Documente Academic
Documente Profesional
Documente Cultură
doi: 10.1111/1365-2478.12161
ABSTRACT
The next generation of seismic imaging algorithms will use full wavefield migration,
which regards multiple scattering as indispensable information. These algorithms
will also include autonomous velocity-updating in the migration process, called joint
migration inversion. Full wavefield migration and joint migration inversion address
industrial requirements to improve the images of highly complex reservoirs as well as
the industrial ambition to produce these images more automatically (automation in
seismic processing).
In these vision papers on seismic imaging, full wavefield migration and joint migration inversion are formulated in terms of a closed-loop, estimation algorithm that
can be physically explained by an iterative double-focusing process (full wavefield
Common Focus Point technology). A critical module in this formulation is forward
modelling, allowing feedback from the migrated output to the unmigrated input
(closing the loop). For this purpose, a full wavefield modelling module has been
developed, which uses an operator description of complex geology. Full wavefield
modelling is pre-eminently suited to function in the feedback path of a closed-loop
migration algorithm.
The Future of Seismic Imaging is presented as a coherent trilogy of papers that
propose the migration framework of the future. In Part I, the theory of full wavefield
modelling is explained, showing the fundamental distinction with the finite-difference
approach. Full wavefield modelling allows the computation of complex shot records
without the specification of velocity and density models. Instead, an operator description of the subsurface is used. The capability of full wavefield modelling is illustrated
with examples. Finally, the theory of full wavefield modelling is extended to full wavefield reverse modelling (FWMod1 ), which allows accurate estimation of (blended)
source properties from (blended) shot records.
INTRODUCTION
In standard migration practice, we have little information
about the inconsistency between output and input: migration
is implemented as an open-loop process. In particular, if we
want to use the information in multiple scattering, a simple
open-loop approach is no longer acceptable. By taking the
open-loop seismic image as the input in a forward modelling
algorithm, we are able to close the loop in migration, so that
we generate numerically simulated measurements in the feed E-mail:
C
back path. Next, iterative minimization of the difference between simulated and real measurements allows us to optimize
the seismic image (see the basic diagram in Fig. 1). Multiples
are an integral part of this process. As I will explain later,
multiples are a blended wavefield phenomenon, the blended
sources being natural. Full wavefield migration (FWM), therefore, is also the obvious solution to migrate manmade blended
shot records.
In closed-loop FWM, forward modelling is a critical process. Errors in the modelling result must be avoided because
they are transferred to errors in the residue (i.e., the difference
a.j.berkhout@tudelft.nl
911
Figure 1 Migration is formulated as a closed-loop process, so that output and input are connected via a feedback loop with a forward modelling
module that transforms subsurface reflectivities into simulated measurements. The residue steers the next iteration.
Figure 2 Wavefields in the same subsurface, but with two fundamentally different descriptions. Although the operator description is different
from the property description, FWMod modelling and FinDif modelling generate the same wavefields. In closed-loop migration, the FinDifrelated property description must be replaced by the FWMod-related operator description.
between measured and simulated data), and therefore, errors may be introduced in the migration output. Today,
finite-difference modelling yields excellent results, but the
subsurface must be described in terms of detailed elastic properties (Moczo et al. 2007). Such a description is outside the
migration framework and, therefore, should not be used at
this stage of seismic processing. In this seismic trilogy on the
future of migration, the usual property description of the subsurface in terms of a detailed velocity and density model is
replaced by an alternative description that makes use of local propagation and scattering operators. The consequence
is that property-driven modelling (in which the algorithm is
based on a differential equation for seismic wavefields) can
C
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
(1a)
M
j+ (zn ; z0 ).
W (zm, zn ) R (zn , zn ) P
(1b)
n=m+1
Vector-matrix equations (1a) and (1b) formulate the simplest version of the discrete scattering integral (first-order
WRW-model). In equations (1a) and (1b), the elements of
C
Figure 3 In primary wavefield migration, there is no gridpoint interaction, so that i) scattering is upward only (no multiples) and ii)
propagating wavefields are not influenced by the scattering process
(no transmission effects). Because of these simplifications, the scattered wavefields are discontinuous in primary wavefield migration,
leading to artefacts in the seismic image.
vector S +j (z0 ) represent the (blended) source array with identification label j at the surface z0 , matrix W+ (zm, z0 ) represents the downward propagation operator between depth
levels z0 and zm, matrix R (zn , zn ) represents the angledependent reflection operator at zn (operator R (zn , zn ) trans j+ (zn ; z0 ) into upgoing wavefield
forms downgoing wavefield P
j+ (zn ; z0 ) by an elastic reflection process), and maR (zn , zn ) P
m
W zn1 , zn and
(1c)
W+ zn , zn1 ,
(1d)
n=1
W+ (zm, z0 ) =
1
n=m
where the columns of W zn1 , zn and W+ zn , zn1 are determined by the local velocities.
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
(2a)
M
j+ (zn ; z0 ),
W (zm, zn ) R (zn , zn ) P
(2b)
n=m+1
(4a)
with
j (z0 ; z0 ) ,
+j (z0 ; z0 ) = S +j (z0 ) + R (z0 , z0 ) P
Q
(4b)
where matrix R (z0 , z0 ) is the reflectivity operator at the sur j (z0 ; z0 ) into downface that transforms upgoing wavefield P
going wavefield R (z0 , z0 ) P j (z0 ; z0 ).
Using equations (4a) and (4b) , the following family of
extended forward models for PWM can be formulated:
a. for the total response (primaries + surface multiples):
+j (z0 ; z0 ) ;
j (z0 , z0 ) = X0 (z0 , z0 ) Q
P
(5a)
(5b)
0,
M
j (z0 ; z0 ) = X0 (z0 , z0 ) R (z0 , z0 ) P j (z0 ; z0 ) .
(5c)
M
(6a)
m=1
m1
I + T (zn , zn )
n=1
W (zn , zn+1 )
(3a)
and
1
I + T+ (zn , zn )
W+ (zm, z0 ) = W+ zm, zm1
+
n=m1
zn , zn1 .
(3b)
C
M
(6b)
m=1
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
and Q
+ ) at the surface z0 . In primary wavefield migration, it
Figure 4 The feedback model, showing the up- and downgoing wavefields ( P
j
j
+ = S + . Additionally, for linearization purposes,
is assumed that the surface-related multiples have been removed from the input, so that Q
j
j
internal multiples are neglected.
simple: replace downgoing source wavefield S +j (z0 ) by down j (z0 ; z0 ), and replace
going reflected wavefield R (z0 , z0 ) P
primary response P0, j (z0 ; z0 ) by surface multiple response
0,
M
j (z0 ; z0 ) .
Importantly, in surface multiple migration, the source
wavelet does not need to be known: both input and output are
given by the measured data (compare equations (5b) and (5c)).
In Part II, it will be shown that FWM can be used to migrate
primaries, surface multiples as well as internal multiples. This
all multiple option in migration demonstrates the problems
with traditional approaches in making large investments to
removing multiples. In fact, as will be explained in Part III,
the best primarymultiple separation is achieved by the full
wavefield migration algorithm.
m1
j (zn ; z0 );
W+ (zm, zn ) R (zn , zn ) P
(7a)
n=0
M
j+ (zn ; z0 ),
W (zm, zn ) R (zn , zn ) P
n=m+1
C
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
(7b)
where the hybrid propagation operators W include the transmission effects; see recursive expressions (3a) and (3b). By
comparing full wavefield model equations (7a) and (7b) with
primary model equations (2a) and (2b), we see the addition
of a second term in each equation.
The second term in expression (7a) has large consequences: it introduces at each depth level an extra reflection
process (quantified by reflection operator R ), which generates the surface (n = 0) and the internal multiples (n > 0). The
first term in expression (7b) includes the response of the lower
half-space z > zM . If in (7b) we choose m = 0, then we obtain
the upgoing wavefield in the reflection measurements at z0 . If
in (7a) we choose m = M, then we obtain the downgoing
wavefield in the transmission measurements at zM .
In the following, we will not make use of full wavefield
equations (7a) and (7b). Instead, we will take the transmission
operators [I + T ] outside the hybrid propagation operators W , and we assign to T+ and T the role of forward
scattering operators (compare these with backward scattering operators R and R ). By doing this, we reintroduce in
our full wavefield model the scatter-free propagation operators W (expressions (1c) and (1d)), and we use both T+ ,
T and R , R as scattering operators. This leads to the preferred scattering formulation that will be the basis of all full
wavefield algorithms in Part II and Part III (Berkhout 2012):
a. for the downgoing wavefields (m = 1, 2, . . . , M):
+
j+ zm
; z0 = W+ zm
, z0 S j (z0 )
P
+
m1
+ + +
W+ zm
, zn S j zn ; z0 ;
(8a)
+
zm
; z0
= W
+
+
zm
, zM
M
j
P
(9a)
and
+
j zn ; z0
S +j zn+ ; z0 = R zn+ , zn+ P
+
j zn ; z0 ,
+ T+ zn+ , zn P
(9b)
where S +j = S j and T = R , T+ = R if we can neglect wave conversion at zn . Equations (9a) and (9b) represent
the combined reflection (first term) and transmission process
(second term) at depth level zn (see Fig. 6).
We will see that the proposed scattering formulation of
the FWMod equations (8a) and (8b) (in which propagation
operators W are free of scattering, and scattering operators
(T, R) are free of propagation) is critical to migrate accurately the seismic response from complex geology. Later in
this paper, it will be shown that equations (8a) and (8b) can
be easily extended to the multi-mode situation, so that both
P- and S-waves are simultaneously taken into account to complement the PP-image with the SP-, PS-, and SS-images (from
one to four subsurface images). Again, in this multi-mode situation, propagation will be scatter-free, and scattering will be
propagation-free.
k+ zm, zm1 Pkj+ zm1 ; z0
W
(zM ; z0 )
+
k+ zm, zm1 Skj
zm1 ; z0 .
W
(10a)
+
W zm
, zn S j zn ; z0 ,
(8b)
n=m+1
C
+
j zn ; z0
S j zn ; z0 = R zn , zn P
+
j zn ; z0
+ T zn , zn+ P
n=0
j
P
In equation (10a), the first term describes wave propagation according to the classical Huygens principle. In gridpoint
k at depth level zm1 , wavefield sample Pkj+ acts as a one-way
secondary point source. Its wavefield is propagated by oper k+ to depth level zm. Similarly, in the second term of
ator W
+
equation (10a), wavefield sample Skj
also acts as a one-way
+
as an extended
secondary point source. I will refer to Skj
Huygens source. Both Huygens sources radiate downward.
For a homogeneous gridpoint, the extended Huygens source
is zero. Using full wavefield forward model equation (8b), we
can write
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
Figure 6 The elastic PP-scattering process at gridpoint k, showing the scattered wavefields S
j in full wavefield migration. Note that in primary
wavefield migration, the reflection operator R is set to zero (no multiples) and the transmission operators T are deleted (transmission effects
are ignored).
j (zm; z0 ) =
P
k zm, zm+1 Pkj zm+1 ; z0
W
and
+ + +
k zn , zn Pkj zn ; z0
S +j zn+ ; z0 =
R
k zm, zm+1 Skj
zm+1 ; z0 .
W
(10b)
+ Tk+ zn+ , zn Pkj+ zn ; z0 ,
(10d)
functions
Again, in the first term, wavefield sample
as a secondary point source according to the classical Huy
represents an extended
gens principle; the second term Skj
Huygens source. Both Huygens sources radiate upward.
The extended Huygens source is zero for a homogeneous
gridpoint.
In Part II, we will see that the extended Huygens sources
can be found in the CFP-gathers, each of which is obtained by
full wavefield focusing at detection. By a minimization process on the CFP-gathers, the Huygens sources are transformed
into the scattering operators, which represent full wavefield
focusing at emission.
+ Tk zn , zn+ Pkj zn+ ; z0
(upward radiating, see equation (9a))
C
(10c)
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
(a)
(b)
+1000
-1000
Cp=2500
Cs=0.000
=2300
Cp =2500
Cs =1500
=2300
Rk
Rk
amplitude
amplitude
Rk
Rk
Rk
Rk
Rk
Rk
(d)
-1000
Rk
Rk
Rk
amplitude
amplitude
Rk
Rk
Cp=2500
Cs=1500
=2300
Rk
Rk
+1000
Cp=2000
Cs= 800
=1800
depth (m)
depth (m)
Cp=2500
Cs=1500
=2300
Rk
-1000
+1000
Cp =2000
Cs =1200
=1800
Rk
Rk
(c)
Rk
Rk
Rk
+1000
Cp =2000
Cs =1500
=1800
depth (m)
depth (m)
Cp =2000
Cs =0.000
=1800
-1000
Rk
Rk
Rk
Figure 7 The difference between reflectivity matrices R and R at one reflector gridpoint. For small offsets and small shear contrasts,
R R , but for large shear contrasts and/or large offsets, the difference becomes significant. As expected, around the critical angle, the
difference is always large.
C
Similarly, if we combine the interpretation of the secondary source arrays with equation (8a), we can conclude
that seismic shot records at zM consist of blended wavefields, in which the blending is natural: blended source array S +j consists of coded source elements at each subsurface
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
j
2. P
zm
; z0
+ + +
j zm1 ; z0
= W+ zm
, zm1 Q
(11a)
j = X0 S +j + X0 R P
j at z0 .
P
+
j zm+1
j zm+1
; z0 = P
; z0 + S j zm+1
; z0
1. Q
(12a)
+
+
j zm+1 ; z0 ,
j zm
; z0 = W zm
, zm+1 Q
2. P
(12b)
(14a)
(11b)
C
j
P
(1)
(1) +
= X0
S j at z0
(14b)
j
P
(2)
X0
(2)
(1)
(2)
j
at z0 ,
S +j + X0 R P
(14c)
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
Figure 8 Illustration of the forward modelling concept in full wavefield migration and joint migration inversion, showing that the response of
a single reflector represents an interference pattern of gridpoint responses, where each gridpoint response is generated by one coded secondary
P + ). Note that the directivity of this source element is given by the angle-dependent reflection property in gridpoint k.
source element ( R
k kj
C
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
Figure 9 Forward extrapolation of the upgoing wavefields (a) and the downgoing wavefields (b) in depth layer (zm1 , zm), where S
j is the
physical scattering at the layer boundaries, and the columns of W are the local propagation operators inside the layer.
Figure 10 Computational diagram for the full wavefield modelling algorithm FWMod, which transforms scattering operators into full wavefields.
Each roundtrip adds one order of scattering, starting with order 1 (primaries only).
provide a desirable solution, because sparsity does not represent the property of a real Earth. Dynamic thresholding in
FWMod is an interesting approach to dealing with crosstalk
in FWM and JMI without making any assumption on a
sparse end result. By using an automatic thresholding process
on the reflectivities, we can obtain a first iteration that shows
the primary GPRs with only the largest reflectivities. In the
next iterations, the threshold is lowered step-by-step, so
C
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
Figure 11 Example of full wavefield modelling (FWMod). In each roundtrip, a new order of multiple scattering is generated. This extra data is
used in the next full wavefield migration (and joint migration inversion) iteration, helping to steer the solution to the correct minimum.
multiples is generated. Hence, the strongest reflectors automatically generate the highest-order multiples. Note the interesting property of the hierarchical modelling strategy: the
strongest GPRs, together with their surface and internal multiples, automatically have priority in the migration process. In
C
FWM and JMI, their responses are subtracted before estimating the weaker ones, and so on. This strategy acts as a type
of L1 constraint in L2-minimization (Daubechies, Defrise and
de Mol 2004). However, hierarchical FWMod brings parameter selection outside the constrained L2-minimization box,
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
Figure 12 Example of hierarchical full wavefield modelling. In each roundtrip, new primaries are included, and a new order of multiple scattering
is generated (compare with Fig. 11). Note that response 1 is not shown (which is only the shallowest reflection).
Figure 13 The coda of a reflective overburden is seriously masking the reservoir response. A hierarchical imaging strategy in full wavefield
migration (and joint migration inversion) solves this problem.
giving the user full control, and any smart selection process
can then be implemented. For instance, in addition to reflector
strength, priority may also be given to shallow versus deep.
This is of particular importance if we are dealing with a highly
reflective overburden, which causes a strong coda on top of
C
the deeper response. Figure 13 illustrates this notorious problem. The example shows that the response of the reservoir
is completely masked by the coda of the overburden. By giving higher priority to the strong reflectors of the overburden,
the crosstalk between the overburden coda and the reservoir
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
j zm; z0
j z0+ ; z0 = W z0+ , zm
I zm, zm P
P
+
m
W z0+ , zn S j zn ; z0 ,
(15a)
n=1
+
j zm
where I is the unity matrix and P
; z0 is the response
of the lower half-space (z > zm). From a physics point of
view, this equation describes forward propagation as a massive upward-moving defocusing process at the detector side,
starting at depth level zm and ending at the surface z0 (m = 1,
2, . . . , M). If we multiply equation (15a) by scatter-free focal
+ + +
, z0 = W zm, z0 , then we can write:
operator F+ zm
+
+ + + +
j z0 ; z0
j zm
; z0 = I zm
, zm F zm, z0 P
P
m
+
F+ zm
, zn S j zn ; z0 ,
(15b)
n=1
j z0+ ; z0 is known. Expression (15b) shows that rewhere P
verse modelling is a massive downward-moving focusing process at the detector side, starting at the surface z0 and ending
at depth level zm (m = 1, 2, . . . , M). Note that the summation
C
M1
W+ zM , zn+ S +j zn+ ; z0 .
(16a)
n=m
Again, from a physics point of view, this equation describes forward propagation as a massive downward-moving
defocusing process, now at the source side, starting at depth
level zm and ending at zM (m = M 1, M 2, . . . , 0).
If we multiply equation (16a) by scatter-free focal operator
+ +
, zM = W zm, zM , then we can write (m = M 1,
F zm
M 2, . . . , 0):
+ + +
j zM ; z0
j+ zm
; z0 = I zm
, zm F zm, zM P
P
M1
+ + +
F zm
, zn S j zn ; z0 ,
(16b)
n=m
j+ zM ; z0 is known. Expression (16b) shows again
where P
that reverse modelling is a massive focusing process, now at
the source side and moving upward, starting at maximum
depth level zM and ending at depth level zm; compare this with
(15b). Note that when we turn around at zM , the starting
j+ zM ; z0 , and when we turn around
wavefield is given by P
j z0+ ; z0 . The next
at z0 , the starting wavefield is again P
roundtrip makes use of improved estimates of the secondary
sources S j at each depth level zm (similar to FWMod) and
j+ z0 ; z0 .
source vector S +j (z0 ) = P
If we compare equation (15a) with (15b) and (16a) with
(16b), we see that the full wavefield forward modelling process
in each roundtrip of FWMod is replaced by the full wavefield
reverse modelling process in FWMod1 : application of W
followed by addition is replaced by application of F = W
followed by subtraction. Figure 14 shows the computational
diagram of FWMod1 (the combination of focusing and removal processes at the detector and source side as described in
equations (15b) and (16b), respectively). Similar to FWMod,
the algorithm is recursive (compare Fig. 14 with Fig. 10, see
also Appendix B).
Physically, one roundtrip of the FWMod1 process can be
j by one,
described by a decrease of the scattering order in P
so that after a roundtrip, the first-order response (primaries)
is transformed into an estimate of the (blended) source,
the second-order response is transformed into the first-order
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
Figure 14 Computational diagram for the full wavefield reverse modelling algorithm FWMod1 , which applies focusing by reverse extrapolation
and removes physical scattering by subtraction at both the detector and the source side. Each roundtrip transforms one order of scattering into
an update of S +
j z0 , starting with order 1 (primaries) in the first roundtrip. Compare with Fig. 10.
(17a)
(17c)
C
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
Figure 15 Computational diagram for the full wavefield redatuming process from z0 to z M , which is a combination of reverse modelling at the
detector side and forward modelling at the source side.
+
P+ zm
; z0 = W+ zm
, z0 S (z0 )
+
m1
+ + +
W+ zm
, zn S zn ; z0
n=0
C
(18a)
M
+
W zm
, zn S zn ; z0
n=m+1
(18b)
M1
+ + +
F zm
, zn S zn ; z0
n=m
(18c)
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
+
+
; z0 = F+ zm
, z0 P (z0 ; z0 )
P zm
m
+
F+ zm
, zn S zn ; z0
n=1
P =
Pp
P
s
, W =
Wpp 0
0 W
ss
(18d)
(19a)
and F = W
Sp
Rpp Rps
P+p
S =
=
S
Rsp Rss
P+
s
s
Tpp Tps
Pp
+
,
Tsp Tss
P
s
(19b)
S+ =
S+p
Rpp Rps
Pp
=
+
Ss
Rsp Rss
P
s
+
+
+
T pp T ps
Pp
+
.
+
+
Tsp Tss
P+
s
(19c)
C
CONCLUSIONS
In this paper, I have proposed a new way of forward modelling, called FWMod, which functions in the feedback path
of the closed-loop, iterative algorithms FWM and JMI. FWMod transforms the output of FWM and JMI into seismic
shot records that include surface and internal multiples (full
wavefield modelling). Optionally, converted waves may be
included in the modelling result.
The traditional FinDif-related presentation of the subsurface in terms of detailed velocity and density is outside
the framework of migration and, therefore, cannot be used in
closed-loop imaging. Instead, FWMod describes the subsurface in terms of propagation and scattering operators. These
operators are not specified by the user, but they are estimated
by FWM and JMI.
In FWMod, each inhomogeneous gridpoint (labelled k)
functions as a two-way secondary source (extended Huygens
source), the source properties being determined by the scat k, T
k] and the one-way incident wavetering operators [ R
+
fields [Pkj , Pkj ]. The combination of the classical and extended Huygens sources generates wavefields that properly
represent the nonlinear scattering properties of any complex
geology.
The phase relationship between gridpoint k and its neighbouring gridpoints is determined by the recursive propaga k ]. A direct relationship exists be k+ , W
tion operators [W
tween these recursive operators and the local velocity properties around gridpoint k. By keeping the scattering operators
outside the propagation operators (the big decoupling), we
k are scattering-free operators with
k+ and W
ensure that W
a unit spatial amplitude spectrum if we neglect inelastic absorption. This choice has far-reaching consequences for migration and inversion algorithms: scattering operators are
determined only by the amplitude properties, and propagation operators are determined only by the phase properties
of the seismic data. The result of this orthogonal property is
that the nonlinearity in migration and inversion is decreased
significantly.
By using the concept of secondary sources ( S j ), each
shot record can be considered to be the response of a blended
array of coded sources ( S j ), at each depth level the uncoded
sources being given by the scattering operators in each gridk], and the blending codes being given by the
k, T
point [ R
incident wavefields (Pkj , Pkj+ ) at these gridpoints.
As explained in this paper, FWMod offers great modelling flexibility and opens up new opportunities in closedloop seismic imaging. For instance, to minimize interference
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
EPILOGUE
In the proposed reformulation of seismic wave theory, the subsurface is not represented by the usual elastic parameters, but
is described in terms of decoupled propagation and scattering operators. The propagation operators are scattering-free
and determine the traveltime properties of the seismic measurements. The scattering operators are propagation-free and
determine the amplitude properties of the seismic measurements. Data-driven propagation operators contain anisotropy
and inelastic absorption; data-driven scattering operators contain angle-dependency and conversion losses.
The dual operator description (propagation, scattering)
allows an extension of the well-known Huygens principle by
C
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
Daubechies I., Defrise M. and de Mol C. 2004. An iterative thresholding algorithm for linear inverse problems with a sparsity constraint. Communications on Pure and Applied Mathematics 57,
14131457.
Lu S., Whitmore N.D., Valenciano A. and Chemingui N. 2011. Imaging of primaries and multiples with 3D SEAM synthetic. 81st SEG
meeting, Expanded Abstracts, 32173221.
Moczo, P., Robertsson, J.O.A. and Eisner L. 2007. The finitedifference time-domain method for modeling of seismic wave propagation. Advances in Geophysics 48, 421516.
Verschuur D.J. and Berkhout A.J. 2011. Seismic migration of blended
shot records with surface-related multiple scattering. Geophysics
76, A7A13.
APPENDIX A
TWO-WAY EXPRESSION OF EXTENDED
HUYGENS SOURCES
The expression of the extended Huygens sources in terms of
one-way wavefields is given by
+
j zn ; z0
S j zn ; z0 = R zn , zn P
+
j zn ; z0 ,
(A1a)
+ T zn , zn+ P
or, if we omit the depth level indication,
j+ + T P
j .
S j = R P
S j =
j+ + P
j ) is the two-way wavefield, and ( P
j+ P
j )
where ( P
is related to the vertical derivative of the two-way wavefield
(vertical depending on the coordinate system). Similarly, for
the downward-radiating version, we can write
1
j + P
j P
j+ + 1 R T+ P
j+ ,
R + T+ P
2
2
j + P
j P
j+ + R P
j+ .
(A2b)
= R+ P
S +j =
REFERENCES
Berkhout A.J. 1982. Seismic Migration, Imaging of Acoustic Energy
by Wavefield Extrapolation, A: Theoretical Aspects, 2nd edn. Elsevier.
Berkhout A.J. 2012. Combining full wavefield migration and full
waveform inversion, a glance into the future of seismic imaging.
Geophysics 77, S43S50.
Berkhout A.J. and Verschuur D.J. 1994. Multiple technology, part
2: migration of multiple reflections. 64th SEG meeting, Expanded
Abstracts, 14971500.
C
(A1b)
(A3a)
and
j P
j+ ),
S +j = R ( P
so that S +j = S j .
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930
(A3b)
APPENDIX B
C
(B1b)
+
+ +
+j zm
j zm+1 ; z0 .
Q
; z0 = F zm
, zm+1 P
(B2a)
with
+
+
+j zm+1
j+ zm+1
; z0 = Q
; z0 S + zm+1
; z0 .
P
(B2b)
Note F = [W ] . In full wavefield migration and joint migration inversion, FWMod and FWmod-1 are the basic computational modules (see Figs 10 and 14).
2014 European Association of Geoscientists & Engineers, Geophysical Prospecting, 62, 911930