first break volume 32, December 2014

technical article

Tutorial: migration imaging conditions

Ian F. Jones1*

Abstract
Migration of seismic data is the process that attempts to build an image of the Earths interior from recorded field data, by
repositioning these data into their true geological position in the subsurface, using various numerical approximations of
either a wave-theoretical or ray-theoretical description of the propagation of sound waves in the subsurface.
This migration can be described as being performed in a number of stages, both for ray and wave-extrapolation based
methods. The final stage of the migration process is that which forms the image, via what is known as an imaging condition. In this tutorial, I will outline the various methods involved in forming imaging conditions, primarily for the case of
wave-extrapolation methods, and describe some of the techniques used to build gathers of pre-stack-migrated data for use
in post-migration velocity analysis.

Introduction
Migration has been in widespread use as an industrial process since the mid-1970s for 2D post-stack data and since
the late 1990s for pre-stack 3D data (e.g. Bancroft 1997,
2007; Jones et al., 2008). Consequently, most geoscientists
are familiar with the underlying concepts and the principles embedded in the various migration approaches used
in time and depth imaging (e.g. Jones and Lambar 2003;
Williamson et al., 2010), for both the ray-based methods
such as Kirchhoff and beam (Popov, 1982; Hill, 1990; Gray
1992, 2004), and to perhaps a lesser extent, the wavefield
extrapolation migration (WEM) methods (Hale, 1991). The
WEM techniques fall into two broad classes, depending on
whether they comprehend vertical changes in direction for
the propagating waves: namely one-way wave-equation
wavefield extrapolation migration (also often referred to as
WEM), and two-way wave-equation reverse-time migration
(RTM) (see for example, Hemon, 1978; McMechan, 1983;
Whitmore, 1983; Beysal et al., 1983; Bednar et al., 2003).
Perhaps less well understood are the various steps
involved in determining how a wavefield propagates through
the subsurface (as represented by a velocity-depth model,
e.g. Jones, 2010) and the subsequent step of building an
image of the associated reflectivity structure of the earth (e.g.
Claerbout, 1971; Leveille et al., 2011; Liu et al., 2011). Here,
I will be reviewing the latter stages of the imaging process:
the so-called imaging condition.

was formulated in a slightly different way by Jon Claerbout

(1971) who noted that if there is a reflector in the subsurface
(an impedance contrast), then at that point in space, when the
downgoing source wavefield hits the reflector, an upcoming
reflection is generated which then travels up to the receivers.
In other words, wherever the source and receiver wavefields
are in the same place at the same time, there must be a reflector. This observation forms the basis of an imaging condition.
This concept underpins how all migration techniques
build an image. For ray theory methods, we perform ray
tracing from the surface source locations down into the
earth for all desired subsurface image points, and ray trace
similarly for the surface receiver locations from subsurface
image points. In practice, for single mode propagation, the
ray paths are interchangeable for the path between a given
surface and subsurface point. For wavefield extrapolation
migration methods, we compute the wavefields for all
propagation times for both waves emitted at the source
location and waves recorded at the receiver position (using
finite difference techniques for example). We compute the
representative downgoing source wavefield by modelling the
response of a specified waveform going into the earth, and at
the same time back-propagate the actual recorded real data
from their surface receiver positions.
For both of these approaches an image is formed where
the downgoing and upcoming wavefields intersect, and in the
following sections I will describe these methods in more detail.

Imaging conditions

The imaging condition for ray methods

To paraphrase Newton, what comes up must have gone

down, and in the context of seismic exploration, for there to
be an upcoming echo from a subsurface reflector we must
have originally had a downgoing sound signal. This notion

In ray-based methods, the migration process is decomposed

into two or more stages (erven, 2001). For Kirchhoff,
this entails the calculation of travel times associated with a
given velocity model from representative surface source and

1
*

ION-GXT, UK.
Corresponding author, E-mail: [email protected]

2014 EAGE www.firstbreak.org

45

technical article

Figure 1 In ray-based methods, rays are traced through a current velocity

model from a surface position sampled on the 2D grid (DX, DY), representing
a source or receiver location (as denoted by the star) to a grid of subsurface
points on the 3D grid (dx, dy, dz).

receiver locations to all subsurface points on a predefined

grid (Figure 1). This step is typically followed by distribution of input data along the computed surfaces of constant
travel time (the isochrones), which represent the transit time
from the source to the receiver, via the subsurface reflection
point, then summation of contributions resulting from all
the input data traces. When a sufficient number of input
traces individual response contributions have been summed,
the imaging condition will eventually build up an image of
the reflector. Figure2a shows an early stage in this summation process; we can still discern the individual migration
responses. Once the process is complete and all input con-

first break volume 32, December 2014

tributions have been summed, the responses merge to build

an image of the reflector (constructive interference) and to
cancel the unwanted portions of the responses (destructive
interference), as seen in Figure2b.
For a beam scheme, the travel time computation is
preceded with a slope field determination (typically using
a slant stack analysis of shot, receiver, or offset gathers),
which facilitates selection of only those travel paths that will
meaningfully contribute to the image. In this case, ideally
the imaging condition is restricted to the Fresnel zone in the
vicinity of the reflector (Popov, 1982; Hill, 1990).
In both the Kirchhoff and beam methods, the step which
forms the image (the imaging condition) is essentially the
summation of all contributions resulting from the input data
traces after they have been spread along portions of the various isochrones. In other words, we rely on the principle of
constructive and destructive interference (stationary phase)
to form the image.
In ray-based methods, the procedure can be performed
for each trace at a time, and also, the image can be formed
for localized subsets of the final image. This feature makes
ray-based methods cost effective (thus desirable) for iterative
velocity model updating. A corollary of this observation is
that surface-offset gathers can easily be computed, rather
than just a final (stacked) image: which is a requirement
for velocity analysis. Surface offset gathers can be later converted to angle gathers, or in beam schemes and Kirchhoff
variants such as common reflection angle migration (Koren
et al., 2007) angle gathers can be formed directly.
The tracing of rays though a velocity model, in advance
of forming the image, has both advantages and disadvantages.
The advantage, just mentioned, is that subsets of the migrated
data can be cost-effectively computed by ray tracing for only

Figure2 The imaging condition for ray-based methods. a) For each input trace (with a contributing wavelet represented by the black dots), a migration response
is formed and added to the output image space (whether it be time or depth). b) When enough elemental responses are added, eventually an image of the
reflectors builds up and the superfluous parts of the responses tend to cancel-out. The solid black line shows the locations of the input trace wavelets (the dots
in Figure2a), and the white line formed by the superposition of responses (the tangential envelope) which is the output migrated result (Figure from Jones
et al., 2008).

46

www.firstbreak.org 2014 EAGE

first break volume 32, December 2014

technical article

the parts of interest (whether they be gathers or subsets of

the image). The disadvantage is that the formation of the
image is decoupled from the velocity model via the traveltime representation of the wave behaviour in the earth. This
decoupling introduces an undesirable resampling of information (converting velocity to equivalent representative travel
times) often on a much coarser grid than the velocity was
sampled at. Additionally, during the subsequent formation of
the image, this travel time information has to be interpolated
back from the sample grid on which it is stored, to the grid on
which we require the image. This has to be done to estimate
the travel times we would have from the actual source and
receiver locations by using those values we computed on the
coarse surface grid (DX, DY in Figure 1), and also for the
desired output migrated samples (typically on a 12.5 m x
12.5m x 5 m grid) given those we computed at spacing dx, dy,
dz (in Figure1): typically 100 m x 100 m x wz (where wz is
less than the water depth for marine data). In addition, using
a ray to represent a wave also limits our inherent resolution to
the size of the Fresnel zone (as in ray theory we are asserting
that every ray is totally independent of any neighbouring ray,
whereas for real waves this is untrue within the Fresnel zone).

The imaging condition for wavefield

extrapolation methods
Wavefield extrapolation migration of shot records involves
extrapolation of a synthetic source wavefield down into
the earth (downward continuation), and at the same time,
extrapolation of the actual real recorded wavefield back
from the receivers into the earth (called upward continuation, as we move energy back towards the source).
Referring back to Claerbouts imaging condition mentioned earlier, for a wavefield extrapolation migration we
can build an image by multiplying the source and receiver
wavefields together at each propagation time step, and where
we have a high amplitude contribution resulting from this
product, there must have been a reflector in the subsurface.
At each propagation time-step, the source-side and receiverside 3D wavefields are multiplied together, and at the end of
the extrapolation process (when we have exhausted all the
useful propagation time) all these hundreds of 3D product
volumes are summed together to form the image contribution resulting from this particular shot record.
This summation of wavefield products is referred to as
the correlation imaging condition in shot migration: the
image is being formed by what is essentially a correlation of
downgoing and upcoming wavefields (e.g. Bancroft, 1997).
This process is repeated for all available shots, and all these
overlapping 3D shot-contribution volumes are summed to
form the full migrated image of the study area. Figure3a is a
cartoon depicting the underlying correlation imaging condition for a shot migration, and Figure3b depicts a synthetic
modelling exercise showing the superposition for the downgoing and upcoming wavefields for a single shot, as well as
the resulting image after summing all shots (Figure3c).

2014 EAGE www.firstbreak.org

Figure 3 a) At each propagation time step, the downgoing and upcoming

(source and receiver) wavefields are multiplied, and then all these 3D product
volumes are summed to make the image contribution for this input 3D shot
gather. This cartoon is a bit misleading, as although the source-side wavefield is computed starting from time=0 in small propagation steps up to the
maximum recording time, the corresponding receiver-side back propagation is
done in reverse: from time=Tmax, back to time=0. Hence, all the receiver-side
time-movie snapshots must first be computed, then reversed in order, prior
to multiplication with the corresponding source-side movie frames (although
there are various computation tricks to save memory while doing this). b) A
synthetic modelling exercise showing the superposition for the downgoing
and upcoming wavefields for a single shot for propagation time 1.94s: the
red-and-black event is the downgoing source contribution, and the black-andwhite overlay is the upcoming receiver wavefield. c) The resulting image after
summing all shots for all time steps: the yellow ray-path denotes the reflection
point corresponding to the yellow circle indicated in 3b.

47

technical article

first break volume 32, December 2014

However, this summation of image contributions for all

propagation time results in an image, and not a gather of
pre-stack traces. There is no pre-stack gather information
associated with such shot migration schemes. Hence, if we
need gathers for post-migration velocity update (Zhou et
al., 2011), we have to introduce a method for forming them.

1D convolutional model analogy

In the 1D convolution model of reflection seismology, we
note that the recorded signal is equal to the Earths reflectivity
response convolved with the downgoing source wavelet. In
the context of the current discussion, the recorded signal is the
upcoming wavefield as measured at the receivers on the surface, and the source is the downgoing wavefield. So, we have:
Upcoming = Downgoing * Reflectivity

(1)

In order to estimate the reflectivity in this 1D case, we perform some form of trace-by-trace deconvolution, essentially
dividing by the recorded upcoming signal by the downgoing
source wavelet term:
Reflectivity = Upcoming / Downgoing 

(2)

However, in implementing the shot migration imaging condition, as described in the previous section, we are performing
a multiplication of the upcoming and downgoing 3D wavefields:
Reflectivity = Upcoming * Downgoing

(3)

As described in equation (2) we should actually be dividing

these two terms (division in the frequency domain or deconvolution in the time domain). In other words, in the tradition
of all good geophysicists, we take the liberty of making the
approximation that division can be replaced by multiplication! In order to recover from this embarrassing approximation, and atone for other sins, we can divide the product in
equation (3) by the square of the downgoing wavefield (plus
a small constant, epsilon, to help avoid division by zero).
Reflectivity = (Upcoming * Downgoing) /
(Downgoing2 + epsilon) (4)
The reason for trying to avoid performing this division
directly is that small terms in the denominator lead to
numerical instability, so the multiplication trick allows us
to see where we have a subsurface reflector (in accordance
with Claerbouts imaging principle). The subsequent division
in equation (3) attempts to balance the amplitudes. This latter division is referred to as illumination compensation (e.g.
Schleicher et al., 2007; Liu et al., 2007).
We can try to form this imaging condition more correctly
with a form of inversion or deconvolution (a least-squares
or deconvolutional imaging condition, e.g. Schuster, 1997;

48

Figure 4 One-way shot migration imaging condition for a simple reflector.

a) Downgoing source-side wavefield for one-way propagation, b) Upcoming
receiver-side wavefield, c) Imaging condition from multiplying both wavefields together to form a contribution to the final image.

Nemeth et al., 1999; Guiton et al., 2006), but this tends to be

very expensive, so is not currently in widespread industrial use.

Imaging condition artefacts

For a one-way solution of the wave equation using a shot
extrapolation migration scheme (Sava and Fomel, 2006),
the procedure of multiplying the downgoing and upcoming

www.firstbreak.org 2014 EAGE

first break volume 32, December 2014

Figure 5 Two-way shot migration imaging condition for a simple reflector. a)

Downgoing source-side wavefield for two-way propagation this has energy on
the downgoing path, but also creates a contribution back up along the upgoing path. b) Upcoming receiver-side wavefield propagated back into the earth
towards the reflector also reflects back upwards towards the source. c) Imaging
condition from multiplying both wavefields together to form a contribution to
the final image, but we also get an unwanted image contribution that has to be
removed (the grey region along coincident portions of the ray paths).

technical article
upwards and downwards. This permits imaging of steep and
complex geobodies via exploitation of double bounces and
turning rays (e.g. Hale et al., 1992; Bernitsas et al., 1997;
Cavalca and Lailly, 2005). Unfortunately, it also results in some
unwanted side effects, such as spurious strong near-vertical
artefacts emanating from overlapping downgoing and upcoming wavepaths, lateral amplitude terminations on strong vertical velocity boundaries (localized edge effects), and laterally
mispositioned double bounce arrivals (e.g. when we have significant error in the anisotropy parameters). Figure5 indicates
how and where some of these unwanted contributions can
form. The majority of the background low-frequency smear
(the grey regions in Figure5c) is often removed with filtering
(e.g. in the KxKy domain). These artefacts typically only persist
down to a depth where we encounter the first strong velocity
contrasts (where the critical angle may be reached). Here, one
side or the other (from the source and receiver wavefields)
fails to penetrate below that depth for certain angles, hence
this class of artefacts is reduced. Some of these artefacts can
be removed by modifying the imaging condition by employing
directional filters at each time propagation step (Poynting
vector filtering: Yoon and Marfurt, 2006) but this approach is
computationally expensive.
This procedure is shown in more detail in Figures 6 and7.
Figure 6a shows the result of migrating a single shot gather,
where a strong near-vertical artefact is produced outlined
in yellow (corresponding to the unwanted grey smears of
Figure 5c). After migrating all shots and summing their
contributions, a final image is produced (Figure 6b) where
most of the artefacts have been cancelled, and the background
low frequency-wavenumber smear has also been removed by
filtering. Figure7 shows several images for the downgoing and
upcoming wavefields for individual time steps in the propagation process, for the shot record migrated in Figure6a. In these
images, the red-and-black wavefront is the downgoing forward
modelled source wavefield. Superimposed on the figure is the
corresponding upcoming back-propagated receiver wavefield
(in black-and-white). At any given propagation time step,
where these two wavefields overlap with significant amplitude,
we will obtain a contribution to the final image via their multiplication. A summation over all these multiplied time frames
is then performed to produce the image contribution from
this shot. The position of the strong artefact highlighted in
Figure6a is overlain on each of the wavefield snapshots, and it
can be seen that the strong coincident energy of the downgoing
and upcoming wavefields builds to form this unwanted energy
in the series of time frames leading down to the reflector at
3km depth (corresponding to about 1.94s two-way time for
the offset in the shot gather shown).

Deghosting imaging conditions

wavefields at each time step leads to an image of any simple
reflections (Figure4).
For a two-way solution of the wave equation, both the
source-side and the receiver side wavefields can both travel

2014 EAGE www.firstbreak.org

There has been much recent development in the de-ghosting

of marine surface-tow seismic data, whereby the interference patterns of the source-side and receiver-side sea-surface
downgoing reflections (ghosts) are suppressed (e.g. Carlson,

49

technical article

first break volume 32, December 2014

Figure 6 a) The migration of a single shot record from synthetic modelling over a salt diapir, gives rise to an artefact (outlined in yellow). This artefact corresponds
to the grey unwanted regions of Figure 5c. Some of these artefacts will cancel when added to the contribution of many adjacent shot records (as shown in 6b).

Figure7 The origin of this strong artefact seen in Figure6a is highlighted in the wavefield propagation snapshots at times t = 0.77s, 0.99s, 1.21s, 1.61s (a, b, c,
d, respectively).

2007; Zhou et al., 2012). On the receiver side, the seismic

signal we wish to process is the upcoming energy, but from
the sea surface there is always a reverse polarity downgoing
reflected copy of this signal (which becomes more coherent
and pernicious with calmer weather, as the sea surface is then
more mirror-like).

50

By exploiting the upcoming and downgoing aspects of

two-way wavefield extrapolation migration (such as RTM),
both the source and receiver ghost effects can be incorporated into the imaging condition in an attempt to deconvolve
their interference effects during migration (e.g. Soubaras,
2010; Zhang et al., 2012).

www.firstbreak.org 2014 EAGE

first break volume 32, December 2014

Shot record extended imaging conditions

Each of the elemental sub-images resulting from the wavefieldextrapolation migration of an individual shot-record only contains a zero-offset trace: there is no inherent pre-stack gather
resulting from this process. The correlation imaging condition
only produces the image, not the gathers. Hence, to create a
gather (say for use in subsequent velocity analysis or an AVA
study) we need to invoke some additional computational
techniques.
The most widely used of these methods is called an
extended imaging condition. The idea in an extended imaging condition is to shift the downgoing and upcoming 3D
wavefield volumes with respect to each other just before they
are multiplied together at each propagation time step. These
shifted product volumes are then summed as before to form the
shifted image contribution from this particular shot record. This
shifting procedure is repeated several times, so that we end-up
with many 3D imaged volumes for each shot (one for each shift
value), rather than a single image volume for the shot.
If we re-sort these shift-volumes into gathers (where the
horizontal axis is the shift value), then we now have a prestack gather that can be used for velocity analysis and further
post-processing prior to stack (the theory behind this technique
is a form of interferometry). The shifting can be done in four
different ways: laterally in the inline direction, laterally in the
crossline direction, vertically in depth, or finally, in propagation
time. It could also be done in depth with respect to the reflector
normal (in other words, the vector represented by the inline,
crossline, and vertical components), and this vector azimuthangle gather can be computed from the inline and crossline
shift results. The history of these extended imaging condition
techniques dates back to focusing analysis in 2D preSDM (Faye
and Jeannot, 1986; Audebert and Diet, 1990; MacKay and
Abma, 1992) through to the more recent works of Ricket and
Sava, 2002; Sava and Fomel, 2003; 2006; Biondo and Symes
2004; and Xu et al., 2010, among others.
The basic shift-gather is not intuitively very useful, but they
can be converted into subsurface true angle gathers via various
transforms. However, all this shifting and transforming can suffer from aliasing of the underlying data due to poor sampling,
most commonly in the crossline direction (Zhou et al., 2011).
If we use a lateral inline or crossline shift, then the extended
imaging condition gathers are called sub-surface-offset gathers,
and if we shift in propagation time, they are called time-shift
(or delay-time) gathers. Either can be converted to angle gathers ready for residual-moveout (RMO) picking and velocity
updating, or post-processing, etc. Some authors prefer to use
the propagation time delay gathers, as these are thought to be
less prone to aliasing induced error. These time shift gathers are
converted to angle gathers via a tau-p transform and a velocity
scaling procedure.
Figure8a shows an RTM time shift gather from the flank
of a salt diapir. The horizontal axis corresponds to the degree of
time delay relative to the basic (zero shifted) conventional imaging condition. The central trace in the time-shift gather (lag=0)

2014 EAGE www.firstbreak.org

technical article
is simply the migrated trace from the conventional RTM
image. Shifting to positive or negative lags produces the traces
on either side, and as these delays are progressive, the overall

Figure8 a) The time shift gather at the location indicated by the yellow arrow
in Figure 10b. The central trace from this gather (at shift=0) is the actual
seismic trace at that location (as seen in Figure 10a). Shifting the wavefield
to slightly earlier times or slightly later propagation times prior to forming
the image produces the trace to either side of the central zero-shift trace.
This results in the sloping appearance of events in the time-shift gather. The
small segment of uphill trending energy (indicated by the white arrow) is
non-physical and needs to be removed, as does the low-frequency vertical
event on and near to the zero lag trace. b) Converting the vertical axis from
depth to time makes the sloping events appear linear, as velocity model distortion is removed. c) Rotating in accordance to the shift travel time, makes the
gathers look flat, so that in a subsequent tau-p transform the p=0 trace will
correspond to zero angle of incidence at a reflector.

51

technical article
appearance of the time-shift gather is that of sloping events. The
vertical low frequency stripe (on and around the central zero
lag trace) results from the two-way wave equation artefact of
Figure5c. It is interesting to note that this low frequency effect
will only occur if the velocity model is good. If it was far from
the real earth velocity structure, then the forward modelled
downgoing wave would seldom be coincident with the real
receiver wavefield. As mentioned previously, these artefacts
only persist down to a depth where the critical angle is reached,
typically at a large vertical velocity contrast event; in this case
the salt-sediment interface. In Figure8b, the data are stretched
vertically from depth to time so as to remove velocity-induced
distortion of otherwise linear trending events, and then rotated
to make them suitable for transformation to the angle domain
(Figure8c). Figure9 shows the data converted via a tau-p transform (and velocity scaling) to produce angle gathers, which can
be muted to remove unwanted energy prior to conversion back
to the time shift domain (Figure9b).
An example of such processing is shown in Figure 10.
The first image is an unfiltered RTM result for the shallow
seismic section showing the low-frequency background

Figure 9 a) Tau-p transform of the time-shift gather shown in Figure 8c:

energy to the left of the centre corresponds to non-physical events thus is
unwanted: the event circled in white corresponds to the dipping segment
indicated by the arrow in Figure8a. b) The time-shift gather after tau-p filtering and conversion back to the time domain via inverse-tau-p transform.

52

first break volume 32, December 2014

energy (discussed in Figure5c). The general background low

frequency noise is removed with a KxKy filter (figure10b).
However, we still see a strong event emanating from the left
flank of the salt dome which is clearly non-geological. Such

Figure 10 a) Shallow section from the RTM image prior to filtering the backscattered noise; b) Deeper section showing near-vertical RTM artefact emanating from a strong reflector termination (indicated with the yellow arrow). This
image is taken from an early stage of the velocity model building. c) Image
from later stage in the model building after filtering of RTM angle gathers (as
shown in Figure 9b). (From Jones and Davison, 2014: GXT RTM image shown
courtesy of Talisman Sinopec Energy UK and partners GdF-Suez, EON and
Idemitsu. Input data courtesy of CGG).

www.firstbreak.org 2014 EAGE

first break volume 32, December 2014

technical article

Figure11 a)-c): Contributions to the image at a specific specular point on a subsurface reflector from neighbouring shots and receivers for opening angles 0,
10, and 20 respectively. The angle gathers are formed by merging contributions from several shots (d).

unwanted energy can be removed by filtering of the RTM

angle gathers (e.g. Kaelin and Carvajal, 2011; Xie et al.,
2012), producing an acceptable image (Figure10c).

Poynting vector angle gathers

The Umov-Poynting vector (e.g. Yoon et al., 2004; Zhang
and McMechan, 2011) describes the direction a wavefront is
travelling in. In order to determine this information, we need
to take the spatial derivative of the wavefield at all times (the
divergence) and the temporal derivative at all spatial locations. This will tell us which way every point on the wavefront is travelling at all times. Once we have computed the
RTM image (using an imaging condition as described previously), we use the Poynting vector to determine the angular
contribution a given input shot gather has to each point in
the subsurface in this image. Combining the contributions
for all input shots permits building of angle gathers for each
subsurface location (e.g. He, et al., 2012).
Figure 11 depicts a subsurface reflection point on a 300
dipping horizon where the vertical dashed line intersects the
horizon. Any given incident (opening) angle at this subsurface
point will only be illuminated by a single shot and receiver
pair, as indicated in Figures 11a-c for opening angles 00, 100,
and 200 respectively. We can separate-out this part of the RTM

2014 EAGE www.firstbreak.org

image using the Poynting vector information, so as to obtain

the angular contribution at this point from that shot gather.
Combining information from neighbouring shots (Figure 11d)
builds-up the full angle gather which can facilitate analysis of the
subsurface incident angles. The angle described here is not the
reflector dip, but rather the opening angle at the reflector (the
angle of incidence with respect to the normal to the reflector).

The phase bias of the imaging condition

The sum of migration operators forms the impulse response. If
velocity changes across the survey, then each operator has a different shape, and the sum of operators (the impulse response)
has a different asymmetric shape. Even for a time migration
where the individual operators are symmetric (as they assume
a locally 1D earth model) the sum of the differing but individually symmetric operators will be asymmetric (e.g. Jones,
2010). The operators we are summing are however acausal
(as they emanate upwards from the scattering point, with no
contributions from below the reflector) hence they will have a
systematic phase bias in the resulting sum of operators. For a
real reflecting event, we would not expect to have any energy
appearing above the reflector, as that is non-physical (acausal),
rather we only expect to see energy once it has reflected (either
at the horizon or below it). To assess the phase behaviour of

53

technical article

Figure12 a) In migration with constant velocity, a flat layer can be built from
the laterally shifted sum of copies of the operator. So, a way of producing the
response for a single point on this flat layer, would be to take a single operator, and laterally sum all the samples to produce a single trace in the centre of
the operator (b). If the wavelets on the individual traces were spikes (c), then
the summation of traces in the operator would not produce the desired spike
at the reflector location, but rather a strange wavetrain extending upwards
from the reflector (d): on the left is shown the individual spikes on the trace,
and to the right is how this might appear as a continuous waveform.

first break volume 32, December 2014

create gathers. Here, I have outlined the mechanics of these

procedures, but I have not dwelt on aspects related to amplitude preservation, as that is beyond the scope of this work.
However, it should be recognised that it is at the imaging
condition stage of the migration that we have the opportunity to
obtain truly representative amplitudes if the imaging condition is
dealt with appropriately (e.g. Lumley, 1989; Zhang et al., 2005;
Zhang and Sun, 2009; Arntsen et al., 2010; Xu et al., 2010). In
ray-based methods, the amplitude weights during the imaging
phase have been well addressed for many years, and Kirchhoff
amplitudes are considered reliable. However, as indicated in this
brief overview, the imaging condition for wavefield extrapolation methods is difficult to execute in a stable but cost effective
way with current computer resources, hence various compromises are typically made, especially in forming angle gathers.

Acknowledgements
My sincere thanks to Jacques Leveille, John Brittan, and
Tony Martin for helpful suggestions for improvement of this
work, to Mick Sugrue for producing the RTM movie snapshots and to ION for permission to publish.

summing such operators it is instructive to see how we build

the response of a flat layer.
Figure 12 showns the mechanism for the origin of this
phase distortion. The seismic image of a flat layer underlying
a constant velocity medium is constructed from the sum of
laterally shifted copies of the same operator as indicated in
Figure12a. So, a way of producing the response for a single
point on this flat layer is to take a single operator and laterally
sum all its samples to produce a single trace in the centre of
the operator (Figure 12b). Now consider the case where the
wavelet on each individual trace of this operator prior to any
summation has zero phase. We have to consider what happens
to the phase of the resulting wavelet when we sum all those
elemental zero phase wavelets. If the wavelets on the individual
traces were spikes (Figure12c) then the summation of traces
in the operator would not produce the desired spike at the
reflector location, but rather a strange wave-train extending
upwards from the reflector (Figure 12d). In other words, it
looks like an asymmetric acausal operator.
Consequently, the phase of the wavelet constituting the flat
reflector, which comes from the immediate imaging condition
in migration, is not zero phase, but it should be (assuming that
we had performed conversion to zero-phase of the input data).
Consequently, we correct it with what is called the derivative
filter (as it has the same mathematical form as differentiation).
This is also referred to as a rho filter.

References

Discussion

erven, V. [2001] Seismic ray theory. Cambridge University Press.

The final stage in migration, the imaging condition, builds the

actual image. Depending on the migration technique used, this
can produce either the image alone (as in WEM and RTM) or
any combination of images and gathers (as in ray techniques).
For the extrapolation methods we need additional steps to

Claerbout, J.F. [1971] Toward a unifned theory of reflector mapping.

Arntsen, B., Tantserev, E. and Amundsen, L. [2010] True-amplitude crosscorrelation shot-profile imaging condition. 81st Annual International
Meeting, SEG, Expanded Abstracts, 3273-3277.
Audebert, F. and Diet, J.P. [1990] A focus on focusing. 52nd EAGE
Conference & Exhibition, Extended Abstracts, 107-108.
Bancroft, J.C. [1997] A Practical Understanding of Pre/Poststack Vol.1.
SEG.
Bancroft, J.C. [2007] A Practical Understanding of Pre/Poststack Vol.2 (Pre
stack). SEG.
Baysal, E., Kosloff, D.D. and Sherwood, J.W.C. [1983] Reverse time migration. Geophysics, 48, 1514-1524.
Bednar, J.B., Yoon, K., Shin, C. and Lines, L.R. [2003] One Way vs Two Way
Wave Equation Imaging Is Two-Way Worth It? 65th EAGE Conference
& Exhibition, Extended Abstracts, B11.
Bernitsas, N., Sun, J. and Sicking, C. [1997] Prism waves an explanation
for curved seismic horizons below the edge of salt bodies. 59th EAGE
Conference & Exhibition, Extended Abstracts.
Biondi, B. and Symes, W.W. [2004] Angle-domain common-image gathers
for migration velocity analysis by wavefield-continuation imaging.
Geophysics, 69, 1283-1298.
Carlson, D. [2007] Increased resolution of seismic data from a dual sensor
streamer cable. 74th Annual International Meeting, SEG, Expanded
Abstracts.
Cavalca, M., and Lailly, P. [2005] Prismatic reflections for the delineation
of salt bodies. 72nd Annual International Meeting, SEG, Expanded
Abstracts.

54

Geophysics, 36, no. 3, 467-481.

Faye, J.P. and Jeannot, J.P. [1986] Prestack migration velocities from depth
focusing analysis. 56th Annual International Meeting, SEG, Expanded
Abstracts.

www.firstbreak.org 2014 EAGE

technical article

first break volume 32, December 2014

Fletcher, R.F., Fowler, P., Kitchenside, P. and Albertin, U. [2005] Suppressing

artefacts in prestack reverse-time migration. 72nd Annual International
Meeting, SEG, Expanded Abstracts, 20492051.
Gray, S.H. [1992] Frequency-selective design of the Kirchhoff migration
operator. Geophysical Prospecting, 40, 565-572.
Gray, S. [2004] Gaussian beam migration of common-shot records. 71st
Annual International Meeting, SEG, Expanded Abstracts, 953-956.
Guitton, A., Kaelin, B. and Biondi, B. [2006] Least-square attenuation of
reverse-time migration artefacts. 73rd Annual International Meeting,
SEG, Expanded Abstracts, 23482352.
Hale, D. [1991], Stable explicit depth extrapolation of seismic wavefields.
Geophysics, 56, 1770-1777.
Hale, D., Hill, N.R. and Stefani, J. [1992] Imaging salt with turning seismic
waves. Geophysics, 57, no.11, 1453-1462.

Rickett, J. and Sava, P. [2002] Offset and angle-domain common imagepoint gathers for shot-profile migration. Geophysics, 67, 883889.
Sava, P. and Fomel, S. [2003] Angle-domain common-image gathers by
wavefield continuation methods. Geophysics, 68, 10651074.
Sava, P. and Fomel, S., [2006] Time-shift imaging condition in seismic
migration. Geophysics, 71, S209S218.
Sava, P., and Poliannikov, O. [2008] Interferometric imaging condition.
Geophysics, 73, S47S61.
Schleicher, J., Costa, J.C., Novais, A. [2007] A Comparison of Imaging
Conditions forWave-Equation Shot-Profile Migration. http://www.
portalabpg.org.br/PDPetro/4/resumos/4PDPETRO_1_2_0533-1.pdf,
accessed May 2014.
Schuster, G. [1997] Acquisition footprint removal by least square migration.
1997 Annual UTAM Report, 7399.

He, Y, Yoon, K., Cai, J., Yeh, A., Wang, B. and Li, Z. [2012] Practical Aspects

Soubaras, R. [2010] Deghosting by joint deconvolution of a migration and

of Subsalt Tomography Using Reverse Time Migration Based Angle

a mirror migration. 81st Annual International Meeting, SEG, Expanded

Gathers. 74 EAGE Conference & Exhibition, Extended Abstracts.

th

Hemon, C. [1978] Equations donde et modeles. Geophysical Prospecting,

26, 790821.

Abstracts, 34063410.
Whitmore, N.D. [1983] Iterative depth migration by backward time propagation. 53rd Annual International Meeting, SEG, Expanded Abstracts.

Hill, N.R. [1990] Gaussian beam migration. Geophysics, 55 (11), 14161428.

Williamson, P., Wang, B., Bevc, D., Jones, I.F. [2010] Full wave-equation

Jeannot, J.P. [1988] Full prestack versus shot record migration: Practical

methods for complex imaging challenges. The Leading Edge, 29 (3),

aspects. 58th Annual International Meeting, SEG, Expanded Abstracts.

264268.

Jones, I.F. [2010] An introduction to velocity model building. EAGE, Houten.

Xie, J., Whiteside, W., and Wang, B. [2012] Remove RTM Artifacts Using

Jones, I.F. and Davison, I. [2014] Seismic imaging in and around salt bodies.

Delayed Imaging Time Gathers. 83rd Annual International Meeting,

Accepted for SEG Interpretation.

Jones, I.F. and Lambar, G. [2003] Wave equation versus ray based imaging.
First Break, 21 (2), 1113.
Jones, I.F., Bloor, R.I., Biondi, B. and Etgen J.T. [2008] Pre-stack depth migration and velocity model building. SEG Geophysics Reprints, 25, Series
Editor: Michael A. Pelissier, 840 pages.
Kaelin, B. and Carvajal, C. [2011] Eliminating imaging artifacts in RTM
using pre-stack gathers. 82nd Annual International Meeting, SEG,
Expanded Abstracts.

SEG, Expanded Abstracts.

Xu, S., Zhang, Y. and Tang, B. [2010] 3D common image gathers from
reverse time migration. 81st Annual International Meeting, SEG,
Expanded Abstracts, 32573262.
Yoon, K., Marfurt, K.J. and Starr, W. [2004] Challenges in reversetime migration. 75th Annual International Meeting, SEG, Expanded
Abstracts, 10571060.
Yoon, K. and Marfurt, K.J. [2006] Reverse-time migration using the
Poynting vector. Exploration Geophysics, 37, 102107.

Koren, Z., Ravve, I., Bartana, A. and Kosloff, D. [2007] Local angle domain

Zhang, Q. and McMechan, G.A. [2011] Direct vector-field method to

in seismic imaging. 69th EAGE Conference & Exhibition, Extended

obtain angle-domain common image gathers for isotropic acoustic and

Abstracts, P287.

elastic reverse-time migration. Geophysics, 76, WB135WB149.

Leveille, J.P., Jones, I.F., Zhou, Z-Z., Wang, B., Liu, F. [2011] Subsalt Imaging

Zhang, Y., Roberts, G. and Khalil, A. [2012] Compensating for source

for Exploration, Production and Development. A Review. Geophysics,

and receiver ghost effects in reverse time migration. 83rd Annual

76, WB3WB20.

International Meeting, SEG, Expanded Abstracts.

Liu, F., Zhang, G., Morton, S.A. and Leveille, J.P. [2011] An effective imaging

Zhang, Y. and J. Sun, [2009] Practical issues of reverse-time migration: True

condition for reverse-time migration using wavefield decomposition.

amplitude gathers, noise removal and harmonic-source encoding. First

Geophysics, 76, S29S39.

Loewenthal, D., Stoffa, P.L. and Faria, E.L. [1987] Suppressing the unwanted
reflections of the full wave equations. Geophysics, 52, 10071012.
Lumley, D.E. [1989] Kirchhoff prestack depth migration: Imaging conditions
and amplitude recovery. 59th Annual International Meeting, SEG,
Expanded Abstracts, 13361339.
MacKay, S. and Abma, R. [1992] Imaging and velocity estimation with
depth-focusing analysis. Geophysics, 57, 16081622.
McMechan, G.A. [1983] Migration by extrapolation of time-dependent
boundary values. Geophysical Prospecting, 31, 413420.
Nemeth, T., Wu, C. and Schuster, G.T. [1999] Least-squares migration of

Break, 26, 1925.

Zhang, Y., Zhang, G. and Bleistein, N. [2005] Theory of true amplitude
one-way wave equations and true amplitude common-shot migration.
Geophysics, 70, E1E10.
Zhou, Z.Z., Howard, M. and Mifflin, C. [2011] Use of RTM full 3D
subsurface angle gathers for subsalt velocity update and image
optimization. Case study at Shenzi field, Geophysics, 76, WB27
WB39.
Zhou, Z.Z., Cvetkovic, M. Xu, B. and Fontana, P. [2012] Analysis of a
broadband processing technology applicable to conventional streamer
data. First Break, 30 (10), 7782.

incomplete reflection data. Geophysics, 64, 20822.

Popov, M.M. [1982] A new method of computation of wave fields using
Gaussian beams. WaveMotion, 4, 8597.

2014 EAGE www.firstbreak.org

Received: 3 July 2014; Accepted: 30 July 2014.

doi: 10.3997/1365-2397.2014017

55