Zhang weglein-2008

Direct non-linear inversion of 1D acoustic media using inverse scattering
subseries
H. Zhang†and A B. Weglein
†Presently at ConocoPhillips
Abstract
A task specific multi-parameter 1
direct non-linear inversion subseries of the inverse scattering
series is derived and tested for a velocity and density varying 1D acoustic medium. There are
various linear estimate solutions in the literature ( e.g., Raz, 1981, Clayton and Stolt, 1981, Stolt
and Weglein, 1985) that assume an adequate estimate of medium properties above any given
target reflector. However, this is the first seismic inversion method that: (1) neither assumes
nor requires an adequate estimate of medium properties above any reflector, and (2) does not
assume that the changes in physical properties satisfies a linear relationship to the reflection data
at the target, and (3) the most importantly the method stands alone in being a direct inversion.
The meaning of direct is that there are formulas that explicitly solve for and directly output
these physical properties, without e.g., search algorithms and optimization schemes, and proxies
that typically characterize indirect methods. Numerical test results indicate that one term
beyond linear provides added value beyond standard linear techniques and common practice for
estimating changes in physical properties at a target. Imaging and inversion for a two parameter
medium directly in terms of data and reference properties is much more complicated than that of
the one parameter case. The message delivered from this study extends and progresses beyond
the earlier one communicated in a one parameter velocity only medium, and serves as a necessary
and critical step in, and guide for, the development of the yet more complicated and realistic
elastic isotropic direct depth imaging and non-linear parameter estimation. Three important
issues are identified and/or further progressed within the multi-parameter acoustic context and
scope of this paper: (1) the concept of purposeful perturbation, (2) how the inverse series
addresses the phenomena in linear inversion known as leakage, and (3) special parameter for
linear inversion that is immune to linear inversion leakage, and the very significant implication of
the latter result for direct depth imaging without the velocity model, are presented and discussed
by analyzing these new two parameter non-linear direct inversion formulas and methods.
Introduction
The objective of seismic exploration is to determine the location (imaging) and mechanical proper-
ties (inversion) of subsurface targets to identify hydrocarbon resources in the earth using recorded
data. The inverse scattering series has a tremendous generality and comprehensiveness allow-
ing many distinct traditional processing objectives to be achieved within a single framework, but
without the traditional need to provide information about the properties that govern actual wave
propagation in the earth. It begins with scattering theory, which is the relationship between the
1
Within the context and scope of this paper, a multi-parameter medium is an acoustic medium where the velocity
and/or density can vary.
184

Direct non-linear inversion of 1D acoustic media using inverse scattering subseries MOSRP07
perturbation or alteration in a medium’s properties and the concomitant perturbation or change
in the wave field. The relationship between those two changes is always non-linear. Any change
in a medium will result in a change in the wave-field that is non-linearly related to that physical
property change.
In this paper we examine the relationship between the perturbation in a medium and the pertur-
bation in a wave field for the case of a 1D variable velocity and variable density acoustic medium.
We assume the original unperturbed medium is a homogeneous whole-space. We further assume
that free surface and internal multiples have been removed (see, e.g., Weglein et al., 2003). And
we assume that we are recording primaries, and our objectives are to: (1) locate reflectors and (2)
determine medium properties of the actual medium. In this paper we present: (1) the first deriva-
tion of equations to directly achieve those two distinct objectives for a one dimensional velocity and
density varying acoustic medium, and (2) we then reduce this general formalism to the special case
of a single horizontal reflector, where the acoustic medium above the reflector is known, but the
objective is to determine the acoustic properties of the half-space below the reflector. For the latter
single reflector case, the recorded data have a non-linear relationship with the property changes
across this reflector. Current inversion methods include: (1) the linear approximation (e.g., Clayton
and Stolt, 1981; Weglein and Stolt, 1992) which is often useful, especially in the presence of small
earth property changes across the boundary and/or small angle reflections, and (2) indirect model
matching methods with global searching (e.g., Tarantola et al., 1984; Sen and Stoffa, 1995) which
define an objective function assumed to be minimized when the best fitting model is obtained.
The assumptions of the former methods (like the small contrast assumptions) are often violated in
practice and can cause erroneous predictions; the latter category usually involves a significant and
often daunting computation effort (especially in multi-D cases) and/or sometimes have reported
erroneous or ambiguous results.
In this paper, a more comprehensive multi-parameter multi-dimensional direct non-linear inversion
framework is developed based on the inverse scattering task-specific subseries (see, e.g., Weglein
et al., 2003). In order to provide more accurate and reliable target identification especially with
large contrast, large angle target geometry, we isolated the inverse scattering subseries responsible
for non-linear amplitude inversion of data.
The original inverse scattering series research aimed at separating imaging and inversion tasks on
primaries was developed for a 1D acoustic one parameter case (constant density medium, only
velocity variable in depth) and a plane wave at normal incidence (Weglein et al., 2002; Shaw et al.,
2003). In this paper we move a step closer to seismic exploration relevance by extending that earlier
work to a multi-parameter case — two parameter case (velocity and density vary vertically in depth)
and allowing for point sources and receivers over a 1D acoustic medium. Clayton and Stolt (1981)
gave a two parameter linear inversion solution for 2D acoustic media (velocity and density vary
both vertically and laterally). In this paper, we use the same parameters but concentrate on 1D
acoustic media to derive the direct non-linear inversion solution. In the application of the direct
non-linear inverse algorithm, we move one step each time (e.g., from one parameter 1D acoustic
case to two parameter 1D acoustic case, or to one parameter 2D acoustic case, instead of ‘jumping’
directly to two parameter 2D acoustic case) so that we can solve the problem step by step and
learn lessons from each step which would guide us to step further towards our goal of greater
185

realism and increased reliable prediction. For one parameter 1D and 2D acoustic media, some work
on direct non-linear imaging with reference velocity is presented by Shaw (2005) and Liu et al.
(2005). It has been shown in this paper that imaging and inversion for two parameter medium
are much more complicated compared to one parameter case, although it seems like just simply
adding one parameter. Examples of the new inverse issues that arise in a two parameter world (and
needed responses) that have no one parameter analogue are leakage, purposeful perturbation for
that issue, and the identification of the special parameter for inversion that avoids leakage, and the
conceptual insights that this understanding provides for our campaign to address pressing imaging
and inversion challenges.
For the direct non-linear inversion solution obtained in this paper, the tasks for imaging-only and
inversion-only terms are separated. Tests with analytic data indicate significant added value for
parameter predictions, beyond linear estimates, in terms of both the proximity to actual value and
the increased range of angles over which the improved estimates are useful.
A closed form of the inversion terms for the one-interface case is also obtained. This closed form
might be useful in predicting the precritical data using the postcritical data.
A special parameter ∆c (∆c = c−c0) (P-wave velocity change across an interface) is also found. Its
Born inversion (∆c)1 always has the right sign. That is, the sign of (∆c)1 is always the same as that
of ∆c. In practice, it could be very useful to know whether the velocity increases or decreases across
the interface. After changing parameters, from α (relative changes in P-wave bulk modulus) and β
(relative changes in density) to velocity and β, another form of the non-linear solution is obtained.
There is no leakage correction (please see details in the section on three important messages) in
this solution. This new form clearly indicates that the imaging terms care only about velocity
errors. The mislocation is due to the wrong velocity. This is suggestive of possible generalization
to multi-D medium, and also of possible model-type independent imaging which only depends on
velocity changes.
The following section is a brief introduction of the inverse scattering subseries. We then gave the
one dimensional multi-parameter acoustic derivation in detail, and that is followed by the numerical
tests for the single reflector case. We also provided a further discussion about the special physical
non-leaking acoustic parameter.
Inverse scattering subseries
Scattering theory relates the perturbation (the difference between the reference and actual medium
properties) to the scattered wave field (the difference between the reference medium’s and the
actual medium’s wave field). It is therefore reasonable that in discussing scattering theory, we
begin with the basic wave equations governing the wave propagation in the actual and reference
medium, respectively 2,
LG = δ, (1)
2
In this introductory math development, we follow closely Weglein et al. (1997); Weglein et al. (2002); Weglein
et al. (2003).
186

L0G0 = δ, (2)
where L and L0 are respectively the differential operators that describe wave propagation in the
actual and reference medium, and G and G0 are the corresponding Green’s operators. The δ on
the right hand side of both equations is a Dirac delta operator and represents an impulsive source.
The perturbation is defined as V = L0 − L. The Lippmann-Schwinger equation,
G = G0 + G0V G, (3)
relates G, G0 and V (see, e.g., Taylor, 1972). Iterating this equation back into itself generates the
forward scattering series
G = G0 + G0V G0 + G0V G0V G0 + · · · . (4)
Then the scattered field ψs ≡ G − G0 can be written as
ψs = G0V G0 + G0V G0V G0 + · · ·
= (ψs)1 + (ψs)2 + · · · , (5)
where (ψs)n is the portion of ψs that is nth order in V . The measured values of ψs are the data,
D, where
D = (ψs)ms = (ψs)on the measurement surface.
In the inverse scattering series, expanding V as a series in orders of D,
V = V1 + V2 + V3 + · · · , (6)
where the subscript “i” in Vi (i=1, 2, 3, ...) denotes the portion of V i-th order in the data.
Substituting Eq. (6) into Eq. (5), and evaluating Eq. (5) on the measurement surface yields
D = [G0(V1 + V2 + · · · )G0]ms + [G0(V1 + V2 + · · · )G0(V1 + V2 + · · · )G0]ms + · · · . (7)
Setting terms of equal order in the data equal, leads to the equations that determine V1, V2, . . .
directly from D and G0.
D = [G0V1G0]ms, (8)
0 = [G0V2G0]ms + [G0V1G0V1G0]ms, (9)
0 =[G0V3G0]ms + [G0V1G0V2G0]ms + [G0V2G0V1G0]ms
+ [G0V1G0V1G0V1G0]ms, (10)
etc. Equations (8) ∼ (10) permit the sequential calculation of V1, V2, . . ., and, hence, achieve
full inversion for V (see Eq. 6) from the recorded data D and the reference wave field (i.e., the
Green’s operator of the reference medium) G0. Therefore, the inverse scattering series is a multi-
D inversion procedure that directly determines physical properties using only reflection data and
reference medium information.
187

Derivation of α1, β1 and α2, β2
In this section, we will consider a 1D acoustic two parameter earth model (e.g. bulk modulus and
density or velocity and density). We start with the 3D acoustic wave equations in the actual and
reference medium (Clayton and Stolt, 1981; Weglein et al., 1997)
ω2
K(r)
+ ·
1
ρ(r)
G(r, rs; ω) = δ(r − rs), (11)
ω2
K0(r)
+ ·
1
ρ0(r)
G0(r, rs; ω) = δ(r − rs), (12)
where G(r, rs; ω) and G0(r, rs; ω) are respectively the free-space causal Green’s functions that
describe wave propagation in the actual and reference medium. K = c2ρ, is P-wave bulk modulus,
c is P-wave velocity and ρ is the density. The quantities with subscript “0” are for the reference
medium, and those without the subscript are for the actual medium. The perturbation is
V = L0 − L =
ω2α
K0
+ ·
β
ρ0
, (13)
where α = 1 − K0
K and β = 1 − ρ0
ρ are the two parameters we choose to do the inversion. Assuming
both ρ0 and c0 are constants, Eq. (12) becomes
ω2
c2
0
+ 2
G0(r, rs; ω) = ρ0δ(r − rs). (14)
For the 1-D case, the perturbation V has the following form
V (z, ) =
ω2α(z)
K0
+
1
ρ0
β(z)
∂2
∂x2
+
1
ρ0
∂
∂z
β(z)
∂
∂z
. (15)
V (z, ), α(z) and β(z) can be expanded respectively as
V (z, ) = V1(z, ) + V2(z, ) + · · · , (16)
α(z) = α1(z) + α2(z) + · · · , (17)
β(z) = β1(z) + β2(z) + · · · . (18)
Where the subscript “i” in Vi, αi and βi (i=1, 2, 3, ...) denote the portion of those quantities i-th
order in the data.Then we have
V1(z, ) =
ω2α1(z)
K0
+
1
ρ0
β1(z)
∂2
∂x2
+
1
ρ0
∂
∂z
β1(z)
∂
∂z
, (19)
V2(z, ) =
ω2α2(z)
K0
+
1
ρ0
β2(z)
∂2
∂x2
+
1
ρ0
∂
∂z
β2(z)
∂
∂z
, (20)
....
188

Substituting Eq. (19) into Eq. (8), we can get the linear solution for α1 and β1 in the frequency
domain
D(qg, θ, zg, zs) = −
ρ0
4
e−iqg(zs+zg) 1
cos2 θ
α1(−2qg) + (1 − tan2
θ)β1(−2qg) , (21)
where the subscripts s and g denote source and receiver quantities respectively, and qg, θ and
k = ω/c0 shown in Fig. 1, have the following relations (Matson, 1997)
qg = qs = k cos θ,
kg = ks = k sin θ.
111 ,, Kc
gq k
gk
zz
000 ,, Kc
Figure 1: The relationship between qg, kg and θ.
Similarly, substituting Eq. (20) into Eq. (9), we can get the solution for α2(z) and β2(z) as a
function of α1(z) and β1(z)
1
cos2 θ
α2(z) + (1 − tan2
θ)β2(z) = −
1
2 cos4 θ
α2
1(z) −
1
2
(1 + tan4
θ)β2
1(z) +
tan2 θ
cos2 θ
α1(z)β1(z)
−
1
2 cos4 θ
α1(z)
z
0
dz [α1(z ) − β1(z )]
+
1
2
(tan4
θ − 1)β1(z)
z
0
dz [α1(z ) − β1(z )], (22)
where α1(z) = dα1(z)
dz , β1(z) = dβ1(z)
dz .
The ﬁrst two parameter direct non-linear inversion of 1D acoustic media for a 2D experiment has
been obtained. As shown in Eq. (21) and Eq. (22), given two diﬀerent angles θ, we can determine
189

α1, β1 and then α2, β2. For a single-interface example, it can be shown that only the first three
terms on the right hand side contribute to parameter predictions, while the last two terms perform
imaging in depth since they will be zero after the integration across the interface (see the section
on three important messages). Therefore, in this solution, the tasks for imaging-only and inversion-
only terms are separated.
For the θ = 0 and constant density case, Eq. (22) reduces to the non-linear solution for 1D one
parameter normal incidence case (e.g., Shaw, 2005)
α2(z) = −
1
2

α2
1(z) + α1(z)
z
−∞
dz α1(z )

 . (23)
If another choice of free parameter other than θ (e.g., ω or kh) is selected, then the functional
form between the data and the first order perturbation Eq. (21) would change. Furthermore, the
relationship between the first and second order perturbation Eq. (22) would, then, also be different,
and new analysis would be required for the purpose of identifying specific task separated terms.
Empirically, the choice of θ as free parameter (for a 1D medium) is particularly well suited for
allowing a task separated identification of terms in the inverse series.
There are several important messages that exist in Eq. (21) and Eq. (22): (1) purposeful perturba-
tion, (2) leakage, and (3) the special parameter for inversion. These three concepts will be discussed
later in this paper. In Eq. (21), it seems simple and straightforward to use data at two angles in
order to obtain α1 and β1. This is what we do in this paper. However, by doing this, it requires a
whole new understanding of the definition of “the data”. That is part of the discoveries of on-going
research activities by Weglein et al. (2007). The imaging algorithm given by Liu et al. (2005) has
been generalized to the two parameter case by Weglein et al. (2007) based on the understanding
of Eq. (22).
A special case: one-interface model
In this section, we derive a closed form for the inversion-only terms. From this closed form, we can
easily get the same inversion terms as those in Eqs. (21) and (22). We also show some numerical
tests using analytic data. From the numerical results, we see how the corresponding non-linear
terms contribute to the parameter predictions such as the relative changes in the P-wave bulk
modulus α = ∆K
K , density β = ∆ρ
ρ , impedance ∆I
I and velocity ∆c
c .
Closed form for the inversion terms
1. Incident angle not greater than critical angle, i.e. θ ≤ θc
For a single interface example, the reflection coefficient has the following form (Keys, 1989)
R(θ) =
(ρ1/ρ0)(c1/c0) 1 − sin2
θ − 1 − (c2
1/c2
0) sin2
θ
(ρ1/ρ0)(c1/c0) 1 − sin2
θ + 1 − (c2
1/c2
0) sin2
θ
. (24)
190

After adding 1 on both sides of Eq. (24), we can get
1 + R(θ) =
2 cos θ
cos θ + (ρ0/ρ1) c2
0/c2
1 − sin2
θ
. (25)
Then, using the deﬁnitions of α = 1 − K0
K1
= 1 −
ρ0c2
0
ρ1c2
1
and β = 1 − ρ0
ρ1
, Eq. (25) becomes
4R(θ)
(1 + R(θ))2 =
α
cos2 θ
+ (1 − tan2
θ)β −
αβ
cos2 θ
+ β2
tan2
θ, (26)
which is the closed form we derived for the one interface two parameter acoustic inversion-only
terms.
2. Incident angle greater than critical angle, i.e. θ > θc
For θ > θc, Eq. (24) becomes
R(θ) =
(ρ1/ρ0)(c1/c0) 1 − sin2
θ − i (c2
1/c2
0) sin2
θ − 1
(ρ1/ρ0)(c1/c0) 1 − sin2
θ + i (c2
1/c2
0) sin2
θ − 1
. (27)
Then, Eq. (25) becomes
1 + R(θ) =
2 cos θ
cos θ + i (ρ0/ρ1) sin2
θ − c2
0/c2
1
, (28)
which leads to the same closed form as Eq. (26)
4R(θ)
(1 + R(θ))2 =
α
cos2 θ
+ (1 − tan2
θ)β −
αβ
cos2 θ
+ β2
tan2
θ.
As we see, this closed form is valid for all incident angles.
In addition, for normal incidence (θ = 0) and constant density (β = 0) media, the closed form Eq.
(26) will be reduced to
α =
4R
(1 + R)2 . (29)
This represents the relationship between α and R for the one parameter 1D acoustic constant
density medium and 1D normal incidence obtained in Innanen (2003). In this case, α becomes
1 − c2
0/c2
1 and R becomes (c1 − c0) / (c1 + c0).
3. Derivation of the inversion terms from the closed form
From the closed form Eq. (26), using Taylor expansion on the left hand side
1
(1 + R(θ))2 = 1 − R(θ) + R2
(θ) − . . .
2
,
191

and setting the terms of equal order in the data equal, we have
α1
cos2 θ
+ (1 − tan2
θ)β1 = 4R(θ), (30)
α2
cos2 θ
+ (1 − tan2
θ)β2 = −
1
2
α2
1
cos4 θ
−
1
2
(1 + tan4
θ)β2
1 +
tan2 θ
cos2 θ
α1β1. (31)
For a one-interface example (in Fig. 2), Eqs. (21) and (22) will respectively reduce to the same
form as Eqs. (30) and (31), which is shown below.
Assume the interface surface is at depth z = a, and suppose zs = zg = 0.
000 ,, Kc
111 ,, Kc
zz
xx
aa
00
Figure 2: 1D one-interface acoustic model.
Using the analytic data (Clayton and Stolt, 1981; Weglein et al., 1986),
D(qg, θ) = ρ0R(θ)
e2iqga
4πiqg
, (32)
and substituting Eq. (32) into Eq. (21), after Fourier transformation over 2qg, for z > a and ﬁxed
θ, we get
1
cos2 θ
α1(z) + (1 − tan2
θ)β1(z) = 4R(θ)H(z − a). (33)
Also, the non-linear solution Eq. (22) will reduce to
1
cos2 θ
α2(z) + (1 − tan2
θ)β2(z) = −
1
2 cos4 θ
α2
1(z) −
1
2
(1 + tan4
θ)β2
1(z)
+
tan2 θ
cos2 θ
α1(z)β1(z), (34)
The two equations Eqs. (33) and (34) agree with Eqs. (30) and (31), respectively.
192

Numerical tests
From Eq. (33), we choose two different angles to solve for α1 and β1
β1(θ1, θ2) = 4
R(θ1) cos2 θ1 − R(θ2) cos2 θ2
cos(2θ1) − cos(2θ2)
, (35)
α1(θ1, θ2) = β1(θ1, θ2) + 4
R(θ1) − R(θ2)
tan2 θ1 − tan2 θ2
. (36)
Similarly, from Eq. (34), given two different angles we can solve for α2 and β2 in terms of α1 and
β1
β2(θ1, θ2) = −
1
2
α2
1
1
cos2 θ1
−
1
cos2 θ2
+ α1β1 tan2
θ1 − tan2
θ2 −
1
2
β2
1
× cos2
θ1 − cos2
θ2 +
sin4
θ1
cos2 θ1
−
sin4
θ2
cos2 θ2
/ [cos(2θ1) − cos(2θ2)] , (37)
α2(θ1, θ2) =β2(θ1, θ2) + −
1
2
α2
1
1
cos4 θ1
−
1
cos4 θ2
+ α1β1
tan2 θ1
cos2 θ1
−
tan2 θ2
cos2 θ2
−
1
2
β2
1 tan4
θ1 − tan4
θ2 / tan2
θ1 − tan2
θ2 ; (38)
where α1 and β1 in Eqs. (37) and (38) denote α1(θ1, θ2) and β1(θ1, θ2), respectively.
For a specific model, ρ0 = 1.0g/cm3, ρ1 = 1.1g/cm3, c0 = 1500m/s and c1 = 1700m/s, in the
following figures we give the results for the relative changes in the P-wave bulk modulus α = ∆K
K ,
density β = ∆ρ
ρ , impedance ∆I
I and velocity ∆c
c corresponding to different pairs of θ1 and
θ2.
From Fig. 3, we can see that when we add α2 to α1, the result is much closer to the exact value
of α. Furthermore, the result is better behaved; i.e., the plot surface becomes flatter, over a larger
range of precritical angles. Similarly, as shown in Fig. 4, the results of β1 + β2 are much better
than those of β1. In addition, the sign of β1 is wrong at some angles, while, the results for β1 + β2
always have the right sign. So after including β2, the sign of the density is corrected, which is very
important in the earth identification, and also the results of ∆I
I (see Fig. 5 ) and ∆c
c (see Fig. 6)
are much closer to their exact values respectively compared to the linear results.
Especially, the values of ∆c
c 1
are always greater than zero, that is, the sign of (∆c)1 is always
positive, which is the same as that of the exact value ∆c. We will further discuss this in the next
section.
Three important messages
As mentioned before, in general, since the relationship between data and target property changes
is non-linear, linear inversion will produce errors in target property prediction. When one actual
193

property change is zero, the linear prediction of the change can be non-zero. Also, when the actual
change is positive, the predicted linear approximation can be negative. There is a special parameter
for linear inversion of acoustic media, that never suffers the latter problem.
From Eq. (24) we can see that when c0 = c1, the reflection coefficient is independent of θ, then
from the linear form Eq. (36), we have
∆c
c 1
=
1
2
(α1 − β1) = 0 when ∆c = 0,
i.e., when ∆c = 0, (∆c)1 = 0. This generalizes to (∆c)1 > 0 when ∆c > 0, or (∆c)1 < 0 when
∆c < 0, as well. This can be shown mathematically (See Appendix B for details).
Therefore, we can, first, get the right sign of the relative change in P-wave velocity from the linear
inversion (∆c)1, then, get more accurate values by including non-linear terms.
Another interesting point is that the image does not move when the velocity does not change across
an interface, i.e., c0 = c1, since, in this situation, the integrands of imaging terms α1 − β1 in Eq.
(22) are zero. We can see this more explicitly when we change the two parameters α and β to ∆c
c
and β. Using the two relationships below (See details in Appendix A)
∆c
c 1
=
1
2
(α1 − β1),
and
∆c
c 2
=
1
2
1
4
(α1 + β1)2
− β2
1 + (α2 − β2) ,
rewriting Eq. (22) as
1
cos2 θ
∆c
c 2
(z) + β2(z) =
cos2 θ − 2
2 cos4 θ
∆c
c
2
1
(z) −
1
2
β2
1(z)
−
1
cos4 θ
∆c
c 1
(z)
z
0
dz
∆c
c 1
−
1
cos2 θ
β1(z)
z
0
dz
∆c
c 1
. (39)
This equation indicates two important concepts. One is leakage: there is no leakage correction at
all in this expression. Here the leakage means that, if the actual value of α (relative changes in
P-wave bulk modulus) is zero, its linear approximation α1 could be non-zero since α and β are
coupled together (like the coupled term α1β1 in Eq. 22) and α1 could get leakage values from β1.
While in Eq. (39), no such coupled term is present at all and thus, if the actual changes in the
velocity are zero, then its linear inversion ∆c
c 1
would be zero and there would be no leakage from
β1. This leakage issue or coupled term has no analogue in the 1D one parameter acoustic case
(Eq. 23) since in this case we only have one parameter and there is no other parameter to leak
194

into. In other words, in the one parameter (velocity) case, each ‘jump’ in the amplitude of the data
(primaries only) corresponds to each wrong location with a wrong amplitude for the parameter
predicted in the linear inverse step; while in the two parameter case of this paper, each ‘jump’ in
the data no longer has the simple one-to-one relationship with the amplitude and location of the
two parameters.
The other concept is purposeful perturbation. The integrand ∆c
c 1
of the imaging terms clearly
tells that if we have the right velocity, the imaging terms will automatically be zero even without
doing any integration; otherwise, if we do not have the right velocity, these imaging terms would
be used to move the interface closer to the right location from the wrong location. The conclusion
from this equation is that the depth imaging terms depend only on the velocity errors.
Conclusion
In this paper, we derive the first two parameter direct non-linear inversion solution for 1D acoustic
media with a 2D experiment. Numerical tests show that the terms beyond linearity in earth
property identification subseries provide added value. Although the model we used in the numerical
tests is simple, the potential within Eqs. (21) and (22) applies to more complex models since the
inverse scattering series is a direct inversion procedure which inverts data directly without knowing
the specific properties above the target.
As shown above, adding one parameter in the wave equation makes the problem much more com-
plicated in comparison with the one parameter case. Three important concepts (purposeful pertur-
bation, leakage and special parameter for inversion) have been discussed and how they relate to the
linear and non-linear results for parameter estimation, addressing leakage, and imaging. Further
progress on these issues is being carried out with on-going research.
The work presented in this paper is an important step forward for imaging without the velocity
model, and target identification for the minimally acceptable elastic isotropic target. In this paper
for the first time the general one-dimensional formalism for a depth varying acoustic medium is
presented for depth imaging and direct parameter estimation, without needing to determine medium
velocity properties that govern actual wave propagation for depth imaging, or what medium is above
a target to be identified. The encouraging numerical results motivated us to move one step further
— extension of our work to the isotropic elastic case (see, e.g., Boyse and Keller, 1986) using three
parameters. The companion and sequel paper to this one provides that extension.
Acknowledgements
We thank all sponsors of M-OSRP and we are grateful that Robert Keys and Douglas Foster for
valuable discussions.
195

Appendix A
In this appendix, we derive the expressions of ∆c
c 1
, ∆c
c 2
, ∆I
I 1
and ∆I
I 2
in terms of α1, β1
and α2, β2. Deﬁne ∆c = c − c0, ∆I = I − I0, ∆K = K − K0 and ∆ρ = ρ − ρ0.
Since K = c2ρ, then we have
(c − ∆c)2
=
K − ∆K
ρ − ∆ρ
.
Divided by c2, the equation above will become
2
∆c
c
−
∆c
c
2
=
∆K
K − ∆ρ
ρ
1 − ∆ρ
ρ
.
Remember that α = ∆K
K and β = ∆ρ
ρ , the equation above can be rewritten as
2
∆c
c
−
∆c
c
2
=
α − β
1 − β
.
Then we have
2
∆c
c
−
∆c
c
2
= (α − β)(1 + β + β2
+ · · · ), (40)
where the series expansion is valid for |β| < 1.
Similar to Eqs. (17) and (18), ∆c
c can be expanded as
∆c
c
=
∆c
c 1
+
∆c
c 2
+ · · · . (41)
Then substitute Eqs. (41), (17) and (18) into Eq. (40), and set those terms of equal order equal
on both sides of Eq. (40), we can get
∆c
c 1
=
1
2
(α1 − β1), (42)
and
∆c
c 2
=
1
2
1
4
(α1 + β1)2
− β2
1 + (α2 − β2) . (43)
Similarly, using I = cρ, we have
(I − ∆I)2
= (K − ∆K)(ρ − ∆ρ).
Divided by I2, the equation above will become
2
∆I
I
−
∆I
I
2
= α + β − αβ. (44)
196

Expanding ∆I
I as
∆I
I
=
∆I
I 1
+
∆I
I 2
+ · · · , (45)
and substitute Eqs. (45), (17) and (18) into Eq. (44), setting those terms of equal order equal on
both sides of Eq. (44), we can get
∆I
I 1
=
1
2
(α1 + β1), (46)
and
∆I
I 2
=
1
2
1
4
(α1 − β1)2
+ (α2 + β2) . (47)
Appendix B
In this appendix, we show that ∆c
c 1
has the same sign as ∆c. For the single interface example,
from Eqs. (36) and (42), we have
∆c
c 1
= 2
R(θ1) − R(θ2)
tan2 θ1 − tan2 θ2
.
The reﬂection coeﬃcient is
R(θ) =
(ρ1/ρ0)(c1/c0) 1 − sin2
θ − 1 − (c2
1/c2
0) sin2
θ
(ρ1/ρ0)(c1/c0) 1 − sin2
θ + 1 − (c2
1/c2
0) sin2
θ
.
Let
A(θ) = (ρ1/ρ0)(c1/c0) 1 − sin2
θ,
B(θ) = 1 − (c2
1/c2
0) sin2
θ.
Then
R(θ1) − R(θ2) = 2
A(θ1)B(θ2) − B(θ1)A(θ2)
[A(θ1) + B(θ1)] [A(θ2) + B(θ2)]
,
where the denominator is greater than zero. The numerator is
2 [A(θ1)B(θ2) − B(θ1)A(θ2)] =2(ρ1/ρ0)(c1/c0) 1 − sin2
θ1 1 − (c2
1/c2
0) sin2
θ2
− 1 − sin2
θ2 1 − (c2
1/c2
0) sin2
θ1 .
Let
C = 1 − sin2
θ1 1 − (c2
1/c2
0) sin2
θ2,
D = 1 − sin2
θ2 1 − (c2
1/c2
0) sin2
θ1.
197

Then,
C2
− D2
=
c2
1
c2
0
− 1 (sin2
θ1 − sin2
θ2).
When c1 > c0 and θ1 > θ2 , we have (Noticing that both C and D are positive.)
c2
1
c2
0
− 1 (sin2
θ1 − sin2
θ2) > 0,
so
R(θ1) − R(θ2) > 0;
Similarly, when c1 < c0 and θ1 > θ2 , we have
c2
1
c2
0
− 1 (sin2
θ1 − sin2
θ2) < 0,
so
R(θ1) − R(θ2) < 0.
Remembering that ∆c
c 1
= 2 R(θ1)−R(θ2)
tan2 θ1−tan2 θ2
. So for c1 > c0, (∆c)1 > 0 and for c1 < c0, (∆c)1 < 0 .
References
Boyse, W. E. and J. B. Keller. “Inverse elastic scattering in three dimensions.” J. Acoust. Soc.
Am. 79 (1986): 215–218.
Clayton, R. W. and R. H. Stolt. “A Born-WKBJ inversion method for acoustic reflection data.”
Geophysics 46 (1981): 1559–1567.
Innanen, Kristopher. A. Methods for the treatment of acoustic and absorptive/dispersive wave field
measurements. PhD thesis, University of British Columbia, 2003.
Keys, R. G. “Polarity reversals in reflections from layered media.” Geophysics 54 (1989): 900–905.
Liu, F., A. B. Weglein K. A. Innanen, and B. G. Nita. “Extension of the non-linear depth imaging
capability of the inverse scattering series to multidimensional media: strategies and numerical
results.” 9th Ann. Cong. SBGf, Expanded Abstracts. . SBGf, 2005.
Matson, K. H. An inverse-scattering series method for attenuating elastic multiples from multi-
component land and ocean bottom seismic data. PhD thesis, University of British Columbia,
1997.
Raz, S. “Direct reconstruction of velocity and density profiles from scattered field data.” Geophysics
46 (1981): 832–836.
Sen, M. and P. L. Stoffa. Global Optimization Methods in Geophysical Inversion. Amsterdam:
Elsevier, 1995.
198

Shaw, S. A. An inverse scattering series algorithm for depth imaging of reflection data from a
layered acoustic medium with an unknown velocity model. PhD thesis, University of Houston,
2005.
Shaw, S. A., A. B. Weglein, D. J. Foster, K. H. Matson, and R. G. Keys. “Isolation of a leading
order depth imaging series and analysis of its convergence properties.” M-OSRP Annual Report
2 (2003): 157–195.
Stolt, R. H. and A. B. Weglein. “Migration and inversion of seismic data.” Geophysics 50 (1985):
2458–2472.
Tarantola, A., A. Nercessian, and O. Gauthier. “Nonlinear Inversion of Seismic Reflection Data.”
54rd Annual Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts. . Soc. Expl. Geophys.,
1984. 645–649.
Taylor, J. R. Scattering theory: the quantum theory on nonrelativistic collisions. Wiley, New York,
1972.
Weglein, A. B., F. V. Araújo, P. M. Carvalho, R. H. Stolt, K. H. Matson, R. T. Coates, D. Corrigan,
D. J. Foster, S. A. Shaw, and H. Zhang. “Inverse scattering series and seismic exploration.”
Inverse Problems 19 (2003): R27–R83.
Weglein, A. B., D. J. Foster, K. H. Matson, S. A. Shaw, P. M. Carvalho, and D. Corrigan. “Predict-
ing the correct spatial location of reflectors without knowing or determining the precise medium
and wave velocity: initial concept, algorithm and analytic and numerical example.” Journal of
Seismic Exploration 10 (2002): 367–382.
Weglein, A. B., F. A. Gasparotto, P. M. Carvalho, and R. H. Stolt. “An inverse-scattering series
method for attenuating multiples in seismic reflection data.” Geophysics 62 (1997): 1975–1989.
Weglein, A. B. and R. H. Stolt. 1992 “Approaches on linear and non-linear migration-inversion.”.
Personal Communication.
Weglein, A. B., P. B. Violette, and T. H. Keho. “Using multiparameter Born theory to obtain
certain exact multiparameter inversion goals.” Geophysics 51 (1986): 1069–1074.
199

Figure 3: α1 (top) and α1 + α2 (bottom) displayed as a function of two diﬀerent angles. The graphs on
the right are the corresponding contour plots of the graphs on the left. In this example, the exact
value of α is 0.292.
200

Figure 4: β1 (top) and β1 + β2 (bottom). In this example, the exact value of β is 0.09.
201

Figure 5: Linear approximation to relative change in impedance (see details in Appendix A) ∆I
I 1
=
1
2 (α1 + β1) (top). Sum of linear and ﬁrst non-linear terms ∆I
I 1
+ ∆I
I 2
= ∆I
I 1
+
1
2
1
4 (α1 − β1)2
+ (α2 + β2) (bottom). In this example, the exact value of ∆I
I is 0.198.
202

Figure 6: Linear approximation to relative change in velocity (see details in Appendix A) ∆c
c 1
=
1
2 (α1 − β1) (top). Sum of linear and ﬁrst non-linear terms ∆c
c 1
+ ∆c
c 2
= ∆c
c 1
+
1
2
1
4 (α1 + β1)2
− β2
1 + (α2 − β2) (bottom). In this example, the exact value of ∆c
c is 0.118.
203

Zhang weglein-2008

More Related Content

What's hot (20)

Viewers also liked (14)

Similar to Zhang weglein-2008 (20)

More from Arthur Weglein (19)

Zhang weglein-2008