CWP-139

DOE/ER14079-27
July 1993

Dynamic Ray Tracing and its

Application in Triangulated Media

Andreas Rfiger

Master of Science Thesis

Geophysics

DISCLAIMER

This report was prepared as an account of work sponsored-by an agoncy of the United States
Government. N¢ithor the United States Government nor any agency thereof, nor any of their
¢mployecs, mak_:sany warranty, express or implied, or assumes any legal liability or responsi-
bility for the accuracy, completeness, or usefulness of any information, apparatus, product, or
process disclosed:, or represents that its use would not infringe privately owned rights, Refer-
once herein to any specific commercial product, process, or service by trade name, trademark,
manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recom-
mendation, or favoring by the United States Government or any agency thereof. The views
and opinions of authors expressed heroin do not necessarily state or roflect those of the
United States Govornmcntor any agoncy thereof.

Center for Wave Phenonlen_t

Colorado S('hool of Mines
' O
Golden, C()lorad 80401 _ _ _,_,:,_:,..........
 ABSTRACT

Hale and Cohen (1991) developed software to generate two-dimensional computer

models of complex geology. Their method uses a triangulation technique designed to
support efficient and accurate computation of seismic wavefields for models of the earth's
interior. Subsequently, Hale (1991) used this triangulation aI)proach to perform dynamic
ray tracing and create synthetic seismograms based on the method of Gaussian beams.
Here, I extend this methodology to allow an increased variety of ray-theoretical exper-
iments. Specifically, the developed program GBmod (Gaussian Beam MODeling)can
produce arbitrary multiple sequences and incorporate attenuation and density variations.
In addition, I have added an option to perform Fresnel-volume ray tracing ((_ervenS, and
Soares, 1992). Corrections for reflection and transmission losses at interfaces, and for
two-and-one-half-dimensional (2.5-D) spreading are included. However, desI)ite these en-
hancements, difficulties remain in attempts to compute accurate synthetic seismograms
if strong lateral velocity inhomogeneities are present. Here, these problems are discussed
and, to a certain extent, reduced. I provide example computations of high-frequency
seismograms based on the method of Gaussian beams to exhibit the advaratages and dis-
advantages of the proposed modeling method and illustrate new features 5:)r both surface
and vertical seismic profiling (VSP) acquisition geometries.
 TABLE OF CONTENTS

ABSTRACT ..................................... i

ACKNOWLEDGMENTS ............................. iv

Chapter 1 INTRODUCTION ......................... 1

Chapter 2 OVERVIEW ON RAY THEORY AND INTRODUCTION

TO DYNAMIC RAY TRACING ............... 5

2.1 The Eikonal Equation ............................ 5

2.2 Kinematic Ray Tracing Equations ...................... 7
2.3 Dynamic Ray-Tracing Equations ....................... 9
2.4 Linearizing the Dynamic Ray Tracing Equations ............. 12
2.5 Physical Meaning of Q and P ........................ 13
2.6 Two- and Two-And-One-Half-Dimensional Models ............. 15

Chapter 3 MODEL REPRESENTATION ................. 21

3.1 Motivation ................................... 21

3.2 Reflection and Transmission Sequences ................... 22
3.3 VSP Modeling Experiment .......................... 25

Chapter 4 APPLICATION OF DYNAMIC RAY TRACING: RAY DATA

AND FRESNEL VOLUMES .................. 33

4.1 Phase-front Curvature and Paraxial Traveltimes .............. 33

4.2 Ray Amplitudes ................................ 34
4.3 Attenuation .................................. 36
4.4 Two.And_One.Half.Dimensional Spreading ................. 39
4.5 Fresnel-Volume Ray Tracing ......................... 39

Chapter 5 THE GAUSSIAN BEAM METHOD ............. 43

5.1 Why Do We Use the Gaussian Beam Method? ............... 43

5.2 Gaussian Beam Theory ............................ 43
5.3 Difficulties with the Gaussian Beam Approach ............... 47
5.4 Basic Assumptions Inherent to the Beam Approach ............ 53
5.5 Gaussian Beams Across Interfaces ...................... 54
5.6 Near-critical Transmission of Gaussian Beams ............... 55
5.7 The Overthrust l_lo(lel Exl)eriment ..................... 58

ii
 5.8 Application of Gaussian Beam Modeling In Complex Media ....... 63
5.9 Quo Vadis Gaussian Beams? ......................... 67

Chapter 6 CONCLUSION .......................... 69

6.1 Summary ................................... 69

6.2 Future Work .................................. 70

Appendix A SOLUTION OF THE TWO-DIMENSIONAL DYNAMIC

RAY TRACING EQUATIONS ................ 71

Appendix B MATRIX Q AND ITS RELATION TO THE RAY JACO-

BIAN ................................. 73

Appendix C TRANSFORMATION OF GAUSSIAN BEAM AMPLITUDES

ACROSS INTERFACES .................... 76

iii
 ACKNOWLEDGMENTS

Thanks to my advisor Dr. Jack Cohen, Dr. Steve Pruess and Dr. Ken Larner for
critiquing and proofreading this thesis and serving in my committee.
Special thanks to my friends Ted Shuck, Lydia Deng, Herman Jaramillo, Norman
Ettrich and Michel Dietrich for their support in difficult moments of this research.
I am especially in debts to my wonderful wife Christa, who changed her life to enable
my studies here in Golden, Colorado,. Without her support, this research would not have
been possible.
Financial support for this work was provided in part by the United States Depart-
mrnt of Energy, Grant Number DE..FG02-89ER14079. (This support does not constitute
an endorsement by DOE of the views expressed in this paper.) Support was also provided
by the members of the Consortium Project on Seismic Inverse Methods for Complex
Structures at the Center for Wave Phenomena, Colorado School of Mines. I likewise
gratefully acknowledge the Fulbright Commission in Bonn, Germany, who honored me
with a scholarship for my first year of studies ia the United States. The computing fa-
cilities necessary for this research are provided by the Center for Geoscience Computing,
Colorado School of Mines.

iv
 Chapter 1

INTRODUCTION

New triangulation methods for representing subsurface models in computers have

helped to overcome limitations inherent in previous computer models of the earth's inte-
rior. A mesh of triangles can flexibly characterize complex subsurface models that can,
for example, include velocity lenses and overhanging dome structures. Additionally, Hale
and Cohen (1991) designed spatial data structures that contain the adjacency topology
(Weiler, 1988) of the model, so that computing time is not wasted in searching for the
triangles and model parameters in the vicinity of an arbitrary point in the model.
The combination of dynamic ray tracing ((_erven_,, 1985a) and the triangulation
technique is very promising. Dynamic ray tracing is an important extension of the clas-
si(;al kinematic ray tracing. It provides useful dynamic properties such as geometrical
spreading and wavefront-curvature information along the rays. Hale (1991) took advan-
tage of the dynamic ray tracing results to generate synthetic seismograms based on the
method of Gaussian beams ((_erven_,, et al., 1982).
In this thesis, I present an introduction to the theory of dynamic ray tracing and dis-
cuss several applications of this technique. I investigate in detail one specific application,
the Gaussian beam method, and address its current limitations.
In the era of high-sl)eed computing, forward modeling b_me(lon ray-theoretical meth-
ods competes with finite-difference approaches (e.g., Fei, 1993). The latter allows spec-
ification of model parameters such as the seismic w_locity at each point in the grid(led
model. If the grid spacing is chosen fine enough, the accuracy of the calculated syn-
thetic seismograms is very good and synthetic sections generated with a finite-difference
m(,thod provide all types of seismic events simultaneously. However, this often makes it
difficult to identify reflections from interfaces of interest. Moreover, the computing speed
of finite-(liffercnce methods in about two orders of magnitude slower than for ray tracing
rXl)eriments. This significantly reduces the application of finite-difference methods as an
int(,ri)retation tool.
If certain rcquir(_m(:nts are met, ray-theoretical techniques an(l, in particular, the
(lynamic ray-traciug method (1o have a future in seismic modeling. Several criteria may
be _stal)lished to cOral)are the l)(,rfi)rmanc(; of either method (see Table 1.1). H(;re, based
on my experience, I a.ssign subjectiv(, grades fr()m 1 to 10 to the l)erfi)rmance of fil)ite
(liff(,rence mettm(ls (FD), Hale's Gaussian b(_am mo(lcling l)ackage (oldGBmod) an(l its
rxt(,nsion (newGBmod), whi(!h I intro(luce in this pal)er.

, Flexibility : The first ('ritcrion assesses tlm model coml)lrxity an_l thr wtrirty of
ray mo(les that can br trace(l. I havr a(l(lcd an option to Sl)(;cify an attenuation
fitctor and (lensity for (,a(,h g(;ologi(, block, i.e., for each ar(;a I)ouncl(,_l l)y interfac(;s.
N'lor('ov(,r, t h_)softwar(, now all()ws tim illtr()(lu('tion of a l)roa(lt'r class of rayl)aths.
1
 FLEXIBILITY DIAGNOSTIC ACCURACY SPEED
!

FD 10 5 9 3
i ,i i,i ,,,,,i, ,,, ,,i, i _ i, i i i i i i

oldGBmod 6 7 5 10

newGBmod 8 9 7 9

poor excellent
12345678910

Table 1.1. A coml)arison of finite-difference methods (FD), Hale's Gaussian beam mod-
eling package (oldGBmod) and its extension (newGBmod), which is introduced in this
paper. Details are given in the text.
 That is, as shown in Figure 1.1, interfaces no longer are limited to being either
reflectors or transmitters of seismic energy, but now can be both, in a specified
sequence. This generalization broadens the applicability of the modeling. For
example, it is now possible to include multiples.

Distance(krn)
0 5 10 15 20

10

FI(_. 1.1. Ray tracing in a complex subsurface model, This model includes sharp-edged
interfaces and lens-type structures, Lithological interfaces are represented by black lines,
The shading denotes the seismic velocity field. Also shown as black lines are raypaths
corresponding to one type of interface multiple. The white lines are edges of auxiliary
triangles used in the model building.

e Diagnostics : The second criterion evaluates the cal_ability of tile method to act
as a tool for interpretation and understanding of wave propagation in complex
structures. Several new options have been included and will be presented in this
paper. One of the new features, called physical ray tracing or _'esncl.volume my
tracing (_erven_ and Soares, 1992), may be used to study the resolution and the
validity conditions of the ray method.

e Accuracy : Third, accurate seismograms have to be computed. The computation

of the reflection and transmission losses at interfaces as well as correction for 2.5-
D amplitude sl)reading improve tile accuracy of the Gaussian beam ray tracing.
However, problems remain when dealing with significant iahomogeneities in the

3
 model. Some limitations are intrinsic to ray methods; others are specific to the
Gaussian beam approach. Both are discussed later in this paper.

• Speed : The fourth criterion, computational aspects, underlines one of the big
advantages of ray methods. The new options of the ray-tracing package such as
the computation of the out-of-plane spreading, energy partitioning at interfaces,
and attenuation, can be performed while still maintaining the attractive feature of
computational ei_lciency with limited computer storage requirements.

To better understand the advantages and disadvantages of dynamic ray tracing, I illus-
trate the new options for both surface and VSP acquisition geometries. A comparison
with VSP data obtained in a physical modeling laboratory supports benefits of the ray-
theoretical modeling.
Modeling using the Gaussian beam method is only one possible application of the
dynamic ray-traced data generated along the ray. Alternative methods (e.g., the paraxial
methods) can also use these data to create synthetic seismograms. The new modifications
to the program GBmod, such as the tracing of multiples, computation of out-of-plane
spreading, or reflection and trallsniission correction are evaluated independently of the
Gaussiaxl beam procedure.
 Chapter 2

OVERVIEW ON RAY THEORY AND INTRODUCTION TO DYNAMIC

RAY TRACING

' to"
Thischaptergivesan m "'
r auctlon tovariousaspects of raytheory.Althoughmost
ofthederivationscan be foundintextbooks(e.g., Aki and Richa.rds,
1980),I decidedto
go intosome detailwheneveritseemed necessary. My goalistoexplainthisfascinating
theoryinsucha way thatstudents and scientists
from neighboringdisciplines
can follo_v,
or evenbetter,findinterestinthecombinationofphysics and mathematicspresented.
As a starting
point,I first consider an isotropic,perfectly
elasticinhomogeneous
medium inthreedimensions. LaterIdecrease the complexity towardsour needsforthe
i modelingprogram. In thefollowing derivations,
ifnot statedotherwise,I make use of
rectangularcartesian
coordinates withthenotation xi,withi- I,2,3.

2.1 The Eikonai Equation

Solving the cikonal equation by tile method of characteristics provides a mathe-

matical derivation of the ray tracing equations for determining the ray path and the
properties of the wavefront traveling along the ray. Other derivations based on energy
considerations or oll variational calculus such as Fermat's principle often offer a better
physical insight; however, they must be applied to longitudinal and shear waves sepa-
rately, This separation cannot be performed in tile general c_e, and strictly holds only
for homogeneous media.
To study the raypath, the aml)litudcs and tile shape of tile wavefront propagating
along a ray, it is therefore useful to start with the clastodynamic equation of motion.

a,j_= p u,_,tt, (2.1)

where u, and o'ij denote the cartesian components of the displacement field and the stress
tensor, respectively. As throughout this thesis, the Einstein summation convention is
used. Moreover, we use a commonly accepted notation for the partial derivatives with
r,,spcct to cartesian Oml)on(_nts x, and tram t, e.g. a,#_ = _,u_,,t = _. p is the
(l(,nsity of the medium. As shown in the literature (e.g., Aki and Richards, 1980), the
isotropic stress tcnsor can be expressed in terms of u, and the Lamd constants A and #
a.s follows:
+ uj,,)
a,j= AbOuk,A{-p(u_,) , (2.2)
where hi1 denotes the I(roncck(,r-(lclta symbol:
 6_j_ 01 if/_j.
ifi=j

One possible trial solution of tile elastodynamic equation of motion is motivated by

the solution of the homogeneous c_e and has the form

,_(_.x,)= e-'_('_'(''))
_
n=O :i_ ' (2.3)
We could also use all equivalent time-domain approach, but this form of ansatz in the
frequency domain is conceptually simpler, especially if we consider tile medium properties
such as density and Lain6 constants to be independent of time. Note that only tile
amplitude coefficients Uk{") and the eikonal r depend on spatial coordinates. This ray
series solution is written in inverse powers of frequency w. If we are interested in the
i high frequencies, we need only consider the first few terms in the series. Specifically,
in practical applications such a_ our modeling, only the n = 0 term is used. In this
case, U2e) will be called Uk. If we are interested in higher-order waves, more than the
b_e) term is needed. The best known example of such is probably the head-wave. Its
first non-vanishing amplitude coefficient is U_t). A detailed discussion for the interested
reader may be found in (_erven# and Ravindra (1971) and Bleistein (1984). A simple,
but tedious, substitution of (2.3) and (2.2) into the equation of motion (2.1) will lead to
the eikonal equation and transport equation. After rearranging the result in expre 8'stuns
with the same powers of (iw), we obtain the basic system of equations for the ray method:

o = (i_)'_N,(V_)
+ (i_)M,(V,)+ O(1),
where N,(Uj) =- Uj ((A + p)r_r 0 + l_6,jr, kr,k - P60) ;

+,_,r_u,+ l_j(rov,+ r.,vi).

To solve for r and U, independent of frequency, we must solve the following two equations:

Y,(Vj) = O, (2.4)

M,(U_)=0. (2.5)
Equation (2.4) can be identified as an eigenvalue problem with three mutually or-
thogonal eigonvectors. Two of these eigenw_ctors correspond to two identical eigenvalues
and span a plane orthogonal to the ray. The third eigenvector is polarized parallel to the
ray.
Analytical expressions can be fou.d for the eigenvalues and le_(l to the eikotml
equation for each wavetype:

r,ir,i = (1),
t_s '

r,_r,_ =
(1), , (2.6)

vs
= y/_p
a represents the seismic velocity of tile wavetype polarized perpendicular to the
ray and ve = V/_ . 'the velocity of the wave polarized parallel to tile ray. Thus, in
isotropic, inhoinogeneous media, high frequency wavefields can be decoupled into two
independently traveling wavetypes.
A more detailed discussion on the derivation of the eikonal equation is provided, for
example, in (_erven_ (1987). Equation (2.5) will be used to evaluate the amplitude of
the wave.

2.2 Kinematic l_y Tracing Equations

The mathematical, straightforward way to derive the solution of tile eikonal equation
is called the method of characteristics.
1
r(xi),i r(x_),i = va(xj)

is a first order, non-linear partial differential equation of the Hamilton-Jacobi type. We

can rewrite tile equation as

1_,_
(,. _,)__ _ = o. (2.7)
If we introduce the slowness vector p,:
1
p_= r, ; Ipl- -,
V

the Hamilton-Jacobi equation takes the form

H(p,, x,) = nt (p_p,),,n (t).

v = 0, (2.8)
Using the method of characteristics (e.g., Bleistein, 1984), we obtain the f() llowmg
equations, which are called the ray trace equations:

du = v"dr, (2.9)
dx, 1
--
du = p, _ ' (2.x0)
dp__j.= . 10v (2.11)
du v "+10xi '

l
 The solution of equations (2.10) and (2.11) represents the trajectory xi and the
distribution of Pi along the ray as a function of a monotonically increasing independent
variable u. The solution xi(u) is called the raypath. Note that we are free to choose n.
If we set n equal to zero, then u is the traveltime v along the ray. If we assign to n the
value 1, (2.9) is of the form:
du = v dr = ds.

Here, s is called the arclength along the ray. s is measured from some reference point
on the ray. However, (2.9) suggests that the simplest form can be _hieved by setting n
equal to 2. For this specification, we follow Bleistein (1984) and set

du = v 2 dr = da. (2.12)

The nonphysical quantity a has units of [t_"g'h21

t time J" The ray trace system may now be
written in the convenient form"

dxi
--da = Pi ,
dpi 1 1
d"'a = "_('_),i ,
dr
--
da = v -2. (2.13)

The first equation in (2.13) does not contain any velocity dependence. As a conse-
quence, we are able to find simple analytic expressions for rays in some special situations.
For example, if the gradient of v -2 in the medium is constant, we can solve for the slow-
ness components pi analytically. The inverse of the velocity squared, the sloth, can then
be written in the form
1
v-_ - s(x, z) = soo+ s,i xi,
and the equation for the slowness vector yields:
1
p,(o)=p,(oo)+ _._(_-_o).
Using this equation, we can write for the coordinates and the traveltime along the raypath
the following expressions:

1 (a -
•,(o) = _,(oo)+ p,(_o)(_- _o)+ -4_,_ a0) 2
_(o) = _(_0)+ [_00+ _,,_,(o0)](_ - o0)
1
+ 5[_.,p,(o0)](_- _0)2
+ l_[S,, 2] (a - a0) 3. (2.14)

The simplicity of the solutions and the fact that polynomiais required to get slowness
8
 and traveltime can be evaluated very efficiently on the computer, make media with a
constant gradient in sloth very attractive.
This approach to solving for the slowness distribution and the raypath lacks direct
physical justification. However, the Hamilton-Jacobi equation (2.8) can also be derived
by applying classical mechanics theory (e.g., Goldstein, 1950, Chapter 9). Considering
the propagation of waves as movement of particles also leads to an expression equivalent
to the eikonal equation. In this sense, ray theory is only a classical-mechanics approxima-
tion of wave propagation. Therefore, observations obtained by using the ray-theoretical
approach outlined in this chapter are only valid if effects involving the wavelength of the
signal are negligible. This is the physical basis of the well-known restriction (see Bleis-
tein, 1984) that the length scale of the medium must be much larger than the wavelength
of the signal.

2.3 Dynamic Ray-Tracing Equations

The system of equations (2.13) can be used to evaluate slowness and traveltime along
a raypath. In this chapter, however, we advance one step further and try to compute
properties of the wavefield close to the ray. The cartesian coordinate system is not very
convenient for this application. Better suited are orthogonal coordinate systems that
move along the ray and whose basis vectors coincide with the polarisation vectors of P
and S waves. Such systems are described by (_erven3_ and Hron (1980) or Hubral (1980),
and have been used to investigate geometrical spreading and curvature along the ray.
For each ray, we introduce an orthogonal curvilinear coordinate system (qt, q2, s)
connected to the ray. In this so-called ray-centered coordinate system, the coordinate s
represents the monotonically increasing arclength along the ray. Parameters ql and q_
form a 2-D cartesian coordin :_tesystem in a plane orthogonal to the ray at fixed arclength.
As sketched in Figure 2.1, the basis vectors are the unit tangent t to the ray at given
s and the unit vectors _l and e2 perpendicular to the ray. They can be computed by
making use of the normal and binormal to the ray at fixed s and by integrating over the
torsion of the ray trajectory.
Any radius vector ri from the origin 0 of the coordinate system (ql = q2 = 0, s = so)
to a point S in the vicinity of the ray can now be expressed ill the form:

q2, = 0, + +
The vector basis (el, e2, _ is right-handed. The scale factors for this curvilinear coordinate
system are:
hq_ = hq2 =1 , h, = h ,
where h is given by:

ql =q2 =0 ql =q2 =0

If we define capital letter indices to have the values 1 and 2, we can rewrite this equation
9
 FIC. 2.1. The ray-centered coordinate system in 3-D uses as basis vectors the unit
tangent t to the ray at given s and the unit vectors el and e2 perpendicular to the ray.

as:

h_-l+q,
(v-, qK=O /21 ,
We are now prepared to express the earlier-derived eikonal equation (2.10) in ray-centered
coordinates:

= + +_ N =V2(q;,q_,_)" (2.10)
It is our intention to approximate the wavefront or the traveltime field close to a selected
central ray. In the ray-centered coordinate system this involves only a Taylor expansion
in the el and e2 directions. We will expand our expressions up to the second order in ql
and q2. Hence, the results of our computatiop,_ are exact only at the central ray itself,
and the approximation holds for only small values of ql and q2, To facilitate the next
derivations, some commonly-used notation (e.g. (_erven:_ [1987]) is introduced.

_(s) = [ _(q_,q2,s)]_.=0,

,,,(s)= .....Oq,
........,,,,.=o
[ Ov(ql, q2, s) ] '
[ O_.v(q,,q_,,)
]
For fixed arclength, the Taylor expansion of the velocity field close to a central ray is
1
v(ql , q2, s) _, v(s) + v,K qK + "_v,KL qK qt,.

10
 After nmltiplying by h2 and expanding to second order in qt and q2, (2.16) takes the
forln

+ + _s = v2(s ) v3(s ) V,KL q," qL. (2.17)

Equation (2.17) describes the traveltime field in the paraxial vicinity of the ray up to the
second order in ql and q2. The solution r(ql, q2, s) can be expressed as a Taylor series in

ql and q2. 1
r(ql, q2, s) = r(0, 0, s) + _ ql qJ M_j . (2.18)
Here we denote the symmetric matrix of second derivatives of tile traveltime field by
M(s),

: Oq,
Oqj
[ O'r(q,,q2,s) ]

The Taylor expansion of r(ql, q2, s) does not include a linear term (0r(_"))qK:0 because
wavefronts are orthogonal to rays in isotropic media. However, outside the central ray,
we find from equation (2.18) that

Or(ql,q_,s)
= Mts"qK,
Oqt
Or(ql , q2, s) Or(O, O,s) 1 dMta
= + ql q3 " .
Os Os "2 ds

These relations are inserted in (2.17) and tile assumption is made that h 2 is close to
unity, i.e.,

q,. (v-' 0q_) ql¢.._-O <<1. (2.i9)

Equation (2.19) defines tile paraxial vicinity of the ray. Under this assumption, the
paraxial approximation of the eikonal equation in ray-centered coordinates reduces to
the simple equation
elM I V
d"_ + vM_ + v _ = O, (2.20)
or, if we choose a (see equation (2.12)) as the monotonically increasing parameter along
the ray:
elM M2 1_.V
d--_"+ + v3 = O, (2.21)
where

v = v j(o)= [ 02v(q' , q'_,a) ] q .-0

is the matrix of second derivatives of the velocity field at points along the ray under
investigation. Equation (2.21), the so-called dynamic ray tracing equation, is a first-

11
order differential equation that nmst be solved along a known ray. This equation is a
nonlinear Ricatti-type differential equation and has no closed-form solution.

2.4 Linearizing the Dynamic Ray Tracing Equations

To facilitate the procedure of solving the Ricatti equation, several authors (e.g.,
_ervenS, and Hron [1980]) proposed two substitutions to obtain a new set of linear dif-
ferential equations of first-order. Following their lead, we eliminate the nonlinear term
by setting
dQ Q-1 (2.22)
M(a) = -d--_a .

Introducing the (2 x 2)- matrix P = _ , one obtains a set of eight linear, first-order

differential equations dQ = p ; dP _ lv
d"a" d'a"= v3 Q' (2.23)
If we define matrices X and S by

then equation
X=
(0o)(0
Q_ Q2_
Pll
P21 P22

(2.23) can be represented

s=

in the form
-
0

_v,21
il l
-_v,12
00)
o

-- _V,22
0 1
0 0
0 0
,

d X = SX.
da (2.24)

Both columns of X must satisfy the same set of equations, in other words, we must only
solve one system of four differential equations twice with different initial conditions.
The solution space of a system of the form

d Y=Sy,
da
with Y being a (4 x 1) matrix, is spanned by four linearly independent solution vectors.
If we solve the system for four linearly independent intrinsic initial choices of Y(ao) as,
for example,

{Yl(a°)'Y2(a°)'Ya(a°)'Y't(a°)
then the solution space is determined.
}= o0
0
'
(o)(o)(0)}
0
0
'

For convenience, the four solution vectors are

o1
0
' o0
1
'

grouped together as columns of the fundamental matrix II(ao, a). Every solution W(a)
can now be computed by a simple matrix multiplication of I_ with a (4 x 1) matrix C(a0)
12 I"
of specific initial conditions:

W(a)= rl(a0,a).C(a0).
Thus, as soon as II(a0, a) is known, we can fil,_dsolutions of the dynamic ray tracing sys-
tem analytically for any initial condition specified at a0, without repeating the dynamic
ray tracing,
This is the general formulation of dynamic ray tracing. However, our focus of atten-
tion is not the most general three-dimensional case. We make use of (2.23) as a starting
point to investigate the behavior of some important physical quantities in more restricted
media.

2.5 Physical Meaning of Q and P

Before we start to interpret the meaning of Q and P, we should recall the three
important coordinate systems used in the derivations. This choice of coordinates, as seen
in Figure (2.2), will be applied throughout the following sections.

• Cartesian coordinates xi

• Ray coordinates "yi= (')'l: 7_, "Ya= a). A/!= ('Yl,"y2)may be identified as the takeoff
angle and the azimuth of the ray at an initial point and specify a ray in a three-
dimensional medium. The monotonically increasing parameter along the ray, a,
can be evaluated using

a = ao +
L
v2dr = ao +
L yds.

• Orthogonal ray-centered coordinates q_ = (ql,q2,qa = s) along the ray. ql and q_

span a cartesian coordinate system perpendicular to the ray at fixed arclength s.

So far, the matrices P and Q appear as mathematical, abstract quantities. However,

(_erven_, (1987) shows that they have associated physical meanings. Specifically, if at
any point of the ray, the (2 x 2) matrices P and Q are initialized as the transformation
matrices from ray parameters to ray-centered components of the slowness vector and from
ray parameters to ray-centered coordinates, then they will represent these transformation
matrices as long as the velocity varies continuously:

0_,,] (2.25)

it is worthwhile to empha,_izethe meaning of the matrix elements of P and q. For

example, according to the first equation in (2.25), Qll is the change of the trajectory
in the ell (lirection for a infinitesimal variation _71 ia the takeoff angle. This change is
m(,a,_ur(,d perl)endi(:uhtr to the unperturlwd r_kvin the el direction Slmlla'ly, Pl_ is the
change of the first compoacnt of the slowness vector due to an infinitesimal change in
13
 azimuth Y2
'X3

FIG. 2.2. Our specific choices of coordinate systems, ql and qa correspond to tile ray-
centered coordinate systexn, "rtand 72 are the ray coordinates, x_, x_ and :ra represent
the cartesian coordinate frame.

azimuth and so on. in the next section, these meanings are illustrated and discussed in
more detail.
Once P and Q are evaluated along the ray, they can be used to compute additional
physical properties along the ray. Recalling the introduction of both matrices (2.22), it
is straightforward to show how the matrix of second clerivatiw_sof the traveltime field M
is obtained, lu a similar manner, the (2 × 2) matrices K and R of curvature and rmlii of
curvttture of thewavefront coxBl)uted.
tire
dq
M = d-'g'q -I ,
K = v(a), M ,
R = K -l ,
1.t
2.6 Two- and Two-And-One-ltalf-D|menslonal Models

A significant simplification call be achieved by considering a two dimensional model.

One ray coordinate and tile initial point is now sufficient to define a ray, and any point
S clone to a ray can be described by two ray-centered coordinates. This situation is
sketched in Figure 2.3.

0
x1

S(s,n)
n

x3

FIG. 2.3. A ray in a 2D-medium '8 l _ well defined by a starting point O and one ray
coordinate. Here this coordinate is chosen to be the takeoff angle. Any point S close to
the ray is a fimction of the ray-centered coordinate s and the distance perpendicular to
the ray u.

In this 2-D ea.se, q and P are re(luted to simple scalars q and p because all derivatiw,s
with respect to :t_ and "_ vanish. The dynamic ray tracing syst(,m (2.23) reduces to only
two linear, ordinary differential equations:

dq dp 1
d"-a= p ' ,l-'_ = --_v,llq . (2.26)

Whenever the gradient of v -_ in the medium is constant, this system h_m analytic
solutions, which are shown explicitly in Appendix A or in Hale (1991).
There is one severe disadvantage in using the 2-D approach. The elimination of the
second out-of-l)lane coordinate ilul)licitly introduces an infinite extension of tile l)hysical
l)r¢)l)erties of the som'ce in the ;r_ direction. The point sourc(_,is thereby converted into a
line source which gcm,ratt's ('ylin¢lrical wav(,fi'onts. The geometrical sl)r('a(ling and l)lu_e
behavior of cylindrical waves diff_,rs signiticantly from those (hw to a point source. The
amplitude ('aa I)e r(,mow,cl i)y consid(_ring a thr(,¢,-(limensi¢)md medium with
lat(;ral and v(,rtical v(,locity variatioas rc,,_tri(,t_,(lto a sp¢,('ifi(,(ll)hm(,. No v(,h)city variation
is allow(_cliu the dir_,c'tiou normal to this lflaao. This nituati()u, _ illustrat(,d in Figure 2.4,
15
is referred to ms two.and.one.half dimensional. Any ray with its initial slowness vector
in the (xl,x3) plane will stay in this plane. Thus, except for the three-dimensional
spreaxling, the problem of interest is essentially two-dimensional, Two excellent papers
on this subject are Bleisteiu (1986) and Docherty (1987). The geologic interpretation
of this s't!luatlon
........ is a seismic expenmeut carried out in the dip direct|on. It is obvious
that derivatives of tile velocity field with respect to q,_are zero if tile basis vector _ of
Figure 2.1 points ill the x2 direction.

FIG. 2.4. Example of a common-striko medium. Dark slmdiug dellotos a high seismic
wave velocity. The velocity field may only change in .rl and xa ditetti"
' era, with no velocity
¢,hanges in the x2 direction. Any ray with its initial slowness vector in the (xt,x:j) plane
will stay in this plane, Thus, except for the thrcc-dimonsiomd spreading, tile problem of
iutorest is essentially two-dimcusiomd. This sitmttion is referrod to ,Ls two.and.one.half
dimensional,

16
 Equation (2.23) reduces significantly for tile 2.5 DcMe,

= Pta = - v,ttQt2 (2,27)

= =0
-P l =0
Let us put a point source into this medium. The choice of coordinates is the same
as that sketched in Figure (2.2), Imagine a ray that starts with the initial slowness
vector perpendicular to the xa axis. This ray will stay in a plane perpendicular to the
z_ axis because there are no changes of velocity in the xa direction. This also implies
that _ always points in the out-of-plane direction. Now, we infinitesimally vary the ray
coordinates and consider the values of our transformation matrices Q and P at the source
itself, Qll is the change of the trajectory in the qt direction for a infinitesimal variation
b'Ylin the takeoff angle. Because we deal with a point source, we certainly will not find
any offset 6ql at the source itself. The same argument holds for all elements QIJ, On the
other hand, the direction of the ray and thus the slowness vector are obviously functions
of _'Ytand 6_'a, The change of the first ray-centered component of t,he slowness vector by
varying the takeoff angle by _'II is _Pl ffi _671. Similarly we can get P_a. However, if
we change the azimuth, we do not get an3; change in the first component, The same is
true for of variations of the takeoff angle on the pa component In summary, we find:

Q.(oo) = 0;
= o;
P,j(oo) = 0;
1
r'll(ao) - ----- '
-,,(ao)'
1
P,n(o'o) : t'(#o)sin'yl, (2,28)

To verify these results qualitatively, we can compute the matrix of curvature at the
i)oint source K(oo),
X(ao) = V(ao), P(ao)Q-i(ao) _ oo .
As _,XlWCt_,d,we find infinite curvature of tlw waveffont at the source.
Applying the initial conditions, (2.28) reduces the systom (2.27) to two equa-
tions: Clearly, the s_,t of equations for Qat att{l Pat yields the solutions Q_t(a) = 0
and l_t(#) = 0, The same, though less obvious, is true for Qta att(l Pta. This set rol).
resents a hontogtmeous system with zero initial conditions, Siml)h, intograti,n h_,lps to
solve for Q_t att(l p,n.

17
 Qz_(a) = ff o Pzz(a')da',
= ,%(oo)(o- oo).
The solution for Qtt and Pit is the same as for tile tw_dimensional case. in other
words, if we have available a routine which evaluates the dynamic ray-tr_e equation in
two dimensions', then a simple computation for Qz_ solves our two-and-half-dimensional
problem for a point source. Specifically, this correction can be easily introduced if we use
as our monotonically increasing parameter along the ray. A similar discussion can be
found in (_erven_ (1987).
As shown in Chapter 4, the above derived properties of the 2.5-D dynamic ray
tr_ing imply the same out.of-plane spre_ing correction for the ray amplitudes as that
derived by Bieisteln (1980).

!8
 Chapter 3

MODEL REPRESENTATION

The triangulation method for constructing and representing subsurface models is

mainly built upon the research of Hale and Cohen (199!). In order to trace a broader
variety of ray modes and to allow VSP experiments, I develop a more dyn_fie way to
deflnlng interfaces than ill the previous approach. In this cllapter, after giving some moti-
vation for using tile triangulation method, I describe and illustrate the new modifications
to tile progrmn GBmod.

3.1 Motivation

There exist a variety of methods to represent geologic subsurf_e models in the

computer. One p_ibility would be to introduce a uniform rectangular-grid sampling of
sui_urface parameters such as th' spell of _ismic waves. To understand the behavior of
physical phenomena, such as the propagation of sound in the subsurface, one has to apply
a mathematical operator on this grid. This appro_h of.linite.differcncing the wavefleld
is inefficient and may involve millions of floating point calculations. Less cost effective is
the numerical integration of the sty-trace equations on these uniformly sampled grids to
illw'stigate the high-frequency part of the wavefleld. Most comptlter software programs
designed to study the propagation of seismic energy using the method of ray tracing,
however, now use a more efficient approach. Their teclanique is to decompose tile model
isjto layers with the same seismic velocity or the,same gradient in seismic velocity. The
itlterfawesseparating these layers are represented mathematic,ally _ smooth, single- valued
ftttlctions of depth. UtJorttmately, the earth's interior and tmpeclally most the areas of
itit_,restto exploration geophysicists are not at all smooth. After geologically simple and
accessible sedimetltary regions hay, been explored for hydroc'arbon resources, Jilterest
itJc'reMesill understaliditlg the propagation of seismic waves in comple× geology such as
that sketched ill Figure 3.1.
Triaslgles arc well suited to reprcsesitmod_,l_of this complexity. Overhanging struc-
tures, fatdted interfa_'es slid lens-type struettlres can be easily included il_to the model.
Nloreover, such a chara_'tcrization allows olie to store the lmranteters of the model in
stwh a way, that valu_,sozi th,, _th4,rsid,_of the,edges of the triangle can be determim_d
withot_t searchiitg throtigh the, mod_,l. This property is especially valt_able for efltcie_t
ray trax_i_g.

19
 I................
...... O

F1¢3.3,1. Idealized diapiric salt structure showing common types of hydrocarbon traps
(after %arpock and Bischke, 1990).

3.2 Reflection and 'Pransmlssion Sequences

Hale (1991) demonstrated how complicated models such as tile salt dome in Fig-
ure 3.2 can be triangulated for efficient ray tracing. To yield analytic solutions to kine-
matic and dynamic ray-tracing equations (see Chapter 2), tile velocity field in each tri-
angle is constrained to have a constaut gra_lient in sloth. The velocity field hms to be
prest'ribedby _slgning velocity values at the vertices of each triangle or, alternatively,
by providing tile sloth values for an eutire geologic block bounded by interfaces, The
latter appro_h h_ been used to define the velocity field of the salt dome in the left part
of Figure 3.2.
The previous version of the modeling code GBmod was designed to generate re-
flections from only a siugle selected reflector during auy given run of the program, The
appro_'h preseuted here is more dynamic. The generation of multiple raypaths, for exam-
pie, requires a certain flexibility in the dcfluition of the iaterf_es. Previously, any given
iutcrface could either reflect or transmit seismic energy, but not both; this deflnitiou of
iuterfaees did not allow oue to trace multiple reflections. As il!ustratcd in Figure 3.3, to
ol_tain selected multiples, a s_,queaceof reflections and transmissious has to be dcflued
at etwh interface of igiterest. I have installed this new fc,ttturein the ray-tracing code.
For each reflector, a 1_]:lcctiltgsequeT_cceaa uow be deflxlcd by the user of this modeling
software.
The two differ,,ut families of nmltiplc,sin Figure 3,3 have been created by instructions
similar to those iu other Uuix 13ouruecommand shells. To geueratt, the multiple raypath
between interface 2 and 3 show|t ,m the left side of the figure, the shell parameter
statements are the foll,_wiug:

2O
 Distance(kin)
0 I 2 3 4 5 6
I I,
I , I _ _ I I
ilm[],r , i .............

4
FIG. 3,2. TrianguhLt(,(l model of a salt (l()me. Note that tilt; mo(lel extends to the right of
the portio,_ displayed here. The slla(iing denotes the seismic velocity. Tile specific form
and the a_signing of the velocity field are discussed in the text.

21
 Distance(km)
0 0.5 1.0 1.5 2.0
0 _..... _:

C_

FIC. 3.3. Raypaths for two different families of multiples. Interfaces of interest are
labeled. The shading of the different layers indicates that the seismic wave velocity
varies with both depth and lateral position.

22
 refseq=2,0,100 \ # refl./transm, sequence for interface 2
refseq=3,1,1 \ # refl./transm, sequence for interface 3

The first number after tile equality sign denotes tile interface', the following numbers
signify the sequence of reflections and transmissions at this interface: "0" denotes trans-
mission, "1" reflection. If no reflection sequence is defined for a reflector, tl,e interface
is considered to be only transmitting. In the above case, the first hit on interface 2 will
lead to a transmission of the seismic energy, and the second arrival generates a reflection.
Finally, the ray propagates towards the surface. For the type of multiple displayed on
the right side of Figure 3.3, the following statements are used:

refseq=l,O,l,O \ # refl./transm, sequence for interface I

refseq=2,1,O,O \ # refl./transm,sequence for interface 2
refseq=3,1 \ # refl./transm,sequence for interface 3

3.3 VSP Modeling Experiment

Recordings from VSP experiments are routinely used to improve the estimation of
subsurface properties close to a well. VSP data sets are often more difficult to interpret
than are surface data. While common midpoint gathers of surface data bear a certain
similarity to actual subsurface structure, the patterns in VSP sections are quite different
from the actual configuration in the subsurface. Generation of VSP synthetic data helps
to overcome this interpretation problem by providing a tool to help identify events on
recorded VSP seismic sections and to check the validity of existing imaging software.

shot _ well

FIG. 3.4. Fault-model VSP experiment with one shot-point. The generated seismic
wavefield is recorded by geophones in a well. Figures 3.8 and 3.9 show synthetic data
and physical-modeling data for this model.

23
 The software described in this paper can be used to generate data for VSP experi-
ments. Consider the fault model shown in Figure 3.4. This model was used by Jaramillo
(1993) of the VSP research group at Colorado School of Mines both to analyze the quality
of the generated synthetic and physical model data and to use the modeling software as
an interpretation tool for the Yucca Mountain Project, The ray-tracing code provides an
additional source of model data.

0 2 distance{kin) 4 6
i =, I ...... i
O- v
y

i,

, 2-
i £

in

FIG,3.5. One family of rays generated by the modeling program. The event is recorded
by receivers in the vertical borehole. The corresponding reflection in the synthetic seismic
section in Figure 3.8 is labeled i.

Figure 3.5 shows one family of rays generated by tile modeling software. Two re-
flections are considered in the ray propagation: the first is at the left boundary of the
model, and the second at the hanging wall. The rays finally are recorded by receivers in
the borehole.
If the left boundary of the nlodel is labeled 1, the faulted interface 2, and the
borehole 3, the UNIX command shell for this experiment contains the lines:
24
 a b

o d
FIG. 3.6. Four different ray types are shown. In example a, both direct arrivals and
primaries from the foot wall and the hanging wall are present, b displays bottom reflec-
tions and direct arri'"-_ls,c surface multiples and d bottom multiples. The corresponding
seismic events are labeled accordingly in the seismic section shown in Figure 3.8. Note
that although the ray density (i.e., increment in takeoff angle) at the source is identical
in euch figure, the ray coverage at the borehole is different.

25
 0 4 tJ | 4

0 4 4
O-

g h
FIG. 3.7. The modeling software allows tracing of rayl_aths that reflect fi'om tile sides of
the model and that cross the borehoh;. Rays that are reflected back from the boundary of
the model, and pa.,_sthe receiver line a second time are finally recorded by the geophones
in the borehole (see events g and h in Figure 3.8. As can be seen in e and f, rays with
overcritical('
' incidence are not transmitted.

26
 depthof receiver(kin)
1 2 3

FIG. 3.8. Syxlthetic data generated by tile extexvlecl versioa of Hale's acoustic mod,;l-
ing progx'axn, Tile ray families px'esented in the previous two flgux'esg_,xwrate seismic
r_,sponses that are labelecl accorciingly here.

27
 depthof receiver(km)
1 2 3

FIO, 3,9, Physical modeling data set after apl)lying a waw,form filter to obtain the full
P-wave amplitude contributi(, )n to tile data, Cmnl)arison with the synthetic section helps
identify most of the dominant coherent eventM.

28
 refseq=t,l \ # refl./transm, sequence for the left boundary
refseq=2,t \ # refl./transm, sequence for the fault
refseq=S,-t \ # refl./transm./stop sequence for the borehole

I have previously introduced the designations, "1" for reflection and "0" for transmis-
sion; here, I introduce "-1" to denote the stopping of tile ray. Thus, the first intersection
of tile raypath with tile boreho!e terminates the tr_tng of this ray. Information including
the traveltime, corresponding amplitude, and llidex of the stopping interface is stored in
a file. By default, the rays are stopped at the boundaries of the model. Only rays with
the correct stopping index contribute to the events on the seismic section and, optionally,
can be displayed. Another feature allows one to select only direct arrivals or only pri-
maries. ThelJe new enhancements proved to be helpful ill interpreting tile physical data
gathered in the Yucca Mountain Project,
The synthetic seismic response for tile specific event shown in Figure 3.5 is labeled
event I in Figure 3.8. This reflection can also be identified in the recorded P.wave
physical modeling data (Figure 3.9). In Figures 3.6 and 3.7, a series of ray families is
displayed for tile same geometry. The letter labels on the models ill Figures 3.6 and 3.7
correspond to tile evellts in Figure 3.8. In this form, one can conveniently compare the
synthetic section with tile physical modeling data to aid in identifying individual events.
The modeling software now allows tracing of all rays of int,,rest, it is possible to
specify reflections from tile vertical and horizontal boundaries of tile model, features that
are unavoidably recorded in physical modeling experiments; multiples call be conveniently
all,fined; and rays may also be traced across the borehole before they encounter the
re,.l,iw, r lille a second time and arc recorded (see ew, nts g and h).
Dilferelwes in Figllres 3.8 and 3.9 call be attributed to a number of factors. Among
tlwm is the fact that tlw ray.trace code presently generates only a_'oustic events, whereas
the physical model is 2-D ehmtic. For this VSP experiment, the COmlmter run time,
incltlding the automated triallgtllation, tile trowing of ,lille different ray modes, and tile
gt,m,ratioll of tlw syllthctic seismic se,,tions took 23 se,'oljds of CPU time on an IBM
RISC System/G000 Model 520 workstation,

29
 Chapter 4

APPLICATION OF DYNAMIC RAY TRACING: RAY DATA AND

FRESNEL VOLUMES

In this chapter, the theory introduced in Chapter 2 will be applied to tile dynamic
ray-tr_ing modeling software. Tile media considered are two-dimension_ or two-mtd-
one-h_f.dinlensional. Starting with elastic wave propagation, 1 finally r,_uce the com.
plexity of modeling by considering only acoustic media, The main differenct_ between
elMtic sad acoustic waw, propagation are, however, discu_ed. Tile dynamic ray tr_ing
data can be applied to both the _oustic and the elMtic case,

4.1 Phase-front Curvature and Paraxlal Traveltlmea

The le_ling.order solution of the eikon_ _uation in ray-centered coordinates (2,17)

can be used to compute lutditional quantitie_ along the rays, Equation (2.22) shows that
the phase-front curvature matrix M can be evaluated by M ffi pQ-I with Q and P
drtermined by

P(#) ) ffi rl(#0,o),

(Q(#) P(#o) )
(Q(a°) '
In tw_>.dimensionalco:nputations, the quantities M, p and q are the scalar equivalents
of matrices M, P and Q.
Knowledge of the pha.qe-frontcurvature M is exploited for v_rious applications. For
example, it can be used to compute the traveltime in the vicinity of tire ray. This is of
practical importance for the nlodcli|tg software because, in general, rays do not emerge
xac tly at the po_ition of the receivers.
t

Consider Figure (4.i) showing receiver position R and the endpoint of a ray E. _s
dcnoteu the difft,rc:tce in arclcllgth, and n is the perpendicular distance from the receiver
to the ray. Assuming the ¢lymunicray-triwing dattt to be known at potltt E, the _ro_iat
travcltime T(R, E) can be approxillmted I,y:

T(R,E) _ T(E)+ _
OsAs + I ---O'_T
(As)2+ !
20s_ 2 .2M(E)

T(E)+ As- ; ; +I
,,(Ei 2 ,'_(_7 0"-"_ 2 ''_ M(E). (4.1)

The Taylor expasision (4.1) inw,lw,s the gradient of velocity in ray direction. While
this term is often :wglected in the extrapolation of traw_ltimes iu smooth media, it is
important if one considers more complex models, in these media, rays often emerge at the
surftwe with angles close to horizolittd and yield significant values of ,_s, As discussed in
30
 FIG;.4.1. The approximate traveltime at any point S tn the vicinity of the r_ can be
evahtated by umiltgthe dynamic r_ tracing data at point Os.

Ch,tpter 5, the same is trtte for tile Gausstan beam method, becau_ this method reqt,ire,
traveltime-extrapolation far away from the ray if strong velocity variations are print.
1 implemented this term lit the program GBmod. However,despite th_ modifications,
note that this approach is valid only for a smooth velocity field tn the vicinity of ray ends
altd receivers. Strong velocity gr_ie|tts or jttmpMtntvelocity lead to ln_cttrate partial
traveltime data.

4.2 Ray Amplitudes

ht Chapter 2, I showed that, to first order accttr,wy, high-freqttezlt'ypressure waves

(P-waves) and shear wttves (S-waves) propagate independently through smooth m_la,
In other words, the amplitttde ve¢,t¢,rU_°)(ffiU,) in eqtiation (2.3) can be reprem,_nt_ in
the f_Jrn|:
= . + . + °'
.i,
where_i,_ and iHpan theray-_'entered coordinate system,Ret'all thatthi_Hy_t_m,
tc_gether withthekhwmatic_z1_! dyn_mic,.ray tr_ciz_g eqli_tlons,_'_nbe derived from
_'qlttttioil(2,4), Sew,rnl textbooks (e,g,, Aki allcl Ric'hitrds, 1980',(_erveltp, 1985_t)show
_,xpiicitly how e¢luati(m (2,5) _,_tnb_, used to solve for the co|npone|tts UI_),U__ altd
U;I_. This derivati_ ix t,xliotts a_td tint_,-¢.onsttnti|tgand will not be repeated here. The

3!
8ol.ti.n obtaim,d t. of tile fidlowt.g for.l:

Here, i,, p a.d J de.ore velocity, de.slty a.d the ray Jacobla., r_pe(,tlvely, Velocity and
the ray J_obla. are different for P-waves _d S-wav_, U(q)(#o) Is a q.a.tlty I.fl.enc_
b) the _.rce radlatio, patter., The ve_'torial_qdltude for P-wa_s is

U,= U_')i,,,
' ' _ l_P.-wa_,
L J. Jp__.__,
The S-wave c_ be repr_.tmt by

[:, - Ui,),_,+ _J"),_,;,,= ,,,...., J = ,&...,

I. the ray-ce.ter_l coordl||ate system (:i, _, i), the S-wav_ havl, two d_o.pled compo.
nents perpeudlc.h..r to the ray,
In _o.stlc media of coilstailt de.sity, the solutio, of the tra.sport equ&tlo, h_ to
|w modified, b_a._ it is pr_ure (or th,, p.teutial .f u,) that is o_rve_l,

Dy.amic ray trwi.g is well ..iti,d to ha.dle tile elMtlc, three-dlme.sio.al (,_, I. f_wt,
re.st a.thors deal excl.sively with tile elastic c_. Computationally, ti|e evaluation
.f _! a.d _ alI|(ltl|e i.trod.ctlo, of boti| P-wav_ a.d $-waw, veloeltlt_ are the main
c'o.shleratlous .,q.ired i. extc.mll.g the ._'o.stle c_, However,to r_luce the complexity
.f the foilowi.g deri_tlo., a.d t(_t tile .l)l)lh'ahlllty of tile method for str.ct.r_ of
higher complexity, we restrict o.rselw.s t. lu,..stie w,wefiehls,
I.._o.stic media, the first wrm of tlm rIkvwries yh,lds

TI., ray J_v,ol,ia. J a.d it. relati,.= to the res.lt, of the dy.ami(' r=¢ytr_u'i,q{arc,derlw,d
i. Appe.dlx B, N.te that the cht)ice .f c,)ordi.ate systems dilfers from tlmse i. tile
liwrat.re (for ¢_()ml).t=ttiomdr(,=_().s, # a.d .or the ar(,Lh..Kths is .wd = the r...i.g
parameteralo.g tl., r.y), C(m.ideri.g the .,s.lt derivedi. the .l)l)emlix, the first term
i. t]., ray _,rie_yi,,](l_

"(#)= _ ,,(,,u)'_,1,t Q(.) ' L,(#.),,,xp(-,,_¢,

- r(.))),
H,'.('e, if we k._,w *'("o), tit(, w,l.city ,tl tl., m.tr,'e, tit,, w,h..tly ,tt the r.y e.d _,(#) _..1
32
the dynamic l,_v ill.acingresults Q and P, we cam approxinlately solve tile wave eqtiatlon
along a ray and in its pal,axial vicinity, The corresponding _iuation for two-di.lensiond
m_lla Invol_ the _alal, q.antlty q evaluat_ from C_l.atlon (2,26) instem| of the matrix
q.
.(.)..i_.i_..o,,,(.,,(,_,(,))).
_I,,(.o)_!v(")
(,.,)
4.8 Attenuation

Quality factor
0 40 80 120 160 200

_ u
1'5i_, 'i ....... [ U

''0"
'.... ['"; . U

0 I00 4OO tm Im tO_

Fla, 4,2, Syntheti,_ _IsmoKram. disl)lay,.d _ a f.nctlol= of q.ality l'm.torQ, The receiw,r
is situated is a hoinc)gcii_)tlsm_!itlnl, 5 km fronl a poillt source, Velocity of the m_iium
is 2,5 kins, Only =io|=,,a.salallen.alien is evahmted. The solid llne in the right di_r_t
represents allalytic'ally comlmted pressure amplit.des at p,,ak frequency', the tl,iangl_
denote modded p,.tk-rrc,qt.,n,T anll)litt.h,s. The amplitudes are _'aled l,elatlve to the
amplitudev,dtleobtailwdill all non.al_ur!Jingll|_lilull.

Re('or(h,(lHeihnii('{hztaare iiilhi,,n(',,,Ihy _|i,,ht_ticI)ropertit,sof the sllt_l|rt'iwe, l

_d(h,dan Ol)tiOll th.t =dlow.oile t{,)_siKn a ('()ll_ttLlitattvnltation vl_lile to e!i4tJi geologic
bh-'k i)oull(le(lby tllli'l.fttl'l'N, The tlllt)h, lnl.,ltt_tion of ilOll(,all_tal al_orl)tion (t.e., oltly tile
re(Jill,tilth of tlie fUlllllitlllte i_ ('on.i(h,r,,(l, iljl_t' ¢'Jlall_t'_i dUO to &ttl_lllUttion al,e lle_lecte(I)
illtO the i)roKr!_lll GBlli()(I il_ iI!alle l,itt{'iellt Jl.y ilitro(lllctltK a ('Oliii)lex v,q(_(.ity IINtll_ the
D,:bl/e app_v_.l{_tio. (,%1
iilh,r, 1983),

- _=---. (1+ --),

33
 Qualityfactor
0 25O 5O0
• F ,rE
': iI
I '
IL

: !
I

i_ ' = '

IT. t _D_1)-D
i ,
DID)))j)I'I'D
l
!. Ii i
I
'_
: I
tl
. .,j.j

FIG, 4,3, Sy==theticH_,ismograms di.I)layed= a fl===ctioa of (ll=_dityfactorQ. The receiver

is situated i==_ h(m=¢)gem,ot=s medittm, ,5 km fro=== a poit=t su.rce, A zero-l)h=e Ri¢_ker
wavelet is propagated from tlle sot|roe to the receivers, Velocity of the medium is 2..5
kin/s, Causal absorl)tio|=is ili¢'l!Id,,daI.l disturbs the Rickc,rwavelet For low values of
Q, one obserw,s a siK|dflca.t time-delay aild wave-f,)rm dist,)rtion,

34
where v,t.l is tile specified real-valued seisnlic velocity, and Q is tile quality factor for
dissipative wave propagation. For _ denoting radial frequency and s, the monotonically
increasing arclength along tile ray, the amplitude decay Ad,_ due to attenuation is given
by

ad'*"' = exP (-_/,, "2Qt,,,.,

l ds'_ffiexp(-_t')'
]
For a constant Q in each triangle, the global absorption factor
IgN

t'ffi f, o 2Qv,o.a
1 dsffi _i=l t,

can be obtained as a sinlple by-product of the ray tracing. Here, N is tile number
of triangles traversed and Qi and ti are the quality factor and the traveltime within
the ith triangle. Figure 4.2 is a display of the _istnie response of a homogeneous, ab-
sorbing medium with vma = 2.5 km/s. The receiver is situated 5 km from the source.
Each trace corresponds to a different value of quality factor Q. The right diagram shows
the dominant-frequency amplitude w_rsusQ. The solid line represents the analytically
computed values, the triangle symbols denote results computed by the Gaussian beam
modeling. The values are normalized by setting the amplitude for the perfectly elastic
medium to unity. Note that the accuracy increases with decreasing absorption in the
medium, However, even for small values of Q, the comparison is satisfactory,
Multiplication of the amplitudes evaluated for non-absorbing media by the ampli-
tude decay factor Ad,_,,yin the frequency domain yields noncausal results. I implemented
a phase-correction proposed by (_erven_ (1987) to simulate causal absorption (see Fig-
ure 4,3), For decreasing Q, one recognizes a similar decay in amplitude as in the noncausal
case, but also a del_y of the causal signal towards larger traveltime. A more detailed ex-
amination shows that the signals broaden and become slightly asymmetric. The efficiency
of the noncausal treatment in dissipative models is essentially the same as in perfectly
ehu_tic models. The computational cost of including causal attenuation effects, however,
is significantly higher; therefore, in the Gaussian beam modeling program GBmod, the
causal feature is optional.

4.4 Two-And.One-Half-Dimensional Spreading

Ih the program, I included a computationally sitnple procedure to correct for gee.

metrical spreading for a point source in the medium. Previously, the modeling yielded
aml_litudes for a line source with cylindrical-wave expansion. I used the three.dimensional
dyuami¢, ray-tracing equation to derive ' the n_te
,'.s sary amplitude ¢_orreettc
' • '_n (see Chapter
2). Consider variations of the properties of the medium to be restricted to the (:_t,x:,)-
l_lane, with no variations in the xa-direction allowed. As derived in equation (4.3), the
pressure amplitude is anti.prolmrtiotml to the square-root of the determinat of matrix

35
 Receiver-Source
Distance(km) Receiver-Source
Distence(km)
5 10 5 10
0 " 0

1 - ...... 1+ I =!r"_--.

°
E2
P
_
-_1
2....
. D

3. "-- 3. " +

FIG. 4.4. Seismic sections generated by Gaussian beam modeling in homogeneous media.
The right side simulates the energy spreading due to a point source; the left time section
shows the response due to a line source. Below, the seismic traces of each section are
added to help see the difference in amplitude decay.

36
Q, In 2.5-D media, this expression reduces to

Using equation (2.29), the factor that multiplies the original two-dimensional solution is
proportional to

(I£ )-"
v ds = (a - ao)-½,

where a is a running parameter of units [length2/time], and s is the arclength increasing

from so to s along the ray. Figure 4.4 emphasizes the difference in amplitude decay for
homogeneous 2-D and 2.5-D media. To help see the spreading behavior, the 10 traces
of each seismic section are summed and displayed in the lower part of the figure. The
amplitude decay is proportional to r-½ for the line source and r -1 for the point source,
where r is the distance traveled.

4.5 Fresnel-Volume Ray Tracing

The term Fcesnel.volume was introduced by Kravtsov and Orlov (1990) to generalize
the concept of Fresnel-zones. These authors suggested that rays be considered as physical
objects rather than volumeless trajectories. Stated differently, all points within a region
in the vicinity of the ray, the Fresnel-volume, influence the wave propagation associated
with this ray. As illustrated in Figure 4.5, a point F belongs to the Fresnel-volume of the

FIG. 4.5. The Fresnel-volume for a ray from A to B. The point OF marks tile intersection
of the ray with the plane E. Points such as F influence the wavefield along the ray if the
Fresnel condition (4.5) is satisfied.

37
 ray from A to B if the Presnel condition (4.5) is satisfied (see also _ervenp and Soares
[1992]), that is
1
It(f, A) + r(f, B) - r(B,A)[ <: _T. (4.5)
Here, T denotes the dominant period of the signal, and r(F,A), r(F, B) and r(B,A) are
the traveltimes from F to A, F to B and B to A, respectively. Fresnel-volumes depend
on the velocity field close to the ray as well as on the dominant frequency of the signal.
I used recent work by (_erven:_and Soares (1992) to supplement the dynamic ray-
tracing routine in the modeling code GBmod with an option to compute the two-
dimensional paraxial approximation of the Fresnel-volume. This may be performed ef-
ficiently and accurately by using the elements of the ray-propagator matrix, which are
evaluated in the dynamic ray tracing. Fresnel-volume ray tracing (or physical ray trac.

Distance
(km)
0 5 10 15 20

10

FIG. 4.6. The Fresnel-volume for a ray traveling through an inhomogeneous medium. A
signal frequency of 5 Hz is considered. Intersections of the Fresnel-volume with the edges
of the lens and the fault indicate a violation of the ray-theoretical assumptions.

ing) offers several applications for the field of exploration geophysics. For example, study
of the resolution of the seismic method can be extended to complicated media, and the
validity conditions of ray methods can be stated explicitly in terms of the radius of the
Fresnel-zone at any point along a specified raypath (see Kravtsov and Orlov, 1990). In
Figure 4.6, the plot of the Fresnel-volume (actually, Fresnel area, for these 2-D models)
for a signal frequency of 5 Hz indicates that the ray-theoretical a,_sumptions are violated
38
becausetheedgesofthefaultand thevelocitylensiIltersect
theFresnel-volume.The
wavefleld
alongtherayis,inthissense,
obstructed.The wavefleldrecorded
at there-
ceiver
isinfluenced
by thezonecovered
by thisFresnel-volume
and notonlyby theline
traced
by theclassical
ray.

39
 Chapter 5

THE GAUSSIAN BEAM METHOD

5.1 Why Do We Use the Gaussian Beam Method?

So far, the majority of issues addressed in this thesis involved the tracing of rays
and the computation of helpful information such as out-of-plane spreading or Fresnel-
volumes. In the modeling software GBmod, these quantities and the results of kinematic
and dynamic ray tracing are evaluated at the ray ends and stored in a file. Several
methods have been proposed to use these data to generate synthetic seismograms, among
them are the pam_al method (e.g. Beydoun and Keho, 1987) and the Gaussian beam
method (e.g., (_erveny, et al, 1982).
The Gaussian beam method (GBM) was first introduced to the field of geophysics
by Popov (1982) and _erven_ et al. (1982). These authors proposed to expand a source
wavefield into Gaussian beams, which are individually computed from the source to their
endpoints. The seismic wavefleldat each receiver is then evaluated by integrating over all
beams in the vicinity of the receiver. Tile GBM therefore considers not only the informa-
tion provided by the nearest beam, but the information of all beams in the neighborhood.
This procedure has several advantages. First, rays (which are the support of the beams)
are no longer required to stop at the exact position of each receiver; thus time-consuming
two-point ray tracing can be avoided. Second, the GBM yields stable results in regions of
the wavefield where the standard ray theory fails (e.g., caustics, shadow zones and critical
distance). Third, unlike seismograms computed by conventional ray tracing techniques,
tile GBM synthetic data are less influenced by minor details in the model representa-
tion. These advantages are discussed and illustrated for example in Cerven:_, (1985b),
where more references call be found. This chapter gives an introduction to the theory of
Gaussian beams. A good knowledge of the theory is necessary to best understand the
limitations of the method when dealing with complex subsurface structures.

5.2 Gausslan Beam Theory

Two main derivations of the GBM are found in the literature. The first appro_h,
the so-called parabolic wave equation method, uses the reduction of the waveequation to
a parabolic equation for each individual wave type. This procedure is described in detail
in (_erven_ and P/_cn lk (1983) and has the advantage of exphc]tly deriving the Gaussian
beams as the high-frequency asymptotic solutions of the wave equation close to rays.
Here, I use a different way to introduce the G13M.As shown in _erven:_ (1985a), the
( !vnam]c
_. , ray-tracing procedure introduced
o in Chapter 2 can be generalized by allowing
the eikonal fimction v(x,) to be complex valued, More strictly, the traveltime is real

4O
valuedattheraysand complexvaluedoutside
therays.Undercertain
conditions,
which
willbeintroduced
below, thehigh-frequency
solution
oftheelastodynamic
equation
with
a complex-valued
phasefunctioniscalled
a Gaussianbeam.Inotherwords,thebeam
can be considered
ms a paraxial
rayapproximation
witha complexvaluedtraveltime.
Recallthat,fortwo-dimensional
media,thetraveltime
ill
thevicinity
ofa raycan be
expressedintheform 'I

r(S)_ r(Os)+ 2#I n2 , (5.1)

whereM isthe phrase.front
curvatureevaluatedalongtheray,Os istheorthogonM
projection
ofS on therayand n istheperpendiculardistance
to theray.In classical
dynamicraytracing (paraxial
raytracing),
M isrealvalued,whereasintheGausslan
beam method,M iscomplex.

M ffiRe(M) + i Ira(M).

For Ira(M) ffi 0, the Gaussian beam solution reduces to the paraxial ray solution,
As shown in Chapter 2, the seismic ray-theoretical pressure at a receiver S in a two.
dimensional acoustic medium call be described by:

u(s)
= u0 .
_0_ e'*wlt'1"lS))

q and q0aredynamicray-tracing
quantities
corresponding
tothematrixQ inthethree-
dimensionalcase.Afterexpandingthetraveltime
r according
toequation(5.1),
the
exponential
canbe written
intheform

exp [-iw (t- r(Os) - _M(Os) . n')].

Allowing M to be complex yields

1
..)]. ..]
Introducing the frequency-dependent beam width
, i

i,e., the distance from the central ray at which the amplitude of the Gaussian beam is
_-times the amplitude on the central ray, and the curvature of the wavefront

K(Os) = -v Re(M),

equation (5.2) takes the form

41
 u(s)-- L _ --i., t r(os)
Vq-U° - -.......................
._ ,_xp- . (s,_)
A sketch of tile amplitude profile perpendicular to the central ray is shown in Fig-
ure 5.1.

FIG. 5.1. Amplitude profile of a Gaussian beam

Solutions of tile form (5.3) are called Gaussian beams if the following two assump-
tions are satisfied along the ray:
e hn(M) > 0

,q#0.
The first assumption guarantees the concentration of the solution close to the central ray.
Second, quantity q descrlbtng the geometrical preadmg must be nonzero aud finite along
! * S 0

the ray; thus Gaussian beams, unlike the conventional ray solutions, are nonsiligular even
at caustics of the wavefleld.
The fundamental matrix

I'I(trO, tr) -- ( p!
q,(a) q_(a) )
(tr)p_(a) '
42
i.e., the set of linearly independent, real dynamic ray-tracing solutions for tile two-
dimensional case (equations [2.26]), with

II(ao,ao)= (,0) 0 i '

can be directly applied to solve for tile complex quantity M. Analogous to the three-
dimensional case studied in Chapter 2, any solution p(a) mid q(a) call be evaluated as a
linear combination of tile dynamic ray tr_ing results :

p(_) = _tpl(_r)+ :_(o),

q(_) ffi z_q_(_)+ :_q_(o),
where, in general, zt and z2 are complex valued constants. Introducing the complex
valued beam parameter t = :|/zz yields the following equation for M :

M ffi p/q = tp! + p_

¢qt + q'J

The beam width L and the curvature It" can now be expressed in terms of the complex
parameter e ffi t t + ie,_ :

L ffi _/__2.
I itl ii iii

...........!_!qi_+ q_)_2+ ('_"qt)_' '

ui ii LII I I[ i + m t

(5.4)
It" m --V _....=_:: ..........
(tlql + q_)_+ (tzqt)_ '
Recall that qt(a0) -- 1, pt(ao) ffi 0, q2(a0) - 0, Pz(ao) ffi 1; so that the following beam
width and curvature are found at tile source:

V w _2
e,1
K = -,, ;i
In summary, the GBM can be seen as one specific application of tile dynamic ray-
tracing results. For each beam, the dynamic ray-tracing data qi, q2, pt and p2 are weighted
by a complex parameter _. The choice of _ determines the frequency-dependent width of
the beam _ well as the ph_me-front curvature.

5.3 Difficulties with the Gausslan Beam Approach

Several meth()ds have been proposed to use kinematic and dynamic ray trazing
data to generate synthetic seismograms, among thetn are the l)araxial method and the
Gaussian beam method. Both of these methods share the advantage that time-consunfing

43
two-point ray tr_ing Is act required. In the partial method, the _ismtc rospon_ at
a receiver is calculated by extrapolating the information provided by the nearest ray.
The GBM uses a weighted sum of information of all rays in the vicinity of the receiver.
Although this procedure is computatlonally more dlfllodt and less e_cient, it offe_ the
_w.atage that the results are reg_dareven in regions, where the conventional ray methods
fail (e.g., caustics or shadow zones). These situations are extensively di_uss_ in the
literature (e.g. Weber, 1988; K_tner and Fritsche, 1988; M_ller, 1984), and several
exlunples of computations of GBM results are shown. None of the_ reference, however,
provides examples of using the GBM in models of signiflcaat complexity, such as in
Figure 5.2, nor do they consider refl_tions from more than one target. To improve our
understanding of the difficulties that influence the Gauss|an be_ solutions for models
with strong inhomogeneities, a quick review of the Gaussian beam summation appro_h
is necessary.

Distance(km)
0 1 2 3

FIG. 5.2. Overthrust model. Within e_h layer, the seismic velocity increases with depth,
Gauss|an beam modeling in media of this complexity ha_ not yet boca considered in the
literature.

Gaussian beams represent high.frequency asymptotic solutions of the wave equation.

Th(,so solutions arc concentrated close to rays, Similar to the cla:_sicalray methods, whore
44
the source waveflchl is expanded into r_vs, in tile GBM the source flc,ld is expandml
into Gau_tan betuus. Several degreesof freedomarc, available to lu'hteve an pJ_curate
expansion:

. the number of beams used in the expulsion

o the width of e_h beam

s the phi-front curvature of each beam,

Phi-front curvature, i.e., the _cond derivative of traveltlme ev_uat_ at the ray, and
beam width may vary from beam to beam and must be specified at one point, wherever
desired, along the ray, The width and the numberof be,uns determine the be,un cov©_age
in the area of interest, Since the wave equation is llne_, synthetic seismograms may be
computed at my point P in the medium as a weighted sum over all beams.

u(r) =/. (5.7)

where u(P) is the solution at the receiver, and ue(P) is the _iutlon of the beaun with
take-off angle @,evaluated at P. The common procedure for estimating the complex-
valued weighting function @(_) is to compare integral (5.7) with the exit solution of
the wave _luation in homogeneous media for frequency w --t co (_erven_, et ILl,1982).
Additionally, it is assumed that, at the source, tile phaak't.frontcurvature and the width
of the beam are tile same for all take-off angl_ or, as considered by Miiller (1984), vary
linearly as a function of ¢_,
Hale (1091) followed the commonly di_ussed appro_h in the literature of setting
the curvature of the phase.front of tile beam to zero at the source; i,e,, phase-fronts
arc planar at the source. The iutroductlon of beams with plane pham,-fionts at the
source does not include any error; however, that choice reduces the variety of p_Ible
eXlmUslonsof a waveflehl into Gaussian beams. According to equations (5.5) and (5,6),
the only parameter left to be specified is ra, ca specifies tile beam width at the m_urce
(see Figure 5.3). The source beam width, evaluated at peak-frequency, will be referred
to as the _am l_,rcm,cter.
To illustrate dimculties inherent to the GBM, a simple experiment is sketched in
Figure 5.4. Rays arc traced from a single shotlmint and ar,, reflected from an interface
at the depth of 2,5 kin, Synthetic scis|nograms for various choices of beam parameter
are shown in Figure 5.5, Note that the choic,_of beam parameter has influence on the
gen,,rated time sections. Even in this simple model, a poor choice of beam parameter
yields spurious arrivals aud abnormal amplitude behavior.
This phenomenon m_v be understood better by considering Figure ,5,0, The diagram
shows the dominant frequeuey beam widths at the ray euds ,_ a flmction of source-to-
receiver offset, The wrong choice of beam parameter yields beam widths that are too
broad at the ray ends, For the beam parameter of 0,1 kin, the width of the beams
at the ray ends even excc,eds the horizontal dimensions of the model, This le,_ls to
spurious arrivals because the influeuce of the beam is nc_tlimited t_Jthe vic'inity of its
45
 roeelvem

beamwidthat ray ende

*r eource beam width or "beam parameter"

FIG, 5,3, Sketch to ilhlstrate the mea.ing of bva_8 _rorncter and beam _dth at roll
e.d, Note that the rays are the .t,pport of tile beams alld that the amplit.de of a beam
(lecre_t_ expo.e.tlo.aliy away from!the ray, The renpo._ at the receivers is obtained
a weighted ..m over the I)eam.,

46
 shot orltloaloffset

1
orltl sl angle veloolty 2 - 7 krn/s

FIO. 5,4. Geometry of _ experime.t to show the depe.de.ce of the GBM ruults on the
choiceof be_ parameter. Seismic_tio.. for ratios, be_ parametersare displ_
in Figure8.5.

emerging poi.t, The .etsmic _<tion shown in Figure S,Sb was generatmi for a choice
of belun parameter of I kin, As one can _ in Figure 5,6, the widths of the beams at
tlte ray e.d. are much smaller titan in the previous case, hardly exceeding 1.8 km even
for large offsets, As exp_ted, the corresponding seismic _ctlo. looks m'ceptable. Do
the_ r_ults imply that a l.rger beam parameter (i.e,, a broader beam width at the
source) produces more stable res.lt.? A.alysl. of Figures _),5c and 5,Zhlprov_ that this
_tlmptlo. is incorrect, The choice of a 2 km and, more obvlo.sly, the choice of {}km
beam l)araz.eter agai. ge.erate sp.rio.s arrivals (hie to very large beam widths at the
r.y el.is, Them simple experime|it, indicate that a proper choice of beam parameter is
('r.ciai in generating ace.rate seis||togrnms, Specifically, if the beam parameter is too
.stall (i,e,, the beam is too .arrow at the so.roe), a strong spre_iing of the beam along
tile r_ can be exp(,cted,

8,4 Bulc Assumptions l.herent to the Beam Approach

From the .i|.l)le experi..,.ts above, we learned titat tlte co.ce.tration of the beam
along the ce.tral ray is e_e.tlal to generati.g _,c.rate sy.thetic .els|nogra|.., A brief
review of some b,_ic ass.mptlo.s of the dy.a.llc ray tr_ing ,uPl)Ort. tills hypothesis:
• The, dy.a.li_' ray tr_'i.g method Is h_ed on the eiko..l eq._tlon, As discussed i.
Chapter 2, strictly sp,,akl.g the eikonal eqttatlo, dt,s,.rii.,s o.ly the high-fr,_.qtte.cy
47
 a offset (km) b offset (km)
0 5 0 5

'..... ............
' I ' _ t ............
......

,.0
t!3!l]ltt
llitrlItj
tUI'
irlljFi' "•
2,5
• 2,51 IIItlIIIItltlIItlIU!H!IIItlIIIIIfliil_!
Offset(km) Offset(km)
Co 1................
s1 ..................
do i .... ,,,,,.....
s
1 ,,

_,.o !ItI1NI
!.M!M!.M.!
'

iiiiiiiiii j__

_"2 _ 2,°IHIIIIIIIiHIHIIIIIIIHIH
2 -_, 2.S!!!_______!i__I__i___
......
Fla. ,5,5, Sc,iM=nic,' Nt,c'tions('Oml_tit¢_d for tl., exp(,rin.,nt skHc'l.,d in Figure 5.4, The
c'hoi(,cof b,,=t=n pitr_t==.,t,,r is tl., following:(s) 0,1kin; (b) 1.0k==l;(,') 2,0 kin; (d) 5.0 kin.
Th,, arrowspoint to sI IlrlOl!N_trrivatls,Clos,,(,Xttllllllll, tlOll of tl., ,,ncirc,l,,ditr(qB show
ttrtiftu,',ts
¢lt_t,to tht, flt(,t t,hnt h_,_.=l_ haw. bt,cometoo bro=vl.

48
 -S 0 5 10 15
Offset(km)

FIG, 5.6, A more tletailecl mmlysi, of tile res.lt, obtai.ecl i. the experiment sketched
i. Fig.r(, 5,4 aids i. better .tiderstmlcli.g of some of the diifl(,ttltiesof Ga.ssian beam
mc_(leli.g, The ill.grin, clisl)l_v, the tlomitmxit-freq.exicyboron width at the receiver a_
a fttlwtio, t,f offm,.t. Ea('h syml)ol (lexmt(,stt (lifferc.t choi('e of beam parmncter at tile
8()lirce,

49
 part of the wavefield. This assumption is usually stated as the validity condition of
ray theory:
_<<:l,

where _ is the wavelength of the signal and l represents the length scale of the
medium (e.g., Bleistein, 1984). Note that large frequencies yield small beam widths
(equation (5.4)).

• The traveltime in the neighborhood of a central ray is approximated by a Taylor

expansion (equation (2.18)), requiring that the velocity has to vary smoothly rela-
tive to the width of the beam. Moreover, velocity discontinuities within the beam
width, in general, are not seen by the central ray and yield inaccurate traveltime
extrapolations.

• The dynamic ray tracing equation (2.20) has been derived under the paraxial as-
sumption (2.19). In two dimensions, this assumption yields:

• <<i.
n=0

Even for small values of extrapolation distances n away from the central ray, the
paraxial assumption will break down if strong velocity gradients perpendicular to
the ray are present in the medium.

In summary, three basic requirements can be stated:

1. The length scale of the medium must be much larger than the wavelength of the
wavefield considered.

2. The width of the beam should be as narrow as possible along the ray.

3. The velocity has to vary smoothly relative to the width of the beam.

Only if these three criteria are satisfied will the Gaussian beam method accurately
represent the seismic wavefield. The next sections will discuss situations in which one or
more of these constraints are violated.

5.5 Gaussian Beams Across Interfaces

Ray tracing in complex subsurface models requires an appropriate transformation of

kinematic and dynamic ray tracing data across discontinuities of first- and second- order
in velocity. The reflection and transmission of high-frequency Gaussian beams in elastic
media has been investigated by (_erven:_ and P_end{'k (1984). These authors applied a
phase-matching method, i.e., they required the traveltime of incident and reflected or
transmitted wave to be equal along a curved reflector.
Hale (1991) traced rays through models containing velocity discontinuities. If a ray
impinges on a boundary dividing two media with different velocities or densities, one
5O
 0

FIG. 5.7. Sketch of the quantities involved in the transmission of a ray at interface _.
The incidence angle is _',and the refraction angle is _+. The velocity is discontinuous,
jumping from vi" to vi+ across the interface.

must account for reflection and transmission at boundaries in order to properly calculate
the wavefleld amplitude. While Hale implemented (_erven_, and PSen_{k's correction for
the influence of discontinuities on the geometrical spreading, the loss of energy reflected
by impedance discontinuities was not included, i modified the program to compute the
necessary correction factor ART along the ray.
N

i=l

This resultisderivedin detailin Appendix C. The summation isoverallinterfaces

encountered by the ray. At the ith interface, this expression involves the reflection or
transmission coefficient R_ for acoustic pressure waves, which can be evaluated by assum-
ing the continuity of both pressure and the normal component of particle velocity across
an interface (Brekhovskikh, 1960). The meaning of the other quantities in equation (5.8)
is sketched in Figure 5.7. If the ray is reflected back from interface _, the two square-
roots yield unity because both the velocities and the angles of incident and reflected ray
are identical. I also included an option to specify a constant value of density in each geo-
logic block bounded by interfaces. Reflection and transmission coefficients are computed
by considering the density discontinuity across the interface. The additional computing
time required for evaluating the energy partitioning due to reflection and transmission
at interfaces is negligible.

5.6 Near-critical Transmission of Gaussian Beams

There exists a potential danger associated with the transmission of dynamic ray-
tracing data. Consider the experiment in Figure 5.8, where rays are traced through a
salt dome model. The source is situated at 3.2 km horizontal position. The analysis of
the beam width at the ray ends is shown in Figure 5.9. Near-critical rays emerge in the
51
region of 0 to 1.5 km and yield extremely broad beam widths. The rays reflected from
the dome, however, behave well.

Position (km)
0 2 4
0

FIG. 5.8. Rays traced in a salt dome model. The source is situated at 3.2 km horizontal
position. Near-critical rays emerge in the region of 0 to 1.5 km and yield extremely broad
beam widths. As we learned from the example considered in Figure 5.4, these beams will
produce spurious arrivals in the corresponding seismic section.

Artifacts due to near critical transmission can be reduced once we better understand
the reason for this beam spreading. As derived by (_erven:_ and P_en_i'k (1984), the
dynamic ray tracing quantity q is transformed across interfaces according to

q+ =q-

For near-critical transmitted rays, _+ is close to 7r/2 and the transformed value q+
will be very small, q+, in essence, specifies the new initial beam width on the other side
of the discontinuity. According to equation (5.4), a small initial value of q again creates
a small initial beam width and, as in the half-space experiment sketched in Figure 5.4,
this small initial beam width will produce strong beam spreading!
52
 4 :!l!l!l[:!!l!rrT_L_iml!!rl_iJL
IIIII
I!!i_]_
IIIIll
a I II I I I II

! 12- '
W
'
'O
= 10-
A

_ A A M

_ 6- A ', A A

. A • J
8 A •

E &6 &&A
A&&&6 A&8

iiI iiiiii, i ii i!ii iiii

ill!_ii I !I II ;::_:_

0 I 2 3 4
Posltlon (kin)
FIG. 5.9. Analysis of the dominant-frequency beam width at the ray ends in Figure 5,8.
Data acquired in the region beyond the horizontal position of 1.5 km will yield accurate
results.

53
 To avoid spurious arrivals associated with near-critical transmitted rays, I imple-
mented a taper on the transmission coefficients. The amplitudes of these rays are hereby
significantly reduced so as not to influence the seismic section.

5.7 The Overthrust Model Experiment

Additional problems can be expected if we trace rays through media such as the
overthrust model introduced in Figure 5.2. This model is composed of geologic blocks
with different seismic velocities. In each layer, the velocity is increasing with depth, Note
that interfaces with strong curvature and sharp edges are present in the model.
The complexity of the medium can be visualized better by displaying rays and
wavefronts for a shot situated at 1.15km horizontal position (see Figure 5.10). White
lines show rays that are reflected from the bottom of the model and emerge at the surface.
I These rays are the only ones chosen to contribute to seismic data collected at the surface.
Energy that propagated out of tile model is simply represented by black wavefronts, with
tile corresponding rays deleted. Of the 600 rays traced through this model, less than one-
sixth of them end up at the surface; tile others either stop at the boundary of the model
or they are overcritically incident at velocity discontinuities. In the large shadow zones
such as the one between 1.2 km and 1.7 km horizontal position, synthetic data generated
by a classical ray-th oretmal approach would not show any seismic response.
e , e t

To test synthetic data computed for this model, Hale (1991) generated a synthetic
midpoint gather for the midpoint at 1.(3kin, shown in Figure 5.11. Negative and positive
offset represent tile interchange of source and receiver positions. Reciprocity requires
that trace amplitudes be independent of this interchange. The asymmetry in the figure
indicate, that the reciprocity in tile created dataset is violatcd, so that the data are not
accurate. Hale supposed two reasons for the failure. First, the ml'88' mg correction for
reflection and transmission on the amp!ituctes, and, second, the violation of the beam
assumption. The latter requires ttle model parameters to vary smoothly relative to the
wl'CIth of the bcam (e.g., White, et al., 1987). At an interface, for example, the width of
the beam has to be significantly smaller than the radius of curvature of tile interface.
Figure 5.12 displays the result of the common midpoint gather after correcting for
the energy loss due to reflection and transmission. The asymmetry is still strong, and
one cannot recognize any improvement. I evaluated tile beam width at the surface for a
source located at a horizontal distance of 1.6 km (see Figure 5.13). Tim range of beam
widths exceeds the dimensions of the model by about one order of magnitude1 This
indicates that beams attached to rays do not provide an accurate description of the
wavefield. Moreover, I was unable to choose any beam parameter that gave acceptable
beam widths at the surface.

5.8 Application of Gaussian Beam Modeling In Complex Media

Increasing tile complexity of the model will eventually violate the basic assumptions
of the GBM laid out in Section 5.4, Complex media such as tile overthrust model shown

54
 Distance(krn)
0 1 2 3

FIG. 5.10, Rays and wavefronts for a shotpoint situated at 1.15 km horizontal position
ill the overthrust model (Figure 5.2). White lines show rays that are reflected from the
bottom of the model and emerge at the surface. Energy that propagates out of the model
is represented by black wavefronts only, with the corresponding rays deleted,

55
 FIa, 5.il. Common midpoint gather for tile model shown in Figure 5.2. The midpoint
is situated at 1.6 km offset, Negative and positive offset represent the interchange of
source and receiver positions. The asymmetry suggests that the data generated fail to
satisfy reciprocity. In this section, the correction for transmission and reflection of seismic
energy at velocity discontinuities is not included.

56
 Offset (km)
-1.5-1.0-0.5 0 0.5 1.0 1.5
0

FIO, 5.12, Common midpoint gather for the model shown in Figure 5.2, The midpoint
is situated at 1.6 km offset. This section differs from that in Figure 5,11 only in that the
correction for transmission and reflection of seismic energy at velocity discontinuities is
now included. The asymmetry, however, remains.

57
 A A

25
i 20 •
A
A
dk

15
Ak A

10

_kAd k A A A
AA i, A

...... ,, _.., .-r-r__,-_,._

J ii I !l IJ .... _ I I III III III1[ _-
0 0.5 1 1.5 2 2.5 3
Distance (kin)

FIG. 5.13, Analysis of ttle dominant frequency beam widths at tile ray end indicates the
violation of the beam assumptions for the overthrust model (Figure 5,2). Several beam
widths exceed the horizontal dimension of the model by about one order of magnitude.

58
above certainly violate these assumptions; thus we should not be surprised to encounter
discouraging test rcsuits, The question is whether we can apply some means to r_uee
disturbing artifacts and generate more stable r_ults.
i suggest a compromise between the rigorousvalidity conditions of the beam method
and the practical application of beam modeling. Thr_ criteria should be carefully ex-
amined when modeling seismic responses in complex m_ia:

o The complex traveltime should not be extrapolated too far away from the central
ray. More specifically, the range of extrapolation should tlot exc_ the natural
length scale of tile medium. R_eivers should add up contributions from only
nearl_yemerging beams,

, The beam width should not be fixed at the source. In complex media, each beam
eventually encounters a different velocity field and thus r_uires an individual opti.
realbeam parameter. This beam parameter must be chin such that the beam re-
mains as narrow as possible along the ray. Both the location at which to assign this
p_ameter (e.g,, at the source or at the ray ends) and its optimum value have been
investigated by various authors (Weber, 1988; Miiller, 1984; {_erven_, 1988a,b).
The published results, however, were satisfactory only for models with minor in-
homogeneities (Kiistner and Frltzsche, i988). I tested the approach suggested by
_erven:_ (1985b) and found some encouraging results which I print below.

• Each receiver should receive contributions from _veral beams. As suggest_ by

Beydoun and Keho (i087), the GDM generally requires on the order of I0 beams
in the vicinity of the receiver in order to reconstruct the high-frequency part of the
wavefleld.

To analyze the importance of the first criterion, I performed the reciprocity test for
the overthrust model, but this time I allowed only nearby beams to contribute to tile re-
ceivers, One cannot avoid that some beams spread too much, however, their contribution
to the seismic section can be reduced by this means. Figure (5.14) shows a synthetic mid-
point gather for the midpoint at 1,6km horizontal position and a limited extrapolation
range of 0.32 kin, Only beams with their central ray situated closer than 0.32 km from
the receiver contribute to the generated seismic data. The symmetry of the synthetic
midpoiilt gather indeed improw:d. The asymmetric direct-arrival branches disappeared
and noisy artifacts between 1.0 and l.Ss arrival time have vanished, The chara_cteristlcs
of the reflections from the bottom of the model, however, remain essentially unchanged.
The latter suggests that the reflected energy instill influenced by broad beams.
As mentioned above, I used an optimal choice: of beam parameter to generate the
midpoint gathers. The optimal choice is ba_ed on a statistical analyses of the dynamic
ray tracing data cvahlatcd at the ray ends. The optimal choice is the one that on average
produces the minimum beam width at the surface. This analysis is cumbersome and one
would like to have an atltomatic Drocedtlre to compute accurate, seismic data.
The apl)roach l_rOl)OSCd by Ccrveny (1985b) does zlot requlr(, the d_,termination of
an optimal beam l)aramctcr. Here, I)_c,d on (_erveny's empirical results rather than on
59
Fla, 5,14, Ccmmm. midpoint gatl.,r for tl., modt.,l.Imwn i. Fig.r,: 5.2, TI., midpoint in
sit.at,,d _t 1,6 km h.riz.ntal po.itl.n, Only bq.am, with tl.,Ir c_,ntr=dr,_ysit.atc_i elcmer
than 0.32 km from the rt,('t.iw,r('ontrilmtoto the gener,tted m,imld¢_data. TI.* =ymmctri(_
dirq,_'tarrival br,tn(.|.,s and noisy artif,tcts between 1,0 and l,Ss arrival time v'_nisht.d,
Th(, ('harnx't_,rintic,s
of the r(,flcctimmfrom th(, Imttum of tht_.m(),h,I,h.wovor, remain
im('hax=g,,(l,

0
 Offset(kin)
-1.5-1.0-0.5 0 0.5 1.0 1.5

F!o, 5,15, Common nltdpoint gather for the model shown in Figure 5,2. Tile midpoint tm
situated at 1.6 km horizontal position. Only beams with their central ray situated closer
than 0.32 km from tile receiver coutrlbute to the generated seismic data. The beam
widths are chosen for ea_,h beam iadividually M a function of the dynamic ray tracing
data qi ,_j, pi ,l_j.

0!
 Off:et (kin)
-1.5-i.0-0.5 0 0.5 1.0 1.5

Fla. 5,16, C()Illllll)li I,li(li)(Jillt KtLtl.'r. 5)r till, l.(_(ll,I HII()W.i. FiK!lrt:5.2, TI._ lIli(ll)()illt
iM_litllttt!,d ttt 1,6 kill lt()ri;_()iltttlI)t)Niti()ll, ()lily I)t,_lll_twitil tl.,ir (_(,lltrttlrtty 14itutLtt_(l
l'l()Ml,rthttli 0.32kill fr!)lll till' re(,ejvvr(.Ol|tril)!ltit,t() tills' gi,lli,rttt(_(l _l,iHlni(' (ltLtlt. Tile
I)i'ltlll wi(ltilH art, ('ll()Mt'llfl)r ¢,_u'liI)q,
alll iltclivi(liltLllytU_a I'l|l|('ti¢)ll ()f t.ll(_ (lyl|atlli('_ rlLy
trtl_,il|K(ll_ltt¢11,ttj, Pl, P3, hl('ri,tu4il|gthe illlllll)t,r ()f rtl_'l_trtl_,li(lt,llr(ltlgh the lll()(le!by It
fau'tl)rc)ffive (tu4(,()li|l)l_rt,(l witll tilt, rl_y_lHilowiiill Fig!|re 5.|()) in|l)r()vt,Ntl.; ('Olltilluity
c)ftile rl,_('{)r(ll,(l
Sc,iNll|ic'rt,til,l'ti()llH.
theoretical derivations, tile beam widths are chosen for each beam individually as a func-
tion of tile dynamic ray tracing data qt,qa, Pt,_. The function @(¢) responsible for an
appropriate weighting of the bealns is al)proximately derived by an asymptotic evaluation
of integral (5.7), followed by a comparison with ray-theoretical results. I implemented
this idea of automatically prescribing the beam width. The resulting midpoint gather tn
Figure (5.15) is quite symmetric; however, at least two features are still disturbing. First,
the direct _rivals corresponding to negative offsets are weaker than the ones correspond-
tng to positive offsets. The offset is negative if tile receiver is situated on the right aide of
the midpoint. Rays traveling ne_ horizontal from right to left are overcritlcally incident
on velocity discontinuities and are not propagated hrther (recall from Chapter 2 that
the geometrical optics approach does not handle the propagation of het_atwaves). This
is the reason why far more direct rays emerge at receivers with positive offset, S_ond,
tile reflections from tile bottom of tlte model arc erratic for the offsets between -1,5 km
and -0.5 kin, A closer examination of the contributing central rays reveals that only very
few rays contribute to these offsets. Moreover,most of these rays went through caustics
of tile wavefield and produce phase shifts. This situation violates the third criterion es-
tablished in this section. In fact, as seen in Figure (5.16), increasing the number of rays
traced through tile model by a factor of 5 (as compared to the rays shown in Figure 5.5)
improves the continuity of the recorded seismic reflections.
It is not my intention to overestimate the accuracy of the results generated by ap-
plying the three criteria outlined above. In particular, the tools suggested to improve the
results are based on experience rather than on a strict mathematical basis. However, in
my subjective opinion, the means proposed to increase the data quality yield satisfactory
reciprocity in the generated common midpoint gathers, indicating that tile computed
soisnlic sections are more _curate. This ix especially evident if one recalls that the basic
asSUml)tionof the GBM are severely violated in the model considered.

5,9 Quo Vadls Gausslan Beams?

Research on Gaussian beams started apl_roxitnately 10 years ago with the prospect
of solving many problems inh_,rent to classical ray theory, Few authors highlighted the
ambiguity associated with a freely varying beam-width parameter, In practice, this
parameter has been defined somewhat arbitrarily, mainly adjusted to mininlize errors
in the beam supcrlmsition (lqlimc,_, 1989) or tuned to minimize errors associated with
w,locity inhomc_geneitics (Kiistncr and Fritsche, 1988). Despite these efforts, the results
lmblished in tile literature are limited to models with simple velocity structures; none of
the references provides modeling results for media with complexity comparable to that
of the overthrust ntodcl shown here.
In the previous section, 1 suggested a procedure to avoid some of the artifacts as.
sociatod with the beam.sproacling problem. However, only a more theoretical derivation
has the potential of artually solving this difficulty.
Based on my experience with the Gaussian beam procedure, I suggest two main
directions for future research which might help to iml}row_the l_orformance of Gaussian

63
beam modeling:

• A more sophisticated choice of beam parameter has to be developed. The parameter

determining the beam width could for example be related to the physical concept
of Fresnel volumes. However, it might then be necessary to specify a different beam
parameter (specifying the phase-front curvature and beam width) at well chosen
locations along each ray, that is, as a function of ¢. In addition, we should consider
a weighting function that varies with ¢, that is, (I,(¢) in equation (5.7) must be
evaluated.

• Smoothing of the velocity field could reduce the spreading problem of the beams.
This apprc, ach is motivated by the good performance of the Gaussian beam migra-
tion procedure (e.g., Hill, 1990; Hale, 1992), where model velocities are smoothed.

64
 Chapter 6

CONCLUSION

6.1 Summary

The purpose of this research has been to enhance the usefulness of the dynamic ray
tracing procedure both for modeling of seismic responses and for interpretation of previ-
ously acquired data. Several new implementations to the dynamic ray tracing software
have been introduced to improve the flexibility and accuracy of the proposed modeling
method.
The modeling software GBmod now allows the introduction of a broader class of
raypaths, For example, it is now possible to trace multiples in surface or VSP experi-
ments. Additionally, I added options to include attenuation and density variations. The
influence of energy partitioning at interfaces and of 3-D geometrical spreading in 2-D
models can now be simulated efficiently. A new feature called Fresnei-volume ray tracing
has been included to study the resolution and the validity conditions of ray-theoretical
methods. All these modifications can be used while maintaining the attractive feature of
computational speed and limited computer storage.
The Gaussian beam method (GBM), one important application of the dynamic ray-
tracing procedure, has been studied in Chapter 5. The theory of beam modeling is
accurate and efficient as long as models with only weak inhomogeneities are considered.
A simple experiment and a summary of the intrinsic assumptions of the GBM help to
understand difficulties when applying the method to more complex velocity structures.
The GBM, as currently developed, cannot properly handle models of high complexity.
Tile limitations are similar to those of classical ray theory. However, I proposed some
useful criteria to avoid most of the artifacts associated with the violation of the beam
assumptions.
Modeling using GBM is only one possible application of the dynamic ray-traced
data generated along the ray. Alternative methods (e.g., the paraxial methods) can
also use these data to create synthetic seismograms. The new modifications such as the
tracing of multiples, computation of out-of-plane spreading, or reflection and transmission
correction are evaluated independently of the Gaussian beam procedure.

6.2 Future Work

The GBM will be used for modeling of seismic wavefields in models with significant
inhomogeneities only when we achieve a focusing of the beams along the ray. In this
case, an extension of the GBM modeling code to elastic media would be worthwhile. The
required changes are not difficult. An interactive procedure for the modeling and a more

65
, involved extension to three-dimensional models could then be considered.
Tile extension to three dimensions in usefld not only for nlo(leling of seismic data
by tile Gaussian beam procedure. New methods designed to efficiently simulate seismic
wavefields in triangulated subsurface models such as wavefront tr_u:ing (e.g., Coultrip,
1993) could be considered and exten(led to three-dimensional media, _mwell.

06
 Appendix A

SOLUTION OF THE TWO-DIMENSIONAL DYNAMIC RAY TRACING

EQUATIONS

The dynamic ray tracing system (2.26)

1
dq ffip
da ' dp =
da _V, llq ' (A-l)

can be solved analytically in media with a constant gradient in 1Iv 2, i,e., the sloth can
be represented in the form:

v-2(x,z) E s(x,z) = Soo+ s,_x + s,,z. (A-2)

Hale (1991) used a symbolic computer language to verify the following solutions for the
coupled system of differential equations (A-l):

q(_)= _ 1 {[ _o+ s,(a-a0)

-2_.... [q(a0)+p(a0)(a-a0)]

p(_) = _ [,o+,,(_ - _o)]p(_o)

+ + k_ - 2,, (_- _o)q(ao)

- [s' + 2s'(a-a°)] q(a)2s' (A-3)

where so, sl, and s2 are constants defined by :

_o --- _[x(ao),
z(ao)],
Sl _ s,_p_(ao) + s,zp_(ao) ,

s_ = _l(s_'+s,_')
so that
s = s[x(a),z(a)] = _0+ s,(_ -_0) + s_(a- _o)_.
For a given set of initial conditions q(ao) and p(ao), (A-3) can be used to obtain the
dynamic ray-tracing data at any value of a. In the ray-tracing part of the modeling, the
quantities q(a) and p(a) need only be evaluated for the intersections with triangle edges.
To continue the dynamic ray tracing, q and p are transformed across the interface by
67
applying a phase-matching procedure. Tile transformed values then serve as new initial
conditions for system (A-i).

68
 Appendix B

MATRIX Q AND ITS RELATION TO THE RAY JACOBIAN

In classicalray theory,geometrical spreadingisdescribed rayJacobianJ. In three

dimensions, the Jacobianisa measureofthecross-sectional areaof a ray tube.Due to
thechoiceofcoordinate systemsintroduced inChapter2 (recall thatour monotonically
increasing parameteralongthe rayisa and not the arclength s),a strictderivation of
therelation betweenthematrixQ and theray J_obian isnecessary.
Assume a family of four rays f/(Tl, 72), fl(Tl + 671,7_), f/(Tl, "Y2+ &Y2)and f/(71 +
&Yl,_f2+ 672) forming a ray tube for a fixed value of a (Figure B-l). The ray parameters
_fl and 72 can for example be identified as the takeoff angle and the azimuth of the ray.

FIG. B-I. A ray tube formed by a familyoffourrays.dA represents a cross-sectional

areaspanned by the raysfora fixedvalueof a. To obtainthe corresponding cross-
sectionalsurfaccareaforfixedvalueof traveltimer, one has to projectthe vectorial
surfaceelementdA ontothetangent

The vectorial surface element dA is projected onto the tangent t to obtain the
corresponding cross-sectional surface area for a fixed value of traveltime 7. This area
can be considered an differential element of the wavefield and is directly related to the
geometrical spreading.
As shown in Bleistein (1984), the differential subsurface element d_ can be expressed
a8

d'A = \071 x 072]_,=#, d71d72,

where × denotes the cross product. The magnitude of this vector represents the differ-
ential area dA. The area dA (r), i.e., the area corresponding to a fixed value of traveltime

69
can be simply evaluated by using the dot-product:

dA (r) = dA. t, = _,0./l x 0"y2] a=ot d'yld72 v O"-'a

= v o(xl,x_,x3)r
Using the definition of the ray Jacobian, dA {r) = J dTtd?2, yields

O(xl,x_,xa)
J=v 0(_1,72,a) --vdet(T)'

T, the transformation matrix from ray parameters to the cartesian system, can be ex-
pressed by
T=H.(_,
where
H = (o(xl'x,?,',,x_) (cg(ql,q2,s)
, )
Both the cartesian and the ray-centered coordinate systems are orthogonal; thus det I-I =
1. (_ is a (3 × 3) matrix with the upper left (2 × 2) submatrix representing the matrix
Q calculated during the dynamic ray tracing. In other words, the Jacobian J can be
written as:

(oo) 0 o

= v.N.det _
= detQ. (B-l)

Relation (B-I) is important in calculation of the amplitudes along each individual

ray. The geometrical spreading can be simply evaluated by forming the determinant of
the dynamic ray tracing matrix Q. Note that the sign of J may be different at different
sections along a ray separated by caustics. (B-1) will also be used to evaluate the energy
loss due to out-of-plane spreading.

7O
 Appendix C

TRANSFORMATION OF GAUSSIAN BEAM AMPLITUDES ACROSS

INTERFACES

FIG. C-1. Sketch of the quantities involved in the transmission of a beam at interface
_. The incidence angle of the central ray at point P+ is c_-,and the refraction angle is
_+. The velocity is discontinuous, jumping from v(e_-) to v(P_) across the interface.

Figure C-1 sketches the situation depicting the influence of a curved first order
discontinuity on the amplitude of a lligh-frequency acoustic Gaussian beam. The notation
is similar to that used in Chapter 5. Let us first consider the case in which the central ray
of a Gaussian beam interacts with one interface El. The expression for the ray theoretical
amplitude between two points O and P_" has been derived in Chapter 4 and is of the
form:

U(p.{)=Uo._V(P_)_v(Oi q(PT)
q(O) '
Boundary conditions have to be applied to transform the amplitudes across the interface

U(P_') = U(P_'). HI ,

71
where Rl denotes the acoustic reflection or transmission coefficient at the interface. If
we continue tile ray tracing to point O,, we find:

1 _1
,,(P_) qiO,I "

= Vo.v(Pt)_ ;i_i _q(O.)

q(Pr)R, '
This result may be generalized by introducing n curved interfaces:

U(O,)= U0._ v(O) q(O-'-',5

,., _ V(Pti q(Pi'---_
'

0erven_, and Paen_i'k (i984) derived the expression for the transformation of q across
interfaces.
_o,,(,_i*)
q(Pi
+)= q(Pi')
'cos(ai.
). (C-2)
For a reflection from the i-th interface, tile last two square-roots in equation (C-l) yield
unity. It is fortunate that the transformation of q is independent of the beam width, thus
enabling us to compute the necessary reflection and transmission correction as a simple
by-product of the ray tracing. Equation (C-2) has been derived under the assumption,
that the radius of curvature of the interface is much larger than the width of the beam.
Under this constraint, our final expression for multiply reflected or transmitted Gaussian
beams is:

=.o..I
,(o.) .I,,(o)/I
,,(o) I0(pc).i
_q(O.)'_.,
._,,(p_)
_i_,,_(,_.)).
(c.3)
Unlike the classical ray-theoretical amplitudes, this expression involves two complex
variables q(O,) and q(O). The reflection or transmission coefficient Ri at the ith interface
for acoustic pressure waves can be evaluated by assuming the continuity of both pressure
and the normal component of particle velocity across an interface (Brekhovskikh, 1960).

72
b