited nature of the 15-degree al- than 20 ms (Figure 4.3-6) and postcursive dispersion at
gorithm, however, causes undermigration whatever the small depth steps less than 20 ms (Figure 4.3-7). Hence,
depth step size (Figure 4.3-5). taking smaller depth steps does not necessarily mean a
Migration 523

FIG. 4.3-5. Tests for extrapolation depth step size in 15-


degree finite-difference migration: (a) a zero-offset section
that contains a diffraction hyperbola with 2500-m/s velocity,
FIG. 4.3-4. Tests for extrapolation depth step size in 15- (b) desired migration; 15-degree finite-difference migration
degree finite-difference migration: (a) a zero-offset section using (c) 4-ms (d) 8-ms, (e) 12-ms, and (f) 16-ms depth step.
that contains a diffraction hyperbola with 2500-m/s velocity,
(b) desired migration; 15-degree finite-difference migration
using (c) 20-ms (d) 40-ms, (e) 60-ms, and (f) 80-ms depth
step.
524 Seismic Data Analysis

FIG. 4.3-7. Tests for extrapolation depth step size in 15-


degree finite-difference migration: (a) a zero-offset section
that contains dipping events with 3500-m/s velocity, (b) de-
FIG. 4.3-6. Tests for extrapolation depth step size in 15-
sired migration; 15-degree finite-difference migration using
degree finite-difference migration: (a) a zero-offset section
(c) 4-ms (d) 8-ms, (e) 12-ms, and (f) 16-ms depth step.
that contains dipping events with 3500-m/s velocity, (b) de-
sired migration; 15-degree finite-difference migration using
(c) 20-ms (d) 40-ms, (e) 60-ms, and (f) 80-ms depth step.
Migration 525

better quality migration free of the artifacts that occur by the 15-degree algorithm as much as they would be
with the finite-difference method. by an algorithm with no dip limitation (compare with
Figures 4.3-8 and 4.3-9 show the migrations of the Figure 4.2-13).
stacked section in Figure 4.3-2a with five different depth Velocity error test results on field data using the
steps using the 15-degree implicit scheme. Note that 15-degree implicit scheme are shown in Figures 4.3-14
as the depth step size is increased, the dipping event and 4.3-15. Figure 4.3-16 is a sketch of the under- and
off the flank of the salt diapir is more undermigrated overmigration effects. As noted above with the dipping
and the diffraction off the tip of the diapir is less col-
events model (Figures 4.3-12 and 4.3-13), when using
lapsed. Dispersion along the diffraction hyperbola is ap-
velocities greater than medium velocities, overmigration
parent at larger depth steps (Figure 4.3-9). Again, this
is not as pronounced with the 15-degree finite-difference
phenomenon is caused by the differencing approxima-
tions to the differential operators used in the design of migration as it is with a 90-degree algorithm, such as
a finite-difference algorithm. the Kirchhoff or phase-shift method. On the other hand,
when using velocities lower than medium velocities, the
undermigration effect is more pronounced with the 15-
degree finite-difference migration in comparison with a
Velocity Errors
90-degree algorithm (compare Figure 4.3-16 with Figure
4.5-13).
Figure 4.3-10 shows the diffraction hyperbola and its
At first, it may appear to be sensible to compensate
migration using the 2500 m/s medium velocity and 5,
for the undermigration caused by a low-dip algorithm
10, and 20 percent lower velocities. When velocities
by adjusting migration velocities. For example, the best
lower than medium velocity are used, the diffraction
hyperbola gets undermigrated by the 15-degree algo- match between the desired migration and the 15-degree
rithm more than it would be by an algorithm with no finite-difference results for the dipping event in Figure
dip limitation (compare with Figure 4.2-10). 4.3-16 occurs when 10 percent higher velocities are used
Figure 4.3-11 shows the diffraction hyperbola and in the finite-difference migration. While for one dip this
its migration using the 2500 m/s medium velocity and adjustment may be acceptable, for another dip it may
5, 10, and 20 percent higher velocities. When velocities not be. Therefore, deficiencies of a migration algorithm
higher than medium velocity are used, the diffraction should not be compensated for by making modifications
hyperbola gets overmigrated less by the 15-degree al- to the velocity field for migration.
gorithm than it would be by an algorithm with no dip
limitation (compare with Figure 4.2-11). Moreover, note
the increase in dispersive noise as a result of overmigra-
Cascaded Migration
tion.
Figure 4.3-12 is the dipping events model with mi-
grations using the 3500 m/s medium velocity, and 5, To compensate for the inherent undermigration by
10, and 20 percent lower velocities. For comparison, la- the 15-degree finite-difference migration, Larner and
bel the correct position of the event with the steepest Beasley (1990) proposed performing migration using
dip from the desired migration on the results of migra- the 15-degree equation, repeatedly — the input to the
tion with different velocities. As in any other migration next migration stage being the output from the previ-
method, velocity errors cause events to be mispositioned ous stage. Such cascaded application of the 15-degree
at increasingly steeper dips. The undermigration effect migration is demonstrated in Figure 4.3-17. Migration
of lower velocities is reinforced by the inherently un- of the constant-velocity diffraction hyperbola using the
dermigrating nature of the 15-degree algorithm. As a 15-degree equation only once yields the familiar re-
result, dipping events are undermigrated more by the sult of unfocused energy accompanied with dispersive
15-degree algorithm in contrast than by an algorithm
noise (Figure 4.3-17c). A cascaded application of the
with no dip limitation (compare with Figure 4.2-12).
15-degree equation produces improved focusing of the
Figure 4.3-13 is the dipping events model with mi-
energy at the apex of the hyperbola. The more the num-
grations using the 3500 m/s medium velocity, and 5,
10, and 20 percent higher velocities. For comparison, ber of stages in the cascaded migration the more the im-
again, label the the correct position of the event with provement in focusing the energy (Figures 4.3-17d,e,f).
the steepest dip from the desired migration on the re- An interesting theoretical observation is that cas-
sults of migration with different velocities. The overmi- caded migration using a dip-limited algorithm, such as
gration effect of higher velocities is counteracted by the the 15-degree finite-difference scheme, actually requires
inherently undermigrating nature of the 15-degree algo- a depth step size that is greater than the optimal depth
rithm. As a result, dipping events are not overmigrated size for a single-stage application of the algorithm. In
(text continues on p. 530)
526 Seismic Data Analysis

FIG. 4.3-8. Tests for extrapolation depth step size in 15-degree finite-difference migration: (a) Desired migration using the
phase-shift method, (b) 4-ms depth step, and (c) 20-ms depth step. The input CMP stack is shown in Figure 4.3-2a.

FIG. 4.3-9. Tests for extrapolation depth step size in 15-degree finite-difference migration: (a) 40-ms depth step, (b) 60-ms
depth step, and (c) 80-ms depth step. The input CMP stack is shown in Figure 4.3-2a and the desired migration is shown in
Figure 4.3-2b.
Migration 527

FIG. 4.3-10. Tests for velocity errors in 15-degree finite- FIG. 4.3-11. Tests for velocity errors in 15-degree finite-
difference migration: (a) a zero-offset section that contains difference migration: (a) a zero-offset section that contains
a diffraction hyperbola with 2500-m/s velocity, (b) desired a diffraction hyperbola with 2500-m/s velocity, (b) desired
migration; 15-degree finite-difference migration using (c) the migration; 15-degree finite-difference migration using (c) the
medium velocity of 2500 m/s, (d) 5 percent lower, (e) 10 medium velocity of 2500 m/s, (d) 5 percent higher, (e) 10
percent lower, and (f) 20 percent lower velocity. Depth step percent higher, and (f) 20 percent higher velocity. Depth
size is 20 ms. step size is 20 ms.
528 Seismic Data Analysis

FIG. 4.3-13. Tests for velocity errors in 15-degree finite-


difference migration: (a) a zero-offset section that con-
FIG. 4.3-12. Tests for velocity errors in 15-degree finite- tains dipping events with 3500-m/s velocity, (b) desired mi-
difference migration: (a) a zero-offset section that con- gration; 15-degree finite-difference migration using (c) the
tains dipping events with 3500-m/s velocity, (b) desired mi- medium velocity of 3500 m/s, (d) 5 percent higher, (e) 10
gration; 15-degree finite-difference migration using (c) the percent higher, and (f) 20 percent higher velocity. Depth
medium velocity of 3500 m/s, (d) 5 percent lower, (e) 10 step size is 20 ms.
percent lower, and (f) 20 percent lower velocity. Depth step
size is 20 ms.
Migration 529

FIG. 4.3-14. Tests for velocity errors in 15-degree finite-difference migration using (a) 95 percent, (b) 90 percent, and (c) 80
percent of rms velocities. The input stacked section is shown in Figure 4.3-2a and the desired migration using the phase-shift
method is shown in Figure 4.3-2b. Depth step size is 20 ms.

FIG. 4.3-15. Tests for velocity errors in 15-degree finite-difference migration using (a) 105 percent, (b) 110 percent, and
(c) 120 percent of rms velocities. The input stacked section is shown in Figure 4.3-2a and the desired migration using the
phase-shift method is shown in Figure 4.3-2b. Depth step size is 20 ms.
530 Seismic Data Analysis

4.3-19). While this observation can be verified by the-


ory, the situation can also be remedied by a cleverly
implemented form of cascaded migration with a con-
stant velocity used in each stage (Larner and Beasley,
1990).
Actually, it turns out that cascaded migration the-
ory dictates constant velocity to be used in each stage.
If a variable velocity is used, then a 90-degree algorithm
such as the phase-shift method is required in lieu of a
dip-limited algorithm for each stage. While the cascaded
application of a dip-limited finite-difference algorithm
with a variable velocity causes overmigration (Figure
4.3-19c), the cascaded application of the phase-shift al-
gorithm with no dip limit yields an accurate image (Fig-
ure 4.3-20e).
Since the advancements made in practical im-
plementation of steep-dip implicit and explicit finite-
difference schemes, practical use of cascaded migration,
however, has been limited.

Reverse Time Migration

In Section 4.1, a migration algorithm based on extrap-


FIG. 4.3-16. The undermigration and overmigration ef- olation back in time while using the stacked section to
fects from Figures 4.3-14 and 4.3-15. B = dipping event be the boundary condition at z = 0 was discussed. The
before and A = dipping event after desired migration, D = impulse response of this algorithm, which is known as
diffraction before and D = diffraction after 15-degree finite-
reverse time migration, is shown in Figure 4.3-21. Note
difference migrations, and L = percent lower velocities, and
that the algorithm can handle dips up to 90 degrees
H = percent higher velocities.
with the accuracy of phase-shift migration. The impor-
tant consideration is that the extrapolation step ∆t in
fact, the more the number of stages in the cascade, the reverse time migration must be taken quite small, usu-
coarser the depth step size to achieve better focusing of ally a fraction of the input temporal sampling interval.
the energy (Figure 4.3-17). This then makes the algorithm computationally inten-
Cascaded migration of the constant-velocity zero- sive.
offset section that contains dipping events is shown Figure 4.3-22 shows a portion of a CMP-stacked
in Figure 4.3-18. Again, the 15-degree finite-difference section and its reverse time migration. The steep flanks
migration yields the familiar result of undermigrated of the salt diapirs have been imaged accurately, enabling
steeply dipping events accompanied by the dispersive delineation of the geometry of the top-salt boundary
noise. When applied in a cascaded manner, the algo- with confidence. Reverse time migration, albeit its sim-
rithm positions the steeply dipping events more accu- ple and elegant implementation (Section 4.1), has not
rately. With a sufficient number of cascades (Figure 4.3- been used widely in practice. Again, this is primar-
18f), the 15-degree algorithm can actually position the ily because it requires very small extrapolation step in
events as accurately as a 90-degree algorithm applied time, which increases the computational cost of the al-
only once (Figure 4.3-18b). For comparison, the event gorithm.
with the steepest dip (AB) is labeled on the desired
migration (Figure 4.3-18b) and the cascaded migration
(Figure 4.3-18f). 4.4 FREQUENCY-SPACE MIGRATION
Unfortunately, the encouraging results from cas- IN PRACTICE
caded migration using the 15-degree algorithm shown
in Figures 4.3-17 and 4.3-18 are only attainable for a The basis of the steep-dip implicit algorithms is the
constant-velocity medium. In case of a medium with dispersion relation of equation (4-18). Finite-difference
vertically varying velocity, the cascaded application of schemes with steep-dip accuracy are implemented con-
the 15-degree algorithm causes overmigration (Figure veniently in the frequency-space domain. An important
Migration 531

FIG. 4.3-17. (a) A zero-offset section that contains a diffraction hyperbola with 2500-m/s velocity, (b) desired migration
using the phase-shift method, (c) 15-degree finite-difference migration using a depth step of 20 ms; cascaded application of the
15-degree migration using (d) 4 cascades with 80-ms depth step, (e) 10 cascades with 200-ms depth step, and (f) 20 cascades
with 400-ms depth step.
532 Seismic Data Analysis

FIG. 4.3-18. (a) A zero-offset section that contains dipping events with 3500-m/s velocity, (b) desired migration using the
phase-shift method, (c) 15-degree finite-difference migration using a depth step of 20 ms; cascaded application of the 15-degree
migration using (d) 4 cascades with 80-ms depth step, (e) 10 cascades with 200-ms depth step, and (f) 20 cascades with 400-ms
depth step. For comparison, event AB with the steepest dip is labeled on the desired migration (b) and the cascaded migration
(f).
Migration 533

FIG. 4.3-19. (a) A zero-offset section that contains three diffraction hyperbolas with a vertically varying velocity, (b) 15-
degree finite-difference migration using a depth step of 20 ms; the output from the last stage of cascaded application of the
15-degree migration using (c) 4 cascades with 80-ms depth step, (d) 10 cascades with 200-ms depth step, (e) 20 cascades with
400-ms depth step, and (f) desired migration using the phase-shift method.
534 Seismic Data Analysis

FIG. 4.3-20. (a) A zero-offset section that contains three diffraction hyperbolas with a vertically varying velocity; four-stage
cascaded migration using the phase shift-method with 80-ms depth step: (b) first-stage, (c) second stage, (d) third stage, (e)
fourth stage, and (f) desired migration using the phase-shift method only once with 20-ms depth step. Compare the fourth
stage (e) with the output of the four-stage cascaded migration using a dip-limited algorithm as in Figure 4.3-19c.
Migration 535

be accurate for dips up to 65 degrees by tuning some co-


efficients (Section D.4). Higher-order operators can be
obtained by the successive application of a number of
operators like the 45-degree operator (Ma, 1981) with
a different set of coefficients (Lee and Suh, 1985). As
shown in Figure 4.4-1, with increasing dip accuracy,
the impulse responses of the algorithms approach the
shape of a semicircle. However, branches in the impulse
response associated with evanescent energy remain.
FIG. 4.3-21. Impulse response of a reverse time migration Figure 4.4-2 shows migration of a constant-velocity
algorithm. diffraction hyperbola using the 65-, 80-, 87-, and 90-
degree implicit schemes (Section D.4). While the fo-
cusing is better than that achieved by the 15-degree
advantage of the implicit method is its exceptional abil- implicit scheme (Figure 4.3-4c), note that the disper-
ity to handle velocity variations, whether vertical or sive noise still persists in the image obtained from the
lateral. Its accuracy for the lateral velocity problem re- 65-degree implicit scheme (Figure 4.4-2c). It is evident
sults from the fact that the time shift associated with that the quality of focusing from the 80-degree im-
the thin-lens term (equation 4-16b) can be implemented plicit scheme is superior (Figure 4.4-2d). The 87- and
exactly in the frequency domain. For these reasons, the 90-degree schemes have caused overmigration of the
algorithm is most appropriate for depth migration to diffraction hyperbola (Figures 4.4-2e,f).
image targets beneath complex structures (Chapter 8). The overmigration effect also can be observed on
The frequency-space, sometimes referred to as ω−x the results from the constant-velocity dipping events
or f − x, migration also has the important operational model in Figure 4.4-3. In fact, the dispersive noise that
advantage that each frequency can be processed sep- accompanies the steeply dipping events is present in all
arately. This property can reduce computer memory cases. The response of an implicit scheme is the product
requirements significantly and, thus, decrease input- of a complicated interplay of various parameters (Sec-
output operations for large data sets. Also, in frequency- tion D.6) — depth step size, sampling intervals in space
space migration, some accuracy features can be conve- and time, dip angle, velocity and frequency. Dispersive
niently implemented. For example, wave extrapolation noise, and under- or overmigration characteristics of im-
can be limited to a specified signal bandwidth. Each fre- plicit schemes depend on the specific implementation.
quency component can, in principle, be downward con- Figure 4.4-4 shows the stacked data migrated us-
tinued using an optimum depth step size that yields a ing three different approximations in frequency-space
minimum acceptable phase error, leading to a minimum domain — 15, 45, and 65 degrees. Note that by higher-
amount of dispersive noise on the migrated section. degree approximation, the collapse of the diffraction be-
comes complete, and the steeply dipping event is mi-
grated more accurately. Compare these results with the
Steep-Dip Implicit Methods desired migration in Figure 4.3-2b. Also note the similar
results obtained from the 15-degree time-space (t − x)
Figure 4.4-1 shows the impulse responses of a series of algorithm (Figure 4.3-2c) and the 15-degree frequency-
implicit frequency-space finite-difference schemes with space (ω − x) algorithm (Figure 4.4-4a).
different degrees of dip accuracy. Whether it is im- Figure 4.4-5 shows a field data example of
plemented in the time-space domain (Figure 4.3-1) or frequency-space implicit finite-difference migrations
frequency-space domain (Figure 4.4-1), the 15-degree with different degrees of accuracy. Compare the re-
algorithm yields an elliptic impulse response. The 45- sults with the desired migration in Figure 4.2-15b and
degree algorithm yields an impulse response in the note that the 80-degree scheme probably produces the
shape of a heart. most preferred image of the salt dome as compared
The 15-degree equation is derived from the Tay- to that from the 65-degree scheme. The schemes with
lor expansion of the dispersion relation (equation 4- steeper dip accuracy (87- and 90-degree schemes), how-
14a). The 45-degre equation is based on the continued ever, yield marginal improvements over the 80-degree
fractions expansion (equation 4-18), which allows wider scheme. Often the 65-degree scheme produces accept-
angle approximations. Kjartansson (1979) implemented able results, and the 80-degree scheme, which requires
the 45-degree equation for migration of stacked data. twice the computational effort, is used occasionally in
The 45-degree equation (4-18) can be upgraded to practice.
536 Seismic Data Analysis
Migration 537

FIG. 4.4-1. Impulse responses of the frequency-space implicit schemes with various degrees of approximations to the one-way
scalar wave equation. (See Section D.4 for the theoretical basis of these responses.)

Depth Step Size typical of finite-difference schemes, persists to varying


degrees irrespective of the depth step size. At smaller
Figure 4.4-6 shows the constant-velocity diffraction hy- depth steps, more energy collapses to the apex (Figure
perbola and the steep-dip implicit 65-degree finite- 4.4-7f).
difference migrations using four different depth steps. As in the case of the parabolic approximation (Fig-
Large depth steps cause undermigration as well as kinks ure 4.3-5), the dispersive noise that accompanies the
along the flank of the diffraction curve (especially ap- undermigrated energy (Figures 4.4-6 and 4.4-7) is an
parent in 60- and 80-ms cases). The dispersive noise, effect of approximating differential operators with dif-
538 Seismic Data Analysis

FIG. 4.4-2. (a) A zero-offset section that contains a diffraction hyperbola with 2500-m/s velocity, (b) desired migration using
the phase-shift method; frequency-space finite-difference migrations using (c) the 65-degree, (d) 80-degree, (e) 87-degree, and
(f) 90-degree implicit scheme.
Migration 539

FIG. 4.4-3. (a) A zero-offset section that contains dipping events with 3500-m/s velocity, (b) desired migration using the
phase-shift method; frequency-space finite-difference migrations using (c) the 65-degree, (d) 80-degree, (e) 87-degree, and (f)
90-degree implicit scheme.
540 Seismic Data Analysis

FIG. 4.4-4. Frequency-space finite-difference migrations of the CMP stack in Figure 4.3-2a using implicit finite-difference
schemes based on (a) the 15-degree, (b) 45-degree, and (c) 65-degree approximation. Note the increasingly better imaging
obtained using higher angle approximations. The desired migration is shown in Figure 4.3-2b.

ference operators. With reasonable depth steps (20 ms), creasing depth step size causes more and more undermi-
the undermigration caused by the dip limitation of the gration at increasingly steep dips. The waveform along
65-degree algorithm (Figure 4.4-7f) is less pronounced reflectors is dispersed at steep dips and large depth
as compared to that of the 15-degree algorithm (Figure steps. Kinks occur along reflectors at discrete intervals
4.3-4c). At large depth steps, the steep-dip accuracy is that correspond to the depth step size. Note that the
compromised, and the difference between the two algo- kinks are more pronounced at increasingly steeper dips.
rithms becomes less distinguishable (compare Figures It is apparent from Figure 4.4-9 that migration
4.4-6f and 4.3-4f). with 20-ms depth step has the least dispersion with the
The response of a finite-difference algorithm to a least undermigration — optimum accuracy in event po-
diffraction or a dipping event depends upon the type sitioning. Further decreasing depth step size actually
of differencing scheme and dip limitation. At large causes overmigration as for the diffraction hyperbola
depth steps, the steep-dip algorithm undermigrates the (Figure 4.4-7).
diffraction hyperbola (Figure 4.4-6) as does the 15- The response of a finite-difference algorithm to dip-
degree scheme (Figure 4.3-4). At small depth steps, the ping events depends again on the approximation made
steep-dip algorithm causes slight overmigration of the to the scalar wave equation. For instance, with small
diffraction hyperbola (Figure 4.4-7c,d), unlike the 15- depth steps less than one-half the dominant period
degree scheme which causes undermigration whatever of the reflection events, the steep-dip implicit scheme
the depth step size (Figure 4.3-5). causes overmigration of the reflection with the steep-
Figure 4.4-8 shows the constant-velocity dipping est dip (Figure 4.4-9c). The 15-degreee implicit scheme,
events model and the steep-dip implicit finite-difference on the other hand, causes postcursive dispersion along
migration results using four different depth step sizes. the steeply dipping reflectors (Figure 4.3-7c). Hence,
The response of the steep-dip implicit scheme is quite taking smaller depth steps does not necessarily mean
similar to that of the 15-degree scheme. Specifically, in- a better quality migration free of the artifacts that oc-
Migration 541

FIG. 4.4-5. Frequency-space finite-difference migrations of the CMP stack in Figure 4.2-15a using, from top to bottom, the
65-degree, 80-degree, 87-degree, and 90-degree implicit schemes. The desired migration using the phase-shift method is shown
in Figure 4.2-15b.
542 Seismic Data Analysis

FIG. 4.4-7. Tests for extrapolation depth step size in 65-


degree frequency-space implicit finite-difference migration:
(a) a zero-offset section that contains a diffraction hyperbola
FIG. 4.4-6. Tests for extrapolation depth step size in 65- with 2500-m/s velocity, (b) desired migration; 65-degree
degree frequency-space implicit finite-difference migration: finite-difference migrations using (c) 8-ms, (d) 12-ms, (e)
(a) a zero-offset section that contains a diffraction hyperbola 16-ms, and (f) 20-ms depth step.
with 2500-m/s velocity, (b) desired migration; 65-degree
finite-difference migrations using (c) 32-ms, (d) 40-ms, (e)
60-ms, and (f) 80-ms depth step.
Migration 543

FIG. 4.4-9. Tests for extrapolation depth step size in 65-


degree frequency-space implicit finite-difference migration:
(a) a zero-offset section that contains dipping events with
3500-m/s velocity, (b) desired migration; 65-degree finite-
difference migrations using (c) 8-ms, (d) 12-ms, (e) 16-ms,
and (f) 20-ms depth step.
FIG. 4.4-8. Tests for extrapolation depth step size in 65-
degree frequency-space implicit finite-difference migration:
(a) a zero-offset section that contains dipping events with
3500-m/s velocity, (b) desired migration; 65-degree finite-
difference migrations using (c) 32-ms, (d) 40-ms, (e) 60-ms,
and (f) 80-ms depth step.
544 Seismic Data Analysis

cur with the finite-difference method. With large depth than it would be by an algorithm with no dip limitation
steps, both schemes cause undermigration accompanied (compare with Figure 4.2-10).
by precursive dispersion along the dipping events (Fig- Figure 4.4-13 shows the constant-velocity diffrac-
ures 4.3-6 and 4.4-8). tion hyperbola and its migrations using the medium
Theoretically, the 65-degree differential equation velocity, and 5, 10, and 20 percent higher velocities.
is more accurate than the 15-degree differential equa- When velocities higher than medium velocity are used,
tion. However, once discretized, the difference in per- the diffraction hyperbola is overmigrated (Figure 4.4-
formance between these two equations can be less (Diet 13d,e,f), more so than with the 15-degree equation (Fig-
and Lailly, 1984). A good finite-difference migration ure 4.3-11d,e,f). On the other hand, the diffraction hy-
program uses differencing schemes that maintain the perbola gets overmigrated by the steep-dip algorithm
dip accuracy implied by the corresponding differential less than it would be by an algorithm with no dip limi-
equation. tation (compare with Figure 4.2-11). Whatever the ve-
The main point to remember is that migration of locity used for migration, dispersive noise is present per-
steep dips generally requires a small depth step size. sistently in finite-difference results (Figures 4.4-12 and
Practical considerations suggest that depth step size 4.4-13).
should be between one-half and one-full dominant pe- Figure 4.4-14 shows the constant-velocity dipping
riod of the seismic data to be migrated (from 20 to 40 events model with migrations using the medium veloc-
ms), depending on the steepness of the dips in the data. ity, and velocities that are 5, 10, and 20 percent lower.
A higher order approximation, such as the 65- For comparison, label the correct position of the event
degree equation, provides a smaller range of choice for with the steepest dip from the desired migration on the
optimum depth size as compared to the 15-degree equa- results of migration with different velocities. As in any
tion. In particular, note that the optimum depth step other migration method, velocity errors cause events to
size is 20 ms for the dipping events model shown in be increasingly more mispositioned at steeper dips. The
Figures 4.4-8 and 4.4-9; any departure from this value undermigration effect of lower velocities is reinforced by
causes either undermigration or overmigration. The re- the inherently undermigrating nature of the steep-dip
sults with the 20-ms depth step in Figures 4.3-4 and algorithm, although not as much as in the case of the
4.4-9 convincingly show that the 65-degree algorithm
15-degree algorithm (Figure 4.3-12). As a result, dip-
can migrate steeper dips and collapse diffractions more
ping events are more undermigrated by the steep-dip
accurately than the 15-degree equation.
algorithm in contrast to an algorithm with no dip limi-
Figures 4.4-10 and 4.4-11 show migrations of the
tation (compare with Figure 4.2-12).
stacked section in Figure 4.2-15a with eight different
Figure 4.4-15 is the dipping events model with mi-
depth steps using a steep-dip implicit scheme. Note that
grations using the 3500 m/s medium velocity, and 5,
as the depth step size is increased, more undermigra-
10, and 20 percent higher velocities. For comparison,
tion occurs. Dispersion along the undermigrated event
again, label the correct position of the event with the
associated with the steep left flank of the salt dome
steepest dip from the desired migration on the results
is apparent at larger depth steps. A 20-ms depth step
of migration with different velocities. The overmigration
size, which is equivalent to the usual case of one-half
effect of higher velocities is counteracted by the inher-
the dominant period of seismic data, generally is an ac-
ceptable choice for most of the implicit finite-difference ently undermigrating nature of the steep-dip algorithm,
schemes. although not as much as in the case of the 15-degree al-
gorithm (Figure 4.3-13). As a result, dipping events are
not overmigrated by the steep-dip algorithm as much as
Velocity Errors they would be by an algorithm with no dip limitation
(compare with Figure 4.2-13).
We now examine the response of a steep-dip implicit The nature of the precursive and postcursive dis-
algorithm to velocity errors. Figure 4.4-12 shows the persion along steeply dipping events in the output from
constant-velocity diffraction hyperbola and its migra- finite-difference migration, theoretically depends on the
tions using the medium velocity, and 5, 10, and 20 type of differencing scheme used in approximating the
percent lower velocities. When velocities lower than differential operators associated with the scalar wave
medium velocity are used, the diffraction hyperbola is equation. In principle, an implicit finite-difference mi-
undermigrated (Figure 4.4-12d,e,f), but not as much gration implemented in the frequency-space domain
as in the case of the 15-degree equation (Figure 4.3- yields less dispersion compared to an implicit scheme
10d,e,f). On the other hand, the diffraction hyperbola implemented in the time-space domain. This is because
gets undermigrated by the steep-dip algorithm more the former requires differencing in x and z, while the
Migration 545

FIG. 4.4-10. Tests for extrapolation depth step size in 65-degree frequency-space implicit finite-difference migration using,
from top to bottom, 32-ms, 40-ms, 60-ms, and 80-ms depth steps. The input CMP stack is shown in Figure 4.2-15a, and the
desired migration using the phase-shift method is shown in Figure 4.2-15b.
546 Seismic Data Analysis

FIG. 4.4-11. Tests for extrapolation depth step size in 65-degree frequency-space implicit finite-difference migration using,
from top to bottom, 8-ms, 12-ms, 16-ms, and 20-ms depth steps. The input CMP stack is shown in Figure 4.2-15a, and the
desired migration using the phase-shift method is shown in Figure 4.2-15b.
Migration 547

FIG. 4.4-13. Tests for velocity errors in 65-degree


frequency-space implicit finite-difference migration: (a) a
FIG. 4.4-12. Tests for velocity errors in 65-degree
zero-offset section that contains a diffraction hyperbola with
frequency-space implicit finite-difference migration: (a) a
2500-m/s velocity, (b) desired migration; 65-degree finite-
zero-offset section that contains a diffraction hyperbola with
difference migration using (c) the medium velocity of 2500
2500-m/s velocity, (b) desired migration; 65-degree finite-
m/s, (d) 5 percent higher, (e) 10 percent higher, and (f) 20
difference migration using (c) the medium velocity of 2500
percent higher velocity. Depth step size is 20 ms.
m/s, (d) 5 percent lower, (e) 10 percent lower, and (f) 20
percent lower velocity. Depth step size is 20 ms.
548 Seismic Data Analysis

FIG. 4.4-15. Tests for velocity errors in 65-degree


FIG. 4.4-14. Tests for velocity errors in 65-degree frequency-space implicit finite-difference migration: (a)
frequency-space implicit finite-difference migration: (a) a zero-offset section that contains dipping events with
a zero-offset section that contains dipping events with 3500-m/s velocity, (b) desired migration; 65-degree finite-
3500-m/s velocity, (b) desired migration; 65-degree finite- difference migration using (c) the medium velocity of 3500
difference migration using (c) the medium velocity of 3500 m/s, (d) 5 percent higher, (e) 10 percent higher, and (f) 20
m/s, (d) 5 percent lower, (e) 10 percent lower, and (f) 20 percent higher velocity. Depth step size is 20 ms.
percent lower velocity. Depth step size is 20 ms.
Migration 549

latter requires differencing in t in addition to x and accuracy, and a 25-point filter with a 70-degree accu-
z. In practice, however, dispersive noise contaminates racy. Note that the explicit extrapolation filter cuts off
migration results from implicit schemes almost without beyond the specified dip limit. In contrast, the implicit
exception. extrapolation treats the evanescent energy as though it
Velocity error tests on field data using the 65- is part of the propagating energy (Figure 4.4-1).
degree implicit scheme are shown in Figures 4.4-16 and
4.4-17. With velocities erroneously too low, note the
pronounced undermigration of the steep salt flank. With Dip Limits of Extrapolation Filters
velocities erroneously too high, note the overmigration
of the steep salt flank and the crossing of reflections at
Figure 4.4-19 shows a zero-offset section that contains a
the vicinity of the crest of the salt structure.
diffraction hyperbola and its migrations using the 30-,
50- and 70-degree explicit schemes. Also shown is the
desired migration using the phase-shift method. Note
Steep-Dip Explicit Methods that the low-dip algorithm causes undermigration of the
diffraction hyperbola, while the steep-dip algorithm fo-
Although implicit schemes are always stable and nat- cuses the energy at the apex better than a steep-dip
urally suitable to accommodate lateral velocity varia- implicit scheme with equivalent dip accuracy (Figure
tions, they have been known to have some unfavorable 4.4-7).
aspects. Specifically, dispersive noise caused by the dif- Explicit schemes are designed based on a specified
ferencing approximations to differential operators does cutoff wavenumber kx (Section D.5), beyond which the
indeed deteriorate the quality of migration of real data amplitude spectrum of the filter is set to zero. Figure
(Figures 4.3-8, 4.3-9, 4.4-10, and 4.4-11). An equally im- 4.4-20 shows the f − k spectra of the results of migra-
portant concern about implicit schemes is the fact that tion of the diffraction hyperbola as shown in Figure 4.4-
they never meet the expected theoretical dip accuracy. 19 using the three explicit extrapolators. Note that the
A 45-degree algorithm can actually provide only a 5- lower the dip limit, the lower the cutoff wavenumber.
degree acccuracy for certain frequencies and velocities. The steep-dip algorithm with a 70-degree dip limit is
Because of a complex interaction of temporal and spa- able to replicate the amplitude response characteristics
tial sampling rates, dip, velocity, and frequency, an op- of the desired migration using the phase-shift method.
timum depth step size that yields minimum phase and This of course is at the expense of using a long, complex
amplitude errors for an entire data set is never easy to convolutional filter.
specify (Section D.6). Figure 4.4-21 shows a zero-offset section that con-
While implicit schemes are guaranteed to be sta- tains a set of dipping events and its migrations using
ble whatever the depth step size, stability of explicit the 30-, 50- and 70-degree explicit schemes. Also shown
schemes requires a depth step that is sufficiently small is the desired migration using the phase-shift method.
(Section D.6). Putting aside the stability issue, explicit For comparison, label the the correct position of the
schemes implemented in the frequency-space domain event with the steepest dip from the desired migration
(Section D.5) can alleviate some of these deficiencies on the results of migration with different velocities. The
of implicit schemes. steep-dip algorithm positions the events nearly as good
Refer to the impulse responses shown in Figure 4.4- as the desired migration using the phase-shift method
18 and note that explicit schemes do not produce any and better than a steep-dip implicit scheme with equiv-
visible dispersion along steep dips. Explicit schemes in- alent dip accuracy (Figure 4.4-9).
volve convolution of a symmetric complex filter in the Figure 4.4-22 shows the f − k spectra of the results
frequency-space domain with the wavefield to perform of migration of the dipping events as shown in Figure
extrapolation in depth. In contrast, implicit schemes 4.4-21 using the three explicit extrapolators. The ex-
can be computationally intensive for large volumes of trapolation filter with the 30-degree dip limit has trun-
data since they require solving complex tridiagonal cated the steeply dipping events at high wavenumbers
equations associated with the differencing in the x di- kx . Except for very high wavenumbers, the extrapola-
rection. tion filter with the 70-degree dip limit is able to repli-
Steep-dip accuracy in explicit schemes is attained cate the amplitude response characteristics of the de-
by increasing the extrapolation filter length. The im- sired migration using the phase-shift method. Keep in
pulse responses shown in Figure 4.4-18 are associated mind that we achieve stability of explicit schemes at the
with three different extrapolators — a 7-point filter with expense of truncating the response at wavenumbers kx
a 30-degree accuracy, an 11-point filter with a 50-degree above a specified cutoff value (Section D.5).
550 Seismic Data Analysis

FIG. 4.4-16. Tests for velocity errors in 65-degree frequency-space implicit finite-difference migration using interval velocities
derived from, from top to bottom, 100, 95, 90, and 80 percent of rms velocities. The input stacked section is shown in Figure
4.2-15a, and the desired migration using the phase-shift method is shown in Figure 4.2-15b. Depth step size is 20 ms.
Migration 551

FIG. 4.4-17. Tests for velocity errors in 65-degree frequency-space implicit finite-difference migration using interval velocities
derived from, from top to bottom, 100, 105, 110, and 120 percent of rms velocities. The input stacked section is shown in
Figure 4.2-15a, and the desired migration using the phase-shift method is shown in Figure 4.2-15b. Depth step size is 20 ms.
552 Seismic Data Analysis

FIG. 4.4-18. (a) Impulse response of a desired migration algorithm using the phase-shift method; impulse responses of (b)
a 30-degree, (c) 50-degree, and (d) 70-degree frequency-space explicit scheme for migration.

A field data example of a stacked section which Velocity Errors


has been migrated using the 30-degree, 50-degree and
70-degree extrapoaltion filters is shown in Figure 4.4-23.
The steep flanks of the salt diapirs are clearly better im- Figure 4.4-25 shows migrations of a zero-offset section
aged by the steep-dip extrapolation filter. For compari- that contains a diffraction hyperbola using a frequency-
son, Figure 4.4-24 shows the desired migration using the space explicit algorithm based on 30-degree, 50-degree
phase-shift method. Although the length of the extrap- and 70-degree extrapolation filters and a velocity that
olation filter for a steep-dip algorithm is much longer is 90 percent of the medium velocity. For comparison,
than that for a low-dip extrapolation filter, the benefit desired migration using the phase-shift method with the
of using the former is indisputably demonstrated by the medium velocity and 90 percent of the medium velocity
field data example shown in Figure 4.4-23. are also shown in the same figure. Migration with an
Migration 553

FIG. 4.4-19. (a) A zero-offset section that contains a


diffraction hyperbola with 2500-m/s velocity, (b) desired
migration using the phase-shift method; migrations using
frequency-space explicit schemes with (c) 30-degree, (d) 50-
degree, and (e) 70-degree accuracy.

FIG. 4.4-20. The f −k spectra of the sections in Figure 4.4-


19: (a) A zero-offset section that contains a diffraction hy-
perbola with 2500-m/s velocity, (b) desired migration using
the phase-shift method; migrations using frequency-space
explicit schemes with (c) 30-degree, (d) 50-degree, and (e)
70-degree accuracy.
554 Seismic Data Analysis

FIG. 4.4-21. (a) A zero-offset section that contains dipping


events with 3500-m/s velocity, (b) desired migration using
the phase-shift method; migrations using frequency-space
explicit schemes with (c) 30-degree, (d) 50-degree, and (e)
70-degree accuracy.
FIG. 4.4-22. The f − k spectra of the sections in Fig-
erroneously low velocity yields the undermigrated form ure 4.4-21: (a) A zero-offset section that contains dipping
of the diffraction hyperbola as seen in Figure 4.4-25c. events with 3500-m/s velocity, (b) desired migration using
Note, however, the dip-limited explicit schemes appear the phase-shift method; migrations using frequency-space
to cause less undermigration compared to the phase- explicit schemes with (c) 30-degree, (d) 50-degree, and (e)
shift method with 90-degree accuracy. This behavior is 70-degree accuracy.
in contradiction to intuition — the undermigration ef-
fect of an erroneously low velocity is reinforced by a the steep limbs of the undermigrated diffraction hyper-
dip-limited algorithm. In fact, this intuitive effect was bola are truncated (Figure 4.4-25). This in turn makes
demonstrated by the steep-dip implicit scheme (Figure the result of migration using lower velocity appear less
4.4-12). undermigrated in case of an explicit scheme compared
The deceptive behavior of the explicit schemes to the case of the phase-shift method with 90-degree ac-
that contradicts our intuition can be explained by the curacy. In fact, if we apply a wavenumber filter to reject
fact that these schemes filter out the energy at high the high wavenumbers from the output of phase-shift
wavenumbers (Figures 4.4-19 and 4.4-20). As a result, migration (Figure 4.4-25c), the result would resemble
Migration 555
556 Seismic Data Analysis

FIG. 4.4-24. Migration of the CMP stack shown in Figure 4.3-22a using the phase-shift method.
Migration 557

FIG. 4.4-25. Tests for velocity errors in frequency-space FIG. 4.4-26. Tests for velocity errors in frequency-space
explicit migration: (a) a zero-offset section that contains a
explicit migration: (a) a zero-offset section that contains a
diffraction hyperbola with 2500-m/s velocity, (b) desired mi-
gration using the phase-shift method, and (c) phase-shift diffraction hyperbola with 2500-m/s velocity, (b) desired mi-
migration using 10 percent lower velocity; migrations us- gration using the phase-shift method, and (c) phase-shift
ing 10 percent lower velocity and frequency-space explicit migration using 10 percent lower velocity; migrations us-
schemes with (d) 30-degree, (e) 50-degree, and (f) 70-degree ing 10 percent higher velocity and frequency-space explicit
accuracy. schemes with (d) 30-degree, (e) 50-degree, and (f) 70-degree
accuracy.
the output of an explicit scheme (Figures 4.4-25d,e,f).
Also note from Figure 4.4-25 that the low-dip ex-
plicit scheme manifests the effect of undermigration frequency-space explicit algorithm based on 30-degree,
much less than the steep-dip explicit scheme. This is 50-degree and 70-degree extrapolation filters and a ve-
because the wavenumber filtering effect is more se- locity that is 110 percent of the medium velocity. Again,
vere for the low-dip explicit scheme (Figure 4.4-20). for comparison, desired migration using the phase-shift
Figure 4.4-26 shows migrations of a zero-offset method with the medium velocity and 110 percent of
section that contains a diffraction hyperbola using a the medium velocity are also shown in the same figure.
558 Seismic Data Analysis

FIG. 4.4-28. Tests for velocity errors in frequency-space


explicit migration: (a) A zero-offset section that contains
FIG. 4.4-27. Tests for velocity errors in frequency-space ex- dipping events with 3500-m/s velocity, (b) desired migration
plicit migration: (a) a zero-offset section that contains dip- using the phase-shift method, and (c) phase-shift migration
ping events with 3500-m/s velocity, (b) desired migration using 10 percent higher velocity; migrations using 10 percent
using the phase-shift method, and (c) phase-shift migration higher velocity and frequency-space explicit schemes with
using 10 percent lower velocity; migrations using 10 percent (d) 30-degree, (e) 50-degree, and (f) 70-degree accuracy.
lower velocity and frequency-space explicit schemes with (d)
30-degree, (e) 50-degree, and (f) 70-degree accuracy.
Migration 559

Migration with an erroneously high velocity yields the ated by using a steep-dip explicit scheme (Figure 4.4-
overmigrated form of the diffraction hyperbola as seen 32j). A steep-dip implicit scheme, on the other hand,
in Figure 4.4-26c. can actually overshoot in the opposite direction and
We make the following observations from Figure cause overmigration (Figure 4.4-32f). The differencing
4.4-26: approximations are manifested by the dispersive noise
(Figure 4.4-32c).
(a) The dip-limited explicit schemes, much like the im- Figure 4.4-33 shows the compilation of the results
plicit schemes (Figure 4.4-13), cause less overmi- of migration of a zero-offset section that contains a set
gration compared to the phase-shift method with of dipping events. Again, for comparison, desired mi-
90-degree accuracy. gration using the phase-shift method is included in the
(b) The low-dip explicit scheme manifests the effect of panel. The dip-limited nature of the implicit and ex-
overmigration much less than the steep-dip explicit plicit schemes is manifested by the undermigration of
scheme. the steeply dipping events (Figures 4.4-33c,g,h). This
(c) The wavenumber filtering effect (Figure 4.4-20) fur- effect is alleviated by using a steep-dip explicit scheme
ther truncates the steep limbs of the overmigrated (Figure 4.4-33j). A steep-dip implicit scheme, on the
hyperbola. other hand, can actually overshoot in the opposite di-
rection and cause overmigration (Figure 4.4-33f).The
The interplay of the three factors results in the differencing approximations are manifested by the dis-
response to velocity errors by the explicit schemes as persive noise accompanying the steeply dipping events
shown in Figure 4.4-26. (Figure 4.4-33c,d,e,f).
Tests for velocity errors are repeated for a zero-
offset section that contains a set of dipping events as
shown in Figures 4.4-27 and 4.4-28. For comparison, la- 4.5 FREQUENCY-WAVENUMBER
bel the correct position of the event with the steepest
MIGRATION IN PRACTICE
dip from the desired migration on the results of migra-
tion with different velocities. The residual diffractions
off the end of the dipping reflectors on the migrated Two different methods of migration are implemented in
sections from the explicit schemes have been truncated the frequency-wavenumber domain. The Stolt method is
by the wavenumber filtering effect of the extrapolation exact for a constant-velocity medium, while the phase-
filters. This filtering effect is most prominent in the case shift method is exact for a medium with vertical ve-
of the explicit scheme with the low-dip limit and erro- locity variations. The Stolt method can be extended
neously high velocity (Figure 4.4-28d). to the case of a medium with lateral velocity varia-
Field data examples for tests of velocity errors for tions judged to be acceptable for time migration. The
the explicit schemes are shown in Figures 4.4-29, 4.4- phase-shift method also can be extended to handle
30, and 4.4-31. First, note the better imaging of the lateral velocity variations. One extension, phase-shift-
salt flanks by the steep-dip explicit scheme compared plus-interpolation scheme, accommodates lateral varia-
to the low-dip explicit scheme (Figure 4.4-29). The un- tions in velocity by interpolating between the results
dermigration effect of erroneously low velocities (Figure of migration using a group of vertically-varying ve-
4.4-30) and the overmigration effect of erroneously high locity functions. Another extension, phase-shift-plus-
velocities (Figure 4.4-31) may be compared with the correction, applies an additional extrapolation opera-
results of migration using optimum velocities (Figure tor at each depth step to account for differences be-
4.4-29). tween the laterally varying velocity field and vertically
We shall complete this section by reviewing the varying-only velocity function used for phase-shift mi-
performance of Kirchhoff summation, finite-difference gration. The phase-shift method also has been extended
frequency-space implicit, and frequency-space explicit to image turning waves associated with salt overhang
schemes with various dip limits. Figure 4.4-32 shows the structures.
compilation of the results of migration of a zero-offset
section that contains a diffraction hyperbola. For com-
parison, desired migration using the phase-shift method Maximum Dip to Migrate
is included in the panel. The dip-limited nature of the
implicit and explicit schemes is manifested by the in- The phase-shift method of migration (Section 4.1 and
complete focusing of the energy at the apex of the Section D.7) allows vertical variations in velocity and
diffraction hyperbola (Figures 4.4-32c,g,h). The under- is accurate for up to dips of 90 degrees. Figure 4.5-1
migration effect caused by the dip limitation is allevi- shows the impulse response of the phase-shift algorithm.
(text continues on p. 565)
560 Seismic Data Analysis

FIG. 4.4-29. From top to bottom, desired migration as in Figure 4.2-15b and frequency-space explicit migration with 30-
degree, 50-degree, and 70-degree accuracy, and using interval velocities derived from 100 percent of the rms velocities. The
input CMP stack is shown in Figure 4.2-15a.
Migration 561

FIG. 4.4-30. From top to bottom, desired migration as in Figure 4.2-15b and tests for velocity errors in frequency-space
explicit migration with 30-degree, 50-degree, and 70-degree accuracy, and using interval velocities derived from 90 percent of
rms velocities. The input CMP stack is shown in Figure 4.2-15a.
562 Seismic Data Analysis

FIG. 4.4-31. From top to bottom, desired migration as in Figure 4.2-15b and tests for velocity errors in frequency-space
explicit migration with 30-degree, 50-degree, and 70-degree accuracy, and using interval velocities derived from 110 percent
of rms velocities. The input CMP stack is shown in Figure 4.2-15a.
Migration 563

FIG. 4.4-32. Summary of the results of migration of a zero-offset section that contains a diffraction hyperbola with 2500-
m/s velocity as in (a): (b) Kirchhoff migration; frequency-space implicit finite-difference migration with (c) 65-degree, (d)
80-degree, (e) 87-degree, and (f) 90-degree accuracy; frequency-space explicit migration with (g) 30-degree, (h) 50-degree, and
(i) 70-degree accuracy; and (j) frequency-wavenumber phase-shift migration.
564 Seismic Data Analysis

FIG. 4.4-33. Summary of the results of migration of a zero-offset section that contains a set of dipping events with 3500-
m/s velocity as in (a): (b) Kirchhoff migration; frequency-space implicit finite-difference migration with (c) 65-degree, (d)
80-degree, (e) 87-degree, and (f) 90-degree accuracy; frequency-space explicit migration with (g) 30-degree, (h) 50-degree, and
(i) 70-degree accuracy; and (j) frequency-wavenumber phase-shift migration.
Migration 565

FIG. 4.5-1. Impulse response of phase-shift migration has


a semicircular shape.

FIG. 4.5-2. The impulse response of the f −k migration op-


erator has a truncated semicircular shape when a maximum
dip limit is imposed. For comparison, the desired response
shape has been superimposed on the f − k responses.

Clearly, for a constant-velocity medium, this respone is


equivalent to that of the Stolt migration. The impulse
response shown in Figure 4.5-1 is considered to be the
desired impulse response for 2-D zero-offset migration,
and as such, responses of all migration algorithms dis-
cussed in this chapter are benchmarked against it.
As with the Kirchhoff summation method, migra- FIG. 4.5-3. Tests for maximum dip to migrate in phase-
tion with the phase-shift method can be limited to shift migration: (a) a zero-offset section that contains dip-
smaller dips by truncating the semicircular wavefront ping events with 3500-m/s velocity, (b) desired migration;
(Figure 4.5-2). This dip filtering capability is useful in phase-shift migrations using (c) 2-ms/trace, (d) 4-ms/trace,
rejecting coherent noise from the stacked section while (e) 8-ms/trace, and (f) 16-ms/trace maximum dip limit.
migrating the data. If migration is constrained to small
dip values, then the steeply dipping reflectors may be
filtered out unintentionally. Edge effects also are pro- sults shown in Figure 4.5-3. Note that steep dips greater
nounced when a very narrow range of dips is passed. than the specified maximum dip to migrate have been
Note the linear streaks on the impulse response with a annihilated. On the field data example shown in Figure
dip limit of 2 ms/trace (Figure 4.5-2). 4.5-4, severe dip filtering action of the 2 ms/trace max-
The dip-filtering action caused by imposing a dip imum dip has caused smearing and eliminated virtually
limit on the impulse response also is visible on the re- all of the signal contained in the section.
566 Seismic Data Analysis

FIG. 4.5-4. Tests for maximum dip to migrate in phase-shift migration: A low value for maximum dip to migrate can be
hazardous. All dips of interest must be preserved during migration.

Depth Step Size results (Figure 4.4-8). As with the finite-difference algo-
rithms, the problem occurs along the steeper dips first;
Figure 4.5-5 shows a zero-offset section that contains
therefore, the steep dips require smaller depth steps
a set of dipping events migrated with the phase-shift (Figure 4.5-5).
method using different depth step sizes. Since the phase- Because of the band-limited nature of seismic data,
shift method is based on the dispersion relation given by very small depth steps are not needed. From Figure
equation (4-13b) of the exact one-way wave equation, 4.5-5, note that migration with a 20-ms depth step pro-
we do not expect undermigration. However, we see dis- duces a section without spurious kinks along the reflec-
continuities along the reflectors at intervals equal to the tors; this is comparable to the desired migration using
depth step size, which is similar to the finite-difference a depth step equal to the temporal sampling interval.
Migration 567

the positioning of events. The only problem with large


depth steps is the kinks along the steep dips. In prin-
ciple, as long as there is no aliasing in the z-direction,
the depth step kinks can be eliminated by a local inter-
polation process. In practice, as for the implicit finite-
difference methods, the depth step size used in migra-
tion with the phase-shift method typically is taken be-
tween the half-and full-dominant period of the wave-
field — 20 to 40 ms, depending on steepness of the dips
present in the section.

Velocity Errors

We now examine the response of phase-shift migration


to velocity errors. Figure 4.5-7 shows the diffraction hy-
perbola and phase-shift migrations using the 2500 m/s
medium velocity, and 5, 10, and 20 percent lower ve-
locities. The lower the velocity, the more the diffrac-
tion hyperbola is undermigrated. These results are used
as a benchmark to evaluate the response of the other
migration algorithms discussed in this chapter. Specif-
ically, compare Figure 4.5-7 with Figures 4.2-10 (the
case of Kirchhoff migration), 4.3-10 (the case of implicit
finite-difference migration), 4.4-12 (the case of implicit
frequency-space migration), and 4.4-25 (the case of ex-
plicit frequency-space migration), note how the various
algorithms respond to velocity errors with significant
differences.
Figure 4.5-8 shows the same diffraction hyperbola
and phase-shift migrations using the 2500 m/s medium
velocity, and 5, 10, and 20 percent higher velocities.
The higher the velocity, the more the diffraction hyper-
bola is overmigrated. These results are used to bench-
mark the response to velocity errors by the other migra-
tion algorithms — Figures 4.2-11 (the case of Kirchhoff
migration), 4.3-11 (the case of implicit finite-difference
migration), 4.4-13 (the case of implicit frequency-space
migration), and 4.4-26 (the case of explicit frequency-
space migration).
The under- and overmigration effects caused by the
use of erroneously low or high velocities on the dip-
ping events model are seen in Figures 4.5-9 and 4.5-
10, respectively. Label the correct position of the event
FIG. 4.5-5. Tests for extrapolation depth step size in phase-
with the steepest dip from the desired migration on
shift migration: (a) a zero-offset section that contains dip-
the results of migrations with different velocities, and
ping events with 3500-m/s velocity, (b) desired migration;
phase-shift migrations using (c) 20-ms (d) 40-ms, (e) 60-ms,
note the event mispositioning caused by erroneously low
and (f) 80-ms depth step. and high velocities. Recall that sensitivity of migration
to velocity errors can be measured quantitatively via
equations 4-1 and 4-2. The test results shown in Fig-
Depth step size tests on field data are shown in ures 4.5-9 and 4.5-10 are used to benchmark the re-
Figure 4.5-6. Unlike the finite-difference results (Figure sponse to velocity errors by the other migration algo-
4.4-10), the phase-shift migration with different depth rithms — Figures 4.2-12 and 4.2-13 (the cases of Kirch-
step sizes produces equally adequate results in terms of hoff migration), 4.3-12 and 4.3-13 (the cases of implicit
568 Seismic Data Analysis

FIG. 4.5-6. Tests for extrapolation depth step size in phase-shift migration: Note the kinks along steep dips with large depth
steps.

finite-difference migration), 4.4-14 and 4.4-15 (the cases degrees (Figure 4.5-1). Of course, we must also remind
of implicit frequency-space migration), and 4.4-27 and ourselves of the fact that the phase-shift method is lim-
4.4-28 (the cases of explicit frequency-space migration). ited to velocities that vary only in the vertical direction.
An aspect of phase-shift migration uniquely dif- Figures 4.5-11 and 4.5-12 are field data examples
ferent from others is its exceptional quality of output. of phase-shift migration using erroneously low and high
As noted from Figures 4.5-7, 4.5-8, 4.5-9, and 4.5-10, velocities, respectively. Figure 4.5-13 shows a sketch of
phase-shift migration produces no dispersive noise since the combined results of these migrations. Clearly, ve-
it does not involve any differencing of differential oper- locities that are too low cause undermigration of the
ators. Instead, the entire design of extrapolation oper- steeply dipping event that defines the flank of the salt
ator and application of migration are in the frquency- diapir and incomplete collapse of the diffraction off the
wavenumber domain. The results do not suffer from any tip of the salt diapir. Velocities that are too high cause
dip limitation since the phase-shift method is based on overmigration as manifested by the crossing events at
an extrapolation filter that is exact for all dips up to 90 the vicinity of the crest of the salt diapir.
Migration 569

FIG. 4.5-7. Tests for velocity errors in phase-shift migra-


tion: (a) a zero-offset section that contains a diffraction
hyperbola with 2500-m/s velocity, (b) desired migration; FIG. 4.5-8. Tests for velocity errors in phase-shift migra-
phase-shift migration using (c) the medium velocity of 2500 tion: (a) a zero-offset section that contains a diffraction
m/s, (d) 5 percent lower, (e) 10 percent lower, and (f) 20 hyperbola with 2500-m/s velocity, (b) desired migration;
percent lower velocity. phase-shift migration using (c) the medium velocity of 2500
m/s, (d) 5 percent higher, (e) 10 percent higher, and (f) 20
percent higher velocity.
Figure 4.5-14 shows the results of phase-shift mi-
gration of the stacked section in Figure 4.2-15 using Figure 4.5-15 shows results of phase-shift migration
velocities lower than what may be optimum for imag- of the stacked section in Figure 4.2-15 using velocities
ing. The under-migration of the left flank of the salt higher than what may be optimum for imaging. Migra-
dome is not so evident. However, the steeply dipping tion with erroneously high velocities (110 and 120 per-
reflection off the right flank intersects the gently dip- cent of optimum velocities) shows signs of overmigration
ping reflections associated with the surrounding strata in the form of crossing events along the left flank of the
— an indication of undermigration. salt diapir.
570 Seismic Data Analysis

FIG. 4.5-9. Tests for velocity errors in phase-shift migra- FIG. 4.5-10. Tests for velocity errors in phase-shift migra-
tion: (a) a zero-offset section that contains dipping events tion: (a) a zero-offset section that contains dipping events
with 3500-m/s velocity, (b) desired migration; phase-shift with 3500-m/s velocity, (b) desired migration; phase-shift
migration using (c) the medium velocity of 3500 m/s, (d) 5 migration using (c) the medium velocity of 3500 m/s, (d)
percent lower, (e) 10 percent lower, and (f) 20 percent lower 5 percent higher, (e) 10 percent higher, and (f) 20 percent
velocity. higher velocity.
Migration 571

FIG. 4.5-11. Tests for velocity errors in phase-shift migration using interval velocities derived from (a) 95 percent, (b) 90
percent, and (c) 80 percent of rms velocities. The input stacked section is shown in Figure 4.3-2a, and the desired migration
using the phase-shift method is shown in Figure 4.3-2b.

FIG. 4.5-12. Tests for velocity errors in phase-shift migration using interval velocities derived from (a) 105 percent, (b) 110
percent, and (c) 120 percent of rms velocities. The input stacked section is shown in Figure 4.3-2a, and the desired migration
using the phase-shift method is shown in Figure 4.3-2b.
572 Seismic Data Analysis

to a scalar (Section D.7). The theoretical range of W is


between 0 and 2.
To understand the effects of the stretch factor, re-
fer to the impulse responses in Figure 4.5-16, where a
single, isolated wavelet on a single trace is migrated us-
ing various stretch factors. Here, W = 1 corresponds
to the exact constant-velocity Stolt algorithm. So, set-
ting W < 1 compresses the impulse response inward
along its steep flanks, while setting W > 1 opens it up.
Thus, the value of W partially controls the aperture of
the generalized Stolt algorithm. The farther W is from
1, the more limited the aperture becomes. A value of
W < 1 implies undermigration at steeper dips, while a
value of W > 1 implies overmigration at steeper dips,
if the medium velocity is constant.
Although not strictly implied by the impulse re-
sponses in Figure 4.5-16, when using a stretch factor
different from 1, the Stolt algorithm tries to emulate a
wavefront in a variable velocity medium (Stolt, 1978),
while compromising on the ability to migrate steeper
dips. Experience has proven that the Stolt migration
with stretch produces acceptable results provided ve-
locity variations are within limits of time migration.
Consider the zero-offset section and the migration
results in Figure 4.5-17. Stretch factor W = 1 produces
the best migrated section because the zero-offset sec-
tion was modeled using a constant-velocity value. For
0 < W < 1, the algorithm produces an undermigrated
FIG. 4.5-13. The combined results of migrations from Fig-
section, while for 1 < W < 2, it produces an over-
ures 4.5-11 and 4.5-12. B = dipping event before and A =
dipping event after desired migration; D = diffraction, and
migrated section. These observations are in agreement
L = percent lower velocities and H = percent higher veloc- with the impulse responses in Figure 4.5-16. The near-
ities. vertical streaks in the section with W = 1.95 represent
wraparound artifacts.
The generalized Stolt algorithm produces the best
result when W = 1, provided the medium velocity is
constant. Since this is never the case, we should examine
Stolt Stretch Factor the algorithm for a vertically varying velocity medium.
Figure 4.5-18 shows the impulse responses for different
As discussed in Section 4.1, the generalized Stolt values of W . Velocity varies linearly from t = 0 to t = 4 s
method of migration involves converting the time sec- between 2000 and 4000 m/s. For different W values, the
tion to an approximately constant-velocity section, portions of the wavefronts that best match the desired
which then is migrated by the constant-velocity Stolt migration using the phase-shift method are between the
algorithm. This conversion is essentially stretching in solid lines. For a vertically varying velocity medium,
the vertical (time) direction. Once the section is mi- W = 1 is no longer the desired factor. In Figure 4.5-18,
grated in the stretched domain, it is converted back to accuracy over the widest range of dip angles with the
the original time domain. The generalized Stolt method Stolt method is attained when W = 0.6. In general, the
must be distinguished from the constant-velocity algo- larger the velocity gradient, the farther the optimum W
rithm. The constant-velocity algorithm is accurate for is from 1. Strictly speaking, the optimum value for W
dips up to 90 degrees for a constant-velocity medium. is even different at different times.
The generalized method approximately accounts for ve- In practice, wavefront plots, like those in Figure
locity variations by stretching the section. 4.5-18, can be generated using both the phase-shift and
Stretching is defined by the stretch factor W . In his Stolt methods for a vertically varying regional velocity
original paper, Stolt (1978) discusses implementation of function. The W factor that yields the best fit at the
the W factor. Although W is a complicated function of largest angular aperture is used then to migrate the
velocity and stretch coordinate variables, it often is set data with the Stolt method.
Migration 573

FIG. 4.5-14. Tests for velocity errors in phase-shift migration using interval velocities derived from 100, 95, 90, and 80 percent
of rms velocities (from top to bottom). The input stacked section is shown in Figure 4.2-15a, and the desired migration using
the phase-shift method is shown in Figure 4.2-15b.
574 Seismic Data Analysis

FIG. 4.5-15. Tests for velocity errors in phase-shift migration using interval velocities derived from 100, 105, 110, and 120
percent of rms velocities (from top to bottom). The input stacked section is shown in Figure 4.2-15a, and the desired migration
using the phase-shift method is shown in Figure 4.2-15b.
Migration 575

domain suffers from wraparound effects both along the


time and space axes.
Figure 4.5-22 shows a zero-offset section that con-
tains a diffraction hyperbola and its migration using
frequency-wavenumber migration based on the phase-
shift method. When plotted with a very high display
gain, we observe the energy in the migrated section
bouncing off the edges of the section both in the time
and space directions.
A field data example of the wraparound effect on
frequency-wavenumber migration is shown in Figure
4.5-23. This is the same section as in Figure 4.2-15b,
except that it has been displayed using a very high gain.
The energy above the water bottom is associated with
FIG. 4.5-16. Tests for the stretch factor in Stolt migration: the wraparound effect in the time and space directions.
By varying the stretch factor W , the impulse response of the The wraparound noise actually exists within the im-
exact 90-degree migration operator (semicircle) is modified. age portion of the section, also. A way to reduce the
For comparison, the desired response has been superimposed wraparound effect is to pad the data with zeros along
on the Stolt migration impulse responses. the axis of Fourier transformation. For frequency-space
migration, data must be padded along the time axis;
To circumvent the difficulty of defining an optimum and for frequency-wavenumber migration, data must be
stretch factor W , Beasley and Lynn (1992) suggested padded along both the time and space axes.
applying the constant-velocity Stolt migration in a cas- A field data example of the wraparound effect on
caded manner. The idea is based on a clever represen- frequency-space migration is shown in Figure 4.5-24.
tation of a vertically varying velocity function by a set This is the same section as in Figure 4.4-11d, except
of constant velocities, which are then used to perform that it has been displayed using a very high gain. The
cascaded migration (Section 4.3). Since each migration energy above the water bottom is associated with the
stage is done by using a constant velocity, the stretch wraparound effect in the time direction.
factor W is by default set to 1. Of course, representa-
tion of a vertically varying velocity function by a set of
constant velocities is only an approximation that can Residual Migration
be valid for small vertical gradients.
Figures 4.5-19 and 4.5-20 show Stolt migrations of
The constant-velocity Stolt algorithm has useful ap-
the CMP stack in Figure 4.3-2a using different values
plications in residual migration as described here and
of the W factor. Migration velocities are varied only
migration velocity analysis as described in Chapter 5.
in the vertical direction. Figure 4.5-21 is a sketch of
Consider a zero-offset section that contains a set of dip-
the migration results for the diffraction D off the tip
ping events as shown in Figure 4.5-25a. The desired
of the salt diapir and steeply dipping event B off the
migration is obtained by using the medium velocity
flank of the salt diapir. The best match between the
of v = 3500 m/s as shown in Figure 4.5-25b. Sup-
desired migration and the Stolt method with stretch is
pose, instead, that you migrate using a velocity of 3000
for W = 0.5.
m/s. The resulting migrated section is shown in Fig-
ure 4.5-25c. Label the event with the steepest dip (AB)
from the desired migration on the migrated section with
Wraparound v1 = 3000 m/s, and note the undermigration of the dip-
ping events. By migrating the already migrated section
Wraparound is the effect of finite data length in time using a velocity of v2 = v 2 − v12 = 1802 m/s (Section
and space on a migration algorithm implemented in D.8), we get the section shown in Figure 4.5-25d. Note
the Fourier transform domain. A migration algorithm that this section obtained from the two-stage migration
implemented in the time-space domain does not suffer using velocities 3500 m/s and 1802 m/s is equivalent to
from wraparound effect. But a migration algorithm im- the one-stage migration using a velocity of 3500 m/s.
plemented in the frequency-space domain suffers from The second stage of the two-stage migration using a ve-
wraparound along the time axis. Similarly, a migration locity of 1802 m/s is called residual migration (Rothman
algorithm implemented in the frequency-wavenumber et al., 1985).
576 Seismic Data Analysis

FIG. 4.5-17. Tests for the stretch factor in Stolt migration: W < 1 causes undermigration, and W > 1 causes overmigration.
(Modeling courtesy Union Oil Company.)

So, what is the practical use of residual migration? 4.5-26e) using the appropriate residual velocity (Sec-
It can improve upon the result of migration using a dip- tion D.8) and the 15-degree finite-difference algorithm.
limited finite-difference algorithm. Figure 4.5-26a shows When compared with the single-stage finite-difference
a zero-offset section that consists of three point scatter- migration (Figure 4.5-26b), note the superior perfor-
ers in a layered earth model with vertically varying ve- mance of the residual migration. Also compare this
locity field. A 15-degree dip-limited finite-difference mi- with the desired migration using the phase-shift method
gration has difficulty collapsing these diffractions (Fig- (Figure 4.5-26c). The important point to keep in mind
ure 4.5-26b). Now, first migrate the zero-offset section is that input to residual migration (the second stage)
with the constant-velocity Stolt algorithm using the must be data which have been migrated (first stage)
lowest value, 2000 m/s, in the vertically varying ve- using a constant velocity (Rothman et al., 1985).
locity function. The result is shown in Figure 4.5-26d. A field data example is shown in Figure 4.5-27
Then, take this section and migrate it again (Figure with a sketch of the migration results in Figure 4.5-28.
Migration 577

FIG. 4.5-18. Tests for the stretch factor in Stolt migration: The medium velocity varies vertically from 2000 m/s at t = 0
to 4000 m/s at t = 4 s.

The single-stage 15-degree finite-difference result shows volves application of a dip-limited algorithm, such as an
the typical undermigrated character (Figure 4.3-3). The implicit finite-difference scheme, repeatedly. Whereas
1500-m/s constant-velocity Stolt migration followed by residual migration involves the application of a dip-
the finite-difference migration seems to produce an out- limited algorithm only once to data which already have
put that is reasonably close to the desired migration. been migrated using a constant-velocity Stolt migra-
A limitation of residual migration is that an ade- tion. In practice, Stolt migration can be replaced with
quate migration is not always achieved since the first- phase-shift migration and a vertically varying velocity
stage migration requires constant velocity which may be function with a gently varying gradient to accommodate
far off from the velocity field associated with the data. the constant-velocity requirement for the first-stage mi-
This is the case in Figure 4.5-27, since after residual mi- gration.
gration, the dipping event still is slightly undermigrated Whether it is residual or cascaded migration, the
(see the sketch in Figure 4.5-28). Undermigration occurs theoretical requirement is for constant velocity to be
because the apparent dip perceived by the second-stage used at each stage preceding the last stage. Departure
migration still may be too large to be handled accu- from this restriction will always limit the implementa-
rately. From equation (D-8c) note that the lower the tion of residual and cascaded migration. Figure 4.5-29
velocity used in migration, the smaller the dip that is shows a zero-offset section that contains a set of dipping
perceived by migration. If the residual velocity func- events with 3500-m/s constant velocity. First, migrate
tion given by equation (D-96b) is not too different from with a constant velocity of 2500 m/s; then, perform a
the original velocity function because of a large vertical cascade of four migrations with appropriate residual ve-
gradient, then residual migration may not be adequate. locities (Section D.8) to obtain the accurate image that
Residual migration is different from cascaded mi- is equivalent to the result from the desired migration us-
gration that is discussed in Section 4.3. The latter in- ing the phase-shift method applied only once with the
578 Seismic Data Analysis

FIG. 4.5-19. Tests for the stretch factor W in Stolt migration: (a) Desired migration, (b) W = 0.001, and (c) W = 0.5. The
input CMP stack is shown in Figure 4.3-2a.

FIG. 4.4.5-20. Tests for the stretch factor W in Stolt migration: (a) W = 0.7, (b) W = 1, and (c) W = 1.5. The input CMP
stack is shown in Figure 4.3-2a and the desired migration is shown in Figure 4.5-19a.
Migration 579

FIG. 4.5-21. The combined results of the migrations from


Figures 4.5-19 and 4.5-20. B = dipping event before and A
= dipping event after desired migration, D = diffraction be-
fore and D = diffraction after Stolt migration with stretch.
Numbers represent different stretch factors W .
FIG. 4.5-22. Wraparound effect in frequency-wavenumber
migration: (a) a zero-offset section that contains a diffrac-
medium velocity. When the same exercise is repeated tion hyperbola with 2500-m/s velocity, (b) desired migration
using a dip-limited implicit finite-difference algorithm, using the phase-shift method, and (c) the same as in (b) but
results are not satisfactory even with constant velocity with a display gain that is 100 times greater than that used
in (b).
(Figure 4.5-30). When a variable velocity is used at each
stage preceding the last stage the dip-limited algorithm
causes overmigration (Figure 4.3-19), while the phase-
shift method with no dip limit yields accurate result
(Figure 4.3-20). 4.6 FURTHER ASPECTS OF MIGRATION
Despite the limitations mentioned above, residual IN PRACTICE
migration is used in practice in the following mode:
In this section, we shall discuss the effects of spatial
aliasing, random noise and profile length on migration,
(a) Perform phase-shift migration using a vertically
and migration from topography. Spatial aliasing is a
varying velocity function v̄(z), which is obtained
direct result of undersampling of recorded data (Section
by averaging the spatially varying velocity field 1.2). Because of spatial aliasing, migration can perceive
v(x, z) and modifying it to meet the requirement events with steep dips at high frequencies as different
that v̄(z) < v(x, z). from the actual dips in ms/trace. As a result, migration
(b) Follow with a residual migration using a dip- mispositions the aliased frequency components of the
limited implicit or explicit frequency-space finite- dipping events.
difference migration with a residual velocity field Random noise usually is more prominent in the
equal to v 2 (x, z) − v̄ 2 (z) (Section D.8). deep part of a stacked section, just where velocities also
580 Seismic Data Analysis

FIG. 4.5-23. Wraparound effect in frequency-wavenumber migration: The input stacked section is shown in Figure 4.2-15a
and the desired migration using the phase-shift method is shown in Figure 4.2-15b.

FIG. 4.5-24. Wraparound effect in frequency-space migration: The input stacked section is shown in Figure 4.2-15a and the
desired migration using the phase-shift method is shown in Figure 4.2-15b.
Migration 581

reflection from that target appears on your unmigrated


stacked section.
Irregular topography associated with areas sub-
jected to overthurst tectonics has to be accounted for
during migration if surface elevation changes are rapid
along the line traverse. Migration algorithms, with the
exception of the Kirchhoff summation and the constant-
velocity Stolt method, are all based on wave extrapo-
lation from one flat depth level to another. A CMP-
stacked section is assumed to be equivalent to a zero-
offset wavefield and usually is referenced to a flat da-
tum. In the presence of severe topography, one needs
to account for the difference between the elevation pro-
file and the reference datum. Otherwise, if the reference
datum is above the surface elevation, to a migration al-
gorithm, events appear deeper than they are, and thus
are overmigrated. If, on the other hand, the reference
datum is below the surface elevation, events appear to
a migration algorithm shallower than they actually are,
and thus are undermigrated.

Migration and Spatial Aliasing

The concept of spatial aliasing is presented in Section


1.2. Here, we shall examine the effect of spatial aliasing
on migration. Figure 4.6-1 shows a zero-offset section
that contains a diffraction hyperbola with 2500-m/s ve-
locity and 12.5-m trace spacing. By discarding every
other trace, obtain another zero-offset section with 25-
m trace spacing. Repeat the procedure to obtain the
FIG. 4.5-25. Principle of residual migration: (a) a zero- zero-offset sections with 50-m and 100-m trace spacings
offset section that contains dipping events with 3500-m/s (Figure 4.6-1).
velocity, (b) desired migration using the phase-shift method The f − k spectra of the zero-offset sections with
with the medium velocity of 3500 m/s, (c) first-pass mi- the four different trace spacings are displayed in Fig-
gration of (a) using constant-velocity Stolt migration with ure 4.6-2. The diffraction hyperbola with 12.5-m trace
3000-m/s velocity, and (d) second-pass migration of the out- spacing maps onto an inverted triangular area in the
put from the first-pass migration as in (c) using a 15-degree f − k plane (Section 4.1). The Nyquist wavenumber is
finite-difference migration with the residual velocity of 1802 40 cycles/km and the bandwidth is given by the corner
m/s (equation D-96b). frequencies 6, 12 - 36, 48 Hz for the passband region
of the spectrum. (See Figure 1.1-26 for the definition of
are generally higher. This results in random noise or- corner frequencies.) The red is associated with the flat
ganized along wavefront arches, commonly referred to part of the passband region and the blue is associated
as smiles. This organized noise corrupts the migrated with the taper zone.
primary energy not just in the deep part of the sec- The f − k spectrum of the zero-offset section with
tion but also has detrimental effect on shallow data in 25-m trace spacing (Figure 4.6-1) indicates spatial alias-
a migrated section. ing beyond approximately 24 Hz (Figure 4.6-2). Conse-
Line length and location of the line traverse at the quently, the triangular shape of the passband region
surface relative to the location of the target in the sub- in the f − k plane that defines the diffraction hyper-
surface have a direct effect on the useability of the a mi- bola is corrupted around the edges near the Nyquist
grated section. Usually a line traverse longer than the wavenumber of 20 cycles/km. At a coarser trace spacing
spatial extent of the subusrface target is needed (Figure of 50 m, which corresponds to a Nyquist wavenumber
4.1-1). Keep in mind that your target does not neces- of 10 cycles/km, the triangular shape in the spectrum
sarily lie directly beneath the CMP location where the is preserved below the threshold frequency for aliasing,
582 Seismic Data Analysis

FIG. 4.5-26. Principle of residual migration: (a) a zero-offset section with vertically varying velocities, (b) 15-degree finite-
difference migration, (c) desired migration using the phase-shift method, (d) first-pass migration of (a) using constant-velocity
Stolt migration with 2000-m/s velocity, and (e) second-pass migration of the output from the first-pass migration as in (d)
using a 15-degree finite-difference migration with the residual velocity function computed by equation (D-96b).

approximately 12 Hz, only. Finally, at trace spacing of unaliased portion of the energy is of course mapped onto
100 m, which correpsonds to a Nyquist wavenumber of the apex. The more frequency components are spatially
5 cycles/km, the triangular shape is obliterated, com- aliased, the less energy at lower frequencies is mapped
pletely (Figure 4.6-2). onto the apex.
Figure 4.6-3 shows the results of Kirchhoff migra- As discussed in Section 4.1, the triangular area on
tion of the zero-offset sections in Figure 4.6-1. Fre- the f −k plane associated with the diffraction hyperbola
quency components that are spatially aliased are per- (Figure 4.6-2) is mapped onto the circular area on the
ceived by migration with a dip different from the actual
f −k plane associated with the migrated section (Figure
dip along the diffraction hyperbola. Normally, energy is
4.6-4). Ideally the area in the f −k plane of the migrated
moved in the up-dip direction along the diffraction hy-
section should be semicircular in shape. Because the
perbola and is mapped onto the apex. However, in each
migrated section, the spatially aliased part of the energy diffraction hyperbola is defined within a finite spatial
is split away from the flanks of the diffraction hyperbola aperture (Figure 4.6-1), there is an implicitly imposed
and mapped onto the regions to the left and right of the dip limit on migration. As a result, the semicircular area
flanks. The aliased energy is dispersed. Since each fre- is notched on either side (Figure 4.6-4).
quency component of the aliased energy is perceived to Spatial aliasing corrupts the semicircular shape of
have a different dip by migration, the displacement of the f − k spectrum on both ends of the spectrum at the
the energy after migration is frequency dependent. The vicinity of the Nyquist wavenumber. In case of severe
Migration 583

FIG. 4.5-27. Residual migration applied to field data: first pass using constant-velocity Stolt migration (1500 m/s), and
second pass using a 15-degree finite-difference migration of the result from the first pass.
584 Seismic Data Analysis

It is instructive to note that the diffraction energy


appears slightly undermigrated with 12.5-m trace spac-
ing, but is overmigrated with 25-m and 50-m trace spac-
ings. As discussed in Sections 4.3 and 4.4, the fidelity of
migration by finite-difference schemes is dictated by an
intricately complex interplay between the various pa-
rameters — spatial and temporal sampling intervals,
dip, frequency, and velocity. Depending on the values
of these parameters, one scheme may cause undermi-
gration in one case and overmigration in another case.
Figure 4.6-7 shows the f −k spectra of the migrated
sections in Figure 4.6-6. Note that implicit frequency-
space migration can create high-frequency noise beyond
the passband of the input data. Note also that spatial
aliasing combined with the inherent dispersive effect of
finite-difference schemes corrupt the semicircular shape
of the f − k spectrum on both ends of the spectrum.
Figure 4.6-8 shows the results of the migration of
the zero-offset sections in Figure 4.6-1 using an explicit
frequency-space finite-difference scheme (Section 4.4).
There appears to be no aliasing noise in either sections
with 12.5-m and 25-m trace spacings. Also note that,
compared to the results of Kirchhoff migration (Figure
4.6-2), there is less aliasing noise in the sections with 50-
m and 100-m trace spacings. These observations can be
verified by referring to the f − k spectra shown in Fig-
ure 4.6-9. Explicit schemes, by the design criterion, at-
FIG. 4.5-28. The combined migration results from Figure
tenuate energy associated with wavenumbers kx above
4.5-27. B = dipping event before and A = dipping event
after desired migration; D = diffraction; A1, D1 = after the a specified cutoff wavenumber defined by a fraction of
first pass; A2, D2 = after the second pass. the Nyquist wavenumber. This effectively removes part
of the aliased energy that maps onto the spectral re-
gion above the cutoff wavenumber associated with the
undersampling, spatially aliased frequency components
extrapolation filter for the explicit scheme. Note from
invade much of the f −k plane as shown in Figure 4.6-4.
the f −k spectrum in Figure 4.6-9 that almost all of the
Aside from the spatial aliasing noise, dispersive
aliased energy has been filtered out for the case of the
noise also is seen on data migrated with a dip-limited 25-m trace spacing. This is why aliasing noise is absent
finite-difference algorithm (Section 4.3). Figure 4.6-5 in the corresponding migrated section in Figure 4.6-8.
shows migration of a zero-offset section that contains Despite the wavenumber filtering effect of the explicit
a diffraction hyperbola using the 15-degree implicit scheme, however, much of the aliased noise remains in
finite-difference method. Note the undermigration of the sections with the 50-m and 100-m trace spacings.
the diffraction hyperbola that is caused by the 15-degree Figure 4.6-10 shows the results of phase-shift mi-
dip limitation, the dispersive noise A that is caused by gration of the zero-offset sections in Figure 4.6-1, and
the finite-difference approximations, and the spatially Figure 4.6-11 shows the corresponding f − k spectra.
aliased energy B that splits away from the unaliased These results are used as a benchmark to evaluate the
part that collapses to the apex. results obtained from the other migration algorithms
Figure 4.6-6 shows the results of migration of the (Figures 4.6-3 through 4.6-9). Except for the aliasing
zero-offset sections in Figure 4.6-1 using an implicit noise, phase-shift migration produces no artifacts.
frequency-space finite-difference scheme (Section 4.4). The experiments described above clearly demon-
Note the dispersive noise caused by the finite-difference strate that all migration algorithms suffer from saptial
approximations in the section with 12.5-m trace spac- aliasing. We now examine the effect of spatial alias-
ing. The dispersive noise in the sections with 25-m, ing on migration using a dipping events model. Figure
50-m, and 100-m trace spacings, however, is attributed 4.6-12 shows a zero-offset section that contains a set of
largely to spatial aliasing. dipping events with 3500-m/s velocity and 25-m trace
(text continues on p. 609)
Migration 585

FIG. 4.5-29. Multiple passes of residual migration: (a) a zero-offset section that contains dipping events with 3500-m/s
velocity, (b) desired migration using the phase-shift method with the medium velocity of 3500 m/s, (c) first-pass migration
of the zero-offset section in (a) using the phase-shift method with a constant velocity of 2500 m/s, (d) second-pass migration
of the output from the first-pass migration as in (c) using the phase-shift method with a constant residual velocity of 1145
m/s, (e) third-pass migration of the output from the second-pass migration as in (d) using the phase-shift method with a
constant residual velocity of 1198 m/s, (f) fourth-pass migration of the output from the third-pass migration as in (e) using
the phase-shift method with a constant residual velocity of 1250 m/s, and (g) fifth-pass migration of the output from the
fourth-pass migration as in (f) using the phase-shift method with a constant residual velocity of 1299 m/s.
586 Seismic Data Analysis

FIG. 4.5-30. Multiple passes of residual migration: (a) a zero-offset section that contains dipping events with 3500-m/s
velocity, (b) desired migration using the phase-shift method with the medium velocity of 3500 m/s, (c) first-pass migration
of the zero-offset section in (a) using the 65-degree implicit method with a constant velocity of 2500 m/s, (d) second-pass
migration of the output from the first-pass migration as in (c) using the 65-degree implicit method with a constant residual
velocity of 1145 m/s, (e) third-pass migration of the output from the second-pass migration as in (d) using the 65-degree
implicit method with a constant residual velocity of 1198 m/s, (f) fourth-pass migration of the output from the third-pass
migration as in (e) using the 65-degree implicit method with a constant residual velocity of 1250 m/s, and (g) fifth-pass
migration of the output from the fourth-pass migration as in (f) using the 65-degree implicit method with a constant residual
velocity of 1299 m/s.
Migration 587

FIG. 4.6-1. Zero-offset sections, which contain a diffraction hyperbola with 2500-m/s velocity, with trace spacings, from top
to bottom, 12.5 m, 25 m, 50 m, and 100 m.
588 Seismic Data Analysis

FIG. 4.6-2. The f − k spectra of the zero-offset sections in Figure 4.6-1 with trace spacings, from top to bottom, 12.5 m,
25 m, 50 m, and 100 m.
Migration 589

FIG. 4.6-3. Kirchhoff migrations of the zero-offset sections in Figure 4.6-1 with trace spacings, from top to bottom, 12.5 m,
25 m, 50 m, and 100 m.
590 Seismic Data Analysis

FIG. 4.6-4. The f − k spectra of the outputs from the Kirchhoff migrations in Figure 4.6-3 with trace spacings, from top to
bottom, 12.5 m, 25 m, 50 m, and 100 m.
Migration 591

FIG. 4.6-5. 15-degree finite-difference migrations of the zero-offset sections in Figure 4.6-1 with trace spacings, from top to
bottom, 12.5 m, 25 m, 50 m, and 100 m.
592 Seismic Data Analysis

FIG. 4.6-6. 65-degree implicit frequency-space migrations of the zero-offset sections in Figure 4.6-1 with trace spacings, from
top to bottom, 12.5 m, 25 m, 50 m, and 100 m.
Migration 593

FIG. 4.6-7. The f − k spectra of the outputs from the 65-degree implicit frequency-space migrations in Figure 4.6-6 with
trace spacings, from top to bottom, 12.5 m, 25 m, 50 m, and 100 m.
594 Seismic Data Analysis

FIG. 4.6-8. 70-degree explicit frequency-space migrations of the zero-offset sections in Figure 4.6-1 with trace spacings, from
top to bottom, 12.5 m, 25 m, 50 m, and 100 m.
Migration 595

FIG. 4.6-9. The f − k spectra of the outputs from the steep-dip explicit frequency-space migrations in Figure 4.6-8 with
trace spacings, from top to bottom, 12.5 m, 25 m, 50 m, and 100 m.
596 Seismic Data Analysis

FIG. 4.6-10. Phase-shift migrations of the zero-offset sections in Figure 4.6-1 with trace spacings, from top to bottom,
12.5 m, 25 m, 50 m, and 100 m.
Migration 597

Hz
60

40

20

-40 0 40
cycles/km
Hz
60

40

20

-20 0 20
Hz
60

40

20

-10 0 10
Hz
60

40

20

-5 0 5

FIG. 4.6-11. The f − k spectra of the outputs from the phase-shift migrations in Figure 4.6-10 with trace spacings, from
top to bottom, 12.5 m, 25 m, 50 m, and 100 m.
598 Seismic Data Analysis

FIG. 4.6-12. Zero-offset sections, which contain dipping events with 3500-m/s velocity, with trace spacings, from top to
bottom, 25 m, 50 m, and 100 m.
Migration 599

FIG. 4.6-13. The f − k spectra of the zero-offset sections in Figure 4.6-12 with trace spacings, from top to bottom, 25 m,
50 m, and 100 m.
600 Seismic Data Analysis

FIG. 4.6-14. Kirchhoff migrations of the zero-offset sections in Figure 4.6-12 with trace spacings, from top to bottom, 25 m,
50 m, and 100 m.
Migration 601

FIG. 4.6-15. The f − k spectra of the outputs from the Kirchhoff migrations in Figure 4.6-14 with trace spacings, from top
to bottom, 25 m, 50 m, and 100 m.
602 Seismic Data Analysis

FIG. 4.6-16. 15-degree finite-difference migrations of the zero-offset sections in Figure 4.6-12 with trace spacings, from top
to bottom, 25 m, 50 m, and 100 m.
Migration 603

FIG. 4.6-17. 65-degree implicit frequency-space migrations of the zero-offset sections in Figure 4.6-12 with trace spacings,
from top to bottom, 25 m, 50 m, and 100 m.
604 Seismic Data Analysis

FIG. 4.6-18. The f − k spectra of the outputs from the 65-degree implicit frequency-space migrations in Figure 4.6-17 with
trace spacings, from top to bottom, 25 m, 50 m, and 100 m.
Migration 605

FIG. 4.6-19. 70-degree explicit frequency-space migrations of the zero-offset sections in Figure 4.6-12 with trace spacings,
from top to bottom, 25 m, 50 m, and 100 m.
606 Seismic Data Analysis

FIG. 4.6-20. The f − k spectra of the outputs from the steep-dip explicit frequency-space migrations in Figure 4.6-19 with
trace spacings, from top to bottom, 25 m, 50 m, and 100 m.
Migration 607

FIG. 4.6-21. Phase-shift migrations of the zero-offset sections in Figure 4.6-12 with trace spacings, from top to bottom,
25 m, 50 m, and 100 m.
608 Seismic Data Analysis

FIG. 4.6-22. The f − k spectra of the outputs from the phase-shift migrations in Figure 4.6-21 with trace spacings, from
top to bottom, 25 m, 50 m, and 100 m.
Migration 609

spacing. By discarding every other trace, obtain another been aliased twice (Figure 4.6-3). Because of the com-
zero-offset section with 50-m trace spacing. Repeat the plexity of aliasing, the noise essentially disperses over
procedure to obtain the zero-offset section with 100-m the whole of the section.
trace spacing (Figure 4.6-12). An interesting observation on the migrated sections
The f − k spectra of the zero-offset sections with in Figure 4.6-14 relates to the energy in the region above
the three different trace spacings are displayed in Figure 1 s. In Kirchhoff migration, amplitudes in the input sec-
4.6-13. The dipping events with 25-m trace spacing map tion are summed along a hyperbolic summation path
onto a series of radial lines in the f − k plane (Section and placed at the apex of the hyperbola. Imagine a sum-
4.1). The Nyquist wavenumber is 20 cycles/km and the mation path whose apex is situated at a time less than 1
bandwidth is given by the corner frequencies 6, 12 - 36, s. There will be some energy placed at this apex location
48 Hz for the passband region of the spectrum. As for since the flanks of the summation path under consider-
the diffraction hyperbola model (Figure 4.6-3), the red ation will intersect through traces with nonzero sample
is associated with the flat part of the passband region values. This situation is encountered when migrating
and the blue is associated with the taper zone. marine data using Kirchhoff summation. Normally, the
The f − k spectrum of the zero-offset section with migrated section is muted above the water bottom to
50-m trace spacing (Figure 4.6-12), which corresponds remove the noise created by migration within the water
to a Nyquist wavenumber of 10 cycles/km, indicates layer.
spatial aliasing beyond approximately 24 Hz (Figure Besides data aliasing, there is also the problem of
4.6-13). Consequently, the aliased segments of the radial operator aliasing. In particular, for a low-velocity hy-
lines map onto the left quadrant of the f − k spectrum. perbola or for a hyperbola with its apex situated at
At a coarser trace spacing of 100 m, which corresponds shallow times, Kirchhoff summation may require more
to a Nyquist wavenumber of 5 cycles/km, spatial alias- than one sample per trace. This results in some energy
ing occurs first at approximately 12 Hz. Then, some in the form of precursors above the migrated sea-bottom
reflection, when only one point per trace is included in
of the energy already aliased becomes aliased for the
the summation.
second time at approximately 36 Hz. Moreover, part of
Figure 4.6-15 shows the f − k spectra of the mi-
the aliased energy is remapped onto the right quadrant
grated sections in Figure 4.6-14. As discussed in Section
(Figure 4.6-13).
4.1, migration rotates the radial lines on the f − k plane
Figure 4.6-14 shows the results of Kirchhoff mi-
associated with the dipping events.
gration of the zero-offset sections in Figure 4.6-12. No
Figure 4.6-16 shows migration of a zero-offset sec-
aliasing noise is present on the migrated section with
tion that contains a set of dipping events using the 15-
25-m trace spacing. Next, consider the migrated sec-
degree implicit finite-difference method. Note the un-
tion with 50-m trace spacing. Frequency components dermigration of the steeply dipping events caused by the
that are spatially aliased are perceived by migration to 15-degree dip limitation, the dispersive noise A caused
dip in the direction opposite to the actual dips of the by the finite-difference approximations, and the spa-
events. Normally, energy is moved in the up-dip direc- tially aliased energy B that splits away from the un-
tion, in this case from right to left as seen in the mi- aliased part and moves in the opposite direction.
grated section with 25-m trace spacing. However, in the Figure 4.6-17 shows the results of migration of the
migrated section with 50-m trace spacing, the spatially zero-offset sections in Figure 4.6-12 using an implicit
aliased part of the energy is split away from the dip- frequency-space finite-difference scheme (Section 4.4).
ping events and moved from left to right. Note that the Note the dispersive noise caused by the finite-difference
aliased energy is dispersed. As for the diffraction hyper- approximations in the section with 25-m trace spacing.
bola model (Figure 4.6-2), each frequency component of In the sections with 50-m and 100-m trace spacings, two
the aliased energy is perceived to have a different dip sets of dispersive noise can be distinguished — one that
by migration, the displacement of the energy after mi- is caused by the finite-difference approximations and
gration is frequency dependent. The unaliased portion the other caused by spatial aliasing. At coarse spatial
of the energy is of course moved from right to left and sampling, the steeply dipping events are faintly detected
positioned accurately. The more frequency components on the migrated section.
are spatially aliased, the less energy at lower frequencies Figure 4.6-18 shows the f − k spectra of the mi-
is mapped to the correct position. grated sections in Figure 4.6-17. As in the case of the
Finally, consider the case of the migrated section diffraction hyperbola (Figure 4.6-7), implicit frequency-
with 100-m trace spacing (Figure 4.6-14). Note that space migration can create high-frequency noise beyond
there exists aliasing noise not only to the right of the the passband of the input data.
dipping events but also in the left-most portion of the Figure 4.6-19 shows the results of the migration of
section. The latter is associated with the energy that has the zero-offset sections in Figure 4.6-12 using an explicit
610 Seismic Data Analysis

frequency-space finite-difference scheme (Section 4.4). the section with 50-m trace spacing, and the section
There is no aliasing noise in the section with 25-m trace with 100-m trace spacing does not even provide an im-
spacing. But there is precursive dispersion along the age of the gently dipping reflections. This is because of
steeply dipping events because of the inherent nature the aliasing noise associated with the steep flanks of the
of the explicit scheme used here. This dispersion is not salt diapir corrupting the surrounding reflections. Spa-
as severe as that observed on the result from the implicit tial aliasing not only adversely affects the quality of the
scheme (Figure 4.6-17). image associated with a dipping event that is aliased,
The corresponding f − k spectra shown in Figure but it also can obliterate other nonaliased events in the
4.6-20 explains why there is less aliasing noise on the data.
migrated sections in Figure 4.6-19 compared to those Figure 4.6-28 through 4.6-30 show migrations of
from Kirchhoff summation (Figure 4.6-14). As for the the stacked sections in Figures 4.6-25 and 4.6-26 using
case of the diffraction hyperbola (Figure 4.6-9), explicit steep-dip frequency-space implicit and explicit schemes,
schemes attenuate energy associated with wavenumbers and the phase-shift method. Similar conclusions are
kx above a specified cutoff wavenumber. This effectively drawn for the Kirchhoff summation results shown in
removes part of the aliased energy that maps onto the Figure 4.6-27. Differences in terms of delineation of
spectral region above the cutoff wavenumber associated the salt boundary and the surrounding strata are at-
with the extrapolation filter for the explicit scheme. tributable to the manner in which these algorithms are
Note from the f − k spectrum in Figure 4.6-20 that implemented and how they treat the velocity field for
a significant portion of the aliased energy in the left migration. For instance, when examining the results
quadrant has been filtered out for the case of the 50-m from the frequency-space implicit scheme (Figure 4.6-
trace spacing. Despite the wavenumber filtering effect of 28), one must keep in mind the effect of spatial aliasing
the explicit scheme, however, much of the aliased noise combined with the effect of undermigration caused by
remains in the section with the 100-m trace spacing.
the dip-limited nature of the algorithm and the effect of
Figure 4.6-21 shows the results of the phase-shift
dispersion caused by finite-difference approximations.
migration of the zero-offset sections in Figure 4.6-12,
What is the remedy for spatial aliasing noise in
and Figure 4.6-22 shows the corresponding f −k spectra.
migration? Arrange the sequence of the sections in Fig-
These results are used as a benchmark to evaluate the
ures 4.6-27 through 4.6-30 from a coarser to a finer trace
results obtained from the other migration algorithms
spacing. Note that the deleterious effect of spatial alias-
(Figures 4.6-14 through 4.6-20). Except for the aliasing
ing in migration disappears as we go to finer trace spac-
noise, phase-shift migration produces no artifacts.
ings. To avoid spatial aliasing, we must record with suf-
The effect of spatial aliasing on migration of field
ficiently fine CMP trace interval or interpolate the data
data, to begin with, is demonstrated in Figures 4.6-23
and 4.6-24. We see the original stacked section and its that have been recorded with coarse spatial sampling.
resampled versions at coarser trace spacings. From the Most modern surveys are conducted using spatial
migrations of these four stacked sections with coarser sampling rates that are perfectly adequate to meet ex-
trace spacings, note the loss of spatial resolution. The ploration and development objectives. If we are dealing
nearly flat events are not adversely affected by spatial with vintage data with coarse spatial sampling, there
aliasing, while the steeply dipping reflection off the right are two ways to circumvent the effect of spatial alias-
flank of the salt diapir can only be detected on the ing. The first approach would be to filter out the aliased
migrated section with very coarse sampling by a low- frequencies. This is undesirable, since it severely limits
frequency, weak-amplitude event. The diffraction en- vertical and lateral resolutions (Section 11.1). The sec-
ergy off the tip of the salt diapir is largely dispersed ond approach would be to do trace interpolation before
into the region with nearly flat events to the right of migration. In Sections 7.2 and G.5, we discuss interpo-
the diapir. lation of aliased data. Note from section 1.3 that the
We now examine the response of the various migra- smaller the trace interval, the higher the Nyquist in
tion algorithms to spatial alaising using the data shown the spatial wavenumber direction (Figures 1.3-10 and
in Figures 4.6-25 and 4.6-26. The stacked data are asso- 1.3-11), and thus, the less likelihood of aliasing high-
ciated with the same line sampled at four different trace frequency data.
spacings — 12.5, 25, 50, and 100 m. Figure 4.6-27 shows A schematic illustration of the spatial aliasing phe-
migrations of the stacked sections using Kirchhoff sum- nomenon is shown in Figure 4.6-31. Start with the
mation. The section with 12.5-m trace spacing provides spectral bandwidth that spans COA in the spatial
a crisp image of the salt diapir, while the sections with wavenumber axis, where A is the location of the Nyquist
coarser trace spacings degrade gradually. Specifically, it wavenumber, and ON in the temporal frequency axis,
is almost impossible to delineate the salt boundary on where N is the location of the Nyquist frequency. Dip
(text continues on p. 619)
Migration 611

FIG. 4.6-23. A portion of a CMP-stacked section spatially sampled at four different trace spacings.
612 Seismic Data Analysis

FIG. 4.6-24. Phase-shift migrations of the stacked sections in Figure 4.6-23 sampled at four different trace spacings.
Migration 613

FIG. 4.6-25. A CMP-stacked section spatially sampled at 12.5-m (top) and 25-m (bottom) trace spacings.
614 Seismic Data Analysis

FIG. 4.6-26. A CMP-stacked section as in Figure 4.6-25 spatially sampled at 50-m (top) and 100-m (bottom) trace spacings.
Migration 615

FIG. 4.6-27. Kirchhoff migrations of the stacked sections in Figure 4.6-25 and 4.6-26 with trace spacings, from top to bottom,
12.5 m, 25 m, 50 m, and 100 m.
616 Seismic Data Analysis

FIG. 4.6-28. 65-degree implicit frequency-space migrations of the stacked sections in Figure 4.6-25 and 4.6-26 with trace
spacings, from top to bottom, 12.5 m, 25 m, 50 m, and 100 m.
Migration 617

FIG. 4.6-29. 70-degree explicit frequency-space migrations of the stacked sections in Figure 4.6-25 and 4.6-26 with trace
spacings, from top to bottom, 12.5 m, 25 m, 50 m, and 100 m.
618 Seismic Data Analysis

FIG. 4.6-30. Phase-shift migrations of the stacked sections in Figure 4.6-25 and 4.6-26 with trace spacings, from top to
bottom, 12.5 m, 25 m, 50 m, and 100 m.
Migration 619

FIG. 4.6-31. Two dipping events in the f − k domain. See


text for details.

components 1 and 2 are aliased beyond frequency val-


ues AT and AS, respectively. Extend the wavenumber
axis to DOB by making the trace interval half of the
original. Event 1 no longer is spatially aliased within
the frequency bandwidth ON. Event 2 still is aliased
beyond the frequency value BV. However, at this point
FIG. 4.6-32. Response of migration to random noise: (a)
and beyond, there may be no significant energy, so fur- zero-offset section with random noise, only, (b) frequency-
ther extension of the wavenumber axis may not be nec- wavenumber migration.
essary. Another important point is that if the temporal
frequency band only extended up to OG to start with,
extension of the wavenumber axis to DOB also would
result in Event 2 being unaliased. Thus, the amount
of trace interpolation that is required also depends on
temporal bandwidth as well as on structural dip.
Trace interpolation often is necessary when dealing
with 3-D data and old data recorded with a large group
interval. In a typical 3-D survey, the inline trace interval
may be as little as 12.5 m, while the trace interval in the
crossline direction, for some old data, may be as much
as 100 m. Therefore, interpolation is required before mi-
gration in the crossline direction. You do not necessarily
interpolate down to the inline trace spacing; instead, de-
pending on the maximum structural dip and velocity in
the area, the optimum trace spacing for interpolation in
the crossline direction can be computed using equation
(1-8). Section 7.2 provides more information on trace
interpolation in relation to 3-D migration.

Migration and Random Noise

Figure 4.6-32 shows a section that contains band-


limited random noise uncorrelated from trace to trace
and its migration using the phase-shift method. Veloc-
ity increases linearly from 2000 m/s at the top to 4000 FIG. 4.6-33. (a) A deeper portion of a CMP-stacked sec-
m/s at the bottom of the section. The amplitude and tion with significant noise level, (b) the same portion after
frequency characteristics of the input section in Figure migration.
620 Seismic Data Analysis

FIG. 4.6-34. Migrations of a portion of a CMP stack with different lateral extents.

FIG. 4.6-35. Edge effects in migration: (a) a portion of a CMP stack, (b) the same portion after migration, and (c) a sketch
that illustrates the edge effect problem using the semicircle superposition method of migration. See text for A, B, C, and D.
Migration 621

FIG. 4.6-36. A CMP stack (a) before and (b) after migration.

4.6-32 are virtually unchanged in the interior portion organizes random noise, it does not attenuate it. A dip-
of the migrated section. However, note the smearing limited migration algorithm acts upon the random noise
of amplitudes at the bottom and side boundaries after like a dip filter and removes the noise energy beyond
migration. the dip limit much like shown in Figure 6.2-2. The dip-
Ambient noise commonly dominates the deep por- limited algorithm also attenuates unaliased linear noise
tion of a stacked section where velocities are high. with a dip steeper than the dip limit.
Therefore, organization of random noise caused by mi-
gration generally is more severe in the deeper part of a
stacked section. A field data example is shown in Fig- Migration and Line Length
ure 4.6-33. In addition to smearing effects, the migrated
section also has smiles, which are caused by sparsely dis- For one reason or another, a seismic line may have to
tributed bursts of amplitude in the input section. Keep be recorded in the field with a shorter length than de-
in mind that a single spike on the time section migrates sired. To see the effect of line length on migration, we
to a semicircle on the depth section. will examine the migrations of the decreasing lengths of
We already have seen the adverse effect of an im- the same CMP stack (Figure 4.6-34). Migration of the
proper choice of aperture width in Kirchhoff summation smaller portions, BD and CD, results in an increasingly
(Figure 4.2-7). A narrow aperture can introduce strong smeared section, particularly in the deeper parts. We
smearing as spurious, nearly horizontal events. A similar conclude that short seismic lines really are not suited
effect occurs for all types of migration algorithms if the for migration.
maximum dip to migrate is severely restricted (Figure If the line traverse is too short, two effects occur.
4.5-4). It is a misconception to imagine that migration First, there is not enough space in the section for dip-
attenuates random noise and improves signal-to-noise ping events to move during migration. This problem
ratio. Instead, one must keep in mind that migration may be alleviated by padding the stacked section with
622 Seismic Data Analysis
Migration 623
624 Seismic Data Analysis

FIG. 4.6-39. A sketch of the events after migration from the flat reference datum level as in Figure 4.6-37 denoted by the
dotted segments and migration from the floating datum as in Figure 4.6-38 denoted by the solid segments. The area covered
corresponds to the upper central portions of the sections in Figures 4.6-37 and 4.6-38 between midpoints A and B.

zero traces on both sides before migration. Second, side Figure 4.6-35 shows a section with significant
boundary effects contaminate a significant portion of smearing caused by side boundary effects. The wave-
the migrated section. The real solution to circumvent front character that dominates the left boundary of the
the boundary effects is to record data with sufficient migrated section down to the bottom of the mute zone
line length. can be explained using the principle of semicircle super-
With a general idea of structural dip in an area, position for migration. Consider a dipping event A that
the geophysicist must consider the additional spatial extends down to the edge of the section as in the sketch
extent that is required by migration. (Refer to the dis- in Figure 4.6-35. After migration B, note the remainder
cussion on Figure 4.1-1.) This is especially important in of the semicircular wavefront C on the left side. This
3-D surveys in which the surface areal coverage must be wavefront did not cancel out during superposition be-
extended beyond the subsurface areal coverage so that cause no data were available beyond the left boundary
steep dips and structural discontinuities can be recorded of the section.
and imaged properly (Section 7.1). The problem with Another source of edge effects is the presence of am-
3-D is that cost increases as the square of the survey plitude bursts at or near the edge of the stacked section
dimension, so that temptation to record too small a associated with a low signal-to-noise ratio that results
survey is great. from low fold. The edge effects on the left boundary
Regardless of line lengths, there are additional below the mute zone in the migrated section (Figure
problems associated with the side boundaries of the 4.6-35) probably stem from the lack of amplitude bal-
stacked section input to migration. All migration al- ance on the CMP stacked section. The latter is caused
gorithms implicitly make some assumption about the by changes in fold at the end of the line.
nature of data outside the side boundaries of the input Figure 4.6-36 shows a CMP stack with an imbri-
stacked section. The simple assumptions, zero ampli- cate structure associated with overthrust tectonics. Af-
tude or zero gradient at the side boundaries of the sec- ter migration, note that there are two zones with no
tion, cause data that should migrate past the edge to be reflections. The zone to the left of CMP 100 resulted
reflected back into the section. To prevent this, traces of from finite line length. Specifically, the events on the
zero amplitude often are appended to the edges of the left flank of the imbricate structure are migrated to the
input section. This allows the dipping events to move right in the up-dip direction, thus leaving behind a zone
freely into the zero-amplitude region during migration. of no events into which no energy is moved since the line
If the events that would migrate off the input section are ends to the left of the structure. The zone of no events
not needed, they often are suppressed by using absorb- between CMP 200 and 300 is a direct consequence of
ing side boundary conditions (Clayton and Engquist, the overthrusting that has given rise to a culminating
1980). structure with very steep, almost overturned events. Mi-
Migration 625

FIG. 4.6-40. Principles of migration from topography. See text for details.
626 Seismic Data Analysis

gration of such steep dips is possible only if they exist to the flat datum. To circumvent this tedious task, mi-
in the recorded data and are imaged using algorithms gration from an irregular topography is done either by
which handle dips beyond 90 degrees. the zero-velocity trick (Beasley and Lynn, 1992) or the
zero-wavefield trick (Reshef, 1991).
In the first approach by Beasley and Lynn (1992),
Migration from Topography a zero velocity value is assigned to the region between
the floating datum and the flat datum, which is specified
A CMP-stacked section is assumed to be equivalent to above the floating datum. Just as it should be in con-
a zero-offset wavefield and usually is referenced to a flat ventional processing, the stacked section is referenced to
datum. Figure 4.6-37 shows a migrated CMP stacked the flat datum, and the velocities are referenced to the
section associated with a seismic line that follows a tra- floating datum. Extrapolate the stacked section down
verse with severe topography from an area with over- one depth step using the zero velocity as part of the
thrust tectonics. Migration was done from a flat refer- migration process. This amounts to a simple vertical
ence datum. When migrating data recorded over such time shift. If the depth level intersects the floating da-
an irregular and severe topography, however, one needs tum profile, then invoke the diffraction term (equation
to account for the difference between the elevation pro- 4-16a) for the traces in the stacked section that coin-
file and the reference datum. Otherwise, events appear cide with the intersection points. Continue the extrap-
to a migration algorithm shallower than they actually olation process from one depth level to the next while
are if the flat reference datum is below the elevation turning on the diffraction term for those traces which
profile, and thus are undermigrated. If the flat reference coincide with the intersection points of the depth levels
datum is above the elevation profile, then events appear and floating datum profile.
to a migration algorithm deeper than they actually are, In the second approach by Reshef (1991), to start
and thus are overmigrated. with, a zero wavefield is assigned to the flat datum level
Figure 4.6-38 shows the same data as in Figure z0 in Figure 4.6-40. The stacked section and the veloc-
4.6-37 with migration from the floating datum which ities are referenced to the floating datum. Extrapolate
is a smooth form the elevation profile. A sketch of key this wavefield down one depth level to z1 using a ve-
events between midpoints A and B from both sections locity the same as that just below the floating datum
is provided in Figure 4.6-39. The flat reference datum in profile. Import the traces 1 and 2 from the stacked sec-
this case is above the elevation profile. Hence, migration tion at the intersection points of the depth level and the
from the flat datum causes overmigration as denoted by floating datum profile, and insert them to the section
the dotted interpretation segments. By migrating the referenced to the depth level z1 . Now, extrapolate down
data from the floating datum and interpreting the re- to the next depth level z2 , and import traces 3, 4, 5, and
sulting section, we obtain the solid segments in Figure 6 from the stacked section, and insert them into section
4.6-39. referenced to the depth level z2 . At each depth level,
Migration algorithms, with the exception of Kirch- while new traces are imported from the original stacked
hoff summation and the constant-velocity Stolt method, section, the previously imported traces are subsjected
are all based on wave extrapolation from one flat depth to wave extrapolation as symbolized by the small arcs
level to another. To accommodate an irregular topog-
associated with each time sample.
raphy, the following formal approach can be used:

(a) Stack the data referenced to the floating datum and


assume it to be the zero-offset wavefield recorded EXERCISES
along the floating datum profile.
(b) Apply wave-equation datuming (Section 8.1) to ex- Exercise 4-1. Consider the special case of a 90-
trapolate the zero-offset wavefield as defined in (a) degree dipping reflector in Figure 4.1-1a. Sketch the cor-
from the floating datum to a flat datum above using responding zero-offset time section.
a velocity the same as that just below the floating Exercise 4-2. Consider Huygens’ secondary
datum. sources along a dipping reflector. Sketch the zero-offset
(c) Migrate the output wavefield from (b) using a pre- section by superimposing the individual responses from
ferred migration algorithm. these sources. Remember, to do zero-offset modeling,
you must map a point in the x − z plane to a hyperbola
The stacking velocity field required in step (a) is in the x − t plane with its apex as the input point.
referenced to the floating datum. The migration ve- Exercise 4-3. For which case is spatial aliasing a
locity field required in step (c) would need to be de- more serious problem, the low-velocity or high-velocity
rived by redefining the stacking velocities with respect medium?
Migration 627

FIG. 4.E-1. See Exercise 4-4.

Exercise 4-4. Locate the dipping event AA on Exercise 4-7. Suppose you specified the wrong
the migrated section in Figure 4.E-1. trace spacing in your migration. What effect does it
Exercise 4-5. A point in the x−t plane is mapped have, overmigration or undermigration? Assume that
onto a semicircle in the x − z plane. Where does it map you supplied the wrong sampling rate in time. What
on the x − τ plane, where τ = 2z/ v. effect does it have on migration output?
Exercise 4-8. Suppose you want to do zero-offset
Exercise 4-6. Refer to Figure 4.1-14. It suggests
recording of the steep flank of a salt dome. Which case
that if the subsurface consisted of a semicircular reflec- would require a longer line length when the medium ve-
tor (b), then the zero-offset response would be as in locity along the raypath is (a) constant or (b) vertically
(a). What would the subsurface be like if you obtained increasing?
(a) using a source-receiver pair with a finite separation Exercise 4-9. How would Figure 4.3-6 look if you
between them? (See Figures D-5 and D-6.) used a 15-degree phase-shift migration algorithm?
Appendix D
MATHEMATICAL FOUNDATION OF MIGRATION

D.1 Wavefield Extrapolation and Migration

A fundamental equation of reflection seismology is the double-square-root (DSR) equation. This


equation describes downward continuation of both shots and receivers into the earth. It is exact
for all dips and offsets. Neglecting the velocity gradient dv(z)/dz makes the DSR equation also
applicable to a stratified earth. The DSR equation can be extended, with some approximation,
to treat weak lateral velocity variations. A comprehensive mathematical treatise of the DSR
equation is found in Claerbout (1985).
The basic 2-D theory for wavefield extrapolation is presented here. Then, using the DSR
equation, a rigorous analysis of conventional seismic data processing is made. We show that
conventional implementation of the DSR equation requires zero-dip and zero-offset assumptions.
Before discussing the DSR equation, we review the basic theory for wavefield extrapolation.
Once the extrapolation equations are developed, they can be used with the imaging principle to
migrate 2-D or 3-D prestack and poststack data.
Start with the 2-D scalar wave equation, which describes propagation of a compressional
wavefield P (x, z, t) in a medium with constant material density and compressional wave velocity
v(x, z):

∂2 ∂2 1 ∂2
2
+ 2− 2 2 P (x, z, t) = 0, (D − 1)
∂x ∂z v ∂t
where x is the horizontal spatial axis, z is the depth axis (positive downward), and t is time.
Given the upcoming seismic wavefield P (x, 0, t), which is recorded at the surface, we want to
determine reflectivity P (x, z, 0). This requires extrapolating the surface wavefield to depth z,
then collecting it at t = 0. The process of obtaining the earth’s reflectivity P (x, z, t = 0) from
the observed wavefield P (x, z = 0, t) at the surface z = 0 is called migration, and the reverse
process is called modeling (Figure D-1).
It is advantageous to decompose the wavefield into monochromatic plane waves with dif-
ferent angles of propagation from the vertical. Therefore, we will work in the Fourier transform
domain whenever possible. The wavefield can always be Fourier transformed over time t. If
there is no lateral velocity variation, then the wavefield also can be Fourier transformed over
the horizontal axis x. Thus,

P (kx , z, ω) = P (x, z, t) exp(ikx − iωt) dx dt, (D − 2a)

and inversely,

P (x, z, t) = P (kx , z, ω) exp(−ikx + iωt) dkx dω. (D − 2b)

When the differential operator in equation (D-1) is applied to equation (D-2b), we get
∂2 ω2
P (kx , z, ω) + − kx2 P (kx , z, ω) = 0. (D − 3)
∂z 2 v2
Although v can be varied with depth z in equation (D-3), for now we assume a constant
velocity case. The stratified earth case is considered later in this appendix. Equation (D-3) has
two solutions, one for upcoming waves, the other for downgoing waves. The upcoming wave
solution to equation (D-3) is recognized as
Mathematical Foundation of Migration 629

FIG. D-1. Relationship between migration and wavefield modeling (see Section D.1).

ω2
P (kx , z, ω) = P (kx , z = 0, ω) exp −i − kx2 z . (D − 4)
v2
Equation (D-4) also is the solution to the following one-way wave equation:
∂ ω2
P (kx , z, ω) = −i − kx2 P (kx , z, ω). (D − 5)
∂z v2
This solution can be verified by substituting equation (D-4) into equation (D-5).
We define the vertical wavenumber as
2
ω vkx
kz = 1− . (D − 6)
v ω
Equation (D-6) often is called the dispersion relation of the one-way scalar wave equation. By
using this expression, equation (D-4) takes the simple form
P (kx , z, ω) = P (kx , z = 0, ω) exp (−ikz z). (D − 7)
To determine the reflectivity P (x, z, 0) from the wavefield recorded at the earth’s surface
P (x, 0, t), proceed as follows:

(a) Perform a 2-D Fourier transform over x and t to get P (kx , 0, ω).
(b) Multiply by the all-pass filter exp(−ikz z) to obtain the wavefield P (kx , z, ω) at depth z.
(c) Perform summation over ω to obtain P (kx , z, 0).
(d) Finally, inverse Fourier tranform over kx to obtain the earth’s image P (x, z, 0) at that depth.

For the constant velocity case P (kx , kz , 0) can be computed by a direct mapping in the
transform domain from (kx , ω) to (kx , kz ) using equation (D-6) (Stolt, 1978).
The main objective here is to interpret equation (D-7) as a tool for downward extrapolating
wavefields given at the surface. While the mathematical development of the process presented
is simple, its physical basis is not obvious. In an effort to develop a physical motivation for
equation (D-7), a simpler derivation follows.
Given the upcoming wavefield P (x, 0, t) recorded at the surface, we can decompose it into
monochromatic plane waves, each traveling at a different angle from the vertical. We identify
these plane waves by attaching each one to a unique (kx , ω) pair. This plane-wave decomposition
is equivalent to Fourier transforming the wavefield to yield P (kx , 0, ω).
Now consider one of these plane waves as shown in Figure D-2. Imagine that this plane wave
passed point P at t = 0, traveled upward, and was recorded by a receiver at surface point G at
630 Seismic Data Analysis

FIG. D-2. Geometry for wavefield extrapolation (see Section D.1) (Yilmaz, 1979).

time t. For reflector mapping, we need to take the energy located at point G on the wavefront at
time t, back to its position at t = 0 — to reflection point P . To return the energy to P , it makes
sense to follow the same raypath used for outward propagation from P . The fact that the same
path is used means that downward continuation does not alter the horizontal wavenumber kx .
Suppose that the wavefront is moved to a depth ∆z = GG beneath the receiver at G so
that the waveform at G now is at G . If a receiver were buried at G , it would have recorded
the plane wave at t − ∆t, where ∆t is the traveltime between G and G . In other words, moving
the receiver at G vertically down a distance ∆z to a new location G changes the traveltime
along the raypath from G to P by −∆t.
From the geometry of Figure D-2, we have
∆z
∆t = cos θ, (D − 8a)
v
where v/ cos θ is the vertical phase velocity. We know the kx and ω values for the plane wave.
Suppose the distance between G and G is one wavelength λ. At time t − ∆t, the wavefront
intersects the x-axis at distance λx from G. From the geometric relation in Figure D-2, we have
λ
= sin θ. (D − 8b)
λx
By using the definitions λ = 2π/ (ω/v), λx = 2π/ kx , and equation (D-8b), we obtain
vkx
sin θ = (D − 8c)
ω
and
2
vkx
cos θ = 1− , (D − 8d)
ω
where ω/v is the wavenumber along the raypath. By substituting equation (D-8d) into equation
(D-8a), we have
2
1 vkx
∆t = 1− ∆z. (D − 8e)
v ω
Mathematical Foundation of Migration 631

As we move down, we do not want to change the amplitude of the plane wave. Given the
change in traveltime, ∆t by equation (D-8e), the corresponding phase shift is −ω∆t. At each
∆z step of descent, we may propagate the plane wave with a different velocity v(z). The total
phase shift to which the waveform was subjected with arrival at P is − ω dt.
To compute the wavefield at P , we use equation (D-8e) and multiply the transformed surface
wavefield P (kx , 0, ω) by
P z 2
ω v(z) kx
exp −i ω dt = exp −i 1− dz . (D − 9)
G 0 v(z) ω
Equation (D-9) is the same operator used in equation (D-7), except that equation (D-7) was
derived for constant v.
We now return to the more mathematical discussion and consider the stratified earth with a
velocity v(z). Since we have not Fourier transformed P (x, z, t) over z, the one-way wave equation
(D-5) also is valid for v(z):

∂ ω2
P (kx , z, ω) = −i − kx2 P (kx , z, ω), (D − 10)
∂z v 2(z)
in which case equation (D-6) becomes
2
ω v(z)kx
kz (z) = 1− . (D − 11)
v(z) ω
Substitution verifies that equation (D-10) has the following solution:
z
P (kx , z, ω) = P (kx , 0, ω) exp −i kz (z) dz . (D − 12)
0

We must check whether this solution satisfies the two-way scalar wave equation (D-3). By
differentiating equation (D-10) and using equation (D-11), we have
∂2 dkz (z) dv(z) ∂
2
P = −i P − ikz (z) P, (D − 13a)
∂z dv dz ∂z
where P = P (kx , z, ω). By substituting equation (D-10) for ∂P / dz, we obtain
∂2 dkz (z) dv(z)
P = −i P − kz2(z) P. (D − 13b)
∂z 2 dv dz
If the velocity gradient dv(z)/ dz is ignored, then the first term on the right side drops out. The
final expression then is
∂2
P + kz2(z) P = 0. (D − 13c)
∂z 2
When equation (D-11) is substituted into this expression, we get:
∂2 ω2
2
P+ 2 − kx2 P = 0, (D − 14)
∂z v (z)
which is identical to equation (D-3) where velocity can be varied with depth z.
So far, we have shown that a 2-D wavefield recorded at the earth’s surface can be extrap-
olated downward using the phase-shift operator given by equation (D-9). Wave extrapolation
can be done through either a constant-velocity medium (equation D-7) or a vertically varying
velocity medium (equation D-12). Seismic imaging is not complete until a stopping condition is
imposed on downward continuation. The process of downward continuation is terminated when
the clock, which measures t − dt, reads zero traveltime.
The concepts described above can be used to downward continue a complete seismic exper-
iment that involves many shots and receivers. The vertical wavenumber given by equation (D-6)
632 Seismic Data Analysis

is expressed as a sum of two square roots, one associated with downward continuation of shots
and the other associated with downward continuation of receivers:
ω
kz = DSR(G, S), (D − 15)
v
where v is the medium velocity that can be varied in depth z, and DSR stands for the double
square root (Claerbout, 1985):

DSR(G, S) = 1 − G2 + 1 − S2, (D − 16)


where G and S are the normalized receiver and shot wavenumbers, kg and ks , respectively:
vkg
G= (D − 17a)
ω
and
vks
S= . (D − 17b)
ω
The newly defined vertical wavenumber in equation (D-15) is inserted into the extrapolation
equation (D-7) to obtain
P (kg , ks , z, ω) = P (kg , ks , 0, ω) exp(−ikz z), (D − 18)
where P (kg , ks , 0, ω) is the Fourier transform of the prestack data P (s, g, z = 0, t) in shot-
receiver coordinates. This new extrapolation equation then can be used to downward continue
common-shot gathers, and by way of reciprocity, common-receiver gathers.
The DSR equation (D-16) is separable in terms of shot and receiver wavenumbers. This
separation means that we can start with the wavefields recorded at the surface as common-shot
gathers and use the first part of the DSR operator to downward continue the receivers to depth
∆z. Then, we can sort the already downward-continued wavefields into common-receiver gathers
and use the second part of the DSR operator to downward continue the shots to depth ∆z. By
alternating between common-receiver and common-shot gathers, the entire seismic experiment
(whole line) can be downward continued until imaging is accomplished (Figure D-3).
Although no approximation, besides the stratified earth assumption, is made in this alter-
nating downward-continuation scheme, it is computationally exhausting. In fact, most of today’s
seismic data processing is done in midpoint-(half) offset (y, h) coordinates, rather than in shot-
receiver (s, g) coordinates. Therefore, we will put DSR as defined by equation (D-16) into the
(y, h) coordinates.
The following coordinate transformation is required:
1
y= (g + s) (D − 19a)
2
and
1
h= (g − s). (D − 19b)
2
After the transformation (Claerbout, 1985), we obtain
G=Y +H (D − 20a)
and
S = Y − H, (D − 20b)
where Y and H are the normalized midpoint and offset wavenumbers, respectively:
vky
Y = (D − 21a)

and
vkh
H= . (D − 21b)

Mathematical Foundation of Migration 633

FIG. D-3. A flow diagram of shot-geophone migration (see Section D.1).

By substituting equations (D-21a) and (D-21b) into equation (D-16), the DSR equation
takes the following form in midpoint-offset coordinates:
DSR(Y, H) = 1 − (Y + H)2 + 1 − (Y − H)2 . (D − 22)
The vertical wavenumber (equation D-15) now is expressed in terms of normalized midpoint-
offset wavenumbers Y and H:
ω
kz = DSR(Y, H). (D − 23)
v
The newly defined vertical wavenumber in equation (D-23) is inserted into the extrapolation
equation (D-18):
P (ky , kh , z, ω) = P (ky , kh , 0, ω) exp(−ikz z), (D − 24)
where P (ky , kh , 0, ω) is the Fourier transform of the prestack data P (y, h, z = 0, t) in midpoint-
offset coordinates.
Figure D-4 shows the ω − ky plane for a specific value of z and h, and the ky − z plane for a
specific value of ω and h. The radial line A corresponds to ky = 2ω/v. The region in the ω − ky
plane below the radial line corresponds to the evanescent energy and that above the radial line
corresponds to the propagating energy. The same transition between the two regions also are
noted in the ky − z plane. The zero-offset case (Figure D-4c) clearly shows the evanescent energy
to the right of the point on the ky axis labeled as ky = 2ω/v dying off rapidly with depth.
The width of the propagation region stays constant with depth. The nonzero-offset case shown
in Figure D-4d, however, indicates that the width of the propagation region varies with depth
— zero at the surface z = 0 and approaching rapidly to the zero-offset case immediately at
shallow depths. The physical interpretation of this depth-dependency is quite intuitive — the
normalized offset wavenumber H becomes increasingly less significant at greater depths, and the
normalized midpoint wavenumber Y becomes the dominating wavenumber. More specifically,
on a CMP gather, moveout decreases with depth which implies nearly zero H.
Figure D-5 shows the response characteristics of the dispersion relation defined by equation
(D-23). Note the semi-elliptical wavefronts in the y − z plane for a single frequency ω; while in
the y − t plane, note the table-top traveltime trajectories. The equations for the wavefront and
traveltime trajectories are derived in Section D.2 using stationary phase approximations.
Note that in equation (D-16), the terms with different spatial wavenumbers are separable.
However, we have lost the property of separation in equation (D-22) because the operators in
Y and H are strongly coupled. As a result, the Taylor series expansions of the square roots in
634 Seismic Data Analysis

FIG. D-4. Real part of the ω − ky plane at z = 200 m: (a) DSR (equation D-23) with h = 0 m, and
(b) DSR with h = 400 m. Real part of the ky − z plane at ω = 16 m: (c) DSR (equation D-23) with
h = 0 m, and (d) DSR with h = 400 m (Yilmaz, 1979).

equation (D-22) yield terms that contain cross-products of the two wavenumbers. The penalty
for processing in the conventional coordinate system (y, h, t) is that strong coupling in the
extrapolation operator requires the entire prestack data set to be handled at the same time for
each depth step.
Conventional processing comprises two important steps. First, the data are organized into
common-midpoint (CMP) gathers, and normal-moveout (NMO) correction is applied to each
Mathematical Foundation of Migration 635

FIG. D-5. The response characteristics of the DSR operator (equation D-23) (Yilmaz, 1979). (a) Real
part of the y − z plane at 16 Hz and h = 400 m. Note the semielliptical wavefronts. (b) Real part of the
y − z plane at t = 1024 ms, h = 400 m. Because of the wraparound in h, we observe two wavefronts,
one for h = 400 m and one for h = 0. (c) Real part of the y − t plane at z = 200, 400, 600, and 800
m superimposed. These are the table-top trajectories for h = 400 m. The loci of the arrival times are
determined by a stationary-phase approximation to DSR (see Section D.2) (Clayton, 1978). Periodicity
in y and t result from approximating Fourier integrals by sums.

gather. The time shift ∆t = t(h) − t0 associated with the NMO correction is given by
2
2h
∆t = t0 1+ −1 , (D − 25)
vt0
where t(h) is the two-way traveltime for a given (half) offset h, and t0 is the corresponding two-
way zero-offset time. Here, v is the root-mean-square (rms) velocity at t0 . Equation (D-25) is
based on the stratified earth (zero-dip) assumption. After NMO correction, traces of the CMP
gather are stacked. This not only reduces data volume, but also enhances the signal-to-noise
ratio.
Second, the CMP stack is migrated as if it were the zero-offset wavefield generated by
exploding reflectors (Section 4.0). The equation used for the downward extrapolation portion of
migration is the solution to the one-way wave equation (equation D-12).
636 Seismic Data Analysis

To account for the one-way traveltime of the exploding reflectors model, the velocity used
in extrapolation is taken as half the medium velocity. Thus, the vertical wavenumber given by
equation (D-11) is expressed as
2
2ω vky
kz = 1− , (D − 26)
v 2ω
where v can be varied with depth z. (Since migration is done in midpoint space, kx has been
replaced with ky .)
Some of the current migration techniques based on wave extrapolation use certain rational
approximations to equation (D-26), while some implement the exact form in the frequency-
wavenumber domain (Section 4.1).
Conventional processing theory has an advantage over the exact theory represented by
the DSR equation in midpoint-offset space (equation D-22). Unlike the approach based on the
DSR equation, the conventional approach is composed of two separable operators — the NMO
correction and stack applied in offset space, and migration applied in midpoint space. However,
this advantage is based on zero-dip and zero-offset assumptions.
Where do we go from here? On the one hand, we have an exact theory that can handle all
dips and offset angles, but is difficult to implement. On the other hand, we have a conventional
approach that has the convenient property of separation, but is based on the zero-dip and
zero-offset assumptions.
To examine the relationship between the two approaches, we return to the exact theory
and make the same two assumptions that underlie the conventional approach. The zero-dip
assumption implies that the earth model is stratified in y−t domain. The seismic energy recorded
over such an earth is concentrated completely at the zero midpoint wavenumber ky = 0. This
suggests that we set the normalized wavenumber Y equal to zero in DSR as defined by equation
(D-22). The resulting operator here is defined as the stacking (St) operator

St(H) = 2 1 − H 2 . (D − 27)
As shown in Section D.2, the NMO shift given by equation (D-25) is a stationary-phase
approximation to equation (D-27) (Clayton, 1978). The St(H) operator condenses primary
information on a CMP gather down to zero offset. After applying this operator on a CMP
gather, we may keep the zero-offset trace and abandon all other offsets. Since a CMP stack
can be regarded as a zero-offset wavefield, equation (D-27) is a zero-dip NMO and stack-type
operator.
Application of the zero-offset (h = 0) assumption into the DSR operator is more subtle.
On a CMP gather at and near h = 0, energy essentially is concentrated at zero value of the
offset wavenumber kh = 0. In fact, NMO correction tries to push the primary energy on a
CMP gather toward kh = 0. Therefore, by setting the normalized offset wavenumber H = 0 in
equation (D-22), the exploding reflectors (ER) migration operator can be expressed as

ER(Y ) = 2 1 − Y 2. (D − 28)
Equation (D-28) is an approximation, because setting H = 0 is not exactly the same as
setting h = 0. By setting H = 0 in equation (D-22) and using the dispersion relation given by
equation (D-23), we have the zero-offset vertical wavenumber

kz = 1 − Y 2. (D − 29)
v
By substituting the definition for Y from equation (D-21a) into equation (D-29), we obtain
2
2ω vky
kz = 1− , (D − 30)
v 2ω
Mathematical Foundation of Migration 637

FIG. D-6. The response characteristics of the exploding reflectors operator ER(Y ) (equation D-29)
(Yilmaz, 1979). (a) Real part of the y − z plane at 16 Hz. Note the circular wavefronts. (b) Real part
of the y − z plane at t = 1024 ms. (c) Real part of the y − t plane at z = 200, 400, 600, and 800 m
superimposed. These are the hyperbolic trajectories. The loci of the arrival times are determined via
the stationary-phase approximation to ER(Y ) (see section D.2) (Clayton, 1978). Periodicity in y and t
result from approximating Fourier integrals by sums.

which is identical to equation (D-26). We conclude that the zero-offset migration operator ER(Y )
(equation D-28) derived from the DSR equation (D-22) is identical to the migration operator
that is based on the exploding reflectors model of conventional processing.
Figure D-6 shows the response characteristics of the dispersion relation defined by equation
(D-30). Note the semicircular wavefronts in the y − z plane and the hyperbolic traveltime curves
in the y − t plane. Compare these with the response of the complete DSR operator for the
nonzero-offset case in Figure D-5. The equations for the wavefront and traveltime trajectories
are derived in Section D.2 using stationary phase approximations.
638 Seismic Data Analysis

D.2 Stationary Phase Approximations

In this section, we shall use the method of stationary phase to derive the traveltime equa-
tion inferred by the double square root equation for nonzero-offset source-receiver separation.
Consider the double square root operator defined by equation (D-22) with the wavenumbers
Y and H defined by equations (D-21a,b). We want to operate on the transformed wavefield
P (ky , kh , z = 0, ω) with the DSR operator. Subsequent inverse Fourier transformation will yield
the wavefield P (y, h, z, t):

P (y, h, z, t) = P (ky , kh , z = 0, ω) exp (iΦz) dky dkh dω, (D − 31)

where the total phase Φ, normalized with respect to z, is given by


ω y h t
Φ = − DSR(Y, H) − ky − kh + ω . (D − 32)
v z z z
The main contribution to integration in equation (D-31) occurs when the phase stays nearly
constant. We therefore determine the variation of the phase with respect to variables ky , kh , and
ω
∂Φ ω ∂DSR ∂Y y
=− − , (D − 33a)
∂ky v ∂Y ∂ky z

∂Φ ω ∂DSR ∂H h
=− − , (D − 33b)
∂kh v ∂H ∂kh z
and
∂Φ 1 ω ∂DSR ∂H ∂DSR ∂Y t
= − DSR − + + , (D − 33c)
∂ω v v ∂H ∂ω ∂Y ∂ω z
and set each variation to zero. Substitute equation (D-32) and carry out the differentiations in
equations (D-33a,b,c) to obtain
1 G 1 S y
√ + √ = , (D − 34a)
2 1 − G2 2 1 − S2 z

1 G 1 S h
√ − √ = , (D − 34b)
2 1 − G2 2 1 − S2 z
and
1 1 vt
√ +√ = , (D − 34c)
1−G 2 1−S 2 z
where G and S are defined by equations (D-20a,b).
Now, eliminate G and S amongst equations (D-34a,b,c) to get the final expression from
stationary phase approximation to the double square root equation as
2 2
y+h + z2 + y−h + z 2 = vt. (D − 35)
This is the equation of an ellipse in the y − z plane at constant t (Section E.5). Figure D-5 shows
the elliptic wavefront and the table-top traveltime trajectory described by equation (D-35).
When equation (D-35) is specialized to the zero-offset case, h = 0, we obtain
vt
y2 + z2 = , (D − 36)
2
which is a circle in the y − z plane at constant t and a hyperbola in the y − t plane at constant
z. Figure D-6 shows the circular wavefront and the hyperbolic traveltime trajectory described
by equation (D-36).
Mathematical Foundation of Migration 639

We now consider the stacking operator defined by equation (D-27). The total phase is given
by
ω h t
Φ = − St(H) − kh + ω . (D − 37)
v z z
Differentiate equation (D-37) with respect to kh and ω and set the results to zero to obtain
H h
√ = (D − 38a)
1−H 2 z
and
1 vt
√ = . (D − 38b)
1−H 2 2z
Now, eliminate H between equations (D-38a) and (D-38b) to obtain the stationary phase ap-
proximation to the stacking operator:
vt
h2 + z 2 =
. (D − 39a)
2
Define the zero-offset time as t0 = 2z/v and substitute into equation (D-39a) to get
2
vt0 vt
h2 + = . (D − 39b)
2 2
Finally, rearrange to get the equation for normal moveout:
2
2h
∆tN M O = t0 1+ −1 , (D − 40)
vt0
where ∆tN M O = t − t0 . This is the same equation as equation (3-2b) in the main text with offset
defined as x = 2h.

D.3 The Parabolic Approximation

Start with the dispersion relation defined by equation (D-6) recast for the exploding reflectors
model for which v is replaced with v/2 to obtain
2
2ω vkx
kz = 1− . (D − 41)
v 2ω
Then apply Taylor series expansion and retain the first two terms:
2
2ω 1 vkx
kz = 1− . (D − 42a)
v 2 2ω
By simplifying, we get the dispersion relation associated with the parabolic equation
2ω vkx2
kz = − . (D − 42b)
v 4ω
By operating on the wavefield P (x, z, t) and replacing −ikz P with ∂P/∂z, we write the corre-
sponding differential equation as
∂P 2ω vkx2
= −i − P. (D − 43)
∂z v 4ω
Derivation of equation (D-43) is based on the constant-velocity assumption. Nevertheless,
just as we did for the 90-degree one-way wave equation (D-10), the 15-degree one-way wave
640 Seismic Data Analysis

equation (D-43) can be recast using a vertically varying velocity function v(z). Going one step
further, once equation (D-43) is inverse Fourier transformed from the horizontal wavenumber kx
to the horizontal axis x, we will replace v(z) with a laterally varying velocity function v(x, z).
Theoretically, this may not be permissible, but in practice, its validity is widely accepted.
The effect of translation is removed by retardation (Figure 4.1-18):
z
dz
τ =2 , (D − 44)
0 v̄(z)
where v̄(z) generally is chosen to be a horizontal average of v(x, z). The time shift defined by
equation (D-44) is equivalent to a phase shift in the frequency domain. Therefore, the actual
wavefield P is related to the time-shifted wavefield Q by
P = Q exp (−iωτ ). (D − 45)
By differentiating with respect to z, we get
∂P ∂ 2ω
= −i Q exp(−iωτ ). (D − 46)
∂z ∂z v̄(z)
Finally, by substituting equations (D-45) and (D-46) into equation (D-43), we obtain
∂Q vk 2 1 1
= i x Q + i2ω − Q. (D − 47a)
∂z 4ω v̄(z) v
The first term on the right side of this equation is called the diffraction term and the second
term is called the thin-lens term.
Consider the special case of v = v̄(z). The thin-lens term then vanishes and we are left with
∂Q vk 2
= i x Q. (D − 47b)
∂z 4ω
After inverse Fourier transforming, we obtain the parabolic differential equation
∂2Q v ∂2Q
= , (D − 48)
∂z∂t 4 ∂x2
where, in practice, velocity can be varied horizontally as well as vertically.
In some practical implementations of equation (D-48), downward continuation is performed
in τ rather than z. The migrated section then is displayed in time τ and the process is called time
migration. The two variables z and τ are related by equation (D-44). The dispersion relation
in equation (D-41) for the scalar wave equation in terms of ωτ = vkz / 2, the Fourier dual of τ ,
takes the form
2
vkx
ωτ = ω 1− . (D − 49a)

By squaring both sides, we get the equation of an ellipse in the ωτ − kx plane.
The dispersion relation in equation (D-42b) for the parabolic equation, also expressed in
terms of ωτ , takes the form
v 2 kx2
ωτ = ω − , (D − 49b)

which is the equation of a parabola in the ωτ − kx , plane. The first term is associated with a
vertical time shift that can be removed by retardation. To obtain the differential equation in
terms of τ , equivalent to equation (D-48), apply the second term on the right side of equation
(D-49b) to the retarded wavefield Q as defined by equation (D-45) and inverse Fourier transform
∂2Q v2 ∂ 2 Q
= . (D − 50)
∂τ ∂t 8 ∂x2
This equation is the basis for the 15-degree time migration algorithms.
Mathematical Foundation of Migration 641

The dispersion relations given by equations (D-49a) and (D-49b) are plotted in Figure D-7
for constant velocity v and input frequency ω = F E. Curve 1 is associated with the 90-degree
scalar wave equation and is the same as shown in Figure 4.1-25, except that the latter is in terms
of kz . Point A is mapped onto point B after migration with the 90-degree equation (D-49a).
The same point A is mapped onto point C after migration with the 15-degree equation (D-49b).
The dispersion curve 2 for the parabolic equation increasingly departs from the dispersion curve
1 for the exact wave equation as the dip gets steeper. (The dip before migration is measured as
the angle between the vertical axis FE and radial direction F A.) Thus, the parabolic equation
causes more and more undermigration as dip increases.

D.4 Frequency-Space Implicit Schemes

Higher-order approximations to the dispersion relation (equation D-41) of the one-way equa-
tion can be found by continued fraction expansion (Claerbout, 1985). When equation (D-41) is
rewritten, we have

kz = R, (D − 51a)
v
where
R= 1 − X 2, (D − 51b)
with
vkx
X= . (D − 51c)

The various orders of approximations to equation (D-51b) are defined by the following
recurrence relation
X2
Rn+1 = 1 − , (D − 52)
1 + Rn
with the initial value R0 = 1. By setting n = 0 in equation (D-52), we have
X2
R1 = 1 − . (D − 53)
2
Then, substitution of equation (D-53) into equation (D-51a) yields
2ω X2
kz = 1− , (D − 54)
v 2
which is the same equation obtained with the parabolic approximation, equation (D-42a).
The next higher-order expansion is obtained by setting n = 1 in equation (D-52) and using
equation (D-53) to obtain
X2
R2 = 1 − , (D − 55)
X2
2−
2
This is referred to as the 45-degree approximation to equation (D-41).
Ma (1981) discovered that the recurrence relation for the continuous fraction expansion can
be expressed as ratios of two polynomials for the even-ordered expansions. He also showed that
the expression can be split into the following partial fractions:
n
αi X 2
R2n = 1 − . (D − 56)
i=1
1 − βi X 2
642 Seismic Data Analysis

FIG. D-7. Dispersion relations for the 90-degree equation (D-49a), 15-degree equation (D-49b) and
the 45-degree equation (D-60a) with α1 = 0.5 and β1 = 0.25, plotted on the ωτ − kx plane for a given
input frequency, where ω = F E and velocity is v. Input frequency ω = F E = AP is mapped on to
output frequency ωτ which is PB, PC, and PD for the 90-degree, 15-degree, and 45-degree√equations,
respectively.
√ The wavenumber at the intersection of the curves along the kx axis is 2ω/v, 2 2ω/v, and
(4ω)/( 3v) for the 90-degree, 15-degree, and 45-degree equations, respectively.

For example, when n = 1,


α1 X 2
R2 = 1 − , (D − 57)
1 − β1 X 2
which is equivalent to the 45-degree expansion in equation (D-55) when α1 = 0.5 and β1 = 0.25.
Note that expansions R4 , R6 , R8 , · · · are made up of sums of the 45-degree term, each with
a different set of coefficients, αi and βi . Lee and Suh (1985) minimized the difference in the
least-squares sense between R of equation (D-51a) and R2n of equation (D-56) for a specified
dip angle and derived optimal coefficients (αi , βi ) for up to the 10th order (Table D-1).
Table D-1. Coefficients of optimized, fractioned, one-way wave equations (Lee and Suh, 1985).

Order, 2n Accuracy Deg. αi βi

2 45 0.5 0.25
2 65 0.478242060 0.376369527
4 80 0.040315157 0.873981642
0.457289566 0.222691983
6 87 0.004210420 0.972926132
0.081312882 0.744418059
0.414236605 0.150843924
8 90- 0.000523275 0.994065088
0.014853510 0.919432661
0.117592008 0.614520676
0.367013245 0.105756624
8 90 0.000153427 0.997370236
0.004172967 0.964827992
0.033860918 0.824918565
0.143798076 0.483340757
0.318013812 0.073588213
Mathematical Foundation of Migration 643

We now derive the differential equation associated with the 45-degree dispersion relation.
By substituting equation (D-57) and the definition for X given by equation (51c) into equation
(D-51a), we obtain

2ω α1 v 2 kx2
1
kz = 1− . (D − 58)
v 4ω 2 β1 v 2 kx2
1−
4ω 2
The first term is a vertical shift that can be removed by retardation in the same manner as
for the 15-degree approximation (equations D-42 through D-46). By simplifying the remaining
terms of equation (D-58), we obtain
β1 v 1 2ω
kz kx2 − kx2 − kz = 0. (D − 59a)
α1 2ω α1 v
Finally, by operating on the retarded wavefield Q(x, ω, z) (equation D-45), we obtain
β1 v ∂ 3 Q ∂2Q 2ω ∂Q
i 2
− 2
+i = 0. (D − 59b)
2α1 ω ∂z∂x ∂x α1 v ∂z
Kjartansson (1979) implemented equation (D-59b) for 45-degree modeling and migration in the
frequency-space domain. Migration in the frequency-space domain (commonly known as the
ω − x algorithm) involves two interleaved operations:

(a) a time shift based on equation (D-45), which is velocity-independent for time migration and
velocity-dependent for depth migration, and
(b) focusing the diffraction energy using equation (D-59b).

Once you have a code for the basic 45-degree operator, it is easy to implement the higher-
order approximations that are given by equation (D-56) with the associated coefficients in Table
D-1. Note that the difference between the 45-degree and 65-degree algorithms is the values used
for coefficients (α1 , β1 ). Also note that the 15-degree equation (D-47b) is obtained from equation
(D-59b) by setting α1 = 0.5 and β1 = 0.
The dispersion relation in equation (D-58) also can be expressed in terms of ωτ = vkz /2,
the Fourier dual of the time variable τ associated with the migrated data:
α1 ωv 2 kx2
ωτ = ω − . (D − 60a)
4ω 2 − β1 v 2 kx2
The first term is associated with a vertical time shift that can be removed by retardation as for
the 15-degree equation (D-49b).
To obtain the differential equation in terms of τ , equivalent to equation (D-59b), apply the
second term on the right side of equation (D-60a) to the retarded wavefield Q as defined by
equation (D-45) and inverse Fourier transform:
β1 ∂ 3 Q ∂2Q 4ω ∂Q
i − +i = 0. (D − 60b)
α1 ω ∂τ ∂x2 ∂x2 α1 v 2 ∂τ
This equation is the basis for the 45-degree and related steep-dip implicit finite-difference
frequency-space time migration algorithms.
The dispersion relation given by equation (D-60a) for the 45-degree equation, in which
α1 = 0.5 and β1 = 0.25, is plotted in Figure D-7 for constant velocity v and input frequency
ω = F E. Point A is mapped onto point D after migration with the 45-degree equation (D-60a).
The dispersion curve 3 for the 45-degree equation lies somewhere between those of the 90-degree
equation (D-49a) (curve 1) and the 15-degree equation (D-49b) (curve 2).
Figure 4.4-1 shows the impulse responses of various approximations to the one-way disper-
sion relation based on equation (D-56). Note that the wavefronts become increasingly closer to
a semicircle as the higher-order terms are included in equation (D-56). The 15-degree approxi-
mation yields an elliptical wavefront. The 45-degree approximation yields an impulse response
644 Seismic Data Analysis

of a heart shape. Refer to Section 4.4 for the practical aspects of 2-D frequency-space steep-dip
time migration, and Section 7.3 for its application to 3-D migration.

D.5 Stable Explicit Extrapolation

The exact extrapolation filter for a specific frequency ω and velocity v is expressed in the
frequency-wavenumber domain as
2
2ω vkx
D(kx ) = exp −i 1− ∆z . (D − 61)
v 2ω
The objective is to find, again for a specific frequency ω and velocity v, an explicit filter with
complex coefficients h(x) in the frequency-space domain such that, when Fourier transformed
to the frequency-wavenumber domain, the difference between the actual transform H(kx ) and
the desired transform D(kx ) of equation (D-61) is minimum. To ensure stability, we impose the
constraint that the amplitude of H(kx ) is never greater than unity within the propagation region
kx ≤ (2ω/v).
Consider the Fourier transform H(kx ) of the actual extrapolation filter h(x), which we will
write in discrete form as hn :
N
H(kx ) = h0 + 2 hn cos(nkx ), (D − 62)
n=1

where 2N + 1 is the length of the symmetric filter hn .


To determine the filter coefficients hn , perform Taylor series expansion of the exact D(kx )
and the actual H(kx ) extrapolators given by equations (D-61) and (D-62), respectively, and
match the coefficients of the terms in each series at kx = 0 (Holberg, 1988). Such a direct match
of the coefficients in the Taylor series results in a response H(kx ) whose amplitude is greater
than unity beyond a certain kx , thus violating the stability constraint that the amplitude of
H(kx ) is never greater than unity within the propagation region kx ≤ (2ω/v).
A way to circumvent the unstable response of the conventional Taylor series method is to
match the first M < N coefficients in the series expansion and set the remaining N − M to zero
(Hale, 1991). This modified Taylor series method proceeds as follows.
Let the filter coefficients hn be defined by the series
M
hn = c0 + 2 cm bmn , (D − 63)
m=1

where
2πmn
bmn = 2 cos . (D − 64)
N
By way of equations (D-63) and (D-64), equation (D-62) takes the form

H(kx ) = c0 1 + 2 cos(nkx )
n
. (D − 65)
2πmn
+2 cm 1 + 2 cos cos(nkx )
m n
N

Equation (D-65) can be written as a single summation


M
H(kx ) = cm Bm (kx ), (D − 66)
m=0
Mathematical Foundation of Migration 645

where
B0 (kx ) = 1 + 2 cos(nkx ), (D − 67a)
n

and
2πmn
Bm (kx ) = 1 + 2 cos cos(nkx ). (D − 67b)
n
N

Now, perform the Taylor series expansion of the exact extrapolator given by equation (D-
61):
∂D(kx ) kx2 ∂ 2 D(kx )
D(kx ) = D(0) + kx + + ···, (D − 68)
∂kx 2 ∂kx2
where the derivatives are evaluated at kx = 0, and thus the terms associated with the odd
derivatives vanish.
Matching the terms of the actual extrapolator H(kx ) with those of the desired extrapolator
D(kx ) is equivalent to matching their even derivatives
M
2j
cm Bm (0) = D2j (0), j = 0, 1, 2, . . . , M, (D − 69)
m=0
2j
where the derivative terms Bm are

B02j (0) = (−1)j 1 + 2 n2j , (D − 70a)


n=1

and
2j 2πmn 2j
Bm (0) = 2(−1)j 1 + 2 cos n . (D − 70b)
n=1
N

Equation (D-69) represents a system of M linear equations that can be solved for the
coefficients cm . The extrapolation filter coefficients hn then can be computed by substituting
the solution of equation (D-69) into equation (D-63).
The modified Taylor expansion yields filter coefficients hn with its response H(kx ) that
vanishes beyond a cutoff value for kx in the wavenumber domain that depends on the scalar
M . Also, the response of the extrapolator based on the modified Taylor expansion satisfies the
stability constraint H(kx ) < 1 in the passband region of the filter. The cutoff wavenumber kx
determines the maximum dip accuracy of the extrapolation filter. The larger the number of filter
coefficients 2N + 1, the steeper the dip accuracy. In practice, extrapolation filter lengths 7, 11
and 25 are often associated with 30-, 50- and 70-degree dip accuracies.
An alternative method for computing the filter coefficients hn with more accuracy at large
wavenumbers kx is based on the Remez exchange algorithm (Soubaras, 1996). The objective is
to minimize the error function
E(kx ) = W (kx )[D(kx ) − H(kx )], (D − 71)
where W (kx ) is a weighting function, such that
||E||∞ = max|E(kx )| (D − 72)
is minimum. The criterion ||E||∞ means that the actual response H(kx ) is equiripple. Specifically,
the response H(kx ) has extrema of the same absolute value with alternating signs within the
passband region specified by a cutoff wavenumber kx . The objective of the error function is to
minimize the deviation of the extrema from the desired value of unity.
646 Seismic Data Analysis

D.6 Optimum Depth Step

The objective is to specify an optimum depth step size that yields minimum-phase errors in
wave extrapolation as part of the design of finite-difference migration algorithms. First, we shall
review implicit and explicit schemes. Then, we shall derive equations for optimum depth size
for frequency-wavenumber implicit schemes.
Start with the one-way wave equation (D-5):

P (kx , z, ω) = −ikz P (kx , z = 0, ω), (D − 73)
∂z
whose solution is
P (z + ∆z) = P (z) exp −ikz ∆z , (D − 74)
where, for simplicity, kx and ω variables are omitted from P , and
2
2ω vkx
kz = 1− . (D − 75)
v 2ω
The phase of the exact extrapolation operator
exp (−ikz ∆z) (D − 76a)
is
Φ = kz ∆z, (D − 76b)
and its amplitude is
A = 1. (D − 76c)
Discretize the one-way wave equation (D-73):
δP
= −ikz P . (D − 77)
δz
Apply differencing approximation using the explicit scheme
P (z + ∆z) − P (z)
= −ikz P (z), (D − 78a)
∆z
and the implicit scheme
P (z + ∆z) − P (z) P (z + ∆z) − P (z)
= −ikz . (D − 78b)
∆z 2
Rewrite the explicit equation (D-78a) as
P (z + ∆z) = P (z) 1 − ikz ∆z . (D − 79a)
The phase of the explicit extrapolation operator
(1 − ikz ∆z) (D − 79b)
is
Φ = tan−1 kz ∆z , (D − 79c)
and its amplitude is
2
 = 1 + kz ∆z . (D − 79d)
Note that the amplitude of the explicit operator is greater than unity and grows from one
extrapolation step to the next. In fact, large depth steps cause the operator to yield unstable
results early in the extrapolation process. In general, explicit schemes tend to be unstable unless
special design considerations are made (Section D.5).
Mathematical Foundation of Migration 647

Rewrite the implicit equation (D-78b) as


1 − ikz ∆z/2
P (z + ∆z) = P (z) . (D − 80a)
1 + ikz ∆z/2
The phase of the implicit extrapolation operator
1 − ikz ∆z/2
(D − 80b)
1 + ikz ∆z/2
is
kz ∆z
Φ = tan−1 2 , (D − 80c)
1 − kz ∆z/2
and its amplitude is
 = 1. (D − 80d)
Note that the amplitude of the implicit scheme is unity. This means that implicit schemes are
unconditionally stable.
We now specialize the phase of the frequency-space implicit scheme given by equation (D-
80b) for the 65-degree dispersion relation as in equation (D-58) and obtain
k̂z ∆z
Φ = tan−1 2 , (D − 81a)
1 − k̂z ∆z/2
where
αk̂x2
k̂z = , (D − 81b)
2ω v
− β k̂x2
v 2ω
with α and β specified as in Table D-1 for the 65-degree scheme. Equation (D-81a) is the time-
retarded form of equation (D-58) and hence corresponds to the second term of the latter. The
discrete forms of the transform variables are (Claerbout, 1976):
ω
sin2 ∆x sin θ
4 v
k̂x2 = , (D − 81c)
∆x2 ω
1 − 4bsin2 ∆x sin θ
v
and
2 ω∆t
ω̂ = tan . (D − 81d)
∆t 2
The scalar b in equation (D-81c) is set to a value between 1/12 and 1/6. The phase error for the
65-degree implicit scheme then is given by
∆Φ = Φ − Φ, (D − 82)
where Φ and Φ are given by equations (D-81a) and (D-76a), respectively. Refer to the equations
(D-81a,b,c,d) and note that the phase error depends on ∆x, ∆t, v, θ, ω, and the depth step ∆z.
Figure D-8 shows contour plots of the phase error defined by equation (D-82) for three
ω/v values as a function of dip angle and depth step size. Note the complicated behavior of
the contours which indicates that the optimum depth step size is associated with a complicated
interdependence of the various parameters — dip, frequency, velocity, spatial, and temporal
sampling rates. Therefore, in practice, migration algoirthms that require wave extrapolation at
discrete depth steps usually do not incorporate an automated estimation of optimum depth steps.
Instead, a constant value between one-half and one dominant period, 20 to 40 ms, depending
on maximum reflector dip is specified in practice. See Section 4.4 for the practical aspects of
frequency-space migration.
648 Seismic Data Analysis

FIG. D-8. Phase error contours in degrees associated with the 65-degree implicit extrapolator (see
Section D.6) for a range of dip angles θ versus depth step sizes ∆z, and for specific ratios of ω/v, from
top to bottom, 0.001 (low frequency or high velocity), 0.06 (medium frequency and velocity) and 0.2
(high frequency or low velocity). The parameter b in equation (D-81c) was set to 0.14 for all three cases.
(Computation by Dave Nichols.)
Mathematical Foundation of Migration 649

D.7 Frequency-Wavenumber Migration

We start with the solution of the scalar wave equation for the zero-offset wavefield as given
by equation (D-7) and assume a horizontally layered earth model associated with a vertically
varying velocity function v(z). By inverse Fourier transforming equation (D-7), we have

P (x, z, t) = P (kx , 0, ω) exp(−ikz z) exp(−ikx x + iωt) dkx dω, (D − 83a)

where kz is defined by equation (D-6) adapted to the exploding reflectors model by replacing v
with v/2:
2
2ω vkx
kz = 1− . (D − 83b)
v 2ω
The imaging principle t = 0 then is applied to get the migrated section P (x, z, t = 0),

P (x, z, t = 0) = P (kx , 0, ω) exp(−ikx x − ikz z) dkx dω. (D − 84)

This is the equation for the phase-shift method (Gazdag, 1978). Equation (D-84) involves inte-
gration over frequency and inverse Fourier transformation along midpoint axis x. Refer to Figure
4.1-29 for a flow diagram of phase-shift migration.
We now consider the special case of constant velocity v. Stolt (1978) devised a migration
technique that involves an efficient mapping in the 2-D Fourier transform domain from temporal
frequency ω to vertical wavenumber kz . We rewrite equation (D-83b) to get
v
ω= kx2 + kz2 . (D − 85)
2
By keeping the horizontal wavenumber kx unchanged and differentiating, we obtain
v kz
dω = dkz . (D − 86)
2 kx2
+ kz2
When equations (D-85) and (D-86) are substituted into equation (D-84), we get
v kz v
P (x, z, t = 0) = P kx , 0, kx2 + kz2 exp(−ikx x − ikz z) dkx dkz .
2 kx2+ kz2 2
(D − 87)

This is the equation for constant-velocity Stolt migration. It involves two operations in the
f − k domain. First, the temporal frequency ω is mapped onto the vertical wavenumber kz via
equation (D-85). This is the same as mapping point B onto point B in Figure 4.1-25. Second,
the amplitudes are scaled by the quantity
v kz
S= ,
2 kx2 + kz2
which is equivalent to the obliquity factor associated with Kirchhoff migration (Section 4.1).
Refer to Figure 4.1-30 for a flow diagram of constant-velocity Stolt migration.
To extend the algorithm to the variable-velocity case, yet retain efficiency, Stolt (1978)
did a coordinate transformation that involves stretching the time axis to make the scalar wave
equation velocity independent. A summary of the theoretical procedure is given here. Consider
wavefield P (x, z, t) and the transformed wavefield P (x, d, T ):
P (x, z, t) = P (x, d, T ),
where T is the stretched time axis and d is the output variable (equivalent of z) for migration
in the stretched coordinate system. Here, the x variable is identical in both coordinate systems.
650 Seismic Data Analysis

This coordinate transformation basically is equivalent to stretching the data using the rms
velocities
t 1/2
1 2
T (t) = 2 dt vrms (t ) t , (D − 88)
c 0

where
t
2 1
vrms (t) = v 2 (t ) dt , (D − 89)
t 0

and c is an arbitrary reference velocity used to maintain the vertical axis as time after the
coordinate transformation from t to T . After some tedious algebra, the time-retarded scalar
wave equation takes the following form in the stretched coordinates (Stolt, 1978):
∂2P ∂2P 2 ∂2P
+ W + , (D − 90)
∂x2 ∂d2 c ∂d∂T
where W is a complicated function of velocity and coordinate variables. In practice, it normally
is set to a constant between 0 and 1. The procedure for Stolt migration with stretch follows:
(a) Start with the stacked section, which is assumed to be a zero-offset section P (x, z = 0, t).
(b) Convert this time section to stretched section P (x, d = 0, T ) by the coordinate transforma-
tion (equation D-88).
(c) 2-D Fourier transform the stretched section P (kx , d = 0, ωT ).
(d) Apply the following mapping function to perform migration:

1 ωT 1 ωT2
kd = 1− − − W kx2 . (D − 91)
W c W c2
This equation is based on the dispersion relation of the retarded wave equation in the
stretched coordinates (equation D-90). The expression for the output from migration is
(omitting the heavy algebra)
c 1 ckd
P (kx , kd , 0) = (1 − W + ) P kx , 0, (1 − W + K) , (D − 92a)
2−W K 2−W
where
1
K= . (D − 92b)
k2
1 + (2 − W ) x2
kd

(e) 2-D inverse Fourier transform the migrated section in the stretched coordinates P (x, d, T =
0).
(f) Convert back to the familiar space-time coordinates P (x, z, t = 0). This is the final migrated
section.
To derive the equations for Stolt migration with the output in time τ = 2z/v, use the
dispersion relation as in equation (D-49a) in lieu of equation (D-83b) and replace the vertical
wavenumber kz with the output frequency ωτ = vkz /2 in equations (D-84) through (D-88).
When W = 1, equation (D-91) takes the simple form

ωT2
kd = − kx2 , (D − 93)
c2
which makes mapping of equation (D-91) equivalent to the constant-velocity Stolt algorithm.
Note that Stolt migration with stretch tries to handle velocity variations. However, it is no
substitute for depth migration; it only accommodates velocity variations that can be handled by
time migration. Figure D-9 shows the flow diagram for Stolt migration with stretch. See Section
4.5 for the practical aspects of frequency-wavenumber migration.
Mathematical Foundation of Migration 651

FIG. D-9. Flowchart for Stolt migration with stretch.

D.8 Residual Migration

The relationship between the output and input temporal frequencies for a 90-degree time mi-
gration operator (rewritten from equation D-49a) is given by
v 2 kx2
ωτ = ω2 − , (D − 94)
4
where ωτ = vkz / 2 is the output frequency, ω is the input frequency, v is the true medium
velocity, and kx is the midpoint wavenumber. Equation (D-94) refers to single-pass migration if
the migration velocity is the same as the medium velocity.
Consider a two-pass migration, first with velocity v1 , followed by a second migration with
velocity v2 . The output frequency from the first pass is
v12 kx2
ω1 = ω2 − , (D − 95a)
4
which is the input frequency for the second pass
v22 kx2
ω2 = ω12 − , (D − 95b)
4
The horizontal wavenumber kx is fixed, since it is invariant under migration.
If the output frequency from the single-pass migration given by equation (D-94) is set equal
to that from the two-pass migration given by equation (D-95b), then the necessary relationship
between the residual migration velocity v2 and the medium velocity v can be established as
v 2 = v12 + v22 . (D − 96a)
652 Seismic Data Analysis

One practical scheme for residual migration is implemented as follows:


(a) The constant-velocity Stolt migration with a stretch factor of W = 1 is used for the first
pass. Velocity v1 in equation (D-96a) is chosen as the minimum value in the actual velocity
field.
(b) For the second pass, a dip-limited finite-difference migration can be used. The first-pass
migration brings the dips down to within the range that the second-pass finite-difference
migration can accommodate, accurately. Remember that the velocity field for the second
pass is computed by the the relation

v2 = v 2 − v12 , (D − 96b)
where v, v1 , and v2 are the original, first- and second-pass migration velocities, respectively. A
multiple-pass application of residual migration is referred to as cascaded migration. Refer to
Sections 4.3 and 4.5 for the practical aspects of cascaded migration and residual migration,
respectively.

REFERENCES

Baysal, E., Kosloff, D., and Sherwood, J.W.C., 1983, Reverse time migration: Geophysics, 48,
1514-1524.
Beasley, C., Lynn, W. S., Larner, K. L. and Nguyen, H. V., 1992, Extended Stolt f − k migration:
62nd Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 551-553.
Beasley, C. and Lynn, W. S., 1992, The zero-velocity layer: migration from irregular surfaces:
Geophysics, 57, 1435-1443.
Berkhout, A.J., 1980, Seismic migration – Imaging of acoustic energy by wavefield extrapolation:
Elsevier Science Publ. Co., Inc.
Berryhill, J.R., 1979, Wave-equation datuming: Geophysics, 44, 1329-1333.
Chun, J.H. and Jacewitz, C., 1981, Fundamentals of frequency-domain migration: Geophysics, 46,
717-732.
Claerbout, J.F., 1976, Fundamentals of geophysical data processing: McGraw-Hill Book Co.
Claerbout, J.F., 1985, Imaging the earth’s interior: Blackwell Scientific Publications.
Claerbout, J.F. and Doherty, S.M., 1972, Downward continuation of moveout-corrected seismo-
grams: Geophysics, 37, 741-768.
Clayton, R., 1978, Common midpoint migration: Stanford Expl. Proj., Rep. No. 14, Stanford
University.
Clayton, R. and Engquist, B., 1980, Absorbing side boundary conditions for wave-equation migra-
tion: Geophysics, 45, 895-904.
Deregowski, S. and Rocca, F., 1981, Geometrical optics and wave theory for constant-offset sections
in layered media: Geophys. Prosp., 29, 374-387.
Diet, J.P. and Lailly, P., 1984, Choice of scheme and parameters for optimal finite-difference mi-
gration in 2-D: 54th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 454-456.
Doherty, S.M. and Claerbout, J.F., 1974, Velocity analysis based on the wave equation: Stanford
Expl. Proj., Rep. No. 1, Stanford University.
Gazdag, J., 1978, Wave-equation migration by phase shift: Geophysics, 43, 1342-1351.
Gazdag, J. and Squazzero, P., 1984, Migration of seismic data by phase shift plus interpolation:
Geophysics, 49, 124-131.
Hale, I. D., 1991, Stable explicit depth extrapolation of seismic wavefields: Geophysics, 56, 1770-
1777.
Holberg, O., 1988, Towards optimum one-way wave propagation: Geophys. Prosp., 36, 99-114.
Kjartansson, E., 1979, Modeling and migration by the monochromatic 45-degree equation: Stanford
Exploration Project Report No. 15, Stanford University.
Kjartansson, E. and Rocca, F., 1979, The exploding reflector model and laterally variable media:
Stanford Exploration Project Report No. 16, Stanford University.
Mathematical Foundation of Migration 653

Larner, K. L. and Beasley, C., 1990, Cascaded migrations: improving the accuracy of finite-
difference migration: Geophysics, 52, 618-643.
Lee, M.W. and Suh, S.H., 1985, Optimization of one-way wave equations: Geophysics, 50, 1634-
1637.
Loewenthal, D., Lu, L., Roberson, R., and Sherwood, J.W.C., 1976, The wave equation applied to
migration: Geophys. Prosp., 24, 380-399.
Ma, Z., 1981, Finite-difference migration with higher-order approximation: Presented at the 1981
Joint Mtg. Chinese Geophys. Soc. and Soc. Explor. Geophys.
Reshef, M., 1991, Migration from irregular topography: Geophysics, 50, 1333-1335.
Rothman, D., Levin, S., and Rocca, F., 1985, Residual migration: Applications and limitations:
Geophysics, 50, 110-126.
Schneider, W., 1978, Integral formulation for migration in two and three dimensions: Geophysics,
43, 49-76.
Soubaras, R., 1992, Explicit 3-D migration using equiripple polynomial expansion and Laplacian
synthesis: 62nd Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 905-908.
Soubaras, R., 1996, Explicit 3-D migration using equiripple polynomial expansion and Laplacian
synthesis: Geophysics, 61, 1386-1393.
Stolt, R.H., 1978, Migration by Fourier transform: Geophysics, 43, 23-48.
Stolt, R.H. and Benson, A.K., 1986, Seismic migration: theory and practice: Geophysical Press.
Taner, M.T., Cook, E.E., Neidell, N.S., 1982, Paleo-seismic and color acoustic impedance sections:
applications of downward continuation in structural and stratigraphic context: 52nd Ann. In-
ternat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 110-111.
Taner, M.T. and Koehler, F., 1977, Wave-equation migration: Technical brochure, Seiscom-Delta,
Inc.
Yilmaz, O., 1979, Prestack partial migration: Ph.D. thesis, Stanford University.

You might also like