Beruflich Dokumente
Kultur Dokumente
10.1190/1.3052116
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
Manuscript received by the Editor 20 May 2008; revised manuscript received 8 October 2008; published online 11 February 2009.
1
CGGVeritas, Calgary, Alberta, Canada. Email: sam.gray@cggveritas.com.
2
Colorado School of Mines, Department of Geophysics, Center for Wave Phenomena, Golden, Colorado, U.S.A. Email: normblei@gmail.com; norm@dix
.mines.edu.
© 2009 Society of Exploration Geophysicists. All rights reserved.
S11
S12 Gray and Bleistein
for example, to both common-offset and common-shot data; its only hoff migration; but, like our method, they provide greater accuracy
real requirement is that all input data 共e.g., all offsets or all shots兲 and allow for imaging into ADCIGs.
containing information from a particular image location be included GRT migration takes a dual approach to Kirchhoff migration but
to provide accurate amplitudes from that location. Zhang et al. obtains the same result. As described by Miller et al. 共1987兲, GRT
共2007兲 apply this Jacobian as a postmigration mapping from shot- migration reconstructs each subsurface scatterer by accumulating
domain common-image gathers 共SDCIGs兲, whose traces are in-
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
reflection data from all isochrons that intersect at the scatterer loca-
dexed by the lateral distance between image location and source lo- tion. At actual reflector locations, the live data that dominate the sum
cation, to angle-domain common-image gathers 共ADCIGs兲, whose are from the isochron that passes tangent to the reflector surface. The
traces are indexed by the opening angle at the imaged reflectors. conceptual distinction of GRT migration lies in performing the im-
This approach to true-amplitude imaging can be applied to any aging operation in the subsurface, as opposed to Kirchhoff migra-
migration method — in particular, to Gaussian-beam migration. In tion’s summation of data expressed in terms of surface coordinates.
addition, because the ray-based Gaussian beam contains the angle Although the isochrons are determined by the acquisition geometry
information as the imaging takes place, we can apply the mapping as well as the subsurface velocity structure, the quality of the image
implicitly during imaging, migrating directly into ADCIGs. and its amplitudes is determined by the geometry of the isochrons at
We present a derivation of true-amplitude Gaussian-beam migra- the scatterer locations. As with Kirchhoff and Gaussian-beam mi-
tion that follows those of Zhang et al. 共2007兲 and Hill 共2001兲. The grations, a reflector is well imaged only if a portion of an isochron
general framework applies to any migration method: crosscorrelat- passes tangent to it. In Gaussian-beam migration, the isochrons are
ing downward-continued wavefields from source and receiver loca- composed of local portions, each determined by the real part of the
tions. We write the wavefields in terms of Green’s functions for the traveltime surface for one of the localized Gaussian beams.
wave equation, and we express the Green’s functions as Gaussian- Several workers 共e.g., Deng and McMechan, 2007兲 mention
beam expansions. Our derivation is in terms of wavefields, but a shortcomings of true-amplitude migration, either from complexities
modification of the Green’s functions by Hill 共2001兲 allows us to in the migration velocity leading to uneven subsurface illumination
consider how migration acts on single traces 共even after they have or from elastic or anelastic effects that lie outside acoustic inverse
been slant stacked into neighboring beam center locations兲, as in theory. Such statements apply equally to all migration methods.
Kirchhoff migration. The derivation contains details of the steepest- Also, shortcomings in processing before migration, or interpretation
descent calculation used by Hill 共2001兲 to reduce the number of inte- after migration, can compromise the fidelity of migrated amplitudes.
grals, which correspond to loops in the computer implementation. 共In fact, the simpler of our two examples illustrates the effect of one
共We perform this analysis only for the 2D case; the corresponding of these problems — incomplete illumination resulting from finite
3D analysis is beyond the mathematical scope of this paper.兲 recording or migration aperture — on migrated amplitudes.兲 Al-
Zhang et al. 共2007兲 show that the crosscorrelation imaging condi- though these shortcomings imply limitations in our ability to ana-
tion is appropriate for true-amplitude migration when it is followed lyze migrated amplitudes except in the simplest situations, we take
by mapping from distance to angle. Here, we show that the same is the position that migrated amplitudes should be as accurate as possi-
true for Gaussian-beam migration. In addition, we show that the ble within the acoustic limits assumed by most migration methods. If
crosscorrelation imaging condition is appropriate for true-amplitude the cost of including accurate migration amplitudes is only a small
Gaussian-beam migration when imaging directly into ADCIGs us- part of the total migration cost, the effort of including accurate am-
ing ray information available to the migration, without the need to plitudes is warranted.
image into intermediate SDCIGs. We also show the relationship be-
tween the crosscorrelation imaging condition and the deconvolution TRUE-AMPLITUDE GAUSSIAN-BEAM
imaging condition usually used for true-amplitude Kirchhoff migra- MIGRATION
tion. A very simple example explains this relationship.
Because the action of Gaussian-beam migration on a single input
The 2D formula — Crosscorrelation imaging condition
trace can be analyzed in a manner similar to Kirchhoff and general- Prestack Gaussian-beam migration was developed first for marine
ized Radon transform 共GRT兲 migrations, we mention some connec- streamer-style common-offset, common-azimuth records 共Hill,
tions with those methods. As described by Hanitzsch 共1997兲, early 2001; Albertin et al., 2004兲 and later for common-shot records
versions of true-amplitude Kirchhoff migration relied on Bleistein’s 共Nowack et al., 2003; Gray, 2005兲. Because our presentation relies
共1987兲 modification of Beylkin’s 共1985兲 theory of imaging disconti- on observations about true-amplitude wave-equation migration, we
nuities. Bleistein 共1987兲 treats the imaging problem as an inverse begin with recorded data in the form of an actual wavefield, i.e.,
problem that integrates reflection data over acquisition surface coor- common-shot records. Although this approach differs from Hill’s
dinates. He incorporates Beylkin’s 共1985兲 geometric term into his 共2001兲 derivation of Gaussian-beam migration, most of the steps are
solution for the weight factor that makes the composition of the in- identical. In fact, if we consider the process that the method applies
version operator with the forward modeling operator a formal identi- to any given input trace in a well-sampled survey, we find that the or-
ty, i.e., the inverse operator applied to reflection data yields a reflec- der of the input data is irrelevant; we can apply the method to input
tivity model. traces gathered in any order with essentially identical migrated re-
Many subsequent true-amplitude Kirchhoff migration results rely sults. If the input traces are well sampled spatially 共receiver loca-
on this approach, which is limited to situations where the wavefields tions and shot locations at regular inline x and crossline y intervals
are well behaved 共i.e., away from caustics兲. Later generalizations al- with spacing that is fine enough to allow the unaliased slant stack of
low for imaging in greater structural complexity by removing the the highest temporal frequencies present in the recorded data at the
caustic restriction and allowing for multiple arrivals 共e.g., Xu et al., highest value of horizontal slowness兲 and if the subsurface illumina-
2001; Bleistein et al., 2005兲. These generalizations are much more tion at an image location from the downward-continued wavefields
complicated than conventional single-arrival true-amplitude Kirch- from source and receiver locations is nonzero, then Gaussian-beam
True-amplitude Gaussian-beam migration S13
migration will produce a migrated common-image gather 共CIG兲 at corded wavefield pU, the response to a 2D line source, is dimension-
that location that allows amplitude-versus-angle 共AVA兲 analysis. less. Inserting expressions 2 and 3 into equation 1 results in
In a second departure from Hill’s 共2001兲 derivation, we derive
true-amplitude Gaussian-beam migration in two dimensions only
and present the 3D expression without derivation. For efficiency, the
2D and 3D derivations rely on a saddle-point 共or steepest-descent兲
I共x;xs兲 ⳱ⳮ4 冕 d i 兩 兩
cos s
Vs
G*共x;xs ; 兲
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
Gaussian beams, a type of dynamic ray tracing, allows time t, which The choice of Green’s function determines the migration method.
appears in the phase, to be a complex quantity: t ⳱ tRe Ⳮ itIm, where Choosing a Green’s function of the form AART exp关i 兴, where AART
i ⳱ 冑ⳮ1. By contrast, the dynamic ray tracing used by standard and are both real, will lead to Kirchhoff migration. Choosing a nu-
Kirchhoff migration requires time along a raypath to remain real: t merical wavefield extrapolator based on a one-way wave equation
⳱ tRe.兲 The 2D derivation requires a saddle-point integral in a single will lead to a wave-equation migration.
complex variable, and the 3D derivation requires the integral in two Instead of these choices, we write the Green’s function as a sum-
complex variables. Because the theory for saddle-point integrals in mation of Gaussian beams 共Hill, 1990, 2001兲:
more than one complex variable is very technical without shedding
冕
light on the problem at hand, we present it elsewhere.
i dpx
We begin with the expression for a single migrated 2D shot record G共x;x⬘ ; 兲 ⳱ uGB共x;x⬘ ; 兲
with shot location xs ⳱ 共xs,0兲, receiver locations xr ⳱ 共xr,0兲, and 4 pz
冕
image locations x ⳱ 共x,z兲, using the imaging condition that is the
i dpx
crosscorrelation of the downward-continued upgoing and downgo- ⳱ AGB exp关i TGB兴. 共5兲
ing wavefields 共Zhang et al., 2007兲: 4 pz
冕
conditions for the dynamic ray-tracing equations that ensure 共1兲 lo-
⳱ⳮ2 dxr pU共xr ;xs ; 兲 calized planar beams uGB at initial source and receiver locations and
共2兲 regular, nonsingular behavior of the individual beams through-
out the subsurface. These depend on sensible choices for parameters
cos r w0, the initial half-width of the Gaussian beams, and r, the refer-
⫻ i G*共x;xr ; 兲, 共2兲
Vr ence frequency. 共A typical value for r is 2 ⫻ 10 Hz, and a typical
value for w0 is one wavelength at the reference frequency.兲
cos s The first of these conditions allows us to think of decomposing the
* 共x;x ; 兲 ⳱ ⳮ2
pD s
冑ⳮ i AART,s exp关ⳮ i s兴 source and recorded wavefields pD and pU into local plane-wave
Vs components that match the initial directions of the Gaussian beams.
cos s In particular, it allows us to consider propagating these components
⳱2 i G*共x;xs ; 兲. 共3兲 into the subsurface using individual beams uGB. If this can be accom-
Vs
plished, then we can migrate the local plane-wave components from
Here, Vs and Vr are surface velocities at the source and receiver a set of surface locations that is much sparser than the original set of
points; s and r are takeoff angles of the rays from source and re- source and receiver locations. Hill 共1990, 2001兲 provides recipes for
ceiver points to image location x; AART,s, AART,r, and s, r are real- this procedure, which is the heart of Gaussian-beam migration. In
valued amplitudes and traveltimes from asymptotic ray theory; and essence, Gaussian-beam migration replaces the migration of indi-
G共x;x⬘ ; 兲 is the Green’s function for the wave equation. The di- vidual data traces over all directions into the subsurface by the
mensionalities of pU and pD* are different: in two dimensions, the migration of localized, directional components of the wavefield with
source wavefield pD* has dimensions of inverse length, but the re- unique initial directions into the subsurface. Thus, at selected beam-
S14 Gray and Bleistein
center locations, local slant stacks are performed and the data mi-
grated over those initial directions using Gaussian beams as the
propagators.
Ds共L,pLx, 兲 ⳱ 冏 冏 冕
r
3/2
dxr pU共xr ;xs ; 兲
To make this rigorous, we first describe the local slant stack, then ⫻ exp关ⳮi pLx共xr ⳮ L兲兴
冋冏冏 册
the partition of unity that permits each input trace to contribute to a
共xr ⳮ L兲2
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
number of beam centers with weights that sum to unity. The local ⫻ exp ⳮ 共8兲
slant stack acts as a simple delay applied to an input trace; the delay r 2w20
accounts for the traveltime difference between a plane wave arriving
at the actual receiver location and the same plane wave arriving at and where AL and TL are Gaussian-beam amplitude and time from
the beam center location L ⳱ 共L,0兲. In the frequency domain, this is beam center location L. Notation for receiver slowness changes
accomplished by applying the phase shift exp关ⳮi px共xr ⳮ L兲兴 to from pr to pL to emphasize that all receiver-side rays emanate from
the input trace whose receiver is located at 共xr,0兲. Hill 共2001兲 incor- L.
porates this phase shift as a modification to his Green’s function ex- Next, we transform from source and receiver slownesses ps and pL
pansion 5, and Gray 共2005兲 suggests a modification to this phase to midpoint and offset slownesses pm and ph 共Hill, 2001; Gray, 2005兲,
shift when the elevations of L and xr are different. 共Effectively, this
factor is used to approximate the wavefield resulting from a point pmx ⳱ pLx Ⳮ psx , 共9兲
source at 共x⬘,0兲 when the source of ray propagation is at a nearby lo-
cation 共L,0兲.兲 phx ⳱ pLx ⳮ psx ,
Because the Gaussian-beam wavefront is perfectly planar only at
its initial location on the beam center and gradually accumulates to arrive at the migration formula:
nonzero curvature away from the beam center, approximating the
共nearly linear兲 delay by the linear slant stack incurs some kinematic
error; however, this error usually is negligible. Potentially more seri-
ous is the amplitude error incurred when the origin of the Green’s
I共x;xs兲 ⳱
8
⌬L r
2 冑2 w0 ⫻ 兺
L
冕 d i 冕冕 dpmxdphx
psz pLz
function 共the receiver location兲 is far from the initial point of the ray cos s cos L * *
tracing 共the beam center兲. However, Hill’s choice of partition of uni- As AL exp关ⳮ i 共Ts* Ⳮ TL*兲兴
Vs VL
ty ensures that the traces far from a beam center are downweighted
exponentially relative to traces close to a beam center, whose ampli- ⫻Ds共L,pLx, 兲. 共10兲
tude error is small. We use this approximation in our synthetic exam-
ples; its error is not enough to degrade the migrated amplitudes sig- This formula expresses a migrated record as a sum over beam-center
nificantly. locations, where each beam center contributes partial images from
The partition of unity is given by Hill 共1990兲 as many directions. We note again that A and T are complex valued be-
cause of our use of Green’s functions composed of Gaussian beams.
1⬇
⌬L
冑2 w0 冑冏 冏 兺 冋 冏 冏
r L
exp ⳮ
共xr ⳮ L兲2
r 2w20
册
, 共6兲
To reduce the computational load of equation 10, Hill 共2001兲 sug-
gests evaluating the integral over phx with its steepest-descent ap-
proximation. 共If A and T were both real valued, we would instead use
a stationary-phase approximation. In the 2D case, in fact, the steep-
where ⌬L is the spacing between beam-center locations.Asimple in- est-descent approximation to the integral agrees formally with a sta-
terpretation of this formula is as a discrete approximation, with ap- tionary-phase approximation, with the normally real quantities in
propriate change of variable, to the formula for the infinite integral of the stationary-phase formula replaced by complex quantities.兲
a Gaussian function: 兰ⳮ ⬁ 冑
⬁ exp共ⳮx /2兲dx ⳱ 2 . Other partitions of
2 In Hill’s 共2001兲 common-offset derivation, local slant-stacked re-
unity are possible, but equation 6 has benign effects on amplitude flection data D depend directly on pmx; but in the present common-
and it generalizes easily to three dimensions. shot formulation, D depends on pmx through pLx. In the steepest-de-
When we insert equations 5 共modified by the phase shift to ac- scent calculation for fixed pmx, we find the value of phx that minimiz-
count for the difference between receiver and beam-center loca- es the imaginary part of the total traveltime T ⳱ Ts Ⳮ TL. This sta-
0
tions兲 and 6 into equation 4 and rearrange terms, we obtain tionary value phx is used with pmx in equation 9 to determine critical
0 0
values psx , pLx ; these are used to evaluate D at the time corresponding
0 0 0
to the total time along beams psx and pLx . The critical values psx ,
⌬L r
I共x;xs兲 ⳱ pLx are also related to vertical slownesses psz ⳱ cos s /Vs, pLz
0 0 0
冕 冕
show the details of the steepest-descent integral in Appendix A; in-
cos s dpsx *
⫻兺 d i A exp关ⳮ i Ts*兴 serting its result into equation 10 leads to
Vs psz s
冕 冕
L
⌬L r
⫻
cos L
VL
冕 dpLx *
A exp关ⳮ i TL*兴Ds共L,pLx, 兲,
pLz L
I共x;xs兲 ⳱ 兺
8 2w 0 L
d 冑i dpmx
where T* ⳱ Ts* Ⳮ TL*. In this equation, As* and Ts* are evaluated Equations 2–6 are substituted into equation 13, and source and re-
along beam psx 0
; and D, AL*, and TL* are evaluated along beam pLx
0
. The ceiver slownesses ps, pr are transformed to midpoint and offset slow-
real parts of Ts* and TL* identify traveltimes from source and receiver nesses pm, ph in the numerator exactly as in equation 9 to arrive at a
beam-center locations to the image location. migration formula:
The second derivative, T*⬙共phx 0
兲, is assumed to be nonzero; it is
⌬L r cos s
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
R共x;xs兲 ⳱
1 cos s
2 Vs
冕 d i
pU共x;xs ; 兲pD
* 共x;x ; 兲
pD共x;xs ; 兲pD
s
* 共x;x ; 兲
s
.
migration to local slant-stack migration, variations of which have
gained popularity in recent years.
As a consequence of the second interpretation, the order of the in-
共13兲 put traces is largely irrelevant. For example, the same derivations
S16 Gray and Bleistein
can be applied to common-offset input data volumes 共Hill, 2001兲. * 兴兩exp关ⳮi T*兴
As*AL*兩det关Ts,ij
⫻ Ds共L,pLx
0 0
,pLy , 兲.
兩As兩2冑det关Tij*兴
For each input trace and a given beam center and incident and emer-
gence angles, the initial delay associates the trace with the incident
and emergence angles at the source and receiver locations corre- 共17兲
sponding to the beam center 共Hill, 2001, Figure 2b兲. At the end of the
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
procedure, which includes mapping the migrated data to opening an- In equations 16 and 17,
冏 冏 冕冕
gles, each input trace will have been migrated in a true-amplitude
sense. True-amplitude SDCIGs will be output when the deconvolu- 3
Ds共L,pLx,pLy, 兲 ⳱ dxrdy r pU共xr ;xs ; 兲
tion imaging condition is applied; true-amplitude ADCIGs will be r
output when the crosscorrelation imaging condition is applied.
⫻ exp关i pL · 共xr ⳮ L兲兴
冋冏冏 册
As an implementation detail, we note the amplitudes A in formu-
las 11 and 14 are complex. These normalized amplitudes are given 兩xr ⳮ L兩2
in Appendix B as 冑VQ0 /V0Q, where V is local wavespeed, Q is a ⫻ exp ⳮ 共18兲
complex dynamic ray-tracing quantity, and subscript zero refers to r 2w20
the initial location of the ray. Although most of the other quantities is the 2D generalization of the weighted slant stack of equation 8, L
are real or complex in some familiar fashion, complex-valued Q is ⳱ 共Lx,Ly,0兲, xr ⳱ 共xr,y r,0兲, pL ⳱ 共pLx,pLy,pLz兲, and the sum is a
relatively unfamiliar. In Gaussian-beam migration, it is real initially double sum over Lx and Ly. Furthermore, Tij* ⳱ 关 2T*„ph0 …/ phx phy兴
and gradually acquires a nonzero imaginary part as a ray propagates is the Hessian matrix of second derivatives of T* evaluated at critical
from the source or the receiver beam center into the subsurface. The point ph0 , replacing T*⬙ in the steepest-descent equation 12, and Ts,ij
*
phases of Q from the source and receiver then combine with the is a similar Hessian for Ts*. Under our assumption that Qs and QL are
phase of exp关ⳮi 共Ts* Ⳮ TL*兲兴/冑T*⬙共p0h兲 to form a total phase shift diagonal, the determinant of Tij* is given by
冋 册
that needs to be applied to the local slant-stack data D while comput-
ing the image. rw20 rw20
det关Tij*兴 ⳱ det Im关Qs兴Qsⳮ1 Ⳮ Im关QL兴QLⳮ1 ,
Vs psz VL pLz
The 3D true-amplitude migration formulas 共19兲
Except for the presence of quantities in all three spatial dimen- * is the same as equation 19 without the
and the determinant of Ts,ij
sions, the derivations of 3D true-amplitude migration formulas pro- second term on the right-hand side. Complex matrices Qs and QL are
ceed as for the 2D derivations. Hill 共2001兲 and Zhang et al. 共2007兲 scalar multiples of the 2 ⫻ 2 identity matrix. The scalars have the
provide a template for these derivations. As mentioned, however, the same values as solutions Qs and QL of the 2D dynamic ray equations,
steepest-descent integrals that make up the final step are more com- making it easy to evaluate the determinant numerically.
plicated in the 3D case. Here, we present the final migration expres-
sions for both crosscorrelation and deconvolution imaging condi-
tions. As a technical condition, in three dimensions, P and Q are 2D Mapping surface offset to opening angles
complex matrices that are diagonal initially 共Hill, 2001兲; the matrix With SDCIGs, it is easier to attach a physical meaning to the de-
quotient PQⳮ1 must remain diagonally dominant for the following convolution imaging condition than to the crosscorrelation imaging
formulas to hold. We force this condition by assuming that second condition; understanding this meaning helps in mapping Gaussian-
spatial derivatives of velocity appearing in the coupled differential beam migrated data from distance to subsurface opening angle. The
equations for P and Q can be neglected. This is equivalent to assum- deconvolution imaging condition forms the quotient of the upgoing
ing that the velocity is a piecewise linear function of the spatial coor- 共reflected兲 wavefield with the downgoing 共incident兲 wavefield. At a
dinates except for isolated negligible terms near velocity interfaces. reflector location, the quotient is the reflection coefficient at the an-
If the velocity function is smoothed so that its relative variations on gle of incidence, or half-opening angle, measured at a particular lat-
the order of a wavelength are small, this is not a damaging assump- eral offset from image point to source location. This heuristic obser-
tion for kinematic ray behavior, although it can affect ray amplitudes vation can be justified rigorously and used to help form ADCIGs for
beneath significant velocity boundaries, such as salt interfaces. Gaussian-beam migration.
Neglecting the second spatial derivatives of velocity leads to sca- With Gaussian-beam migration, we know the locations of the
lar matrices P and Q, which facilitates the evaluation of the steepest- source and receiver beam centers. If, at a reflector, we also know the
descent integrals. Then the 3D migration expressions for crosscorre- ray directions from those locations, we can assign the migrated am-
lation and deconvolution imaging conditions are, respectively, plitude to the opening angle. Gaussian-beam migration provides this
冕 冕冕
information as it migrates data: The subsurface ray angles at image
冑3V共x兲⌬Lx⌬Ly r2
I共x;xs兲 ⳱ ⳮ
16 2w20
兺L d dpmxdpmy
locations from source and receiver beam centers can be determined
and applied after migration to map offsets in SDCIGs to angles in
ADCIGs. This is different from 共and more cumbersome than兲 apply-
As*AL* exp关ⳮi T*兴 ing ray directions during imaging to produce ADCIGs without going
⫻ Ds共L,pLx
0 0
,pLy , 兲, 共16兲 through the intermediate step of producing SDCIGs 共Gray, 2007兲.
冑det关Tij*兴 Where true-amplitude migration is valid, these ADCIGs are suitable
for AVA analysis. In the absence of the mapping from offset to inci-
冑3⌬Lx⌬Ly r2 cos s
冕 冕冕
dence angle, Gaussian-beam migration using the deconvolution im-
R共x;xs兲 ⳱ ⳮ
16 2w20 Vs
兺L d dpmxdpmy aging condition produces SDCIGs that can be applied in amplitude-
versus-offset 共AVO兲 studies in areas of mild structural complexity.
True-amplitude Gaussian-beam migration S17
In the wave-equation migration formulation of Zhang et al. computer memory and disk needed to hold the larger gathers. As our
共2007兲, the crosscorrelation imaging condition 1 provides SDCIGs; use of wide-azimuth data increases, so will our use of CIGs indexed
the migrated amplitude in a particular offset bin is directly propor- by two quantities — either x- and y-offset or opening and azimuth
tional to the product of angle-dependent reflection coefficient multi- angles.
plied by the angular width of the offset bin. In particular, the angular
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
ed record is equivalent to an SDCIG. To convert the gather into an with the same fidelity as in Figure 1, with aperture truncation effects
ADCIG, we must perform one of the mappings proposed by Zhang that are more visible because of the relatively larger amplitudes pro-
et al. 共2007兲. In this constant-velocity, horizontal reflector case, duced at greater distances.
equation 20 provides an analytic relative decay function for the am- As mentioned, when the crosscorrelation imaging condition pro-
plitudes, vertically and laterally. In two dimensions, squared ampli- vides SDCIGs, the migrated amplitude in a particular distance bin is
is proportional to 1/冑x2 Ⳮ z2 and cos s ⳱ z/冑x2 Ⳮ z2, so
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
2
tude AART,s directly proportional to the product of angle-dependent reflection
the crosscorrelation imaging condition produces a constant angle- coefficient multiplied by angular width of the distance bin. The same
dependent reflection coefficient scaled by a factor proportional to the imaging condition can be used with subsurface angle information
decay function z/共x2 Ⳮ z2兲. available in Gaussian-beam migration to produce ADCIGs directly
Figure 1b shows the migrated amplitudes along the reflectors; in the migration. This is done by 共1兲 summing the ray angles from the
these amplitudes are the approximations, generated by the migra- source and beam center at each image location to find the opening
tion, to the decay function. Figure 1b also illustrates the exact decay angle for migrating a sample onto an image location and 共2兲 placing
function for the shallowest reflector, showing good agreement with the migrated amplitude into the bin corresponding to that opening
the migrated amplitudes. Migration aperture truncation artifacts be- angle. Then the amplitude in an angle bin is proportional to the prod-
gin to interfere with the migrated amplitudes at distances corre- uct of angle-dependent reflection coefficient multiplied by angular
sponding to half-opening angles approaching 60°, and migrated am- width of the angle bin. In other words, the laterally and vertically
plitudes decay rapidly beyond those distances, nullifying any benefit varying normalization of equation 20 becomes proportional to the
of amplitude preservation. width of the angle bins. Because the angle bins are all the same size,
In Figure 2 共deconvolution imaging condition兲, the goal is to pro- the normalization becomes a constant, so the crosscorrelation imag-
duce amplitudes proportional to reflection coefficients, without any ing condition is appropriate for migrating directly in the subsurface
vertically and laterally varying scaling. This goal is accomplished opening angle domain. 共For the same reason, the deconvolution im-
a) a)
Figure 3(a)
1000
2000
Depth (m)
3000
4000
5000
0 80
Angle (°)
b) b)
aging condition produces ADGICs whose traces have the inverse records, using the crosscorrelation imaging condition. Figure 3
of the decay function implied by equation 20. Thus, using the cross- shows an ADCIG at a single location. Disappointingly, the ampli-
correlation imaging condition produces a decay at far offsets in the tude behavior is worse than for the SDCIG amplitudes of Figure 1.
SDCIGs as in Figure 1, and using the deconvolution imaging condi- This problem is caused by the granularity of the opening-angle bin-
tion produces an unwanted extra gain in ADCIGs at large angles.兲 ning. Specifically, the small number of migration grid points in each
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
We illustrate the crosscorrelation imaging condition on angle angle bin at the smallest depths causes the angular width of the offset
gathers in Figure 3. Instead of mapping the migrated record of Figure bins in equation 20 to be calculated with significant jitter. Far from
1 from distance to angle as Zhang et al. 共2007兲 do, we build ADCIGs the source, where the angle bins contain a large number of image lo-
directly by migrating a range of 共translationally invariant兲 shot cations, this jitter decreases.
Although the aperture truncation effects slightly visible in Figure
1b and very visible in Figure 2b are present in Figure 3b, their effect
is dominated by the inaccuracy of the numerical approximation to
equation 20. As this example shows, these truncation effects can be
hard to detect 共especially when the crosscorrelation imaging condi-
tion is used兲, and they are not always considered when amplitude
analysis on CIGs is performed. However, as Figure 2b shows, aper-
ture truncation effects may cause significant errors on amplitude
analysis.
Our second example uses the Sigsbee2a model data set 共Paffen-
holz, 2001兲. We produced the migrated stack of Figure 4 by using the
crosscorrelation imaging condition to migrate the input traces into
ADCIGs, followed by stack. We show some ADCIGs in Figure 5,
corresponding to those shown by Zhang et al. 共2007兲. Qualitatively,
these gathers resemble those obtained by true-amplitude wave-
equation migration, but there are differences in amplitude details.
Generally, the Gaussian-beam migration amplitudes are weaker be-
neath salt than the wave-equation migration amplitudes. Because of
the complexity of the model, it is impossible to determine which
Figure 4. Stacked Gaussian-beam migrated image of the Sigsbee2a method produces more correct amplitudes, but the generally lower
data set using the true-amplitude crosscorrelation imaging condi- noise level in the wave-equation migrated gathers and stack suggest
tion. that the relative amplitudes produced by true-amplitude wave-equa-
tion migration are better.
CONCLUSIONS
We have derived true-amplitude versions of Gaussian-beam mi-
gration. Our derivation combines two migration methods 共classical
Gaussian-beam migration and true-amplitude wave-equation mi-
gration兲, applying a commonly used criterion for true-amplitude mi-
gration. This application produces expressions 11 and 15 in two di-
mensions and expressions 16 and 17 in three dimensions for true-
amplitude migration with crosscorrelation and deconvolution imag-
ing conditions. Our final expressions are similar to published results
on Gaussian-beam migration, with modified migration weights. The
modifications arise from a detailed analysis of a steepest-descent
calculation used to approximate a computationally intensive inte-
gral; the modified weights prove useful in migration. Migration
weights obtained without using the results of the steepest-descent
calculation produce migrated amplitudes that are similar to those
produced here, modified by a slowly varying scale factor. That is, the
2D terms involving T*⬙共p0h兲 in equation 12 and the 3D terms involv-
ing detTij* in equation 19 vary slowly in the far field of the source
and receiver beam-center locations, where the high-frequency ap-
proximation of the steepest-descent calculation is valid. Therefore,
neglecting these terms produces an amplitude error that is slowly
varying and can be difficult to measure. This illustrates that high-fre-
Figure 5. ADCIGs at five locations from true-amplitude Gaussian- quency asymptotic methods such as steepest descent produce first-
beam migration. Several diffraction and reflection events are order effects on the kinematic behavior of wavefields but only sec-
marked as in Zhang et al. 共2007兲. ond-order effects on the dynamic behavior.
S20 Gray and Bleistein
We also have provided physical interpretations of the amplitudes We proceed by assuming there is a point zs on the interval of inte-
on migrated traces using both imaging conditions when migrating gration for which
into CIGs indexed by surface distance or using subsurface angle.
The deconvolution imaging condition produces reflection coeffi- d⌿ 共zs兲 d2⌿ 共zs兲
⳱ 0, ⌿⬙ ⳱ ⫽ 0. 共A-2兲
cients 共multiplied by a constant factor兲 when migrating into SD- dz dz2
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
with Arg denoting the angle that picks out this principal argument of 1
the second derivative making an acute angle with the direction of the x共0兲 ⳱ x⬘, p共0兲 ⳱ 共sin ,cos 兲, 共0兲 ⳱ 0.
V共0兲
positive x-axis. Given this information, we can apply the method of
steepest descent 共Bleistein 1984, his equation 7.3.11兲 to the integral 共B-2兲
in equation A-1 to obtain the contribution to the asymptotic expan-
In addition, quantities PGB and QGB used in Gaussian-beam migra-
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
冑 冋 册
dQGB dPGB 1 2V
2 Arg关⌿ ⬙兴 ⳱ V共s兲PGB共s兲, ⳱ⳮ 2 QGB共s兲,
I共 兲 ⬇ f共zs兲exp ⳮ⌿ 共zs兲 ⳮ ds ds V 共s兲 n2
兩 ⌿ ⬙兩 2
冑
rw20 i
2 QGB共0兲 ⳱ , PGB共0兲 ⳱ . 共B-3兲
⳱ f共zs兲exp关ⳮ⌿ 共zs兲兴, ⬎ 0. 共A-7兲 V共0兲, V共0兲
⌿ ⬙
We distinguish between the complex functions QGB and PGB for
Gaussian beams and the corresponding real solutions of asymptotic
That is, with the form of the original integral in equation A-1, the
ray theory, QART and PART. The latter satisfy the same system of equa-
phase shift in the exponent provides just the right factor to yield a
tions B-2, but the initial conditions change as follows:
complex square root of ⌿ ⬙, using as the principal argument the
choice that makes an acute angle with the direction of the original 1
contour of integration. If is negative, the formula will contain the QART共0兲 ⳱ 0, PART共0兲 ⳱ . 共B-4兲
V共0兲
appropriate phase shift to transform this expansion into its complex
conjugate. By comparing these initial conditions with those for QGB and PGB in
Equation A-7, with phx substituted for z, 共cos s /Vs兲共cos L /VL兲 equation B-3, we conclude that
⫻共As*AL* /psz pLz兲Ds substituted for f, and iT* substituted for ⌿ , yields
equation 11 as an approximate evaluation of equation 10. QART共s兲 ⳱ Im关QGB共s兲兴, PART ⳱ Im关PGB共s兲兴. 共B-5兲
We need this relationship in the following.
A⳱ 冑 V共s兲Q共0兲
V共0兲Q共s兲
, ⌿ 共px兲 ⳱ ⳮ iT共px兲. 共B-6兲
In addition to its dependence on slowness px, the function ⌿ depends
1 on location x. However, this dependence does not concern us here,
T ⬅ T共s,n兲 ⳱ 共s兲 Ⳮ PQⳮ1n2 . 共B-1兲 so all indicated derivatives of ⌿ and T* are with respect to px. As in
2
Appendix A, the saddle point is determined by setting
The quantities Q and P are determined along the central ray as solu- d⌿ dT
tions of a system of dynamic ray equations, with ray-centered coor- ⳱ ⳮi ⳱ 0. 共B-7兲
dpx dpx
dinates s 共measured along the central ray from its initial location at s
⳱ 0兲 and n 共measured perpendicular to the central ray at points s兲. For homogeneous media, one of the central rays passes through
Within a particular Gaussian beam, complex amplitude and time A subsurface location x; the saddle point occurs at this central ray. This
and T are functions of position; several Gaussian beams from a given is true because the imaginary part of T is zero for this central ray; that
source or receiver location might strike a particular subsurface loca- is, the imaginary part of T is a minimum and, correspondingly, the
tion, so A and T are also functions of slowness px. This dependence real part of T is a maximum for this critical slowness value, which we
of A and T on px interests us here. The kinematic ray equations and call px0. This is equally true for heterogeneous media, except in two
initial conditions 共at s ⳱ 0兲 for position x, time , and slowness vec- cases: 共1兲 when no central ray from source or receiver beam center
tor p ⳱ ⵜ are location x⬘ passes close to x 共a shadow zone兲 and 共2兲 when, in the dis-
冉 冊
crete sampling of central rays used in the sum approximating the in-
dx dp 1 d 1 tegral, one or more central rays from x⬘ passes near the location x
⳱ V共s兲p, ⳱ ⵜ , ⳱ ,
ds ds V共s兲 ds V共s兲 without actually passing through it. Where an actual saddle point oc-
S22 Gray and Bleistein
curs, we can apply the steepest-descent formula of Appendix A to QART共s兲 QGB共0兲 Im关QGB共s兲兴 QGB共0兲
obtain the leading order asymptotic expansion of the Gaussian-beam T ⬙共p0x 兲 ⳱ ⳱
integral:
QGB共s兲 pz V共0兲
2
QGB共s兲 pz2V共0兲
冕
mum value of imaginary time is not close to zero, the central ray is
not close to a stationary value and the steepest descent calculation is
in error. However, the large value of imaginary time will supply a J共x;xs兲 ⳱ dphxAs*AL* exp关ⳮi T*共phx兲兴Ds共L,prx, 兲
large exponential decay to the wavefield contribution from this cen-
tral ray, and both the contribution to the wavefield and the error will 共B-12兲
be negligible. In such quasi-saddle point cases, we can perform the with T*共phx兲 ⳱ Ts* Ⳮ TL*. Here, J is the innermost integral 共for fixed
steepest-descent integral as if the saddle point were a true one. pmx兲 in equation 10, and we have neglected the dependence of the
At a true saddle point identified by the critical slowness value px0, functions on all variables except the variable of integration phx. In
the complex traveltime becomes real: particular, we have neglected the dependence on the other variable
of integration pmx. The asymptotic expansion that we derive here
T共p0x 兲 ⳱ 共s兲, 共B-9兲 must be applied for each pmx.
in agreement with the traveltime obtained by asymptotic ray theory. We propose to apply the general formula of equation A-7 for ap-
In fact, the asymptotic expansion of the Green’s function in equation proximating the integral J in equation B-12, with ⌿ ⳱ iT * 共phx兲. As
B-8 should agree, at least to leading order, with the expansion ob- happened for the calculation of the asymptotic Green’s function
tained by asymptotic ray theory. The leading order term of that B-11, there may be true simple saddle points 共for which the method
asymptotic expansion is of steepest descent is completely valid兲, quasi-saddle points, or
冋 册
shadow zones. If x is in a shadow zone of either xs or L, no integra-
冑
G共x;x⬘ ; 兲 ⬇ . 共B-10兲 共with no central ray pairs actually striking x兲, there will be a quasi-
2 兩 兩QART共s兲 saddle point. Then we can use the same argument as above to show
2 that the error in using the steepest-descent method to evaluate inte-
V共s兲
gral J is negligible compared with the error in the method itself. As in
When we compare this asymptotic expression for the Green’s func- the evaluation of the Green’s function, we are justified in using the
tion with the one in equation B-8, we obtain an indirect evaluation of steepest-descent method for both true saddle points and quasi-saddle
T⬙共px0兲, namely, 0
points. Doing this yields a critical value phx for which
True-amplitude Gaussian-beam migration S23
J共x;xs兲 ⬇ 冑 2
0 As AL Ds共L,prx, 兲
兩 兩T* ⬙共phx 兲
* *
REFERENCES
冋 册
Albertin, U., D. Yingst, P. Kitchenside, and V. Tcheverda, 2004, True-ampli-
tude beam migration: 74th Annual International Meeting, SEG, Expanded
Abstracts, 398–401.
⫻ exp ⳮ i T *共phx
0
兲Ⳮi sgn共 兲 . Beylkin, G., 1985, Imaging of discontinuities in the inverse scattering prob-
4
Downloaded 08/18/19 to 103.47.158.161. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
冋 册
physics, 70, no.4, S71–S77.
1 Im共Qs兲 1 Im共QL兲 ——–, 2007, Angle gathers for Gaussian beam migration: 69th Annual Con-
T *⬙共phx
0
兲 ⳱ ⳮ rw20 Ⳮ 2 2 . ference and Exhibition, EAGE, Extended Abstracts.
Vs2 psz
2
Qs Vr prz QL Hanitzsch, C., 1997, Comparison of weights in prestack amplitude-preserv-
ing Kirchhoff depth migration: Geophysics, 62, 1812–1816.
共B-14兲 Hill, N. R., 1990, Gaussian beam migration: Geophysics, 55, 1416–1428.
——–, 2001, Prestack Gaussian-beam depth migration: Geophysics, 66,
Equation B-14 evaluates T*⬙ in the asymptotic expression for J in 1240–1250.
equation B-12. We then substitute this expression into equation 11 to Keho, T. H., and W. B. Beydoun, 1988, Paraxial ray Kirchhoff migration:
Geophysics, 53, 1540–1546.
reduce the double integral over pmx and phx into a single integral over Miller, D., M. Oristaglio, and G. Beylkin, 1987, Anew slant on seismic imag-
pmx, which is equation 12. ing: Migration and integral geometry: Geophysics, 52, 943–964.
To complete the discussion, we must mention that the incident or Nowack, R. L., M. K. Sen, and P. L. Stoffa, 2003, Gaussian beam migration
for sparse common-shot and common-receiver data: 73rd Annual Interna-
reflected wavefield might have a caustic of central rays at some im- tional Meeting, SEG, Expanded Abstracts, 1114–1117.
age locations. Even though the regular behavior of the critical pair共s兲 Paffenholz, J., 2001, Sigsbee2 synthetic data set: Image quality as function of
migration algorithm and velocity model error: 71st Annual International
of Gaussian beams will produce nonzero amplitudes at such loca- Meeting, SEG, Workshop W-5.
tions, the interpretation of the peak amplitude at those image loca- Xu, S., H. Chauris, G. Lambaré, and M. Noble, 2001, Common-angle migra-
tions in terms of the geometric optics reflection coefficient will not tion: A strategy for imaging complex media: Geophysics, 66, 1877–1894.
Zhang, Y., S. Xu, N. Bleistein, and G. Zhang, 2007, True-amplitude, angle-
be valid. Briefly, such locations will produce nonsimple saddle domain, common-image gathers from one-way wave-equation migra-
points, violating the validity of the steepest-descent analysis. tions: Geophysics, 72, no.1, S49–S58.