GEOPHYSICS. VOL. 76, NO. 5 (SEPTEMBER-OCTOBER 2011); P. WB109–WB118, 13 FIGS.

10.1190/GEO2010-0403.1
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Multiarrival Kirchhoff beam migration

Jonathan Liu1 and Gopal Palacharla2

such as subsalt areas because only single arrivals are used in the
ABSTRACT imaging process. Using only single arrivals indicates there is only
one traveltime value associated with propagating from one surface
Kirchhoff-type prestack depth migration is the method location to one image location in the subsurface. In conventional
most popular for outputting offset gathers for velocity- approaches, rays are traced from a surface point to imaging points
model updating because of its flexibility and efficiency. in the subsurface. The traveltime table for imaging points is gen-
However, conventional implementations of Kirchhoff erated by ray tracing, and the input seismic trace is projected to
migration use only single arrivals. This limits its ability to these image points according to the calculated traveltimes. When
image complex structures such as subsalt areas. We use the more than one ray passes through an image point, only one raypath
beam methodology to develop a multiarrival Kirchhoff beam may be chosen (usually either the first arrival or the most energetic
migration. The theory and algorithm of our beam migration arrival). In this way, only one seismic arrival is associated with each
are analogs to Gaussian beam migration, but we focus on image point. Therefore, energy corresponding to other arrivals may
attaining kinematic accuracy and implementation efficiency. be missing or mispositioned during the imaging process. Using
The input wavefield of every common offset panel is multiarrival traveltimes from a source (or detector) to an image
decomposed into local plane waves at beam centers on the point may overcome the problems associated with using a single-
acquisition surface by local slant stacking. Each plane wave arrival traveltime (Xu et al., 2001). However, in general, this
contributes a potential single-arrival in Kirchhoff migration. approach is impractical in a production environment because multi-
In this way, our method is able to handle multiarrivals valued traveltimes are difficult to store and interpolate (Gray
caused by model complexity and, therefore, to overcome the et al., 2002).
limitation of conventional single-arrival Kirchhoff migra- Beam migration has advantages in handling multiarrival energy
tion. The choice of the width of the beam is critical to the (Hill, 1990, 2001), so we use the beam methodology to perform
implementation of beam migration. We provide a formula Kirchhoff migration with multiple arrivals. In our approach, the
for optimal beam width that achieves both accuracy and input wavefield near a surface point is decomposed into local plane
efficiency when the velocity model is reasonably smooth. waves by local slant stacking, and each plane wave contributes a
The resulting structural imaging in subsalt and other struc- potential single arrival in Kirchhoff migration. For each plane wave,
turally complex areas is of better quality than that from a central ray is traced from the surface point to an imaging point in
single-arrival Kirchhoff migration. the subsurface, and the traveltime in a neighborhood along the cen-
tral ray is calculated. Each input seismic plane wave is projected to
an image point according to the calculated traveltime. Because each
plane wave corresponds to one arrival and each subsurface point
INTRODUCTION may have several associated plane waves, our imaging process is
able to use and preserve the energy of multiarrivals. In this way,
Even today, as wave-equation migration becomes more and more a seismic trace can contribute many times to an image point through
efficient and practical in 3D depth imaging of complex structures, the multiarrivals, with each arrival being associated with one plane
Kirchhoff migration still retains its irreplaceable advantages of wave. To maintain accuracy and efficiency, each plane wave pro-
flexibility and efficiency for outputting offset/angle gathers for ve- pagates only within a beam, which is defined by a neighborhood
locity model updating. However, conventional implementations of along the central ray. The choice of the width of the beam is critical
Kirchhoff migration limit its ability to image complex structures to the implementation of Kirchhoff beam migration. We choose an

Manuscript received by the Editor 13 December 2010; revised manuscript received 1 April 2011; published online 21 November 2011.
1
ExxonMobil Upstream Research Company, Houston, Texas, USA. E-mail: [email protected].
2
ExxonMobil Exploration Company, Houston, Texas, USA. E-mail: [email protected].
© 2011 Society of Exploration Geophysicists. All rights reserved.

WB109
 WB110 Liu and Palacharla

optimal beam width using the criterion that the summation of all necessary for such processing. The input data are sorted into
beams is sufficient to provide minimum coverage of the imaging common-offset panels, with one trace per bin center in each panel.
area. We derive a formula that computes the optimal beam width To obtain better slant-stacking quality, we recommend performing a
from kinematic ray tracing. Although our formula is derived under partial NMO correction to a single offset value for each offset panel.
the assumption of a simplified velocity model (laterally invariant, The distance between beam centers is wavelength dependent, and a
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

mild gradient), it computes a beam width that is suitable for use with suggested value is
even more complex velocity fields.
Beam migration requires a uniformly sampled input wavefield. V avg
d bm ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; (1)
That means all offset values in a common-offset panel must be 2f min f max
exactly the same — a situation that rarely happens in field data.
The ideal solution is to regularize the input data to generate “perfect” where V avg is the average velocity, and f min and f max are the refer-
singlefold and single-offset common offset volumes. However, this ence minimum and maximum ordinal frequencies (not necessary to
approach may have too much computational cost, which is a subject exactly match the bandwidth in the input seismic data). Usually, we
of industry research today. Here, we simply apply partial normal take f min ¼ 6.25 Hz and f max ¼ 60 Hz. At each beam center, the
moveout (NMO) to map an approximate common-offset panel to localized wavefield is decomposed into plane waves (beams), with
a uniform-common-offset panel. This approximation works best if the beam-angle sampling (in slowness) determined by
the variation of offsets in the common-offset panel is reasonably sffiffiffiffiffiffiffiffiffi
small, dip angles are mild, and the azimuth distribution is narrow. 1 f min
The mechanics of our beam migration here are similar to those in Δp ¼ : (2)
Gaussian beam migration, but there are some differences: 2V avg f max

1) Gaussian beam migration uses a Gaussian function to decom- Compared to Gaussian beam migration (Hill, 1990), the derivation
pose and taper the input wavefield into local plane wave. Our of equations 1 and 2 is more empirical and focused on gaining the
beam migration uses a cosine square function to taper the input computational efficiency. For example, we use a value of d bm that is
p ffiffiffi
wavefield. 2 times the value used in the Gaussian beam migration defined by
2) Gaussian beam migration uses the Gaussian-shape function to Hill (1990). More widely spaced beam centers lead to less compu-
determine the beam width that controls the extrapolation region tational cost in our beam migration implementation. To avoid sharp
from central rays to the imaging domain. Our beam migration truncation edges, frequency-dependent tapering (for example, using
uses an “optimal” beam width to achieve both accuracy and a cosine square basis function) is recommended for the localized
efficiency. wavefield decomposition. Compared to the Gaussian taper, the co-
3) The derivation of formulas in Gaussian beam migration more sine square taper also preserves the partition of unity, but has
strictly follows mathematical and physical principles. Our beam a smaller beam width, so it requires fewer traces in the slant stacking
migration uses some empirical formulas to gain efficiency and and reduces the computational cost. The proof of the partition of
may not preserve amplitudes as well as Gaussian beam migra- unity for the cosine square taper (and many other types of tapers)
tion does. The main goal of our beam migration is to form struc- is shown in Appendix A.
turally correct images; the image amplitudes are secondary.
In this paper, we state key steps involved in implementing our Ray tracing of central rays
beam migration, and explain the theoretical principles of those Each beam angle corresponds to one central ray, with the raypath
steps. We illustrate the success of our beam migration by applying and traveltime determined by the standard kinematic ray equations
to two 2D synthetic data sets (2D Sigsbee and Marmousi2) and one (see Červený et al., 1982):
filed data example.
dx
METHOD ¼ Vp (3)
ds
The major steps in beam migration are:
dp 1
1) local plane-wave decomposition of input data, ¼ − 2 ∇V (4)
ds V
2) ray tracing of central rays,
3) traveltime extrapolation from central rays to the imaging dt 1
domain, ¼ ; (5)
4) application of the imaging condition,
ds V
5) selection of central ray pairs, where x is a point on the ray, p is the slowness vector, s is the path
6) amplitude weighting treatment, and length along the ray, V is the velocity, and t is the traveltime along
7) summation to form the image. the ray. There are two reasons we do not use dynamic ray tracing to
Each of the steps is described in additional detail below. compute the wave front curvature for traveltime extrapolation and
amplitude calculation, as is done in Gaussian beam migration. First,
Plane-wave decomposition of input data we have found dynamic ray tracing to be unstable when the velocity
model has rapid variations, and the instability limits the accuracy of
The input seismic data can be decomposed into local plane waves traveltime extrapolation and amplitude calculation. Second, we only
at beam centers by using local slant stacking. Data regularization is need the relative amplitudes to be sufficiently accurate to provide
 Multiarrival Kirchhoff beam migration WB111

migrated gathers for velocity-model updating. Equations 3–5 are distance between an image point and the central ray is less than wo
only valid in isotropic media. For anisotropic media, those equa- (see Figure 2). In this way, the summation of all beams is sufficient
tions should be modified according to the paper by Zhu et al. to provide minimum coverage of the imaging area. The optimal
(2007). choice of beam width not only reduces the computational cost,
but also reduces the error in traveltime extrapolation (because
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Traveltime extrapolation from central rays to beams the image point is closer to the central ray).
Suppose that x0 is a point on the central ray. For any point x near
Imaging condition and selection of central ray pairs
x0 , the traveltime to x can be approximated by the Taylor series
At each midpoint of a prestack common-offset data set, the data
Δs ðΔsÞ2 ∂V 1 2 are decomposed into plane waves and the central rays are traced at
tðxÞ ¼ tðx0 Þ þ − þ r mr ; (6)
V 2V 2 ∂s 2 the corresponding source and geophone locations; the following
imaging conditions must be addressed:
where Δs is the distance between x0 and the projection of x onto the
ray, r is the distance from x to the ray, and mr is a quantity related to
the wave front curvature (see Figure 1). As a result, mr can be ap- ts þ tg ¼ t (12)
proximated as
σ
mr ¼ ; (7) and
σ2 þ σ02
where
Z
σ¼ Vds; (8)
ray

V avg 2
σ 0 ¼ cos2 ðθ0 Þ ; (9)
2πf min

and θ0 is the ray takeoff angle at the surface. The computation of the
term mr in equation 7 only needs kinematic ray tracing, which
makes equation 7 a good approximation when the velocity model
is sufficiently smooth. The derivation of equation 7 is described in
Appendix B. To save computational time and memory usage, the
traveltime extrapolation should be implemented on a coarse grid
(we recommend 100-m spacing), and later those traveltimes can
be interpolated onto a finer grid during imaging.

Optimal choice of beam width

Suppose that the takeoff angle of a central ray is θ0 , and the emer- Figure 1. Traveltime extrapolation from x0 (a point on the central
ray) to a nearby point x within the beam. This figure illustrates the
gence point on the ray is x. Then the distance between two adjacent terms used in equation 6.
rays at x is given by
 
 ∂x 
Δθ0  
 ∂θ ; (10)
0

where Δθ0 is the separation between ray takeoff angles for two
∂x
adjacent central rays. The accurate computation of ∂θ 0
can be com-
plicated and unstable for an inhomogeneous velocity model. When
the velocity variation is mild and only depth dependent, a simplified
formula for expression 10 is
 
 ∂x 

wo ≡ 2Δθ0   ≈ 2Δθ0 σ ; (11)
∂θ  0 V 0

where V 0 is the velocity at the ray starting point, and σ is defined in

equation 8. The extra factor of two allows two beams from adjacent
central rays to have a sufficient overlap even when the velocity var-
iation is reasonably large. The derivation of equation 11 is described
in the Appendix C. The quantity wo can be calculated during ray Figure 2. Determining the optimal beam width. The quantity wo
tracing by using equation 11. The beam width is set such that the can be calculated during ray tracing by using equation 11.
 WB112 Liu and Palacharla

p0s þ p0g ¼ p0 ; (13) in amplitude terms in our beam migration. The values of r s and r g
are limited by
where t is the traveltime of the recorded data, t s is the traveltime
from the source to the image point, t g is the traveltime from the r s 2 wo;s −2 þ r g 2 wo;g −2 ≤ 2: (15)
geophone to the image point, p0 is the slowness vector at the mid-
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

point, p0s is the slowness vector at the source, and p0g is the slowness Equation 14 is directly analogous to Hill’s (2001) saddle point
vector at the geophone (shown in Figure 3). Each slowness vector analysis in which he uses minimum imaginary traveltime as his
corresponds to a central ray from the source or geophone. For a selection criterion.
given image point, if more than one pair of central rays from
source/geophone satisfies equation 13, the one that maximizes Amplitude weight
 
π 2 The cosine function in equation 14 provides tapering from the
as;g ≡ cos2 ðr w −2 þ r g 2 wo;g −2 Þ (14)
4 s o;s central ray to the beam edge, so as;g can be used as an amplitude
weight for imaging. A smaller distance between the image point and
is chosen. Here r s and rg are the distances from the image point to the central ray implies that a higher amplitude weight (a larger value
the central rays from the source and the geophone, respectively, and of as;g ) should be assigned for imaging. In addition, we use another
wo;s and wo;g are the quantities defined in equation 11 on the central amplitude factor
rays from the source and the geophone, respectively (shown in  
Figure 4). Maximizing as;g implies that the imaged point is closest 1
cs;g ≡ cos ðθg − θs Þ ; (16)
to the selected central pair. From equation 14, as;g will also be used 2

where θs and θg are ray emergence angles from source and receiver,
respectively. One benefit of the amplitude term in equation 16 is to
attenuate the contribution of wide-angle post-critical energy to
imaging.

Summation to form the image

The summation to form the image is a trivial operation for multi-
arrival Kirchhoff beam migration. Compared to the single-arrival
Kirchhoff migration, we perform one more loop for all slowness
vectors at a beam center (midpoint).

Processing flow
Figure 3. The imaging condition in beam migration. The traveltime The processing flow for multiarrival Kirchhoff beam migration
in the recorded data is equal to the traveltime from the source to the can be represented as follows:
image point plus the traveltime from the geophone to the image
point; the slowness vector at the midpoint is equal to the slowness 1. Prepare a common-offset panel, velocity model, and input/
vector at the source plus the slowness vector at the geophone.
output information.
2. At each beam center (midpoint):
2.1. Decompose the wavefield into local plane waves (beams)
at the midpoint.
2.2. At the source, for each p0s ,
2.2.1. Trace a central ray.
2.2.2. Extrapolate traveltime ts from the central ray to the
beam and compute r s wo;s −1 .
2.3. At the geophone, for each p0g ,
2.3.1. Trace a central ray.
2.3.2. Extrapolate traveltime t g from the central ray to the
beam and compute r g wo;g −1 .
2.4. For each midpoint beam with slowness p0 ,

2.4.1. Select pairs of central rays from source/geophone

Figure 4. Selection of traveltimes. For a given image point, if more such that equation 13 holds.
than one pair of central rays satisfies the slowness imaging condi- 2.4.2. Compute t s þ tg and as;g , and replace the previous
tion, we choose the pair closest to the image point. values if as;g is larger than the previous value.
 Multiarrival Kirchhoff beam migration WB113

2.4.3. Migrate the plane wave onto each image point by beam migration provides a much cleaner image in the subsalt, and
using traveltime t s þ t g and amplitude as;g cs;g , and faults and diffractors are well defined (Figure 7). Figure 8 shows the
sum the image. 2D Marmousi2 velocity model. Strong velocity contrasts in the cen-
tral part are likely to produce multiple arrivals. Figure 9 shows the
Step 2.1 enables multiarrivals to imaging compared to conventional
result of single-arrival (the most energetic arrival) Kirchhoff migra-
single-arrival Kirchhoff migration.
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tion. The anticlines in the central deep part (highlighted) are blurred
and distorted. In contrast, the multiarrival Kirchhoff beam migration
RESULTS is able to resolve the anticlines (shown in Figure 10).
We applied our multiarrival Kirchhoff beam migration to two We also applied our beam migration to a 3D marine data set. The
synthetic data sets: Sigsbee and Marmousi. Figure 5 shows the re- single-arrival Kirchhoff migration showed strong artifacts in the
flectivity map of the 2D Sigsbee data set. Note the faults and point subsalt (shown in Figure 11) because the salt dome caused multi-
diffractors (at a depth of 25,000 ft) in the subsalt region. Figure 6 pathing that was not handled well by single-arrival migration. In
shows the result of single-arrival (the most energetic arrival) contrast, the beam migration provided a better and cleaner image
Kirchhoff migration. Due to the single-arrival limitation, faults (shown in Figure 12) because of its capability of correctly handling
are poorly defined, diffractors are not well resolved, and strong ar- multipathing energy through the salt dome. The side-by-side com-
tifacts are present in the subsalt. In contrast, multiarrival Kirchhoff parison of the zoomed section is shown in Figure 13.

Figure 5. Reflectivity map of the Sigsbee data set.

Note the faults and point diffractors (at 25,000 ft
depth) in the subsalt region.

Figure 6. Single-arrival Kirchhoff migration of

the Sigsbee data set of Figure 5. Note that faults
are poorly defined, diffractors are not well re-
solved, and strong artifacts are present in the
subsalt.
 WB114 Liu and Palacharla

Figure 7. Multiarrival Kirchhoff beam migration of the

Sigsbee data set of Figure 5. In comparison to Figure 6,
note the much cleaner image in the subsalt, and the better
defined faults and diffractors.
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 8. Velocity model of the Marmousi data set. The

strong velocity contrasts in the central part are likely to
produce multiple arrivals.

Figure 9. Single-arrival Kirchhoff migration of the

Marmousi data set of Figure 8. The anticlines in the cen-
tral deep part (highlighted) are blurred and distorted.
 Multiarrival Kirchhoff beam migration WB115

Figure 10. Multiarrival Kirchhoff-beam migration of the

Marmousi data set of Figure 8. In contrast to Figure 9, the
multiarrival Kirchhoff beam migration is able to resolve
the anticlines.
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 11. Single-arrival Kirchhoff migration of the real

data set. The migration artifacts deteriorate the image in
the highlighted target zone.

Figure 12. Multiarrival Kirchhoff-beam migration of the

real data set of Figure 8. In contrast to Figure 11, the
multiarrival Kirchhoff beam migration is able to provide
better image in the target zone.
 WB116 Liu and Palacharla
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 13. The side-by-side comparison of Figure 11 and Figure 12, in the zoomed section: (a) the result of single-arrival Kirchhoff migration;
(b) the result of multiarrival Kirchhoff-beam migration.

CONCLUSIONS where f is ordinal frequency, x is the spatial variable, xm is beam

center, a is a constant scale to be determined, and g is a smooth
We incorporated beam technology into Kirchhoff migration and function. Then for any x and f , we have
developed a multiarrival Kirchhoff beam migration method that pro-
vides better image quality in regions with complex velocity struc- X
∞
a X ∞
tures. The theory and algorithm of our beam migration are analogs Gðx; xm ; f ÞΔxm ¼ pffiffiffi gðf ðx − xm Þ2 ÞΔxm
to Gaussian beam migration, but we focus on attaining kinematic f m¼−∞
m¼−∞
Z ∞ Z ∞ (A-2)
accuracy and implementation efficiency. The method was success- a 2 2
fully applied to the 2D synthetic data sets (2D Sigsbee and ≈ pffiffiffi gðf ðx − xm Þ Þdxm ¼ a gðξ Þdξ.
f −∞ −∞
Marmousi2) and 3D field data example, and produced images that
are superior to those from conventional single-arrival Kirchhoff mi- pffiffiffi
gration. The optimal choice of the beam width reduces computa- In the last step, changing variable ξ ¼ f ðx − xm Þ is used. The last
tional cost and error in traveltime extrapolation from a central term of equation A-2 is independent of x and f and equals 1 if 1a ¼
ray. Unlike conventional Kirchhoff migration, data input to multi- ∫∞ 2
−∞ gðξ Þdξ is chosen, which proves the partition of unity for any

arrival Kirchhoff beam migration should be regularized to achieve type of taper function defined in equation A-1. A suggested taper
accurate local slant stacking. function (cosine square) is

8  
< pa ffi cos2 π2 Vf min f2 ðx − xm Þ2
APPENDIX A Gðx; xm ; f Þ ¼
f avg
; (A-3)
:
PARTITION OF UNITY OF TAPER FUNCTION 0; if f min f
V avg 2
ðx − xm Þ2 ≥ 1

Suppose a taper function, G, is frequency dependent in such a

way: where f min is the reference minimum ordinal frequency, and V avg is
the average velocity. In the 3D case, a taper function can be repre-
sented by the product of two 2D taper functions shown in equa-
a
Gðx; xm ; f Þ ≡ pffiffiffi gðf ðx − xm Þ2 Þ; (A-1) tion A-1 (one is in crossline direction and the other one is in
f inline direction).
 Multiarrival Kirchhoff beam migration WB117

APPENDIX B APPENDIX C
DERIVATION OF EQUATION FOR TRAVELTIME DERIVATION OF FORMULA OF BEAM WIDTH
EXTRAPOLATION
When the velocity is depth-only dependent, equations 3 and 4
The wavefront curvature in 2D dynamic ray tracing can be become
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

computed by
dx
dQ ¼ VðzÞpx (C-1)
¼ VP (B-1) ds
ds

dz
dP 1 ∂2 V ¼ VðzÞpz . (C-2)
¼ − 2 2 Q; (B-2) ds
dQ V ∂n
Where px is a constant and
where s and n are ray-centered coordinates (Hill, 1990). In
equation 7, mr is evaluated by (Červený et al., 1982) sin θ0
px ¼ . (C-3)
  V0
P
mr ¼ Re : (B-3)
Q Integrating equations C-1 and C-2, we have
Z
If the velocity field is sufficiently smooth in the direction of n sin θ0
(perpendicular to the raypath), the second-order derivatives in x ¼ px vds ¼ σ; (C-4)
V0
equation B-2 will be ignored, and we have
Z and
Q ¼ Q0 þ P0 Vds (B-4)
ray Z
cos θ0
z¼ pz Vds ≈ σ. (C-5)
and P ¼ P0 , where P0 is the initial value of P and Q0 is the initial V0
value of Q. The convenient choices of P0 and Q0 (see Hill, 1990) are
This approximation is valid if the angle is small or the velocity
V avg 2 change is mild. Using the above two equations, we have
Q0 ¼ cos2 ðθ0 Þ (B-5)
2πV 0 f min
∂x cos θ0
¼ σ; (C-6)
∂θ0 V0
i
P0 ¼ : (B-6)
V0
∂z sin θ0
≈− σ; (C-7)
Here we use the initial beam width as ∂θ0 V0

V avg
w0 ¼ cosðθ0 Þ
2πf min
: (B-7)   sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2
 ∂x  ∂x ∂z σ
 ¼ þ ≈ : (C-8)
 ∂θ  ∂θ ∂θ v
Compared to Hill’s result, we have added the extra factor, cosðθ0 Þ 0 0 0 0
for w0 in equation B-7. A smaller value of w0 will increase the beam
width more rapidly as the beam propagates. For far angles (large This proves equation 11.
θ0 ), the beam width should increase more rapidly with propagation
than it does for near angles due to the bigger separation of raypaths ACKNOWLEDGMENTS
(Δθ0 ¼ V 0 Δp∕ cosðθ0 Þ). Substituting the values of P and Q into
equation B-2, and by using the definitions in equations 8 and 9, The authors thank ExxonMobil for permission to publish these
we have results. We also wish to acknowledge Peter Traynin for his provid-
ing 2D synthetic data sets.
     
P0 i i
mr ¼ Re ¼ Re ¼ Re REFERENCES
Q0 þ P0 σ V 0 Q0 þ iσ σ 0 þ iσ
σ Červený, V., M. M. Popov, and I. Pšenčík, 1982, Computation of wave fields
¼ 2 : in inhomogeneous media — Gaussian beam approach: Journal of the
σ þ σ0 2 Royal Astronomical Society of Canada, 70, 109-128, doi: 10.1111/
j.1365-246X.1982.tb06394.x.
Gray, S. H., C. Notfors, and N. Bleistein, 2002, Imaging using multi-
This proves equation 7. arrivals: Gaussian beams or multi-arrival Kirchhoff?: 72nd Annual
 WB118 Liu and Palacharla

International Meeting, SEG, Expanded Abstracts, 1117-1120, doi: Xu, S., H. Chauris, G. Lambare, and M. Noble, 2001, Common-angle
10.1190/1.1816843. migration: A strategy for imaging complex media: Geophysics, 66,
Hill, N. R., 1990, Gaussian beam migration: Geophysics, 55, 1416-1428, 1877-1894, doi: 10.1190/1.1487131.
doi: 10.1190/1.1442788. Zhu, T., S. H. Gray, and D. Wang, 2007, Prestack Gaussian-beam depth
Hill, N. R., 2001, Prestack Gaussian-beam depth migration: Geophysics, 66, migration in anisotropic media: Geophysics, 72, no. 3S133-S138,
1240-1250, doi: 10.1190/1.1487071. doi: 10.1190/1.2711423.
Downloaded 04/23/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/