Chapter 4
A Seismic Inversion Problem for an
Anisotropic, Inhomogeneous Medium
Dhavide Aruliah1 , Mikalai Birukou2 , Neville Dubash3 , Leslie Fairbairn4 , Jeff Grossman5 ,
Katherine Hegewisch6 , Namyong Lee7 , Robin Swain8 , Carlos Trenado9 ,
Industrial Contacts:Gary Billings10 , Michael Slawinski (Mentor)8 , Chad Wheaton (Mentor)8 ,
4.1 Introduction
In the petroleum industry, many geoscientists rely on information provided by seismic data.
Based on the interpretation of seismic data, geophysicists and geologists attempt to identify and
target areas of economic importance, specifically hydrocarbon traps. We wish to extract from
the data as much information as possible.
A key factor in the successful processing and interpretation of seismic data is knowing certain
rock physics properties, in particular the elastic moduli of the section of earth through which
seismic waves propagate. One means of attempting to quantify the rock physics properties is
through a vertical seismic profile (VSP), wherein seismic sources are located on the surface and
receivers are placed in a well-bore.
Traditionally, only traveltimes were measured at the receivers, but technological advances
now allow measurement of the polarization angles (i.e., particle displacement) as well. This
provides an additional piece of information towards a practical means of accurately determining
the physical parameters that characterize the medium.
1
University of Toronto
2
University of Waterloo
3
University of British Columbia
4
Simon Fraser University
5
University of Calgary
6
Washington State University
7
Minnesota State University
8
Memorial University
9
New Mexico State University
10
Talisman Energy
57
58 CHAPTER 4. SEISMIC INVERSION PROBLEM
Source (0,0)
Upper Medium
vertically nonuniform
elliptically anisotropic
Point of Refraction
x
Layer of Interest
Receiver (X,Z)
uniform
anisotropic
In this report, we consider the propagation of seismic waves through a medium that can be
subdivided into of two distinct parts. The upper part is assumed to be azimuthally symmetric,
linearly nonuniform with increasing depth, and the velocity dependance with direction consistent
with elliptical anisotropy. The lower part, which is the layer of interest, is assumed to also be
azimuthally symmetric, but uniform and nonelliptically anisotropic. Despite nonellipticity, we
assume the angular dependance of the velocity can be described by a convex curve.
Our goal is to produce a single source-single receiver model which uses modern seismic
measurements to determine the elastic moduli of the lower media. Once known, geoscientists
could better describe the angular dependance of the velocity in the layer of interest and also
would have some clues at to the actual material composing it.
4.2 Formulation of the Problem
We begin by assuming that the area of study consists of layered sedimentary rocks. Rotational
symmetry in such a case effectively reduces problem to two dimensions. Since, in the case of
rational symmetry, energy will propagate between source and receiver in a vertical plane, we
can define our coordinate system such that our x and z coordinate axes contain the plane of
propagation.
We start with a vertical cross section of the earth with a seismic source on the surface and a
receiver located below. We define the horizontal offset between source and receiver to be X, and
the vertical offset to be Z. That is, we define the source to be at the origin of our coordinate
system, (0, 0), and the receiver to be at the location (X, Z). Furthermore, we assume that the
subsurface is divided into two distinct media with the depth to the interface between them to
be H. Finally, we define the horizontal offset between the source and the point of refraction of
the seismic ray to be x.
4.2. FORMULATION OF THE PROBLEM 59
4.2.1 Assumptions
We assume that the seismic source is of high frequency with a wavelength shorter than all
characteristic lengths in the layers of interest. With this assumption, it follows, from asymptotic
ray theory, that the signal travels along a ray, which is a trajectory of the stationary traveltime.
Thus, we assume that the rays obey the variational principle of Fermat.
We also assume that the medium through which the seismic waves propagate is a perfectly
linear elastic continuum. Hence the stress-strain relationships are described by Hookes law.
4.2.2 Upper Medium
We consider the upper medium, 0 z H, to be horizontally stratified {(i.e., sedimentary
rocks)} and both vertically nonuniform and elliptically anisotropic. Vertical nonuniformity is
often the result of varying compaction of the subsurface and manifests itself as a linear increase
in {vertical} velocity with depth. That is, Vz (z) = a + bz. In the case of elliptical anisotropy we
can define a single parameter, , that describes the relationship between vertical and horizontal
velocity1 . Specifically,
V2V2
= x 2 z . (4.1)
2Vz
Combining the effects of both nonuniformity and anisotropy results in a ray velocity profile2 of
the form s
1 + 2
V (z, ; a, b, ) = (a + bz) (4.2)
1 + 2 cos2 ()
where a, b, and are predetermined.
Then, by applying Noethers Theorem it can be shown that the ray form of Snells Law leads
to the expression for the ray parameter
2x
p= p . (4.3)
(x2 + (1 + 2) z 2 ) (4a2 (1 + 2) + 4ab (1 + 2) z + b2 (x2 + (1 + 2) z 2 ))
Applying the variational principle of Fermat, given the above assumptions, the traveltime
function3 of a ray traveling from a source at (0, 0) to a point (x, H) is
p
1 1 + 1 p(x)2 a2 (1 + 2)
t1 (x, z; a, b, ) = ln p . (4.4)
b 1 + 1 p(x)2 (a + bz)(1 + 2)
It has been assumed that the parameters a, b and , which characterize the upper medium,
are known. These values are obtained by considering traveltimes from source to receivers located
in the upper medium, and applying a best-fit method (e.g., conjugate-gradient optimization) to
the traveltime expression, t1.
1
Thomsen, L.,(1986) Weak Elastic Anisotropy, Geophysics, 51, pp. 1954 - 1966.
2
Affarwala, B., et. al., (1997) Inversion of Anisotropic Velocity Parameter. Proceedings of the first PIMS
Industrial Solving Workshop, PIMS, pp. 76-89.
3
Slawinski, M., Wheaton, C., personal communication. 2002
60 CHAPTER 4. SEISMIC INVERSION PROBLEM
4.2.3 Layer of Interest
We consider the layer of interest, H z Z, to be uniform, perfectly elastic, and transversely
isotropic. Uniformity implies that all rays will be straight lines from the refraction point to
the receiver. Elasticity and transverse isotropy imply that under Hookes Law, which give the
relationship between stress, , and strain, , via the elastic moduli, Cij ,
i = Cij j , i, j [1, ..., 6],
there will be five independent elasticity constants. Furthermore, if we focus solely on compressional-
wave data, this number further reduces to four. In other words, only C11 , C13 , C33 , C44 will be
independent and need to be solved for in this particular case.
In the equations above, we have made reference to the ray angle, . However, the angle that
is measured in field observations is not the ray angle. In fact, in the presence of anisotropy there
are three distinct angles associated with wave propagation which must be carefully distinguished.
The first is the ray angle, , which is associated with the direction of energy propagation. In the
case of uniform media where raypaths are straight, it is equal to the angle between the vertical
and the radian to a point on the wavefront (see Figure 2). If we were to {then} construct a
perpendicular to the wavefront at end of the radius, the angle between the perpendicular and
the vertical would be the phase angle, . Thirdly, we have the displacement (or polarization)
angle, , which describes the direction of particle displacement as the wavefront propagates.
This last angle is obtained by measuring three orthogonal displacements at the receiver. It can
be shown that both the ray and polarization angles, and , respectively, are related to the
phase angle by the following expressions4 ,
1 v
tan() + v
tan() = tan() v
(4.5)
1 v
4
Arnold, V.I., (1989) Mathematical Methods of Classical Mechanics (2nd ed.): Springer-Verlag, pp. 248 -
252.
4.3. METHOD OF SOLUTION 61
v 2 C44 sin2 () C33 cos2 ()
tan() = (4.6)
(C13 + C44 ) sin() cos()
Our goal is to formulate and solve an inverse problem, in which a, b, , X, Z, T ,and are the
inputs and the elasticity constants, C11 , C13 , C33 , C44 , of the layer of interest are the outputs.
4.3 Method of Solution
4.3.1 Solving the Forward Problem
Before we attempt to tackle the inverse problem, we consider the forward problem. By starting
with the forward problem we will be able to generate input values for the inverse problem while
at the same time knowing what outputs the inverse should give. We begin by fixing the values
of a, b, and for the upper medium. Also, we set the refraction point, x, the mass density of
the lower layer of interest, , the elastic moduli, Cij , and the total traveltime, T,from source
to receiver. Using the equations above, we are then able find the refraction point, x, and the
polarization angle, .
First, we calculate the horizontal ray slowness, p, in the upper medium by substituting the
refraction point (x, H) into equation (4.3)
2x
p(x) = p
(x + (1 + 2)H )(4a (1 + 2) + 4ab(1 + 2)H + b2 (x2 + 1 + (1 + 2)H 2
2 2 2
and also the traveltime in the upper medium using equation (4.4)
p
1 1 + 1 p(x) a (1 + 2)
2 2
t1 (x, z; a, b, ) = ln p .
b 1 + 1 p(x)2 (a + bz)(1 + 2)
By Snells law, the tangential component of phase slowness, p, will be conserved across the
interface. If and v() are the phase angle and phase velocity, respectfully, for the lower layer
we have
sin()
= p.
v()
The phase velocity5 , v(), is given by
(C33 C11 ) cos2 () + C11 + C44 +
v 2 () = (4.7)
2
with
= [(C11 C33 ) cos2 () (C11 + C44 )]2
4[C33 C44 cos4 () (2C13 C44 C11 C33 + C13
2
) cos2 () sin2 () + C11 C44 sin4 ()].
5
Daley, P.F. & Hron, F., 1977. Reflection and Transmission Coefficients for Transversely Isotropic Media,
Bull. Seism. Soc. Am., 67, 661-675.
62 CHAPTER 4. SEISMIC INVERSION PROBLEM
At this point we can solve for the phase angle and thereby the phase velocity in the lower
medium. Once we have the phase velocity, we can solve for the ray angle using equation (4.5),
1 v
tan() + v
tan() = tan() v
.
1 v
We then substitute the phase angle into the ray velocity expression, can find the ray velocity,
V , using
2
2 2 v
V () = v () + . (4.8)
Since the layer of interest is uniform, the ray path is a straight line and the ray velocity in
the direction of the ray angle will be constant. Thus the traveltime, t2 , in the layer of interest is
p
(X x)2 + (Z H)2
t2 = (4.9)
V ()
and the total travel time, T , is
T = t1 (x) + t2 (x).
Substituting the phase angle into expression (4.6) we get the polarization angle, ,
v 2 C44 sin2 () C33 cos2 ()
tan() = .
(C13 + C44 ) sin() cos()
4.3.2 Solving the Inverse Problem
Next we attempt to solve the inverse problem directly by combining the expressions described
above into a system of equations. Since the rays will be straight in the lower medium we can
reexpress the equations above to get the following system of nonlinear equations,
X x tan() + v1 v
=0 (4.10)
ZH 1 tan()
v
v
p
(X x)2 + (Z H)2
T t1 (x) = 0
V ()
p(x)v(; Cij ) sin() = 0
v 2 C44 sin2 () C33 cos2 ()
tan() = 0.
(C13 + C44 ) sin() cos()
Unfortunately we have the problem of an underdetermined system with six unknowns
(x, , C11 , C13 , C33 , C44 ) but only four equations. However, the industrial contacts on the team
tell us that reasonable estimates of C33 and C44 are not completely out of reach. Therefore we
proceed with the added the restriction that C33 and C44 are both known.
4.4. FUTURE WORK 63
This system can now be solved using either Jacobi Iteration or the Conjugate-gradient
Method. Results
As a starting point for the forward problem, we used the following the parameter values
which are typical in actual settings,
a = 2000 C11 = 3.13e10 Z = 1154.46
b = 0.3 C13 = 0.34e10 X = 1057.03
= 0.2 C33 = 2.25e10 H = 700.
= 2310 C44 = 0.65e10
Using the forward problem, we generated the corresponding values of T, . Then, we used our
inverse problem code to try to regenerate the values for x, C11 , C13 , C33 , C44 . For exact values of
the input values, T, , the system converged to the correct values of Cij . Unfortunately however,
for even small perturbations of the inputs (consistent with experimental error) a solution was
often not possible and when solutions were found, they tended to be widely different from the
true values. Due to the time constraints of the workshop we were not able proceed with an in-
depth error analysis. However, it was noted that provided the experimental errors were random
(as opposed to systematic) a collection of outputs obtained by considering numerous source-
receiver setups, could be used to estimate the true Cij s. By taking the best fit for each set of
estimates of Cij , the results tended to be reasonably close to the true values.
4.4 Future Work
Through our work on this problem we feel that our progress was marked by raising more
questions than answers. As a starting point for future or continued work on this project we
would suggest that an in-depth error analysis be performed. By doing so, hidden redundancies
in the formulations and suitable initial/boundary conditions may be identified to combat the
sensitivity to errors and non uniqueness of solutions.
More importantly, however, we feel that the problem would be best approached by consid-
ering not one, but two sources on the surface (Figure 3). By doing so this would result in eight
equations in eight unknowns, negating the need for a priori knowledge of C33 and C44 .
64 CHAPTER 4. SEISMIC INVERSION PROBLEM
Source 1
Source 2
Upper Medium
inhomogeneous
elliptically anisotropic
Lower Medium 1
homogeneous 2
weakly anisotropic Receiver
Adding a second source to equations (4.10) would give,
1 v
X x1 tan(1 ) + v1 11
0= (4.11)
ZH 1 tan(
v1
1 ) v1
1
1 v
X x2 tan(2 ) + v2 22
0=
ZH 1 tan(
v2
2 ) v2
2
p
(X x1 ) + (Z H)2
2
0 = T1 t1 (x1 )
V1 (1 )
p
(X x2 )2 + (Z H)2
0 = T2 t2 (x2 )
V2 (2 )
0 = p1 (x1 )v1 (1 ; Cij ) sin(1 )
0 = p2 (x2 )v2 (; Cij ) sin(2 )
v12 C44 sin2 (1 ) C33 cos2 (1 )
0 = tan(1 )
(C13 + C44 ) sin(1 ) cos(1 )
v C44 sin2 (2 ) C33 cos2 (2 )
2
0 = tan(2 ) 2
(C13 + C44 ) sin(2 ) cos(2 )
4.5. SAMPLE FILE USED TO RUN SIMULATIONS 65
4.5 Sample File Used to Run Simulations
Throughout the coding of the equations presented above, a combination of MatlabT M and
MapleT M code was used (depending on who owned the laptop we were using at the time).
Following is the coding done using Maple T M to manipulate the input values as stated above.
The first code was used to input values to be solved for in the inverse model, and generate data
for the inverse model.
>restart:Digits:=40:
Forward Model where Cijs Known and Refraction Point (xr,zr)
Known.
>
># In this program we leave the Z of the receiver position a free parameter and solve for it
in the process. We only set the horizontal offset of the receiver with respect to the refraction
point, which we fix. The origin is assumed to be at the source.
># Enter in the input values and solve for receiver location, total traveltime, and polarization
angle at receiver
>a:=2000: b:=0.8: chi:=0.3: zr:=800: roe:=2310: # zr=depth of interface/refraction point.
>C11:=3.13*1010: C13:=0.34*1010: C33:=2.25*1010: C44:=0.65*1010:
&&&&&&&&&&&&&&&&&&&&&&&&&&&
>X:=2600: xr:=1154.46: # X=receiver depth, xr= refraction point.
&&&&&&&&&&&&&&&&&&&&&&&&&&&
>
># Traveltime in the upper medium
>p:=2*xr/sqrt((xr2+(1+2*chi)*zr2)*(a2*(4+8*chi)+4*a*b*(1+2*chi)*zr+b2*(xr2+(1+2*chi)*zr2))):
>t1:=(1/b)*ln( (a+b*zr)/a * (1+sqrt(1-p2*a2*(1+2*chi)))/(1+sqrt(1-p2*(a+b*zr)2*(1+2*chi)))):
>
># Calculate the phase velocity and its derivative with respect to phase angle, theta
>Delta:=((C11-C33)*cos(theta)2-(C11+C44))2-4*(C33*C44*cos(theta)4-(2*C13*C44-C11*C33+C132)*
cos(theta)2*sin(theta)2+C11*C44*sin(theta)4):
>v:= sqrt((C11+C44+(C33-C11)*cos(theta)2+sqrt(Delta))/(2*roe)):
The Phase Angle for the ray traveling down from the refraction point (xr,zr):
will have problems if p*v >1. arcsin will then be complex (ie, Shadow Zone.) When you fix
the refraction point, snells law must have a solution for theta (a phase angle must exist). Since
you are using p, conserved from upper medium, combined with v(theta) for the lower medium,
if your refraction point is such that your angle of incidence is beyond the critical angle, snells
law will not have a solution in the lower medium because there will be NO TRANSMITION.
>eqn:=theta=arcsin(p*v):
>theta:=fsolve(eqn,theta,0..Pi/2);
>
># Solve for the Ray Angle from refraction point to receiver at (X,Z).
>rayang:=arctan((tan(theta)+1/v*dv)/(1-tan(theta)/v* dv)):
>rayangle:=evalf(%):
>
66 CHAPTER 4. SEISMIC INVERSION PROBLEM
># Calculate the polarization angle along the ray in the lower medium
>phi:=arctan( (roe*v2-C44*sin(theta)2-C33*cos(theta)2)/((C13+C44)*sin(theta)*cos(theta))):
>
># Solve for the Ray velocity
>rayvelocity:=sqrt(v2+dv2):
Solve for the traveltime in the lower medium, t2.
Start by fixing the offset of x-coordinate of the receiver position from the refraction point,
xr, than calculate the traveltime and z-position of receiver.
>xo:=X-xr:
>w:=xo/sin(rayangle):
>t2:=w/rayvelocity:
>Z:=zr+xo*cot(rayangle):
>T:=t1+t2:
=====================================================
Data to be used for the inverse problem is as follows,
C33, C44 Values;
>C33:=C33; C44:=C44;
Receiver location,
>X,Z:=X,Z;
Total Traveltime...
>TravelTime:=T;
And Polarizaton at the receiver is....
>phi:=phi; evalf(phi*180/Pi);
=======================================================
Inverse Code Should Give,
>xr:=xr; C11:=C11; C13:=C13;
The second code that is presented is a formulation of the inverse problem which relies on
MapleT M s fsolve numerical solver to determine the values of the elastic moduli.
4.5. SAMPLE FILE USED TO RUN SIMULATIONS 67
Inverse Model for Traveltimes and Polarization Angles for 1 Source / 1
Receiver Setup.
>
># List the Input Data for the inverse problem.
>restart:
>Digits:=40:
>a:=2000: b:=0.8: chi:=0.3: zr:=700: roe:=2310:
>C33:=2.25*1010: C44:=0.65*1010:
>X := 1057.0356:
>Z := 729.57398:
>T1 := .48321603:
>phi1 := 1.24456:
>p1:=2*xr1/sqrt((xr12+(1+2*chi)*zr2)*(a2*(4+8*chi)+4*a*b*(1+2*chi)*zr+b2*(xr12+(1+2*chi)*zr2))):
>Delta1:=((C11-C33)*cos(theta1)2-(C11+C44))2-4*(C33*C44*cos(theta1)4-(2*C13*C44-C11*C33+C132)*
cos(theta1)2*sin(theta1)2+C11*C44*sin(theta1)4):
>v1:=sqrt((C11+C44+(C33-C11)*cos(theta1)2+sqrt(Delta1))/(2*roe)):
>dv1:=diff(v1,theta1):
>t1a:=(1/b)*ln(abs((a+b*zr)/a*(1+sqrt(1-p12*a2*(1+2*chi)))/(1+sqrt(1-p12*(a+b*zr)2*(1+2*chi))))):
># System of Equations to solve for C11,C13,C33,C44,xr1,theta1
>eqn1:=v1*p1-sin(theta1)=0:
>eqn2:=T1-sqrt(((X-xr1)2+(Z-zr)2)/(v12+dv12))-t1a=0:
>eqn3:= tan(phi1)=( (roe*v12-C44*sin(theta1)2-C33*cos(theta1)2)/((C13+C44)*sin(theta1)*cos(theta1))):
>eqn4:= (X-xr1)/(Z-zr)=((tan(theta1)+1/v1*dv1)/(1-tan(theta1)/v1*dv1)):
>sys:=[eqn1,eqn2,eqn3,eqn4]:
>sols:=fsolve({eqn1,eqn2,eqn3,eqn4},{xr1,C11,C13,C33,C44,theta1});
>assign(sols);
>xr1:=xr1; C11:=C11; C13:=C13; C33:=C33; C44:=C44;
68 CHAPTER 4. SEISMIC INVERSION PROBLEM
Bibliography
[1] Affarwala, B., et. al., (1997) Inversion of Anisotropic Velocity Parameter. Proceedings of
the first PIMS Industrial Solving Workshop, PIMS, pp. 76-89.
[2] Arnold, V.I., (1989) Mathematical Methods of Classical Mechanincs (2nd edition): Springer
- Verlag
[3] Cerveny, V., (2001) Seismic Ray Theory: Cambridge University Press.
[4] Daley, P.F. & Hron, F., (1977) Reflection and Transmission Coefficients for Transversely
Isotropic Media, Bull. Seism. Soc. Am., 67, 661-675.
[5] Slawinski, M., (2002) Waves and Rays in Elastic Continua, in print.
[6] Thomsen, L.,(1986) Weak Elastic Anisotropy, Geophysics, 51, pp. 1954 - 1966.
69
