Anatomy Inverse Problems

GEOPHYSICS, VOL. 65, NO. 6 (NOVEMBER-DECEMBER 2000); P. 17081710, 4 FIGS.
The Anatomy of Inverse Problems

John A. Scales
and Roel Snieder
A major task of geophysics is to make quantitative state-

ments about the interior of the earth. For this reason, inverse
problems are an important area of geophysical research and
industrial application. Figure 1 shows how many texts present
inverse problems. The earth model is an element of a mathe-
matical space that contains all allowable parameterizations of
the earths properties (or at least those properties relevant to a
given experiment); this space is referred to as model space. The
physics of theproblemdetermines whichdatad correspondtoa
given model m. The problemof computing the model response
(synthetic data) given a model is called the forward problem.
The corresponding data reside in a mathematical space that is
called data space. In many applications, one records the data,
and the goal is to nd the corresponding model. The task is
called the inverse problem, as shown in Figure 1.
Unfortunately, Figure 1 is wrong. There is a simple reason
for this. In general the model that one seeks is a continuous
function of the space variables with innitely many degrees of
freedom. For example, the 3-D velocity structure in the earth
has innitely many degrees of freedom. On the other hand, the
data space is always of nite dimension because any real exper-
iment can only result in a nite number of measurements. A
simple count of variables shows that the mapping fromthe data
to a model cannot be unique; or equivalently, there must be el-
ements of the model space that have no inuence on the data.
This lack of uniqueness is apparent even for problems involv-
ing idealized, noise-free measurements. The problem only be-
comes worse when the uncertainties of real measurements are
takenintoaccount. Althoughtheuniqueness questionis ahotly
debated issue in the mathematical literature on inverse prob-
lems, it is largely irrelevant for practical inverse problems be-
cause they invariably have a nonunique solution (if by solution
we mean an earth model). It is this nonuniqueness that makes
Figure 1 deceptive, because the arrowpointing fromdata space
to model space suggests that a unique model corresponds to
every data set.
Amore realistic scheme of inverse problems is shown in Fig-
ure 2. Given a model m, the physics of the problem determines
the predicted data d; this is called the forward problem. For
a given data set, one determines an estimated model m. We
Manuscript received by the Editor June 8, 2000; revised manuscript received August 23, 2000.
Colorado School of Mines, Department of Geophysics and Center for Wave Phenomena, Golden, Colorado 80401. E-mail: jscales@mines.edu,
rsnieder@mines.edu
c 2000 Society of Exploration Geophysicists. All rights reserved.
refer to this as the model estimation problem. (Later, we will
consider the generalization to the problem of estimating prop-
erties of models, rather than the models themselves.) Note that
there may be many reasonable model estimates for a givendata
set and that the estimation procedure may be nonlinear even
when the forward problem is linear. Thus, the mean of a set
of numbers is a linear function of the numbers, whereas the
median is a nonlinear function. Yet both the median and the
mean may be reasonable estimators of the center of the set
of numbers. Part of the art of solving inverse problems comes
from the need to dene what it means for an estimate to be
reasonable.
Since the mapping from data space to model space is
nonunique, theestimatedmodel mayalsodependonthedetails
of the algorithm that one has used for the estimation problem
as well as on the regularization and model parameterization
that has been used. In general, the estimated model m differs
from the true model m. For example, in seismic inversion the
estimated model may be a blurred version of the true model. In
addition, the data are always contaminated with errors; these
errors represent an additional source of discrepancy between
the estimated model and the true model.
One is not nished when the estimated model is constructed;
it is essential to somehow quantify the error between the esti-
mated model and the true model. This is called the appraisal
problem. Inthis problem, one determines the uncertainty inthe
estimated model. This uncertainty has a statistical component
related to the propagation of errors in the data, and a deter-
ministic component that accounts for the nite resolution that
is attained in the model estimation, as well as systematic errors
in the problem.
For linear inverse problems, resolution kernels and con-
dence set analysis are powerful tools for formulating the ap-
praisal problems, and the theory of linear error propagation is
sufciently well developed to account for the errors in the esti-
matedmodel duetotheerrors inthedata. For nonlinear inverse
problems, the only tool available for the appraisal problem
may be Bayesian inversion where one estimates the statistical
properties of the model when the data and other knowledge
are combined in a statistical sense by repeated sampling of the
1708
The Anatomy of Inverse Problems 1709
model space (e.g., Mosegaard and Tarantola, 1995; Gouveia
andScales, 1998). These techniques are only applicable toprac-
tical problems where the number of parameters is relatively
small. This is partly due to computational cost of such methods,
but is also related in a very subtle way to the behavior of prob-
abilities in high dimensional spaces (see Scales and Tenorio,
2001). For large-scale inverse problems such as the determi-
nation of 3-D earth structure, Monte Carlo sampling methods
are not feasible. This means that there is presently no opera-
tional theory to account for the appraisal problemof nonlinear
inverse problems with large number of parameters (Snieder,
1998). Developing such a theory is a theoretical and practi-
cal challenge that is much more important than establishing
uniqueness proofs of idealized mathematical problems.
In practice, one solves inverse problems with a certain goal.
For example, one may use an estimated model obtained from
an inversion of seismic data as a basis for deciding where to
drill or how to optimally exploit a reservoir. In practice, one is
never interested in the seismic velocity at a certain spot in the
subsurface, but for a seismic interpreter it is crucial to know
whether at a certain place a syncline or an anticline is present.
This means that for practical inverse problems one is interested
in patterns that can be used in a meaningful way for making
decisions. In practice, these decisions are usually not based ex-
clusively on the estimated model m, but involve the integration
of other data as well as human expertise. In addition, the uncer-
tainty in the estimated model is an important factor in making
decisions. Thus, Figure 3 gives a more realistic view on inverse
problems, because it shows explicitly that decisions rather than
model estimation is the endpoint of practical inverse problems.
It is interestingtoconsider therelationbetweentheappraisal
problem and the process of decision making. An important as-
pect of the appraisal problem is the statistical treatment of
error propagation. This means that the appraisal problem has
probability as an important component. Decisions are usually
FIG. 1. The conventional view of inverse problems: nd the
model that predicts the measurements.
FIG. 2. An improved view of inverse problems.
based on risk rather than probability. This may be economic
risk (where to drill), environmental risk (what happens when
pollutants spread), or even academic risk (am I sure enough
to publish my results). Risk is always concerned with proba-
bility plus another component such as prot or environmental
impact. This means that in this stage those who produce the
models and their uncertainty must interface with others for
making decisions effectively.
Figure 3 offers an overview of the landscape of geophysical
inversion. Those working on wave propagation problems focus
on the forward problem. Those focused on the development of
migration algorithms work on the estimation problem. Statisti-
cians and a limited number of scientists in the inverse problem
community are concerned with the appraisal problem. Seismic
interpretation is usually based on an estimated model, and so
can be thought of as a decision-making problem. It is illuminat-
ing to see how different researchers work on different parts of
this problem. One may wonder whether these research efforts
couldbemoreeffectivewhentheir activities areseeninthecon-
text of the anatomy of the inverse problemas showninFigure 3.
In Figure 3, the estimated model m forms an essential part
of the inversion process. But is it necessary or desirable to pro-
duce an estimated model in the process of inversion of data?
We, of course, are conditioned to produce models from our
data. However, given the fact that this estimated model dif-
fers from the true model, one can be led astray by features in
the estimated model that are artifacts of the inversion process.
Another view on inverse problems is given in Figure 4 where
the goal is to determine the range of models that are consistent
with the data (as well as other information). This range of mod-
els can simply be a box within which all the models are believed
to t the data (there being no comparative relation among the
models in the box), or it could be a probability distribution on
the space of models P(m) (in which case one can speak of the
best or most probable model). Both the box and the probabil-
ity are determined by the data, the uncertainties and whatever
other data-independent information is available. As we have
discussed previously (Scales and Snieder, 1997), people taking
the former approach are called frequentists (this includes most
statisticians), whereas people taking the latter approach are
called Bayesians (this includes most geophysicists). In either
case, the estimation and appraisal problem is replaced by the
inference problem. Inference means characterizing somehow
the set of models that explain the data (and satisfy whatever
other information is available). One can then use this box, or
the probability distribution P(m), as a basis for making deci-
sions. The tutorial by Scales and Tenorio (2001) shows a nice
example of a toy inverse problems tackled from a Bayesian
and a frequentist point of view.
The reader may be put off by the idea of solving inverse prob-
lems without constructing a model. Our minds are conditioned
to making models fromdata. However, we have seen that these
estimated models can only be the endpoint of research when
one ignores the fact that models are being produced with the
goal of making decisions. When seen in this larger context (Fig-
ures 3 or 4), it is worth considering whether one is not better
served by knowing the probability distribution of the set of
models than by knowing a single estimated model and a mea-
sure of its uncertainty.
It is interesting to consider how we would use a probabil-
ity density function in a high-dimensional model space. The
1710 Scales and Snieder
FIG. 3. The inverse problem as part of a decision-making
process.
FIG. 4. The inverse problem as an inference problem.
simplest approach is to compute the mean and variance of each
model parameter, and one can visualize this information rel-
atively easily. However, as we have seen, the patterns in the
model are much more interesting than the estimates of one
model parameter. Assessing the robustness of certain patterns
in the model is much more difcult, especially since this entails
the use of the correlation between different model parameters.
In a high-dimensional model space it is extremely difcult to
characterize andinterpret the correlations of the model param-
eters. In order to interrogate the resulting probability density
function of the model in a meaningful way, research is needed
into cluster and feature analysis of (possibly multimodal) prob-
ability density functions of many degrees of freedom, as well
as the development of an interface between the exploration
of these high-dimensional functions and the decision-making
process.
It will be clear from this that in order to treat inverse prob-
lems in ways that are different from current practice requires
signicant theoretical and numerical advances. However, clear
strategies for the optimal use of inverse problems in the pro-
cess of decision making should also be an important item on
the agenda for researchers in inverse problems.
REFERENCES
Gouveia, W. P., and Scales, J. A., 1998, Bayesian seismic waveform in-
version: Parameter estimation and uncertainty analysis: J. Geophys.
Res., 103, 27592779.
Mosegaard, K., and Tarantola, A., 1995, Monte Carlo sampling of so-
lutions to inverse problems: J. Geophys. Res., 100, 1243112447.
Scales, J. A., and Snieder, R., 1997, To Bayes or not to Bayes: Geo-
physics, 62, 10451046.
Scales, J. A., and Tenorio, L., 2001, Prior information and uncer-
tainty in inverse problems: Geophysics, tentatively scheduled for
March/April.
Snieder, R., 1998, The role of nonlinearity in inverse problems: Inverse
Problems, 14, 387404.
GEOPHYSICS, VOL. 65, NO. 6 (NOVEMBER-DECEMBER 2000); P. 17111725, 9 FIGS., 3 TABLES.
Anisotropic approximations for mudrocks:
A seismic laboratory study
Morten Jakobsen
and Tor Arne Johansen
ABSTRACT
Anexperimental technique for determining the in-situ
elastic properties of mudrocks with (horizontal) align-
ments in the microstructure is used to study the accuracy
of a set of three nestedscalar anisotropic approximations
for transversely isotropic (TI) media. Each subsequent
approximation adds one more velocity parameter and
includes the previous as a special case. These approxima-
tions are convenient and robust because of their close re-
lationship to standard geophysical measurements. There
exists no good theory to predict the effects of an im-
posed stress on the elasticity of mudrocks. In this study,
the tensor of elastic moduli of a single test specimen of
mudrock subjected to an anisotropic stress eld is de-
termined from ultrasonic group velocity measurements
involving pointlike transducers. The mechanical char-
acterization (performed at constant pore pressure) is
accompanied by detailed microscopic observations and
analysis. The method was used to obtain accurate elas-
tic constants for ve well-dened mudrocks with differ-
ent degrees of anisotropy. These elastic constants were
then used to study the accuracy of the three anisotropic
approximations. We have found that mudrocks can be
signicantly anelliptic. The agreement between the sec-
ond anisotropic approximation and the exact phase ve-
locity was found to be very good for qP-waves. For
qSV-waves agreement was not as good, and the third
anisotropic approximation is required to obtain reason-
able accuracy.
INTRODUCTION
The assumption that sedimentary rocks behave elastically
as an isotropic medium is a major source of error in the seis-
mic imaging of subsurface features in the earth (Sayers, 1995;
Rowbotham, 1997). The nature of depositional processes tends
to produce transverse isotropy with a vertical symmetry axis
Manuscript received by the Editor October 12, 1998; revised manuscript received March 7, 2000.
University of Bergen, Institute of Solid Earth Physics, Allegt. 41, N-5007 Bergen, Norway. E-mail: morten@if.uib.no; torarne@ if.uib.no.
in undisturbed, horizontal, plane-layered, sedimentary rocks.
Shale anisotropy, in particular, has been identied at all scales
from ultrasonic to seismic (Hornby et al., 1995) because of the
partial alignment of plate-like clay particles within the rock.
The inuence of the intrinsic anisotropy on the seismic ve-
locities can often be quite strong in comparison with multi-
layering effects (see Werner and Shapiro, 1998). Clearly, the
inherent anisotropy of sedimentary rocks should be taken into
account in interpreting seismic eld data. For velocity analysis
andimaging, however, thefull complexityof ahexagonal model
is often unnecessary (e.g., Dellinger et al., 1993; Schoenberg
et al., 1996; Alkhalifah et al., 1996).
Dellinger et al. (1993) have introduced a set of three nested
scalar anisotropic approximations for transversely isotropic
(TI) media. Each subsequent approximation adds one more
velocity parameter and includes the previous as a special case.
These approximations are convenient and robust because of
their close relationship to standard geophysical measurements.
Sayers (1993) demonstrates the applicability of the approxi-
mations by using his simple model of the elastic anisotropy
of shales. The purpose of this study is to check the accuracy
of the approximations for anisotropic mudrocks using accu-
rate elastic constants obtained from ultrasonic wave speeds
measured on core samples under simulated in-situ stress con-
ditions. This is important since mudrocks are the most abun-
dant of all lithologies, constituting some 75% of sedimentary
rock sequences. We dene a mudrock as a sedimentary rock
containing admixtures of very ne sand-, silt-, and clay-sized
grains. Fissilemudrocks arereferredtoas shales. [Someauthors
use shale as a synonym for mudrock (Leeder, 1982).]
The elastic stiffnesses of TI rocks are often determined by
using the pulse-transmission technique on sets of plugs cored
at different known orientations with respect to the bedding
normal (Jones and Wang, 1981; Hornby, 1994; Dellinger and
Vernik, 1994; Johnston and Christensen, 1995). Convention-
ally, ultrasonic transducers are in contact with the at sides
of the specimens, and phase velocities are measured. In this
method, problems often arise because of insufcient control
with core inhomogeneities (Vernik and Liu, 1997).
1711
1712 Jakobsen and Johansen
In the method of Jakobsen and Johansen (2001) used in this
study, wave-speed data are obtained on a single test specimen
cut normal to bedding, thereby limiting the effects of inherent
material variability from specimen to specimen. In addition to
the ultrasonic transducers used in the conventional procedure,
several pointlike transducers are used to obtain wave speeds
in the bedding plane and in an off-symmetry direction of a
zonal section. The tensor of elastic moduli is recovered from
the wave speeds assuming that ultrasonic group velocities are
measured. The method is then applied to check the anisotropic
approximations of Dellinger et al. (1993) for ve North Sea
samples.
ANISOTROPIC APPROXIMATIONS FOR TI MEDIA
In the rst anisotropic approximation of Dellinger et al.
(1993), the phase velocity squared W() is given by
W() = W
z
C
w
+ W
x
S
w
, (1)
where W
z
and W
x
are the vertical and horizontal wave speeds
squared, C
w
= cos
2
, S
w
= sin
2
, and is the angle the wave
normal makes with the vertical. Equation (1) is consistent with
a phase velocity surface being ellipsoidal.
In the second anisotropic approximation of Dellinger et al.
(1993), the phase velocity squared is given by
W() =
(W
z
C
w
)
2
+(1 +W
x NMO
/W
x
)(W
z
C
w
)(W
x
S
w
) +(W
x
S
w
)
2
W
z
C
w
+W
x
S
w
,
(2)
where W
x NMO
is thesquareof theshort-spreadnormal moveout
(NMO) velocity we use every day in surface-to-surface data
processing. If W
x NMO
is equal to W
x
, then equation (2) reduces
to equation (1).
In the third anisotropic approximation of Dellinger et al.
(1993), the phase velocity squared is given by
W() =
(W
z
C
w
)
3
+(2 + W
x NMO
/W
x
)(W
z
C
w
)
2
(W
x
S
w
)
(W
z
C
w
+ W
x
S
w
)
2
+
(2 +W
z NMO
/W
z
)(W
z
C
w
)(W
x
S
w
)
2
+(W
x
S
w
)
3
(W
z
C
w
+W
x
S
w
)
2
,
(3)
where W
z NMO
is the square of the moveout velocity measured
for near-horizontal propagation, such as might be found in a
cross-borehole experiment. If W
z NMO
is equal to W
z
, then equa-
tion (3) reduces to equation (2).
In terms of the density normalized stiffnesses W
i j
=c
i j
/
of a medium of hexagonal symmetry (Musgrave, 1970), the
anisotropic velocity parameters are given for qP-waves by
(Dellinger et al., 1993)
W
x
= W
11
, (4)
W
z
= W
33
, (5)
W
x NMO
= W
55
+(W
13
+ W
55
)
2
(W
33
W
55
), (6)
W
z NMO
= W
55
+(W
13
+ W
55
)
2
(W
11
W
55
). (7)
For qSV-waves they are given by
W
x
= W
55
, (8)
W
z
= W
55
, (9)
W
x NMO
= W
11
(W
13
+ W
55
)
2
(W
33
W
55
), (10)
W
z NMO
= W
33
(W
13
+ W
55
)
2
(W
11
W
55
). (11)
DETERMINATION OF ELASTIC PROPERTIES
OF MUDROCKS
Wave-speed measurements
Norsk Hydro collected 30-m-long sections of core from a
depth of about 2250 m in well 31/4-A-04, which crosses a ma-
jor fault zone in the Brage eld, North Sea. The samples were
selected and preserved at the rig site during the coring op-
eration. The 4-inch-diameter core was cut using an oil-based
drilling mud. The oil in this drilling uid was used to ll the
void between the core and the core barrel. To prevent the core
from drying out, it was ensured that no air was trapped inside
the core barrel. The core samples were then brought onshore
for storage.
It is important to leave the initial pore uid content intact to
obtain realistic acoustic measurements in the laboratory. Sed-
imentary rocks dehydrate rapidly (within hours) at ambient
conditions and thus detoriate. To minimize the dehydration,
steps were taken to minimize the specimens exposure to air
during storage and preparation. An alternative approach is to
restore the initial saturation by wetting the specimen with syn-
thetic pore uid. This method was not used since many miner-
als have water as an integral part of their structure and drying
cracks may form irreversibly during initial dehydration.
For ve different core depths, a single cylindrical specimen
was cut normal to bedding, the diameter and length being nom-
inally 38 mm and 6080 mm, respectively. End planes were
ground parallel to within 0.1 mmand planar to within 0.05 mm.
The specimen was analyzed in terms of density, porosity, and
mineralogy, as described in Table 1. Scanning electron micro-
scope (SEM) microphotography was used to analyze the mi-
crostructure of the specimen (see Figure 1).
The specimen was intrumented with a permeable sleeve
(strips of lter paper with sand and oil) to lower the pore uid
draining distance and to increase the draining area. With this
technique, anacceptabletime(hours) was requiredfor thepore
pressure to equilibrate after changes in applied stress, as dis-
cussed in Hornby (1994).
The specimen was then instrumented with piezoelectric
crystals (see below or Jakobsen and Johansen, 2001) to help
determine the overall properties of the argillaceous rock, as-
sumed to behave elastically as a mediumof hexagonal symme-
try (Kaarsberg, 1959).
Before placing the (instrumented) specimen in the pressure
vessel (described in Jakobsen and Johansen, 1999), it was jack-
eted with a rubber sleeve to isolate the pore-uid-pressure sys-
tem from the conning-pressure system. The measurements
were performed with separate control of conning-uid pres-
sure, axial load, and pore-uid pressure. To simulate in-situ
stress conditions, we used
11
=13 MPa and
33
=20 MPa,
where
11
and
33
denote the horizontal andvertical stress com-
ponents, respectively. The pressure in the pore-uid reservoir
Velocity Anisotropy in Mudrocks 1713
was kept constant at 1 MPa. Extensiometers (LVDTs incontact
with the jacketed specimen) were used to measure axial and
radial strain, and this information was used to correct velocity
measurements for length changes.
To measure wave speeds, we used the pulse-transmission
technique (e.g., Birch, 1960). The source was excited by the
output port of a Panametrics 500PR, and the receiver was con-
nected to the input port. The waveform from the receiving
transducer was digitally recorded using a Phillips PM3352A
oscilloscope. The signal was averaged 100 times to increase
the S/N ratio and transferred to a computer (using Lab View)
for further processing. Aschematic of the transducer/specimen
system is shown in Figure 2.
End-cap assemblies composed of 22-mm-wide transducers
(500 kHz) were used to generate and receive ultrasonic pulses
propagating along the bedding normal (Figure 3a). By per-
forming a reference measurement with no specimen in place,
the delay in the buffer rods was found (Jakobsen and Johansen,
2001).
Specially designed transducers composed of 5-mm-wide
PZT crystals (600 kHz) were glued on at tracks made on the
curved side of the cylindrical specimen to generate and receive
ultrasonic pulses propagating in the bedding plane (Figure 3b)
and in an off-symmetry direction (Figure 3c).
Our choice for the geometry of the transducer/specimen sys-
tem has implications for the type of velocity measured. In me-
dia of hexagonal symmetry, the phase and group (energy) ve-
locities of a wave are not, in general, equal. Only along the
zonal axis and in the basal plane are the two velocities equal.
For propagation along the principal symmetry axes, so-called
pure modes are generated. This allows us to use large transduc-
ers along the bedding normal and smaller ones in the bedding
plane.
Jakobsen and Johansen (2001) have developed a numerical
model and have used it to demonstrate that group velocities
canbeextractedfromtraveltimes measuredinanoff-symmetry
direction in a specimen instrumented with 5-mm-wide trans-
Table 1. Description of the mudrocks examined in this study. The density was obtained from measurements of the volume and
weight of a cylindrical test specimen. The porosity was determined using a standard helium expansion technique. The mineralogy
was estimated using energy dispersive spectroscopy in an electron microscope.
Core depth (m) 3492 3506 3525 3536 3564
Density (kg/m
3
) 2428 2444 2379 2441 2408
Porosity (%) 9.9 15.7 7.9 13.9 6.5
Mineralogy (vol) % % % % %
Chlorite 4 20 19 27 15
Illite/muscovite 9 17 8 10 15
Kaolinite 4 14 23 14 6
Glauconite 2 5 8 2
Biotite 1 2 4 3 3
Quartz 2 20 19 11 32
Albite 6 5 2 7
Ankerite 2
Barite 3
Calcite 50 0 1
Garnet 3
Gypsum 1 0
K-feldspar 1 7 7 4 9
Siderite 8
Titanite 9
Other 14 12 10 10 10
ducers. [The work of Dellinger and Vernik (1994) is not rel-
evant here, since they were dealing with a different source-
receiver geometry.] The analysis of Jakobsen and Johansen is
based on simple ray theory and can be summarized as follows:
The source and receiver transducers were modeled as a series
of point transducers lying very close to one another. The result
was a waveform that incorporated the effects of nite-width
transducers. From analysis of synthetic waveforms and inspec-
tion of associated wave surfaces, it was concluded that the
rst-break traveltimes in off-axis experiments yield group ve-
locities, provided that the sourcereceiver separation is mea-
sured fromedge to edge as shown in Figure 2. This is equivalent
to replacing the nite-width transducers with effective point
transducers. A certain amount of energy (number of rays) is
needed before the rst break can be detected, suggesting
there is some uncertainty in the location of the effective point
transducers. This source of error was taken into account
in the Monte Carlo simulations discussed below.
We use reference axes coincident with rock axes so that Ox
3
lies along the main symmetry axis and Ox
1
, Ox
2
complete an
orthogonal triad. We denote by v
i j
the speed of a pure mode
propagating along axis Ox
i
and polarized along axis Ox
j
. The
group speed of a quasi-compressional wave propagating at an
angle to the axis Ox
3
is denoted by V
q P
.
The wave speeds measured as a function of applied stress,
shown in Table 2 (v
11
>v
33
, v
31
v
32
v
13
< v
12
), are consistent
with the assumption of transverse isotropy with the main sym-
metry axis perpendicular to the (layering or bedding) plane of
circular symmetry (see Schoenberg, 1994). The conclusion that
v
31
is similar to v
32
is not affected by the sometimes large errors
in these wave speeds since the waveforms were similar for the
two orthogonal polarization directions (see Figure 3a). From
Figure 4, it is apparent that the wave speeds are increasing as
a function of increasing stress. This can be explained by a re-
duction in void space and by increased contact between grains
(Jones and Wang, 1981). The anisotropy of the specimens will
be discussed in the following.
FIG. 1. Caption-see next page
Wave-speed inversion
For a medium of hexagonal symmetry (Musgrave, 1970),
the nonzero elastic stiffnesses c
i j
are c
11
=c
22
, c
33
, c
44
=c
55
,
c
13
=c
23
, and c
12
=(c
11
2c
66
). The moduli c
11
, c
33
, c
55
, and
c
66
can be determined using wave-speed data measured along
the principal axis as follows:
c
11
= v
2
11
, (12)
c
33
= v
2
33
, (13)
c
55
= v
2
13
, (14)
c
66
= v
2
12
, (15)
where is the density. The modulus c
13
requires at least one
off-axis measurement for its determination. If a group wave
FIG. 1. Looking for alignments in the microstructure of ve core samples (mudrocks) fromvarious depths in a North Sea well. Thin
sections were taken in a plane perpendicular to bedding. For all core depths, some alignments in the structure can be seen on the
(2.18 times) natural size pictures (smaller gures). For all core depths but 3536 m, alignments in the structure can hardly be seen
on the scanning electron micrographs (larger gures). Some of the pictures show that microcracks have been formed and that the
statistical distribution of heterogeneities is not locally uniform, suggesting that the ultrasonic acoustic properties of mudrocks can
be both anisotropic and weakly inhomogeneous. Core depths are (a) 3492 m, (b) 3506 m, (c) 3525 m, (d) 3536 m, and (e) 3564 m.
speed V
q P
(with group angle ) has been measured, c
13
may be
obtained from
c
13
= N(, c
11
, c
33
, c
55
, V
q P
, ), (16)
where N denotes a numerical procedure formulated for gen-
eral anisotropic media by Every and Sachse (1990) and im-
plemented for media of hexagonal symmetry by Jakobsen and
Johansen (2001). The procedure corresponds to a two-stage
optimization method in which c
13
is varied so as to obtain a
t of measured to calculated group velocity. At each iterative
step a numerical minimization process is used to nd the phase
angle and speed of waves having the required ray direction.
The above approach results in very accurate estimates of the
principal stiffnesses c
11
, c
33
, c
55
, and c
66
but less accurate es-
timates of the critical stiffness c
13
. Very accurate estimates of
c
13
could be obtained if a larger set of group velocities was
measured. These velocities would correspond to qP-modes
measured in any direction or to qSV-modes associated with
unfolded regions of the ray surface. Each data point would
have its own standard deviation, and we would minimize a
well-known quantity called the chi-squared. So far, we have
not been able to work with more than one off-axis receiver.
FIG. 2. Schematic of a specimen cut normal to bedding and
instrumented with ultrasonic transducers. The vector ending
at X
3
lies along the main symmetry axis.
We think that the situation with several qP-receivers can be
realized in future studies, although we have not solved the
practical problems that occur when we try to work with sev-
eral off-axis receivers inside the pressure vessel. It should be
noted that no qSV-modes were detected using our small trans-
ducers. The qSV-modes are probably strongly contaminated by
qP-modes.
To get an idea of how accurate c
13
can be estimated (now
and in the future), we performed Monte Carlo simulations of
hypothetical experiments involving one or more qP-receivers
(Jakobsen and Johansen, 2001). First we generated 1000 noisy
wave speed data sets; the statistical errors in the wave speeds
were uncorrelated and normal distributed with a standard de-
viation of 1% for the principal wave speeds and 3% for the
off-axis wave speed. Then we inverted each data set to obtain
its elastic stiffnesses, and we analyzed the distribution of the
results in terms of mean values and standard deviations. In the
case of (1, 2, 3, 4) randomly located qP-receivers, the uncer-
tainty in the critical stiffness c
13
is (28, 25, 20, 17)%. The reader
may think that Hornby (1998) is claiming larger uncertain-
ties without the qSV-constraint on c
13
. But the reader should
not forget that Hornby estimated elastic parameters by using
phase slowness measurements (involving several specimen ori-
entations) and obtained error bars via a simple perturbation
approach.
The recovered elastic stiffnesses for the Brage mudrocks are
shown in Table 3. Clearly, most of the specimens are strongly
anisotropic, as canbe notedonthe wave surfaces reconstructed
in Figure 6. However, for comparison and quantication of
anisotropy, it is convenient to redene the ve independent
stiffnesses in terms of two vertical wave speeds and three di-
mensionless anisotropy parameters. These anisotropy param-
eters are given by (Thomsen, 1986):
=
c
11
c
33
2c
33
, (17)
=
c
66
c
44
2c
44
, (18)
=
(c
13
+c
44
)
2
(c
33
c
44
)
2
2c
33
(c
33
c
44
)
. (19)
We remember that and describe the anisotropy of P- and
S-waves, respectively, while is a critical factor that depends
on the shape of the qP- and qS-wave surfaces. If ( ) >0,
as in Table 3, the anellipticity is considered positive. This
means that the qP-wave slowness curve bulges out fromthe el-
lipse connecting the vertical and horizontal P-wave slownesses
and that the qSV-wave slowness curve is pushed inward from
the circle that connects its horizontal and vertical slownesses
(Schoenberg, 1994; Tsvankin 1996). Elliptical anisotropy can
be used in both the phase and group domains as a rst step
away from isotropy (Dellinger et al., 1993).
As an aside, let us discuss the stress dependence of the
velocity anisotropy in our mudrocks. From Table 3 it is ap-
parent that the P-wave anisotropy (Thomsens parameter )
(normally) decreases as a function of increasing hydrostatic
stress. This can be explained by the closure of cracks and pores
(Lo et al., 1986; Hudson, 1991; Hornby, 1994). The P-wave
anisotropy also decreases as a function of increasing axial load
(at constant hydrostatic pressure), suggesting that bedding-
parallel microcracks are getting closed. The variation in the
S-wave anisotropy (Thomsens parameter ) as a function of
triaxial stress is smaller than expected, and the uncertainty in
Thomsens parameter (see Figure 7) is to large to draw rm
conclusions about its stress dependence. At the estimated in-
situ conditions, open microcracks may still contribute to the
observed anisotropy. Bedding-parallel microcracks could ex-
ist in situ in most mature source shales undergoing the major
stage of hydrocarbon generation and migration (Vernik, 1994).
Some of the microcracks may have been created in the pro-
cess of recovering core, suggesting that our laboratory tech-
nique may overestimate the in-situ elastic anisotropy of the
Brage mudrocks. Nevertheless, the in-situ stiffnesses may be
of great value in future modeling studies, and they are used
here to test our favorite anisotropic approximations for TI
media.
FIG. 3. Typical ultrasonic waveforms recorded in this experi-
ment. (a) Measurements along the bedding normal. (b) Mea-
surements in the bedding plane. (c) Measurements in an
off-symmetry direction. The pulser/receiver settings varied
with the measurement direction. This is one reason why the
off-axis signal is often associated with a lower frequency than
the other signals. The off-axis signal is a function of the input
electrical transient and the off-axis response of the piezoelec-
tric transducer.
TEST OF THE ANISOTROPIC APPROXIMATIONS
Figures 8 and 9 compare the three successive scalar ani-
sotropic approximations of Dellinger et al. (1993) with the qP-
and qSV-wave phase velocities for the Brage mudrocks, ob-
tained using the densities in Table 1 and the in-situ stiffnesses
in Table 3. The agreement of the rst anisotropic (elliptical) ap-
proximation and the exact result is relatively poor, particularly
for qSV-waves.
The agreement between the second anisotropic approxima-
tion and the exact result is good for qP-waves, particularly for
near-vertical and near-horizontal wave propagation. For qSV-
waves the agreement between the second anisotropic approx-
imation and the exact result is better for near-vertical prop-
agation than for all other angles of propagation. The third
anisotropic approximationworks well for qSV-waves. Our nd-
ings essentially agree with those of Sayers (1993).
Table 2. Ultrasonic wave speeds (in units of m/s) measured on the mudrocks of Table 1 as a function of stress components
ij
(inGPa) with1 MPa pore pressure. We expect that v
11
/v
11
< 1%, v
12
/v
12
< 1%, v
13
/v
13
< 1%, v
31
/v
31
< 5%, v
32
/v
32
< 5%,
v
33
/v
33
< 1%, and V
qP
/V
qP
3%, where indicates a standard deviation.
Depth
11

33
v
11
v
12
v
13
v
31
v
32
v
33
V
qp
(49
0
)
3492 1.5 1.5 3444 2103 1955 1950 1950 2860 3113
4.0 4.0 3507 2007 1901 1900 1900 2966 3175
7.0 7.0 3607 2097 1971 1950 1950 3096 3287
10.0 10.0 3677 2162 1948 1950 1950 3179 3297
13.0 13.0 3725 2183 2058 2050 2050 3226 3413
13.0 16.5 3713 2183 1968 1950 1950 3270 3540
13.0 20.0 3740 2192 2067 2050 2050 3302 3470
3506 1.5 1.5 3383 1892 1685 1700 1700 2990 2953
4.0 4.0 3400 1922 1737 1750 1750 2998 2898
7.0 7.0 3514 1966 1767 1750 1750 3067 3019
10.0 10.0 3539 2016 1800 1800 1800 3160 3082
13.0 13.0 3551 2029 1828 1850 1850 3176 3114
13.0 16.5 3586 2056 1856 1850 1850 3253 3156
13.0 20.0 3593 2051 1863 1850 1850 3266 3218
3525 1.5 1.5 3156 1897 1846 1850 1850 2867 2995
4.0 4.0 3223 1914 1874 1850 1850 2938 3060
7.0 7.0 3286 1977 1939 1950 1950 3032 3151
10.0 10.0 3344 2028 1978 2000 2000 3139 3180
13.0 13.0 3375 2063 2007 2000 2000 3181 3246
13.0 16.5 3382 2078 2003 2000 2000 3243 3277
13.0 20.0 3416 2092 2028 2050 2050 3263 3313
3536 1.5 1.5 3610 2201 1516 1500 1500 2813 2903
4.0 4.0 3633 2211 1525 1550 1550 2851 2999
7.0 7.0 3661 2221 1532 1550 1550 2902 3054
10.0 10.0 3677 2220 1544 1550 1550 2924 3061
13.0 13.0 3691 2235 1545 1550 1550 2951 3139
13.0 16.5 3698 2222 1547 1550 1550 2973 3095
13.0 20.0 3717 2217 1563 1550 1550 2995 3125
3564 1.5 1.5 3341 1961 1890 1900 1900 2810 3020
4.0 4.0 3400 1985 1894 1900 1900 2894 3049
7.0 7.0 3481 2012 1924 1900 1900 2997 3163
10.0 10.0 3512 2053 1953 1950 1950 3022 3173
13.0 13.0 3558 2085 1982 2000 2000 3069 3208
13.0 16.5 3572 2094 1997 2000 2000 3097 3216
13.0 20.0 3592 2094 1995 2000 2000 3136 3246
CONCLUSIONS AND DISCUSSION
A procedure has been described by which an economic
characterization of rocks with (horizontal) alignments in the
microstructure can be made, and this is important when
interpreting seismic eld data. The advantage of this proce-
dure is that the elastic stiffnesses of a TI rock can be deter-
mined using a single test specimen, limiting the effect of in-
herent material variability from specimen to specimen. This
same advantage becomes important in cases where specimen
orientation is limited because of size and availability.
The procedure has beenimplementedsuccessfully inthe lab-
oratory on a set of ve mudrocks taken from a borehole in the
North Sea basin. The elastic properties of the mudrocks were
found to be a function of conning pressure and axial load.
The ultrasonic measurements were performed at constant pore
pressure. Our characterization of mechanical properties was
accompanied by detailed microscopic observations and anal-
ysis. The experimental data are useful in characterizing the
anisotropic elastic properties of sedimentary rocks in terms of
density, porosity, mineralogy, and microstructure.
The elastic stiffnesses recovered at simulated in-situ
(anisotropic) stress conditions were used to test the predic-
tions of a set of three scalar anisotropic approximations for TI
media introduced by Dellinger et al. (1993). The results of this
study suggest that the second and third anisotropic approxi-
mations should continue to nd use in replacing TI dispersion
relations in modeling, migration, and inversion programs. We
do not recommend using an elliptical approximation for the
elastic anisotropy of mudrocks.
ACKOWLEDGMENTS
We thank Norsk Hydro for their hospitality and help and for
permission to publish their data. Geir stby provided valu-
able help in the seismic laboratory. Harald Flesche provided
the mineralogy data and the SEM pictures. Dag kland gave
us information about in-situ stress conditions. The Norwegian
Research Council and the consortium for Seismic Reservoir
Characterization at the University of Bergen are thanked for
economic support.
REFERENCES
Alkhalifah, T., Tsvankin, I., and Larner, K., 1996, Velocity analysis
and imaging in transversely isotropic media: Methodology and a
case study: The Leading Edge, 15, 371378.
FIG. 4. Ultrasonic wave speeds (compressional modes) in the
Brage mudrocks measured as a function of effective vertical
stress given as P
c
+P
a
P
p
, where P
c
is the conning pressure,
P
a
is the axial load, and P
p
is the pore pressure. Here, P
a
= 0
for P
c
< 13 MPa and P
p
= 1 MPa. Core depths are (a) 3492 m,
(b) 3506 m, (c) 3525 m, (d) 3536 m, and (e) 3564 m.
Birch, F., 1960, The velocity of compressional waves in rocks to 10
kilobars, part 1: J. Geophys. Res., 65, 10831102.
Dellinger, J., andVernik, L., 1994, Dotraveltimes inpulse-transmission
experiments yield anisotropic group or phase velocities?: Geo-
physics, 59, 17741779.
Dellinger, J., Muir, F., and Karrenbach, M., 1993, Anelliptic approxi-
mations for TI media: J. Seis. Expl., 2, 2340.
Every, A. G., and Sachse, W., 1990, Determination of the elastic con-
stants of anisotropic solids from acoustic-wave group-velocity mea-
surements: Phys. Rev. B, 42, 81968205.
Hornby, B., 1994, The elastic properties of shales: Ph.D. thesis, Univ.
Cambridge.
1998, Experimental laboratory determination of the dynamic
elastic properties of wet, drained shales: J. Geophys. Res., 103,
29 94529 964.
Hornby, B. E., Miller, D. E., Esmersoy, C., and Christie, P. A. F., 1995,
Ultrasonic-to-seismic measurements of shale anisotropy in a North
Sea well: 65th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded
Abstracts, 1721.
Hudson, J. A., 1991, Overall properties of heterogeneous material:
Table 3. Elastic stiffnesses and anisotropy parameters characterizing the mudrocks. The stiffnesses (in GPa) were recovered
from the wave speeds in Table 2, and the dimensionless anisotropy parameters were calculated using the formulas of Thomsen
(1986). On the basis of Monte Carlo simulations, we expect that c
11
/c
11
< 2%, c
33
/c
33
< 2%, c
55
/c
55
< 2%, c
66
/c
66
< 2%,
and c
13
/c
13
28%, where indicates a standard deviation. The uncertainty in the anisotropy parameters is indicated in Figure 7.
Depth
11

33
c
11
c
33
c
55
c
66
c
13

3492 1.5 1.5 28.8 19.9 9.3 10.7 3.8 0.22 0.08 0.14
4.0 4.0 29.9 21.4 8.8 9.8 5.4 0.20 0.06 0.08
7.0 7.0 31.6 23.3 9.4 10.7 5.7 0.18 0.07 0.05
10.0 10.0 32.8 24.5 9.2 11.3 5.2 0.17 0.11 0.04
13.0 13.0 33.7 25.3 10.3 11.6 6.0 0.17 0.06 0.05
13.0 16.5 33.5 26.0 9.4 11.6 11.8 0.14 0.12 0.20
13.0 20.0 34.0 26.5 10.4 11.7 6.9 0.14 0.06 0.05
3506 1.5 1.5 28.0 21.8 6.9 8.7 3.5 0.14 0.13 0.18
4.0 4.0 28.3 22.0 7.4 9.0 0.80 0.14 0.11 0.23
7.0 7.0 30.2 23.0 7.6 9.4 2.3 0.16 0.12 0.20
10.0 10.0 30.6 24.4 7.9 9.9 2.7 0.13 0.13 0.20
13.0 13.0 30.8 24.7 8.2 10.1 2.9 0.12 0.12 0.18
13.0 16.5 31.4 25.9 8.4 10.3 2.8 0.11 0.11 0.20
13.0 20.0 31.5 26.1 8.5 10.3 4.4 0.10 0.11 0.16
3525 1.5 1.5 23.7 19.6 8.1 8.6 4.5 0.10 0.03 0.06
4.0 4.0 24.7 20.5 8.3 8.7 4.8 0.10 0.02 0.05
7.0 7.0 25.7 21.9 8.9 9.3 5.2 0.09 0.02 0.05
10.0 10.0 26.6 23.4 9.3 9.8 4.1 0.07 0.03 0.03
13.0 13.0 27.1 24.1 9.6 10.1 5.0 0.06 0.03 0.00
13.0 16.5 27.2 25.0 9.5 10.3 5.6 0.04 0.04 0.02
13.0 20.0 27.8 25.3 9.8 10.4 5.8 0.05 0.03 0.00
3536 1.5 1.5 31.8 19.3 5.6 11.8 4.7 0.32 0.55 0.15
4.0 4.0 32.3 19.9 5.7 11.9 6.8 0.31 0.54 0.08
7.0 7.0 32.8 20.6 5.7 12.1 7.6 0.30 0.55 0.07
10.0 10.0 33.1 20.9 5.8 12.0 7.3 0.29 0.53 0.09
13.0 13.0 33.3 21.3 5.8 12.2 9.3 0.28 0.55 0.02
13.0 16.5 33.4 21.6 5.9 12.1 7.7 0.27 0.53 0.09
13.0 20.0 33.8 21.9 6.0 12.0 8.0 0.27 0.50 0.08
3564 1.5 1.5 26.9 19.0 8.6 9.3 3.4 0.21 0.04 0.09
4.0 4.0 27.8 20.2 8.6 9.5 3.1 0.19 0.05 0.00
7.0 7.0 29.2 21.6 8.9 9.7 4.6 0.18 0.04 0.04
10.0 10.0 29.7 22.0 9.2 10.1 3.8 0.18 0.05 0.01
13.0 13.0 30.5 22.7 9.5 10.5 3.6 0.17 0.05 0.00
13.0 16.5 30.7 23.1 9.6 10.6 3.3 0.16 0.05 0.03
13.0 20.0 31.1 23.7 9.6 10.6 3.8 0.16 0.05 0.03
Geophys. J. Internat, 107, 505511.
Jakobsen, M., and Johansen, T. A., 2001, Determination of the elas-
tic properties of shales using single test specimens, in Hood, J. A.,
Ed., Advances in anisotropyselected theory, modeling, and case
studies: Soc. Expl. Geophys., scheduled for early 2001.
Johnston, J. E., and Christensen, N. I., 1995, Seismic anisotropy of
shales: J. Geophys. Res., 100, 55916003.
Jones, L. E. A., and Wang, H. F., 1981, Ultrasonic velocities in Creta-
ceous shales from the Williston basin: Geophysics, 46, 288296.
Kaarsberg, E. A., 1959, Introductory studies of natural and articial
argillaceous aggregates by sound propagation and X-ray diffraction
methods: J. Geol., 67, 447472.
Leeder, M. R., 1982, Sedimentology: Allen and Unwin Ltd.
Lo, T., Coyner, K. B., and Toksoz, M. N., 1986, Experimental deter-
mination of elastic anisotropy of Berea Sandstone, Chicopee Shale,
and Chelmsford Granite: Geophysics, 51, 164171.
Musgrave, M. J. P., 1970, Crystal acoustics: Holden-Day.
Rowbotham, P. S., 1997, Anisotropic migration of coincident VSP
and cross-hole seismic reection surveys: Geophys. Prosp., 45, 683
699.
Sayers, C. M., 1993, Anelliptic approximations for shales: J. Seis. Expl.,
2, 319331.
1995, Anisotropic velocity analysis: Geophys. Prosp., 43, 541
568.
Schoenberg, M., 1994, Transversely isotropic media equivalent to thin
isotropic layers: Geophys. Prosp., 42, 885915.
Schoenberg, M., Muir, F., andSayers, C. M., 1996, Introducing ANNIE:
A simple three-parameter anisotropic velocity model for shales: J.
Seis. Expl., 5, 3549.
Thomsen, L., 1986, Weak elastic anisotropy: Geophysics, 51, 1954
1966.
Tsvankin, I., 1996, P-wave signatures and notation for transversely
isotropic media: An overview: Geophysics, 61, 467483.
Vernik, L., 1994, Hydrocarbon-generation-induced microcracking of
source rocks: Geophysics, 59, 555563.
Vernik, L., and Liu, X., 1997, Velocity anisotropy in shales: A petro-
physical study: Geophysics, 62, 521532.
Werner, U., and Shapiro, S. A., 1998, Intrinsic anisotropy and thin
multilayeringTwo anisotropy effects combined: Geophys. J. Inter-
nat, 132, 363373.
FIG. 5. Same as in Figure 4, but for shear waves.
FIG. 6. Wave surfaces of the Brage mudrocks reconstructed us-
ing the densities in Table 1 and the in-situ stiffnesses in Table 3.
The curves marked with l and u were made using the lower and
upper bounds on c
13
, respectively. The three sheets of the wave
surface (using the mean value of c
13
) have been overlain by ul-
trasonic group-velocity measurements (Table 2). Dashed lines
indicate SH-modes. Core depths are (a) 3492 m, (b) 3506 m,
(c) 3525 m, (d) 3536 m, and (e) 3564 m.
FIG. 7. Indicating the uncertainty of Thomsens (1986) anisotropy parameters for the core samples under simulated in-situ stress.
The error bars were found using a simple Monte Carlo routine.
FIG. 8. Comparison of the exact (solid lines) qP-mode phase
velocities of the Brage mudrocks with the rst (dotted
lines), second (dashed-dotted lines), and third (dashed lines)
anisotropic approximations of Dellinger et al. (1993). The
largest mismatch between the anisotropic approximations and
the exact result is found for the 3506-mspecimen. Core depths
are (a) 3492 m, (b) 3506 m, (c) 3525 m, (d) 3536 m, and (e)
3564 m.
FIG. 9. Same as in Figure 8 but for the signicantly more
anisotropic qSV-mode. Core depths are (a) 3492 m, (b) 3506 m,
(c) 3525 m, (d) 3536 m, and (e) 3564 m.
Potential eld from a dc current source arbitrarily
located in a nonuniform layered medium
H edison Kiuity Sato
ABSTRACT
In this paper, I evaluate the potential eld due to a dc
current source located anywhere within a horizontally
layeredspace, all layers possessing exponentially varying
resistivities. The solution takes the formof Hankel trans-
forms with their kernel expressions containing functions
dened by recursion formulas. The resulting expressions
can be used to model any kind of resistivity array.
Specializing the general solution to full- or half-space
models possessing exponential dependence, I nd terms
I interpret as primary and secondary contributions.
Analytic expressions for the secondary electrical eld
revealed an error in the Stoyer and Wait solution for a
half-space.
To test the derived n-layer solution based on recur-
sion formulas, I modeled resistivity logs for both homo-
geneous layers and intercalated homogeneous and het-
erogeneous layers. Curves from the theory reproduced
asymmetries and ripples observed in real resistivity logs.
INTRODUCTION
Many authors have investigated the nature of the potential
eld due to a dc current source at the surface of an earth whose
resistivities vary continuously with depth. For the particular
case of exponential dependency of resistivities, Sunde (1949),
Paul and Banerjee (1970), and Stoyer and Wait (1977) have
studied the half-space model. Similar results, but for a layered
model, are reported by Banerjee et al. (1980a,b), Raghuwanshi
and Singh (1986), and Kim and Lee (1996). The last authors
foundarecursionrelationfor thekernels of aHankel transform
expression for the potential. All these investigations located
the current source at the surface, limiting the cases that could
be analyzedtoelectrical methods using electrode arrays spread
on the earths surface.
Investigations using buried electrodes are reported for earth
models having homogeneous layers. Alfano (1962) considers
Manuscript received by the Editor December 29, 1997; revised manuscript received February 25, 2000.
CPGG/UFBA, Campus Univ. da Federa c ao, Instituto de Geoci encias, Salvador, Bahia 40170-290, Brazil. E-mail: sato@cpgg.ufba.br.
a three-layer model and demonstrates that, by using buried
electrodes, the uncertainty in the interpretation of resistivity
soundings can be reduced. Daniels (1978) presents an analyti-
cal solution for the n-layer model and applies his solution to the
modeling of normal logs. Baumgartner (1996) uses submerged
electrodes to study underwater formations.
In this paper, I develop the analytical solution for the poten-
tial eld resulting from a dc current source buried in a layered
space, where the layer electrical resistivities vary exponentially
with depth. I limit myself to modeling normal logs for two the-
oretical situations, but my solution can model any kind of re-
sistivity array.
THE MODEL AND BASIC EQUATIONS
As shown in Figure 1, my system is an innite medium sliced
by n parallel horizontal planes, with the resulting consecutive
layers numberedfrom0 ton. Assuming a cylindrical coordinate
systemwhose z-axis is perpendicular totheinterfaceplanes and
pointing downward, I assume (1) planes dened by equations
z =z
0
, z =z
1
, . . . , z =z
n1
, (2) a current source with intensity I
located at (0, z
c
) in the mth layer, (3) an observer located at
(r, z), and (4) electrical resistivity of the i th layer as a function
of z given by
i
(z) =
i
exp(
i
z), (1)
where
i
and
i
are characteristic constants of the i th layer.
Introducing an articial interface dened by the plane z =z
c
,
the mth layer is divided into two layers, increasing the total
number of layers to n +2. With this procedure, two functions
rather than one describe the potential in the mth layer. In this
new system, the layers are designated by 0, 1, . . . , m, m
, . . . ,
n 1, n, where m represents the portionof the former mthlayer
bounded by the planes z =z
m1
and z =z
c
and where m
is the
complementary portion bounded by planes z =z
c
and z =z
m
.
The thickness of each layer is dened as shown in Figure 1.
The potential V
i
in each layer, except at the point (0, z
c
),
satises the equation
2
V
i
r
2
+
1
r
V
i
r
+

2
V
i
z
2
+
1
i
z
V
i
z
= 0, (2)
1726
dc Field in Nonuniform Layered Medium 1727
where
i
=1/
i
is the electrical conductivity. The solution of
this differential equation is well known. Following Kim and
Lee (1996), I write
V
i
=
_

0
_
A
i
e
+
i
z
+ B
i
e
i
z
_
J
0
(r) d. (3)
In equation (3), J
0
is the zero-order Bessel function, A
i
and B
i
are functions of to be determined, and
+
i
=
_
i
+
_
2
i
+4
2
__
2, (4)
i
=
_
i

_
2
i
+4
2
__
2. (5)
To compact the subsequent formulas,
i
is dened as
i
=
+
i

i
. (6)
BOUNDARY CONDITIONS
Convergence at innity
To guarantee the convergence of the potential V
0
when z
,
B
0
= 0. (7)
Similarly, the convergence of the potential V
n
when z +
is guaranteed by making
A
n
= 0. (8)
FIG. 1. Layered-space model.
Conditions at the interfaces
Except at thearticial interface z =z
c
, theelectrical potential
and the normal current density are continuous at the interfaces
z =z
i
, where i =0, 1, 2, . . . , n 1. The unknown functions A
i
,
B
i
of the potential V
i
and A
i +1
, B
i +1
of the potential V
i +1
satisfy
the equations
A
i
e
+
i
z
i
+ B
i
e
i
z
i
A
i +1
e
+
i +1
z
i
B
i +1
e
i +1
z
i
= 0,
(9)
1
i
(z
i
)
_
A
i
+
i
e
+
i
z
i
+ B
i
i
e
i
z
i
_
i +1
(z
i
)
_
A
i +1
+
i +1
e
+
i +1
z
i
+ B
i +1
i +1
e
i +1
z
i
_
= 0. (10)
Because of the split made in layer m, special care must be
taken with equations (9) and (10) at the interfaces bounding
that layer. At interface z =z
m1
, i =m 1 but i +1 is m. At
interface z =z
m
, i is m
while i +1 =m +1.
Conditions related to the current source
Thephysical constraints fromthecurrent sourcearehonored
with an approach similar to that of Sato (1996). The electrical
potentials V
m
and V
m
satisfy two physical conditions. The rst
condition is the continuity of the potential at the articial plane
interface z =z
c
, except at r =0. Then, the unknown functions
A
m
, B
m
, A
m
, and B
m
satisfy the expression
A
m
e
+
m
z
c
+ B
m
e
m
z
c
A
m
e
+
m
z
c
B
m
e
m
z
c
= 0. (11)
The second condition is charge conservation. Consider a
cylinder volume having height 2h and radius , its center co-
inciding with the current source of intensity I , as shown in
Figure 2. The total current owing out of the surface S of the
cylinder must be equal to the current intensity for any h and
values. In the limiting case h 0, the current through the
lateral cylindrical surface will tend to zero, assuming that the
current density has a nite value. Since the total current re-
duces to the sum of the current owing through the circular
base surfaces, the following equation holds:
lim
h0
_
_
2
=0
_

r=0
j
m
z=z
c
h
( z r dr d)
+
_
2
=0
_

r=0
j
m
z=z
c
+h
( z r dr d)
_
= I, (12)
FIG. 2. Cylindrical surface S around the current source.
1728 Sato
where z is the unit vector in the z-direction and j
i
=
i
V
i
,
the electrical current density in the layer i .
Using equations (1) and (3) to evaluate j
m
and j
m
in equa-
tion (12), integrating over and r, and rearranging the terms,
I obtain
_

0
2
c
I
3/2
_
A
m
+
m
e
+
m
z
c
+ B
m
m
e
m
z
c
A
m
+
m
e
+
m
z
c
B
m
m
e
m
z
c
_
()
1/2
J
1
() d =
1/2
, (13)
with
c
=1/
m
(z
c
), the electrical conductivity at the current
source vertical position. This equation must be satised for
any radius and resembles a Hankel transform integral. Using
Erd elyi (1954, p. 22), I have
A
m
+
m
e
+
m
z
c
+ B
m
m
e
m
z
c
A
m
+
m
e
+
m
z
c
B
m
m
e
+
m
z
c
=
I
2
c
. (14)
SOLUTION OF THE PROBLEM
The potential V
i
depends on knowing the pair of functions
A
i
, B
i
. In the present problem, n +2 pairs of A
i
, B
i
are dened
and their relations are subjected to the boundary conditions
derived in the previous section. These relations comprise a set
of n +2 linear equations.
As long as each one of these equations is related to a specic
z-coordinate, the set can be organized with increasing z-values.
Equation (7) is taken as the rst equation and equation (8) as
the last one. Between those extremes, the second through the
penultimate equations are (a) expressions (9) and (10) written
for i =0, . . . , m 1, (b) equations (11) and (14), and (c) equa-
tions (9) and (10) written for i =m, . . . , n 1. Organized in this
way, the set of equations is downsized recursively to a system
of two linear equations written in terms of the A
i
and B
i
pair.
When solved, the two linear equations determine the potential
in the i -layer.
Actually, the downsizing process consists of two sets of steps.
One starts withthe rst equationandconstitutes the downward
recurrence, while the second, starting with the last equation,
constitutes the upward recurrence. Both recurrences are de-
veloped in Appendices A and B.
Each recursion process has one or two stages, depending on
the observer position relative to the current source location.
For the points having z z
c
, the downward recurrence has one
stage andthe upwardhas two. Conversely, for the points having
z z
c
, the downward recurrence has two stages and the upward
has just one.
Potential expressions for points z
<
z
c
The pair of equations (A-3) and (B-10) can be solved for A
i
and B
i
, and the potential V
i
, valid for the layers 0, 1, . . . , m 1,
m, can be expressed as
V
i
=
I
2
c
_

0
1 g
i
exp[
i
(z z
i 1
)]
1 g
i
G
i
exp[
i
(z
i
z
i 1
)]
e
+
i
(z
i
z)
R
i
m
J
0
(r) d, (15)
remembering that z
i
=z
c
when i = m. Moreover, g
i
is obtained
from the stage 1 downward recurrence formulas (A-4) to
(A-7), and G
i
and R
i
is obtained from the stage 2 upward
recurrence formulas (B-11) to (B-13), combined with equa-
tions (B-7) to (B-9) and the stage 1 upward recurrence formu-
las (B-2) to (B-5) evaluated for layer m
.
Potential expressions for points z
>
z
c
In a similar manner, the pair of equations (A-13) and
(B-1) can be solved for A
i
and B
i
. The potential V
i
, valid for
the layers m
, m +1, . . . , n 1, n, becomes
V
i
=
I
2
c
_

0
1 G
i
exp[
i
(z
i
z)]
1 g
i
G
i
exp[
i
(z
i
z
i 1
)]
e
i
(zz
i 1
)
r
i
+
m
J
0
(r) d, (16)
remembering that z
i 1
=z
c
when i =m
. Furthermore, G
i
is ob-
tained from the stage 1 upward recurrence formulas (B-2) to
(B-5), and g
i
and r
i
are obtained from the stage 2 downward
recurrence formulas (A-14) to (A-16), combined with equa-
tions (A-9) to (A-11) and the stage 1 downward recurrence
formulas (A-4) to (A-7) evaluated for layer m.
HALF-SPACE AND FULL-SPACE MODELS
If I assume a two-layer case with a nonconductive layer 0
(representing the air) and a current source located at z
c
z
0
,
onlythetwopotentials V
1
and V
1
needbeconsidered. Following
the directions giveninthe previous section, these potentials are
V
1
=
I
2
c
_
_

0
e
+
1
(z
c
z)
+
1

1
J
0
(r) d
_

0
_
+
1
_
1
_
e
+
1
(z
c
z
0
)+
1
(zz
0
)
+
1

1
J
0
(r) d
_
(17)
and
V
1
=
I
2
c
_
_

0
e
1
(zz
c
)
+
1

1
J
0
(r) d
_

0
_
+
1
_
1
_
e
+
1
(z
c
z
0
)+
1
(zz
0
)
+
1

1
J
0
(r) d
_
.
(18)
Both equations (17) and (18) are a sum of two integral ex-
pressions. In the equations the second integral expressions are
equal and depend on z
0
, the coordinate of the air interface.
The rst integral in equations (17) and (18) does not depend
on z
0
. The second integral tends to zero when the air inter-
face recedes to , meaning that this integral represents a
secondary potential (V
1,S
). In both equations (17) and (18),
the rst integrals represent the primary potential (V
1,P
), the
potential attributable to a current source in a space having ex-
ponentially varying conductivity and not the potential from
a current source in a homogeneous space. Inserting the de-
nitions for
+
1
and
1
, I nd that both primary potential ex-
pressions can be integrated (Erd elyi, 1954) and represented
by a single expression:
V
1,P
=
I e
(zz
c
)
1
/2
4
c
e
|
1
/2|
r
2
+(zz
c
)
2
_
r
2
+(z z
c
)
2
. (19)
Moreover, considering the denitions for
+
1
and
1
, recogniz-
ing that
2
/
1
=
+
1
, and dening Z =z
c
z
0
+z z
0
, the
components E
1,S|r
and E
1,S|z
of the electrical eld associated
with the secondary potential can be obtained by taking deriva-
tives of the secondary potential in equations (17) or (18). The
expressions for the E
1,S|r
and E
1,S|z
components contain only
tabulated integrals (Erd elyi, 1954). Finally, I write them as
E
1,S|r
=
I e
(zz
c
)
1
/2
4
c
_
(|
1
| +
1
)
e
|
1
/2|Z
r
_
|
1
| +

1
Z
r
2
+Z
2
_
e
|
1
/2|
r
2
+Z
2
r
+
re
|
1
/2|
r
2
+Z
2
3
r
2
+Z
2
+|
1
/2|
re
|
1
/2|
r
2
+Z
2
r
2
+Z
2
_
,
E
1,S|z
=
I e
(zz
c
)
1
/2
4
c
e
|
1
/2|
r
2
+Z
2
r
2
+Z
2
1
2
+
Z[1 +|
1
/2|
r
2
+Z
2
]
r
2
+Z
2
_
. (20)
The term (|
1
| +
1
) exp(|
1
/2|Z)/r in the E
1,S|r
expres-
sion (20) was erroneously dropped in the analysis of Stoyer
and Wait (1977). This term is null only when
1
0; otherwise,
it is a function of (1/r), implying that its associated poten-
tial is r-logarithm dependent. This peculiarity is analogous to
the asymptotic form for the electrical eld far from a current
source located in a conductive layer overlying an innitely re-
sistive substratum. The result can be extrapolated and used as
a reminder when manipulating the n-layer earth model with
its deepest layer characterized by increasing resistivity with
depth (exponential dependency or not). The logarithm term
can appear in the potential values because it was referenced to
innity. If adequately accounted for, this fact does not consti-
tute a problemin situations involving a difference of potentials
between points, like those found in congurations using two
potential electrodes.
NUMERICAL RESULTS
I used two numerical tests to check the theoretical results,
modeling a polepole apparent resistivity sounding and two
resistivity logs. To evaluate the Hankel transform integrals, I
calledonthe digital ltering techniques of Anderson(1975). To
model the resistivity logs, I specied r-values that were small
compared with the tool size (16 or 64 inches), around 0.001 m
for the 16-inch normal tool and 0.01 m for the 64-inch tool.
These r-values represent a very small deviation of the log tool
alignment from the vertical. I conrmed the validity of my
procedure by comparing the resistivity log values in a homo-
geneous mediumobtained using the analytical expression with
those fromthe Hankel transformintegral formula evaluatedby
the digital ltering techniques. Tothe eye, the rst test achieved
results that exactly match the polepole apparent resistivity re-
sponses for the models used by Kim and Lee (1996).
In the second test, 16- and 64-inch normal resistivity logs
were computed for two earth models. The rst model had only
homogeneous layers; its resistivity logs are shown in Figure 3.
These log curves resemble those observed in the normal resis-
tivity logs shown in Daniels (1978) and in Jorden and Campbell
(1986) for a thick-bed, homogeneous, three-layer model.
The second model had both inhomogeneous (varying expo-
nentially with depth) and homogeneous layers. Its resistivity
logs are shown in Figure 4. In addition to the typical shape of
the resistivity curve associated with the homogeneous layers
located in the 5080 m interval of the model, other features
appeared when continuous depth variations were introduced.
These included the asymmetrical shapes strongly controlled
by the exponential depth dependency in the 80150 m in-
terval and ripples in the interval from 150165 m associated
with the fast variation of the resistivity with depth, combined
with the large distance between the electrodes in the 64-inch
tool.
CONCLUSIONS
I have found a theoretical solution for the multilayer
variable-resistivity geoelectric problem. My solution repro-
duces the characteristic shapes expectedfromidealizedmodels
of resistivity logs in a layered earth and are validated quantita-
tively by comparison with published results. This comparison
shows that the approach used to handle the physical conditions
imposed by the current source was correct.
FIG. 3. Apparent resistivities for 16- and 64-inch normal log
tools in an earth model composed of homogeneous layers.
1730 Sato
FIG. 4. Apparent resistivities for 16- and 64-inch normal log
tools in an earth model composed of layers having homoge-
neous or exponential depth-dependent resistivity.
Since no restrictions are imposed on the location of the ob-
server and the current source, the results can be used to eval-
uate the response of layered models, horizontally or vertically
stratied, for any kind of tool or array used in direct or in-
verse electrical modeling and tomography. My results offer a
new way to overcome the difculties (Van Dam, 1976) that
APPENDIX A
DOWNWARD RECURRENCE
My objective is to establish a linear equation with variables A
i
and B
i
, incorporating the physical and geometric parameters
of the layers and interfaces above layer i .
Stage 1
This stage involves only equations with variables A
i
and B
i
, for
i =0, 1, . . . , m 1, m
. Actually, equation (7), conditioning the

potential at , is the rst equation of the set. Equation (7),
along with the boundary condition equations (9) and (10) at
the interface z =z
0
, can be simplied to an equation with the
variables A
1
and B
1
. Repeating the process with equations (9)
and (10) at the interface z =z
1
, and so on, the variables A
i 1
and B
i 1
satisfy an equation resembling
x
i 1
A
i 1
+ B
i 1
= 0. (A-1)
Equation (A-1), combined with equations (9) and (10) at the
interface z =z
i 1
, gives an equation with the variables A
i
and
B
i
that can be written as
1

+
i
+
i 1
s
i,i 1
1 x
i 1
exp(
i 1
z
i 1
)
1
_
i 1
_
+
i 1
_
x
i 1
exp(
i 1
z
i 1
)
1

+
i 1
s
i,i 1
1 x
i 1
exp(
i 1
z
i 1
)
1
_
i 1
_
+
i 1
_
x
i 1
exp(
i 1
z
i 1
)
A
i
exp(
i
z
i 1
) + B
i
= 0, (A-2)
arise in interpreting electrical data related to gradual changes
of groundwater salinity with depth.
ACKNOWLEDGMENTS
The author was supported in this research by a fellowship
from Conselho de Desenvolvimento Cientco e Tecnol ogico
(CNPq), Brazil. I thank professors Irshad Mufti and Ben
Clennel, who carefully read this paper and made suggestions
which certainly improved this work. Also, I extend my grati-
tude to the SEG associate editor and the reviewers for their
constructive comments and discussions. Finally, I express my
deepest thanks to Dr. Frank Levins.
REFERENCES
Alfano, L., 1962, Geoelectrical prospecting with underground elec-
trodes: Geophys. Prosp., 10, 290303.
Anderson, W. L., 1975, Improved digital lters for evaluating Fourier
and Hankel transformintegrals: U.S. Geol. Survey Technical Report
PB 242 800.
Banerjee, B., Sengupta, B. J., and Pal, B. P., 1980a, Apparent resistivity
of a multilayered earth with a layer having exponentiality varying
conductivity: Geophys. Prosp., 28, 435452.
1980b, Resistivity sounding on a multilayered earth containing
transition layers: Geophys. Prosp., 28, 750758.
Baumgartner, F., 1996, A new method for geoelectrical investigations
underwater: Geophys. Prosp., 44, 7198.
Daniels, J. J., 1978, Interpretation of buried electrode resistivity data
using a layered earth model: Geophysics, 43, 9881001.
Erd elyi, A., 1954, Tables of integral transforms II: McGraw-Hill.
Jorden, J. R., and Campbell, F. L., 1986, Well logging IIElectric and
acoustic logging: Soc. Petr. Eng.
Kim, H., andLee, K., 1996, Responseof amultilayeredearthwithlayers
having exponentially varying resistivities: Geophysics, 61, 180191.
Paul, M. K., and Banerjee, B., 1970, Electrical potential due to a
point source uponmodels of continuously varying conductivity: Pure
Appl. Geophys., 80, 218237.
Raghuwanshi, S. S., and Singh, B., 1986, Resistivity sounding on a hori-
zontally stratied multi-layered earth: Geophys. Prosp., 34, 409423.
Sato, H. K., 1996, Electrical current source in the interior of horizontal
layers having conductivities varying potentially with depth: Ph.D.
thesis, Federal Univ. Bahia (in Portuguese).
Stoyer, C. H., and Wait, J. R., 1977, Resistivity probing of an expo-
nential earth with a homogeneous overburden: Geoexploration,
15, 1118.
Sunde, E. D., 1949, Earth conduction effects in transmission systems:
Van Nostrand.
Van Dam, J. C., 1976, Possibilities and limitations of resistivity method
of geoelectrical prospecting in the solution of geohydrological prob-
lems: Geoexploration, 14, 179193.
where s
i,i 1
=
i 1
(z
i 1
)/
i
(z
i 1
). Inspecting this equation and
comparing it to equation (A-1), I found it convenient to dene
x
i 1
=g
i 1
exp(
i 1
z
i 2
). Then equation (A-2) can be rewrit-
ten in a generalized form,
g
i
A
i
exp(
i
z
i 1
) + B
i
=0, (3)
with the following recursion relations:
g
i
=
1
_
+
i
_
+
i 1
_
s
i,i 1
f
i
1
_
i
_
+
i 1
_
s
i,i 1
f
i
, (4)
f
i
=
1 g
i 1
exp(
i 1
h
i 1
)
1
_
i 1
_
+
i 1
_
g
i 1
exp(
i 1
h
i 1
)
, (5)
.
.
.
f
1
= 1, (6)
g
0
= 0. (7)
Stage 2
Equation (A-3) evaluated for i = m, combined with equa-
tions (11) and (14), both related to the physical constraints
from the current source, gives the rst equation of the second
set of equations in the downward recurrence:
g
m
A
m
exp(
m
z
c
) + B
m
=
I
2
c
+
m
r
m
exp
_
m
z
c
_
,
(8)
where
g
m
=
1 f
m
1
_
m
_
+
m
_
f
m
, (9)
r
m
=
f
m
1
_
m
_
+
m
_
f
m
, (10)
f
m
=
1 g
m
exp(
m
h
m
)
1
_
m
_
+
m
_
g
m
exp(
m
h
m
)
. (11)
Equation (A-8) includes a nonnull independent term. There-
fore, the successive combinations with the continuity equa-
tions (9) and(10) at the interfaces z =z
m
, . . . , z
n1
give anequa-
tion like
x
i 1
A
i 1
+ B
i 1
= y
i 1
. (12)
Using a procedure similar to that of stage 1, I can write an
equationwiththe variables A
i
and B
i
, validfor the layers i =m
,
m +1, . . . , n 1, n, in the following form:
g
i
A
i
exp(
i
z
i 1
) + B
i
=
I
2
c
+
m
r
i
exp
_
i
z
i 1
_
,
(13)
where g
i
andr
i
are dened by the following recursion relations:
g
i
=
1
_
+
i
_
+
i 1
_
s
i,i 1
f
i
1
_
i
_
+
i 1
_
s
i,i 1
f
i
, (14)
r
i
= r
i 1
exp
_
i 1
h
i 1
_
_
1
_
i 1
_
+
i 1
_
f
i
1
_
i
_
+
i 1
_
s
i,i 1
f
i
_
,
(15)
f
i
=
1 g
i 1
exp(
i 1
h
i 1
)
1
_
i 1
_
+
i 1
_
g
i 1
exp(
i 1
h
i 1
)
, (16)
and h
m
should be interpreted as h
m
, as well as z
m1
as z
c
.
APPENDIX B
UPWARD RECURRENCE
In a way similar to that of Appendix Abut working in the op-
posite direction, my objective is to establish an equation with
the variables A
i
and B
i
, incorporating the physical and geo-
metric parameters of the layers and interfaces below layer i .
Stage 1
Equation (8) is the rst equation of the set with the variables
A
i
and B
i
for i =n, n 1, . . . , m +1, m
. By combining equa-
tion (8) with equations (9) and (10), evaluated at the interfaces
z =z
n1
, z =z
n2
, etc., I write the following equation:
A
i
+ G
i
B
i
exp(
i
z
i
) = 0, (B-1)
with the recursion relations
G
i
=
1
_
i
_
i +1
_
S
i,i +1
F
i
1
_
+
i
_
i +1
_
S
i,i +1
F
i
, (B-2)
F
i
=
1 G
i +1
exp(
i +1
h
i +1
)
1
_
+
i +1
_
i +1
_
G
i +1
exp(
i +1
h
i +1
)
, (B-3)
.
.
.
F
n1
= 1, (B-4)
G
n
= 0, (B-5)
where S
i,i +1
=
i +1
(z
i
)/
i
(z
i
).
Stage 2
Taking equation (B-1) evaluated for i =m
, along with equa-

tions (11) and (14) relative to the current source, I write the
rst equation of the second set of equations in the upward re-
currence as
A
m
+ G
m
B
m
exp(
m
z
c
)
=
I
2
c
m
R
m
exp
_
+
m
z
c
_
, (B-6)
where
G
m
=
1 F
m
1
_
+
m
_
m
_
F
m
, (B-7)
R
m
=
F
m
1
_
+
m
_
m
_
F
m
, (B-8)
F
m
=
1 G
m
exp(
m
h
m
)
1
_
+
m
_
m
_
G
m
exp(
m
h
m
)
. (B-9)
Equation (B-6) has a nonnull independent term. Therefore,
successive combinations with the continuity equations (9) and
1732 Sato
(10) at the interfaces at z =z
m1
, . . . , 0 will give equations rep-
resented by
A
i
+ G
i
B
i
exp(
i
z
i
) =
I
2
c
m
R
i
exp
_
+
i
z
i
_
(B-10)
with the following recurrence relations:
G
i
=
1
_
i
_
i +1
_
S
i,i +1
F
i
1
_
+
i
_
i +1
_
S
i,i +1
F
i
, (B-11)
R
i
= R
i +1
exp
_
+
i +1
h
i +1
_
_
1
_
+
i +1
_
i +1
_
F
i
1
_
+
i
/
i +1
_
S
i,i +1
F
i
_
,
(B-12)
F
i
=
1 G
i +1
exp(
i +1
h
i +1
)
1
_
+
i +1
_
i +1
_
G
i +1
exp(
i +1
h
i +1
)
. (B-13)
Here, h
m
should be interpreted as h
m
and z
m
as z
c
.
A dual-grid nonlinear inversion technique with applications
to the interpretation of dc resistivity data
Carlos Torres-Verdn
, Vladimir L. Druskin
, Sheng Fang
,
Leonid A. Knizhnerman
, and Alberto Malinverno
ABSTRACT
We develop a solution to the nonlinear inverse prob-
lem via a cascade sequence of auxiliary least-squares
minimizations. The auxiliary minimizations are nonlin-
ear inverse problems themselves, except that they are
implementedwithanapproximate forwardproblemthat
is at least an order of magnitude faster to solve than
the algorithm used to simulate the measurements. Any
given auxiliary minimization in the cascade is fully self-
contained and yields a solution of the unknown model
parameters. This solution, in turn, is used to perform a
numerical simulation of the data to quantify its agree-
ment with the input measurements. If the difference
between the measured data and the simulated data is
above the estimated noise threshold, it is input as data to
the subsequent auxiliary inverse problemin the cascade.
Otherwise, the inversion is brought to a successful com-
pletion. We describe the theory andoperating conditions
under which this cascade inversion approach converges
to a solution equivalent to that of a single inversion im-
plemented with the original forward problem.
Depending on the choice of approximate forward
problem and regardless of the computer algorithm em-
ployed to solve the inversion, the cascade sequence of
auxiliary minimizations can be made many times more
efcient than a single inversion performed with the orig-
inal forward problem. In this paper, we consider a prac-
tical and exible way to construct the approximate for-
ward problem by using a subset of the nite-difference
grid used to simulate the measurements numerically, i.e.,
a dual-grid construction approach.
Numerical examples of performance are described
based on the inversion of cross- and single-well 2.5-D
direct-current (dc) resistivity data. Our results suggest
that the proposed dual-grid minimization technique may
be able to improve greatly the performance of even
the most effective multidimensional nonlinear inversion
procedures currently in use.
INTRODUCTION
Inversion of geophysical data is a central component of the
noninvasive probing and characterization of the subsurface.
Regardless of the computer resources devoted to this task,
there is a permanent need for more efcient and more ro-
bust procedures that can map large volumes of data accurately
into space and/or time distributions of the material property
underlying the physics of the measurements.
One popular inversion procedure consists of minimizing a
quadratic cost function that emphasizes the sumof the squared
differences between the measured and the simulated data. Be-
cause in most cases the relationship between the property dis-
tribution function and the simulated data is nonlinear, the min-
Manuscript received by the Editor November 4, 1996; revised manuscript received December 30,1999.
University of Texas at Austin, Department of Petroleum and Geosystems Engineering, Austin, Texas 78712-1061. E-mail: cverd n@pe.utexas.edu.
Schlumberger-Doll Research, Old Quarry Road, Ridgeeld, Connecticut 06877. E-mail: druskin@ridgeeld.sdr.slb.com; alberto@ridgeeld.
sdr.slb.com.
Baker Atlas, 10205 Westheimer, Houston, Texas 77042. E-mail: sheng.fang@waii.com.
Central Geophysical Expedition, Narodnogo Opolcheniya Street., House 40, Building 3, Moscow 123298, Russia. E-mail: mmd@cge.ru.
imization is performed with a nonlinear search technique con-
structed by way of a suitable number of sequential linear steps
(or iterations) which eventually trace the road toward an ex-
tremum of the cost function. The way this search is performed
is usually a compromise between efciency (mainly speed) of
the computations and stability (convergence) of the method. A
common approach is referred to as the Gauss-Newton method,
in which, at a given step in the search procedure, only rst-
order variations of the cost function are used to direct the
search toward the next stop point. This requires that a matrix
of rst-order variations of the simulated data with respect to
a variation in the model parameters be computed at each iter-
ation, the so-called Jacobian, or sensitivity, matrix. Generally,
1733
1734 Torres-Verdn et al.
computing the Jacobian matrix at each linear step represents
the most time-consuming element of the minimization (see,
for instance, Oristaglio and Worthington, 1980). It is, then, not
surprising that a plethora of procedures has been put forth to
approximate, bypass, and/or economize the effect the Jacobian
matrix has at every linear step in the search for the extremum
(see, for instance, Smith and Booker, 1991; Ellis et al., 1993;
Mackie and Madden, 1993; Torres-Verd n and Habashy, 1994;
and Farquharson and Oldenburg, 1995, among others).
Whereas the efcient computationof the Jacobianmatrix is a
subject of much-needed research, comparatively less emphasis
has been placed on deriving alternative minimization schemes
that could be less taxing in the search of the extremum. In this
paper, we develop and test one such alternative scheme. The
proposed method consists of a cascade sequence of auxiliary
least-squares minimizations. These auxiliary minimizations are
nonlinear inverse problems themselves, except that they are
implemented with an approximate forward problem that is
much faster to solve than the forward problemused to simulate
the measurements numerically. Once the approximate forward
problemhas been constructed, it remains unchanged for all the
auxiliary minimizations in the cascade. Subsequently, by solv-
ing each auxiliary minimization, we obtain a solution to the
unknown model parameters. This solution, in turn, is used to
simulate the measurements with the original forward problem.
If the difference between the simulated data and the measured
data falls below a specied threshold, then the cascade stops
and the current model solution is taken as the output of the
inversion. Otherwise, this difference is input as data for the
subsequent auxiliary minimization in the cascade. We derive
operating conditions under which this cascade process not only
converges but also converges to the same extremum as a min-
imization scheme implemented solely with the exact forward
problem and regardless of the computer algorithm chosen to
perform the minimization. The main advantage of the cascade
inversion procedure thus lies in the fact that the exact forward
problem needs to be computed only as many times as auxiliary
minimizations are solved before the cascade reaches conver-
gence. Moreover, at no point in this process does one need to
perform a minimization with the exact forward code. This es-
sentially does away with the need to compute the effect of the
sensitivity matrix at each stop point in the iterative solution of
the full nonlinear inverse problem.
By way of example, in this paper we consider the construc-
tion of the approximate forward problem by taking a subset of
the nite-difference grid used to simulate the measurements
numerically. This is why we have used the term dual-grid non-
linear inversionto designate the proposed inversion procedure.
The approach of choosing a subset of the original grid allows
great exibility to control the efciency and convergence prop-
erties of the cascade. Our presentation concludes with several
examples of parametric inversion drawn from the interpreta-
tion of single- and crosswell direct-current (dc) resistivity data
simulated under the assumption of 2-D variations of electrical
conductivity.
A TEST EXAMPLE
To focus the discussion on a practical geophysical inverse
problem, let us consider the example described in Figure 1:
An array of ve dc electric current sources is deployed at
10-m spacings along the borehole indicated on the left side
of the diagram. These sources are used to generate a spatial
distribution of electric potentials in the surrounding rock
formations. Sampling these potentials is the rst step toward
estimating the distribution of electrical conductivity in the
surrounding formations. To that effect, we consider an array
of 18 contact electrodes deployed along a second borehole
located on the right-hand side of Figure 1. The spacing
between contiguous electrodes is not uniform. All the voltage
measurements acquired with the array of contact electrodes
are referred to a common ground, or reference potential,
located at innity. Finally, a third borehole is assumed to
be located midway between the borehole with the current
sources and the borehole with the contact electrodes. This
third borehole is not equipped with electric sources or with
contact electrodes and is used solely to inject saltwater into
the surrounding rock formations within the depth interval
from 30 m to 80 m. An experiment with similar characteristics
was designed and implemented at the Richmond Field Station
of the University of California, Berkeley (Bevc and Morrison,
1991), with the intent of monitoring in time the diffusion of the
injected saltwater into an existing aquifer. In the subsurface
model shown in Figure 1 and merely for the sake of simplicity,
FIG. 1. Geometric description of the subsurface test model.
Thegureshows threeboreholes withinthesamecross-section.
To the left, ve dc electric current sources are deployed at a
uniform depth interval of 10 m. Contact-point electrodes are
deployed along the well shown on the right of the gure. The
third borehole is located midway between the borehole with
the sources and the borehole with the contact electrodes and is
usedtoinject saline water intothe surrounding rockformations
along the depth interval from 30 m to 80 m. Notice that the
contact electrodes are not assumed to be spaced evenly; there
is a total of 18 of them.
Dual-grid Nonlinear Inversion 1735
we make the additional geometric assumption that the injected
water diffuses into the surrounding formations in a radially
symmetric manner. This corresponds to a 2-D model of elec-
trical conductivity. However, the electric potential generated
by the sources remains a 3-D scalar function, whereupon
the geometric complexity of the simulation problem can be
referred to more properly as 2.5-D.
The objective of the dc electrical experiment described in
Figure 1 is to generate electric potentials in the subsurface with
each current source acting separately, and to measure these
potentials at each of the existing contact electrodes. The set of
electric potentials measured with all the possible combinations
of source-electrode pairs constitutes the data from which the
distribution of electrical resistivity between wells can be esti-
mated in principle via an inversion process. An additional data
set is gathered along the borehole with the electric sources by
placing contact electrodes at source locations.
Synthetic (numerical) and eld data examples of limited-
angle dc electrical impedance tomography suggest that the
inversion is severely ill-conditioned and hence at best can
estimate a few salient features of the distribution of electrical
resistivity between wells (see, for instance, Daily and Owen,
1991; Daily et al., 1992; and Sasaki, 1992, among others).
Therefore, in an attempt to reduce the number of unknown
variables in the problem but without sacricing generality, we
have taken the approach of parameterizing the prole of water
invasion with only a few variables. One such prole of water
invasion is illustrated in Figure 2. The simplied way we have
FIG. 2. A sample description of the electrical resistivity model
assumed after the inception of water injection. The injection of
water into the background resistivity model shown in Figure 1
gives risetoavertical variationintheradius of invasionbecause
of existing vertical variations in permeability. We consider six
such variations of 1-D permeabilities along the depth interval
where water injection takes place. It also is assumed that the
invasion is axially symmetric and that the invaded rocks exhibit
a uniform electrical resistivity of 0.5 m.
chosen to describe the electrical resistivity distribution consists
of assuming that the rocks saturated with saltwater exhibit
a uniform electrical resistivity regardless of the resistivity
of the uninvaded rock matrix. This is a justied assumption
because, according to Archies law, when saltwater is made to
ll the available pore space in permeable rocks, the electrical
resistivity of the former becomes the dominant contribution
to the effective rock resistivity, especially with large values of
porosity. In actual eld experiments, it is desirable to make
the injected saltwater as conductive as possible to increase
the resistivity contrast with the surrounding rock formations
and hence maximize the anomalous dc response. Assuming
arbitrarily that the electrical resistivity of the invaded rocks
is equal to 0.5 .m, the remaining unknown variables are the
invasion radii at the various depth levels of the water-injection
experiment. As indicated in Figure 2, we have chosen arbi-
trarily to parameterize the vertical segment of water injection
with six such radii of invasion. By choosing this simplied
parameterization, we also intend to concentrate the discussion
on the mechanical details of the inversion rather than on
the trade-offs between resolution and uncertainty that are at
the heart of any geophysical inverse problem.
NUMERICAL SIMULATION OF THE 2.5-D DC PROBLEM
The differential equation satised by the electric potential,
U(r), is given by
[(r)U(r)] = J
S
(r), (1)
where r is the observation point, (r) is the distribution of
electrical conductivity, and J
S
(r) is the impressed dc current
source. In this paper, a solution of equation (1) is approached
by using a nite-difference formulation in cylindrical coordi-
nates via the spectral Lanczos decomposition method (SLDM)
(Druskin and Knizhnerman, 1995). Details of this highly ef-
cient method of solution are discussed in Appendix A. Ex-
tensive self-consistency checks and benchmarking exercises
with homogeneous and 1-D models of electrical conductivity
showedthat our simulations were accurate withinless than1%.
The results from one such exercise are shown in Figure 3, with
plots of percent errors obtained by simulating the single- and
crosswell potentials resulting from a single current source. For
these simulations, we considered the layered sequence of beds
shown in Figure 1 and an electric source located at a depth of
45 m. We computed the potentials resulting from the layered-
bedmodel withanaccurate 1-Dalgorithm, andthese potentials
served as reference to calculate the percent errors caused by
the SLDMalgorithm. The plots shown in Figure 3 indicate per-
cent errors larger for the single-well case than for the crosswell
case, but in both cases the errors are less than 1% in absolute
value. A single-source simulation on a 2-D grid of size 41 49
was clocked at 10 seconds of CPU time when implemented in
a Silicon Graphics Power Challenge machine.
Figure 4 is a plot of the electric potential U(r) in volts sim-
ulated along the same model cross-section shown in Figure 1
when the dc current source is placed at a depth of 45 m within
the borehole shown on the left side of the same gure. The
values of electric potential in Figure 4 are described as a color
map and with a logarithmic scale (log
10
base) in an effort to dis-
play somewhat uniformly the wide range of numbers spanned
by U(r). Figure 5, on the other hand, is a plot constructed in
FIG. 3. Accuracy assessment of the SLDM numerical simula-
tion algorithm. The plots describe percent differences between
the electric potentials simulated with the SLDMalgorithmand
the same potentials simulated with a 1-D algorithm. The sub-
surface model of electrical resistivity is the layered sequence
of beds shown in Figure 1, and the electric source is located at
a depth of 45 m. Percent potential differences are shown for
both single- and crosswell measurements.
FIG. 4. Electric potential in volts (log
10
-base) simulated for the
1-D resistivity model shown in Figure 1 and along the same
cross-section. The dc current source is located at a depth of
45 m in the borehole shown on the left in Figure 1.
the same way as Figure 4 but describing the electric potential
simulatedfor themodel of water injectionshowninFigure2. Fi-
nally, Figure 6 displays the relative percent differences between
the previous two simulations. Notice that the largest and most
prominent potential anomaly occurs at the boundaries of the
largest invasion radius. The remaining boundaries, at best, are
suggested slightly by the distribution of anomalous potential.
NONLINEAR INVERSION AS MINIMIZATION
Let m be the size-N vector of unknown variables that fully
describes the underlying electrical resistivity distribution (in
this case, the radii of water invasion), and m
R
a reference vec-
tor of the same size as m which has been determined from
some a priori information. Often, the estimation (inversion) of
mfrom the measured data is undertaken by solving the equiv-
alent problem of minimizing a cost function dened with some
metric. Let C(m) be one such quadratic cost function given by
(Torres-Verdn and Habashy, 1994)
2C(m) =
__
_
W
d
_
d(m) d
obs
__
_
2
2
_
+W
m
(mm
R
)
2
, (2)
where
FIG. 5. Electric potential in volts (log
10
-base) simulated for the
water-ood resistivity model shown in Figure 2 and along the
same cross-section. The dc current source is located at a depth
of 45 m in the borehole shown on the left of Figure 2 (compare
with Figure 4).
d
obs
is the measured data vector, usually corrupted with some
level of noise, W
d
is the inverse of the data covariance matrix,
2
is the prescribed value of data mist, d(m) is the data vector
numericallysimulatedfor specic values of m, W
m
is theinverse
of the model covariance matrix, and is a Lagrange multiplier
or regularization parameter.
A common technique used to nd a stationary point (or
extremum), m, where the cost function attains a minimum,
consists of performing a Gauss-Newton xed-point iteration
search (Gill et al., 1981). This method disregards second-order
variations of the cost function in the vicinity of a local iteration
point. The corresponding iterated formula can be written as
m
k+1
=
_
J
T
(m
k
) W
T
d
W
d
J(m
k
) +W
T
m
W
m
_
1
_
J
T
(m
k
) W
T
d
W
d
_
d(m
k
) d
obs
+J(m
k
) m
k
_
+W
T
m
W
m
m
R
_
, (3)
subject to
l
i
m
k+1
i
u
i
.
In the above expression, the superscript k is used as an iteration
count, the superscript T denotes transpose, and J(m) is the
FIG. 6. Relative percent difference between the electric poten-
tial simulated for the water-ood resistivity model shown in
Figure 2 and the electric potential simulated for the 1-D back-
ground shown in Figure 1. The dc current source is located at
a depth of 45 m in the borehole shown on the left in Figure 1.
Jacobian matrix of C(m), given by
J(m) =
_
_
_
_
_
_
_
_
_
_
_
d
1
/m
1
d
1
/m
d
1
/m
N
.
.
.
.
.
.
.
.
.
.
.
.
d
j
/m
1
d
j
/m
d
j
/m
N
.
.
.
.
.
.
.
.
.
.
.
.
d
M
/m
1
d
M
/m
d
M
/m
N
_
_
MN
The upper and lower bounds enforced on m
k+1
are intended
to have the iterated solution yield only physically consistent
values (in the problem at hand, for instance, a trivial choice of
lower bound consists of restricting the unknown invasion radii
to take only positive numbers).
When the linear system of equations embodied in equa-
tion (3) is solved for subsequent values of m in the search
of a minimum of the quadratic cost function, evaluating the
Jacobian matrix remains the most computationally demanding
operation. The xed-point iteration search for a minimum of
C(m) is concludedwhenthe measureddata have beent within
the prescribed tolerance,
2
.
There are, of course, alternative implementations of the
Gauss-Newton algorithmthat, under particular conditions, can
be more efcient than the general Gauss-Newton procedure.
For instance, Sasaki (1992) has used a formulation in which
the source-potential reciprocity is invoked to reduce the num-
ber of computations needed to assemble the Jacobian matrix.
This procedure becomes extremely efcient when the mea-
surements consist of a large number of independent source
locations. Steepest descent and/or conjugate gradient proce-
dures also have been proposed by Mackie and Madden (1993),
in which at most only a few elements of the Jacobian matrix
are needed (and certainly no inversion of the Jacobian matrix
is needed) at each step of the nonlinear inversion. All that is
required is the product of the Jacobian matrix times a vector.
Albeit efcient in their computations, such procedures may
take a very large number of linear steps before converging
to the minimum of the cost function. The specic choice of
a nonlinear minimization algorithm is determined largely by
the underlying physics of the problem, as well as by the mea-
surement conditions and the assumed model parameterization
(number and type of parameters). In this paper, we have no
intention of ponticating a particular adaptation of the Gauss-
Newton or steepest descent procedures to solve the nonlinear
minimizations. Our aim is to take one step back and reformu-
late the inverse problem in a manner that makes it naturally
more efcient regardless of the specic algorithmused to solve
the minimization.
A NEWMINIMIZATION PROCEDURE
Inthe example at hand, the forward-modeling operator d(m)
consists of solving the partial differential equation (1) with a
nite-difference grid. Figure 7 is a graphical description of the
nite-difference grid used to that end. It consists of 41 nodes
in the radial direction (away from the injection well) and 49
in the vertical direction (the gure shows only a fraction of
the grid). This is the actual grid used to perform the numeri-
cal simulations of the potential shown in Figures 3 through 5.
The grid boundaries coincided with the coordinates = 6 km
and z = 6 km, respectively. Because we have assumed axial
symmetry in the electrical-conductivity distribution, the solu-
tionalgorithmdescribedinAppendix Arequires discretization
only along the (, z) plane ( 0). This is why we are showing
only the nite-difference mesh along the plane in Figure 7.
Usingshorthandnotation, let us identifytheinverseoperator
with the formula
m = d
1
(d
obs
), (4)
where d
1
(d
obs
) is in general nonlinear. In analogy with the
forward and inverse operators dened above, we introduce
auxiliary forward and inverse operators denoted as
d = h( m) (5)
and
m = h
1
(

d), (6)
respectively, where

d is a modied measurement vector yet to
be dened. The symbol m is used here only for labeling pur-
poses to identify when a model vector has been derived from
the operation h
1
(

d). Heuristically, there are two important
conditions we would like to impose on the construction of the
above auxiliary forward and inverse operators:
1) h(m) should be much faster to compute than d(m), and
2) h
1
(

d) should be bounded in a special way (to be dened
shortly).
FIG. 7. Detail of the 2-D nite-difference grid used to per-
form the numerical simulation of single- and crosswell dc re-
sistivity measurements. The complete grid consist of 41 nodes
in the vertical direction and 49 in the radial direction, with
outer boundaries extending to 6 km in the radial direction and
6 km in the vertical direction. The circles indicate voltage
measurement points.
Condition (1) is highly desirable from a practical point of
view, because if a relationship exists between h
1
and d
1
, then
solving for the former via repeated calls to h(m) could save,
in principle, signicant computer time. Condition (2), on the
other hand, not only secures the possibility of inverting h(m)
in a stable manner, but if the bound is designed properly, it also
could enforce a stable contractive mapping of the differences
between h(m) and d(m), as is shown next.
The objective pursued here is to construct a series of approx-
imations m to m by performing inversions with the operator
h
1
(

d). On convergence, such series of approximations should
yield m equal to m, but the computer efciency still should
be in favor of performing the repeated inversions of m. We
construct this series of approximations through a xed-point
iteration recurrence given by
m
k+1
= h
1
[h( m
k
) d( m
k
) +d
obs
] (7)
where, once again, the superscript k is used to denote iteration
count. We rst remark that if the above xed-point iteration is
to converge, then, on convergence,
m
k+1
= m
k
.
Recall also that by denition
m
k
= h
1
[h( m
k
)],
whereupon it follows that
d
obs
d( m
k
) = 0.
If d
1
(d
obs
) has a solution equal to m, then it is obvious that
m
k
= m.
We therefore conclude that if the xed-point iteration given
by equation (7) is a contractive mapping, it converges to the
solution of the original inverse problem d
1
(d
obs
). By direct
inspection of equation (7), it also follows that
d = h( m
k
) d( m
k
) +d
obs
. (8)
The remaining important issue todeal withhere concerns the
operating conditions that will guarantee the xed-point itera-
tion given by equation (7) to converge. Evidently, the smaller
the Euclidean norm of the difference [h(m) d(m)], the closer
to global convergence the xed-point iteration will be, but this
is hardly a quantitative statement. In Appendix B, we prove
that convergence is guaranteed when the following bounding
condition is satised:
h
1
(

d) [h(m) d(m)] 1, (9)
where the operator designates the Jacobian matrix (Fr echet
derivative) of the corresponding vector function; accordingly,
using the notation introduced earlier, d(m) =J(m). Apart
from being a rigorous design restriction, the above bounding
condition is not necessarily a practical recipe to construct the
auxiliary forward-modeling operator h(m). The message is sim-
ply that the closer the Jacobianmatrix of h(m) is tothe Jacobian
matrix of d(m), the closer to global convergence the recurrence
(7) will be, provided that h
1
(
d) is bounded appropriately as
well.
We summarize the proposed nonlinear inversion procedure
with the following sequence of steps:
1) Construct the auxiliary forward-modeling operator h(m)
to yield to a faster method of solution than the one used
to compute d(m), but subject to the condition
h
1
(
d) [h(m) d(m)] 1.
This construction customarily would be achieved by trial
anderror, but wouldbedoneonlyonce(andevenperhaps
only once for a large variety of data sets).
2) Provide an initial guess for m.
3) Compute d(m); if the Euclidean norm of [d(m) d
obs
] is
below the specied threshold, then the inversion stops and
the current value of mis accepted as the extremumof the
least-squares cost function. Otherwise, compute
d = h(m) d(m) +d
obs
.
4) Find the extremum of the auxiliary cost function
2C(m) =
_
W
d
[h(m)

d]
2
2
_
+W
m
(mm
R
)
2
.
5) Return to step (3).
Because the modied data vector

d is reset every time the
above process reaches step (3) before convergence, each of
the auxiliary cost functions of step (4) is dened in a cascade
fashion whereby a given cost function depends on the results
obtained from minimizing the previous cost function in the
cascade. This is why we have adopted the term cascade mini-
mization to refer to the new inversion procedure developed in
this paper.
The critical step in the cascade minimization is the construc-
tion of the auxiliary operator h(m), which remains unchanged
throughout this process. Without loss of generality, in this pa-
per we have chosen to dene the auxiliary (approximate) for-
ward operator by using a subset of the nite-difference grid
used to simulate d(m). The specic subset of the original nite-
difference grid should be chosen to expedite signicantly the
computation of h(m). Additionally, the choice of grid subset
should abide by condition (9). For the problem at hand, exten-
sive experimentation by trial and error with numerical simula-
tions and checks to bounding condition (9) led us to the use of
thenite-differencegridillustratedinFigure8for thecomputa-
tionof h(m). The mainfeature of this gridis that withinthe area
where the conductivity distributionis expectedtochange inthe
course of the minimization, the grid is as spatially ne as the
grid shown in Figure 7 that is used for the computation of d(m).
Outside this area, however, the grid is assigned larger spatial
steps in the radial and vertical directions to reach the same
boundaries used in the construction of the ne grid. There is a
total of 27 nodes in the vertical direction and 26 in the radial di-
rection for the coarse grid (compare with 41 49 for the ne
grid). Anumerical simulation with this coarse grid was clocked
at approximately 2 seconds of CPU time when implemented
in a Silicon Graphics Power Challenge machine (compare with
10 seconds of CPU time for the ne grid). Minimization of the
auxiliary least-squares cost functions was performed with the
Gauss-Newton procedure described by equation (3). We also
found that for the problem at hand, the computation of the
Jacobian matrix within a given auxiliary minimization could be
simplied by using the Broydens update procedure described
inAppendix C. Finally, the attending linear systemof equations
(3) was solved by enforcing upper and lower bound constraints
of 0 mand 60 m, respectively, using a constrained least-squares
algorithm from Lawson and Hanson (1974).
EXAMPLES OF PARAMETRIC INVERSION
The foregoing nonlinear inversion procedure was imple-
mented in the estimation of radii of water invasion for the
hypothetical experiment described in Figures 1 and 2. As men-
tioned earlier, there are six unknown radii of invasion, and
data consist of single- and crosswell dc resistivity experiments
with 18 contact electrodes and ve dc current sources. In the
estimation of the radii of water invasion, we also contaminated
the simulated data with various amounts of random Gaussian
noise. Figure 9 shows plots of the simulated voltages plus ad-
ditive noise. Noise was synthesized numerically with a zero-
mean Gaussian random number generator of standard devia-
tion equal to a given percentage of voltage amplitude. Three
curves are shown in that gure, for noise percentages of 0%,
2%, and 5%, respectively. The data shown correspond to cross-
well measurements simulated for a dc current source located
at a depth of 45 m, and the underlying resistivity model is the
one described in Figure 2.
Figure10shows thevertical proles of invasionradii inverted
for the resistivity model of Figure 2. Four panels of results are
shown in that gure, with the true invasion prole plotted
for reference with a solid line and the inverted prole with
a dashed line. The upper left panel shows the initial prole
FIG. 8. Detail of the 2-D nite-difference grid used to solve
the approximate forward problem associated with the auxilary
least-squares cost functions. The complete grid consists of 27
nodes in the vertical direction and 26 in the radial direction,
with outer boundaries extending to 6 kmin the radial direction
and 6 km in the vertical direction (compare with Figure 7).
used to initialize the inversion (in this case uniform), and the
second, third, and fourth panels show the vertical prole of
invasion radii inverted from single-well, crosswell, and com-
bined (single- plus crosswell) dc resistivity data, respectively.
In this case, no noise was added to the simulated data before
performing the inversions, and the data were t to within the
accuracy inherent to the numerical simulations. Notice that
FIG. 9. Plot of the electric potential in volts simulated along the
line of contact electrodes shown in Figure 2 for a dc current
source located at the depth of 45 m in the borehole to the
left of the same gure. The underlying resistivity model also
is described in Figure 2. Curves are shown for three different
noise percentages added to the data. The noise was synthesized
with a zero-mean random Gaussian number generator.
FIG. 10. Plots of the vertical prole of invasion radii inverted
from noise-free dc resistivity data. In all panels, the original
prole is plotted with a solid line and the inverted prole with
a dashed line. Different panels show the proles inverted from
single-well, crosswell, andcombined(single- plus crosswell) re-
sistivity data. The upper left panel displays the prole of inva-
sion radii used to initialize the inversions. Each single- or cross-
well experiment consisted of a set of ve dc current sources and
18 contact electrodes, deployed as described in Figure 2.
the vertical prole of invasion radii inverted from the cross-
well data is slightly closer to the true prole than the prole
inverted from the single-well data. On the other hand, the pro-
le inverted from both single- and crosswell data is superior to
either prole inverted separately. We also observe that in all
cases, the largest invasion prole is the one that is estimated
more accurately. Conversely, the shortest radii of invasion are
the ones reconstructed most poorly. This behavior is consistent
with the distribution of anomalous electric potential shown in
Figure 6, where the largest potential anomaly by far occurs at
and about the boundary of the largest invasion radius. The fact
that even with noise-free data the inverted prole does not per-
fectlymatchtheoriginal proleindicates that thequadratic cost
function does exhibit an extremum; but in the neighborhood
of that extremum, the cost function is extremely at. This ex-
plains why multiple solution possibilities can t the data within
the accuracy of the forward-modeling code.
Figure 11 describes the route to convergence of the nonlin-
ear inversion procedure used to obtain one of the proles of
invasion radii shown in Figure 10. In Figure 11a, the relative
data mist is plotted as a function of the auxiliary least-squares
function dened in the cascade. We computed the relative data
mist with the formula
_
_
W
d
[d(m) d
obs
]
_
_
2
_
_
W
d
d
obs
_
_
2
,
where the data-weighting matrix W
d
is a diagonal matrix with
elements equal to the inverse of the measurement times the
standard deviation of the noise (in the noise-free case, W
d
is
FIG. 11. Plots of data mist versus iteration number in the
search of the extremum of the cost function. (a) is the nor-
malized data mist evaluated with the ne grid shown in Fig-
ure 7, and the abscissas identify the corresponding auxiliary
least-squares cost function. (b) shows the normalized data mis-
t as a function of the iteration number within one of the auxil-
iary cost functions constructed with the coarse grid illustrated
in Figure 8.
set to a diagonal matrix with elements equal to the inverse
of the measurement). By so doing, the mist errors, in effect,
were expressed as relative errors, with the much desirable con-
sequence that the inversions were not biasedstrongly tot only
the largest amplitude variations in the data. Each of the auxil-
iary cost functions is identied with a number that corresponds
to its location in the cascade. Figure 11b is a plot of the data
mist with respect to the iteration number within one of such
auxiliary cost functions. For this case, the auxiliary cost function
was minimized after 17 iterations, despite the abrupt backward
change at iterations 3 and 4. The minimization was performed
as described in the previous section, using the coarse nite-
difference grid shown in Figure 8. We also encountered cases
in which the auxiliary cost function was minimized in as few
as ve iterations, especially when the global data mist shown
in Figure 11b was close to its expected lower bound. Because
the number of data was much larger than the number of un-
knowns, we had no need to add a regularization term to the
auxiliary cost function to obtain a unique solution. In the case
of noisy measurements, we implemented a simple regulariza-
tion technique which consisted in setting the matrix W
m
equal
to a unity diagonal matrix in equation (2). This corresponds
to the well-known Wiener regularization scheme (Treitel and
Lines, 1982). We further set the regularization parameter, ,
to be a small percentage of the ratio between the largest and
smallest eigenvalues of matrix J
T
(m
k
) W
T
d
W
d
J(m
k
) in equa-
tion (3). The value of this percentage was chosen arbitrarily to
be proportional to the estimated noise level.
In Figure 11a, the nonlinear inversion was completed with
only four calls to the forward-modeling code implemented on
the ne grid. The complete cascade inversion was performed in
approximately 17 minutes of CPU time on a Silicon Graphics
Power Challenge machine. This represented nearly a 12-fold
savings in CPU time with respect to a single nonlinear inver-
sion implemented with the original nite-difference grid. In
both experiments, i.e., (1) using a Gauss-Newton minimization
procedure implemented entirely with the ne nite-difference
grid (including the computation of the Jacobian matrix), and
(2) using the cascade inversion with an approximate forward
problem realized with the coarse grid, model results agreed
within less than 1%. Sample plots of data residuals are shown
inFigure 12 for the case of a dc current source locatedat a depth
of 45 m. These residuals were constructed by subtracting the
voltages fed to the inversion as data from the corresponding
voltages simulatednumerically for the invertedproles of inva-
sion radii shown in Figure 10. The panel on the right is a plot of
the data residuals along the contact electrodes opposite to the
dc source, i.e., for the case of a crosswell measurement system.
On the other hand, data residuals along the contact electrodes
within the same borehole, i.e., for a single-well measurement
system, are plotted in the panel on the left. We remark that the
residuals are approximately two orders of magnitude smaller
in amplitude than the input data.
The effect of noisy data on the inversions becomes evident in
the results describedby Figure 13 and14. Inrendering bothsets
of inversions, zero-mean random Gaussian noise was added to
the data, as described graphically in Figure 9, with standard de-
viations equal to 2%and 5%, respectively, of the measurement
amplitude. These inversions show a drastic reduction in the
ability of noisy data to resolve the short invasion radii in com-
parison with the resolution available from noise-free data. For
the most part, only the largest invasion radius remains close to
its true value. Such behavior becomes most evident when only
single-well data are used as data to perform the inversions.
CONCLUSIONS
The algorithm for nonlinear inversion developed in this pa-
per is an attempt to cast standard minimization procedures in
the formof a cascade sequence of simpler, hence more efcient,
nonlinear inversions. We have chosen to construct these simple
FIG. 12. Data mist residuals rendered by the inversions of
invasion radii shown in Figure 10. In this subset, the dc current
source is located at a depth of 45 m within the borehole shown
onthe left inFigure 2. The panel onthe right is a plot of the data
residuals along the line of contact electrodes in the opposite
borehole (crosswell experiment). Data residuals for contact
electrodes within the same borehole (single-well experiment)
are shown in the panel on the left.
from 2% noisy dc resistivity data. Plotting conventions are as
described in Figure 10. Noise added to the measurements was
synthesized with a zero-mean random Gaussian number gen-
erator of standard deviation equal to 2% of the measurement
amplitude (compare with Figures 10 and 14).
from 5% noisy dc resistivity data. Plotting conventions are as
described in Figure 10. Noise added to the measurements was
synthesized with a zero-mean random Gaussian number gen-
erator of standard deviation equal to 5% of the measurement
amplitude (compare with Figures 10 and 13).
nonlinear inversions by using an approximate solution to the
original forwardproblem. Specically, inthis paper we have ex-
plored the possibility of constructing an approximate forward
problem by taking a subset of the nite-difference grid that is
used to simulate the measurements numerically. Other approx-
imate forward problems could be used for the same purpose
as long as they abide by the necessary requirements of con-
vergence. Approximations based on integral-equation formu-
lations such as the ones reported by Habashy et al. (1993) and
Torres-Verd n and Habashy (1994), for instance, could take the
role of the auxiliary forward-modeling operator in the cascade.
However, the dual-grid approach proposed here is attractive be-
cause it allows great exibility to adjust the degree of accuracy
required to achieve a given rate of convergence. We also en-
vision that the same cascade procedure can be formulated to
performthe inversion over a sequence of several subsets of the
original nite-difference grid with different degrees of relative
coarseness. Albeit promising, this possible extension to the
dual-grid approach remains a subject for future research. An
inversion strategy such as that could well nd its best applica-
tion in the formulation of 3-D inverse problems, in which re-
peated access to the forward-modeling code can be extremely
taxing, even in computers with parallel architecture (see, for
instance, Newman and Alumbaugh, 1996).
Throughthe simple examples reportedinthis paper, we have
shown that the dual-grid inversion approach is superior to the
minimization performed solely on the grid designed to repro-
duce the measurements; the difference was in the order of a
10-foldreductioninCPUtime. This observationstemmedfrom
simple dc resistivity experiments involving very few unknowns
and little or no need of regularization. However, several issues
at stake need to be further examined, particularly regarding
the link between grid design and convergence rate of the aux-
iliary nonlinear inversions. The approach taken in this paper
consisted of reducing the number of nodes outside the area of
interest, but there is reason to think that even across the area of
interest, thenite-differencegridcouldbemadecoarser. This is
also a subject for future research that will be related intimately
to the specic nature of the partial differential equation that
governs the measurements.
ACKNOWLEDGMENTS
Metin Karakas of Schlumberger-Doll Research shared with
us many valuable insights on the physics of uid ow in porous
media, particularly in the context of water-injection experi-
ments. We are obliged to Greg Newman, David Alumbaugh,
and Ki Ha Lee for constructive technical and editorial sugges-
tions that have improved the quality of this paper.
REFERENCES
Allers, A., Sezginer, A., and Druskin, V. L., 1994, Solution of
2.5-dimensional problems using the Lanczos decomposition: Radio
Sci., 29, 955963.
Bevc, D., and Morrison, H. F., 1991, Borehole-to-surface electrical re-
sistivity monitoring of a salt water injection experiment: Geophysics,
56, 769777.
Daily, W., and Owen, E., 1991, Cross borehole resistivity tomography:
Geophysics, 56, 12281235.
Daily, W., Ramirez, A., LaBrecque, D., and Nitao, J., 1992, Electrical
resistivity tomography of vadose water movement: Water Resources
Research, 28, 14291442.
Dennis, J. E. Jr., and Schnabel, R. B., 1983, Numerical methods for
unconstrained optimization and nonlinear equations: Prentice-Hall,
Inc.
Druskin, V. L., and Knizhnerman, L. A., 1989, Two polynomial meth-
ods of calculating functions of symmetric matrices: J. Comput. Math.
Math. Phys., 29, 112121.
1995, Krylov subspace approximation of eigenpairs and matrix
functions in exact and computer arithmetic: Num. Linear Algebra
and Appl., 2, 205217.
1998, ExtendedKrylovsubspaces: Approximationof thematrix
square root and related functions: SIAM J. Matrix Analysis Appl.,
19, 755771.
Ellis, R. G., Farquharson, C. G., and Oldenburg, D. W., 1993, Approxi-
mate inverse mapping inversion of the COPROD2 data: J. Geomag.
Geoelectr., 45, 10011012.
Farquharson, C. G., and Oldenburg, D. W., 1995, Approximate sensi-
tivities for the multi-dimensional electromagnetic inverse problem:
International Symposium on Three-Dimensional Electromagnetics,
Schlumberger-Doll Research, Expanded Abstracts, 335345.
Gill, P. E., Murray, W., and Wright, M. H., 1981, Practical optimization:
Academic Press Inc.
Habashy, T. M., Groom, R. W., andSpies, B., 1993, BeyondtheBornand
Rytov approximations: A nonlinear approach to electromagnetic
scattering: J. Geophys. Res., 98-B2, 17591775.
Lawson, C. L., and Hanson, R. J., 1974, Solving least squares problems:
Prentice-Hall, Inc.
Mackie, R. L., and Madden, T. R., 1993, Three-dimensional magne-
totelluric inversion using conjugate gradients: Geophys. J. Internat.,
115, 215229.
Newman, G. A., and Alumbaugh, D. L., 1996, Three-dimensional elec-
tromagnetic modeling and inversion on massively parallel comput-
ers: Research Report SAND960582, Sandia National Laboratories.
Oristaglio, M. L., and Worthington, M. H., 1980, Inversion of surface
and borehole electromagnetic data for two-dimensional electrical
conductivity models: Geophys. Prosp., 24, 633657.
Parlett, B. N., 1980, The symmetric eigenvalue problem: Prentice-Hall,
Inc.
Sasaki, Y., 1992, Resolution of resistivity tomography inferred from
numerical simulation: Geophys. Prosp., 40, 453463.
Scales, L. E., 1985, Introduction to non-linear optimization: Macmillan
Publ. Ltd.
Smith, J. T., and Booker, J. R., 1991, Rapid inversion of two- and three-
dimensional magnetotelluric data: J. Geophys. Res., 96, 39053922.
Tamarchenko, T., 1988, Fast algorithms of electromagnetic forward
modeling in complex geometry: Ph.D. thesis, Moscow Geophysical
Prospecting Institute (in Russian).
Torres-Verd n, C., andHabashy, T. M., 1994, Rapid2.5-Dforwardmod-
eling and inversion via a new nonlinear scattering approximation:
Radio Sci., 29, 10511079.
Treitel, S., and Lines, L. R., 1982, Linear inverse theory and deconvo-
lution: Geophysics, 47, 11531159.
APPENDIX A
FINITE-DIFFERENCE SOLUTION OF THE 2.5-D dc PROBLEMFOR CYLINDRICALLY SYMMETRIC ANISOTROPIC
CONDUCTIVITY MODELS
Let us consider a reference frame with cylindrical coordi-
nates (, z, ) chosen such that the z axis coincides with the
water-injection well described in Figure 2. We further assume
that the electrical-conductivity distribution is axially symmet-
ric about the z axis and therefore can be described solely as a
function of the coordinates (, z). However, we do allow both
dc current sources and contact electrode points to be located
anywhere in space. This causes the resulting electric potential
to be a function of all three coordinates (, z, ). Simulation
problems with these characteristics, i.e., 3-D functional behav-
ior and 2-D model complexity, nevertheless can be solved with
2-D formulations, as we are set to show next.
A standard method of solution entails a decomposition in
terms of azimuthal (cylindrical) harmonics. Each one of these
azimuthal harmonics is solved via a 2-D simulation problem,
and the results for all harmonics are superimposed to generate
the complete solution. However, because of the special prop-
erties of the potential eld arising in connection with crosswell
studies posed in a cylindrical coordinate frame, a harmonic de-
composition of this nature may require computing the solution
of an excessive number of harmonics (as many as 8,000 har-
monics to obtain a solution in which the current source and the
contact electrode have the same z and coordinates but differ-
ent coordinates). To circumvent this problem, we resort to an
alternative approach which is numerically more efcient than
harmonic decomposition and is based on the spectral Lanczos
decomposition method (SLDM). This approach has the advan-
tage of implicitly rendering 2.5-D simulation problems into a
single simulation of a 2-D problem.
Let U(, z, ) designate the dc potential resulting from a
point current source ()g(, z) acting in the closed spatial do-
main R
3
, and subject to Dirichlet boundary conditions
along . The conductivity function here is considered to have
the characteristics of an axially symmetric diagonal tensor,
i.e., = (, z) = di ag{
(, z),
z
(, z),
(, z)}, thereby
lending itself to the study of an important class of conductiv-
ity anisotropy. The differential equation satised by U(, z, )
thus is given by
[(, z, ) U(, z, )] = ()g(, z) (A-1)
or, in cylindrical coordinates,
1
_
+

z
_
z
U
z
_
+

2
U
2
= ()g(, z).
(A-2)
By making the change of variable
u(, z, ) = U(, z, )
_
, (A-3)
equation (A-2) can be written as
A[u] +

2
u
2
= ()(, z) 0 2, (A-4)
where Ais a differential operator which only considers varia-
tions of u with respect to and z, given by
A[u(, z, )] =
_

u
_
+
2
_
1
z
_
z

z
_
1
u
_
and
(, z) =
_
g(, z).
It can be shown easily that A[u] is symmetric and negative
denite.
To solve for u(, z, ), we approximate the operator A
using a ve-point second-order nite-difference stencil on a
2-D grid consisting of n nodes. Let matrix Adenote the nite-
difference discretization of the differential operator A. Equa-
tion (A-4) thus becomes an ordinary linear differential equa-
tion for the unknown nth dimensional vector u() constructed
as the approximation of u(, z, ) at the n nodes of the 2-D
grid. By taking into account the driving term on the right-hand
side of equation (A-4), as well as the azimuthal symmetry of
u(, z, ), i.e.,
u() = u(2 ),
oneobtains thefollowingmatrixfunctional representationof u:
u() = 0.5
_
e
A
+e
(2+)
A
_
_
I e
2
A
_
_
1
0 , (A-5)
where is the nth dimensional representation of function
(, z) over the same 2-D grid. The computation of u() via
equation (A-5) requires the evaluation of matrix
Aand of
its subsequent exponentiation e
A
for every possible value
of . This computation can be performed in principle if one
solves the eigenvalue problem for matrix A. After solving
this eigenvalue problem, one could solve for vector u() for as
many values of as needed without signicant additional com-
putations. We also remark that having solved the eigenvalue
problem for matrix Ain principle would allow one to solve for
vector u() in response to several values of vector without
an appreciable increase in computer operations. Each vector
would correspond to a specic electric-source location in
the (, z) plane. If the electric-source location varies only with
respect to the coordinate, one would not have to modify the
vector as long as the electric source remains a point source
(or a nite set of source points).
Althoughobtaining a solutionof vector u() via a solutionof
the eigenvalue problem of matrix A provides valuable insight
to the problem, a numerical solution implemented in this way
would be impractical because of the often large size of matrix
A. To solve this problem, we make use instead of the SLDM
as described by Druskin and Knizhnerman (1995). With the
SLDM, vector u() in equation (A-5) can be written explicitly
as
u() u
m
() = 0.5Q
_
e
H
+e
(2+)
H
_
H
_
I e
2
H
_
_
1
e
1
, (A-6)
where Q and H are n m and m m matrices, respectively,
obtained after performing m steps of the Lanczos recurrence
process on matrix A and vector ; vector e
1
is the rst unit
vector in R
m
. Under the Lanczos recurrence process (Parlett,
1980), matrix H is symmetric and tridiagonal and is obtained
from a Gram-Schmidt orthogonalization process of A. Matrix
Q, on the other hand, will lose orthogonality because of round-
off errors when the Lanczos recursions are performed in nite
precision, but a landmark theorem by Druskin and Knizhner-
man (1995) shows that this phenomenon is not detrimental to
convergence.
From a computational viewpoint, the advantage of equa-
tion (A-6) over equation (A-5) is that although matrix A is
sparse, matrix H is only tridiagonal, and therefore solving
the eigenvalue problem of H is substantially more efcient
than solving the eigenvalue problem of A. In turn, at a given
m-step of the Lanczos process, solving the eigenvalue problem
for matrix H enables one to obtain a solution for u
m
(). Yet
in like manner with the eigenvalue problem of matrix A, one
of the the most important features of the SLDM solution is
that once the eigenvalue problem for matrix H is solved, re-
sults for additional values of can be obtained with practically
no overhead in computer efciency. It is this very feature of
equation (A-6) that makes SLDM superior in principle to any
alternative iterative matrix solver such as conjugate gradient,
for instance. However, because matrices Q and H are derived
from the repeated projections of vector onto matrix A, one
has to recompute both these matrices every time one changes
the source vector , i.e., every time one changes the (, z) loca-
tion of the electric source. Although this would not be the case
withanumerical solutionobtainedviathesolutionof theeigen-
value problemof matrix A, the SLDMsolutionstill canprovide
a signicant edge in computer efciency when a solution is re-
quired for a large collection of source vectors . Tamarchenko
(1988) and Allers et al. (1994) have developed a similar solu-
tion of equation (A-1) in Cartesian coordinates. In cylindrical
coordinates, however, the matrix operator A can be severely
ill conditioned because of the essential singularity at = 0.
Of course, one would expect that convergence of u
m
() to-
ward u() could be achieved for m < n. Otherwise, the SLDM
solution would run the risk of becoming as inefcient as a di-
rect numerical solution based on the solution of the eigenvalue
problem of matrix A. The convergence properties of SLDM,
whenoperatedoncertainmatrixfunctionals, havebeenstudied
at length by Druskin and Knizhnerman (1989). In their stud-
ies, the number of steps m required to achieve convergence
are dictated mainly by the condition number of matrix A as
well as by the matrix functional at stake. In an effort to im-
prove the convergence properties of certain matrix function-
als such as the one given by

A and which appears in equa-
tion (A-5), Druskin and Knizhnerman (1998) have advanced a
novel recurrence scheme which, in similar manner to the Lanc-
zos method, is based on a Krylov subspace representation of
the matrix vector pair (A, ), but which considers in addition
the extended Krylov subspace spanned by the pair (A
1
).
Druskin and Knizhnerman (1996) have coined the term ex-
tendedKrylovsubspace method(EKSM) todesignatetheir new
recurrence procedure. This extension of the standard Krylov
subspace spanned by (A, ) in principle dramatically improves
the convergence properties of SLDM, even considering the ex-
traexpenses incurredbythepreliminarycomputationof A
1
(via LU factorization and back substitution, for instance).

The FORTRAN code used to perform the numerical simu-
lations reported in this paper is based on an implementation
of the EKSM. Typical simulation errors over a 41 49 grid are
not higher than 1%. A single-source simulation with this size
grid has been clocked at 10 seconds of CPU time in a Silicon
Graphics Power Challenge machine.
APPENDIX B
CONVERGENCE OF THE FIXED-POINT ITERATION
In this appendix, we establish the necessary condition for the
recurrence
m
k+1
= h
1
[h( m
k
) d( m
k
) +d
obs
]
(equation 7) to be convergent. In the Cauchy sense, local con-
vergence is assured whenever
m
k+1
m
k
m
k
m
k1
. (B-1)
From equation (7), we obtain
m
k+1
m
k
= h
1
[h( m
k
) d( m
k
) +d
obs
]
h
1
[h( m
k1
) d( m
k1
) +d
obs
]. (B-2)
When m
k+1
is close to convergence, we can approximate this
last equality as
m
k+1
m
k
h
1
(

d
k
) {[h( m
k
) h( m
k1
)]
[d( m
k
) d( m
k1
)]}, (B-3)
where
d
k
= h( m
k
) d( m
k
) +d
obs
and the operator denotes the Jacobian matrix (Fr echet
derivative) of the corresponding vector function. Further sub-
stitution of
h( m
k
) h( m
k1
) h( m
k
) [ m
k
m
k1
]
and
d( m
k
) d( m
k1
) d( m
k
) [ m
k
m
k1
]
yields
m
k+1
m
k
h
1
(

d
k
) [h( m
k
)d( m
k
)] [ m
k
m
k1
].
(B-4)
Comparison of equations (B-1) and (B-4) then shows that for
the recurrence of equation (7) to converge, the condition
h
1
(

d
k
) [h( m
k
) d( m
k
)] 1
is a necessary requirement.
APPENDIX C
BROYDENS UPDATE FORMULA
Given the relatively small number of unknowns in the in-
verse problem at hand, we implemented a rst-order forward-
difference formula to approximate the entries of the Jacobian
matrix associated with a given auxiliary least-squares mini-
mization, i.e.,
h
j
m
h
j
(m
+m
) h
j
(m
)
m
,
where m
was chosen as small as possible but without com-

promising the intrinsic accuracy of the forward-modeling code
(1%). These entries subsequently were updated as the values
m
changed and the iterations progressed in their way toward

the extremum of the auxiliary cost functions. The above for-
mula requires two forward-modeling runs per unknown. In
addition, we experimented with the fact that the Jacobian ma-
trix as a whole couldbe updatedina simple way fromthe values
taken by the same matrix from previous iterations. For this, we
resorted to Broydens rank-one update formula (Dennis and
Schnabel, 1983; Scales, 1985), i.e.,
J(m
k+1
) J(m
k
) +[h(m
k
) J(m
k
) m
k
]
m
k
T
m
kT
m
k
, (C-1)
where
h(m
k
) = h(m
k+1
) h(m
k
),
m
k
= m
k+1
m
k
,
and as usual, the superscript k is used to denote iteration
count. The derivation of this last equation is based on the as-
sumption that h(m
k
) changes in linear fashion with respect
to m
k
along the step direction m
k
. Often, this remains a
good approximation only if m
k
is close to the extremum of
the least- squares cost function. We have found that Broydens
update formula can be used efciently in conjunction with the
grid subset designed to construct a solution of the auxiliary
forward-modeling operator h(m). On occasion, we also have
used Broydens formula throughout a few iterations. After ob-
serving no signicant convergence toward the minimum, we
then proceeded to reset the Jacobian matrix by performing
once again the numerical simulations required in computing its
entries. This combinationof procedures provedtobeextremely
efcient.
GEOPHYSICS, VOL. 65, NO. 6 (NOVEMBER-DECEMBER 2000); P. 17461757, 16 FIGS., 1 TABLE.
Quasi-analytical approximations and series
in electromagnetic modeling
Michael S. Zhdanov
, Vladimir I. Dmitriev
,
Sheng Fang
, and G abor Hurs an
ABSTRACT
The quasi-linear approximation for electromagnetic
forward modeling is based on the assumption that the
anomalous electrical eld within an inhomogeneous do-
main is linearly proportional to the background (nor-
mal) eld through an electrical reectivity tensor

. In
the original formulation of the quasi-linear approxima-
tion,

was determined by solving a minimization prob-
lem based on an integral equation for the scattering cur-
rents. This approach is much less time-consuming than
the full integral equation method; however, it still re-
quires solution of the corresponding system of linear
equations. In this paper, we present a new approach to
the approximate solution of the integral equation us-
ing

through construction of quasi-analytical expres-
sions for the anomalous electromagnetic eld for 3-D
and 2-D models. Quasi-analytical solutions reduce dra-
matically the computational effort related to forward
electromagnetic modeling of inhomogeneous geoelec-
trical structures. In the last sections of this paper, we
extend the quasi-analytical method using iterations and
develop higher order approximations resulting in quasi-
analytical series whichprovide improvedaccuracy. Com-
putation of these series is based on repetitive applica-
tion of the given integral contraction operator, which
insures rapid convergence to the correct result. Numer-
ical studies demonstrate that quasi-analytical series can
be treated as a newpowerful method of fast but rigorous
forward modeling solution.
INTRODUCTION
Theintegral equation(IE) methodis apowerful tool for elec-
tromagnetic numerical modeling (Hohmann, 1975; Weidelt,
Manuscript received by the Editor July 28, 1998; revised manuscript received March 9, 2000.
University of Utah, Department of Geology and Geophysics, Salt Lake City, Utah 84112-0111. E-mail: mzhdanov@mines,utah.edu.
Moscow State University, Faculty of Computational Math and Cybernetics, Moscow 1169899, Russia.
Formerly University of Utah, Department of Geology and Geophysics, Salt Lake City, Utah; presently Baker Atlas, 10205 Westheimer, Houston,
Texas, 77042. E-mail: sheng.fang@waii.com.
1975; Dmitriev and Pozdnyakova, 1992). This method is based
on expressing the electromagnetic elds in terms of an inte-
gral equation with respect to the excess current within an in-
homogeneity. The integral equation is written as a system of
linear algebraic equations by approximating the excess cur-
rent distribution j
a
by the piecewise constant functions. The
resulting algebraic system is solved numerically (Xiong, 1992).
The main difculty of this technique is the size of the linear
system of equations matrix, which demands excessive com-
puter memory and calculation time to invert. This limita-
tion of the integral equation technique becomes critical in in-
verse problem solution which requires multiple forward mod-
eling calculations for different (updated) geoelectrical model
parameters.
A novel approach to 3-D electromagnetic (EM) modeling
based on linearization of the integral equations for scattered
EM elds has been developed recently by Zhdanov and Fang
(1996a, b, 1997). Within this method, called quasi-linear (QL)
approximation, the excess currents are assumed to be propor-
tional to the background (normal) eld E
b
through an electri-
cal reectivity tensor

. In the original papers on QL approx-
imations, the electrical reectivity tensor was determined by
solving a minimization problem based on an integral equation
for the scattering currents (Zhdanov and Fang, 1996a, b). This
problem is much less time-consuming than the full IE method;
however, it still requires solution of the corresponding system
of linear equations. In this paper, we present a newapproach to
estimating

, whichleads toconstructing quasi-analytical (QA)
expressions for the anomalous electromagnetic eld for 3-D
and 2-D models. We demonstrate also the connection between
the QL and QA approximations and the localized nonlinear
(LN) approximations introduced by Habashy et al. (1993) and
Torres-Verdin and Habashy (1994) and conduct a comparative
study of the accuracy of different approximations.
In the last sections of the paper, we extend the quasi-
analytical method using iterative techniques and develop
1746
Quasi-Analytical Approximations 1747
approximations of the higher orders whichprovide better accu-
racy than the original QA and LN approximations. Combina-
tionof theseiterativesolutions forms thequasi-analytical series
which generate a rigorous solution of EM modeling problems.
TENSOR QUASI-LINEAR EQUATION
Consider a 3-Dgeoelectric model with the background (nor-
mal) complex conductivity
b
and local inhomogeneity D
with the arbitrary spatial variations of complex conductivity
=
b
+ . We assume that =
0
=4 10
7
H/m, where
0
is thefree-spacemagnetic permeability. Themodel is excited
by an electromagnetic eld generated by an arbitrary source
time harmonic as e
i t
. Complex conductivity includes the ef-
fect of displacement currents: = i , where and are
electrical conductivity and dielectric permittivity. The electro-
magnetic elds in this model can be expressed as a sum of the
background (normal) and anomalous elds:
E = E
b
+E
a
, H = H
b
+H
a
, (1)
where the background eld is a eld generated by the given
sources in the model with the background distribution of con-
ductivity
b
, andtheanomalous eldis producedbytheanoma-
lous conductivity distribution .
It is well known that the anomalous eld can be presented as
an integral over the excess currents in inhomogeneous domain
D (Hohmann, 1975; Weidelt, 1975):
E
a
(r
j
) =
___
D
G
E
(r
j
| r)j
a
(r) dv = G
E
(j
a
),
(2)
H
a
(r
j
) =
___
D
G
H
(r
j
| r)j
a
(r) dv = G
H
(j
a
),
where

G
E
(r
j
| r) and

G
H
(r
j
| r) are, respectively, the electric and
magnetic Greens tensors dened for an unbounded conduc-
tive medium with the background conductivity
b
, G
E
and G
H
are the corresponding Greens linear operators, and excess cur-
rent j
a
is determined by the equation
j
a
= E = (E
b
+E
a
). (3)
Using Greens operators, one can calculate the electromag-
netic eld at any point r
j
, if the electric eld is known within
the inhomogeneity:
E(r
j
) = G
E
( E) +E
b
(r
j
), (4)
H(r
j
) = G
H
( E) +H
b
(r
j
). (5)
Expression (4) becomes the integral equation with respect to
electric eld E(r), if r
j
D.
The QL approximation is based on the assumption that the
anomalous eld E
a
inside the inhomogeneous domain is lin-
early proportional to the background eld E
b
through some
tensor

(Zhdanov and Fang, 1996a):
E
a
(r)

(r)E
b
(r). (6)
Substituting formula (6) into formula (4), we obtain the QL
approximation E
a
q1
(r) for the anomalous eld:
E
a
q1
(r
j
) = G
E
( (
I +
(r))E
b
). (7)
Rewriting expression (7) gives the tensor quasi-linear (TQL)
equation with respect to the electrical reectivity tensor

:
(r
j
)E
b
(r
j
) = G
E
[
(r)E
b
] +E
B
(r
j
), (8)
where E
B
(r
j
) denotes the Born approximation:
E
B
(r
j
) = G
E
( E
b
) =
_
D
G
E
(r
j
| r) (r)E
b
(r) dv,
(9)
and G
E
[
(r)E
b
] is a linear operator of

(r):
G
E
[
(r)E
b
] =
_
D
G
E
(r
j
| r)(r)
(r)E
b
(r) dv.
(10)
The original QL approximation (Zhdanov and Fang, 1996a,
b) was based on the numerical solution of a minimization prob-
lem arising from the TQL equation (8):
_
_
(r
j
)E
b
(r
j
) G
E
[
(r)E
b
] E
B
(r
j
)
_
_
= min . (11)
The advantage of this approach is that, by choosing a ne
enough discretization for a function

(r
j
), one can generate
an accurate solution. The disadvantage, however, is that sim-
ilar to the full IE method, the QL approach still requires so-
lution of the corresponding system of linear equations. In this
paper, we develop a new TQL equation solution that results
in analytical expressions for the electrical reectivity tensor
(r
j
). This technique is, obviously, much faster than the origi-
nal QL approximation. However, it may be less accurate than
the corresponding QL approximation with a ne grid for

(r
j
)
discretization. In other words, there is a trade-off between the
simplicity of the approximate solution and its accuracy.
QUASI-ANALYTICAL SOLUTIONS FOR 3-D
ELECTROMAGNETIC FIELD
In this section we analyze different approximate solutions of
the TQL equation (8). The iterative approach to the rigorous
solution of the TQL equation is outlined in Appendix A.
Solution for a scalar reectivity tensor
In the framework of the quasi-linear approach, we may con-
sider the electrical reectivity tensor selected to be a scalar
(Zhdanov and Fang, 1996a),

=
I, where

I is the unity tensor.
In this case, integral equation (8) can be rewritten as
(r
j
)E
b
(r
j
) = G
E
[ E
b
] +E
B
(r
j
). (12)
Following Habashy et al. (1993) and Torres-Verdin and
Habashy (1994), we note that Greens tensor

G
E
(r
j
| r) is sin-
gular at the point where r
j
=r. Therefore, one can expect that
the dominant contribution to the integral G
E
[ E
b
] in equa-
tion (12) is fromsome vicinity of the point r
j
=r. Assuming that
(r
j
) is slowly varying within domain D, we write
(r
j
)E
b
(r
j
) (r
j
)G
E
[ E
b
] +E
B
(r
j
)
= (r
j
)E
B
(r
j
) +E
B
(r
j
). (13)
As we seek a scalar reectivity coefcient , it is useful to
calculate the dot product of both sides of equation (13) and the
1748 Zhdanov et al.
background electric eld:
(r
j
)E
b
(r
j
) E
b
(r
j
)
= (r
j
)E
B
(r
j
) E
b
(r
j
) +E
B
(r
j
) E
b
(r
j
). (14)
Assuming that
E
b
(r
j
) E
b
(r
j
) = 0, (15)
and dividing equation (14) by the square of the background
eld, we obtain
(r
j
) =
g(r
j
)
1 g(r
j
)
, (16)
where
g(r
j
) =
E
B
(r
j
) E
b
(r
j
)
E
b
(r
j
) E
b
(r
j
)
. (17)
Substituting equation (16) into equation (1), we nd
E(r) = E
a
(r) +E
b
(r) [(r) +1]E
b
(r)
=
1
1 g(r)
E
b
(r). (18)
Therefore from equations (4) and (5), we nally determine
E
a
QA
(r
j
) = E(r
j
) E
b
(r
j
)
=
___
D
G
E
(r
j
| r)
(r)
1 g(r)
E
b
(r) dv, (19)
and
H
a
QA
(r
j
) = H(r
j
) H
b
(r
j
)
=
___
D
G
H
(r
j
| r)
(r)
1 g(r)
E
b
(r) dv. (20)
Formulas (19) and (20) give QA solutions for 3-D electro-
magnetic elds. Note that the only difference between the new
QAapproximationandthe Bornapproximation(9) is the pres-
ence of the scalar function [1g(r)]
1
. Hence computationally,
the QA approximation and the Born approximation are prac-
tically the same. On the other hand, we show below that the
QA approximation is more accurate than the Born approx-
imation.
General solution for different polarizations of the anomalous
and background electric elds
The QA solutions developed in the previous section were
based on the assumption that the electrical reectivity was
a scalar. This assumption reduces the areas of practical ap-
plication of the QA approximations because in this case the
anomalous (scattered) eld is polarized in direction parallel to
the background eld within the inhomogeneity. However, in
general, the anomalous eld can be polarized in a different di-
rection than the background eld. To overcome this difculty,
we introduce a tensor quasi-analytical (TQA) approximation
to

, which permits different polarizations for the background
and anomalous (scattered) elds.
We again assume that the product

(r)E
b
(r) is a smoothly
varying function of the coordinates, and it can be taken outside
the integral over the anomalous domain Dwithout substantial
discrepancies. As aresult, weobtainfromtheTQLequation(8)
(r
j
)E
b
(r
j
) G
E
[
I]
(r
j
)E
b
(r
j
) +E
B
(r
j
)
= g(r
j
)
(r
j
)E
b
(r
j
) +E
B
(r
j
),
or
[
I g(r
j
)]
(r
j
)E
b
(r
j
) = E
B
(r
j
), (21)
where
g(r
j
) = G
E
[ (r)
I].
Solving equation (21) yields
(r
j
)E
b
(r
j
) = [
I g(r
j
)]
1
E
B
(r
j
). (22)
Substituting equation (22) into equation (1), we obtain
E(r) = E
a
(r) +E
b
(r) [
(r) +
I]E
b
(r)
= [
I g(r)]
1
E
B
(r) +E
b
(r). (23)
Therefore, from equations (4) and (5) we nd
E
a
TQA
(r
j
) = E(r
j
) E
b
(r
j
) =
___
D
G
E
(r
j
| r) (r)
{[
I g(r)]
1
E
B
(r) +E
b
(r)} dv, (24)
and
H
a
TQA
(r
j
) = H(r
j
) H
b
(r
j
) =
___
D
G
H
(r
j
| r) (r)
{[
I g(r)]
1
E
B
(r) +E
b
(r)} dv, (25)
where
g(r
j
) = G
E
[ (r)
I] =
___
D
G
E
(r
j
| r) (r) dv.
(26)
We call expressions (24) and (25) TQA approximations
for an electromagnetic eld. We show below that this ap-
proximation provides a more accurate solution for a forward
problem than a scalar QA approximation. However, we must
compute the tensor multiplier [
I g(r)]
1
, which is slightly
more time-consuming than calculation of the scalar coefcient
[1 g(r)]
1
.
Quasi-analytical solutions for a 2-D electromagnetic eld
Assume now that both the electromagnetic eld and the
complex conductivity in the geoelectrical model are two-
dimensional (i.e., they vary only along the directions x and z of
some Cartesian systemof coordinates, and are constant in the y
direction). In this case, repeating derivations described above
for the 3-D case, we can obtain the following QA expressions
for a 2-D electromagnetic eld:
E
a
QAy
(r
j
) i
0
__
D
G
b
(r
j
| r)
(r)
1 g(r)
E
b
y
(r) ds, (27)
and similarly for magnetic eld components:
H
a
QAx
(r
j
)
__
D
G
b
(r
j
| r)
z
(r)
1 g(r)
E
b
y
(r) ds, (28)
H
a
QAz
(r
j
)
__
D
G
b
(r
j
| r)
x
(r)
1 g(r)
E
b
y
(r) ds, (29)
where G
b
(r
j
| r) is a 2-D scalar Greens function for an un-
bounded conductive medium with the background conductiv-
ity
b
, and
g(r) =
E
B
y
(r)
E
b
y
(r)
. (30)
These formulas can serve as a neweffective QAtool for both
direct and inverse 2-D electromagnetic problem solutions. Nu-
merical tests demonstrate that these approximations produce
a very accurate result for 2-D models (Dmitriev et al., 1998).
LOCALIZED NONLINEAR APPROXIMATION
The TQA approximation can be treated as a generalization
of the LN approximation introduced by Habashy et al. (1993).
Let us rewrite equation (23) in the form
E(r) = [
I g(r)]
1
[E
B
(r) g(r)E
b
(r)]
+[
I g(r)]
1
E
b
(r). (31)
Taking into account once again that Greens tensor

G
E
(r
j
| r)
exhibits either singularity or a peak at the point where r
j
=r,
one can calculate the Born approximation G
E
[ (r)E
b
(r)] us-
ing the formula
E
B
(r
j
) = G
E
[ (r)E
b
(r)] g(r
j
)E
b
(r
j
).
This result implies
E
B
(r) g(r)E
b
(r) 0, (32)
and is particularly appropriate if the background eld is a
smoothly varying spatial function (Habashy et al., 1993).
Under assumption (32), equation (31) can be rewritten
E(r) = E
a
(r) +E
b
(r) [
I g(r)]
1
E
b
(r). (33)
Therefore, from equations (4) and (5) we nd
E
a
LN
(r
j
) = E(r
j
) E
b
(r
j
)
=
___
D
G
E
(r
j
| r) (r)[
I g(r)]
1
E
b
(r) dv,
(34)
and
H
a
LN
(r
j
) = H(r
j
) H
b
(r
j
)
=
___
D
G
H
(r
j
| r) (r)[
I g(r)]
1
E
b
(r) dv.
(35)
Formulas (34) and (35) express the LNapproximation intro-
duced by Habashy et al. (1993), where
[
I g(r)]
1
=

(r)
is their depolarization tensor.
Thus we can see that the difference between the TQL ap-
proximation and the LN approximation is determined by a
term
E
a
TQA
(r
j
) E
a
LN
(r
j
)
=
___
D
G
E
(r
j
| r) (r)

(r)[E
B
(r) g(r)E
b
(r)] dv.
(36)
Note that both TQA and LN approximations use the same
depolarization tensor

(r), based on the idea of a localized ef-
fect in the Greens integral operator. The only difference is that
in the case of the LN approximation we use this localization
property twice for computing both the depolarization tensor
and the expression for the Born approximation E
B
(r
j
) on the
right-hand side of the TQL equation (8). In the case of TQA
approximation, we use the exact formula for E
B
(r
j
), and we
consider TQA a partially localized approximation. This differ-
ence does affect the accuracy of these two approximations for
different geoelectrical models, illustrated below by numerical
examples.
COMPARATIVE ACCURACY STUDY
To compare the accuracy of the Born, QA, TQA, and LNap-
proximations, we conducted several numerical experiments for
the models presented in Figure 1. Model 1 consists of a conduc-
tive rectangular prismembedded in a homogeneous half-space
excited by a horizontal rectangular loop (Figure 1, top panel).
FIG. 1. 3-Dgeoelectrical models usedfor comparative accuracy
study of different approximations.
1750 Zhdanov et al.
The frequency is 1000 Hz, and the conductivity ratio is 10. The
receivers are located above the body along the y-axis. Model 2
consists of a conductive cube with a resistivity of 1 ohm-m lo-
cated at a depth of 10 m within a homogeneous half-space with
a resistivity of 10 ohm-m (Figure 1, middle panel). The sides
of the cubic prism have a length of 50 m. Model 3 contains the
body with a horizontal size of 100 m100 m, a vertical dimen-
sion of 50 m, and located at a depth of 10 m. The EM eld in
models 2 and 3 is excited by a vertical magnetic dipole located
on the surface of the earth.
Figure 2 shows the real and imaginary parts of the horizon-
tal electric and vertical magnetic components of the scattered
eld computed by solving the full integral equation and the
approximate solutions. The deviations of QA, TQA, and LN
approximations from the true solution are minor, but the
Born approximation fails. The next gure, Figure 3, presents
the same approximate solutions but for an expanded vertical
scale. We cansee nowthe small differences betweenthe various
approximations.
To analyze more carefully the discrepancies in different ap-
proximations, we consider model 2 presented in Figure 1 (mid-
dle panel). In this experiment the transmitter (Tx) and receiver
(Rx) geometry was xed (transmitter-receiver separation was
100 m), with proles run over the conductive prismatic body
withthe center at a depthof 35 mbelowthe originof x and y co-
FIG. 2. Behavior of the anomalous electromagnetic eld com-
ponents computedfor model 1 by solving the full integral equa-
tion, Born approximation, and the scalar QA, TQA, and LN
approximations.
ordinates. For each position of the Tx/Rx system, we computed
the vertical component of the magnetic eld using the rigorous
full IEmethod and three different approximations: (1) the QA
approximation, (2) the LN approximation, and (3) the TQA
approximation. The relative errors of approximate solutions in
comparison with the rigorous solution were calculated as
=
H
appr
H
full
2
H
full
2
100%. (37)
The main goal of this experiment is to demonstrate that the
accuracy of approximation is not only a function of the con-
ductivity contrast, frequency, and size of the anomalous body,
it also depends on the relative locations of the transmitter, re-
ceiver and conductive target.
Figures 4, 5, and 6 are maps of relative errors for QA, LN,
and TQA approximations correspondingly at the receiver for
different positions of the recording system relative to the body
center at a depth of 35 m below the earths surface. Figure 7
shows the proles of errors along the line connecting transmit-
ter and receiver. One can see that for all three approximations
the errors increase when the body is located just under the
transmitter or the receiver. However, the largest discrepancies
occur when the transmitter is near the inhomogeneity, with sig-
nicantly lower (27%) discrepancies for all other locations.
The most accurate result is delivered by TQA approximation
(dotted line in Figure 7). The high level of discrepancies for
FIG. 3. Behavior of the anomalous electromagnetic eld com-
ponents computedfor model 1 by solving the full integral equa-
tion, the scalar QA, TQA, and LN approximations.
QA approximation near the transmitter can be explained by
the fact that the primary electric eld is equal to zero on the
vertical axes passing through the position of the transmitter
dipole. Since the expression for scalar coefcient g(r) in for-
mulas (17) for the QA approximation becomes singular when
E
b
(r) 0, there is a signicant increase of discrepancies in the
near zone below the transmitter.
Another cause of discrepancies in this zone is the fact that
QA approximation is based on a scalar reectivity tensor. In
the area below the transmitter, the primary electric eld forms
a smoke ring blown by the transmitting loop into the earth
(Nabighian, 1979). The conductive body located just belowthe
transmitter distorts this eld through a secondary electric eld,
directed at some angle with the primary eld. The scalar re-
ectivity tensor cannot account for this rotation of the sec-
FIG. 4. Map of the relative errors between the scalar QA ap-
proximation and the full integral equation solution computed
for model 2.
FIG. 5. Map of the relative errors between the LN approxi-
mation and the full integral equation solution computed for
model 2.
ondary elds which generates additional discrepancies in the
approximation. The plots in Figures 47 show that TQA and
LN approximations handle this polarization pretty well. An
especially accurate result is reached by a TQA solutions. The
corresponding discrepancies do not exceed 7%and 15%in the
areas under the receiver and transmitter, respectively.
An increase in discrepancies generated by LN approxima-
tion in comparison with TQA approximation can be explained
bythefact that LNapproximationis sourceindependent. When
the receiver is closer to the transmitter, the source effect be-
comes more signicant, which leads to an increase in discrep-
ancies. The TQAsolution approximates the polarization of the
secondary eld and takes into account the source position. As a
result, it produces a more accurate approximation, so the TQA
approximation is source dependent.
FIG. 6. Map of the relative errors between the tensor QA ap-
proximation and the full integral equation solution computed
for model 2.
FIG. 7. The relative errors of the scalar QA approximation,
TQA approximation, and LN approximations computed for
model 2.
1752 Zhdanov et al.
We analyzed also the effect of the conductivity contrast on
the accuracy of different approximations. In Figure 8, we ex-
amine the relative errors for model 3 shown in Figure 1, bottom
panel. The horizontal size of the conductive body in this model
(100 100 m) is bigger than in model 2. The center of the con-
ductive body is located just below the receiver. We consider
now that the ratio of anomalous conductivity to background
conductivity varies within a range of ve orders of magnitude.
One can see in the plot in Figure 8 that within a conductivity ra-
tio range from10
2
to 30, the best accuracy is provided by TQA
approximations with the discrepancies less than 10%. For the
higher conductivity ratio, the accuracy of all approximations
becomes less than 2030% .
Figures 9 and10 illustrate the effect of frequency onthe accu-
racy of approximation for model 3. The conductive rectangular
prism has a resistivity of 1 ohm-m while the background resis-
FIG. 8. The relative errors of different approximations as a
function of the conductivity ratio computed for model 3.
FIG. 9. The ratio of different approximations to the full integral
equation solution of the scattered H
z
as a function of frequency
computed for model 3 with a conductivity of 10.
tivity of a half-space is 10 ohm-m. The frequency range is from
10
1
up to 10
4
Hz. Figure 9 shows the ratio of the estimated
to true (computed by IE method) amplitudes of the anoma-
lous eld H
z
. Figure 10 presents the difference between the
phases of the anomalous H
z
component computed by true and
approximate solutions.
The most signicant feature of all these plots is the stable
behavior of the QA approximation. The ratio of the approxi-
mated and true amplitudes of the anomalous eld H
z
is equal
to one within the entire frequency range. The phase difference
is also close to zero along the entire horizontal axis of the plot
in Figure 10. At the same time both TQA and LN approxi-
mations produce good amplitude estimate till the frequency of
1 kHz, andgoodphases only for frequencies below100 Hz. This
similarity in the TQA and LN data behavior can be explained
by the fact that both approximations are based on a depolar-
ization tensor calculation, which is independent of the back-
ground eld. Therefore, these approximations cannot take into
account properly the background eld which cause discrepan-
cies inthese approximations. At the same time, Figure 11 shows
that at the frequency range above 100 Hz the induction effects
become strong and the background eld begins to vary signif-
icantly from its static limit. In the case of the QA approxima-
tion, we evaluate more carefully the induction effect because
the background eld is present in the expressions for scalar co-
efcient g in formulas (19) and (20) for QA anomalies. That is
why the QA approximation produces stable results for a wide
frequency range.
QUASI-ANALYTICAL SERIES
The mainlimitationof the QAmethod(as well as QLandLN
approximations) is that it is still an approximate method of 3-D
forward modeling, and its practical application requires addi-
tional control of the approximationdiscrepancies. It is possible,
however, to increase the accuracy of the QAapproximation by
constructing QA approximations of a higher order in a similar
FIG. 10. The difference of the phases between the approximate
and the full integral equation solutions of the scattered H
z
as a
functionof frequencycomputedfor model 3withaconductivity
of 10.
fashion to the QLseries for QLapproximations (Zhdanov and
Fang, 1997).
The QL series were based on a new method of constructing
the converged Born series developed by Singer and Fainberg
(1995) and Pankratov et al. (1995). This method was applied in
Zhdanov and Fang (1997) to construct QL series that almost
always converged. In this paper, we use the same method to
generate QA series and calculate the accuracy of the QA ap-
proximations. These series are built on the QA approximation
as the rst term of the series. As a result, the computation of
QA series becomes even easier and faster than in the case of
QL series, which required the solution of the linear algebraic
equation in the rst step of the iterations. Computation of the
QA series does not involve any system of equation solution. It
is based on repetitive application of the given integral contrac-
tion operator, which insures rapid convergence to the correct
result.
Following Zhdanov and Fang, (1997) we modify Greens op-
erator according to the formula
G
m
( (r)E
b
(r))
=
_
Re
b
G
E
(2
_
Re
b
(r)E
b
(r)) + (r)E
b
(r)
=
_
Re
b
___
D
G
E
(r
j
| r) 2
_
Re
b
(r)E
b
(r) dv
+ (r)E
b
(r). (38)
It was proved in Zhdanov and Fang (1997) that the L
2
norm
of this operator is always less than or equal to one:
G
m
1. (39)
We can rewrite now the integral equation for the anomalous
eld (2) as follows:
aE
a
= C(aE
a
), (40)
where C(aE
a
) is an integral operator of the anomalous eld:
C(aE
a
) = G
m
[aE
a
] +G
m
[aE
b
] aE
b
, (41)
FIG. 11. Mutual coupling ratios of the real and imaginary parts
of the vertical magnetic eld over a homogeneous half-space
of 10 ohm-m as a function of frequency.
and
a =
2Re
b
+
2
Re
b
, =

2Re
b
+
.
The solution of this integral equation can be obtained using
the method of successive iterations, which is governed by the
following equation:
aE
a(N)
= C[aE
a(N1)
], N = 1, 2, 3 . . . (42)
Theseiterations always converge for any lossymediumbecause
C is a contraction operator (Zhdanov and Fang, 1997).
We start iterations with the QA approximation for the
anomalous eld:
E
a(0)
qa
=
g
1 g
E
b
. (43)
In this case, the rst order QA approximation is equal:
aE
a(1)
qa
= C
_
aE
a(0)
qa
_
= G
m
_
aE
a(0)
qa
_
+G
m
[aE
b
] aE
b
. (44)
We will call the rst iteration determined by expression (44) a
modied quasi-analytical approximation (MQA):
E
a(1)
qa
= E
a
MQA
=
1
a
G
m
_
aE
a(0)
qa
_
+
1
a
G
m
[aE
b
] E
b
.
Taking into account the denition of the modied Greens op-
erator (38) and formula (43), we obtain
E
a
MQA
=
2Re
b
2Re
b
+
_
G
E
_
E
a(0)
qa
_
+G
E
_
E
b
__
=
2Re
b
2Re
b
+
G
E
_

1 g
E
b
_
=
2Re
b
2Re
b
+
E
a
MQA
. (45)
Equation (45) shows that the modied QA approximation is
equal to the original QA approximation outside inhomogene-
ity D:
E
a
MQA
(r
j
) = E
a
QA
(r
j
), r
j
/ D, (46)
while they are different inside the geoelectrical inhomogeneity.
The second-order QA approximation is equal to:
aE
a(2)
qa
= C
_
aE
a(1)
qa
_
= (G
m
)
2
_
aE
a(0)
qa
_
+G
m
(aE
Bm
) +aE
Bm
,
where E
Bm
is a modied Born approximation determined by
the formula
E
Bm
=
1
a
G
m
E
b
E
b
=
2Re
b
2Re
b
+
E
B
.
The third-order QA approximation is given by the formula
aE
a(3)
qa
= C
_
aE
a(2)
qa
_
= (G
m
)
3
_
aE
a(0)
qa
_
+(G
m
)
2
(aE
Bm
)
+G
m
(aE
Bm
) +aE
Bm
.
1754 Zhdanov et al.
Finally, the Nth-order QA approximation can be treated as
the sum of N terms of the QA series:
aE
a(N)
qa
=
N1
k=0
(G
m
)
k
(aE
Bm
) +(G
m
)
N
_
aE
a(0)
qa
_
, (47)
where G
m
is the modied Greens operator:
(G
m
)
_
aE
a(0)
qa
_
=
_
Re
b
___
D
G
E
(r
j
| r) 2
_
Re
b
bE
a(0)
qa
dv
+bE
a(0)
qa
=
_
Re
b
___
D
G
E
(r
j
| r) E
a(0)
qa
dv
+

2
Re
b
E
a(0)
qa
(r
j
).
Note, that QA series can be built on TQA and LN approxi-
mations as the rst iterations. We select the approach based on
QA approximation for the sake of simplicity.
ACCURACY ESTIMATION
The accuracy of the QA approximation of the Nth order is
estimated in the same way as in Zhdanov and Fang (1997) by
the formula
N
=
_
_
aE
a
aE
a(N)
qa
_
_
_
_
aE
a(N)
qa
_
_

r
N
, (48)
where E
a(0)
qa
=
ag
1 g
E
b
, and r
N
is the relative convergence rate
of the QA approximations:
r
N
=
_
_
aE
a(N)
qa
aE
a(N1)
qa
_
_
_
_
aE
a(N)
qa
_
_
. (49)
In particular, the accuracy of the original QA approximation
E
a
QA
can be estimated by computing, using the formula
=
_
_
aE
a
aE
a
QA
_
_
_
_
aE
a
QA
_
_

_
_
aE
a
QA
ag
1g
E
b
_
_
_
_
aE
a
QA
_
_
.
(50)
Thus, the accuracy estimation formula for QA solutions is ex-
pressed by the QA solution itself.
Formulas (48) and (50) make it possible to obtain a quan-
titative estimation of the QL approximation accuracy without
direct comparison with the rigorous full IE forward modeling
solution.
NUMERICAL MODELING RESULTS
We developed a computer code based on the QA series for
the electromagnetic eld in a 3-D case. The algorithm was
tested for 3-D geoelectrical models.
Consider a 3-D geoelectrical model (model 4), consisting of
a homogeneous half-space (with resistivity of 100 ohm-m) and
a thin conductive rectangular inclusion with the resistivity of
1 ohm-m (Figure 12). The electromagnetic eld in this model
is excited by a horizontal rectangular loop, located 50 m to the
left of the model, with the loop 10 m on a side and a current of
1 A. We have used the integral equation code for computing
the scattering current in the complex conductivity structure
and the QA series code.
Figure 13 presents maps of the excess electrical currents
distribution within the inhomogeneity obtained by a rigorous
FIG. 12. 3-D geoelectrical model of a thin conductive rectan-
gular body embedded in a homogeneous half-space excited by
a horizontal rectangular loop (model 4).
FIG. 13. Behavior of the scattering currents induced inside the
conductive rectangular body in model 4 obtained by solving
the full integral equation and the approximate solution after
one, 15, and 50 iterations.
integral equation solution and by QAseries of different orders
for a frequency of 1000 Hz. Note that this model is a difcult
one for QA approximation because it contains a conductiv-
ity contrast of 100. Nevertheless, we can see how the currents
computed by QA series converge to the true solution. Fig-
ure 14 presents a map of relative errors in the excess current
calculations between the integral equation technique and the
QA series of the order of 5 and 15. One can observe that the
discrepancies decrease during the iterations.
We appliedthe QAseries solutiontomodel the EMresponse
for themorecomplicatedmodel simulatingtheKambalda-style
nickel sulde deposit in Western Australia (Stolz et al., 1995.).
The sketch of the model is shown in the top panel of Figure 15;
the bottom panel of Figure 15 shows the vertical geoelectrical
cross-section of the model. The model consists of a conduc-
tive overburden above an inclined nickel lens. The conductivity
contrast betweenthe nickel lens andthe host rocks is 10
4
, which
is far beyond the normal limits of QA approximations. Fig-
ure 16 presents the horizontal andvertical anomalous magnetic
elds of forward modeling based on integral equation solution
and QA series of the order of 1, 10, 20, and 50. The very high
conductivity contrast and the proximity of the anomalous body
causes inaccuracies in the approximate solutions. However, the
always convergent series algorithmprovides the correct values.
Table 1 shows a comparison of CPU time for EM modeling
using integral IE (Xiong, 1992) and the QA approximations of
the different orders for the thin sheet and the Kambalda-style
nickel-sulde deposit models. For 1088 cells and 50 iterations
FIG. 14. Maps of the relative residuals of scattering currents
obtained by different orders of QA series with respect to the
full integral equation solution. The number of iterations is 5
(top) and 15 (bottom).
of the QA series, the algorithm spent approximately half the
CPUtime requiredfor the solutionof the full integral equation.
For 4352 cells, the time gain is very signicant. It takes about
2.5 hours for the newcode to run 50 iterations, which generates
the same solution as the integral equation code (Figure 16). It
took more than ve days to reach the same result using full IE
method.
CONCLUSION
We have generalized the QL method of forward modeling
and developed a new approach to calculation of the electrical
Table 1. Comparison of the CPU time (in seconds) for EM
modeling using full integral equation solution and QA ap-
proximations of the different orders for the thin sheet and the
Kambalda-style nickel-sulde deposit models.
Cells Full IE 1st-order 20th-order 50th-order
Cells Full IE QA QA QA
100 30 11 15 21
1088 1264 221 368 589
2176 21 687 459 1326 2967
4352 490 752 911 3931 8827
FIG. 15. Kambalda-style ore deposit model with an inclined
dike structure. The bottom panel shows the vertical resistivity
cross-section.
1756 Zhdanov et al.
FIG. 16. Magnetic eldcomponents obtainedby different num-
bers of iterations of QA series and full integral equation solu-
tions over the Kambalda-style ore deposit model.
reectivity tensor based on the solution of the corresponding
integral equation.
Based on this approach, we introduced the new scalar (QA)
and tensor (TQA) quasi-analytical solutions for electromag-
netic elds in 3-D inhomogeneous media. We demonstrated
also that TQA approximation is a generalization of the local-
ized nonlinear approximation introduced by Habashy et al.
(1993). The new TQA approximation permits different polar-
izations for the background and anomalous (scattered) eld,
which increases the accuracy of approximation for conductiv-
ity contrasts below 30. The comparative accuracy study of the
different approximations demonstrates that TQA approxima-
tion has a superb accuracy, especially in the areas close to the
transmitter and to the anomalous geoelectrical structures. At
the same time, QA approximation generates a stable and ac-
curate result (discrepancies below 3%) for a wide frequency
range (from 10
1
up to 10
4
Hz).
The computational time for QA approximations is compa-
rable with that required for the Born approximation, although
the new approximate solutions are much more accurate. To
generate a rigorous forward-modeling result, we may apply
these approximations iteratively. This approach leads us to a
construction of the QA series.
The developed approximations of the electromagnetic eld
can be used as effective tools for fast 3-D forward modeling.
One of the attractive areas of their application is rapid comput-
ing of the Fr echet derivative for 3-Delectromagnetic inversion.
We improved the accuracy of the QAapproximation by con-
structing QAapproximations of a higher order in a similar way
as has been done for QL approximations in Zhdanov and Fang
(1997). These series are a new fast and accurate method of
3-D EM modeling that accelerate dramatically the solution
of forward EM problems in inhomogeneous 3-D geoelectrical
structures.
ACKNOWLEDGMENTS
The authors acknowledge the support of the University of
Utah Consortium for Electromagnetic Modeling and Inver-
sion (CEMI), which includes Advanced Power Technologies
Inc., Baker Atlas Logging Services, BHP Minerals, Exxon Pro-
duction Research Company, Inco Exploration, Japan National
Oil Corporation, Mindeco, Mobil Exploration and Produc-
tion Technical Center, Naval Research Laboratory, Newmont
Gold Company, Rio Tinto, Shell International Exploration and
Production, Schlumberger-Doll Research, Unocal Geother-
mal Corporation, and Zonge Engineering.
REFERENCES
Dmitriev, V. I., and Pozdnyakova, E. E., 1992, Method and algorithm
for computing the electromagnetic eld in stratied medium with
a local inhomogeneity in an arbitrary layer: Computational Mathe-
matics and Modeling, 3, 181188.
Dmitriev, V. I., Pozdnyakova, E. E., Zhdanov, M. S., and Fang, S., 1998,
Quasi-analytical solutions for the EM eld in inhomogeneous struc-
tures based on a unied iterative quasi-linear method: 68th Ann.
Mtg., Soc. Expl. Geophys., Expanded Abstracts, 444447.
Dmitriev, V. I., and Sedelnikova, A. V., 1992, Iterative method for
computing the anomalous electric eld in a conducting medium:
Computational Mathematics and Modeling, 3, 197203.
Habashy, T. M., Groom, R. W., and Spies, B. R., 1993, Beyond the
Born and Rytov approximations: A nonlinear approach to electro-
magnetic scattering: J. Geophys. Res., 98, B2, 17591775.
Hohmann, G. W., 1975, Three-dimensional induced polarization and
EM modeling: Geophysics, 40, 309324.
Nabighian, M. N., 1979, Quasi-static transient response of a conducting
half-spaceAn approximate representation: Geophysics, 44, 1700
1705.
Pankratov, O. V., Avdeev, D. B., and Kuvshinov, A. V., 1995, Scattering
of electromagnetic eld in inhomogeneous earth. Forward problem
solution: Fizika Zemli, No. 3, 1725.
Singer, B. S., and Fainberg, E. B., 1995, Generalization of iterative dis-
sipative method for modeling electromagnetic elds in nonuniform
media with displacement currents: J. Appl. Geophys., 34, 4146.
Stolz, N., Raiche, A., Sugeng, F., and Macnae, J., 1995, Is full 3-D in-
version necessary for interpreting EM data?: Expl. Geophys., 26,
167171.
Torres-Verdin, C., and Habashy, T. M., 1994, Rapid 2.5-dimensional
forward modeling and inversion via a new nonlinear scattering ap-
proximation, Radio Sci., 29, 10511079.
Weidelt, P., 1975, EMinductioninthree-dimensional structures: J. Geo-
physics, 41, 85109.
Xiong, Z., 1992, EM modeling of three-dimensional structures by the
method of systemiteration using integral equations: Geophysics, 57,
15561561.
Zhdanov, M. S., and Fang, S., 1996a, Quasi-linear approximation in 3-D
EM modeling Geophysics, 61, 646665.
1996b, 3-D quasi-linear electromagnetic inversion: Radio Sci.,
31, 741754.
1997, Quasi-linear series in 3-D EM modeling: Radio Sci., 32,
21672188.
APPENDIX A
ITERATIVE METHOD
We demonstrate here that approximate expressions (16) and
(22) for the electrical reectivity tensor can be treated as the
rst iterations within an iterative solution of the TQLequation
(8) solution.
Let us subtract G
E
[
(r
j
)E
b
] from both sides of equation
(8):
(r
j
)E
b
(r
j
) G
E
_

(r
j
)E
b
_
= G
E
_
(
(r)
(r
j
))E
b
_
+E
B
(r
j
). (A-1)
One can apply an iterative process to solve equation (A-1):
(k+1)
(r
j
)E
b
(r
j
) G
E
_

(k+1)
(r
j
)E
b
_
= G
E
_
(
(k)
(r)
(k)
(r
j
))E
b
_
+E
B
(r
j
), (A-2)
where k is the iteration number and
(0)
(r) = 0. (A-3)
Note that convergence of this iteration process for a specic
geoelectrical model depends on the properties of the operator
A[
(r)
(r
j
)] = G
E
_
(
(r)
(r
j
))E
b
_
. (A-4)
It was demonstrated by Dmitriev and Sedelnikova (1992)
that the L
2
norm of this operator determined on a class of the
slowly varying functions is usually small, which provides the
convergence of the iteration process (A-2). This assumption is
based on the fact that for slowly varying

(r), the difference
[
(r)
(r
j
)] is small if r
j
is close to r, and the kernel

G
E
(r
j
| r)
is small if the distance between the points r
j
and r is large. In
other words, one can consider that operator A[
(r)
(r
j
)], act-
ing on the class of slow varying functions, is a small operator
(has a small L
2
norm).
We demonstrated that by using a modied Greens operator
with the norm less or equal to one (Zhdanov and Fang, 1997),
we can construct the modied tensor QL equation with the
norm of the operator A[
(r)

(r
j
)] always small. So, in this
situation, the iterative process (A-2) will always converge.
In the case of a scalar electrical reectivity tensor, integral
equation (A-1) can be rewritten as
(r
j
)
_
E
b
(r
j
) E
B
(r
j
)
_
= G
E
_
((r) (r
j
))E
b
_
+E
B
(r
j
). (A-5)
Calculating the dot product of both sides of equation (A-5)
and the background electric eld, and dividing the resulting
equation by the square of the background eld, we obtain
(r
j
)[1 g(r
j
)] = A
_
(k)
(r)
(k)
(r
j
)
_
+ g(r
j
),
(A-6)
where g(r
j
) is determined by equation (17), and
A[(r) (r
j
)] =
G
E
_
((r) (r
j
))E
b
_
E
b
(r
j
)
E
b
(r
j
) E
b
(r
j
)
.
(A-7)
The integral equation (A-6) can also be solved iteratively:
(k+1)
(r
j
)[1 g(r
j
)] = A
_
(k)
(r)
(k)
(r
j
)
_
+ g(r
j
).
(A-8)
As we already discussed above, these iterations will con-
verge to a true solution due to the small norm of operator
A[(r) (r
j
)] acting on the class of slowly varying functions.
Note that these iterations will always converge if one uses the
modied Greens operator in equation (A-7).
The rst iteration of equation (A-8) yields
(1)
(r
j
) =
g(r
j
)
1 g(r
j
)
, (A-9)
which coincides with the approximate formula (16).
Thus, we can see that QA approximation can be treated as
the rst iteration in the solution of the TQL equation using the
iterative algorithm (A-8).
Note that in the same way one can demonstrate that TQA
approximation can be treated as the rst iteration in the itera-
tive solution of the TQL equation for tensor

(r).
Directional ltering for linear feature enhancement in geophysical maps
Michael P. Sykes
and Umesh C. Das
ABSTRACT
Geophysical maps of data acquired in ground and air-
borne surveys are extensively used for mineral, ground-
water, and petroleum exploration. Lineaments in these
maps are often indicative of contacts, basement fault-
ing, and other tectonic features of interest. To aid the
interpretation of these maps, a versatile processing tech-
nique of directional ltering, based on the 2-Dnormal
Radon transform, is used to enhance or suppress spe-
cic lineaments. Synthetic data and eld examples us-
ing electromagnetic and radiometric data are used to
demonstrate the superiority of the Radon transform
method over conventional Fourier transform ltering.
The Radon transform technique is shown to be more
versatile and less susceptible to processing artefacts than
the Fourier transform methods.
INTRODUCTION
Geophysical methods such as magnetic, gravity, radiomet-
ric, electrical, and electromagnetic methods acquire measure-
ments on or above the surface of the earth to assist in the inter-
pretation of geological structures in the subsurface. The degree
to which geophysical maps reect the subsurface structure is
limited by data quality and sampling. It is often necessary to
process the data in certain ways to improve the usefulness of
the maps. Traditionally, the processing has been carried out by
the application of lters in the Fourier transform (FT) domain.
Of particular interest to the geologist are the linear anoma-
lies in geophysical maps which may correspond to subsurface
faults, contacts and other tectonic features. In mineral explo-
ration, large ore bodies are often accompanied by stringers,
and many deposits are found along the boundaries between
adjacent regions or in fractured zones (Palacky, 1988). Linea-
ments on geophysical maps are often used to identify areas
of high priority for further exploration. Pawlowski (1997) has
shown the usefulness of the Radon transform (RT) in display-
ing linear features in magnetic data according to their strike
Manuscript received by the Editor August 11, 1998; revised manuscript received April 24, 2000.
Curtin University of Technology, Department of Exploration Geophysics, Perth, Western Australia, Australia.E-mail: sykes@geophy.curtin.edu.au;
das@geophy.curtin.edu.au.
direction. In his paper, a section of the transformed data is
muted out so that features having a particular orientation are
displayed. Yunxuan (1992) applied the Radon transform to
synthetic gravity maps to demonstrate its usefulness for up-
ward and downward continuation and to remove unwanted
linear features. We take these ideas further and demonstrate
the usefulness of the Radon transform to enhance linear fea-
tures in a geophysical map as well as attenuating unwanted
acquisition artifacts.
THE 2-D RADON TRANSFORM
The Radon transform is a numerical process that is useful
for analysing lineaments in a 2-D map. It is an integration pro-
cedure that, in its most general form, maps a shape in the space
domain into a single pixel in the spectral (transform) domain.
In seismic processing, a type of Radon transform, commonly
known as the slant stack, has been used for the generation
of synthetic seismograms (Chapman, 1978), multiple suppres-
sion (Taner, 1980), and ground-roll removal and separation of
refractions (Stoffa et al., 1981). Although the theory of the
Radon transform is well established and is used extensively in
seismic data processing (Duranni and Bisset, 1984; Brysk and
McCowan, 1986; Beylkin, 1987), and many other elds (Deans,
1983), we include a brief introduction for completeness.
The slant-stack RT operates along straight lines which are
dened by their slope p and intercept as shown in Figure 1(a).
Off-end seismic shot records lead to this natural choice of
line parameters since, in these records, there exists an obvi-
ous choice of origin at x =0, t =0 and a preferred direction
of events that limits the range of and p. However, the slant
stack is limited in its ability to transform steeply sloping events
(Toft, 1996), and vertical lines cannot be transformed. To avoid
this limitation, it is usual practice to dene two transform do-
mains: one relative to the x-axis, and the other relative to the
t -axis. In potential eld maps, no such preferred direction ex-
ists. To accommodate the full range of possible strike directions
the normal RT is preferred. In this form, the origin is cho-
sen as the centre of the map and the lines dened by their
normal slope () and offset () parameters [Figure 1(b)].
Integration of the map amplitudes along each line (S lines
1758
Geophysical Map Enhancement 1759
in Figure 1) produces the amplitude of the spectral domain
point located at (, ). It should be noted that the normal RT
and the slant-stack RT are very closely related and perform
the same function equally well. The main difference between
the two forms of RT is that individual points transform into
straight lines when using slant stack, but into sinusoids when
usingthenormal RT. Mathematically, thespectral domainfunc-
tion F(, ) is obtained by applying the normal RT to the
space domain function f (x, y) with the help of the following
equation:
F(, ) =
f (x, y)( x cos y sin ) dx dy,

(1)
where is the Dirac delta function. The delta function ensures
that only points that lie onthe line S contribute tothe transform
domain. The RT maps linear features with signicant strike
length as high-amplitude points at the appropriate location in
FIG. 1. (a) Schematic of the slant-stack Radon transform. The
space domain function f (x, y) is integrated along S lines de-
ned in terms of the parameters and p. The value of the
integral is assigned to the coordinate (, p) in the transform
domain. (b) Schematic of the normal Radon transform. The
space domain function f (x, y) is integrated along S lines de-
ned in terms of the parameters and . The value of the
integral is assigned to the coordinate (, ) in the transform
domain.
the spectral domain. A unique correspondence exists between
space domain linear features and spectral domain high spots
(Robinson, 1982). Thus, the analysis of the linear features in a
2-D map can be carried out by examining the amplitude distri-
bution of the transform domain function. The high amplitudes
in the transformdomain can be selectively amplied or attenu-
ated to produce a corresponding enhancement or suppression
of specic linear events in the space domain.
Inversion of F(, ) to f (x, y) is achieved via the ltered
back-projection method (Toft, 1996), dened as
f (x, y) =
||
F(, )e
i 2
d
e
i 2(x cos +y sin )
d d, (2)
where is a wavenumber vector used for ltering in the Fourier
domain.
SYNTHETIC EXAMPLES
To demonstrate the usefulness of the RT as a data process-
ing tool, we rst use synthetic data and then follow with some
real geophysical maps. The RTand inverse RTare numerical
processes that are dependent on the sampling rate used in the
computation. The implementation of the transform pair is not
exact and will not recover the original amplitudes. Increasing
the sampling rate to produce a more exact result leads to in-
creased CPU time and memory, which may be prohibitive in
some cases. Since we are concerned with image enhancement,
the recovery of exact amplitudes is not so important. However,
it is important that a sufciently dense sampling of the map is
used to avoid aliasing in the transform domain and the gener-
ation of transform artifacts in the new map. In the examples
presentedhere, we followthe sampling requirements described
by Toft (1996).
Transform integrity
Figure 2a presents a very simple test pattern comprised of
vertical and horizontal lines; Figure 2b presents its Radon
transform. The linear features are clearly distinguishable in the
spectral domain as high-amplitude peaks whose locations are
determined by the corresponding line parameters (, ). Note
that the horizontal lines produces two peaks in the transform
domain. This is because has values ranging from /2 to
+/2, both of which represent a horizontal line in the space
domain. This double sampling of horizontal features could
be avoided by restricting the range of values used in the
forward RT, but is considered unnecessary and undesirable.
Parallel linear events in the space domain are transformed
as high-amplitude peaks at different locations in the spectral
domain due to their different offset from the origin. This
separation in the Radon spectral domain enables the linear
features to be processed separately.
Figure 2c contains the map recovered by application of the
forward and inverse RTtransform pair; Figure 2d shows a con-
tour mapof theamplitudedifferences betweenthetwomaps. In
most parts of the recovered map, the amplitude differences are
less than 2% of the original amplitudes. The largest difference
represents less than 10% change. The test pattern represents a
difcult image to process due to the very sharp boundaries. The
contours that represent high-percentage changes occur where
1760 Sykes and Das
there are small differences in the low amplitudes near steep
edges. In most geophysical maps, the amplitude gradients are
less severe and will therefore not suffer the same level of vari-
ation. In terms of visual appearance, the two maps are very
similar. This provides condence that the RT can be used for
image processing without creating excessive distortion of the
original image.
It shouldbe emphasisedthat excessive processing will always
produce deleterious effects on the maps. Care must be taken
to ensure that the features of interest are enhanced as much
as possible without adversely affecting the rest of the map. In
each of the following examples of RT processing, we present
the maps produced using the corresponding Fourier domain
processing for comparison.
FIG. 2. (a) Test pattern comprised of vertical and horizontal features. (b) Radon transform of (a). (c) Map produced by application
of the forward and inverse Radon transform pair. (d) Contour surface of the difference between (a) and (c). Most of the map is
recovered within 2% of the original amplitudes. Larger variations (up to 10%) are observed near the steep edges.
Directional ltering
Directional ltering is useful for removing unwanted fea-
tures in the map that have a particular orientation such as
fence-line and pipeline anomalies in EM and magnetic maps.
Traditionally, directional ltering is achieved by the applica-
tion of a cosine tapered, pass or rejection lter to the Fourier
domain data, in the direction of choice. The lters are often re-
ferred to as pie-slice lters. To test the Radon ltering tech-
nique, we use the map in Figure 2b. Reduction of the amplitude
peaks prior to inversion will result in the attenuation of the cor-
responding linear feature in the recovered map. To this end, we
reduce the peaks that correspond to horizontal lines to a se-
lected level prior to inversion and followby reducing the peaks
that correspond to vertical lines. For the models presented in
this paper, we follow the recommendation of Yunxuan (1992)
and replace the peak amplitudes with the mean value of the
entire RTdata prior toinversion. This helps topreserve the am-
FIG. 3. (a) Horizontal and vertical features attenuated using Radon domain ltering. (b) Particular vertical
features attenuated by Radon domain ltering. (c) Fourier domain direction ltering of horizontal and vertical
features produces signicant ringing.
plitudes in the new map and reduces the severity of transform
artifacts. In this implementation of the RT pair, the choice of
amplitude scaling is not particularly important. What is more
important is the appropriate selection of spectral data points
1762 Sykes and Das
to which the scaling is applied. If insufcient data points are
scaled, the linear feature will not be completely absent from
the recovered map. If too many data points are scaled, other
features in close proximity in the spectral domain may also be
altered. Selection of the optimum scaling window is achieved
by examining the amplitude distribution of the Radon domain
FIG. 4. (a) Test pattern in which linear features cross. (b) Radon transform of the test pattern and (c) the recovered maps with
selected features removed by Radon domain ltering. (d) The maps produced using conventional Fourier domain directional
(pie-slice) ltering. The Radon domain ltering has greater directionality and has very little effect on the unltered lineaments.
data. Tapering of the window edges was considered unneces-
sary since the crude method of block scaling appears to work
adequately for the purpose of image enhancement.
The maps shown in Figure 3a show that the RT ltering is
effective in removing horizontal and vertical features without
adversely affecting the appearance of the map. In addition,
we show in Figure 3b that the RT ltering can be applied to
specic linear features, even if they have the same strike di-
rection. In this gure, only the vertical features on one side of
the map have been attenuated and there is very little effect
on the other features in the map. For comparison, in Figure 3c
we show that conventional directional ltering in the Fourier
domain is very limited in its application to particular features.
Such a lter enhances all of the features having the particular
strike direction by attenuating all others. A cosine roll-off ta-
per is applied to the lter to reduce the generation of artifacts
caused by the ltering. Despite this, the Fourier directional l-
tering has produced noticeably more ringing than the RT
method. The FT method cannot be used to lter out specic
features.
To demonstrate the technique for the case where lineaments
cross, another simple data set was produced (Figure 4a) that
comprised three linear features of different strike but common
amplitude and midpoint. The Radon transform of this data set
FIG. 5. (a) A very noisy space domain map in which a sloping linear feature is barely visible and (b) its corresponding Radon
transform with characteristic lineament peak. (c) The map produced after amplication of the RT peak prior to inversion. (d) The
map produced if the noise is rejected by applying a threshold cutoff to the Radon data prior to inversion. (e) The map produced by
directional ltering in the Fourier domain.
is shown in Figure 4b. The ability of Radon domain ltering
to target selected features separately and remove them from
the map is demonstrated in Figures 4c. As a comparison, di-
rectional ltering was applied to the maps shown in Figure 4a
using the more conventional Fourier domain directional lter-
ing. In Figure 4d, the FT directional ltering can be seen to be
less direction specic than the RT method. The amplitudes of
the unlteredfeatures are affectedmore by the Fourier domain
processing.
Enhancement of linear features
In Figure 5a, we show a very noisy synthetic data map con-
taining a diagonal linear feature whose amplitude is equal to
the maximum noise level. The linear feature is very difcult
to resolve in the space domain due to the noisy background.
In real situations, such a signal-to-noise ratio is unacceptably
small, but it is used here for demonstration purposes. We apply
1764 Sykes and Das
FIG. 6. (a) Computed AEM response of a conductive target using a towed-bird conguration showing the distorted edges (her-
ringbone pattern) due to measurement asymmetry. (b) The corresponding response if a symmetric measuring system is used.
(c) A spectral line (constant ) of the Radon domain obtained from (a) showing rapid amplitude variations due to the herringbone
pattern. (d) The corresponding spectral line obtained from the transform of (b) to be free of the amplitude variations. (e) The
space domain map produced from (a) after ltering in the Radon domain to reduce the spectral line amplitude variations. (f) The
herringbone pattern removed by Fourier domain ltering.
the RT to this data to produce Figure 5b. The linear feature is
nowquite recognisable as a peak in the spectral domain. To en-
hance the appearance of the linear feature in the new map, the
spectral data was scaled to enhance the peak prior to the inver-
sion. This was achieved by selecting a threshold value equal to
the approximate backgroundlevel inthe spectral domain. Data
points having a value larger than the threshold were amplied.
The degree of amplication was varied until a satisfactory re-
sult was achieved. Toproduce the mapof Figure 5c, the spectral
peak was amplied by a factor of two prior to inversion. The
linear feature is much more apparent in the new map than it
was in the original map (cf. Figure 5a).
A corollary of the previous example in which scaling was
applied to enhance the peak in the spectral data is to reduce
the portion of the spectral data which is attributable to the
unwanted nonlinear features (noise) in the space domain. In
this case, the spectral data having values less than the thresh-
old level were zeroed. Figure 5d shows the map produced by
scaling down the background rather than enhancing the peak.
In both cases, the linear event is much more apparent after RT
processing.
As a comparison, a directional cosine-pass lter in the direc-
tion of the lineament was used to produce the image in Fig-
ure 5e. In this case the directional ltering has enhanced the
lineament. However, directional ltering in the Fourier do-
main could not be used as effectively if the map contained
two features having different strike directions. In such a case,
one of the features would be removed from the map alto-
gether. The Radon domain ltering does not suffer from this
restriction.
Removal of data acquisition artifacts
Data maps produced from airborne geophysical measure-
ments often contain artifacts attributable to ight-line direc-
tion. With the xed-wing towed-bird congurations used in
many airborne electromagnetic (AEM) surveys, the asymme-
try of the measuring system often manifests itself as a herring-
bone noise pattern in the mapped data. The jagged appearance
of conductor boundaries, caused by the apparent displacement
of the boundary in the direction of ight, is often called a her-
ringbone pattern. Since most surveys are own in opposite di-
rections along adjacent lines, the displacements alternate from
line to line and produce the herringbone effect. These effects
have a clearly dened direction and will therefore be conned
to a particular region in the RT spectral domain.
As an illustration of herringbone artifacts, a synthetic AEM
map (Figure 6a) of the vertical magnetic eld component
(dB
z
/dt ) was computed for an asymmetric xed-wing towed-
birdconguration(i.e., transmitter andreceiver are at different
altitudes) over a conductive target body in a conductive host
medium (half-space). The presence of these artifacts on a map
reduces boundary denition. In contrast, the calculated AEM
map for a symmetric acquisition system is much smoother and
more clearly represents the boundaries of the conductor (Fig-
ure 6b). The herringbone pattern can been seen in the RT do-
main as rapid variations along spectral lines having values
corresponding to the ight-line direction. Figures 6c and d,
respectively, show a plot of a spectral line obtained from each
map. Removal of the rapid amplitude variations fromFigure 6c
prior toinversionresults inanewmapthat is freeof theherring-
bone pattern (Figure 6e). A variety of techniques were used to
remove the rapid amplitude variations in the RT domain in-
cluding zeroing, scaling, averaging along the line, interpolation,
and 1-D Fourier transform ltering. All techniques were suc-
cessful inproducingamapwithout herringbonepattern. Again,
the most important factor is identication of the portion of the
RT domain that needs to be ltered.
Directional ltering of the ight-line direction by conven-
tional Fourier cosine rejection produced the map shown in
Figure 6f. Comparison of Figures 6e with f provides evidence
that the RT processing is at least as effective in removing her-
FIG. 7. (a) AEM eld data map containing a signicant false
anomaly. (b) The same map after Radon domain ltering to
remove the unwanted feature. Note the minimal effect on the
rest of the map. (c) The same map to which Fourier directional
ltering has been applied. The pie-slice Fourier lter has
severely altered the nature of the map.
1766 Sykes and Das
ringbone pattern as FT directional ltering and has much less
impact on the rest of the map.
FIELD EXAMPLES
We apply the RT enhancement procedures described above
to three different geophysical data maps and compare with
maps produced using traditional Fourier domain processing.
FIG. 8. (a) Total count radiometric map and (b) the same map after scaling of the Radon domain peaks to enhance the lineaments.
(c) and (d) The maps produced by Fourier ltering in the northeast and northwest directions, respectively.
Directional ltering to remove linear features
From inspection of a typical AEM map (Figure 7a), we see
a dominant north-trending feature in the upper part of the
map directed along a ight line. This is believed to be due to
instrument malfunctionandnot a true signal. Withthis inmind,
the peak in the RTdomain corresponding to this linear feature
was reduced. Inversion of the ltered RTdomain produced the
map shown in Figure 7b. With the dominant feature removed
from the map, the underlying features can be more clearly ob-
served.
Fourier cosine rejection applied to the ight line direction
produced the map shown in Figure 7c. The Fourier lter has
removed the unwanted feature but has severely affected the
rest of the map.
Lineament enhancement in radiometric data
In an attempt to enhance linear trends in a typical radiomet-
ric data map (Figure 8a), we applied the Radon transform. As
described earlier, selected amplitudes in the Radon domain
were amplied (doubled in this case). After inversion back
to the spatial domain, the lineaments in the new map (Fig-
ure 8b) are much sharper and more easily seen. As a compari-
son, northwest and northeast direction ltering of the Fourier
domain was used to enhance the lineaments in the map (Fig-
ures 8c and d). Both of these maps are very different from the
original and display lineaments in only one direction.
Removal of acquisition artefacts
Figure 9a presents a portion of a time-domain AEM map
acquired using a towed-bird Geotemsystem. Herringbone pat-
tern, resulting from the asymmetric transmitter-receiver con-
guration, is apparent over most of the map. Figure 9b shows
the improved map after smoothing of the appropriate RT
spectral lines (using a ve-point averaging lter) prior to in-
version back to the space domain. The herringbone noise pat-
tern has been very effectively removed, allowing for improved
interpretation of the remaining anomalies.
Application of a Fourier cosine rejection lter produced the
map shown in Figure 9c. While the FT lter has removed the
herringbones, it has also distorted many of the features dis-
played in the map.
CONCLUSION
It has been shown that the Radon transform is a very use-
ful processing tool for improving the quality of geophysical
maps. The ability of the RT to isolate directional features in
the spectral domain enables each lineament to be processed
separately without signicantly affecting the rest of the map.
Scaling of selected portions of the Radon domain data can be
used to enhance or attenuate specic lineaments present in the
map. A number of different scaling techniques were used, and
there appears to be little advantage of one scaling technique
over another. Selection of the optimum scaling window is best
achieved by inspection of the RT data. Tapered scaling win-
dows were considered unnecessary as the RT technique is very
robust and does not produce transformartifacts to a signicant
degree. However, the inability of the RT to exactly recover the
amplitudes in the reconstructed maps is an area that needs fur-
ther work. This shortcoming can be tolerated in view of the
overall improvement in appearance of the maps.
In the examples presented in this paper, the RT processing
has been shown to be superior to the conventional FT domain
processing for lineament enhancement. The Radon domain l-
tering need only be applied to a small portion of the trans-
formed data. The FT domain processing is not event specic
and is therefore more likely to suffer from the production of
transform artifacts, even when using a tapered lter.
ACKNOWLEDGMENT
The authors express their gratitude to Harold Yarger and the
other reviewers for their helpful suggestions for improvement
of the original manuscript.
REFERENCES
Beylkin, G., 1987, Discrete Radon transform: IEEE Trans. Acoustics,
Speech and Signal Processing, 35, 162172.
Brysk, H., and McCowan, D. W., 1986, A slant-stack procedure for
point-source data: Geophysics, 51, 13701386.
FIG. 9. (a) AEM map with herringbone pattern and (b) the
same map after Radon domain ltering. (c) The map produced
by Fourier ltering in the ight-line direction. The FTlter has
severely distorted the features in the map.
1768 Sykes and Das
Chapman, C. H., 1978, Anewmethod for computing synthetic seismo-
grams: Geophys. J. Roy. Astr. Soc., 54, 482518.
Deans, S. R., 1983, Radon transform and some of its applications: John
Wiley & Sons, Inc.
Durrani, T. S., and Bisset, D., 1984, The Radon transform and its prop-
erties: Geophysics, 49, 11801187.
Palacky, G. J., 1988, Resistivity characteristics of geologic targets, i n
Nabighian, M. N., ed., Electromagnetic methods in applied geo-
physics, Vol. 1, Theory: Soc. Expl. Geophys., 53129.
Pawlowski, R. S., 1997, Use of slant stack for geologic or geophysical
map lineament analysis: Geophysics, 62, 17741778.
Robinson, E. A., 1982, Spectral approach to geophysical inversion by
Lorentz, Fourier and Radon transforms: Proc. IEEE, 70, 10391053.
Stoffa, P. L., Buhl, P., Diebold, J. B., and Wenzel, F., 1981, Direct
mapping of seismic data to the domain of intercept time and ray
parameterA plane-wave decomposition: Geophysics, 46, 255
267.
Taner, M. T., 1980, Long-period sea-oor multiples and their suppres-
sion: Geophys. Prosp., 28, 3048.
Toft, P. A., 1996, The Radon transformtheory and implementation:
Ph.D. thesis, Technical Univ. of Denmark.
Yunxuan, Z., 1992, Application of the Radon transform to the pro-
cessing of airborne geophysical data: Ph.D. thesis, Internat. Inst. for
Aerospace Survey and Earth Sciences.
Gas generation and overpressure: Effects on seismic attributes
Jos e M. Carcione
and Anthony F. Gangi
ABSTRACT
Drilling of deepgas resources is hamperedby highrisk
associated with unexpected overpressure zones. Knowl-
edge of pore pressure using seismic data, as for instance
fromseismic-while-drilling techniques, will help produc-
ers plan the drilling process in real time to control po-
tentially dangerous abnormal pressures.
We assume a simple basin-evolution model with a con-
stant sedimentation rate and a constant geothermal gra-
dient. Oil/gas conversion starts at a given depth in a
reservoir volume sealed with faults whose permeability
is sufciently low so that the increase in pressure caused
by gas generation greatly exceeds the dissipation of pres-
sure by ow. Assuming a rst-order kinetic reaction, with
a reaction rate satisfying the Arrhenius equation, the
oil/gas conversion fraction is calculated. Balancing mass
and volume fractions in the pore space yields the ex-
cess pore pressure and the uid saturations. This excess
pore pressure determines the effective pressure, which
in turn determines the skeleton bulk moduli. If the gen-
erated gas goes into solution in the oil, this effect does
not greatly change the depth and oil/gas conversion frac-
tion for which the hydrostatic pressure approaches the
lithostatic pressure.
The seismic velocities versus pore pressure and differ-
ential pressure are computed by using a model for wave
propagation in a porous medium saturated with oil and
gas. Moreover, the velocities and attenuation factors ver-
sus frequency are obtained by including rock-frame/uid
viscoelastic effects to match ultrasonic experimental ve-
locities. For the basin-evolution model used here, pore
pressure is seismically visible whenthe effective pressure
is less than about 15 MPa and the oil/gas conversion is
about 2.5% percent.
INTRODUCTION
In deeply buried oil reservoirs, oil-to-gas cracking may in-
crease the pore pressure to reach or exceed the lithostatic
pressure (Chaney, 1950; Barker, 1990; Luo and Vasseur, 1996).
Oil can be generated from kerogen-rich source rocks and can
ow through a carrier bed to a sandstone reservoir rock. Ex-
cess pore-uid pressures in sandstone reservoirs are generated
when the rate of volume created by the transformation of oil to
gas is more rapidthanthe rate of volume loss by uidow. If the
reservoir is sealed on all sides by an impermeable shale or lime-
stone, then the condition of a closed system will be satised for
gas generation. Because of the presence of semivertical fault
planes andcompartmentalization, this conditionholds for most
North Sea reservoirs. Moreover, gas generation signicantly
decreases the relative permeability (Luo and Vasseur, 1996).
Berg and Gangi (1999) developed a simple model to cal-
culate the excess pore pressure as a function of the fraction of
kerogenconvertedtooil andthefractionof oil convertedtogas.
The oil/gas conversion ratio is computed as a function of time,
Manuscript received by the Editor September 3, 1998; revised manuscript received October 25, 1999.
Osservatorio Geosico Sperimentale, P.O. Box 2011, 34016 Trieste, Italy. E-mail: jcarcione@ogs.trieste.it.
Department of Geology and Geophysics, Texas A&M University, College Station, Texas 77843-3114. E-mail: gangi@tamu.edu.
for a given sedimentation rate and geothermal gradient. Then
Berg and Gangi (1999) determined the excess pore pressure
resulting from the oil/gas conversion with burial time, which
is derived by balancing mass and volume changes in the pore
space. Carcione (2000) applied this model of the calculation of
the kerogen/oil conversion to obtain the seismic properties of
overpressured shales. Here, we use an extended version of this
model for computing the porosity variations and uid satura-
tions as a function of the excess pore pressure. We consider two
cases, when the generated gas goes into solution in the oil and
whenit remains as free gas. Whengas goes intosolutionindead
oils, it generates live oils. The analysis is based on the equations
for live oil given by Batzle and Wang (1992). The solubilities of
hydrocarbon gases in water are very small compared with the
solubilities of the same gases in oil.
Seismic velocities have been used to predict pore pressure,
with most of the models based on empirical relations (Dutta
and Levin, 1990). Attenuation has been studied mainly at ul-
trasonic frequencies, and a few works have focused on verti-
cal seismic proling (VSP) data and sonic data (Hague, 1981;
1769
1770 Carcione and Gangi
Dasios et al., 1998a). Recently, attenuation obtained from sur-
face seismic data has beenusedsuccessfully topredict lithology
and pore pressure (Helle et al., 1993; Dasios et al., 1998b). In
this work, the seismic properties are calculated by using a mod-
ication of Biots theory for fully saturated porous solids. The
effective parameters for the uid mixture are computed by the
average formulas given by Berryman et al. (1988).
The wave velocities of the dry rock, determined by labora-
tory measurements, give estimates of its pore compressibility
and bulk moduli in the low-frequency (relaxed) regime. Be-
cause dissipation mechanisms are caused mainly by grain/uid
interactions, dry-rock velocities are frequency independent.
On the other hand, measurements of ultrasonic saturated-rock
velocities provide the high-frequency-limit (unrelaxed) veloc-
ities, which can be used to estimate the amount of velocity dis-
persion between the seismic and ultrasonic ranges and, there-
fore, the amount of attenuation. By using this information,
the phase velocities and attenuation factors are computed as a
function of frequency.
GAS GENERATION AND ROCK PROPERTIES
First, We will introduce some useful denitions about the
different pressures considered in this work. Pore pressure, also
known as formation pressure, is the in situ pressure of the u-
ids in the pores. The pore pressure is equal to the hydrostatic
pressure when the pore uids only support the weight of the
overlying pore uids (mainly brine). The lithostatic or conn-
ing pressure results from the weight of overlying sediments,
including the pore uids. In the absence of any state of stress in
the rock, the pore pressure attains lithostatic pressure and the
uids support all the weight. However, fractures perpendicular
tothe minimumcompressive stress directionappear for a given
pore pressure, typically 7090% of the conning pressure. In
this case, the uid escapes from the pores, and pore pressure
decreases. A rock is said to be overpressured when its pore
pressure is signicantly greater than hydrostatic pressure. The
difference between pore pressure and hydrostatic pressure is
called differential pressure. Acoustic and transport properties
of rocks generally depend on effective pressure, a combination
of pore and conning pressures [see equation (20)]. Various
physical processes cause anomalous pressures on an under-
ground uid. The most common causes of overpressure are
compaction disequilibrium and cracking, i.e., oil-to-gas con-
version (Mann and Mackenzie, 1990; Luo and Vasseur, 1996).
Let us assume a reservoir at depth z. The lithostatic pres-
sure for an average sediment density of is equal to p
c
= gz,
where g is the acceleration of gravity. On the other hand, the
hydrostatic pore pressure is approximately p
H
=
w
gz, where
w
is the density of water.
For a constant sediment burial rate, S, and a constant
geothermal gradient, G, the temperature variation of a par-
ticular sediment volume is
T = T
0
+ Gz, z = St, (1)
with a surface temperature T
0
at time t =0. Typical values of G
range from 20 to 30
C/km, and S may range between 0.05 and

0.5 km/m.y. (m.y. = million years).
Assume that at time t
i
, corresponding to depth z
i
, the reser-
voir volume has been saturated with oil which owed from an
adjacent source rock, and that the volume is closed. That is,
the permeability of the sealing faults is sufciently low so that
the rate of pressure increase caused by gas generation greatly
exceeds the dissipation of pressure by ow. Pore-pressure ex-
cess is intended to be above hydrostatic.
Oil/gas generation rate
The mass of convertible oil changes with time t at a rate
proportional tothemass present. Assumingarst-order kinetic
reaction (Luo and Vasseur, 1996; Berg and Gangi, 1999),
dM
o
dt
= r
o
(t )M
o
(t ) (2)
or
M
o
(t ) = M
oi
exp
_
_
t
t
i
r
o
(t ) dt
_
, (3)
where r
o
(t ) is the reaction rate, M
o
(t ) is the mass of convertible
oil at time t , and M
oi
is the initial oil mass. The fraction of oil
converted to gas is F(t ) =[M
oi
M
o
(t )]/M
oi
:
F(t ) =1exp
_
_
t
t
i
r
o
(t
) dt
_
1exp[(t )]. (4)
The reaction rate follows the Arrhenius equation (Luo and
Vasseur, 1996)
r
o
(t ) = A exp[E/RT(t )], (5)
where E is the oil/gas activation energy, R =1.986 cal/mol

K
is the gas constant, A is the oil/gas reaction rate at innite
temperature, and T(t ) is the absolute temperature in

K given
by
T = T
0
+ Ht, H = GS. (6)
With this temperature dependence, the integral (t ) be-
comes
(t ) =
_
t
t
i
r
o
(t
) dt
=
A
H
_
T
T
i
exp(E/RT
) dT
,
T
i
= T
0
+ Ht
i
(7)
or
(t ) =
A
H
_
T
_

1
exp(Ex/RT)
dx
x
2
T
i
_

1
exp(Ex/RT
i
)
dx
x
2
_
. (8)
For values of E/RT greater than 10, the exponential inte-
gral can be approximated by (Gautschi and Cahill, 1964, 248,
Table 5.5)
_

1
exp(Ex/RT)
dx
x
2
=
exp(E/RT)
2 + E/RT
, (9)
Gas Generation, Overpressure, and Seismics 1771
with an error of 1.3% or less. Then the integral becomes
(T(t )) =
A
H
_
T exp(E/RT)
2 + E/RT

T
i
exp(E/RT
i
)
2 + E/RT
i
_
.
(10)
The activation energy and innite-temperature rate used in
this work are E =52 kcal/mol and A =5.510
26
/m.y. (Luo and
Vasseur, 1996; Berg and Gangi, 1999).
Oil/gas conversion factor and excess pore pressure
We are interested primarily in a rst-order solution which
shows the large effect of oil-to-gas conversion on seismic prop-
erties. In this sense, following Berg and Gangi (1999), we as-
sume:
1) The compressibilities of oil and water are independent of
pressureandtemperature, andthat of therockis indepen-
dent of temperature but depends on pressure. That this is
the case can be seen from the results given by Batzle and
Wang (1992) in their Figures 5 and 13, in which they show
that the density is almost a linear function of tempera-
ture and pressure. This means that the mentioned prop-
erties are approximately constant (see also their Figure 7,
in which the oil compressibility remains almost constant
when going from low temperature and low pressure to
high temperature and high pressure).
2) The gas satises the van der Waals equation (Friedman,
1963):
_
p +a
2
g
_
(1 b
g
) =
g
RT, (11)
where p is the gas pressure,
g
is the gas density, T is the
absolute temperature, and R is the gas constant. More-
over, a =0.225 Pa (m
3
/mole)
2
=879.9 MPa (cm
3
/g)
2
,
and b =4.28 10
5
m
3
/mole =2.675 cm
3
/g (one mole of
methane, CH
4
, corresponds to 16 g). The van der Waals
equationis a goodapproximationof the behavior of natu-
ral gas, as shownby Berg andGangi (1999), where the dif-
ferences between the experimental dataas represented
by Standings results (Standing, 1952)and the van der
Waals results are only about 15% over the depths of in-
terest.
3) The initial pore uids are water and oil.
The excess pore pressure at depth z is pp
H
, where p
H
is the
hydrostatic pore pressure at depth z, and p is the pore pressure
when a fraction F of oil has been converted to gas (F =0 and
p = p
i
= p
H
at time t
i
) . The mass balance is
g
V
g
= F
oi
V
oi
, (12)
where
oi
is the initial oil density, V
g
is the converted gas vol-
ume, and V
oi
is the initial oil volume. Assumption (3) implies
V
pi
= V
wi
+ V
oi
or V
pi
/V
wi
= 1 +v, (13)
where V
pi
is the initial pore volume and v =V
oi
/V
wi
.
The compressibilities of the oil, water, and pore space are
dened, respectively, as
c
o
=
1
V
o
dV
o
dp
, c
w
=
1
V
w
dV
w
dp
, c
p
=
1
V
p
dV
p
dp
e
,
(14)
where p
e
is the effective pressure and c
p
can be obtained from
the compressibility at zero pore pressure.
In general, compressional and shear-wave velocities depend
on effective pressure p
e
= p
c
np, where n 1 is the effective
stress coefcient. Note that the effective pressure equals the
conning pressure at zero pore pressure. It is found that n 1
for static measurements of the compressibilities (Zimmerman
et al., 1986), while n is approximately linearly dependent on
the differential pressure p
d
= p
c
p in dynamic experiments
(Gangi and Carlson, 1996; Prasad and Manghnani, 1997):
n = n
0
n
1
p
d
, (15)
where n
0
and n
1
are constant coefcients.
We assume the following functional formfor c
p
as a function
of effective pressure:
c
p
= c
p
+ exp
_
p
e
_
p
_
, (16)
where c
p
, , and p
are coefcients obtained by tting the

experimental data. On the other hand, the thermal expansion
coefcients for oil, water, and pore space are dened, respec-
tively, as
o
=
1
V
o
dV
o
dT
,
w
=
1
V
w
dV
w
dT
,
p
=
1
V
p
dV
p
dT
. (17)
Integration from p
i
(p
ei
) to p (p
e
) and from T
i
to T
i
+T,
where T =T T
i
, yields
V
o
(p, T) = V
oi
[exp(c
o
p +
o
T)],
V
w
(p, T) = V
wi
[exp(c
w
p +
w
T)], (18)
and
V
p
(p, T) = V
pi
{exp[E(p) +
p
T]}, (19)
where
E(p) = c
p
p
e
+p
_
exp
_
p
e
_
p
_
exp
_
p
ei
_
p
__
and p = p p
i
= p p
Hi
. Using equation (15), we can write
the effective pressure as
p
e
= p
c
(n
0
n
1
p
c
)p n
1
p
2
. (20)
Berg and Gangi (1999) assume a constant pore compressibility
c
p
. In this case, E(p) =c
p
p.
In principle, if no gas goes into solution in the oil, the pore
volume at pore pressure p and temperature T is given by
V
p
(p, T) = V
pi
{exp[E(p) +
p
T]}
= V
wi
[exp(c
w
p +
w
T)] +(1 F)V
oi
[exp(c
o
p +
o
T)] + V
g
(p, T), (21)
since (1 F) of the initial oil remains at pressure p. Substitut-
ing equation (12) into equation (21) and dividing through by
V
wi
yields
o
=

0
+(0.00277p 1.71 10
7
p
3
)(
0
1.15)
2
+3.49 10
4
p
0.972 +3.81 10
4
(T +17.78)
1.175
. (25)
_
1 +
1
v
_
{exp[E(p) +
p
T]}
=
1
v
exp(c
w
p +
w
T)
+(1 F)[exp(c
o
p +
o
T)] + F
oi
/
g
. (22)
In the absence of water, 1/v 0.
There are two possibilities when the generated gas goes into
solution: (1) all the gas dissolves in the oil, which occurs at the
beginning of the conversion process down to a critical depth,
and (2) a fraction of gas goes into solution in the oil and the
rest remains as free gas. The process can be approximated, as
illustrated in Figure 1: It begins at stage (1); in stage (2), a
volume V
g
=V
g1
+V
g2
of gas is generated, with V
g1
going im-
mediately into solution and V
g2
remaining as free gas (stage 3).
In the following, we obtain the balance equations, equivalent
to equation (22) and the fractions of solid, water, live oil, and
free gas at stage (3), for the two cases described above.
The volume ratio of liberated gas to remaining oil at atmo-
spheric pressure and 15.6
C is
R
G
= 0.02123G
r
_
p exp
_
4.072
0
0.00377T
__
1.205
, (23)
where G
r
is the gas gravity and
0
is the oil density at atmo-
spheric pressure and temperature (Batzle and Wang, 1992);
R
G
, given in liters of gas/liters of oil, represents the maximum
amount of gas that can be dissolved in the oil. At depth z,
temperature T, and pressure p, the equivalent ratio is
R
G
=

gs
os
R
G
, (24)
where
gs
,
os
and
g
,
o
are the gas and oil densities at the
surface and at depth z, respectively. The gas densities can be
computed from the van der Waals equation (11), and the oil
density is given by
FIG. 1. Oil-to-gas conversion process. Part of the generated gas
goes into solution.
Oil densities are given in g/cm
3
, temperature T in
C, and pres-
sure p in MPa. Because R
G
is the maximumvolume of gas (V
g1
)
that can be dissolved in the remaining volume of dead oil, we
have
R
G
=
_

o
(1 F)
oi
_
V
g1
V
oi
. (26)
The volume of free gas (V
g2
) is equal to the total volume of
generated gas (V
g
) minus the maximum volume of dissolved
gas (V
g1
),
V
g2
= F
oi
g
V
oi
(1 F)
oi
o
R
G
V
oi
=
__
F
(1 F)
o
R
G
_
oi
_
V
oi
V
oi
, (27)
where equation (26) has been used. There is a critical depth
z
c
for which the gas saturates the oil. Above this depth (which
corresponds to a critical pressure p
c
), all the gas goes into solu-
tion in the oil (note that the process is a dynamic one, in which
gas is generated continuously). This occurs when the volume
of free gas [equation (27)] is equal to zero, i.e., when a fraction
F
c
=

g
R
o
+
g
R
G
(28)
of oil has been converted to gas, at the critical pressure p
c
.
Below the critical depth, the pore space will contain live oil
mixed with free gas. To compute z
c
and F
c
, we assume that the
system is subject to the pore-pressure prole obtained from
equation (22), which is the situation shown in stage (2) of Fig-
ure 1, before the gas goes into solution. Then z
c
is the depth
for which =0 [see equation (27)] or F = F
c
. It can be argued
that after the gas goes into solution, the pressure will decrease.
However, if this occurs, part of the gas will go out of solution
immediatley (as free gas) and the pressure will increase again
to the previous value. It can be shown that by choosing, for
instance, the hydrostatic pressure p
H
as the reference pressure
prole, instead of that obtained from equation (22), the criti-
cal depth will be different, but the results will not be affected
greatly.
The density of the live oil at and below the critical depth is
given by
lo
=
G
+(0.00277p 1.71 10
7
p
3
)(
G
1.15)
2
+3.49 10
4
p (29)
with
G
= (
0
+0.0012G
r
R
G
)/B
0
, (30)
the saturation density, and
B
0
= 0.972 +0.00038
_
2.4R
G
_
G
r
0
_
1/2
+ T +17.8
_
1.175
,
(31)
the oil-volume factor (Batzle and Wang, 1992). When comput-
ing the density of live oil, the temperature effect was consid-
ered twice in Batzle and Wang (1982), in B
0
and in their equa-
tion (19) (M. Batzle, personal communication). Equation (29)
gives the correct density. At the initial depth z
i
there is no
generated gas, and the oil density is that of the dead oil
oi
.
When z
i
z z
c
[case (1)], all the gas goes into solution, but
the oil is not saturated because it absorbs less than R
G
liters of
gas; it absorbs (F/F
c
)R
G
. The density,
lo
, of live oil is obtained
by substituting R
G
by (F/F
c
)R
G
in equations (29), (30), and
(31). Because, in this range, the mass of live oil is equal to the
initial mass of dead oil, the volume of live oil is
V
lo
=

oi
lo
V
oi
V
oi
. (32)
Then the balance equation above the critical depth is
_
1 +
1
v
_
{exp[E(p) +
p
T]}
=
1
v
exp(c
w
p +
w
T) +. (33)
We nowconsider case (2), i.e., when z >z
c
andthe pore space is
lledwithlive oil andfree gas. The mass of live oil
lo
V
lo
is equal
to the mass of remaining dead oil (1 F)
oi
V
oi
plus the mass
of gas going into solution. The latter is equal to the maximum
volume of gas V
g1
that can be dissolved in the remaining dead
oil [equation (26)], multiplied by the gas density
g
. Then the
volume of live oil is
V
lo
=
_
oi
lo
(1 F)
_
1 + R
o
__
V
oi
V
oi
. (34)
Below the critical depth, the density of live oil,
lo
, is obtained
from equation (29). Taking into account that the volume of
free gas is given by equation (27) and using (34), the balance
equation is
_
1 +
1
v
_
{exp[E(p) +
p
T]} =
1
v
exp(c
w
p
+
w
T) + + F
oi
/
g
R
G
(1 F)
oi
/
o
. (35)
Equations (33) and (35) relate the pore pressure p to the
converted oil/gas fraction F at the conning pressure p
c
and
temperature T. All these quantities depend on the deposition
time t . Note also that ,
g
, , and R
G
depend on pore pressure
and temperature.
Volume fractions and saturations
The fractions of live oil and water at stage (3), above the
critical depth z
c
(see Figure 1), are equal to the corresponding
volumes divided by the total volume V
s
+V
w
+V
lo
. The volume
of the solid part can be obtained from the initial porosity
i
and the initial pore volume V
pi
, because
i
=V
pi
/(V
pi
+ V
s
);
thus, V
s
=V
pi
(1/
i
1), and by using (13), V
s
/V
oi
=(1 + 1/v)
(1/
i
1). It can be shown easily that the fractions of live oil,
water, and solid are given by
lo
= [(V
s
/V
oi
) +(V
w
/vV
wi
) +]
1
, (36)
w
=
1
v
(V
w
/V
wi
)
lo
, (37)
and
s
= 1
lo
w
, (38)
respectively, where the volume ratios are given by equa-
tion (18).
The total volume below the critical depth is given by
V
s
+V
w
+V
lo
+V
g2
. Using equations (18), (27), and (34), we
obtain the the volume fractions of live oil, water, free gas, and
solid:
lo
= [(V
s
/V
oi
) +(V
w
/vV
wi
) + +]
1
, (39)
w
=
1
v
(V
w
/V
wi
)
lo
, (40)
f g
=

lo
, (41)
and
s
= 1
lo
f g
, (42)
respectively.
The saturations are equal to the corresponding uid volumes
divided by the pore volume or equal to the uid fractions di-
vided by the total porosity , which is given by
=
V
p
V
p
+ V
s
, (43)
where V
s
is the volume of the solid part. Using equation (21)
and the expression for V
s
obtained above, we obtain
=

i
exp[E(p) +
p
T]
1
i
{1 exp[E(p) +
p
T]}
. (44)
Then the free gas, live oil, and water saturations are
S
f g
=
f g
/, S
lo
=
lo
/, and S
w
=
w
/, (45)
respectively.
Let us calculate the saturations and volume fractions when
the gas does not go into solution. As the pore pressure in-
creases from p
i
to p, the pore volume changes from V
pi
to
V
pi
{exp[E(p) +
p
T]} . The saturations are equal tothe cor-
responding volumes divided by the pore volume. Using equa-
tions (13) and (21) gives, for the oil and water saturations,
S
o
=
v
1 +v
(1 F)[exp(c
o
p +
o
T E(p)
p
T)] (46)
and
S
w
=
1
1 +v
[exp(c
w
p +
w
T E(p)
p
T)],
(47)
respectively. The gas saturation is simply
S
g
= 1 S
o
S
w
. (48)
On the other hand, the uid fractions are
o
= S
o
,
w
= S
w
, and
g
= S
g
. (49)
Compressibilities and dry-rock bulk moduli
The isothermal gas bulk modulus K
g
and the gas compress-
ibility c
g
=K
1
g
depend on pressure. The latter can be calcu-
lated from the van der Waals equation as
c
g
=
1
g
p
at constant temperature. It gives
c
g
=
_

g
RT
(1 b
g
)
2
2a
2
g
_
1
. (50)
The pore compressibility c
p
is related closely to the bulk modu-
lus of the dry rock K
m
(the compressibility c
p
is denoted by C
pp
in Zimmerman et al. (1986) and by K
1
p
in Mavko and Mukerji
(1995), and K
m
corresponds to C
1
bc
and K
1
dry
, respectively). Us-
ing the present notation, c
p
can approximately be expressed as
c
p
=
_
1
K
m
1
K
s
_
1
1
K
s
, (51)
where depends on the excess pore pressure. Because dry-
rock wave velocities are practically frequency independent,
the seismic bulk moduli K
m
and
m
versus conning pressure
can be obtained fromlaboratory measurements in dry samples.
If V
P
and V
S
are the experimental compressional and shear
velocities, the moduli are given approximately by
K
m
=(1 )
s
_
V
2
P

4
3
V
2
S
_
,
m
=(1 )
s
V
2
S
, (52)
where
s
is the grain density and is the porosity. We recall
that K
m
is the rock modulus at constant pore pressure, i.e.,
the case when the bulk modulus of the pore uid is negligible
compared with the dry-rock bulk modulus, as, for example, air
at room conditions (Mavko and Mukerji, 1995). Note that gas
at high pressures can have a nonnegligible bulk modulus.
Seismic properties of the rock volume
The seismic velocities of the overpressured porous rock are
computed by using Biots theory of dynamic poroelasticity
(Appendix A). The phase velocities of P- and S-waves are
given by
V
P
=
_
Re
_
1
V
P
__
1
, V
S
=
_
Re
_
1
V
S
__
1
, (53)
where V
P
are the complex velocities of the fast ( +) and slow
( ) waves, V
S
is the complex shear wave velocity, andRe takes
the real part.
EXAMPLE
We assume that the medium is saturated fully with oil and
that before oil/gas conversion occurs, the initial pressure, p
i
,
is hydrostatic. Because there is no water, v , and equa-
tion (22) becomes
exp[E(p) +
p
T] = (1 F)[exp(c
o
p +
o
T)]
+ F
oi
/
g
. (54)
When there is dissolved gas, the balance equations are
exp[E(p) +
p
T] = , z z
c
(55)
and
exp[E(p) +
p
T] = + F
oi
/
g
R
G
(1 F)
oi
/
o
, z >z
c
. (56)
With a surface temperature of 15.6
C, a temperature gradi-
ent G =25
C/km, a sedimentation rate S =0.08 km/m.y., and a

reservoir volume at z
i
=z
1
=2 km, we have t
i
=t
1
=25 m.y.
and T
1
=65.6
C. After 75 m.y., the depth of burial is

z
2
=8 km, t
2
=100 m.y., and T
2
=215.6
C. On the other hand,

if =2.4 g/cm
3
, the conning pressure has increased from
47 MPa to approximately 188 MPa, and the initial pore pres-
sure is p
i
20 MPa (assuming
w
=1 g/cm
3
). If no conversion
takes place, the nal pore pressure would be the hydrostatic
pressure at 8 km, i.e., approximately 78 MPa.
The experimental data for oil-saturated sandstone are avail-
able in Winkler (1985), in his Figures 3 and 4 and Tables 4 and 7.
It is important to understand howthe data were measured. Af-
ter a suitable pore vacuum was achieved, conning pressure
was increased to 40 MPa. Dry-rock wave velocities then were
obtained at successive pressures as the conning pressure de-
creased to 10 MPa. While still in the pressure vessel, the sample
was saturated with oil to a pore pressure of 5 MPa. Differential
pressure p
c
was then increased to 10, 20, and 40 MPa, corre-
sponding to the values (p
c
, p) = (20, 10), (40, 20), and (60, 20)
(in MPa). Note that Winkler calls the differential pressure the
effective stress. The experiments on dry samples correspond to
zero pore pressure. Best-t plots of the dry-rock compressibil-
ity and shear modulus versus conning pressure are
K
1
m
[GPa
1
] = 0.064 +0.122 exp(p
c
[MPa]/6.48)
(57)
and
m
[GPa] = 13.7 8.5 exp(p
c
[MPa]/9.14), (58)
and c
p
in GPa
1
is given by equation (16), with c
p
=0.155,
=0.6, and p
=6.48. The pore compressibility c

p
has been
obtained from equation (51) by assuming that the porosity is
that at hydrostatic pore pressure [this approximation is sup-
ported by experimental data obtained by Domenico (1977)
and Han et al. (1986)]. The best-t plots for c
p
and K
1
m
are
illustrated in Figure 2.
To obtain the moduli for different combinations of the
conning and pore pressures, we should make the substitu-
tion p
c
p
e
= p
c
np, where we assume, following Gangi and
Carlson (1996), that n depends on differential pressure as
n = n
0
n
1
p
d
, n
0
= 1, n
1
= 0.014 MPa
1
.
This dependence of n versus differential pressure is in good
agreement with the experimental values corresponding to
the compressional velocity obtained by Christensen and
Wang (1985) and by Prasad and Manghnani (1997).
Table 1 indicates the properties for Berea sandstone, where
the values correspond to those at the initial (hydrostatic) pore
pressure. The oil and gas viscosities as a function of tempera-
ture and pore pressure are taken fromLuo and Vasseur (1996).
The procedure for computing the phase velocities and atten-
uation factors is the following:
1) Compute the oil/gas conversionfactor as a functionof de-
position time and depth by using equations (4) and (10).
2) Compute the dry-rock bulk moduli (52) from the experi-
mental wave velocities V
P
(p
c
) and V
S
(p
c
). This also gives
the dry-rock bulk moduli as a function of effective pres-
sure. We assume that the porosity is that at hydrostatic
pressure.
3) Compute the initial gas density and the gas density at
pressure p,
g
(p) from the van der Waals equation (11)
and the pore compressibility from equation (51). As be-
fore, we assume that the porosity is that at hydrostatic
pressure.
4) Compute the excess pore pressure corresponding to the
oil/gas conversionfactor F by using equations (54) or (55)
and (56), and the oil and gas saturations by using equa-
tions (45) or (46) and (48). Overpressure occurs when
pore pressure exceeds hydrostatic pressure.
5) The low-frequency wet-rock velocities are computed by
using the porous model (Appendix A), and the dilata-
tional and shear relaxation times are obtained by tting
the experimental high-frequency wet-rock velocities at
full saturation, i.e., at the onset of the oil/gas conversion.
6) Phase velocities and attenuation factors are computed as
Table 1. Material properties for Berea sandstone.
Grain
s
= 2650 kg/m
3
K
s
= 37 GPa
s
= 39 GPa
Fluids
0
= 934 kg/m
3
oi
= 908 kg/m
3
K
o
= 2.16 GPa
o
= 5 10
4
C
1
o
= 800 cP
g
= 0.012 cP
Rock frame
K
m
= 15.04 GPa
m
= 13.16 GPa
= 0.203
= 10
12
m
2
T = 2
p
= 2 10
4
C
1
1 cP = 0.001 Pas.
FIG. 2. Best-t plots of c
p
and K
1
m
obtained from the experi-
mental data for Berea sandstone published by Winkler (1985,
Figures 3 and 4 and Tables 4 and 7).
a function of pore pressure and frequency from equa-
tions (A-15) and (A-16), respectively.
Figure 3 shows the oil/gas conversion factor F [equa-
tion (4)] (a) and its logarithm (b) as a function of depth and
time. The high activation energy requires either a long time or
deep burial, on the order of 4.5 to 5 km, before appreciable
fractions of conversion occur, but signicant fractional conver-
sions occur at 3 km. The pore-pressure buildup with depth is
FIG. 3. Oil/gas conversion factor F [equation (4)] (a) and
log (F) (b) as a function of depth and time.
shown in Figure 4, for all the generated gas out of solution
(continuous line), and for part of the generated gas dissolved
in the oil (broken line). The two cases yield practically the
same result. The pressure increases rapidly for very small frac-
tions of oil converted to gas. This is caused by the expansion
of the pore volume resulting from generation of free gas or
live oil. Beyond the lithostatic pressure, the effective pressure
becomes negative and the pore compressibility increases (see
Figure 2), making the rock highly compliant. This precludes
a rapid increase of the pore pressure, which follows the litho-
static pressure below4 kmdepth. The pore pressure equals the
lithostatic pressure when 2.5% of the oil has been converted
to gas, which occurs at a depth of approximately 4.2 km. In the
case of dissolved gas, the pore pressure equals the lithostatic
pressure without generation of free gas. It is caused by the in-
crease in live-oil volume, since the critical depth is z
c
=5.05 km
and F
c
=20%. If the hydrostatic pressure is used as reference
prole to solve equations (33) and (35), we obtain z
c
=4.77 km
and F
c
=7% and practically the same pore-pressure buildup
shown in Figure 4. Because the results with and without dis-
solvedgas are practically the same, we consider inthe following
analysis the case with no dissolved gas. Figure 5 shows the vari-
ations of the saturations with excess pressure (a) and of the
porosities with depth (b) [(equations (46), (48), and (49), re-
spectively]. The arrows indicate when the pore pressure equals
the lithostatic pressure.
The low-frequency wave velocities versus excess pore pres-
sure and differential pressure are shown in Figure 6. An excess
pore pressure of 60 MPa corresponds to zero differential pres-
sure. The oil/gas conversion starts at a differential pressure of
21.5 MPa, whichcorresponds toanonset time of approximately
FIG. 4. Pore-pressure buildup with depth and deposition time
(continuous line), when all the generated gas is out of solution
(continuous line), and when part of the generated gas is dis-
solved in the oil (broken line). The hydrostatic and lithostatic
pressure are represented by thin lines.
25 million years, when the sandstone is fully saturated with oil.
As can be appreciated, the velocities decrease substantially af-
ter an excess pore pressure of approximately 35 MPa. This re-
sults partly fromthe replacement of oil by gas, but mainly from
the decrease in the dry-rock bulk moduli K
m
and
m
caused by
the decrease in effective pressure.
Winkler (1985) obtained the high-frequency compressional
and shear velocities, 4140 m/s and 2500 m/s, at a differential
pressure of 27.4 MPa. These experimental velocities, corre-
sponding to 400 kHz, are tted with the theoretical veloci-
ties (A-15) to obtain the relaxation times used in equations
(A-10) and (A-11). The pores are saturated with dead oil of
density 890 kg/m
3
, which corresponds to the value reported
by Winkler (1985). Figure 7 shows the normalized compres-
sional (continuous lines) and shear (broken lines) velocities
(a) and attenuations (b) versus frequency. The velocities are
normalized with respect to the low-frequency values 4013 m/s
and 2426 m/s, respectively. The square and circle correspond
FIG. 5. Saturations versus excess pore pressure p p
H
(a) and
porosities versus depth (b). The arrows indicate when the pore
pressure equals the lithostatic pressure.
to the experimental values obtained by Winkler (1985) for the
compressional and shear waves, and the triangle is the P-wave
attenuation for a differential pressure of 40 MPa [attenuation
, indB, is relatedto Q factor as 17.372/(2Q)]. Tot these
values, we use one mechanism for the shear modulus
c
, cen-
tered at 4 kHz and with relaxation times
=4.09 10
5
s and
=3.86 10
5
s (corresponding to a mimimum quality factor
of approximately 34), and a continuous spectrum of mecha-
nisms for M
c
from 1 Hz to 1 MHz and

Q 25. The need for a
continuous spectrum of relaxation mechanisms to explain the
P-wave attenuation values is demonstrated in Winkler (1985),
but because of lack of data for the shear wave, we assume the
simplest model, a single relaxation mechanism.
The Biot relaxation mechanisms, resulting from the relative
motion between the solid and the uids, are important beyond
1 MHz. The peaks at 4 KHz result mainly from the shear vis-
coelastic mechanism.
FIG. 6. Computed low-frequency wave velocities versus excess
pore pressure (a) and differential pressure (b). Frequency is
25 Hz. The broken lines in (b) are the dry-rock wave velocities
derived from equations (57) and (58).
CONCLUSIONS
The variations of P-wave and S-wave velocities have been
determined as a function of excess pressure caused by oil/gas
conversion. The results for a model in which a reservoir vol-
ume is buried at a constant sedimentation rate for a geother-
mal gradient, which is constant in time and depth, show that
the velocities decrease signicantly when only a small amount
(about 2.5%) of the oil in the closed reservoir is converted to
gas. If the gas goes into solution, the results are practically the
same, because the volume increase of the live oil is close to that
of the mixture of dead oil/free gas. Moreover, the differential
pressure vanishes before the oil has absorbed all the possible
generated gas.
FIG. 7. Normalized compressional (continuous line) and shear
(broken line) velocities (a) and attenuations (b) versus fre-
quency, for a differential pressure of 27.4 MPa. The veloci-
ties are normalized with respect to the low-frequency values
4013 m/s and 2426 m/s, respectively. They correspond to the
onset of the oil/gas conversion process. The square and circle
at 400 kHz correspond to the experimental values obtained by
Winkler (1985) for the compressional and shear waves, and the
triangle at 400 kHz is the P-wave attenuation for a differential
pressure of 40 MPa.
That small conversion of oil to gas is sufcient to make the
pore pressure equal to the conning pressure. The large change
in the velocity results mainly from the fact that the dry-rock
moduli are functions of the effective pressure, with the largest
changes occurring at low effective pressures. In fact, porosity
changes and uid moduli and density have a second-order ef-
fect on the velocities. The effective pressure decreases because
the pore pressure increases as a result of conversion of high-
density oil to the low-density gas or low-density live oil, in the
case of dissolved gas. Perceptible changes in velocities occur
when the differential pressure is 15 MPa, at which about 0.6%
of oil has been converted to gas. The changes become signif-
icant when the differential pressure decreases to 10 MPa, at
which about 1% has been converted.
Attenuation also will be a function of effective pressure be-
cause of the changing pore-space geometry. However, exper-
imental data of attenuation for partially saturated rocks as a
function of the effective pressure are not available. Realistic
attenuation values were obtained only for the case with full
oil saturation, for which experimental evidence of the velocity
dispersion is available.
ACKNOWLEDGMENTS
This work was supported by Norsk Hydro a.s. (Bergen) with
funds of the Source Rock project, and by the European
Union, under the project Detection of overpressure zones
with seismic and well data. Thanks go to Hans Helle for im-
portant technical comments.
REFERENCES
Barker, C., 1990, Calculated volume and pressure changes during the
thermal cracking of oil to gas in reservoirs: AAPG Bull., 74, 1254
1261.
Batzle, M., and Wang, Z., 1992, Seismic properties of pore uids: Geo-
physics, 57, 13961408.
Berryman, J. G., Thigpen, L., andChin, R. C. Y., 1988, Bulkelastic wave
propagation in partially saturated porous solids: J. Acoust. Soc. Am.,
84, no. 1, 360373.
Berg, R. R., andGangi, A. F., 1999, Primarymigrationbyoil-generation
microfracturing inlow-permeability source rocks: Applicationtothe
Austin chalk, Texas: AAPG Bull., 83, no. 5, 727756.
Carcione, J. M., 1998, Viscoelastic effective rheologies for modeling
wave propagation in porous media: Geophys. Prosp., 46, 249270.
2000, Amodel for seismic velocityandattenuationinpetroleum
source rocks: Geophysics, 65, 10801092.
Chaney, P. E., 1950, Abnormal pressure and lost circulation: World Oil,
130, 122126.
Christensen, N. I., and Wang, H. F., 1985, The inuence of pore pres-
sure and conning pressure on dynamic elastic properties of Berea
sandstone: Geophysics, 50, 207213.
Dasios, A., Astin, T. R., and McCann, C., 1998a, Compressional-wave
attenuation fromfull waveformsonic data: 60th Ann. Internat. Mtg.,
Europ. Assoc. Geophys. Eng., Expanded Abstracts, 1044.
1998b, Increasing condence in seismic Q measurements: A
comparison of estimates from sonic and surface seismic data: 68th
Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts,
10801083.
Domenico, S. N., 1977, Elastic properties of unconsolidated porous
sand reservoirs: Geophysics, 42, 13391368.
Dutta, N. C., and Levin, F. K., 1990, Geopressure, Geophysical Reprint
Series No. 7, Soc. Expl. Geophys.
Dvorkin, J., Nolen-Hoeksema, R., and Nur, A., 1994, The squirt-ow
mechanism: Macroscopic description: Geophysics, 59, 428438.
Friedman, A. S., 1963, Pressure-volume-temperature relationships of
gases, virial coefcients, in American Institute of Physics handbook:
McGraw-Hill Book Co.
Gangi, A. F., and Carlson, R. L., 1996, An asperity-deformation model
for effective pressure: Tectonophysics, 256, 241251.
Gautschi, W., and Cahill, W. F., 1964, Exponential integral and related
functions, in M. Abramowitz and I. A. Stegun, Eds., Handbook of
mathematical functions with formulas, graphs and mathematical ta-
bles: U. S. Department of Commerce, National Bureau of Standards,
Applied Math. Series 55, S5, 227254.
Hague, P. S., 1981, Measurements of attenuation from vertical seismic
proles: Geophysics, 46, 15481558.
Han, D., Nur, A., and Morgan, D., 1986, Effects of porosity and clay
content onwavevelocities insandstones: Geophysics, 51, 20932107.
Helle, H. B., Inderhaug, O. H., Kovaliev, V. P., Madatov, A.,
and Mitrofanov, G. M., 1993, Complex seismic decomposition
Application to pore pressure prediction, 55th Ann. Internat. Mtg.,
Europ. Assoc. Expl. Geophys., Expanded Abstracts, B005.
Luo, X., and Vasseur, G., 1996, Geopressuring mechanism of or-
ganic matter cracking: Numerical modeling: AAPG Bull., 80, 856
874.
Mann, D. M., and Mackenzie, A. S., 1990, Prediction of pore uid
pressures in sedimentary basins: Marine and Petroleum Geology, 7,
5565.
Mavko, G., and Mukerji, T., 1995, Seismic pore space compressibility
and Gassmanns relation: Geophysics, 60, 17431749.
Prasad, M., and Manghnani, M. H., 1997, Effects of pore and dif-
ferential pressure on compressional wave velocity and quality
factor in Berea and Michigan sandstones: Geophysics, 62, 1163
1176.
Standing, M. B., 1952, Volumetric and phase behavior of oil eld hy-
drocarbon systems: Reinhold.
Winkler, K. W., 1985, Dispersion analysis of velocity and attenuation
in Berea sandstone: J. Geophys. Res., 90, 67936800.
Zimmerman, R. W., Somerton, W. H., and King, M. S., 1986, Com-
pressibility of porous rocks: J. Geophys. Res., 91, 1276512777.
APPENDIX A
SEISMIC PROPERTIES OF A POROUS MEDIUM SATURATED WITH HYDROCARBON AND WATER
Biots theory of dynamic poroelasticity is used to compute
the wave velocities and attenuation factors, where the pore
uid is a mixture of hydrocarbon and water. The complex ve-
locities of the fast ( + sign) and slow ( sign) compressional
waves and shear wave are [see Carcione (1998)],
V
2
P
=
A
_
A
2
4ME
c
2
c
(A-1)
and
V
S
2
=

c
, (A-2)
where
A = M( 2
f
) +
(E +
2
M), (A-3)
c
=
2
f
_
, (A-4)
and
=
T
f

i
2 f
, (A-5)
with and E elastic coefcients, f the frequency, and i =
1.
The sediment density is
= (1 )
s
+
f
where
s
and
f
are the solid and uid densities, respectively; T
is the tortuosity; is the uidviscosity; and is the permeability
of the medium.
The elastic coefcients are given by
E = K
m
+
4
3
, (A-6)
M =
K
2
s
D K
m
, (A-7)
D = K
s
_
1 +
_
K
s
K
1
f
1
__
, (A-8)
= 1
K
m
K
s
, (A-9)
with K
f
the bulk modulus of the mixture hydrocarbon/water.
The stiffness E is the P-wave modulus of the dry skeleton, M
is the elastic coupling modulus between the solid and the uid,
and is the poroelastic coefcient of effective stress.
Wave velocities generally are expected to be lower at low
frequencies, typical of seismic measurements, than at high fre-
quencies, typical of laboratory experiments. Because the mag-
nitude of this effect cannot be described entirely by Biot-
type theories, additional relaxation mechanisms are required
to model the velocity dispersion. Measurements of dry-rock
velocities contain all the information about pore shapes and
pore interactions and their inuence on wave propagation.
Low-frequency wet-rock velocities can be calculated by using
Gassmanns equation, i.e., the low-frequency limit of the dis-
persion relation. High-frequency wet-rock velocities are then
given by the unrelaxed velocities. Because dry-rock velocities
are practically frequency independent, the data can be ob-
tained from laboratory measurements
Viscoelasticity is introduced into the poroelastic equations
for modeling a variety of dissipation mechanisms related to
the skeleton/uid interaction. One of these mechanism is the
squirt-ow (Dvorkin et al., 1994), by which a force applied to
the area of contact between two grains produces a displace-
ment of the surrounding uid in and out of this area. Because
the uid is viscous, the motion is not instantaneous and energy
dissipation occurs. Skeleton/uid mechanisms are modeled by
generalizing M to a frequency-dependent modulus of the form
M
c
= M
_
1 +
2

Q
ln
_
1 +i
2
1 +i
1
__
1
, (A-10)
where
1
and
2
are time constants, with
2
<
1
, and

Q denes
the value of the quality factor, which remains nearly constant
over the selected frequency band. The low-frequency limit of
M
c
is the Biot modulus M.
On the other hand, is generalized to
c
=

L
L
l=1
1 +i
l
1 +i
l
, (A-11)
where
l
and
l
are the relaxation times of the L attenuation
mechanisms.
Themixturehydrocarbon/water behaves as acompositeuid
with properties depending on the constants of the constituents
and their relative concentrations. This problem has been ana-
lyzed by Berryman el al. (1988), and the results are given by
the formulas
K
f
= (S
o
c
o
+ S
w
c
w
)
1
, (A-12)
f
= S
o
o
+ S
w
w
, (A-13)
where
o
is the density of the hydrocarbon,
f
= S
o
o
+ S
w
w
, (A-14)
where
o
and
w
are the viscosities of the hydrocarbon and wa-
ter, respectively. Equation (A-14) is a good approximation for
most values of the saturations.
The phase velocity V
P(S)
is equal to the angular frequency
=2 f divided by the real wavenumber. Then
V
P
=
_
Re
_
1
V
P
__
1
, V
S
=
_
Re
_
1
V
S
__
1
,
(A-15)
where Re takes the real part.
The magnitude of the attenuation factors (in dB) is given by
P
= 17.372
Im
_
V
P
_
Re
_
V
P
_,
S
= 17.372
Im(V
S
)
Re(V
S
)
,
(A-16)
where Im takes the imaginary part.
Biots theory is strictly valid for frequencies below
f
c
=

f
2T
f
. (A-17)
The usefulness of geophone ground-coupling
experiments to seismic data
Guy G. Drijkoningen
ABSTRACT
Ground-coupling devices measure local conditions
around a geophone and are therefore useful for geo-
phone design. For purposes of gathering seismic data,
the data from local ground-coupling experiments must
be related to seismic data. In particular, the geophysicist
wants to know whether a local measurement could help
in detecting a ground-coupling problem or in correcting
the seismic data for it. Two sets of eld experiments have
been carried out to investigate these effects. Our experi-
mental data of coupling-measurement devices showthat
the behavior of well-planted spiked geophones is deter-
mined by shear along the spike, while the behavior of
poorly planted geophones is determined by its weight.
However, when a link is established between the data
from a coupling-measurement device to seismic data,
it becomes clear that the usefulness of this device for
predicting problematic ground-coupling phenomena in
seismic data is very limited.
INTRODUCTION
Ground coupling of geophones is still not a well-understood
problem. Increased interest in single-sensor technology,
three-component technology, four-component ocean-bottom
sensors, and high-resolution seismics requires a better under-
standing of geophone-coupling problems. In this paper, ex-
perimental aspects of geophone coupling are investigated fur-
ther, continuing from reports in the literature. A paper by
Krohn (1984) is the most important in this area. The goal of
this study was to investigate whether and how local ground-
coupling measurements could help in detecting and/or cor-
recting seismic data for ground-coupling problems. To answer
thesequestions, somebasic aspects of couplingrequiredcareful
reinvestigation.
Manuscript received by the Editor February 17, 1999; revised manuscript received November 11, 1999.
Delft University of Technology, Department of AppliedEarthSciences, Mijnbouwstraat 120, 2628 RXDelft, Netherlands. E-mail: g.g.drijkoningen@
ta.tudelft.nl.
A clear distinction is made in this paper between so-called
spike-shear coupling and weight coupling. The former term
refers to the assumed way a spiked geophone is coupled to
the ground and is important for geophone design. In weight
coupling, it is purely the weight of the detecting device that
furnishes the coupling between the geophone and the ground.
This distinction between the two ways of coupling has become
clear after the introduction of a spike in theoretical models by
Tan (1987) and by Rademakers et al. (1996). In this paper, the
focus will be mainly on the distinction between spike-shear and
weight coupling of geophones. This is a common problem that
the practicing geophysicist encounters when a geophone is not
well coupled to the ground.
The paper by Krohn (1984) about experiments on ground
coupling clearly describes the different laboratory and eld
techniques used to determine such coupling. Here, only eld
techniques are described briey and used, with a focus on
how to relate these measurements to seismic data. Measure-
ments have been carried out in different soils, using two types
of coupling-measurement devices. In addition, seismic-shot
records were taken at the same locations. These results have
been analyzed and conclusions have been drawn about the use-
fulness of such measurements to seismic data.
CONCEPTS OF GROUND COUPLING
The rst issue in ground coupling is to dene what geophone
ground coupling means. In the past, some confusion has arisen
because theoretical models did not explain experimental re-
sults about spiked geophones (see, for example, Krohn, 1984).
Thus, theoreticians seemed to have a different meaning for
ground coupling than that used by the practicing geophysicist.
The denitionof geophone groundcoupling usedintheoretical
models (see, for example, Tan, 1987) is as follows: Geophone
ground coupling is the difference between the velocity mea-
sured by the geophone and the velocity of the ground without
the geophone. This denition is useful for the design of geo-
phones so that optimal characteristics can be found. However,
1780
Usefulness of Geophone Coupling Experiments 1781
once an optimal geophone design has been obtained, the prac-
ticing geophysicist must cope with a geophone that is not de-
ployed appropriately. For example, the geophone is not cou-
pled well to its surroundings. In this situation, the above
denition is not appropriate and should be revised as follows:
Bad geophone coupling is the difference between the velocity
as measured by the badly planted geophone and the velocity
as measured by the well-planted geophone.
The theoretical implications of well-coupled and not
well-coupled geophones become clearer when different the-
oretical models are considered. Until the publication of Tans
note (1987), ground coupling was modeled theoretically by
a cylinder resting on a half-space (Lamer, 1970; Hoover and
OBrien, 1980). Although this model is in contradiction to
what has been used for decades in the seismic industry, namely
spiked geophones, it has been used to describe geophone
ground coupling. Krohn (1984) already has mentioned that
the spike should be taken into account in the modeling, but
Tan (1987) was the rst to carry this out. However, Tan (1987)
did not quantify the ground-coupling phenomenon, which was
undertaken by Rademakers et al. (1996). In this latter model,
where the spike is modeled as a cylinder, the nal expression
contains two integrals resulting from the bottom and side of a
cylinder, respectively. The dominant contribution is from the
side. More details can be found in Rademakers (1996). Phys-
ically, we learn from this model that when the geophone is
well coupled, the termwith the shear force along the spike is
dominant. However, whentheforcealongthespikeis removed,
the model no longer holds and the models from Lamer (1970)
and Hoover and OBrien (1980) should be used. In particu-
lar, it is the weight that is dominant, hence the name weight
coupling.
A difference in the response resulting from different models
is caused by the ground coupling showing amplitudes larger
than one around the resonance frequency in previously re-
ported models (Lamer, 1970; Hoover and OBrien, 1980), but
in the model by Rademakers et al. (1996), the ground cou-
pling always has an amplitude less than one. It is hypothesized
that the model of a cylinder resting on a half-space is useful
for weight coupling when the spike of the geophone no longer
provides coupling and is a device that uses its mass to couple
to the ground.
In practice, the problem is associated strongly with the two
types of coupling. Spike-shear coupling generally is well be-
haved for spiked geophones, but weight coupling generally is
not well behaved. Practical examples of both these states
will be given later in this paper. Weight coupling generally
is not well behaved because the contact area between the
geophone (or another sensor) and the ground is rough. Of-
ten, the only way to improve the coupling is to increase the
mass of the geophone so that the contact with the ground be-
comes better (stronger coupling, more regular contact area).
In practice, geophones are kept lightweight because of weight
limits imposed on transporting thousands of geophones. The
requirement for light weight and a good coupling condition
for weight coupling work against each other. We investigated
whether it is possible to detect experimentally whether geo-
phones are spike-shear or weight coupled or, in practical
terms, well or not well coupled. When a badly coupled (that
is, weight-coupled) geophone is detected, the ideal situation
would be to change this weight-coupled geophone to one that
is spike-shear coupled. This question also is addressed in this
paper.
FIELD TECHNIQUES FOR MEASURING GROUND
COUPLING
Different techniques for measuring ground coupling were
described extensively by Krohn (1984). She described labora-
tory (shake-table) and eld tests and compared the two meth-
ods. As she found out, which is also our experience, shake-
table tests show some extra scattering effects resulting from
interference of the soil with the box and other box effects.
Consequently, it was decided to carry out only eld tests. In
our experiments, three methods were deployed, in which one
of the methods was used only to validate another of the three
methods, as described below.
The rst is a standard method available in some seismo-
graphs. In our case, this is a Bison Spectra, which works as fol-
lows: Asignal is generated in the seismograph and sent into the
geophone; the response from the geophone then is recorded.
The second method used a setup similar to that originally de-
ployed by Washburn and Wiley (1941), namely, one geophone
bolted on top of another geophone. In our setup, a cylinder
was constructed between the two geophones, with the mass of
the cylinder equal to the mass of a geophone. The upper geo-
phone is used as an actuator, and the response from the lower
geophone is measured. This method was used only to validate
the response of the third method.
The third method used was a slight modication of the
method described by Hoover and OBrien (1980). A small
weight was dropped from a xed height, and the seismograph
was triggered simultaneously with the release of the weight.
Special care was taken to ensure that the geophone was not
excited outside its linear range. To be sure of the coupling re-
sponse, the second method (double-geophone conguration)
was deployed under the same conditions in the eld as the
weight-drop method. Therefore, the responses from these two
methods should be identical.
EXPERIMENTS
The aims of the experiments were twofold:
1) The rst set of tests (experiment 1) was carried out to
make an inventory of the information one obtains from
the separate measurements, such as coupling conditions
(such as a geophone on its side), or to see whether a
geophone was well coupled to the ground. The separate
measurements were performedmultiple times tobuildup
a statistical basis for computing the mean and variance.
Fromdifferent measurements, it became clear what could
and could not be detected.
2) The second set of tests (experiment 2) was carried out to
investigate the usefulness of ground-coupling measure-
ments in relation to data obtained from seismic shots.
First, the aimwas to see whether the same behavior could
be observed in the measurements of the coupling devices
and in the shot data. If this was the case, then the aim
was to see whether the shot data could be corrected for
this behavior by using the responses from the coupling-
measurement devices.
1782 Drijkoningen
EXPERIMENT 1: LOCAL GROUND-COUPLING
MEASUREMENTS
As mentioned above, these tests were undertaken to see
what could and could not be measured by local ground-
coupling measurements by using the techniques described ear-
lier. Let us rst concentrate on the signal generated by the
seismograph and sent directly into the (receiving) geophone.
An example of this is given in Figure 1. On the four left traces,
the response is measured from well-planted geophones. Note
that the dominant effect in the response is the resonance fre-
quency of the geophone, which is a 10-Hz geophone in this
case. The next four traces in Figure 1 show the response of
badly planted geophones. Bad planting resulted from picking
up a well-planted geophone and letting it drop in the same
hole. Care was taken to create a hole which nearly vertical
(within 5%). Comparing these traces to the responses of the
well-planted geophones, no differences could be observed. De-
termining the average responses and errors from 20 measure-
ments, we nd that the average response of the badly planted
geophone falls within the error bars of the well-planted geo-
phones. Consequently, such a system cannot be used for de-
tecting badly planted geophones.
This method can be used to determine whether a geophone
is planted at all. In the last four traces in Figure 1, the responses
are given from four geophones lying on their side. Note that
the responses are signicantly different fromthose of a planted
FIG. 1. Geophone response to a signal from a Bison seismo-
graph. Traces 14, well planted; traces 58, badlyplanted; traces
912, geophones lying on their side (not planted).
geophone. Such a measurement systemthus can be used for au-
tomatic detection by choosing some appropriate attribute(s).
When the responses of the planted geophones are compared
with those from geophones on their side, a small peak could be
seen in the time responses of the planted geophones at approx-
imately 25 ms. This is caused by an artifact of the geophone,
and separate analysis showed that the different planting of the
geophone did not affect this peak.
To detect and/or correct for well- or badly planted geo-
phones, the other two methods described in the section Field
techniques for measuring ground coupling must be used. We
will describe a series of tests performed on two soils, namely
pure sand, wetted, and pure clay, with some grass on top. The
experiments were repeated 20 times to build up statistical con-
dence. These tests were undertaken only with a small mass
(22.8 g), dropped on the geophone.
In Figure 2, a subset of four responses out of 20 is given for
one geophone being planted well into sandy soil (beach sand),
and also for four responses of badly planted geophones. Some
striking differences can be observed: The energy content (sum
of squared amplitudes) of the badly planted geophone is con-
sistently higher than that of the well-planted geophones; for
the 20 responses, the ratio of the average energies is 2.0. In ad-
dition, the responses of the badly planted geophones contain
more low-frequency energy. The average for the (statistical)
rst moments of the amplitude spectra was 270 Hz for the
well-planted geophone and 163 Hz for the badly planted geo-
phones. This analysis also has been undertaken on 20 responses
obtainedfromthegeophones intheclay. Four of thesearegiven
in Figure 3. The ratio of the average energies is 1.9. The aver-
age of the (statistical) rst moments of the amplitude spectra
FIG. 2. Geophone response to a signal from a small weight
drop for geophones in sand. Traces 14, well planted; traces
58, badly planted.
is now 243 Hz for the well-planted geophones and 151 Hz for
those that were planted badly. These measurements are consis-
tent with the measurements of, for example, Krohn (1984) and
Faber et al. (1994). However, fromour experimental results, we
can claimsomething that these authors did not; that is, only two
states can be discerned. To support this with some statistical
condence, the standard deviation has been determined. The
relative standard deviation (one standard deviation divided by
the average) from the well-coupled geophones in the sand is
8% and for clay is 4%. For the badly planted phones, the de-
viation is 16% for sand and 6% for clay. The higher values of
the standard deviation are well known from experiments by
others. Geophones coupled by their weight need to be made
heavier because the contact area on many types of surfaces
is a serious problem. Another solution is to bury the phones
(Krohn, 1984). Because the amplitude spectra of the well- and
badly planted geophones have their own average and standard
deviation, we can state with statistical condence that there
are only two distinct states. The literature seems to imply
that the coupling could vary continuously, but this was not the
case in our measurements.
Combining these observations with the concepts obtained
from modeling the spike as a cylinder in the ground
(Rademakers et al., 1996), we conclude that well-planted geo-
phones are of the spike-shear coupled type, and badly planted
geophones aretheweight-coupledtype. It shouldbenotedhere
that this observation could be used for making devices that
can determine the coupling locally without using a coupling-
measurement device. Specically, aforceshouldbeappliedthat
would be slightly larger than that necessary to compensate for
the weight of the geophone. Then, if the geophone is not lifted,
FIG. 3. Geophone response toa signal froma small weight drop
for geophones inclay. Traces 14, well planted; traces 58, badly
planted.
it is spike-shear coupled and thus well coupled. However, if the
geophone is lifted, it is weight coupled and thus badly coupled.
Another comparison was performed for the validation of a
new model by Rademakers et al. (1996). Specically, this was
the impedance contrast between the soil and the geophone.
Tan (1987) showed theoretically that the density of the spike
should match the density of the soil to obtain the best coupling.
However, in the model by Rademakers et al. (1996), the effect
of spike density was not dominant. Consequently, two spikes
withdifferent densities (stainless steel andaluminum) were de-
ployed in clay and sand. For each situation, the geophone was
planted well coupled 20 times so that a statistical base could
be gathered. Four from each of these 20 are shown in Figure 4.
The rst eight are the geophones in clay, where the rst four are
with the steel spike and the second four are with the aluminum
spike. Although the weights are different, the responses are
not different statistically. The same conclusion can be drawn
from the responses in the sand. Statistically, there is no differ-
ence. However, when we compare the responses from the clay
with those from the sand, we do see a signicant difference.
The (normalized) energies, with their standard deviation in
parentheses, are: 4.45 (0.47), 4.30 (0.30), 3.26 (0.25), and
3.51 (0.19). This effect also can be seen often in vertical seis-
mic proles (VSPs) where the geophone passes through many
rocks with different characteristics. In conclusion, the effects of
soil variation can be determined with local devices, but the ef-
fect is determined by the characteristics of the soil itself rather
than the difference in characteristic between the soil and the
spike. This conrms the model by Rademakers et al. (1996).
EXPERIMENT 2: LOCAL GROUND-COUPLING
MEASUREMENTS IN RELATION TO SEISMIC-SHOT DATA
In this set of experiments, the usefulness of local techniques
is investigated with respect to their use for seismic-shot data.
As observed in the rst set of experiments, there is a signicant
FIG. 4. Geophone response to a signal from a small weight
drop for geophones in clay and sand, with steel and aluminum
spikes. Traces 14, steel in clay; traces 58, aluminum in clay;
traces 912, steel in sand; traces 1316, aluminum in sand.
1784 Drijkoningen
difference between well- and badly planted geophones in re-
sponses from local techniques (see Figures 2 and 3). In these
experiments, the local techniques are combined with seismic-
shot data to see whether the differences, as seen in the local
measurements, also can be seen in the global (seismic-shot)
measurements. To that end, many experiments have been car-
ried out on types of soils, with four typical soils having been
chosen: peaty soil, sandy soil, gravel, and hard rock. The set
discussed here is only a subset of a larger data set, and the
examples shown are typical.
For each soil type, a set of experiments consisted of a shot
from a small charge located 50 m from a group of single
geophones. The geophones were perpendicular to the shot-
geophone direction so that phase differences would be mini-
mal. Two of the geophones had the weight-drop device on top,
with one well planted and one badly planted. Other geophones
have been used to build up some condence in the results of
the observations. When the data from one shot is analyzed, the
variations seen in the seismic data should then be caused by:
1) The difference in local soil conditions (geology). This
variation is hard to quantify, but a qualitative conclu-
sion can be obtained by looking at the variation of the
responses of the well-planted geophones from the same
seismic shot.
2) The difference in geophone-coupling conditions. This
variation could be determined partly from a comparison
with the other well-planted geophones. It should be real-
ized that coupling could be a 3-Deffect. For bad planting,
the data consisted only of the geophone with a weight-
drop device on top, so that no variation could be deter-
mined in this case.
FIG. 5. Response to a seismic shot and small weight drop for geophones in peaty soil. Left panel: time traces
from seismic shot, well planted (trace 1) and badly planted (trace 2). Upper right panel: amplitude spectra of
traces shown on the left: relative standard deviation (solid thick line), well planted (solid thin line), badly planted
(dotted line). Lower right panel: amplitude spectra of responses to weight drop: well planted (solid line), badly
planted (dotted line).
3) Differences ingeophones. This variationwas testedinthe
laboratory and did not make any signicant contribution
to the total variation.
Let us look at the rst set of data obtained from peaty soil.
This soil was located next to a marsh area near Hannover,
Germany, where measurements were taken in a dried-out por-
tion of the marsh. The different responses are given in Figure 5.
On the left side of the gure, two seismic traces are shown from
the seismic-shot data, where the left one is the well-planted
geophone and the one on the right is the badly planted geo-
phone. One can see the difference in the response resulting
fromthe bad planting: The rst arrival is reverberatory. On the
above right of the gure, the amplitude spectra of these two
traces are given. A difference can be seen between the data at
about 100 Hz. Also shown in this panel is the relative standard
deviation in the amplitude, determined from 14 well-planted
geophones. On the bottom right of the gure, the amplitudes
of the weight-drop responses from these two particular geo-
phones are given. These responses should give the difference
seen in the shot data. In these weight-drop responses, a clear
maximum (resonance) can be seen at about 75 Hz. In addition,
it can be seen that the amplitude of the weight-drop response
for the badly planted geophone has amplitude that is higher
than the amplitude for the well-planted geophone. However,
there are also some differences to be seen between the spectra
of the weight drop and the seismic shot. First, the difference
between the two weight-drop responses is larger than the dif-
ference between the responses to the seismic shot. Moreover,
the weight-drop response for the well-planted geophone still
has a maximum at the same frequency as the response of the
badly planted geophone. This may suggest that the resonance
is purely a soil-related resonance rather than a resonance re-
lated to the coupling condition. In general, it is obvious from
Figure 5 that if we could deconvolve the seismic data of the
bad planting by using the weight-drop data, the result would
be incorrect.
The next set of measurements was carried out on sandy soil,
specically, sand from the beach of Wassenaar, Netherlands,
which was partly wet and fairly compact sand. Again, the dif-
ferent data are plotted next to each other and shown in Fig-
ure 6. The left trace is again from the well-planted geophone,
and the right is from the badly planted geophone. In this case,
there are some minor differences to be seen in the time traces.
This is also the case for the amplitude spectra. Examining the
responses of these geophones to the weight drop, it can be
observed that there is no clear maximum (resonance); the re-
sponses are at. If the maximum is viewed as a resonance, the
badly planted geophone has a slight resonance at about 170 Hz.
In the seismic-shot data, a difference also can be seen around
these frequencies, although it is not very signicant. Again, it
is clear that in this case, if the weight-drop data are used for
correcting the data for bad coupling, the results will not be
correct.
The third set of measurements was carried out in a gravel
deposit near Hannover, Germany. In this area, it was not
easy to plant the geophones well. The data from this area are
shown in Figure 7. The time traces show only minor differ-
ences. However, the amplitude spectra show some clearer dif-
ferences. What is striking in this example is that the amplitude
of the well-planted geophone is higher than the amplitude of
the badly planted geophone. This is probably caused by the
gravel, because it was difcult to establish spike-shear cou-
pling. The responses fromthe weight drop showa resonance at
FIG. 6. Response to seismic shot and small weight drop for geophones in sandy soil. Left panel: time traces from
seismic shot, well planted (trace 1) and badly planted (trace 2). Upper right panel: amplitude spectra of traces
shown on left: relative standard deviation (solid thick line), well planted (solid thin line), badly planted (dotted
line). Lower right panel: amplitude spectra of responses to weight drop: well planted (solid line), badly planted
(dotted line).
about 130 Hz and at frequencies where the seismic data show
the most marked differences. The weight drop indicates that
maximum/resonanceoccurs at approximatelythecorrect range
of frequencies. In addition, the responses of the well-planted
geophone have higher amplitudes than geophones that were
planted badly. Again, if we use the weight-drop data to correct
the seismic data for its plant, the result will be incorrect.
The last set of measurements took place in a limestone mine
near Hannover. The rock outcropped at the surface, and mea-
surements were made by planting the geophones in the cracks
of the limestone. The results are shown in Figure 8. In the
seismic-shot data, a slight difference can be seen between the
responses of the well-coupled and badly coupled geophones.
In the amplitude spectra of the seismic data, the most marked
difference can be seen at about 200 Hz. The amplitude spec-
tra of the weight drop show a maximum (resonance) at about
200 Hz for the well-planted geophone and about 120 Hz for
the badly planted geophone. Therefore, the weight-drop de-
vice predicts a decrease in the resonance frequency for bad
planting, a higher amplitude at the maximum (resonance), but
it does not quantitatively predict the difference in behavior of
the seismic data for good and bad planting. Apossible explana-
tion here could be that the geophones are rocking sideways
because the geophone was planted with only the tip of the
spike into the rock. Because the weight-drop device excites
the geophone only vertically, it does not measure this sideways
movement.
DISCUSSION
One can query why there is no quantitative relation between
the responses of the coupling-measurement device andthe shot
data. The reasons could be:
1786 Drijkoningen
1) Coupling-measurement devices measure coupling con-
ditions and local soil characteristics. It is not possible
to discriminate between the two. The term coupling-
measurement device may be confusing.
2) three-dimensional movement. The coupling-measure-
ment device excites the geophone only in one direction,
but the wave motion fromthe seismic shot is 3-D. Clearly,
this is a restriction of the coupling-measurement device.
A typical problem of nonvertical movement is the rock-
ing (sideways) movement of the geophone. This occurs
when spiked geophones are planted in hard rock. For un-
consolidatedsoils, Krohn(1984) partly solvedthe rocking
problemby recommending that the geophones be buried.
However, 3-D movement will remain. It may be worth-
while to design geophones with lower cross-sensitivity at
the cost of performance in the vertical direction.
3) Although theoretically there is a relation between re-
ection and transmission, it is often difcult to obtain
transmission data from reection data. Washburn and
Wiley (1941) gave a formula that would accomplish this
for the weight-drop device, but it turned out not to be
very useful when applied to seismic-shot data. It may be
better to measure the transmission of the local geophone-
coupling conditions rather than to determine it from re-
ection measurements.
CONCLUSIONS
Local ground-coupling-measurement devices, suchas asmall
weight-drop device, measure directly the most important char-
acteristics of ground coupling. In particular, they show the
FIG. 7. Response to seismic shot and small weight drop for geophones in gravel. Left panel: time traces from
seismic shot, well planted (trace 1) and badly planted (trace 2). Upper right panel has amplitude spectra of traces
shown on the left: relative standard deviation (solid thick line), well planted (solid thin line), and badly planted
(dotted line). Lower right panel has amplitude spectra of responses to weight drop: well planted (solid line),
badly planted (dotted line).
amount of energy transferred, the maxima (resonances), and
the shifts in maxima (resonances). Fromthe measurements dis-
cussedinthis paper, it has become clear what canandcannot be
detected. Inaddition, it was shownbyusingtheweight-dropde-
vice that well-coupledgeophones are the so-calledspike-shear-
coupled geophones, while badly coupled spiked geophones are
the weight-coupled type. One cannot speak of a continuous
variation of coupling conditions, but mainly, two states can
be discerned: spike-shear or weight coupled.
However, when relating the coupling responses to seismic
data, it has become clear that the responses of the coupling-
measurement devices can be used only qualitatively. When it
was used quantitatively, the following problems arose:
1) The maxima (resonances) in the amplitude spectra of the
coupling-measurement device often do not coincide with
the maximum difference seen in the seismic data.
2) A maximum in the response of the coupling-measure-
ment device does not mean it is a resonance of the cou-
pling condition. It also may be a feature that is purely a
characteristic of the topsoil.
3) The differences observed in the responses of the coupling-
measurement device do not explain quantitatively the
differences seen in the seismic data.
In general, one can say that coupling-measurement devices
are very suitable for small eld-laboratory tests, as under-
taken in experiment 1 of this paper. We can learn from such
experiments, for example, that there are two distinct states for
good and bad coupling. However, quantitative use of such de-
vices for seismic data, such as correcting for bad coupling con-
ditions, is very limited.
FIG. 8. Response to seismic shot and small weight drop for geophones in limestone. Left panel: time traces from
seismic shot, well planted (trace 1) and badly planted (trace 2). Upper right panel has amplitude spectra of traces
shown on the left: relative standard deviation (solid thick line), well planted (solid thin line), and badly planted
(dotted line). Lower right panel has amplitude spectra of responses to weight drop: well planted (solid line),
badly planted (dotted line).
ACKNOWLEDGMENTS
The author would like to thank Schlumberger for its co-
operation in this work, specically Dr. Andreas Laake and
Dr. Wilfried Junge. In addition, my colleagues at Delft Univer-
sity of Technology are acknowledged, specically Leode Groot
and Bart Cremers. This work was supported partly by the EC
Thermie Programme (project no. OG150/94DE/UK/NL).
REFERENCES
Faber, K., Maxwell, P. W., and Edelmann, H. A. K., 1994, Recording
reliability in seismic exploration as inuenced by geophone-ground
coupling: 56th Ann. Mtg., Eur. Assn. Expl. Geophys, Expanded
Abstracts B014.
Hoover, G. M., and OBrien, J. T., 1980, The inuence of the planted
geophone on seismic land data: Geophysics, 45, 12391253.
Krohn, C. E., 1984, Geophone ground coupling: Geophysics, 49, 722
731.
Lamer, A., 1970, Couplage sol-geophone: Geophys. Prosp., 18, 300
319.
Rademakers, F., 1996, Geophone ground coupling: New elastic ap-
proach: M.Sc. thesis: Catholic Univ. Leuven and Delft Univ.
Tech.
Rademakers, F., Drijkoningen, G. G., and Fokkema, J. T., 1996, Geo-
phone ground coupling: New elastic approach: 66th Ann. Internat.
Mtg., Soc. Expl. Geophys., Expanded Abstracts, paper 4.6.
Tan, T. H., 1987, Reciprocity theoremapplied to the geophone-ground
coupling problem: Geophysics, 52, 17151717.
Washburn, H., and Wiley, H., 1941, Effect of the placement of a
seismometer on its response characteristic: Geophysics, 6, 116
131.
Estimation of fracture parameters from reection seismic dataPart I:
HTI model due to a single fracture set
Andrey Bakulin
, Vladimir Grechka
, and Ilya Tsvankin
ABSTRACT
The simplest effective model of a formation contain-
ing a single fracture system is transversely isotropic with
a horizontal symmetry axis (HTI). Reection seismic
signatures in HTI media, such as NMO velocity and
amplitude variation with offset (AVO) gradient, can be
conveniently describedby the Thomsen-type anisotropic
parameters
(V)
,
(V)
, and
(V)
. Here, we use the lin-
ear slip theory of Schoenberg and co-workers and the
models developed by Hudson and Thomsen for penny-
shaped cracks to relate the anisotropic parameters to the
physical properties of the fracture network and to devise
fracture characterization procedures based on surface
seismic measurements.
Concise expressions for
(V)
,
(V)
, and
(V)
, linearized
in the crack density, show a substantial difference be-
tween the values of the anisotropic parameters for
isolated uid-lled and dry (gas-lled) penny-shaped
cracks. While the dry-crack model is close to ellipti-
cal with
(V)
(V)
, for thin uid-lled cracks
(V)
0
and the absolute value of
(V)
for typical V
S
/V
P
ratios
in the background is close to the crack density. The
parameters
(V)
and
(V)
for models with partial satu-
ration or hydraulically connected cracks and pores al-
ways lie between the values for dry and isolated uid-
lled cracks. We also demonstrate that all possible pairs
of
(V)
and
(V)
occupy a relatively narrow triangu-
lar area in the [
(V)
,
(V)
]-plane, which can be used to
identify the fracture-induced HTI model from seismic
data.
The parameter
(V)
, along with the fracture orienta-
tion, can be obtained from the P-wave NMO ellipse for
a horizontal reector. Given
(V)
, the NMO velocity of a
dipping event or nonhyperbolic moveout canbe inverted
for
(V)
. The remaining anisotropic coefcient,
(V)
, can
be determined from the constraint on the parameters
of vertically fractured HTI media if an estimate of the
V
S
/V
P
ratio is available. Alternatively, it is possible to
nd
(V)
by combining the NMO ellipse for horizontal
events with the azimuthal variation of the P-wave AVO
gradient. Also, we present a concise approximation for
the AVO gradient of converted (PS) modes and show
that all three relevant anisotropic coefcients of HTI
media can be determined by the joint inversion of the
AVO gradients or NMO velocities of P- and PS-waves.
For purposes of evaluating the properties of the frac-
tures, it is convenient to recalculate the anisotropic
coefcients into the normal (
N
) and tangential (
T
)
weaknesses of the fracture system. If the HTI model
results from penny-shaped cracks,
T
gives a direct
estimate of the crack density and the ratio
N
/
T
is
a sensitive indicator of uid saturation. However, while
there is a substantial difference between the values of
N
/
T
for isolated uid-lled cracks and dry cracks, in-
terpretation of intermediate values of
N
/
T
for porous
rocks requires accounting for the hydraulic interaction
between cracks and pores.
INTRODUCTION
Seismic detection of subsurface fractures has important ap-
plications in characterization of naturally fractured reservoirs.
It is well known that preferential orientation of fracture net-
works makes the medium azimuthally anisotropic with re-
Manuscript received by the Editor April 16, 1999; revised manuscript received February 28, 2000.
Formerly St. Petersburg State University, Department of Geophysics, St. Petersburg, Russia. Presently Schlumberger Cambridge Research, High
Cross, Madingley Road, Cambridge CB 3 0EL, England. E-mail: bakulin@cambridge.scr.slb.com.
Colorado School of Mines, Center for Wave Phenomena, Department of Geophysics, Golden, Colorado 80401-1887. E-mail: vgrechka@dix.mines.
edu; ilya@dix.mines.edu.
c _2000 Society of Exploration Geophysicists. All rights reserved.
spect to seismic wave propagation. Although the presence of
azimuthal anisotropy has a strong inuence on all propaga-
tion modes (i.e., P-, S-, and converted waves), most exist-
ing studies concentrate on analysis of time delays or reec-
tion amplitudes of split shear waves at near-vertical incidence.
1788
Fracture Characterization for HTI Media 1789
While such measurements make it possible to map the orien-
tation and intensity (or crack density) of a vertical fracture
set, they are not sensitive to the fracture inll (content), un-
less the fractures are corrugated (see part III of this series).
Recently, it was demonstrated that the azimuthal dependence
of P-wave signatures has the potential of not only constrain-
ing the fracture orientation (e.g., Corrigan et al., 1996) and
density (Tsvankin, 1997) but also of discriminating between
dry (gas-lled) and uid-lled fractures (Sayers and Rickett,
1997; R uger and Tsvankin, 1997). Further progress in seismic
methods of fracture detection, however, requires a better un-
derstanding of the relationshipbetweenthe anisotropic param-
eters measured from seismic data and the physical properties
of fracture systems.
This is therst of threepapers inwhichwediscuss anisotropic
seismic signatures associated with vertical fracture sets. The
transversely isotropic medium with a horizontal axis of sym-
metry (HTI), addressed here, is the simplest anisotropic model
caused by parallel rotationally invariant vertical fractures in
isotropic host (background) rock. In the two sequel papers we
study more complicated orthorhombic and monoclinic mod-
els needed to describe multiple fracture sets and/or formations
with an anisotropic background. Among the main questions
we will attempt to answer are these:
1) Do different theories of seismic wave propagation in
fractured media (e.g., those by Hudson, 1980, 1981;
Schoenberg, 1980, 1983; Thomsen, 1995; Hudson et al.,
1996) lead to the same effective anisotropic model?
2) What is the relationship between the anisotropic parame-
ters that control seismic signatures and the physical prop-
erties of the fractures?
3) What types of seismic data are needed to obtain informa-
tion about various properties of fracture systems?
4) What characteristics of fractured formations (e.g., crack
density, fracture azimuth, equant porosity, type of inll)
can be evaluated using surface reection data?
While our ultimate goal is to invert seismic data for the
physical characteristics of fracture networks, the data are con-
trolled by the elastic parameters (e.g., by the stiffness tensor)
of the effective anisotropic medium. The effective parameters
depend not only on the orientation, compliances, etc., of the
fractures but also on the elastic properties of the (possibly
anisotropic) host rock. Taken together, the characteristics of
the fractures and host rock form a set of so-called microstruc-
tural parameters. Even though seismic data may provide suf-
cient information to recover the effective anisotropic model, it
may be impossible to obtain the microstructural parameters in-
dividually. Indeed, with increasing complexity of the fractured
model (e.g., in the presence of multiple dipping fracture sets),
the number of microstructural parameters can become arbi-
trarily large, whereas the maximum number of the effective
anisotropic parameters is just 21 for the most general media
of triclinic symmetry. Still, we will show that, for a range of
fracture models, seismic data can be inverted for some par-
ticular microstructural parameters responsible for practically
important properties of the fractured medium.
To identify the fracture parameters constrained by seismic
data, it is useful to compare the results of different approaches
to effective medium theory. Effective models of fractured me-
dia discussed below include those based on parallel innite
fractures with linear slip boundary conditions (Schoenberg,
1980, 1983), isolatedparallel penny-shapedcracks that havethe
form of oblate spheroids (Hudson, 1980, 1981), and partially
saturated penny-shaped cracks or hydraulically connected
cracks and pores (Hudson, 1988; Thomsen, 1995; Hudson et al.,
1996). An important result in establishing the relationships be-
tweendifferent theories is obtainedby Schoenberg andDouma
(1988), who prove that innite parallel fractures with linear slip
(Schoenberg, 1980, 1983) and penny-shaped cracks (Hudson,
1980, 1981) yield the same structure of the effective stiffness
tensor (i.e., it is impossible to distinguish between the two mod-
els just by inspecting the stiffnesses). This is a clear indication
that seismic data are not sensitive to one of the microstruc-
tural parameters: the shape of the fractures. Further general-
izing this result, we demonstrate that the presence of equant
porosity with hydraulic connection to the cracks and/or partial
saturation does not change the structure of the stiffness tensor.
The values of the anisotropic coefcients in this case turn out
to be intermediate between those for isolated uid-lled and
dry cracks.
Reection seismic signatures in anisotropic media are
most concisely described by Thomsen-type dimensionless
anisotropic coefcients, such as the parameters , , and orig-
inally introduced by Thomsen (1986) for transverse isotropy
with a vertical symmetry axis (VTI media). These coefcients
capture the inuence of anisotropy on various seismic signa-
tures and therefore can be determined fromseismic data. Most
existing work on effective parameters of fractured media (e.g.,
Hudson, 1981; Schoenberg, 1983), however, is formulated in
terms of the stiffnesses c
i j
or compliances s
i j
. Better under-
standing of the inuence of cracks on reection seismic signa-
tures requires expressing these results through Thomsen-type
anisotropic coefcients. For the VTI medium due to a single
system of horizontal fractures, Schoenberg and Douma (1988)
obtain the weak-anisotropy approximations for Thomsens
(1986) anisotropic parameters andnote that only twoout of the
three coefcients (, , ) are independent. Thomsen (1995)
presents more general expressions for , , and of a simi-
lar VTI model that contains horizontal fractures hydraulically
connected to pore space.
For HTI media resulting fromvertical, rotationally invariant
fractures, reection traveltimes and amplitudes are convenient
toexpress throughthe Thomsen-type parameters
(V)
,
(V)
, and
(V)
described by R uger (1997) and Tsvankin (1997). [Note
that
(V)
,
(V)
, and
(V)
are different from the generic Thomsen
parameters dened with respect to the horizontal symmetry
axis.] Approximate expressions for
(V)
and
(V)
are derived by
Sayers andRickett (1997), whouse themtostudy the azimuthal
variation of the P-wave amplitude variation with offset (AVO)
response. Here, wegiveasystematic analysis of thedependence
of the anisotropic coefcients of HTI media on the physical
properties of vertical fractures.
After a brief discussion of geological data on fractures, we
review the linear slip theory following the work of Schoenberg
(1983), Schoenberg and Douma (1988), and Schoenberg and
Sayers (1995). This theory, based on physically intuitive rela-
tions between stress and discontinuity in displacement, is for-
mulated in terms of the fracture compliances or weaknesses
and requires no assumptions about the microstructure and
microgeometry of fractures. Then we obtain the anisotropic
coefcients of the effective HTI solid as functions of the
1790 Bakulin et al.
fracture weaknesses and determine the bounds of
(V)
and
(V)
for arbitrary strength of fracturing. For models with penny-
shaped (spheroidal) cracks, the anisotropic coefcients and
weaknesses areexpressedthroughthemicrostructural parame-
ters using the theories of Hudson(Hudson, 1981, 1988; Hudson
et al., 1996) and Thomsen (1995). Extending the results of
Tsvankin (1997) and Contreras et al. (1999), we show that a
complete fracture characterization procedure (including eval-
uation of the uid content) can be based on (1) conventional
P-wave data recorded in wide-azimuth 3-D surveys or (2) the
combination of P-waves and converted PS-waves.
EFFECTIVE MODELS OF FRACTURED MEDIA
Geological background
A concise overview of the properties of natural fracture sys-
tems can be found, for example, in Aguilera (1998). Subsurface
fractures usually occur in large populations or sets with simi-
lar orientations. Fracture openings (apertures) may vary from
very thin (0.0010.01 mm) to relatively wide (0.10.5 mm), but
it is believed that in-situ values do not deviate substantially
from an average of 0.02 to 0.03 mm (Romm, 1985). Aguilera
(1998), however, mentions that in some reservoirs where frac-
tures are propped by partial mineralization, the apertures may
reach 1 inch or more.
An important parameter of a fracture set is the distance be-
tween the adjacent fractures, which is much greater than the
fracture opening. Conventionally, geologists describe fractured
rocks by the so-called fracture intensity,
fracture intensity =
number of fractures
meter
,
with the number of fractures counted in the direction per-
pendicular to the fracture planes. Fracture intensity for very
sparse sets is less than 0.75 m
1
, while for tight sets it may
exceed 10 m
1
(Bagrintseva, 1982). Typical values of fracture
intensity for carbonate reservoirs are estimated as 120 m
1
(Kirkinskaya and Smekhov, 1981).
Since seismic wavelengths are on the order of tens and hun-
dreds of meters, it is clear that
seismic wavelengthfracture spacingfracture opening.
Therefore, in building effective seismological models we can
neglect nite fracture openings and details of the spatial dis-
tributions of fractures and can consider fractured blocks as
equivalent or effective anisotropic solids (i.e., use the long-
wavelength approximation). The parameters of such an effec-
tive model will depend on the orientation and intensity of the
fracture set(s) and the properties of the material lling the
fractures, as well as on the elastic coefcients of the host rock.
Parallel fractures: The linear slip model
To obtain the effective parameters of fractured media,
Schoenberg (1980, 1983) suggests treating the fractures, re-
gardless of their shape and microstructure, as either innitely
thin and highly compliant (soft) layers or planes of weakness
with linear-slip (nonwelded, see below) boundary conditions.
While these two representations are equivalent in the long-
wavelength limit, each is useful in developing effective models
of fractured rock and gaining insight into the physical meaning
of the fracture parameters.
The most straightforward way to derive the parameters of
theeffectivemediumis toapplytheapproachbasedonthethin-
layer model. The exact Backus (1962) averaging procedure for
parallel thin, soft layers embedded in an isotropic matrix leads
to the following simple formof the effective compliance matrix
s (the inverse of the stiffness matrix c):
s = s
b
s
f
, (1)
where s
b
is the compliance matrix of the host rock and s
f
is the
excess compliance associated with the layers (Schoenberg and
Muir, 1989; Molotkov and Bakulin, 1997). Although it would
be more natural to denote stiffness by s and compliance by
c, the notation used here is widely cited in the literature on
elasticity and wave propagation.
It may be proved by means of the reectivity (matrix)
method that for vertical layers orthogonal to the x
1
-axis, s
f
is given by
s
f
=
_
_
_
_
_
_
_
_
_
_
_
s
11 f
0 0 0 s
15 f
s
16 f
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
s
15 f
0 0 0 s
55 f
s
56 f
s
16 f
0 0 0 s
56 f
s
66 f
_
_
_
_
_
_
_
_
_
_
_
. (2)
Here, is the fraction of the total volume occupied by the thin
layers and s
i j
f
are the compliances of the layers material. As
follows from equations (1) and (2), only six compliance ele-
ments contribute to the effective parameters. Although is
assumed to be small, all compliances s
i j
f
are large (i.e., the
material is soft) and the products s
i j
f
are nite. The general
condition of the mediums stability also requires that the 3 3
submatrix composed of the nonzero elements of the compli-
ance matrix (2) be nonnegative denite.
Equation (2) can also be used to describe a set of parallel
fractures of innite extent withthe fracture normal nparallel to
the x
1
-axis. Following Schoenberg (1980), we treat fractures as
planes of weakness with nonwelded boundary conditions. For
the purpose of deriving the effective parameters, the medium
containing parallel planes of weakness is equivalent tothe thin-
layer model discussed above. Therefore, the matrix of the ex-
cess fracture compliances can be written as
s
f
=
_
_
_
_
_
_
_
_
_
_
_
K
N
0 0 0 K
NV
K
NH
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
K
NV
0 0 0 K
V
K
V H
K
NH
0 0 0 K
V H
K
H
_
_
_
_
_
_
_
_
_
_
_
. (3)
The jumps in the stress tensor (
i j
) and displacement vector
(u) across a plane of weakness satisfy the boundary condi-
tions of linear slip (Schoenberg, 1980; Molotkov and Bakulin,
1997):
[
11
] = [
12
] = [
13
] = 0,
[u
1
] = h(K
N
11
K
NH
12
K
NV
13
),
[u
2
] = h(K
NH
11
K
H
12
K
V H
13
),
[u
3
] = h(K
NV
11
K
V H
12
K
V
13
), (4)
where h is the average distance (spacing) between the fractures
and the brackets denote the jump of the corresponding value
across the interface (plane of weakness).
Equations (4) help explain the physical meaning of the com-
pliances K
N
, K
V
, K
H
, K
NV
, K
NH
, and K
V H
. For instance, K
N
is
the normal fracture compliance relating the jump of the nor-
mal (to the fractures) displacement u
1
to the normal stress
11
.
Likewise, K
V
and K
H
are the two shear compliances along
the vertical (x
3
) and horizontal (x
2
) directions. The compliance
K
NV
is the coupling factor between the jump of the normal
displacement u
1
and the shear stress
13
or, equivalently, be-
tween u
3
and
11
. Such a coupling may be caused by a slight
roughness (corrugation) of the fracture surfaces, with peaks
and troughs somewhat offset from one side of the fracture to
the other (Schoenberg and Douma, 1988).
From the physical point of view, a nonzero K
NV
implies that
the normal and shear slips with respect to the fracture plane
(u
1
and u
3
) are coupled. Since a nonzero element K
NV
in ma-
trix (3) yields a nonzero compliance s
15 f
in equation (2), a
fracture set with K
NV
,=0, K
NH
=K
V H
=0 produces a medium
of monoclinic symmetry, and is sometimes called monoclinic
(Schoenberg and Douma, 1988). The same type of coupling is
associated with nonzero values of K
NH
and K
V H
. However, as
shown by Berg et al. (1991), K
V H
(unlike K
NV
and K
NH
) can
always be eliminated by a proper rotation of the coordinate
frame around the direction normal to the fractures.
c = s
1
= c
b
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
( 2)
N

N

N
0 0 0
2
2
2
2
N
0 0 0
2
2
2
2
N
0 0 0
0 0 0 0 0 0
0 0 0 0
T
0
0 0 0 0 0
T
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
, (9)
Using largely physical (rather than rigorous mathematical)
arguments, the above results are generalized by Nichols et al.
(1989) for multiple sets of fractures of innite extent and by
Schoenberg and Sayers (1995) for thin fractures of arbitrary
shape and nite dimensions. In particular, the linear addition
of the fracture compliances to the compliance of the host rock
[equation (1)] is applied by these authors to noninteracting
multiple fracture sets.
Hereafter, we consider the simplest formof the matrix of the
excess fracture compliance by assuming no coupling between
the slips along the coordinate directions and a purely isotropic
microstructure of the fracture surfaces. For fracture sets with
these properties [called rotationally invariant by Schoenberg
and Sayers (1995)], the elements of the compliance matrix sat-
isfy the following relationships (Hsu and Schoenberg, 1993):
K
NV
= K
NH
= K
V H
= 0, K
V
= K
H
(5)
As a result, the matrix of fracture compliances (3) reduces to
s
f
=
_
_
_
_
_
_
_
_
_
_
_
K
N
0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 K
T
0
0 0 0 0 0 K
T
_
_
_
_
_
_
_
_
_
_
_
, (6)
where K
V
=K
H
is denoted by K
T
. The values K
N
and K
T
are
nonnegative and have the physical meaning of the normal and
tangential compliances added by the fractures to the host rock.
For a purely isotropic background model characterized by the
Lam e constants and , Hsu and Schoenberg (1993) introduce
dimensionless quantities
N
=
( 2)K
N
1 ( 2)K
N
, (7)
T
=
K
T
1 K
T
. (8)
From this denition it is clear that both
N
and
T
, which
we call the normal and tangential weaknesses (Bakulin and
Molotkov, 1998), vary from 0 to 1.
Using equations (1) and (6) and inverting the compliance
matrix yields the stiffness matrix of the effective fractured
medium (Schoenberg and Sayers, 1995):
wherec
b
=s
1
b
is thestiffness matrixof the(isotropic) host rock.
If both weaknesses (
N
and
T
) are equal to zero, the medium
contains no fractures, while the weaknesses approaching unity
correspond to the extreme degree of fracturing. Equation (9)
shows that
N
=1(c
11
=0) implies thevanishing P-waveveloc-
ity in the direction normal to the fractures; likewise, for
T
=1
the S-wave velocity across the fractures goes to zero.
Fromequation(9) it canbe inferredthat the stiffnesses of the
effectivemediumarerelatedbyc
22
=c
33
, c
12
=c
13
, c
55
=c
66
, and
c
23
=c
33
2c
44
(note that c
44
is not inuenced by the fractures).
Hence, the effective medium is transversely isotropic with a
horizontal symmetry axis (HTI medium). Although general
HTI models are described by ve independent parameters (c
11
,
c
33
, c
13
, c
44
, and c
55
), the stiffness matrix from equation (9)
1792 Bakulin et al.
depends on just four quantities: and of the host rock and
the dimensionless fracture weaknesses
N
and
T
. Therefore,
there exists a relationship (constraint) between the stiffnesses
that can be written as (Schoenberg and Sayers, 1995)
c
11
c
33
c
2
13
= 2c
44
(c
11
c
13
). (10)
Aligned penny-shaped cracks
One of the simplest models of fractured rock contains a sin-
gle set of parallel penny-shaped cracks (i.e., the cracks have
the form of oblate spheroids) embedded in an isotropic solid.
If the semimajor axis of the spheroid is denoted by a and the
semiminor axis by c, we can introduce the so-called aspect ratio
c/a, which is much smaller than unity for thin cracks. If the
cracks are spheroidal, it is convenient to replace the fracture
(crack) intensity by a similar parameter called the crack den-
sity, dened as e =a
3
), where is the number of cracks per
unit volume and ) denotes volume averaging (Hudson, 1980).
Considering thin, penny-shaped cracks orthogonal to the x
1
-
axis, Hudson (1980, 1981) derives the following expression for
the stiffness matrix c of the effective medium:
c = c
b
_
_
_
_
_
_
_
_
_
_
_
( 2)
2
U
11
( 2)U
11
( 2)U
11
0 0 0
( 2)U
11

2
U
11

2
U
11
0 0 0
( 2)U
11

2
U
11

2
U
11
0 0 0
0 0 0 0 0 0
0 0 0 0
2
U
33
0
0 0 0 0 0
2
U
33
_
_
_
_
_
_
_
_
_
_
_
O(e
2
), (11)
where U
11
and U
33
are dimensionless quantities that depend
on the boundary conditions on the crack faces, inll param-
eters, possible interaction of cracks, and some other factors.
Quadratic and higher order terms in crack density e in equa-
tion (11) are ignored. Although the underlying assumptions of
Hudsons theory differ from those of the linear slip theory, the
effective medium in both cases has the same symmetry (HTI)
with the same constraint (10) on the stiffness components.
Comparison of Hudsons and linear-slip models
Schoenberg and Douma (1988) note an even more profound
similarity between Hudsons and linear-slip models that goes
beyond relationship (10). They point out that matrices (9)
and (11) have the same structure and become identical if the
fracture weaknesses satisfy the following relations:
N
=
( 2)
U
11
e, (12)
T
= U
33
e. (13)
Equations (12) and(13) canbeusedtoobtainexplicit expres-
sions for the weaknesses of dry and uid-lled cracks. Suppose
the fractures are lled with a weak solid with bulk modulus
k
/
and shear modulus
/
(both moduli may be equal to zero).
Substituting the expressions for U
11
and U
33
given by Hudson
(1981) into equations (12) and (13) yields (Schoenberg and
Douma, 1988)
N
=
4e
3g(1 g)
_
1
1
g(1 g)
_
k
/
4/3
/
__
a
c
__,
(14)
T
=
16e
3(3 2g)
_
1
4
(3 2g)
_
__
a
c
__. (15)
The parameter g is dened as
g

2
=
V
2
S
V
2
P
, (16)
where V
P
and V
S
are the P- and S-wave velocities in the back-
ground medium.
For dry (gas-lled) cracks both moduli of the inll material
vanish (k
/
=
/
=0), giving
N
=
4e
3g(1 g)
, (17)
T
=
16e
3(3 2g)
. (18)
If the cracks are lled with uid, the shear modulus is
equal to zero (
/
=0), but the bulk modulus k
/
for water or
oil may be comparable to the shear rigidity of the host
rock. Hence, for thin cracks with a small aspect ratio c/a,
[(k
/
4/3
/
)/](a/c) 1 and
N
goes to zero. However,
T
remains the same as for dry cracks. Therefore, for uid-lled
cracks
N
= 0, (19)
T
=
16e
3(3 2g)
. (20)
Schoenberg andSayers (1995) suggest using the ratio K
N
/K
T
as an indicator of uid content. They note from existing ex-
periments (Pyrak-Nolte et al., 1990a, b; Hsu and Schoenberg,
1993) that this ratio is close to unity for dry cracks and almost
vanishes for uid-lled ones. To conrm this observation, they
infer from Hudsons (1981) theory that for dry cracks
K
N
/K
T
1 /2, (21)
where is Poissons ratio for the host rock.
The exact ratio K
N
/K
T
can be obtained in terms of the weak-
nesses from equations (7), (8), and (16):
K
N
K
T
= g
N
(1
T
)
T
(1
N
)
. (22)
For uid-lled cracks K
N
/K
T
=0 because the weakness
N
goes to zero.
In the limit of small weaknesses (
N
_1 and
T
_1), equa-
tion (22) reduces to
K
N
K
T
g
T
. (23)
For dry cracks, we substitute the weaknesses from equa-
tions (17) and (18) into equation (23) to nd
K
N
K
T
3 2g
4(1 g)
= 1 /2. (24)
Thus, equation (21) is a linearized approximation valid for
small fracture weaknesses (i.e., small crack densities). The
K
N
/K
T
ratio from equation (24) varies between 0.75 and 1
for the whole range of g from 0 to 0.5, which seems to conrm
the observation of Schoenberg and Sayers (1995).
Calculations based on the exact equation (22), however, may
produce K
N
/K
T
ratios substantially exceeding unity, evenif the
crack density is small (Figure 1). The discrepancy between the
exact and weak-anisotropy results is explained by the inuence
of the weakness
N
in the denominator of equation (22). Ac-
cording to Hudsons (1981) theory, for dry cracks the normal
weakness
N
is larger than the shear weakness
T
and the
term 1
N
may be far smaller than unity. It is still unclear
whether Hudsons theory, which produces the large values of
K
N
/K
T
for e =0.05 0.07, remains accurate for these crack
densities (Assad et al., 1993) because none of the existing ex-
perimental measurements (Pyrak-Nolte et al., 1990a,b; Hsu
and Schoenberg, 1993) report values of K
N
signicantly higher
than those of K
T
.
FIG. 1. Ratio of the normal-to-shear fracture compliances in
Hudsons model of dry penny-shaped cracks in isotropic host
rock. Solid linesthe exact K
N
/K
T
ratio for different crack
densities e (marked on the plot). Dashed linelinearized ap-
proximation for K
N
/K
T
from equation (24).
Thomsens model of fractured porous media
The weaknesses
N
and
T
of subsurface fractured rocks
may differ substantially from the above estimates obtained for
the simplied model of dry or uid-lled isolated cracks [equa-
tions (17)(20)]. On the one hand, fractures may be stiffened
by the cementation of crack (fracture) faces or by the rigidity
of the fracture inll. Hudsons theory accounts for these factors
by allowing for a nonzero ratio
/
/in equations (14) and (15).
On the other hand, uid-lled fractures may be weakened by
partial saturation(Hudson, 1988) or uidow(squirt) between
the fractures and pore space (Thomsen, 1995; Hudson et al.,
1996). If the uid can move under stress from the fractures
into hydraulically connected pores, the fractures are no longer
stiff enough to preserve the continuity of the normal (orthogo-
nal to their faces) displacement component. Consequently, the
weakness
N
does not vanish (as it did for isolated cracks com-
pletely lled with uid) and can be expected to take a value
intermediate between zero and the
N
for gas-lled cracks [see
equation(17)]. Thus, theweaknesses
N
for uid- andgas-lled
cracks may be considered as the lower and upper bounds (re-
spectively) of this parameter. This conclusion is supported by
the work of Thomsen (1995), Hudson (1988), and Hudson et al.
(1996).
Thomsen (1995), extending the results of Hoenig (1979),
Budiansky and OConnell (1976), and Hudson (1981), devel-
ops a formalism to account for uid ow between cracks and
spherical (isometric) pores. We consider the case of a small con-
centration of pores [equations (6) and (7) of Thomsen (1995)]
so that their equant porosity can be modeled as a dilute dis-
tribution of spheres in an isotropic background solid. Then
Thomsens theory also results in an effective stiffness matrix
that has the same form as in equation (9). The normal and
tangential weaknesses for Thomsens model are given by
N
= q
Hudson,dry
N
(25)
and
T
=
16e
3(3 2g)
. (26)
An explicit expression for the coefcient q can be found in Ap-
pendix A. Note that equant porosity has no inuence on the
tangential weakness, and equation (26) coincides with equa-
tion (18) for
T
obtained using Hudsons (1981) theory.
Analysis of equations (25), (A-2), and (A-3) shows that in
the presence of equant porosity,
N
does not vanish, even if the
entire pore space is lled with uid. The parameter q becomes
larger with increasing equant porosity, thus making
N
closer
to that of dry cracks. However, q cannot exceed unity [see
equations ((A-1)(A-3)], which means that
N
for Thomsens
model always lies between the values for isolated uid-lled
cracks (
N
=0) and dry cracks.
To illustrate the inuence of equant porosity on the weak-
nesses, we computed
N
as a function of the V
S
/V
P
ratio for a
wide range of equant porosity values (Figure 2). All curves lie
between
N
=0 for isolated uid-lled cracks [equation (19)]
and the dashed line that corresponds to
N
for dry cracks
[equation (17)]. While
N
is almost zero for small equant
porosity
p
=0.01% and
p
=0.1%, it noticeably increases
when the porosity
p
=1% and becomes close to that of dry
cracks as
p
reaches 10%.
1794 Bakulin et al.
In summary, all theories examined above lead to the same
form of the elastic stiffness tensor that describes a system of
parallel, rotationally invariant fractures in a purely isotropic
host rock. The effective anisotropic medium is characterized
by a special type of the HTI symmetry with four independent
stiffness elements. Although each theory operates with a dif-
ferent set of inherent parameters, only four parameter combi-
nations can be extracted from seismic data acquired for such
a model. In principle, four measurements of the elastic coef-
cients may be sufcient to evaluate the Lam e parameters and
of the host rock and two effective fracture parametersthe
weaknesses
N
and
T
. An additional (fth) measurement is
redundant because of constraint (10) and may be used to check
the validity of the assumed model.
The most general description of the effective mediumis pro-
vided by the linear-slip theory, which is directly formulated in
terms of the four measurable parameters. Below we discuss
how to estimate these parameters from reection seismic data
and use themto evaluate the physical properties of the fracture
system.
ANISOTROPIC PARAMETERS OF HTI MEDIA RESULTING
FROM VERTICAL FRACTURES
General fracture set
First, let us consider the most general case of a vertical frac-
ture set with the stiffness tensor expressed through the fracture
weaknesses in equation (9). Seismic signatures in HTI media
are most conveniently described by the vertical velocities (for
instance, the P-wave velocity V
P0
and the velocity V
S
of the
S-wave polarized orthogonal to the cracks) and the following
FIG. 2. Inuence of equant porosity
p
on the normal weak-
ness of a crack system connected to isometric pores (dotted
lines). The dashed line (on top) shows the normal weakness for
isolated dry cracks;
N
=0 corresponds to isolated uid-lled
cracks. The vertical scale is in the units of crack density e. The
parameters of the crack system are the aspect ratio =0.0005,
the volume portion occupied by the cracks
c
=0.01%, and the
crack density e =0.05. The cracks are lled with water that has
the bulk modulus k
/
=2 GPa. The P-wave velocity in the host
rock is V
P
=5 km/s.
set of anisotropic coefcients introduced by R uger (1997) and
Tsvankin (1997) by analogy with Thomsens (1986) parameters
for VTI media:
(V)
c
11
c
33
2c
33
, (27)
(V)
(c
13
c
55
)
2
(c
33
c
55
)
2
2c
33
(c
33
c
55
)
, (28)
(V)
c
66
c
44
2c
44
. (29)
The degree of anellipticity of the model is controlled by the
parameter denoted as
(V)
(Tsvankin, 1997):
(V)

(V)
(V)
1 2
(V)
. (30)
Exact expressions for
(V)
,
(V)
,
(V)
, and
(V)
in terms of the
fracture weaknesses can be derived in a straightforward way
from equation (9) and are given in Appendix B. To gain in-
sight into the relationship between the anisotropic coefcients
and the fracture weaknesses, equations (B-1)(B-4) can be lin-
earized with respect to
N
and
T
under the assumption of
weak anisotropy:
(V)
= 2g(1 g)
N
, (31)
(V)
= 2g[(1 2g)
N

T
], (32)
(V)
=
T
2
, (33)
(V)
= 2g(
T
g
N
). (34)
Analogous expressions for
(V)
and
(V)
in terms of the frac-
ture compliances K
N
and K
T
are given by Sayers and Rickett
(1997).
Since the squared velocity ratio g V
2
S
/V
2
P
<0.5, we infer
from equations (B-1)(B-4), as well as from equations (31)
(34), that
(V)
,
(V)
, and
(V)
for vertical fractures are nonposi-
tive:
(V)
0,
(V)
0, and
(V)
0. (35)
An equivalent form of the expression for
(V)
[derived from
equation (B-4)],
(V)
= (K
T
K
N
)
2
2
2
, (36)
shows that the anellipticity parameter is controlled by the
difference between the tangential and normal compliances
(Schoenberg and Sayers, 1995). Since usually K
N
K
T
,
(V)
is predominantly positive. In the model with gas-lled cracks,
however, Hudsons theory predicts that the K
N
/K
T
ratio may
exceed unity with increasing crack density (Figure 1) and
(V)
may become negative.
As discussed above, the parameters of the crack-induced
HTI model satisfy an additional constraint represented
through the stiffnesses in equation (10). Rewriting this con-
straint in terms of the Thomsen-style anisotropic coefcients
yields (Tsvankin, 1997)
(V)
=
V
2
P0
2V
2
S
(V)
_
2 1
_
f
(V)
_
(V)
1
_
1 2
(V)
_
f
(V)
(V)
f
(V)
(V)
___
f
(V)
_
1 f
(V)
__
, (37)
where V
P0
is the vertical P-wave velocity, V
2
S
is the vertical
velocity of the S-wave polarized orthogonal to the cracks, and
f
(V)
1 V
2
S
/V
2
P0
. In the linearized weak-anisotropy approx-
imation, equation (37) reduces to
(V)
=
1
4g
_
(V)
(V)
1 2g
1 g
_
. (38)
Approximate relations analogous to equation (38) are given
by Schoenberg and Douma (1988) and Thomsen (1995) for TI
media formed by parallel horizontal cracks in an isotropic host
rock.
As discussed by Tsvankin (1997) and Contreras et al. (1999),
equations (37) and (38) suggest that for crack-induced HTI
models the shear-wave splitting parameter [
(V)
[ can be ob-
tained from P-wave moveout data. Indeed, P-wave NMO ve-
locityfromhorizontal anddippingreectors canbeinvertedfor
the vertical velocity V
P0
,
(V)
, and
(V)
(Contreras et al., 1999).
Then, if an estimate of the V
P
/V
S
ratio at vertical incidence is
available,
(V)
can be computed from equation (37).
Possible ranges of anisotropic parameters
The Thomsen-style anisotropic coefcients in the fracture-
induced HTI model are controlled by the weaknesses and the
ratio of and , expressed in our notation by g V
2
S
/V
2
P
. While
the shear-wave splitting parameter
(V)
is equal to
T
/2
[equation (B-3)], the exact expressions (B-1) and (B-2) for
(V)
and
(V)
are more complicated. Therefore, it is instructive to
study the possible ranges of
(V)
and
(V)
by varying parame-
ters
N
,
T
, and g simultaneously within reasonable bounds.
We constrain g by assuming 0.35 V
S
/V
P
0.65 (or, equiv-
alently, 0.12 g 0.42) and use two different ranges for
(V)
:
[
(V)
[ 0.05 (equivalently,
T
0.1) and [
(V)
[ 0.15 (equiv-
alently,
T
0.3). For each pair of g and
T
within the chosen
range, we vary
N
from 0 to 1, unless the maximum possible
value of the ratio K
N
/K
T
=1 is reached (we ignore larger val-
ues of K
N
/K
T
produced by Hudsons theory). Figure 3 shows
FIG. 3. Possible ranges of the parameters
(V)
and
(V)
in HTI media resulting from parallel vertical fractures. (a) [
(V)
[ 0.05
(or
T
0.1); (b) [
(V)
[ 0.15 (
T
0.3). The V
S
/V
P
ratio is constrained by 0.12 g 0.42, and K
N
/K
T
1.
that the possible ranges for both
(V)
and
(V)
are relatively
narrow, especially if [
(V)
[ 0.05. The areas of feasible
(V)
and
(V)
have a quasi-triangular shape and lie above the diagonal
that corresponds to the elliptical model, dened by
(V)
=
(V)
and
(V)
=0 (K
N
=K
T
).
The values of
(V)
and
(V)
in Figure 3 were computed using
the exact equations (B-1) and (B-2), which are valid for any
values of
N
and
T
between 0 and 1 (hence, for any strength
of fracturing). For example,
N
in Figure 3b reaches an un-
commonly large value of 0.92, which corresponds to extremely
compliant fractures that reduce the velocity across them by up
to 75% of the background value.
Isolated penny-shaped cracks
Explicit expressions for dry and uid-lled cracks.For the
special case of thin, isolated, penny-shaped cracks, we can
rewrite equations (31)(34) in terms of the crack density using
Hudson s (1981) theory. To obtain the linearized anisotropic
coefcients for dry (or gas-lled) cracks, we substitute equa-
tions (17) and(18) for the weaknesses intoequations (31)(34):
(V)
=
8
3
e, (39)
(V)
=
8
3
e
_
1
g(1 2g)
(3 2g)(1 g)
_
, (40)
(V)
=
8e
3(3 2g)
, (41)
(V)
=
8
3
e
_
g(1 2g)
(3 2g)(1 g)
_
. (42)
Thus, in the weak-anisotropy approximation all parameters be-
come simple functions of the crack density e and the squared
V
S
/V
P
ratio (g) of the host rock. If we assume that the V
S
/V
P
ratio varies between 0.35 and 0.65, the corresponding ranges
of the anisotropic coefcients are
1796 Bakulin et al.
(V)
= 2.68e,
(V)
= (2.82 0.05)e,
(V)
= (1.10 0.13)e,
(V)
= (0.14 0.05)e. (43)
According to equations (43), the HTI medium resulting
from dry cracks is close to elliptically anisotropic (
(V)
(V)
,
(V)
0), with the parameters largely controlled by the crack
density e. Although the only parameter fully independent of
g is
(V)
, the variations in the other three parameters for the
whole range of g are rather insignicant. Note that the shear-
wave splitting coefcient [
(V)
[ is close to e (Thomsen, 1995;
Tsvankin, 1997), so the time delay between the split S-waves
at vertical incidence gives a good direct estimate of the crack
density (a well-known fact often used in fracture characteriza-
tion).
For uid-lled cracks, we substitute equations (19) and (20)
into equations (31)(34) to obtain
(V)
= 0, (44)
(V)
=
32ge
3(3 2g)
, (45)
(V)
=
8e
3(3 2g)
, (46)
(V)
=
(V)
=
32ge
3(3 2g)
. (47)
Fluid saturation does not change the splitting parameter
[
(V)
[ [compare equations (41) and (46)] but has a strong in-
uence on the other anisotropic coefcients. If the cracks are
completely saturated with uid, the P-wave velocities in the
directions parallel and orthogonal to the cracks are equal to
each other, and
(V)
=0 [equation (44)]. As follows from con-
straint (38), as well as from equations (45) and (46), for van-
ishing
(V)
(V)
=

(V)
4g
. (48)
For a typical value g =0.25,
(V)
and
(V)
become identical.
Therefore, for uid-lled cracks and g =0.25, the absolute val-
ues of
(V)
,
(V)
, and
(V)
are close to each other and to the
crack density e (Tsvankin, 1997; R uger and Tsvankin, 1997).
Inuence of uid content.The inuence of uid satura-
tion on the anisotropic parameters is illustrated in Figure 4. As
mentioned above, the presence of uid has no inuence on the
shear-wave splitting parameter
(V)
because the shear modulus
/
in equation (15) for the weakness
T
goes to zero for both
dry and uid-lled cracks.
In contrast, the values of
(V)
for uid-lled and dry cracks
are substantially different; the same conclusion holds for
(V)
.
The parameter
(V)
is a function of only one weakness (
N
)
[equation (31)], which is responsible for the jump of the nor-
mal displacement across thecrackfaceandthereforeis strongly
dependent on the uid bulk modulus k
/
. Note that the differ-
ence between
(V)
for dry and uid-lled cracks remains the
same (8e/3) for any V
S
/V
P
ratio. For the parameter
(V)
, the
inuence of the crack inll rapidly decreases with the V
S
/V
P
ratio. For the most typical V
S
/V
P
0.5 (g 0.25), the absolute
value of
(V)
for dry cracks is almost three times greater than
that for uid-lled cracks.
For models with hydraulically interconnected cracks and
pores (Thomsen, 1995) or with partially saturated cracks, the
anisotropic parameters always lie between the values for dry
and uid-lled isolated cracks plotted in Figure 4. This follows
fromthe above analysis of equations (25) and(26) andFigure 2.
INVERSION OF SEISMIC SIGNATURES FOR THE
PARAMETERS OF HTI MEDIA
Weaknesses
N
and
T
can be estimated from seismic data
if two of the three anisotropic coefcients (
(V)
,
(V)
,
(V)
) and
the V
S
/V
P
ratio are available. For instance, using the linearized
weak-anisotropy approximations (31) and(32) for
(V)
and
(V)
,
we obtain the weaknesses as
N
=

(V)
2g(1 g)
, (49)
T
=
1
2g
_
1 2g
1 g

(V)
(V)
_
. (50)
Equations (49) and (50) indicate that the inversion for
N
and
T
can be based solely on P-wave traveltime data, which
yield
(V)
and
(V)
(Contreras et al., 1999), assuming that g
was estimated independently (e.g., from well logs in the area).
Other options include using the azimuthal variation of the
P-wave AVO gradient (R uger, 1997; R uger and Tsvankin,
1997) or converted (PS) waves; these various approaches are
discussed in more detail below.
Further interpretation of the weaknesses in terms of the
physical properties of the fracture set is nonunique and can-
not be carried out without making assumptions about the
FIG. 4. Anisotropic parameters of the effective HTI model re-
sulting from isolated penny-shaped cracks embedded in an
isotropic host rock [from equations (39)(42) and (44)(47)].
Solid linesuid-lled cracks. Dashed linesdry (gas-lled)
cracks. The vertical scale is in the units of the crack density e.
microstructure of the fractures and pore space. For penny-
shaped cracks, possibly connected to isometric pores, the tan-
gential weakness
T
depends on only the crack density (i.e.,
on the strength of fracturing) and remains the same for both
dry and uid-lled cracks [equations (18), (20), and (26)]. The
normal weakness
N
(and the ratio
N
/
T
), in contrast, is
also a function of uid content [compare equations (17), (19),
and (25)] and thus provides information about uid saturation
in addition to the crack density (see below). Still, for porous
formations
N
/
T
depends not just onuidsaturationbut also
on the presence of a hydraulic connection between cracks and
pore space.
Reection moveout and AVO gradient of P-waves
P-wave NMO velocity.If the reservoir has a sufcient
thickness, information about fracturing can be obtained from
the azimuthal variation of the interval NMO velocity. In az-
imuthally anisotropic media the conventional Dix differentia-
tion, routinely used to recover the interval velocity, becomes
inaccurate and must be replaced with a more general expres-
siongivenbyGrechka, Tsvankin, andCohen(1999). Adetailed
discussion of the methodology of azimuthal moveout analy-
sis, including a correction for lateral velocity variation, can be
found in Grechka and Tsvankin (1999).
Azimuthal variation of P-wave NMO velocity is described
by an ellipse in the horizontal plane, even for arbitrary
anisotropic heterogeneous media (Grechka and Tsvankin,
1998). In the special case of a plane homogeneous HTI layer,
the NMO ellipse is given by (Tsvankin, 1997)
V
2
P,nmo
() = V
2
P0
1 2
(V)
1 2
(V)
sin
2
, (51)
where V
P0
is the P-wavevertical velocityand is theazimuthof
the common-midpoint (CMP) line with respect to the symme-
try axis (normal to the fractures). Since
(V)
0, the semimajor
axis of the NMO ellipse lies in the fracture plane.
P-wave NMO velocity measurements along three or more
well-separated azimuths can be inverted for the fracture ori-
entation, the velocity V
P0
, and the anisotropic parameter
(V)
.
The value of
(V)
(or
(V)
), also required for estimating frac-
ture weaknesses, can be found from NMO velocities of dip-
ping events (Tsvankin, 1997; Contreras et al., 1999) or from
nonhyperbolic moveout (Al-Dajani and Tsvankin, 1998).
P-wave AVOgradient.Reection coefcients measured at
oblique incidence angles over fractured formations vary with
the azimuth of the sourcereceiver line. This azimuthal depen-
dence can be used in AVOanalysis to obtain information about
the orientation, density, and content of the fractures (R uger
and Tsvankin, 1997; Sayers and Rickett, 1997).
The weak-anisotropy approximation for the plane P-wave
reection coefcient at a small-contrast interface between two
HTI media with the same orientation of the symmetry axis
is derived by R uger (1997) in terms of the Thomsen-style
anisotropic parameters. The behavior of the reection coef-
cient R at small and moderate incidence angles is governed
by the so-called AVO gradientthe initial slope of R plotted
as a function of sin
2
. Here we consider the azimuthally vary-
ing AVO gradient for a P-wave reection from the interface
between isotropic and HTI media. The difference between the
AVO gradients in the directions perpendicular and parallel to
the cracks (B
ani
) can be written as (R uger, 1997; R uger and
Tsvankin, 1997)
B
ani
=

(V)
8g
(V)
2
, (52)
where g has the meaning of the average V
2
S
/V
2
P
for the two half-
spaces. Substituting the approximate expressions (32) and (33)
for
(V)
and
(V)
yields B
ani
as anexplicit functionof the fracture
weaknesses:
B
ani
= g[
T
(1 2g)
N
]. (53)
An equivalent equation in terms of the (dimensional) compli-
ances is given by Sayers and Rickett (1997).
For isolated penny-shaped cracks, the azimuthal variation of
the AVO gradient can be related directly to the crack density.
Equations (40), (41), (45), and (46) for
(V)
and
(V)
obtained
from Hudsons (1981) theory lead to the following expressions
for dry and uid-lled fractures:
B
ani
dry
=
4(8g
2
12g 3)
3(3 2g)(1 g)
e, (54)
B
ani
wet
=
16g
3(3 2g)
e. (55)
Clearly, the magnitude of the azimuthal AVO response in
weakly anisotropic HTI media is proportional tothe crackden-
sity. In addition, the value of B
ani
is strongly dependent on g,
i.e., on the V
S
/V
P
ratio of the host rock (Figure 5). If V
S
/V
P
takes a typical value of 0.55, B
ani
wet
is positive and B
ani
dry
is close to
FIG. 5. Azimuthal variation of the AVO gradient for the
P-wave reection from the interface between isotropic and
HTI media. The difference between the AVO gradients in the
directions perpendicular and parallel to the cracks (B
ani
) was
calculated for dry [equation (54), dashed line] and uid-lled
cracks [equation (55), solid line]. The vertical scale is in the
units of the crack density e. The dots mark the values of B
ani
picked from the curves of the exact reection coefcient for
the crack density e =0.07. The cracks were introduced in the
lower half-space using the rst-order Hudsons (1981) theory.
The upper half-space and the host rock in the lower half-space
have the same velocities, with V
P
=6000 m/s and V
S
chang-
ing in accordance with the V
S
/V
P
ratio on the horizontal axis
[V
S
=V
P
(V
S
/V
P
)]. The densities of the upper and lower media
are 2800 and 2300 kg/m
3
, respectively.
1798 Bakulin et al.
zero. This result agrees with the computations of Strahilevitz
and Gardner (1995), R uger and Tsvankin (1997), and Sayers
and Rickett (1997), who noted that the AVO gradient for dry
cracks is almost independent of azimuth for models with simi-
lar V
S
/V
P
ratios. The relation between B
ani
wet
and B
ani
dry
, however,
changes substantially for higher or lower values of V
S
/V
P
. For
instance, B
ani
dry
2 B
ani
wet
when V
S
/V
P
is close to 0.4. Compari-
son with the exact reection coefcient (Figure 5) shows the
high accuracy of equations (54) and (55) for moderate crack
densities reaching 0.050.07.
Thus, the value of B
ani
, combined with an estimate of the
V
S
/V
P
ratio, may serve as an indicator of uid content. Quan-
titative interpretation of the azimuthal AVO response, how-
ever, requires additional information about the crack density
or anisotropic coefcients. Indeed, if the cracks are dry and
V
S
/V
P
=0.5 0.6, the P-wave AVO gradient does not vary
much with azimuth, irrespective of the degree of fracturing. A
direct estimate of the crack density (i.e.,
T
) can be obtained
from shear-wave splitting at vertical incidence; then B
ani
can
be used to nd
N
[equation (53)] and evaluate uid satura-
tion. Alternatively, as suggestedby R uger andTsvankin(1997),
the azimuthal AVO response can be combined with the results
of azimuthal velocity analysis [equation (51)] to resolve the
anisotropic coefcients
(V)
and
(V)
, which can be further re-
calculated into the fracture weaknesses.
As discussed above, if the cracks are partially saturated or
are connected to pore space,
N
is always greater than zero
but is smaller than the value for dry cracks. Equation (53) indi-
cates that B
ani
in this case falls between the values of B
ani
wet
and
B
ani
dry
for the same V
S
/V
P
ratio. Therefore, the curves shown in
Figure 5 dene the lower and upper bounds of B
ani
for various
microgeometries of the fractured medium.
Distinguishing between dry and uid-lled cracks.Let us
assume that two anisotropic parameters [for example,
(V)
and
(V)
] have been found and an estimate of g is available. Then
it is important to assess whether one can distinguish between
dry and uid-lled cracks in the presence of realistic errors in
(V)
,
(V)
, and V
S
/V
P
ratio (g).
First, consider a fracture set with the density e =7% embed-
ded in a mediumwith V
S
/V
P
=0.5 (g =0.25). For such a model,
the tangential weakness
T
=0.15 for both dry and uid-lled
cracks, while the normal weakness changes from
N
=0.50 if
the cracks are dry (
(V)
=0.21,
(V)
=0.19) to
N
=0 for
uid-lled cracks (
(V)
=0,
(V)
=0.07). We simulated errors
in the data by adding Gaussian noise with the standard devia-
tion =0.05 to the correct values of
(V)
,
(V)
, and V
S
/V
P
and
inverted these three parameters for
N
and
T
using the exact
equations from Appendix B. The inversion results, along with
the input anisotropic coefcients, are marked by small dots in
Figure 6. Clearly, the clouds of estimated weaknesses are well
separated by their values of
N
. This implies that although er-
rors in
(V)
,
(V)
, and V
S
/V
P
on the order of 0.05 0.1 intro-
duce a substantial uncertainty in the value of the crack density
(or
T
), they do not prevent us from separating models with
Figure 7 displays the results of some additional numerical
experiments. As expected, reducing errors in the data leads
to a visible tightening of the clouds of
N
and
T
(compare
Figures 7a and 7b with Figures 6c and 6d).With decreasing data
error, it may become possible to distinguish between fully and
partially saturated fractures (or fractures connected to pore
space).
The sensitivity of the weaknesses to errors in input data (re-
ected in the size of the
N
and
T
clouds) is controlled by
the V
S
/V
P
ratio (Figures 7cf). For a xed magnitude of errors
in
(V)
and
(V)
, the errors in
N
and
T
decrease with in-
creasing V
S
/V
P
ratio. However, the value of
N
for dry cracks
also decreases from
N
=0.7 for V
S
/V
P
=0.4 to
N
=0.4 for
V
S
/V
P
=0.6 (Figures 7d and 7f). As a result, the separation
between
N
for the models with dry and uid-lled cracks
is smaller for higher V
S
/V
P
ratios, and our overall ability to
resolve the uid content is almost independent of the V
S
/V
P
ratio.
Our numerical tests thus suggest that it is feasible to distin-
guish between dry and uid-lled fractures if errors in
(V)
,
(V)
,
and the V
S
/V
P
ratio do not exceed 0.1 and the crack density
reaches 57%.
Reection moveout and AVO gradient of PS-waves
For multicomponent surveys, it is possible to devise a
fracture-detection algorithm based on mode-converted (PS)
data. Since moveout and amplitude information average
medium properties on different scales, combining the P-wave
NMO velocity and AVO gradient in the characterization of
heterogeneous formations may prove difcult. Therefore, it is
advantageous to supplement the P-wave NMO ellipse with re-
ection traveltimes of PS-waves or, alternatively to carry out
joint inversion of the AVO gradients of P- and PS-waves.
One of the split shear waves inHTI media (denotedas S
|
or S
parallel) is polarized in the isotropy (fracture) plane, while the
polarization vector of the second S-wave (S
) lies in the plane

FIG. 6. Noise-contaminated anisotropic coefcients
(V)
and
(V)
for (a) uid-lled and (b) dry cracks (small dots). (c) The
inverted weaknesses
N
and
T
for the uid-lled model from
(a). (d) The inverted
N
and
T
for the dry-crack model from
(b). The large dots indicate the correct parameter values.
formed by the symmetry axis and the slowness vector. At ver-
tical incidence, the S
-wave is polarized perpendicular to the

fractures and travels slower than the S
|
-wave. In each vertical
symmetry plane, the P-wave is coupled to the in-plane polar-
ized S-wave and generates a single converted-mode reection
from a horizontal interface. Thus, in the isotropy (fracture)
plane one should be able to record the reected wave PS
|
,
while in the plane orthogonal to the fractures (the symmetry-
axis plane) anincident P-wave is convertedintothe PS
reec-
tion. Sinceeither oneor theother of thetwoconvertedmodes is
weak in the vicinity of each symmetry plane, extracting the full
azimuthal dependence of the NMO velocity or AVO gradient
of the PS
|
- or PS
-wave requires a better azimuthal coverage

than that for P-wave reections. Here, we restrict the discus-
sion to the reection traveltimes and amplitudes of converted
waves in the vertical symmetry planes of the HTI medium.
PS-wave traveltimes.As follows from the analysis of
Grechka, Theophanis, and Tsvankin (1999) for the more gen-
eral orthorhombic media, the azimuthal variation of the NMO
velocity of each converted wave (PS
|
and PS
) in a horizon-
tal HTI layer is elliptical. By performing moveout (semblance)
analysis inthe symmetry planes, we determine the vertical trav-
eltime and one of the semiaxes of the NMO ellipse for each
FIG. 7. Dependence of the inverted weaknesses
N
and
T
(small dots) for uid-lled (left column) and dry (right col-
umn) cracks on the V
S
/V
P
ratio and on the standard devi-
ation of
(V)
,
(V)
, and V
S
/V
P
from their correct values.
The weaknesses are the same as in Figure 6. (a) and (b):
=0.025, V
S
/V
P
=0.5. (c) and (d): =0.05, V
S
/V
P
=0.4. (e)
and (f): =0.05, V
S
/V
P
=0.6.
converted wave. Note that mode conversions in a horizontal
HTI layer vanish at vertical incidence, and the vertical (zero-
offset) traveltimes of the convertedwaves are foundessentially
by extrapolation using the best-t hyperbola provided by sem-
blance analysis.
Given the P-wave zero-offset reection traveltime, the zero-
offset times of the converted waves from the same interface
can be used to obtain the ratio of the shear-wave vertical ve-
locities [i.e., the parameter
(V)
] and an estimate of the ratio
g V
2
S
/V
2
P
. This implies that the P-wave NMO ellipse, which
provides
(V)
, and the vertical traveltimes of the P-wave and
split converted waves are sufcient to nd
(V)
,
(V)
, and g and,
therefore, both fracture weaknesses.
Symmetry-plane NMO velocities of the converted waves
provide redundant information that may, however, increase
the accuracy of the parameter estimation procedure. In each
vertical symmetry plane the traveltimes of reected waves are
describedbythesameequations as inVTI media, andtheNMO
velocities of P-, S-, and PS-waves are related by the following
Dix-type equation (Tsvankin and Thomsen, 1994):
[V
S,nmo
]
2
=
t
(PS)
0
[V
PS,nmo
]
2
t
(P)
0
[V
P,nmo
]
2
t
(PS)
0
t
(P)
0
, (56)
where t
0
denotes the one-way vertical traveltime for pure (P
and S) reections and the two-way vertical traveltime for the
PS-wave. Equation (56) can be used to obtain the NMO ve-
locities of the reected shear waves S
|
(in the fracture plane)
and S
(in the plane perpendicular to the fractures) from P

and PS data. Since the velocity of each mode in the fracture
plane is independent of the angle with the vertical, the NMO
velocity of the S
|
-wave is equal to the fast shear-wave vertical
velocity V
S1
. The NMOvelocity of the S
-wave in the direction

perpendicular to the fractures is given by (Tsvankin, 1997)
V
S
,nmo
= V
S2
_
1 2
(V)
(57)
and
(V)
_
V
P0
V
S2
_
2
_
(V)
(V)
_
, (58)
where V
P0
and V
S2
are the vertical velocities of the P- and
S
-waves, respectively. If the vertical velocities and

(V)
were
determined from the P-wave NMO ellipse and the vertical
traveltimes, the NMO velocity of the S
-wave yields
(V)
. Both
the fast S-wave velocity V
S1
and
(V)
, however, can be obtained
as well from the vertical traveltimes of the converted waves,
the P-wave NMOellipse, and the constraint on the anisotropic
parameters of HTI media.
PS-wave AVO gradient.For relatively thin target layers,
interval moveout analysis becomes inaccurate; so it may be
advantageous to supplement the P-wave AVO response with
prestackamplitudes of PS-wave reections. Of course, it is also
possible to use nonconverted shear waves, but they are seldom
acquired in exploration surveys.
As before, we consider an interface between isotropic and
HTI media and analyze the magnitude of the azimuthal vari-
ation of the reection coefcient. The results of R uger (1996)
show that the PS reection coefcient at small offsets is pro-
portional to sin , where is the P-wave incidence angle.
1800 Bakulin et al.
Therefore, we dene the PS-wave AVO gradient as the coef-
cient of the sin term and use R ugers (1996) weak-anisotropy
approximations to express the AVO gradient through the
anisotropic parameters. The difference between the AVO gra-
dients of the PS
-wave across the fractures and the PS

|
-wave
parallel to them can be written as
B
ani
=

(V)
2
4(
g g)
(V)
2
2(1
g)
. (59)
Introducing the fracture weaknesses into equation (59) yields
B
ani
=
g
1
g
[
T

g(1 2g)
N
]. (60)
Despite some similarity between equation (60) and the corre-
sponding P-wave expression (53), the factor B
ani
for the P- and
PS-modes is controlled by different combinations of
T
and
N
. Hence, using both P- and PS-waves in azimuthal AVO
analysis has the potential of yielding information on the weak-
nesses
T
and
N
or, equivalently, the crack density and the
uid content of the fractures. Also, as mentioned, the squared
vertical-velocity ratio g may be determined from the P- and
PS-wave traveltimes, unless the reservoir is too thin.
Figure 8 reproduces the numerical example from Figure 5
but this time for PS-waves. The curves were generated by sub-
stituting the values of
N
and
T
for dry (dashed line) and
uid-lled (solid) cracks into equation (60). Note that the accu-
racy of the weak-anisotropy approximation for the PS-wave is
not much lower than that for the P-wave. As expected fromthe
formof the analytic solutions, B
ani
for both P- and PS-waves is
positive in models with isolated uid-lled cracks and changes
sign with increasing V
S
/V
P
for dry cracks. However, in the lat-
ter case the intersection with the horizontal axis for PS-waves
occurs at a lower value of V
S
/V
P
thanfor P-waves. Hence, if the
V
S
/V
P
ratio is close to a typical value of 0.55 and the P-wave
AVO gradient for dry cracks is almost independent of azimuth
FIG. 8. The difference between the converted-wave AVO gra-
dients inthe directions perpendicular (PS
) andparallel (PS
|
)
to the cracks for dry (dashed line) and uid-lled (solid line)
cracks; both curves are computed from the weak-anisotropy
approximation. The dots mark the values of B
ani
picked from
the curves of the exact reection coefcient for the crack den-
sity e =0.07. The model parameters are the same as those in
Figure 5.
(R uger and Tsvankin, 1997; Sayers and Rickett, 1997), the ad-
ditionof PS-waves may helpidentify fracturing and, moreover,
discriminate between dry and uid-lled cracks.
DISCUSSION AND CONCLUSIONS
The linear-slip theory (e.g., Schoenberg and Sayers, 1995),
basedonthe general treatment of fractures as surfaces of weak-
ness inside a solid matrix, provides a convenient framework for
relating seismic signatures tothe properties of fracture systems.
The inherent parameters of the linearslip theory for rotation-
ally invariant fractures are the normal (
N
) and tangential
(
T
) weaknesses, which can be estimated from seismic data.
The theories of Hudson (1981) and Thomsen (1995), designed
for specic physical fracture models involving penny-shaped
cracks, can be used to express
N
and
T
through parameters
dependent onthemicrostructureof cracks andpores. Although
determination of these microstructural parameters from seis-
mic data requires additional information about the medium,
Hudsons and Thomsens models are helpful in guiding the in-
terpretation of the weaknesses in terms of the crack density
and uid content.
Vertical, parallel, rotationally invariant fractures lead to
a particular type of HTI media described by four indepen-
dent parameters. We obtained linearized expressions for the
Thomsen-style anisotropic coefcients of HTI media
(V)
,
(V)
,
and
(V)
in terms of the weaknesses
N
and
T
and the V
S
/V
P
ratiointhebackgroundmedium. Oneof theanisotropic param-
eters (
(V)
,
(V)
, or
(V)
) canbefoundfromtheother twoandthe
V
S
/V
P
ratio. Interestingly,
(V)
,
(V)
, and
(V)
in any fracture-
induced HTI model are always negative, while the anellipticity
parameter
(V)
is predominantly positive. All possible pairs
[
(V)
,
(V)
] for HTI models resulting from parallel vertical frac-
tures belong to a relatively narrow area with a quasi-triangular
shape on the [
(V)
,
(V)
]-plane. These results can be used to
identify fracture-induced HTI media from seismic measure-
ments.
If the model contains isolated penny-shaped (spheroidal)
cracks, the weaknesses and the anisotropic parameters can be
related to the physical properties of the crack system using
Hudsons theory. In the linearized weak-anisotropy approx-
imation, all anisotropic parameters are proportional to the
crack density, with the coefcients controlled by the V
S
/V
P
ratio and, for
(V)
and
(V)
, uid saturation. The shear-wave
splitting parameter [
(V)
[ for dry or uid-lled cracks and a
wide range of the V
S
/V
P
ratios remains close to the crack den-
sity. Therefore, the traveltimes or reection amplitudes of split
S-waves at vertical incidence are strongly dependent on the de-
gree of fracturing but cannot be used to discriminate between
In contrast,
(V)
and
(V)
are sensitive to the presence of uid
in the cracks. If the cracks are dry, the effective medium is
close to elliptical (
(V)
(V)
), with the anisotropic parameters
tightly governed by the crack density and weakly dependent on
the V
S
/V
P
ratio. The absolute value of
(V)
(V)
for dry cracks
is almost three times higher than [
(V)
[ and the crack density.
For isolated uid-lled cracks,
(V)
practically vanishes, while
(V)
for typical V
S
/V
P
0.5 is close to
(V)
and (by absolute
value) to the crack density. The inuence of crack inll on
(V)
, however, decreases with the V
S
/V
P
ratio. For models with
interconnected cracks and pores (Thomsen, 1995) or with
partially saturated cracks,
(V)
and
(V)
always lie between the
values for dry and isolated uid-lled cracks.
The simplest way to estimate
(V)
and the fracture orienta-
tion is to reconstruct the P-wave NMO ellipse from a hori-
zontal reector by means of azimuthal moveout analysis. The
parameter
(V)
can also be obtained from P-wave data using
dipping events or nonhyperbolic moveout. Then
(V)
and
(V)
,
along with the V
S
/V
P
ratio, which must be evaluated sepa-
rately (e.g., from well logs in the area), can be used to nd
(V)
.
Alternatively,
(V)
can be determined directly from prestack
P-wave amplitudes by combining the azimuthal variation of
the AVO gradient with the NMO ellipse for horizontal events.
If convertedwaves are available andthe reservoir is sufciently
thick, it is possible to recover all relevant anisotropic parame-
ters from the zero-offset traveltimes of P- and PS-waves and
either the P-wave NMO ellipse or the AVO gradients of the
P- or PS-waves. For surveys acquired with shear-wave sources,
it is benecial to calibrate P-wave data with split shear waves,
whose polarizations, traveltimes, and AVOresponse give an in-
dependent estimate of the fracture orientation and parameter
(V)
and help to evaluate the V
S
/V
P
ratio.
Any two out of the three anisotropic parameters (
(V)
,
(V)
,
and
(V)
) and the V
S
/V
P
ratio can then be recalculated into the
fracture weaknesses
N
and
T
. Further interpretation of the
weaknesses is not unique and must rely on the assumed phys-
ical model. For penny-shaped cracks, the tangential weakness
is close to twice the crack density, while the normal weakness,
N
, (or the ratio
N
/
T
) is sensitive to uid saturation. Even
for moderate crack densities, there is a substantial difference
between the value of
N
for dry cracks and the vanishing
N
for isolated uid-lled cracks, with intermediate values of the
normal weakness corresponding to partial saturation and/or
the presence of uid ow (squirt) between cracks and pore
space. Therefore,
N
/
T
is a useful indicator of uid content,
although its interpretation may be ambiguous without addi-
tional information.
ACKNOWLEDGMENTS
This research was carried out during a visit of Andrey
Bakulin to the Center for Wave Phenomena (CWP), Colorado
School of Mines, in1998. We are grateful toLeonThomsen(BP
Amoco) and members of the A (nisotropy)-team of CWP for
helpful discussions and to Andreas R uger (Landmark) for his
review of the manuscript. The support for this work was pro-
vided by the members of the Consortium Project on Seismic
Inverse Methods for Complex Structures at CWP and by the
U.S. Department of Energy (award #DE-FG03-98ER14908).
REFERENCES
Aguilera, R., 1998, Geologic aspects of naturally fractured reservoirs:
The Leading Edge, 17, 16671670.
Al-Dajani, A., and Tsvankin, I., 1998, Nonhyperbolic reection move-
out for horizontal transverse isotropy: Geophysics, 63, 17381753.
Assad, J. M., Tatham, R. H., McDonald, J. A., Kusky, T. M., andJech, J.,
1993, Aphysical model studyof scatteringof waves byalignedcracks:
Comparison between experiment and theory: Geophys. Prosp., 41,
323339.
Backus, G. E., 1962, Long-wave elastic anisotropy produced by hori-
zontal layering: J. Geophys. Res., 67, 44274440.
Bagrintseva, K. I., 1982, Fracturing of rocks: Nedra (in Russian).
Bakulin, A. V., andMolotkov, L. A., 1998, Effectivemodels of fractured
and porous media: St. Petersburg Univ. Press (in Russian).
Berg, E., Hood, J., and Fryer, G., 1991, Reduction of the general frac-
ture compliance matrix Z to only ve independent elements: Geo-
phys. J. Internat., 107, 703707.
Budiansky, B., and OConnell, R. J., 1976, Elastic moduli of a cracked
solid: Internat. J. Solids Structures, 12, 8197.
Contreras, P., Grechka, V., and Tsvankin, I., 1999, Moveout inversion
of P-wave data for horizontal transverse isotropy: Geophysics, 64,
12191229.
Corrigan, D., Withers, R., Darnall, J., and Skopinski, T., 1996, Fracture
mapping from azimuthal velocity analysis using 3D surface seismic
data: 66th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Ab-
stracts, 18341837.
Grechka, V., andTsvankin, I., 1998, 3-Ddescriptionof normal moveout
in anisotropic inhomogeneous media: Geophysics, 63, 10791092.
1999, 3-D moveout inversion in azimuthally anisotropic media
with lateral velocity variation: Theory and a case study: Geophysics,
64, 12021218.
Grechka, V., Theophanis, S., and Tsvankin, I., 1999, Joint inversion of
P- and PS-waves in orthorhombic media: Theory and a physical-
modeling study: Geophysics, 64, 146161.
Grechka, V., Tsvankin, I., and Cohen, J. K., 1999, Generalized Dix
equation and analytic treatment of normal-moveout velocity for
anisotropic media: Geophys. Prosp., 47, 117148.
Hoenig, A., 1979, Elastic moduli of the non-randomly cracked body:
Internat. J. Solids Structures, 15, 137154.
Hsu, C.-J., andSchoenberg, M., 1993, Elastic waves throughasimulated
fractured medium: Geophysics, 58, 964977.
Hudson, J. A., 1980, Overall properties of a cracked solid: Math. Proc.
Camb. Phil. Soc., 88, 371384.
1981, Wave speeds and attenuation of elastic waves in material
containing cracks: Geophys. J. Roy. Astr. Soc., 64, 133150.
1988, Seismic wave propagation through material containing
partially saturated cracks: Geophys. J., 92, 3337.
Hudson, J. A., Liu, E., and Crampin, S., 1996, The mechanical proper-
ties of materials with interconnected cracks and pores: Geophys. J.
Internat., 124, 105112.
Kirkinskaya, V. N., and Smekhov, E. M., 1981, Carbonate rocks as
collectors of oil and gas: Nedra (in Russian).
Molotkov, L. A., and Bakulin, A. V., 1997, An effective model of a
fractured medium with fractures modeled by the surfaces of discon-
tinuity of displacements: J. Math. Sci., 86, 27352746.
Nichols, D., Muir, F., and Schoenberg, M., 1989, Elastic properties of
rocks with multiple sets of fractures: 59th Ann. Internat. Mtg., Soc.
Expl. Geophys., Expanded Abstracts, 471474.
Pyrak-Nolte, L. J., Myer, L. R., andCook, N. G. W., 1990a, Transmission
of seismic waves across single natural fractures: J. Geophys. Res., 95,
No. B6, 86178638.
1990b, Anisotropy in seismic velocities and amplitudes from
multiple parallel fractures: J. Geophys. Res., 95, No. B7, 11 345
11 358.
Romm, E. S., 1985, Structural models of rocks pore space: Nedra (in
Russian).
R uger, A., 1996, Reection coefcients and azimuthal AVO analysis
in anisotropic media: Ph.D. thesis, Colorado School of Mines.
1997, P-wave reection coefcients for transversely isotropic
models with vertical and horizontal axis of symmetry: Geophysics,
62, 713722.
R uger, A., and Tsvankin, I., 1997, Using AVO for fracture detection:
Analytic basis and practical solutions: The Leading Edge, 10, 1429
1434.
Sayers, C., and Rickett, J. E., 1997, Azimuthal variation in AVO re-
sponse for fractured gas sands: Geophys. Prosp., 45, 165182.
Schoenberg, M., 1980, Elastic wave behavior across linear slip inter-
faces: J. Acoust. Soc. Am., 68, 15161521.
1983, Reectionof elastic waves fromperiodically stratiedmedia
with interfacial slip: Geophys. Prosp., 31, 265292.
Schoenberg, M., and Douma, J., 1988, Elastic wave propagation in
media with parallel fractures and aligned cracks: Geophys. Prosp.,
36, 571590.
Schoenberg, M., and Muir, F., 1989, A calculus for nely layered
anisotropic media: Geophysics, 54, 581589.
Schoenberg, M., and Sayers, C., 1995, Seismic anisotropy of fractured
rock: Geophysics, 60, 204211.
Strahilevitz, R., and Gardner, G. H. F., 1995, Fracture detection us-
ing P-wave AVO: 65th Ann. Internat. Mtg., Soc. Expl. Geophys.,
Expanded Abstracts, 589591.
1966.
1995, Elastic anisotropy due to aligned cracks in porous rock:
Geophys. Prosp., 43, 805830.
Tsvankin, I., 1997, Reection moveout and parameter estimation for
horizontal transverse isotropy: Geophysics, 62, 614629.
Tsvankin, I., andThomsen, L., 1994, Nonhyperbolic reectionmoveout
in anisotropic media: Geophysics, 59, 12901304.
1802 Bakulin et al.
APPENDIX A
WEAKNESSES FOR THOMSENS MODEL OF FRACTURED POROUS ROCK
Thomsen (1995) has developed a model that accounts for
the effect of uid ow (squirt) between cracks and isometric
pores in porous rocks. He generalizes the results of Hoenig
(1979), Budiansky and OConnell (1976), and Hudson (1981)
for the model that contains two sets of hydraulically connected
inclusions (e.g, the rst set is thin cracks and the second is iso-
metric pores). Here we restrict the discussion to models with a
small concentration of isometric pores [equations (6) and (7)
of Thomsen (1995)], where equant porosity is approximated by
a dilute distribution of spherical pores in an isotropic matrix.
The expressions for the generic Thomsen parameters , ,
and for this medium (Thomsen, 1995) become identical to
those for the linear-slip model (Schoenberg and Douma, 1988),
if we use the following expressions for
N
and
T
:
N
=
4
3
e
(1 g)g
_
1
k
/
2/3
_
D
cp
,
(A-1)
T
=
16
3
e
(3 2g)
.
Here and are the Lam e parameters of the matrix, k
/
is the
bulk modulus of the uid lling the cracks (
/
=0), e is the
crack density, g V
2
S
/V
2
P
, and D
cp
is the so-called uid factor,
designed to account for the interconnection between cracks
and pores. At relatively low frequencies, where the uid has
enough time to move from cracks to pores, D
cp
is given by
(Thomsen, 1995)
D
cp
=
_
1
k
/
2/3
k
/
( 2/3)(
c
p
)
(A
p
p
A
c
e)
_
1
, (A-2)
where
p
and
c
are the fractions of volume occupied by
pores and cracks, respectively [the crack porosity may be ex-
pressed as
c
=(4ec)/(3a)]. Coefcients A
p
and A
c
depend
only on the background V
S
/V
P
ratio: A
p
=(3 2g)/(2g) and
A
c
=(4/9) [(2 3g)/(1 g)] (Thomsen, 1995). Comparing
equation (A-1) with (17), we may represent the normal weak-
ness for the Thomsen model as
N
= q
Hudson, dry
N
, q =
_
1
k
/
2/3
_
D
cp
. (A-3)
APPENDIX B
EXACT EXPRESSIONS FOR THE ANISOTROPIC COEFFICIENTS OF TERMS OF THE WEAKNESSES
The exact expressions for the anisotropic coefcients
(V)
,
(V)
, and
(V)
in terms of the weaknesses
N
and
T
can be
derived from equations (9) and (27)(29) as
(V)
=
2g(1 g)
N
1
N
(1 2g)
2
, (B-1)
(V)
=
2g[(1 2g)
N
T
][1 (1 2g)
N
]
_
1
N
(1 2g)
2
_
_
1
1
1 g
_
T

N
(1 2g)
2
_
_,
(B-2)
and
(V)
=
T
2
. (B-3)
The exact anellipticity coefcient
(V)
is proportional to the
difference between the shear and normal compliances:
(V)
= (K
T
K
N
)
2
2
( )(1 K
N
( 2))
( )( 2)(1 2K
N
)
2
2
(K
T
K
N
)
.
(B-4)
Schoenberg and Sayers (1995) were the rst to point out that
the sign of anellipticity is controlled by the difference K
T
K
N
.
Estimation of fracture parameters from reection seismic dataPart II:
Fractured models with orthorhombic symmetry
Andrey Bakulin
, Vladimir Grechka
, and Ilya Tsvankin
ABSTRACT
Existing geophysical and geological data indicate that
orthorhombic media with a horizontal symmetry plane
should be rather common for naturally fractured reser-
voirs. Here, we consider two orthorhombic models: one
that contains parallel vertical fractures embedded in a
transversely isotropic background with a vertical sym-
metry axis (VTI medium) and the other formed by two
orthogonal sets of rotationally invariant vertical frac-
tures in a purely isotropic host rock.
Using the linear-slip theory, we obtain simple ana-
lytic expressions for the anisotropic coefcients of ef-
fective orthorhombic media. Under the assumptions of
weak anisotropy of the background medium(for the rst
model) and small compliances of the fractures, all ef-
fective anisotropic parameters reduce to the sum of the
background values and the parameters associated with
each fracture set. For the model with a single fracture
system, this result allows us to eliminate the inuence of
the VTI background by evaluating the differences be-
tween the anisotropic parameters dened in the vertical
symmetryplanes. Subsequently, thefractureweaknesses,
which carry information about the density and content
of the fracture network, can be estimated in the same
way as for fracture-induced transverse isotropy with a
horizontal symmetry axis (HTI media) examined in our
previous paper (part I). The parameter estimation pro-
cedure can be based on the azimuthally dependent re-
ection traveltimes and prestack amplitudes of P-waves
alone if an estimate of the ratio of the P- and S-wave
vertical velocities is available. It is benecial, however,
to combine P-wave data with the vertical traveltimes,
NMO velocities, or AVO gradients of mode-converted
(PS) waves.
In each vertical symmetry plane of the model with
two orthogonal fracture sets, the anisotropic parameters
are largely governed by the weaknesses of the fractures
orthogonal to this plane. For weak anisotropy, the frac-
ture sets are essentially decoupled, and their parameters
can be estimated by means of two independently per-
formed HTI inversions. The input data for this model
must include the vertical velocities (or reector depth)
to resolve the anisotropic coefcients in each vertical
symmetry plane rather than just their differences.
We also discuss several criteria that can be used to
distinguish between the models with one and two frac-
ture sets. For example, the semimajor axis of the P-wave
NMO ellipse and the polarization direction of the ver-
tically traveling fast shear wave are always parallel to
each other for a single system of fractures, but they may
become orthogonal in the medium with two fracture
sets.
INTRODUCTION
This is the second part of our series of three papers on char-
acterization of naturally fractured reservoirs using surface seis-
mic data. The main goal of the series is to provide a connection
between fracture detection methods and rock physics models
of fractured media, such as those developed by Hudson (1980,
1981, 1988), Schoenberg (1980, 1983), and Thomsen (1995).
In particular, we are interested in elucidating the dependence
Manuscript received by the Editor April 20, 1999; revised manuscript received March 1, 2000.
Cross, Madingley Road, Cambridge, CB3 0EL, England. E-mail: bakulin@cambridge.scr.slb.com.
Colorado School of Mines, Center for Wave Phenomena, Department of Geophysics, Golden, Colorado 80401-1887. E-mail: vgrechka@dix.mines.
edu; ilya@dix.mines.edu.
of reection seismic signatures on the physical properties of
the fractures and in developing inversion algorithms for esti-
mating fracture parameters using P-wave and converted (PS)
reection data.
In the rst paper of the series (Bakulin et al., 2000; hereafter
referred to as part I), we address these problems for transverse
isotropywithahorizontal axis of symmetry(HTI media), which
describes a single set of vertical, parallel, rotationally invariant
fractures in a purely isotropic background medium. HTI is the
1803
1804 Bakulin et al.
simplest azimuthally anisotropic model that provides valuable
insights into the behavior of seismic signatures over fractured
formations.
However, since the rock matrix usually exhibits some form
of anisotropy, it is difcult to expect the HTI model to be ade-
quate for typical fractured reservoirs. Here, we extend the ap-
proach of part I to more complicated, but likely more realistic,
orthorhombic models. The description of seismic signatures in
orthorhombic media can be simplied signicantly by applying
Tsvankins (1997b) notation based on the analogy between the
symmetry planes of orthorhombic and TI media. For instance,
P-wave velocities and traveltimes are fully controlled by the
P-wave vertical velocity and ve dimensionless anisotropic co-
efcients, rather than by nine stiffnesses in the conventional
notation. Also, this parameterization yields concise exact ex-
pressions for the NMOvelocities of pure modes reected from
a horizontal interface (Grechka and Tsvankin, 1998).
As shown by Grechka et al. (1999), eight out of the nine pa-
rameters of orthorhombic media with a horizontal symmetry
plane can be obtained from the NMO velocities of horizontal
P- and split PS-events, provided the vertical velocities or re-
ector depth are known. Reection moveout of P-waves alone
is sufcient to nd the orientation of the symmetry planes,
the NMO velocities within them, and three anellipticity coef-
cients . The parameters , however, can be determined only
if dipping events or nonhyperbolic moveout is available.
While orthorhombic symmetry may be caused by a variety
of physical reasons, we restrict ourselves to two orthorhombic
models believed to be most common for fractured reservoirs:
(1) a single set of vertical cracks in a transversely isotropic
background with a vertical symmetry axis (VTI) and (2) two
systems of vertical fractures orthogonal to each other in an
isotropic background medium. If the intrinsic anisotropy of
the host rock and the anisotropy induced by the fractures are
weak, the anisotropic coefcients of the background and frac-
tures can be algebraically added in describing the effective
medium. This result provides important insights into the in-
uence of fractures on the effective anisotropy and helps to
generalize the parameter estimation methodology of part I for
orthorhombic media. For both orthorhombic models we re-
late Tsvankins (1997b) anisotropic coefcients to the fracture
compliances and introduce practical fracture characterization
techniques operating with P- and PS-waves.
MODEL 1: ONE SET OF VERTICAL FRACTURES
IN A VTI BACKGROUND
Effective orthorhombic medium
A single system of vertical fractures embedded in a VTI
matrix yields an effective orthorhombic medium in which one
of the symmetry planes coincides with the fracture plane. Ac-
cording to the linear-slip theory (Schoenberg, 1980, 1983), the
effective compliance tensor of such a medium can be obtained
by simply adding the excess fracture compliances to the com-
pliance of the background. [Part I compares the linear-slip
theory with the effective medium theories of Hudson (1980,
1981, 1988) and Thomsen (1995).] The inversion of the result-
ing compliance tensor yields the stiffness tensor (and the cor-
responding two-index stiffness matrix) of the fracture-induced
orthorhombic model. If the fracture faces are perpendicular
to the x
1
-axis, the effective stiffness matrix c has the following
form (Schoenberg and Helbig, 1997; Appendix A):
c =
_
_
_
_
_
_
_
_
_
_
_
c
11
c
12
c
13
0 0 0
c
12
c
22
c
23
0 0 0
c
13
c
23
c
33
0 0 0
0 0 0 c
44
0 0
0 0 0 0 c
55
0
0 0 0 0 0 c
66
_
_
_
_
_
_
_
_
_
_
_
=
_
c
1
0
0 c
2
_
, (1)
where 0 is the 3 3 zero matrix and c
1
and c
2
are given by
c
1
=
_
_
_
_
_
_
_
_
_
_
c
11
b
(1
N
) c
12
b
(1
N
) c
13
b
(1
N
)
c
12
b
(1
N
) c
11
b
N
c
2
12
b
c
11
b
c
13
b
_
1
N
c
12
b
c
11
b
_
c
13
b
(1
N
) c
13
b
_
1
N
c
12
b
c
11
b
_
c
33
b
N
c
2
13
b
c
11
b
_
_
_
_
_
_
_
_
_
_
(2)
and
c
2
=
_
_
_
c
44
b
0 0
0 c
44
b
(1
V
) 0
0 0 c
66
b
(1
H
)
_
_
_. (3)
Here c
i j
b
are the stiffness coefcients of the VTI background
(constrained by c
12
b
=c
11
b
2c
66
b
);
N
,
V
, and
H
are the
dimensionless weaknesses of the fractures (Schoenberg and
Helbig, 1997; BakulinandMolotkov, 1998), whichchange from
zero (no fractures) to unity (extreme fracturing). The tan-
gential weaknesses
V
and
H
provide a measure of crack
density, whereas the normal weakness
N
contains informa-
tion about the uid content of the fractures and possible uid
ow between the fractures and pore space (Schoenberg and
Douma, 1988; part I).
Matrix (1) describes a special type of orthorhombic media
with the stiffnesses satisfying the relation (Schoenberg and
Helbig, 1997)
c
13
(c
22
+c
12
) = c
23
(c
11
+c
12
). (4)
The existence of the additional constraint (4) stems from the
fact that while general orthorhombic media are characterized
by nine independent values of c
i j
, the fracture-induced model
considered here is dened by only eight quantities (for a xed
fracture orientation): ve stiffness coefcients (c
11
b
, c
13
b
, c
33
b
,
c
44
b
, and c
66
b
) of the VTI background and three fracture weak-
nesses (
N
,
V
, and
H
).
The inversion of surface data for the physical parameters of
the fractures requires relating seismic signatures tothe fracture
weaknesses. The results of Grechka and Tsvankin (1999) and
Grechka et al. (1999) show that, in orthorhombic media, such
commonly used signatures as NMO velocities and AVO gradi-
ents are most concisely expressed through the dimensionless
anisotropic coefcients introduced by Tsvankin (1997b). The
denitions of Tsvankins parameters
(1,2)
,
(1,2,3)
, and
(1,2)
in
Fractured Orthorhombic Media 1805
terms of the stiffness elements c
i j
are given in Appendix B.
Substituting equations (1)(3) for the stiffnesses of the
fracture-induced orthorhombic model into equations (B-1)
(B-9) yields the anisotropic coefcients , , and as functions
of
N
,
V
, and
H
.
Weak-anisotropy approximation
The exact expressions for Tsvankins anisotropic parame-
ters in terms of the weaknesses, however, are too compli-
cated to give insight into the inuence of the fractures on
the effective anisotropic model. Therefore, we assume that
Thomsens (1986) coefcients
b
,
b
, and
b
of the VTI back-
ground, along with the weaknesses
N
,
V
, and
H
, are small
quantities of the same order. Linearizationof the effective stiff-
ness matrix in these small quantitites [equation (A-17)] shows
that for weak anisotropy the effective anisotropic coefcients
(1,2)
,
(1,2,3)
, and
(1,2)
should represent the sums of the VTI
background parameters
b
,
b
, and
b
and the corresponding
anisotropic coefcients of the HTI medium resulting from the
fractures in an imaginary isotropic host rock sufciently close
to the VTI background model. This isotropic medium can be
characterized, for instance, by the P- and S-wave velocities V
P
and V
S
equal to the vertical velocities V
P0
b
and V
S0
b
in the VTI
background.
Since this result provides a theoretical basis for most of the
subsequent discussion, we prove it for one of the anisotropic
coefcientsthe parameter
(2)
dened by equation (B-3).
Substituting the stiffnesses from equations (1) and (2) into
equation (B-3) yields
(2)
=
c
11
b
c
33
b

N
_
c
11
b

c
2
13
b
c
11
b
_
2c
33
b
_
1
N
c
2
13
b
c
11
b
c
33
b
_ . (5)
Using the denition of the coefcient
b
in the VTI background
medium [
b
(c
11
b
c
33
b
)/(2c
33
b
); see Thomsen (1986)], equa-
tion (5) can be rewritten as
(2)
=
N
c
2
11
b
c
2
13
b
2c
11
b
c
33
b
1
N
c
2
13
b
c
11
b
c
33
b
. (6)
Linearizing this equation with respect to
b
and
N
, we drop
the anisotropic term in the denominator to obtain
(2)

b
N
c
2
11
b
c
2
13
b
2c
11
b
c
33
b
. (7)
Since (c
2
11
b
c
2
13
b
)/(c
11
b
c
33
b
) is multiplied by the already small
weakness
N
, it can be replaced in the weak anisotropy ap-
proximation by the corresponding isotropic quantity 4g(1 g),
where g V
2
S0
b
/V
2
P0
b
is the ratio of the squared vertical S- and
P-wave velocities in the background. Thus, equation (7) can
be represented as
(2)

b
2g(1 g)
N
. (8)
The combination 2g(1 g)
N
can be recognized as the lin-
earization of the anisotropic coefcient
(V)
(R uger, 1997;
Tsvankin, 1997a) in an HTI mediumdue to a single set of verti-
cal fractures perpendicular to the x
1
-axis that are embedded in
a purely isotropic rock with the squared S-to-P velocity ratio
g (part I). Hence, equation (8) implies that
(2)

b
+
(V)
. (9)
Weconcludethat intheweakanisotropylimit theanisotropic
coefcient
(2)
reduces to the sum of the Thomsen background
parameter
b
and the coefcient
(V)
of the HTI model due to
the fractures embedded in an isotropic medium close enough
to the VTI background medium. Similar linearizations for the
other anisotropic parameters of the effective orthorhombic
model are given below.
Symmetry plane [x
2
, x
3
] parallel to the fractures.The lin-
earized anisotropic coefcients in the symmetry plane [x
2
, x
3
],
derived from equations (B-6)(B-8), have the following form:
(1)
=
b
, (10)
(1)
=
b
, (11)
and
(1)
=
b
+

V

H
2
. (12)
Therefore, for weak anisotropy coefcients
(1)
and
(1)
coin-
cide with those of the VTI background medium. This is an
expected result; if rotationally invariant fractures are intro-
duced into an isotropic matrix, the plane [x
2
, x
3
] represents the
so-called isotropy plane of the effective HTI medium. In the
isotropy plane the velocities of all waves are not inuenced
by the fractures and remain constant for all propagation direc-
tions (e.g., Tsvankin, 1997a). The coefcient
(1)
does not coin-
cide with
b
only because, in contrast to the more conventional
model from part I, the fractures considered here are not rota-
tionally invariant (i.e.,
V
=
H
). The difference between
V
and
H
, however, has no bearing on parameters
(1)
and
(1)
.
In addition to the basic set of anisotropic parameters intro-
duced by Tsvankin (1997b), it is useful to consider the anel-
lipticity coefcients
(1,2,3)
responsible for time processing of
P-wave data (Grechka and Tsvankin, 1999). For weak ani-
sotropy, the coefcient in the [x
2
, x
3
]-plane given by equa-
tion (B-10) is also equal to the background value:
(1)
=
b
. (13)
Symmetry plane [x
1
, x
3
] perpendicular to fractures.The
weak-anisotropy approximations of the anisotropic coef-
cients in the plane [x
1
, x
3
] are given by
(2)
=
b
2g(1 g)
N
, (14)
(2)
=
b
2g[(1 2g)
N
+
V
], (15)
(2)
=
b

H
2
, (16)
and
(2)
=
b
+2g[
V
g
N
]. (17)
Each of the expressions (14)(17) contains two terms with
a distinctly different physical meaning. The rst term is the
corresponding anisotropic coefcient of the VTI background,
1806 Bakulin et al.
while the second term absorbs the inuence of the fractures.
As discussed above for
(2)
, the fracture-related terms in equa-
tions (14)(17) are approximately equal (assuming
H
=
V
)
tothecoefcients
(V)
,
(V)
,
(V)
, and
(V)
(respectively) derived
inpart I for afractureset embeddedinanisotropic background.
The constraint on the effective stiffnesses [equation (4)]
leads to an additional relationship between the anisotropic co-
efcients. In the weak anisotropy limit, equation (4) can be
rewritten as
(2)
(1)
=
1
4g
_
(2)
(1)
(2)
(1)
_
1 2g
1 g
_
. (18)
It is interesting that the background parameters and the tan-
gential compliance
H
have no inuence on constraint (18),
which depends only on the differences between the anisotropic
coefcients in the vertical symmetry planes, i.e., on
(V)
,
(V)
,
and
(V)
. As a result, equation (18) is analogous to the con-
straint given by Tsvankin (1997a) and part I for HTI media
due to rotationally invariant fractures (i.e., for
H
=
V
).
Horizontal symmetry plane [x
1
, x
2
].The only Tsvankins
(1997b) anisotropic coefcient dened in the horizontal plane
is
(3)
[equation (B-9)]. After linearization, it becomes
(3)
= 2g[
N

H
]. (19)
The parameter
(3)
does not contain any background aniso-
tropic coefcients because [x
1
, x
2
] is the isotropy plane of the
VTI medium. Since
(3)
is dened with respect to the x
1
-axis,
which is normal to the fractures, equation (19) coincides with
the expression for the generic Thomsen coefcient obtained
for VTI media due to horizontal fractures by Schoenberg and
Douma (1988).
The linearized anellipticity coefcient
(3)
[equation (B-12)]
has the form
(3)
= 2g[
H
g
N
]. (20)
Estimation of the anisotropic parameters from reection data
Inversion of reection data for the effective parameters
of orthorhombic media is discussed by Grechka and Tsvankin
(1998, 1999), R uger (1998), and Grechka et al. (1999). We now
give a brief overview of their parameter-estimation methods
operating with either wide-azimuth P-wave data or a combina-
tion of azimuthally dependent signatures of P- and PS-waves.
P-wave signatures.NMO velocity of P-waves in a hori-
zontal orthorhombic layer is described by an ellipse with the
axes in the vertical symmetry planes (Grechka and Tsvankin,
1998). Since for both orthorhombic models considered here
the orientation of the symmetry planes is determined by
the strike of the fractures, the axes of the P-wave NMO el-
lipse yield the fracture azimuth(s). In the orthorhombic model
due to a single fracture set,
(1)
>
(2)
[see equation (32)] and
the semimajor axis of the P-wave NMO ellipse points in the
direction of the fracture plane. This result holds for horizontal
transverse isotropy as well (part I).
If
(1)
=
(2)
and the P-wave NMOvelocity froma horizontal
reector is azimuthally independent (i.e., the ellipse degener-
ates into a circle), the azimuths of the symmetry planes can be
obtained from the NMO ellipse of a dipping event (Grechka
and Tsvankin, 1999). P-wave reection traveltimes from dip-
ping interfaces or the azimuthal variation of nonhyperbolic
moveout (Al-Dajani et al., 1998) can also be inverted for the
anellipticity coefcients
(1)
,
(2)
, and
(3)
[equations (13), (17),
and (20)]. Nonnegligible values of both
(1)
and
(2)
help to
detect the presence of anisotropy in the background and dis-
criminate between HTI and orthorhombic models. For HTI
media, all anisotropic coefcients (including ) in one of the
symmetry planes should go to zero.
The semiaxes V
(1)
P,nmo
and V
(2)
P,nmo
of the P-wave NMO el-
lipse from a horizontal reector are given by (Grechka and
Tsvankin, 1998)
V
(i )
P,nmo
= V
P0
_
1 +2
(i )
, (i = 1, 2). (21)
Hence, V
(i )
P,nmo
can be combined with the vertical velocity V
P0
to determine
(1)
and
(2)
. If V
P0
is unknown, the P-wave NMO
ellipse constrains the difference between the two coef-
cients:

_
V
(2)
P,nmo
_
2
_
V
(1)
P,nmo
_
2
_
V
(2)
P,nmo
_
2
+
_
V
(1)
P,nmo
_
2
=

(2)
(1)
1 +
(2)
+
(1)

(2)
(1)
.
(22)
Additional informationfor parameter estimationis provided
by prestack amplitudes of P-waves. In the weak-anisotropy
approximation, the variation of the P-wave amplitude varia-
tion with offset (AVO) gradient between the vertical symmetry
planes is governed by the expression
(2)
(1)
8g(
(2)
(1)
)
(R uger, 1998). Therefore, if the squared velocity ratio g is
known and
(2)
(1)
has been found from the P-wave NMO
ellipse [equation (22)], the AVO gradient yields an estimate of
(2)
(1)
.
P- and PS-wave signatures.Although P-wave data alone
may be used to estimate a subset of the effective parameters of
orthorhombic media, it is highly benecial to combine P-wave
traveltimes or amplitudes with the signatures of shear or con-
verted waves. Since S-waves are not generated in most explo-
ration surveys, we emphasize the joint inversion of P-waves
and P- to S-mode conversions. The vertical traveltimes of P-
and PS-waves give a direct estimate of the vertical-velocity
ratio needed to compute the fracture weaknesses. The shear-
wave splitting parameter at vertical incidence, conventionally
evaluated fromthe time delays between the fast and slowshear
or converted waves, is close to
(2)
(1)
. Also, the difference
between the AVOgradients of the PS-wave in the vertical sym-
metry planes can be combined with the azimuthal variation of
the P-wave AVO gradient or the P-wave NMO ellipse to ob-
tain another estimate of
(2)

(1)
. A more detailed discus-
sion of the azimuthal AVO inversion of PS-waves is given in
part I.
Grechka et al. (1999) outline the following methodology of
the joint moveout inversion of P- and PS-waves. Similar to
pure modes, the azimuthally varying NMO velocity of either
split PS-wave in a horizontal orthorhombic layer is described
by an ellipse aligned with the vertical symmetry planes. The
semiaxes of the NMOellipses of the P- and two split PS-waves
(i.e., the symmetry-plane NMO velocities) can be used to re-
construct the NMO ellipses of the pure shear waves S
1
and S
2
.
Assuming that wave S
1
represents an SV (in-plane polarized)
mode in the [x
2
, x
3
]-plane [the superscript (1)], the semiaxes of
the shear-wave ellipses are expressed as
V
(2)
S1, nmo
= V
S1
_
1 +2
(2)
= V
(1)
S2, nmo
, (23)
V
(1)
S1, nmo
= V
S1
_
1 +2
(1)
, (24)
V
(2)
S2, nmo
= V
S2
_
1 +2
(2)
, (25)
and
V
(1)
S2, nmo
= V
S2
_
1 +2
(1)
= V
(2)
S1, nmo
. (26)
Here
(1)
and
(2)
are the anisotropic coefcients
(1)
_
V
P0
V
S1
_
2
_
(1)
(1)
_
(27)
and
(2)
_
V
P0
V
S2
_
2
_
(2)
(2)
_
, (28)
and V
S1
and V
S2
are the vertical velocities of the split shear
waves S
1
and S
2
:
V
S1
= V
S2
_
1 +2
(1)
1 +2
(2)
(29)
and
V
S2
= V
S0
. (30)
If one of the vertical velocities or the reector depthis known
and the anisotropic parameters
(1,2)
have been obtained from
P-wave moveout, the shear-wave NMO ellipses make it pos-
sible to nd four additional coefcients
(1,2)
and
(1,2)
. The
only anisotropic parameter not constrained by conventional-
spread normal moveout in a horizontal orthorhombic layer is
(3)
.
Estimation of fracture parameters
Inversion for the weaknesses.Since we are mostly inter-
ested in evaluating the properties of the fracture set, it is con-
venient to remove the inuence of the background medium at
the outset of the inversion procedure. Inspection of equations
(10)(17) shows that the contribution of the VTI background
parameters can be eliminated by computing the difference be-
tween the anisotropic coefcients in the planes parallel and
orthogonal to the fractures:
(2)
(1)
= 2g(1 g)
N
, (31)
(2)
(1)
= 2g[(1 2g)
N
+
V
] , (32)
(2)
(1)
=
V
2
, (33)
(2)
(1)
= 2g[
V
g
N
]. (34)
Equations (31)(34) are identical to the expressions for
(V)
,
(V)
,
(V)
, and
(V)
(respectively) in HTI media due to a single
set of vertical rotationally invariant fractures with the weak-
nesses
N
and
V
(part I). Therefore, the weaknesses can be
determined from equations (31)(34) using the HTI expres-
sions described in detail in part I. We emphasize that obtaining
N
and
V
requires knowledge of any two of the differences
(2)
(1)
,
(2)
(1)
,
(2)
(1)
, and
(2)
(1)
.
As discussed,
(2)
(1)
can be estimated from the elon-
gation of the P-wave NMOellipse for horizontal events [equa-
tion(22)]. Also,
(2)
and
(1)
canbe obtainedusing the NMOve-
locities of dipping P-events or nonhyperbolic moveout. Thus,
P-wave moveout data provide sufcient information to deter-
mine the weaknesses
N
and
V
fromequations (32) and (34):
N
=
_
(2)
(1)
_
+
_
(2)
(1)
_
2g(1 g)
, (35)
V
=
1
2(1 g)
_
1 2g
g
_
(2)
(1)
_
(2)
(1)
_
_
.
(36)
In principle, the inversion based on equations (35) and (36)
requires knowledge of the squared vertical-velocity ratio g,
which cannot be found without shear or converted-wave data.
However, numerical tests (and the results of part I) show that
even a relatively rough estimate of g is sufcient for recovering
the weaknesses with acceptable accuracy.
P-wave moveout data may provide not just the difference
between
(2)
and
(1)
but also the individual values of these co-
efcients. Therefore, in addition to substituting
(2)
(1)
into
the equations for the weaknesses, we can use the approximate
relation (13),
(1)
=
b
, to determine the coefcient in the
background VTI model.
If the P-wave moveout information is limited to the NMO
ellipse from a horizontal reector, it is possible to obtain the
difference
(2)
(1)
by including the P-wave AVO gradients
in the directions parallel and perpendicular to the fractures.
The algorithm based on the NMO ellipses of horizontal events
and AVO gradients is identical to the one described in part I
for HTI media. The presence of anisotropy in the background
has no inuence either on estimating
(2)
(1)
and
(2)
(1)
or on inverting these differences [equations (32) and (33)] for
the weaknesses
N
and
V
.
In multicomponent surveys, the vertical traveltimes of PS-
waves (in combination with P-wave data) provide estimates of
g and the splitting coefcient
(2)
(1)
. Thus, another possi-
ble set of input parameters includes the P-wave NMO ellipse
from a horizontal reector (yielding
(2)
(1)
) and the vertical
traveltimes of converted modes.
Note that the weakness
H
does not enter any of the differ-
ences (31)(34), despite the fact that it appears in equations
(12) and (16) for
(1)
and
(2)
. However, even the individual
values of
(1)
and
(2)
constrain the combination
b
H
/2 but
not
H
separately. The only source of information about
H
is
(3)
or
(3)
[equations (19) and (20)], which can be estimated
using the NMO velocities of P-wave reections from dipping
interfaces (Grechka and Tsvankin, 1999):
H
=

(3)
2g
+ g
N
. (37)
If dipping events are not available, the data cannot be in-
verted for the coefcient
(3)
and, therefore, for the weakness
H
. In this case, a reasonable simplifying assumption is that
the fractures are rotationally invariant and
V
=
H
. Then the
number of independent medium parameters reduces to seven,
and
(3)
and
(3)
can be expressed as
(3)
=
(2)
(1)
2
_
(2)
(1)
_
, (38)
(3)
=
(2)
(1)
. (39)
1808 Bakulin et al.
Interpretation of the weaknesses in terms of the physical
properties of the fractures is discussed in part I. For rotation-
ally invariant penny-shaped cracks, the tangential weakness
V
=
H
gives an estimate of the crack density, while the nor-
mal weakness
N
represents a sensitive, albeit nonunique, in-
dicator of uid content.
Numerical examples.Although the weak-anisotropy ap-
proximations provide useful insight into the parameter esti-
mation problem, it is preferable to use the exact equations for
actual inversion. In the examples below, we compute the frac-
ture weaknesses
N
,
V
, and
H
, along with the anisotropic
coefcient
b
of the background, by inverting the values of
[equation (22)],
(1)
,
(2)
, and
(3)
presumably extracted from
P-wave seismic data. The goal of this numerical study is to ex-
amine the sensitivity of the inverted parameters to errors in the
input data and in the parameters of the background medium.
Our nonlinear inversionalgorithmis basedonequations (1)
(3) for the stiffness elements and on the exact expressions for
the anisotropic coefcients given in Appendix B. The weak-
anisotropy approximations (13) and (35)(37) are used only
to obtain the initial guesses for the weaknesses and the co-
efcient
b
. The parameters of the background medium are
g =V
2
S0b
/V
2
P0b
=0.25,
b
=0.2, and
b
= 0.1(
b
canbeexpressed
through
b
and
b
). Although the only background parame-
ter in equations (35)(37) is the squared velocity ratio g, the
anisotropic coefcients
b
and
b
of the VTI medium are con-
tained in the exact equations for the stiffnesses and effective
anisotropic parameters and must be specied for the inversion.
FIG. 1. Numerical inversion of ,
(1)
,
(2)
, and
(3)
for the fracture weaknesses
N
(dotted line),
V
(dash-dotted),
H
(dashed),
and the coefcient
b
of the background medium (solid). Errors of 0.1 were introduced in (a) , (b)
(1)
, (c)
(2)
, and (d)
(3)
,
with the other input parameters held at the correct values. In the absence of errors, the inversion yields the correct values of the
weaknesses and
b
(large dots).
As shown below, however, the inversion results are insensitive
to variations of g,
b
, and
b
within the range important in prac-
tice. The fracture set is dened by the weaknesses
N
=0.5 and
V
=
H
= 0.2, which approximately correspond to values for
penny-shaped gas-lled cracks (see part I).
The inversion results are displayed in Figure 1. For the cor-
rect input parameters (and the correct background parameters
g,
b
, and
b
), the inversion algorithmproduces accurate values
of the weaknesses and
b
. Errors in the measured anisotropic
parameters of up to 0.1 result in similar (often smaller) errors
in the inverted values, indicating that the parameter estimation
procedure is reasonably stable. As expected from the analytic
results, different inverted parameters are most sensitive to er-
rors in different measured quantities. For example, errors in
produce comparable distortions in all three weaknesses, reach-
ing and sometimes exceeding 0.1 (Figure 1a). The anelliptic-
ity coefcient
b
is quite sensitive to errors in
(1)
(Figure 1b)
and is almost independent of the other input parameters, in
accordance with the weak-anisotropy approximation (13). Er-
rors in
(1)
and
(2)
of about 0.1 cause smaller errors (0.05)
in the tangential weaknesses
V
and
H
than in the normal
weakness
N
(0.1; Figures 1b,c). The weakness
H
is sen-
sitive primarily to errors in
(3)
(Figure 1d), as suggested by
equation (37). In the weak-anisotropy limit,
N
and
V
are in-
dependent of
(3)
[equations (35) and (36)], which is conrmed
by Figure 1d.
In the second test, we examine the inuence of errors in
the background parameters g,
b
, and
b
(which were assumed
known a priori in the previous example) on the inversion
results. Substantial errors in g of 0.1, or up to 40%, lead
to relatively small distortions of only about 0.03 in all in-
verted parameters except
N
(for
N
the errors reach 0.1;
Figure 2a). The anisotropic coefcients
b
and
b
have an even
smaller inuence on the estimated quantities, especially on the
tangential compliances (Figure 2b,c). This result is especially
encouraging because
b
and
b
are difcult to estimate from
surface seismic data without information about the vertical ve-
locities or reector depth.
Estimation of background parameters.As discussed
above, theindividual values of theanisotropic parameters
(1,2)
,
(1,2)
, and
(1,2)
can be found by combining P-wave moveout
data with the reection traveltimes of PS-waves (provided the
reector depthis known). Thenit is possible toestimate not just
the fracture weaknesses but also the squared vertical-velocity
ratio g and the background anisotropic parameters
b
,
b
, and
b
[equations (10)(12); to nd
b
, we assume
V
=
H
].
For the numerical example in Figure 3, the input data in-
cluded the vertical and NMOvelocities of P- and split S-waves,
where the shear-wave signatures were supposedly determined
from P and PS data. The NMO velocity of each mode was
computed in three azimuthal directions with a step of 45
and
approximated with an ellipse to nd the NMO velocities in the
symmetry planes. Then the vertical velocities and symmetry-
plane NMOvelocities were combined to determine
(1,2)
,
(1,2)
,
FIG. 2. Inuence of the background parameters (a) g =V
2
S0
b
/V
2
P0
b
, (b)
b
, and (c)
b
on the inversion results from Figure 1 for the
correct values of and
(1,2,3)
. Dotted line
N
; dash-dotted
V
; dashed
H
; solid
b
.
and
(1,2)
from equations (21) and (23)(26). To check the in-
uence of measurement errors, we added Gaussian noise with
a standard deviation of 2% to the vertical and NMO velocities
and performed the inversion for 200 sets of the distorted input
parameters. The algorithm is based on the exact expressions
for the anisotropic coefcients derived fromequations (1)(3),
with the starting model computed using the weak-anisotropy
approximation. Since the inversion errors in Figure 3 are lim-
ited by 0.05, the estimation of both weaknesses and all back-
ground parameters is sufciently stable.
FIG. 3. Inversion of the vertical and NMO velocities of the P-
and split S-waves for the parameters of the fractures and VTI
background. The dots mark the correct values of the parame-
ters; it is assumed that
V
=
H
. The bars correspond to one
standard deviation in the inverted quantities caused by
Gaussian noise in the input data with a standard deviation
of 2%.
1810 Bakulin et al.
MODEL 2: TWO ORTHOGONAL FRACTURE
SETS IN ISOTROPIC ROCK
Another practically important fractured model with or-
thorhombic symmetry is composed of two orthogonal vertical
fracture sets embedded in an isotropic or VTI background. To
simplify the parameter estimation procedure, we assume that
the fractures are rotationally invariant andthe backgroundma-
trixis purelyisotropic. Choosingthe x
1
-axis tobeperpendicular
to the rst set of fractures, we obtain the compliance matrices
for both fracture systems in equations (A-3) and (A-14) where,
because of the rotational invariance, the compliances satisfy
K
V1
=K
H1
K
T1
and K
V2
=K
H2
K
T2
. Building the effective
matrix of the excess compliance [equation (A-12)] and adding
the compliance of the isotropic background yields the effective
stiffness matrix [equation (A-1)]
c =
_
_
_
_
_
_
_
_
_
_
_
c
11
c
12
c
13
0 0 0
c
12
c
22
c
23
0 0 0
c
13
c
23
c
33
0 0 0
0 0 0 c
44
0 0
0 0 0 0 c
55
0
0 0 0 0 0 c
66
_
_
_
_
_
_
_
_
_
_
_
=
_
_
c
1
0
0

c
2
_
_
,
(40)
where 0 is the 3 3 zero matrix and the matrices

c
1
and

c
2
are
given by
c
1
=
1
d
_
_
_
( +2)l
1
m
3
l
1
m
1
l
1
m
2
l
1
m
1
( +2)l
3
m
1
l
2
m
1
l
1
m
2
l
2
m
1
( +2)(l
3
m
3
l
4
)
_
_
_,
(41)
c
2
=
_
_
_
_
_
(1
T2
) 0 0
0 (1
T1
) 0
0 0
(1
T1
)(1
T2
)
(1
T1
T2
)
_
_
_
_
_
.
(42)
Here, and are the Lam e parameters of the background
medium and
l
1
= 1
N1
, l
2
= 1 r
N1
, l
3
= 1 r
2
N1
,
l
4
= 4r
2
g
2
N1
N2
,
m
1
= 1
N2
, m
2
= 1 r
N2
, m
3
= 1 r
2
N2
,
(43)
g = /( +2) = V
2
S
_
V
2
P
, r = 1 2g,
d = 1 r
2
N1
N2
.
The values
Ni
and
Ti
(i =1, 2) are the normal and tangential
fracture weaknesses (see part I) related to the fracture compli-
ances by the equations
Ni
=
K
Ni
( +2)
1 + K
Ni
( +2)
(44)
and
Ti
=
K
Ti
1 + K
Ti
, (45)
which represent a special case (valid for an isotropic back-
ground) of equations (A-4).
Since both fracture planes and the horizontal plane consti-
tute three orthogonal planes of mirror symmetry, the effective
mediummust be orthorhombic with the nine stiffness elements
[equation (40)]. Our model, however, represents a special case
of general orthorhombic media with only six independent pa-
rameters: , ,
N1
,
T1
,
N2
, and
T2
. As follows from equa-
tions (41) and (42), the three additional relationships (con-
straints) between the stiffnesses are
c
12
(c
33
+c
23
) = c
13
(c
22
+c
23
), (46)
2 (c
11
+c
13
) =
_
c
11
c
33
c
2
13
_
_
c
44
+c
55
c
44
c
55
1
c
66
_
, (47)
and
2(c
22
+c
23
) =
_
c
22
c
33
c
2
23
_
_
c
44
+c
55
c
44
c
55
1
c
66
_
. (48)
Models with two identical orthogonal fracture sets, character-
ized by fewer independent parameters, are discussed in Ap-
pendix C.
Anisotropic coefcients in the weak-anisotropy limit
Equations (40)(43), combined with the denitions from
Appendix B, can be used to express Tsvankins (1997b)
anisotropic coefcients in terms of the fracture weaknesses
Ti
and
Ni
. Here we restrict ourselves to linearized expressions
obtained in the limit of small fracture compliances
T1,2
1
and
N1,2
1. The result of this linearization can be predicted
from equation (A-13): since in the linear approximation frac-
ture systems are not inuenced by each other, the effective or-
thorhombic medium is composed of two HTI media produced
by each fracture set individually.
Symmetry plane [x
2
, x
3
].The anisotropic parameters with
the superscript (1) are dened in the symmetry plane [x
2
, x
3
],
which is parallel to the rst set of fractures and orthogo-
nal to the second one. In the absence of the second set, the
[x
2
, x
3
]-plane would coincide with the isotropy plane of the
HTI medium associated with the rst fracture system. There-
fore, we can expect these parameters to be largely inuenced
by the second set of fractures orthogonal to the x
2
-axis. In-
deed, the linearized , , , and coefcients in the [x
2
, x
3
]
plane depend only on the weaknesses
N2
and
T2
:
(1)
= 2g(1 g)
N2
, (49)
(1)
= 2g[(1 2g)
N2
+
T2
], (50)
(1)
=
T2
2
, (51)
(1)
= 2g[
T2
g
N2
], (52)
where g V
2
S
/V
2
P
is the ratio of the squared S- and P-wave ve-
locities in the background. Equations (49)(52) coincide with
the expressions for the anisotropic coefcients
(V)
,
(V)
,
(V)
,
and
(V)
of the HTI model associated with the second fracture
set.
Symmetry plane [x
1
, x
3
].Likewise, the linearized aniso-
tropic coefcients in the symmetry plane [x
1
, x
3
] are governed
by the weaknesses of the rst fracture set. As follows from the
symmetry of the model, the expressions for
(2)
,
(2)
,
(2)
, and
(2)
are fully analogous to equations (49)(52):
(2)
= 2g(1 g)
N1
, (53)
(2)
= 2g[(1 2g)
N1
+
T1
], (54)
(2)
=
T1
2
, (55)
(2)
= 2g[
T1
g
N1
]. (56)
Symmetry plane [x
1
, x
2
].The remaining anisotropic coef-
cient
(3)
dened in the horizontal symmetry plane [x
1
, x
2
] is
given by
(3)
= 2g[
N1
T1
] 2g[(1 2g)
N2
+
T2
]. (57)
The linearized parameter
(3)
is equal to the sum of the generic
Thomsen coefcient caused by horizontal fractures in VTI
media (the term 2g[
N1
T1
]; see Schoenberg and Douma,
1988) and the HTI coefcient
(V)
is due to vertical fractures
(part I). To explain this result, recall that
(3)
is dened with
respect tothe x
1
-axis, whichis orthogonal totherst fractureset
(so this set becomes horizontal if the x
1
-axis is made vertical in
VTI media) and lies in the planes of the second set of fractures
(making this set vertical).
The anellipticity coefcient
(3)
[equation (B-12] in the hor-
izontal plane is
(3)
=
(1)
+
(2)
. (58)
Relations between anisotropic coefcients.The nine stiff-
nesses of the effective orthorhombic model contain only six in-
dependent quantities, whichleads tothe three constraints (46)
(48). Rewriting equations (46) and (47) through Tsvankins
(1997b) parameters yields
(i )
=
1
4g
_
(i )
(i )
1 2g
1 g
_
, (i = 1, 2), (59)
where quadratic and higher-order terms in the anisotropic co-
efcients were dropped. The same result can be obtained from
the linearized expressions for the anisotropic coefcients given
above. Equation (59) is identical to the relationship between
(V)
,
(V)
, and
(V)
of HTI media (Tsvankin, 1997a; part I).
The remaining constraint (48) takes the form
(3)
=
(1)
+
(2)
2
(2)
. (60)
Equations (59) and (60) show that in Tsvankins notation our
orthorhombic model can be fully described by the two vertical
velocities and four anisotropic coefcients
(1,2)
and
(1,2)
.
In the special case of penny-shaped gas-lled fractures, the
normal and tangential compliances are equal to each other
(K
N1
=K
T1
and K
N2
=K
T2
) and, as noticed by Schoenberg and
Sayers (1995), the anellipticity parameters in the symmetry
planes go to zero:
(1)
=
(2)
=
(3)
= 0. (61)
The structure of equations (49)(56) for the anisotropic co-
efcients suggests that estimation of the weaknesses of the
orthogonal fracture sets can be decomposed into two HTI-
type inversions in the vertical symmetry planes discussed in
part I. Here, we verify the accuracy of the weak-anisotropy ap-
proximations and outline two possible strategies for obtaining
the weaknesses from surface reection data. One is based on
azimuthally dependent P-wave reection moveout alone; the
other uses both P- and converted-wave data. In contrast to the
model with a single fracture set, we must know one of the ver-
tical velocities (or reector depth) because the inversion algo-
rithm requires the individual values of Tsvankins anisotropic
coefcients rather than their differences.
P-wave inversion.As for the model with a single fracture
set, the fracture orientation can be determined directly from
the P-wave NMO ellipse (or the NMO ellipses of converted
waves), unless the ellipse degenerates into a circle. This and
some other special cases are discussed in Appendix D. Suppose
the parameters
(1)
and
(2)
have been found using the semiaxes
of the P-wave NMOellipse froma horizontal reector and the
vertical velocity [equations (21)]. Combining the coefcients
with the anellipticity parameters
(1)
and
(2)
(also determined
from P-wave moveout) allows us to solve equations (50), (52),
(54), and (56) for the weaknesses
N1
=
(2)
+
(2)
2g(1 g)
, (62)
T1
=
1
2(1 g)
_
1 2g
g

(2)
(2)
_
, (63)
N2
=
(1)
+
(1)
2g(1 g)
, (64)
and
T2
=
1
2(1 g)
_
1 2g
g

(1)
(1)
_
. (65)
The equations for eachpair of weaknesses [(62) (63) and(64)
(65)] are identical to those obtained in part I for HTI media
due to a single set of rotationally invariant fractures.
To test the accuracy of the weak anisotropy approximations
(62)(65), we computed the exact anisotropic coefcients
(1,2)
and
(1,2)
for three fractured models in Table 1 [using equations
(40)(42) and the denitions from Appendix B] and inverted
them for the weaknesses in the limit of weak anisotropy. Table
1 shows that approximations (62)(65) give reasonably good
estimates of the fracture weaknesses. The only signicant er-
ror, in the tangential weakness of the second fracture set, is
unrelated to the rst set because it remains the same for the
corresponding HTI model (third row in Table 1).
Inversion of P- and PS-wave data.If dipping events are
not available and CMP spreads are not sufciently long for
using nonhyperbolic moveout, it may be possible to nd
the anisotropic coefcients
(1,2)
,
(1,2)
, and
(1,2)
by combin-
ing P- and PS-wave data [equations (23)(26)]. Since these
anisotropic parameters depend on only four fracture weak-
nesses, P- and PS-wave data (including the vertical veloci-
ties or reector depth) provide useful redundancy in the in-
version procedure. In the weak-anisotropy approximation, the
1812 Bakulin et al.
weaknesses can be computed, for example, as
T1
= 2
(2)
= 1
_
V
(2)
S1,nmo
V
S1
_
2
, (66)
T2
= 2
(1)
= 1
_
V
(1)
S2,nmo
V
S2
_
2
, (67)
N1
=
1
1 2g
_
T1
+

(2)
2g
_
, (68)
N2
=
1
1 2g
_
T2
+

(1)
2g
_
. (69)
Figure 4 shows the results of numerical inversion based
on NMO equations (23)(26) and the exact expressions for
Tsvankins anisotropic coefcients in terms of the fracture
weaknesses. Similar to the numerical example in Figure 3,
the vertical velocities and azimuthally varying NMO veloci-
ties of P- and S-waves (input data) were distorted by Gaussian
noise with a standard deviation of 2%. The initial model was
found from the weak-anisotropy approximations (66)(69).
The standard deviations for the inverted weaknesses are rela-
tively small, with errors in the tangential compliances
T1
and
T2
being somewhat lower than those in
N1
and
N2
.
Applying the linear-slip theory developed by
Schoenberg (1980, 1983), we studied two types of orthorhom-
bic media believed to be representative of naturally fractured
reservoirs. The rst model contains a single set of vertical frac-
tures embedded in a VTI background (e.g., the background
Table 1. Comparison of the actual fracture parameters with
those estimated by inverting the exact anisotropic coefcients
[equations (40)(42) and Appendix B] using the linearized
equations (62)(65); the squared velocity ratio g = 0.25. The
rst model is composed of two orthogonal fracture sets in an
isotropic background; thesecondandthirdmodels containonly
one fracture set (i.e., they have the HTI symmetry).
Model Fracture parameters
N1

T1

N2

T2
1 Actual 0.30 0.15 0.60 0.30
Estimated 0.28 0.14 0.66 0.21
2 Actual 0.30 0.15 0 0
Estimated 0.30 0.14 0 0
3 Actual 0 0 0.60 0.30
Estimated 0 0 0.67 0.21
FIG. 4. Inversion of the P- and S-wave vertical and NMO ve-
locities for the model with two orthogonal fracture sets. The
input data were contaminated by Gaussian noise with a stan-
dard deviation of 2%, and the inversion was repeated 200 times
for different realizations of the input parameters. The bars cor-
respond to one standard deviation in the inverted quantities.
anisotropy may result from ne layering), while the second is
produced by two orthogonal systems of rotationally invariant
vertical fractures in an isotropic host rock. For both effective
models we obtained Tsvankins (1997b) anisotropic parame-
ters, whichcapturethecombinations of thestiffness coefcients
responsible for commonly measured seismic signatures.
To gain a better understanding of the inuence of fractures
on the effective medium, we simplied the anisotropic param-
eters under the assumption of weak background and fracture-
induced anisotropy. For the model with a single fracture set,
the anisotropic parameters of the HTI mediumdue to the frac-
tures are added to the background coefcients to produce the
linearized effective parameters of the orthorhombic medium.
Therefore, by computing the difference between the effective
coefcients dened in the vertical symmetry planes, we elim-
inate the inuence of the background and reduce the inverse
problemto that for horizontal transverse isotropy (part I). The
information necessary for this inversion procedure can be ob-
tained from azimuthally dependent P-wave reection travel-
times alone if dipping events or nonhyperbolic moveout are
available (also, it is necessary to have an estimate of the ra-
tio of the P- and S-wave vertical velocities). Alternatively, the
inversioncanbe performedby combining the P-wave NMOel-
lipse froma horizontal reector with other data, such as the az-
imuthally varying P-wave AVO gradient or the vertical travel-
times and, possibly, NMOvelocities of the split converted (PS)
modes. For weak anisotropy, the algorithm based on the NMO
ellipses and AVO gradients of P- and PS-waves reected from
horizontal interfaces is identical to that outlined in part I for
HTI media (i.e., it is independent of the presence of anisotropy
in the background).
In the case of two orthogonal fracture sets, the linearized
expressions for the effective anisotropic coefcients in each
vertical symmetry plane contain only the contribution of the
fractures orthogonal to this plane. As a result, the inversion
for the fracture weaknesses splits into two separate inversion
procedures in the symmetry planes, which can be carried out
using the HTI algorithm of part I. In contrast to the model
with a single fracture set, however, determination of the weak-
nesses of both fracture sets requires knowledge of the vertical
velocities in addition to the azimuthally varying surface seismic
signatures.
Although both effective media considered here have or-
thorhombic symmetry, the results of our analysis indicate sev-
eral possible ways to identify the correct underlying physical
model. In both cases, the stiffness tensor is described by fewer
independent parameters thanfor general orthorhombic media,
and the anisotropic coefcients for the models with one and
two fracture sets satisfy different constraints. Another useful
criterion is the sign of the anisotropic parameters. The coef-
cients
(1,2)
,
(1,2)
, and
(1,2)
for the model with two orthogonal
systems of fractures in an isotropic background are negative
[equations (49)(51) and (53)(55)]. In contrast, and (and
often as well) dened in the fracture plane of the model with
a single fracture set in a VTI background are usually positive.
Also, we can distinguish between the two models by compar-
ing the polarization direction of the vertically traveling fast
shear wave with the orientation of the semimajor axis of the P-
wave NMO ellipse. For one set of fractures (in either isotropic
or VTI background), they always coincide with each other;
for two systems of fractures with different uid content, they
may become orthogonal. This happens, for example, if the two
inequalities
T1
>
T2
and(1 2g)
N1
+
T1
<(1 2g)
N2
+
T2
(or, equivalently,
(2)
>
(1)
) are satised simultaneously.
ACKNOWLEDGMENTS
This research was carried out during a visit of Andrey
Bakulin to the Center for Wave Phenomena (CWP), Col-
orado School of Mines, 1998. We are grateful to members of
the A(nisotropy)-team of CWP for helpful discussions and to
Andreas R uger (Landmark) for his review of the manuscript.
The support for this work was provided by the members of
the Consortium Project on Seismic Inverse Methods for Com-
plex Structures at CWP and by the U.S. Department of Energy
(award #DE-FG03-98ER14908).
REFERENCES
Al-Dajani, A., Tsvankin, I., and Toks oz, N., 1998, Nonhyperbolic
reection moveout in azimuthally anisotropic media: 68th Ann.
Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 1479
1482.
Alkhalifah, T., and Tsvankin, I., 1995, Velocity analysis in transversely
isotropic media: Geophysics, 60, 15501566.
Bakulin, A. V., and Molotkov, L. A., 1998, Effective models of frac-
tured and porous media: St. Petersburg Univ. Press (in Russian).
Bakulin, A., Tsvankin, I., and Grechka, V., 2000, Estimation of fracture
parameters from reection seismic dataPart I: HTI model due to a
single fracture set: Geophysics, 65, 17881802, this issue.
1999, 3-D moveout velocity analysis and parameter estimation
for orthorhombic media: Geophysics, 64, 820837.
Grechka, V., Theophanis, S., and Tsvankin, I., 1999, Joint inversion of
P- and PS-waves in orthorhombic media: Theory and a physical-
modeling study: Geophysics, 64, 146161.
Camb. Phil. Soc., 88, 371384.
Molotkov, L. A., and Bakulin, A. V., 1997, An effective model of a
R uger, A., 1997, P-wave reection coefcients for transversely
isotropic models with vertical and horizontal axis of symmetry: Geo-
physics, 62, 713722.
1998, Variation of P-wave reectivity with offset and azimuth
in anisotropic media: Geophysics, 63, 935947.
1983, Reection of elastic waves from periodically stratied
media with interfacial slip: Geophys. Prosp., 31, 265292.
Schoenberg, M., and Douma, J., 1988, Elastic wave propagation in me-
dia with parallel fractures and aligned cracks: Geophys. Prosp., 36,
571590.
Schoenberg, M., and Helbig, K., 1997, Orthorhombic media: Modeling
elastic wave behavior in a vertically fractured earth: Geophysics, 62,
19541974.
1966.
Tsvankin, I., 1997a, Reection moveout and parameter estimation for
horizontal transverse isotropy: Geophysics, 62, 614629.
1997b, Anisotropic parameters and P-wave velocity for or-
thorhombic media: Geophysics, 62, 12921309.
Winterstein, D. F., 1990, Velocity anisotropy terminology for geophysi-
cists: Geophysics, 55, 10701088.
APPENDIX A
COMPLIANCE FORMALISM FOR FRACTURED MEDIA
Here we review both exact and approximate methods of
obtaining effective parameters of fractured media using the
results of Schoenberg (1980, 1983), Schoenberg and Douma
(1988), Schoenberg and Muir (1989), Nichols et al. (1989), and
Molotkov and Bakulin (1997). A discussion of different ap-
proaches to effective medium theory for fractured models can
be found in part I.
To simplify the derivation of the effective elastic parameters
of fractured media, it is convenient to use the compliances s in-
stead of the stiffnesses c. The effective compliance matrix of a
fractured mediumcan be written as the sumof the background
compliance s
b
and the so-called matrix of excess fracture com-
pliance s
f
:
c
1
s = s
b
+s
f
. (A-1)
If the background is VTI, then the stiffness matrix is given by
c
b
s
1
b
=
_
_
_
_
_
_
_
_
_
_
_
c
11b
c
12b
c
13b
0 0 0
c
12b
c
11b
c
13b
0 0 0
c
13b
c
13b
c
33b
0 0 0
0 0 0 c
44b
0 0
0 0 0 0 c
44b
0
0 0 0 0 0 c
66b
_
_
_
_
_
_
_
_
_
_
_
,
(A-2)
where c
12b
=c
11b
2c
66b
. The matrix s
f
of the excess compli-
ance of a fracture set with the normal in the x
1
-direction can
be written as
s
f
=
_
_
_
_
_
_
_
_
_
_
_
K
N
0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 K
V
0
0 0 0 0 0 K
H
_
_
_
_
_
_
_
_
_
_
_
, (A-3)
where K
N
is the normal fracture compliance and K
V
and K
H
are the two shear compliances in the vertical and horizontal di-
rections. The matrix s
f
is no longer diagonal if the fractures are
corrugated (part I; for more details, see part III of this series).
Since K
V
=K
H
, fractures described by matrix (A-3) are some-
times called orthorhombic (Schoenberg and Douma, 1988).
It is convenient to replace the excess fracture compliances
K
N
, K
V
, and K
H
by the following dimensionless quantities in-
troduced by Schoenberg and Helbig (1997):
N
=
K
N
c
11b
1 + K
N
c
11b
,
V
=
K
V
c
44b
1 + K
V
c
44b
,
(A-4)
H
=
K
H
c
66b
1 + K
H
c
66b
.
1814 Bakulin et al.
Then equation (A-3) becomes
s
f
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
N
c
11b
(1
N
)
0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0

V
c
44b
(1
V
)
0
0 0 0 0 0

H
c
66b
(1
H
)
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
.
(A-5)
We call
N
,
V
, and
H
the normal, vertical, and hori-
zontal weaknesses introduced by the fractures (Bakulin and
Molotkov, 1998). The weaknesses vary from 0 to 1, with the
zero value corresponding to unfractured media and unity de-
scribing heavily fractured media in which the P- (for
N
=1)
or S-wave (for
V
=1 or
H
=1) velocity vanishes for propa-
gation across the fractures (part I).
Substituting equations (A-2) and (A-5) into equation (A-1)
yields the effective stiffness matrix for a single fracture set em-
bedded in a VTI background (Schoenberg and Helbig, 1997):
c =
_
c
1
0
0 c
2
_
, (A-6)
where
c
1
=
_
_
_
_
_
_
_
_
_
_
c
11
b
(1
N
) c
12
b
(1
N
) c
13
b
(1
N
)
c
12
b
(1
N
) c
11
b
N
c
2
12
b
c
11
b
c
13
b
_
1
N
c
12
b
c
11
b
_
c
13
b
(1
N
) c
13
b
_
1
N
c
12
b
c
11
b
_
c
33
b
N
c
2
13
b
c
11
b
_
_
_
_
_
_
_
_
_
_
,
(A-7)
c
2
=
_
_
_
c
44b
0 0
0 c
44b
(1
V
) 0
0 0 c
66b
(1
H
)
_
_
_, (A-8)
and 0 is the 3 3 zero matrix.
Alternatively, the same result can be obtained using the se-
ries
c [s
b
+s
f
]
1
=
__
I +s
f
s
b
1
_
s
b
_
1
= c
b
[I +s
f
c
b
]
1
= c
b
k=0
(s
f
c
b
)
k
, (A-9)
where I is the 6 6 identity matrix. Series (A-9) converges to
equation (A-6) if all eigenvalues of the matrix s
f
c
b
have abso-
lute values less than 1. This is always the case if the weaknesses
N
,
V
, and
H
, which dene nonzero eigenvalues of the ma-
trix s
f
[see equation (A-5)], are sufciently small.
Equation (A-9) can serve as the basis for developing use-
ful approximations for the effective stiffness matrix c. If the
crackdensity (or fracture intensity) is small andthe weaknesses
{
N
,
V
,
H
} 1, we may truncate series (A-9) by keeping
only linear terms with respect to
N
,
V
, and
H
:
c c
b
c
b
s
f
c
b
. (A-10)
It is interesting to note that equation (A-10) becomes exact
if we replace the matrix s
f
[equation (A-5)] by its linearized
version:
s
lin
f
=
_
_
_
_
_
_
_
_
_
_
N
/c
11
b
0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0
V
/c
44
b
0
0 0 0 0 0
H
/c
66
b
_
_
_
_
_
_
_
_
_
_
.
(A-11)
Approximation (A-10) can be used to obtain two impor-
tant results. First, it can be extended in a straightforward way
to multiple fracture sets (Nichols et al., 1989). If the effective
compliance represents the sum of the excess compliances of N
fracture sets,
s
f
=
N
i =1
s
f i
, (A-12)
then for the effective stiffness c we have from equation (A-10)
c c
b
i =1
c
b
s
f i
c
b
. (A-13)
The compliance matrix s
f i
cannot be described by equa-
tion (A-3) if the normal n
i
to the i th fracture set does not coin-
cide with the x
1
-axis. The matrices s
f i
for arbitrary orientation
of n
i
may be obtained from equation (A-3) using the so-called
Bond rotation (Winterstein, 1990). For example, the compli-
ance matrix for a fracture set with the normal n
i
=[0, 1, 0] has
the form
s
f i
=
_
_
_
_
_
_
_
_
_
_
_
0 0 0 0 0 0
0 K
N
0 0 0 0
0 0 0 0 0 0
0 0 0 K
V
0 0
0 0 0 0 0 0
0 0 0 0 0 K
H
_
_
_
_
_
_
_
_
_
_
_
. (A-14)
Second, equation (A-10) is well suited for deriving the weak-
anisotropy approximation for the stiffnesses of effective media
formed by fractures embedded in an anisotropic background.
We assume that the background mediumis weakly anisotropic,
so that
c
b
= c
iso
b
+c
ani
b
, (A-15)
where c
iso
b
is the stiffness matrix of a purely isotropic solid that
approximates c
b
in some sense, c
ani
b
is the anisotropic portion
of c
b
, and 1. We also assume that the elements of the frac-
ture compliance matrix are of the same order as and can be
denoted as s
f
. If we represent the effective stiffness matrix c
in a form similar to equation (A-15), equation (A-10) can be
rewritten as
c
iso
+c
ani
c
iso
b
+c
ani
b

_
c
iso
b
+c
ani
b
_
s
f
_
c
iso
b
+c
ani
b
_
.
(A-16)
Collecting the terms linear in yields
c
ani
= c
ani
b
c
iso
b
s
f
c
iso
b
. (A-17)
This equation, validinthe weakanisotropy limit, canbe loosely
interpreted in the following way: the anisotropy c
ani
of an ef-
fective medium containing fractures in an anisotropic back-
ground described by c
b
is equal to the sum of the background
anisotropy c
ani
b
and the anisotropy caused by the fractures em-
bedded into any isotropic medium with c
iso
b
sufciently close
to c
b
[see equation (A-13)]. The matrix c
iso
b
must satisfy the
inequality |(||c
b
||/||c
iso
b
||) 1| <.
APPENDIX B
ANISOTROPIC PARAMETERS FOR ORTHORHOMBIC MEDIA
The basic set of the anisotropic parameters for orthorhom-
bic media has been introduced by Tsvankin (1997b). His no-
tation contains the vertical velocities of the P-wave and one
of the S-waves and seven dimensionless Thomsen-type (1986)
anisotropic coefcients. The denitions of those parameters in
terms of the stiffnesses c
i j
and density are given below.
V
P0
the P-wave vertical velocity:
V
P0

_
c
33
( is density). (B-1)
V
S0
the velocity of the vertically traveling S-wave polarized
in the x
1
-direction:
V
S0

_
c
55
. (B-2)
(2)
the VTI parameter in the [x
1
, x
3
] symmetry plane
normal to the x
2
-axis [this explains the superscript (2)]:
(2)
c
11
c
33
2c
33
. (B-3)
(2)
1
, x
3
] plane:
(2)
(c
13
+c
55
)
2
(c
33
c
55
)
2
2c
33
(c
33
c
55
)
. (B-4)
(2)
1
, x
3
] plane:
(2)
c
66
c
44
2c
44
. (B-5)
(1)
2
, x
3
] symmetry plane:
(1)
c
22
c
33
2c
33
. (B-6)
(1)
2
, x
3
] plane:
(1)
(c
23
+c
44
)
2
(c
33
c
44
)
2
2c
33
(c
33
c
44
)
. (B-7)
(1)
2
, x
3
] plane:
(1)
c
66
c
55
2c
55
. (B-8)
(3)
1
, x
2
] plane (x
1
plays the
role of the symmetry axis):
(3)
(c
12
+c
66
)
2
(c
11
c
66
)
2
2c
11
(c
11
c
66
)
. (B-9)
These nine parameters fully describe wave propagation in
general orthorhombic media. In particular applications, how-
ever, it is convenient to operate with specic combinations of
Tsvankins parameters. For example, P-wave NMO velocity
from dipping reectors depends on three coefcients , which
determine the anellipticity of the P-wave slowness in the sym-
metry planes (Grechka and Tsvankin, 1999). The denitions of
(1,2,3)
are analogous to that of the AlkhalifahTsvankin (1995)
coefcient in VTI media:
(1)
2
, x
3
] plane:
(1)

(1)
(1)
1 +2
(1)
. (B-10)
(2)
1
, x
3
] plane:
(2)

(2)
(2)
1 +2
(2)
. (B-11)
(3)
1
, x
2
] plane:
(3)

(1)
(2)
(3)
_
1 +2
(2)
_
_
1 +2
(2)
__
1 +2
(3)
_ . (B-12)
Vertical transverse isotropy may be considered as a special
case of orthorhombic media, where
(1)
=
(2)
= , (B-13)
(1)
=
(2)
= , (B-14)
(1)
=
(2)
= , (B-15)
(3)
= 0, (B-16)
and, as a consequence,
(1)
=
(2)
= , (B-17)
(3)
= 0. (B-18)
1816 Bakulin et al.
APPENDIX C
TWO IDENTICAL FRACTURE SETS
If the model includes two identical orthogonal fracture sets
in a purely isotropic background, only four effective stiffnesses
out of nine remain independent because the stiffness matrix
depends on just four quantities: , ,
N1
=
N2
N
, and
T1
=
T2
T
. From equations (40)(43) it follows that in
this case c
11
=c
22
, c
13
=c
23
, and c
44
=c
55
. Constraints (47) and
(48) become identical and, together with equation (46), may
be reduced to
c
12
(c
33
+c
13
) = c
13
(c
11
+c
13
) (C-1)
and
2(c
11
+c
13
)c
44
c
66
= (2c
66
c
44
)
_
c
11
c
33
c
2
13
_
. (C-2)
We can call this medium quasi-VTI because its stiffness ma-
trix is close to the one for vertical transverse isotropy. Un-
like real VTI media, however, the stiffnesses of the quasi-VTI
model satisfy constraints (C-1) and (C-2), which replace the
VTI relationship c
11
=c
12
+ 2c
66
. Indeed, the combination of
the stiffnesses that must vanish in VTI media can be written in
the exact form as
c
11
2c
66
c
12
=
4
2
(K
T
K
N
)
(1 +2K
N
)(1 +2 K
T
)
. (C-3)
Hence, because of the difference betweenthe normal andshear
compliances, themodel withtwoidentical fracturesets does not
have the VTI symmetry.
The anisotropic coefcients of quasi-VTI media are de-
scribed by equations (B-3)(B-5), (B-13)(B-15), and
(3)
=
_
_
_
_
4
2
( +)(K
T
K
N
)
2 +

1 +2( +)K
N
_
_
+2( +)(K
T
K
N
) +
(1 +4K
N
)
1 +2K
N
+2
2
K
T
1 +2( +)K
N
1 +2K
N
_
1
. (C-4)
Interestingly, quasi-VTI media have two additional vertical
symmetry planes at 45
with respect to planes [x

1
, x
3
] and
[x
2
, x
3
]. Clearly, expressions for anisotropic coefcients
(1,2)
,
(1,2)
, and
(1,2)
, which could be dened within these symmetry
planes, are different from those given by equations (49)(58).
Two identical scalar fracture sets
If we further assume that both fracture sets are lled with
gas [i.e., K
N
=K
T
; Schoenberg and Sayers (1995); part I], the
effective medium becomes VTI. The effective stiffnesses given
belowdepend only on the background parameters and and
the weakness
T
=(K
T
)/(1 + K
T
) [equation (45)]:
c
11
= ( +2)
(1
T
)[1 +
T
(3 4g)]
D
, (C-5)
c
13
=
1
2
T
D
, (C-6)
c
33
= ( +2)
1
T
(8g 5)
D
, (C-7)
c
44
= (1
T
), (C-8)
c
66
=
1
T
1 +
T
, (C-9)
where
D (1 +
T
)[1 +
T
(2 3g)]. (C-10)
The ve stiffness elements of this three-parameter VTI
medium are related by two constraints that can be ob-
tained by substituting the VTI relation c
11
=c
12
+2c
66
into
equations (C-1) and (C-2). In terms of Thomsens (1986)
anisotropic coefcients, these constraints take the form
= = 4 (1 +2 )(1 g). (C-11)
The equality = indicates that the effective mediumis ellipti-
cally anisotropic. Inthe weak-anisotropy limit, equation(C-11)
yields
= 4 (1 g). (C-12)
For a typical value g =0.25, = 3 .
APPENDIX D
FRACTURE CHARACTERIZATION FOR TWO ORTHOGONAL FRACTURE SETS: SPECIAL CASES
Azimuthally independent P-wave NMO velocity
In contrast to the case of a single fracture set in an isotropic
or VTI background, the orientation of two orthogonal fracture
systems cannot always be found fromthe P-wave NMOellipse
from a horizontal reector. If
(1)
=
(2)
or, equivalently,
(1 2g)
N1
+
T1
= (1 2g)
N2
+
T2
(D-1)
[equations (50) and (54)], the P-wave NMO ellipse degener-
ates into a circle. This equation does not necessarily imply that
the two systems of fractures are identical because it can be
satised for fracture sets with different crack densities (e.g.,
T1
>
T2
) and different uid saturations (
N1
<
N2
). The
azimuths of fractures in this case can be found from the shear-
wave polarization directions or by using P-wave reections
from dipping interfaces and/or the azimuthal variation of P-
wave nonhyperbolic moveout.
Equal tangential weaknesses
If the tangential weaknesses are equal to each other
(
T1
=
T2
), the anisotropic coefcients
(1)
and
(2)
also be-
come identical [see equations (51) and (55)] and there is no
shear-wave splitting at vertical incidence. Hence, the fracture
orientation cannot be determined from shear-wave splitting.
However, if the fracture sets have different normal weaknesses
(
N1
=
N2
), the fracture azimuths can be found using P-wave
NMO ellipses from a horizontal or a dipping reector.
Identical fracture sets (quasi-VTI medium)
In the case of two identical orthogonal fracture sets, neither
the P-wave NMO ellipse from a horizontal reector (it degen-
erates into a circle because
(1)
=
(2)
) nor the shear-wave po-
larization directions (
(1)
=
(2)
, see Appendix C) can be used
to detect the fracture orientation. Still, since
(3)
(or
(3)
) differs
from zero, the orientation can be found from normal moveout
of dipping P events or from P-wave nonhyperbolic moveout.
The rest of the inversion procedure is similar to that for the
general case of two different fracture systems.
Despite the absence of shear-wave splitting in the vertical
direction, S-waves can still be used for estimating fracture pa-
rameters. The wavefront of the slow shear wave in quasi-VTI
media always has cusps at 45
with respect to the symmetry

planes [x
1
, x
3
] and [x
2
, x
3
] (Figure D-1) if c
11
2c
66
c
12
>0
[or, equivalently, K
T
> K
N
; see equation (C-3) and part I]. This
wavefront structure by itself can be used to identify the sym-
metry directions (and, therefore, the fracture strikes) if a suf-
cient number of azimuthal measurements is available. Once
those directions are found, we can use equations (23)(29) for
FIG. D-1. Group velocity surfaces of the (a) fast and (b) slow shear waves for the model with two identical orthogonal fracture
sets. The background velocities are V
P
=2 km/s and V
S
=1 km/s. The fracture weaknesses
N
=0 and
T
=0.15 approximately
correspond to uid-lled penny-shaped cracks with a crack density of 7%.
the NMOvelocities of the S
1
- and S
2
-waves in the vertical sym-
metry planes and estimate the fracture parameters based on
equations (66) and (68). It is interesting that in quasi-VTI me-
dia the S-wave NMO velocities from a horizontal reector are
dened, strictly speaking, only within the vertical symmetry
planes because the shear-wave singularity makes the S-wave
offset-traveltime curve nondifferentiable in any other azimuth.
Identical gas-lled fracture sets (VTI medium)
Finally, if two identical fracture sets are gas lled, the ef-
fective medium becomes VTI (Appendix C). Since the VTI
model is azimuthally isotropic, it is impossible to nd the frac-
ture orientation from seismic data. The single parameter
T
(or K
T
) needed to describe the fractures can be found fromany
anisotropic coefcient givenby equations (49)(56) if the back-
ground V
S
/V
P
ratio is known. Estimating several anisotropic
coefcients provides a redundancy that can be used to verify
the validity of this model [see equation (C-11)]. Indeed, this
VTI medium is elliptical ( =) and has a specic relationship
(C-12) between and .
Estimation of fracture parameters from reection seismic dataPart III:
Fractured models with monoclinic symmetry
Andrey Bakulin
, Vladimir Grechka
, and Ilya Tsvankin
ABSTRACT
Geophysical and geological data acquired over nat-
urally fractured reservoirs often reveal the presence of
multiple vertical fracture sets. Here, we discuss modeling
and inversion of the effective anisotropic parameters of
two types of fractured media with monoclinic symmetry.
The rst model is formed by two different nonorthog-
onal sets of rotationally invariant vertical fractures in
an isotropic host rock; the other contains a single set of
fractures with microcorrugated faces.
In monoclinic media with two fracture sets, the shear-
wave polarizations at vertical incidence and the orienta-
tion of the NMO ellipses of pure modes in a horizontal
layer are controlled by the fracture azimuths as well as
by their compliances. While the S-wave polarization di-
rections depend only on the tangential compliances, the
axes of the P-wave NMO ellipse are also inuenced by
the normal compliances and therefore have a different
orientation. This yields anapparent discrepancybetween
the principal anisotropy directions obtained using P and
S data that does not exist in orthorhombic media. By
rst using the weak-anisotropy approximation for the
effective anisotropic parameters and then inverting the
exact equations, we devise a complete fracture charac-
terization procedure based on the vertical velocities of
the P- and two split S-waves (or converted PS-waves)
and their NMO ellipses from a horizontal reector. Our
algorithm yields the azimuths and compliances of both
fracture systems as well as the P- and S-wave velocities
in the isotropic background medium.
In the model with a single set of microcorrugated frac-
tures, monoclinic symmetry stems from the coupling be-
tween the normal and tangential (to the fracture faces)
slips, or jumps in displacement. We demonstrate that for
this model the shear-wave splitting coefcient at verti-
cal incidence varies with the uid content of the frac-
tures. Although conventional fracture models that ig-
nore microcorrugation predict no such dependence, our
conclusions are supported by experimental observations
showing that shear-wave splitting for dry cracks may be
substantially greater than that for uid-lled ones.
INTRODUCTION
This work completes our series of three papers on seismic
characterization of naturally fractured reservoirs. The rst two
papers are devoted to the model with a single system of rota-
tionallyinvariant fractures inanisotropic background(Bakulin
et al. 2000a; hereafter referred to as part I) and to fractured
models with orthorhombic symmetry (Bakulin et al. 2000b;
part II). Parts I and II contain a detailed review of recent pub-
lications on fracture characterization that is not repeated here.
All three papers strive to build a bridge between rock physics
theories of fractured media and seismic methods of fracture
detection. They also attempt to develop efcient fracture char-
acterization methodologies based on surface reection data.
Manuscript received by the Editor June 16, 1999; revised manuscript received March 3, 2000.
Cross, Madingley Road, Cambridge CB 3 0EL, England. E-mail: bakulin@cambridge.scr.slb.com.
Colorado School of Mines, Center for Wave Phenomena, Department of Geophysics, Golden, Colorado 80401-1887. E-mail: vgrechka@dix.
mines.edu; ilya@dix.mines.edu.
c _2000 Society of Exploration Geophysicists. All rights reserved.
Theoretical tools (the so-called effective medium theories)
for modeling the seismic response of fractured media have
existed since the early 1980s (e.g., Schoenberg, 1980, 1983;
Hudson, 1980, 1981, 1988; Thomsen, 1995). In particular,
Schoenberg (1980, 1983) and Schoenberg and Muir (1989) sug-
gest treating fractures as highly compliant surfaces inside a
solid host rock. According to their linear-slip theory, the ef-
fective compliance of a rock mass with one or several frac-
ture sets can be found as the sum of the compliances of the
host (background) rock and those of all the fractures. Then
the background and fracture parameters can be related to
the effective Thomsen-type anisotropic coefcients, whichgov-
ern the inuence of anisotropy on various seismic signatures.
1818
Fractured Monoclinic Media 1819
This formalism can be used to invert seismic signatures for
the fracture compliances and to make inferences about the
physical properties of the fracture network. (Schoenberg and
Douma (1988) and part I show that fracture compliances ef-
fectively absorb such information about the microstructure
of fractures as their shape, possible interaction, partial sat-
uration, the presence of equant porosity, etc. This informa-
tion cannot be obtained unambiguously from seismic data
alone.)
In part I we implement these ideas for transversely isotropic
media with a horizontal symmetry axis (HTI) formed by a sin-
gle set of vertical rotationally invariant fractures. A system of
vertical fractures in a VTI (transversely isotropic with a ver-
tical symmetry axis) background or two orthogonal fracture
sets in an isotropic host rock lead to a medium of orthorhom-
bic symmetry examined in part II. Further complications intro-
duced into the model may lower the symmetry of the effective
medium to monoclinic. First experimental evidence of mono-
clinic symmetry in the subsurface is provided by Winterstein
and Meadows (1991), who analyze walkaway vertical seismic
proling (VSP) data over a fractured reservoir.
s
b
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

(3 2)

2(3 2)

2(3 2)
0 0 0

2(3 2)

(3 2)

2(3 2)
0 0 0

2(3 2)

2(3 2)

(3 2)
0 0 0
0 0 0
1
0 0
0 0 0 0
1
0
0 0 0 0 0
1
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
. (2)
Here, we discuss two types of fracture-induced monoclinic
media. The rst model contains two different nonorthogonal
sets of rotationally invariant fractures in an isotropic back-
ground. We prove that all fracture parameters of the effective
monoclinic medium with a horizontal symmetry plane can be
estimated using the vertical and NMO velocities of the P- and
two split S (or PS)-waves reected from horizontal interfaces.
It should be emphasized that shear-wave data alone do not
contain enough information to constrain the model parame-
ters (Liu et al., 1993). Also, we conrm the result of Grechka
et al. (2000) that the polarization directions of the vertically
propagating shear waves generally are not aligned with the
axes of the P-wave NMO ellipse (which cannot happen in the
higher symmetry HTI or orthorhombic media). A similar con-
clusion is drawn by Sayers (1998) who examines the azimuthal
variation of the P-wave phase-velocity function in monoclinic
media. This gives aplausibletheoretical explanationfor thedis-
crepancies in the fracture orientation estimated from P- and
S-wave data by P erez et al. (1999).
The second monoclinic model examined here has a vertical
symmetry plane and consists of a single set of microcorrugated
(rotationally noninvariant) fractures in an isotropic matrix. We
derive the shear-wave splitting coefcient as a function of the
compliances and showthat it is substantially different for uid-
lled and dry fractures.
TWO SETS OF VERTICAL FRACTURES
Effective elastic parameters
Within the framework of the linear-slip theory, the effec-
tive compliance matrix s of a rock with multiple fracture sets
can be determined as the sumof the compliance s
b
of the back-
ground mediumand the fracture compliances s
f
(Nichols et al.,
1989; Schoenberg andMuir, 1989; Molotkov andBakulin, 1997;
Bakulin and Molotkov, 1998). Considering two different arbi-
trarily oriented fracture sets with the compliances s
f 1
and s
f 2
in
a purely isotropic background, we can represent the effective
compliance as
s = s
b
s
f 1
s
f 2
c
1
, (1)
where c is the stiffness matrix of the effective medium. The
compliance s
b
of the isotropic background, written in terms of
the Lam e parameters and , has the form
Assuming that both fracture sets are rotationally invariant,
their matrices s
f i
(i =1, 2) can be described by the normal
and tangential (with respect to the crack faces) compliances
K
Ni
and K
Ti
. If the normal n
i
to the i th fracture set points in
the direction of the x
1
-axis, the compliance matrix is given by
(Schoenberg and Douma, 1988; Schoenberg and Sayers, 1995;
part I)
s
x
1
f i
=
_
_
_
_
_
_
_
_
_
_
_
K
Ni
0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 K
Ti
0
0 0 0 0 0 K
Ti
_
_
_
_
_
_
_
_
_
_
_
. (3)
Rotation of the fracture normal by the angle
i
around the
x
1
-axis changes the matrix s
f i
according to the so-called Bond
transformation,
s
f i
= N(
i
) s
x
1
f i
N
T
(
i
), (4)
where the 6 6 matrix N(N
T
is the transpose) is explicitly writ-
ten in Winterstein (1990). The exact expressions for s
f i
, taken
1820 Bakulin et al.
from Schoenberg et al. (1999), are given in Appendix A. It
seems to be natural to choose the coordinate frame in such a
way that one of the fracture sets is orthogonal to the axis x
1
(or x
2
), as is done by Liu et al. (1993). As we discuss below,
however, a simpler effective stiffness matrix can be obtained
by aligning the horizontal coordinate axes with the polariza-
tion directions of the vertically traveling S-waves. In this case,
neither fracture system is orthogonal to one of the horizontal
coordinate directions.
Equations (1)(4) make it possible to compute the com-
pliance (s) and stiffness (c) matrices of the effective medium
formedby twovertical fracture sets inanisotropic background.
Analysis of the compliance matrix [see equations (A-1)(A-9)
for a single fracture set] shows that this medium is monoclinic
with a horizontal symmetry plane. Below, we discuss the in-
version of the effective anisotropic coefcients of monoclinic
media for the fracture parameters.
Anisotropic coefcients of monoclinic media
Dimensionless anisotropic parameters, rst introduced by
Thomsen (1986) for VTI media, proved to be extremely useful
in seismic velocity analysis and inversion. The main advantage
of Thomsens notation is in capturing the combinations of the
stiffness coefcients responsible for a wide range of seismic
signatures (Tsvankin, 1996). Thomsen-style notation for HTI
media was introduced by R uger (1997) and Tsvankin (1997a).
(Inpart I we discuss the inversionof the anisotropic coefcients
(V)
,
(V)
, and
(V)
of fracture-induced HTI media for the frac-
ture compliances.) Tsvankin (1997b) shows that orthorhombic
media can be conveniently described by seven anisotropic co-
efcients and two vertical velocities (of the P- and one of the
split S-waves), with a total of only six parameters fully respon-
sible for the P-wave kinematics. Tsvankins notation simplies
the expressions for azimuthally varying NMO velocities in or-
thorhombic media (Grechka and Tsvankin, 1998) and helps to
identify the subset of the medium parameters that can be de-
termined using P-wave surface reection data (Grechka and
Tsvankin, 1999).
An extension of Thomsen parameters to monoclinic media
is suggested by Grechka et al. (2000). They note that the ex-
pression for NMO ellipses of P- and S-waves from horizontal
reectors take a particularly simple formif the horizontal coor-
dinate axes x
1
and x
2
coincide with the polarization directions
of the vertically propagating split shear waves S
1
and S
2
. In this
natural coordinate frame the stiffness coefcient c
45
vanishes
(Helbig, 1994; Mensch and Rasolofosaon, 1997),
c
45
= 0, (5)
and the number of independent stiffnesses reduces from 13
to 12.
Grechka et al. (2000) replace the stiffness elements with the
vertical velocities of P-waves (V
P0
) and one of the S-waves
(V
S0
) and 10 anisotropic coefcients denoted as
(1,2)
,
(1,2,3)
,
(1,2)
, and
(1,2,3)
(Appendix B). These parameters can be di-
vided into two different groups. The rst contains the vertical
velocities and the , , and coefcients, which are dened
exactly in the same way as Tsvankins (1997b) parameters for
orthorhombic media. These nine quantities mainly control the
semiaxes of the NMO ellipses of waves P, S
1
, and S
2
reected
fromhorizontal interfaces. The secondgroupincludes the three
coefcients responsible for the rotation of the NMO ellipses
with respect to the coordinate axes (Grechka et al., 2000).
The parameter
(3)
determines the orientation of the P-wave
NMO ellipse, whereas
(1)
and
(2)
govern the rotations of the
S
1
- and S
2
-ellipses, respectively [see equations (B-10)(B-12)].
This denition of the coefcients distinguishes the notation
of Grechka et al. (2000) fromthe Thomsen-style parameteriza-
tion of arbitrary anisotropic media introduced by Mensch and
Rasolofosaon (1997).
Grechka et al. (2000) show that 11 (out of 12) parameters of
monoclinic media [V
P0
, V
S0
,
(1,2)
,
(1,2)
,
(1,2)
, and
(1,2,3)
] can
be estimated in a stable way using the vertical velocities and
NMOellipses fromhorizontal reectors of the P- and two split
S-waves. If the survey is acquired with P-wave sources, pure
shear reections can be replaced by the converted waves PS
1
and PS
2
; this option is especially practical for offshore data.
Our goal is to demonstrate that the 11 effective quantities
listed above can be inverted for the following physical parame-
ters of the model: the P- and S-wave velocities V
P
and V
S
in the
isotropic background, the azimuths
1
and
2
of the normals to
the fracture faces, and dimensionless fracture weaknesses de-
noted by
Ni
and
Ti
(i =1, 2). The weaknesses are related to
the fracture compliances K
Ni
and K
Ti
as (Hsu and Schoenberg,
1993; part I)
Ni
=
( 2) K
Ni
1 ( 2) K
Ni
and
Ti
=
K
Ti
1 K
Ti
,
(i = 1, 2), (6)
where
Ni
and
Ti
are always positive and vary from zero (no
fracturing) to unity (extreme degree of fracturing).
For isolated penny-shaped cracks, vanishing values of the
ratio
Ni
/
Ti
correspond to uid-lled fractures, whereas
Ni
/
Ti
( 2)/ (or K
Ni
K
Ti
) indicate that the cracks
are dry. Also, regardless of the type of crack inll, the tangen-
tial weakness
Ti
is close to twice the crack density e
i
(part I).
If we ignore the inuence of one fracture system on the other
(a reasonable assumption for small crack density), these rela-
tions hold for the weaknesses of each system.
Following Grechka et al. (2000), we choose the axis x
1
to
coincide with the polarization direction of the fast shear wave
S
1
at vertical incidence (Figure 1). In this coordinate frame,
c
45
vanishes [equation (5)], providing an additional constraint
for the fracture parameters. This constraint can be obtained
by analyzing the nonzero elements of the effective compliance
matrix s [equation (1)] and the matrices s
b
[equation (2)], s
f 1
,
and s
f 2
[equations (A-1)(A-9)]. Since s
44
, s
45
, and s
55
are the
only nonzero elements of the fourth and fth columns and the
fourth and fth rows of s, the inverse (stiffness) matrix c has a
block
_
c
44
c
45
c
45
c
55
_
=
_
s
44
s
45
s
45
s
55
_
1
.
Hence, c
45
= 0 requires that
s
45
= 0, (7)
which can be satised only if [see equations (1), (2), and (A-7)]
K
T1
sin 2
1
K
T2
sin 2
2
= 0. (8)
Replacing the compliances with the weaknesses [equation (6)]
yields
T1
1
T1
sin 2
1

T2
1
T2
sin 2
2
= 0. (9)
Thus, we have a total of 12 equations [11 monoclinic param-
eters V
P0
, V
S0
,
(1,2)
,
(1,2)
,
(1,2)
, and
(1,2,3)
plus constraint (9)]
to be solved for eight unknowns [V
P
, V
S
,
Ni
,
Ti
, and
i
(i =1, 2)]. The nonlinear relations between the anisotropic co-
efcients and fracture parameters make it necessary to apply
numerical methods and study the uniqueness of the solution.
As shownbelow, our inversionalgorithmconverges towardval-
ues close to the actual model parameters, given a reasonable
magnitude of errors in the input data.
Equations (8) and (9) allow us to make two important con-
clusions about the fracture orientation even prior to the in-
version. First, equation (9) can be satised only if the fracture
azimuths
1
and
2
(assumed to lie between /2 and /2)
have opposite signs because both
T1
and
T2
are nonnega-
tive (Figure 1). Therefore, the angle between the normals (and
the two crack systems themselves) can be found as
1
2
; the
absolute values of
1
and
2
are equal only if
T1
=
T2
.
Second, equation (8) indicates that for a xed angle between
the fractures, the azimuths
1
and
2
are controlled only by the
ratio of the tangential compliances K
T1
/K
T2
. The shear-wave
polarization directions bisect the angles between the fractures
if K
T1
=K
T2
. To nd the fracture orientation relative to the
shear-wave polarization directions when K
T1
,=K
T2
, we need
to supplement equation (8) with the condition
c
55
> c
44
. (10)
Inequality (10) ensures that the fast shear wave S
1
is polarized
along the x
1
-axis (Figure 1). Using equations (1)(4), we can
rewrite condition (10) in the form
K
T1
cos 2
1
K
T2
cos 2
2
< 0. (11)
FIG. 1. Two sets of parallel vertical fractures form an effective
monoclinic medium. The fracture normals make the angles
1
and
2
with the axis x
1
that coincides with the polarization
direction of the vertically traveling S
1
-wave. To dene
1
and
2
in a unique fashion, we assume that /2
1
</2 and
/2 <
2
/2.
Combined with equation (8), inequality (11) unambiguously
denes the directions of both fracture sets for given values of
K
T1
, K
T2
, and the angle
1
2
.
Let us assume that the rst fracture set has a higher tangen-
tial compliance (K
T1
> K
T2
). Analysis of equations (8) and (11)
shows that in this case the normal to the rst fracture set lies
within the interval /4 <
1
<3/4. Hence, the polarization di-
rection of the fast (S
1
) shear wave (i.e., the x
1
-axis) is always
closer to the strike of the more compliant fractures. Accord-
ing to the quantitative estimates in Figure 2, for K
T1
/K
T2
>3
the S
1
-polarization direction does not deviate by more than
10
from the strike of the rst fracture system. In the limit of

K
T1
K
T2
, the medium becomes effectively HTI, and the po-
larization of the S
1
-wave is parallel to the rst fracture set; this
result is well known for horizontal transverse isotropy (e.g.,
Winterstein, 1990).
Weak-anisotropy approximation.If the fracture density is
small and the weaknesses
Ni
_1 and
Ti
_1, the effective
medium is weakly anisotropic. In this case, it is possible to
simplify the expressions for the anisotropic coefcients by lin-
earizing them in the weaknesses
Ni
and
Ti
(Appendix C).
Equations (C-1)(C-11) show that all , , and coefcients
are even functions of the fracture azimuths
1
and
2
, and in-
formation about the signs of the azimuths can be obtained only
from the coefcients. Since
(1,2,3)
determine the rotation of
the P- and S-wave NMOellipses withrespect tothe shear-wave
polarization directions (Grechka et al., 2000), this result indi-
cates the importance of carefully measuring the orientation of
the elliptical axes.
The linearized expressions (C-1)(C-11) can be used to esti-
mate all fracture parameters following, for instance, the al-
gorithm outlined in Appendix D. To check the accuracy of
the weak-anisotropy approximation, we computed the exact
anisotropic coefcients for the fracture parameters given in
Table 1 and carried out the inversion using the equations from
Appendix D. For typical moderate values of the tangential
FIG. 2. Azimuth of rst fracture system ([
1
90
[) as a func-
tion of the ratio of the tangential compliances K
T1
/K
T2
. Each
curve corresponds toa different angle betweenthe fracture sys-
tems:
1
2
=20
(circles), 40
(triangles), 60
(dashed line),
and 80
(solid line).
1822 Bakulin et al.
compliances for both fractures systems (if the fractures are
penny-shaped, the crack densities are e
1
T1
/2 =0.06 and
e
2
T2
/2 =0.10), our approximation leads to noticeable er-
rors in two weaknesses (
T1
and
N2
). However, it can still
be used to obtain a good initial guess for the nonlinear
inversion.
Special case: Equal tangential weaknesses.It is instructive
to examine the special case of equal tangential weaknesses of
the two fracture systems:
T1
=
T2

T
. (12)
Then, as follows from equation (9),
1
=
2
, (13)
and the S
1
-wave polarization direction bisects the angle be-
tweenthefracturesets (seeFigure2for K
T1
=K
T2
, whichcorre-
sponds to
T1
=
T2
). Substituting relations (12) and (13) into
equations (C-1)(C-11), we obtain equations (E-3)(E-13),
which can be inverted for the fracture parameters in a rela-
tively straightforward fashion. The results of Appendix E indi-
cate that the fracture azimuth and the weaknesses
T
,
N1
,
and
N2
can be found just from the vertical velocities of the
P- and S-waves and the P-wave NMO ellipse; the shear-wave
NMO ellipses provide redundant information.
If the tangential weaknesses of the two fracture sets are
equal, the orientations of the pure-mode NMO ellipses are
related in a simple way to the fracture azimuths. For instance,
the azimuth
P
of the semimajor axis of the P-wave NMO el-
lipse can be written in the weak-anisotropy approximation as
(Grechka et al., 2000)
tan 2
P
=
2
(3)
(2)
(1)
. (14)
Substituting equations (E-7), (E-8), and (E-13) into equa-
tion (14), we nd
tan 2
P
= A
P
tan 2, (15)
where
A
P
=
(
N2
N1
)(1 2g)
(
N1
N2
)(1 2g) 2
T
. (16)
Note that [ A
P
[ 1, which implies that the axes of the P-wave
NMO ellipse deviate from both fracture azimuths (except for
the special case of orthogonal fractures discussed below). Also,
if
N2
,=
N1
, the angle
P
,=0 and the P-wave ellipse is rotated
with respect to the polarization directions of the vertically
propagating shear waves, as notedby Grechka et al. (2000) (see
also Sayers, 1998). This may explain the discrepancies in the
fracture orientation estimated from the P-wave NMO ellipse
Table 1. Comparison of the actual fracture parameters with
those estimated from the exact anisotropic coefcients using
the linearized formulas from Appendix D.
Parameter V
S
/V
P

N1

T1

1

N2

T2

2
Actual 0.5 0.25 0.12 30
0.00 0.20 13
Estimated 0.5 0.20 0.04 40
0.08 0.22 5
and S-wave polarizations described by P erez et al. (1999). The

same conclusion is true for the NMO ellipses of the S
1
- and
S
2
-waves whose rotation angles are determined by the param-
eters
(1)
and
(2)
(Grechka et al., 2000). As follows from equa-
tions (E-11) and (E-12) for the fracture-induced monoclinic
model,
(1)
,=0 and
(2)
,=0.
If the two fracture systems are identical [i.e.,
N1
=
N2
in
addition to
T1
=
T2
], the axes of the P-wave NMO ellipse
are aligned with the shear-wave polarization directions (A
P
=
P
=0). In this case, the medium becomes orthorhombic, and
S-wave polarization vectors (which bisect the fractures) lie in
the vertical symmetry planes of the model. For
N1
=
N2
, all
three coefcients vanish, so the shear-wave NMOellipses are
cooriented with the P-wave ellipse.
Another special case of orthorhombic symmetry is that of
different but orthogonal fracture sets (2 =90
). If only the
normal weaknesses
N1
and
N2
are different (but
T1
=
T2
),
shear waves do not split at vertical incidence and have no de-
ned polarization directions (part II).
For monoclinic media, the axes of the P-wave NMO el-
lipse and the polarization vectors of the vertically traveling
S-waves are parallel only if the ratio of the tangential and nor-
mal compliances is the same for both fracture systems (i.e.,
(3)
=0 if K
N1
/K
T1
=K
N2
/K
T2
or
N1
/
T1
=
N2
/
T2
). This
result, valid only in the weak anisotropy approximation, can
be obtained from equations (8), (9), and (C-11). For example,
(3)
=0 if the normal and tangential compliances are equal to
each other, which agrees with the result of Sayers (1998).
Arbitrary fracture sets.The main signicance of the lin-
earized approximations for the fracture parameters is in pro-
viding insight into the behavior of the anisotropic coefcients
and a good initial model for the inversion procedure. Here
we discuss the inversion of the anisotropic parameters for the
fracture compliances andorientations basedonthe exact equa-
tions (1)(4) for the stiffnesses and the denitions from Ap-
pendix B. We assume that the quantities V
P0
, V
S0
,
(1,2)
,
(1,2)
,
(1,2)
, and
(1,2,3)
have been estimated from the vertical and
NMO velocities of the P- and split S-waves (or converted PS-
waves). To examine the stability of the inversion algorithm, we
introduce errors in all quantities [similar to those shown in Fig-
ures 2 and 4 in Grechka et al. (2000)] caused by Gaussian noise
with a variance of 2% added to the NMO velocities. The vari-
ances of the errors in the effective parameters are as follows:
2% in V
P0
and V
S0
, 0.01 in
(1)
and
(2)
, and 0.03 in all other
anisotropic coefcients. The fracture parameters that give the
best t to the error-contaminated values of the vertical veloc-
ities and anisotropic coefcients are found by minimization
using the simplex method.
Figure 3 displays typical inversion results for the V
P
/V
S
ra-
tio in the isotropic background and weaknesses
Ni
and
Ti
.
The standard deviation in the estimated V
S
/V
P
ratio (3.1%)
is somewhat higher than that in the input vertical velocities
V
P0
and V
S0
but is still quite acceptable. The errors in the
weaknesses
Ti
are about twice as large as those in the in-
put anisotropic coefcients. This error amplication can be
understood from the weak-anisotropy approximations (D-6)
and (D-8), which indicate that
Ti
are derived from the co-
efcients multipliedby a factor of 2; similar results are obtained
for models with one system of fractures in parts I and II.
The accuracy of the inverted normal weaknesses
Ni
is even
lower than that of
Ti
. This can also be explained using the
weak-anisotropy approximations (C-1)(C-11), which contain
terms in the form
Ti
a
Ni
with [a[ <1. Therefore, the same
errors in the input anisotropic coefcients can be compensated
by greater variations in
Ni
compared to those in
Ti
. Even
though the errors in
Ni
are rather signicant, Figure 3 shows
that we can still distinguish between dry fractures [the rst
system with K
N1
K
T1
or
T1
(1
N1
) g
N1
(1
T1
)] and
uid-lled ones with
N2
0. The unphysical values
N2
<0
appear in Figure 3 because random errors added to the data
may produce anisotropic coefcients that do not correspond
to any fractured media.
MICROCORRUGATED FRACTURES AND
FLUID-DEPENDENT SHEAR-WAVE SPLITTING
Shear-wave splitting at vertical incidence traditionally has
beenregardedas themost reliablemeasureof fractureintensity
for a set of parallel vertical fractures embedded in an isotropic
or VTI host rock. The fractional difference betweenthe S-wave
vertical velocities V
S1
and V
S2
is expressedthroughthe so-called
shear-wave splitting parameter
(S)
, dened as
(S)
V
2
S1
V
2
S2
2V
2
S2
V
S1
V
S2
V
S2
. (17)
It has beencommonknowledge among researchers working on
fracture characterization that
(S)
depends only on the crack
density and does not contain information about the uid con-
tent of the fractures. This conclusion seems to have been con-
rmed both theoretically (Hudson, 1981; Thomsen, 1995) and
experimentally (e.g., Martin and Davis, 1987; Winterstein and
Meadows, 1991). Onthe other hand, Guest et al. (1998) present
a case study where the splitting parameter for gas-lled cracks
proves to be signicantly higher than that for brine-lled ones,
which clearly contradicts the existing understanding. Here, we
give a possible theoretical explanation of these observations
by obtaining the effective parameters of a fracture set with
microcorrugated faces. The idea of the theory is to allow the
coupling between the tangential slip along fracture faces (re-
sponsible for shear-wave splitting) and the normal slip known
to depend on the uid content.
FIG. 3. Results of the inversion for the parameters of two frac-
ture systems and the background velocities. The dots mark the
actual values of the fracture parameters; the bars correspondto
one standard deviation in the inverted quantities. Not shown
arethestandarddeviations intheestimatedbackgroundveloci-
ties V
P
and V
S
(2.0%and2.5%, respectively) andinthe fracture
azimuths (9
; the actual values are

1
=30
and
2
=12.8
).
Linear-slip model for fractures with microcorrugated faces
As in the previous section, our analysis is based on the linear-
slip theory of Schoenberg (1980), Schoenberg and Muir (1989),
and Schoenberg and Sayers (1995). For simplicity, we examine
a single systemof parallel fractures with the normal n =[1, 0, 0]
in an isotropic background medium. The matrix of the excess
fracture compliance in this case has the form (Schoenberg and
Douma, 1988; part I)
s
f
=
_
_
_
_
_
_
_
_
_
_
_
K
N
0 0 0 K
NV
K
NH
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
K
NV
0 0 0 K
V
K
V H
K
NH
0 0 0 K
V H
K
H
_
_
_
_
_
_
_
_
_
_
_
, (18)
where K
N
is the normal fracture compliance that relates the
jump of the displacement normal to the fracture (i.e., the nor-
mal slip) to the normal stress in the direction [1, 0, 0]. The val-
ues K
V
and K
H
are the tangential compliances relating the slips
and stresses in the vertical [0, 0, 1] (K
V
) and horizontal [0, 1, 0]
(K
H
) directions. The compliance K
NV
couples the normal slip
to the tangential vertical stress or, equivalently, the tangential
slip in the direction [0, 0, 1] to the normal stress. Likewise, the
compliances K
NH
and K
V H
couple the horizontal stress in the
[0, 1, 0] direction to the normal and vertical slips.
The conventional conclusion about the shear-wave splitting
coefcient
(S)
being independent of fracture inll is based
on the assumption that the normal and tangential slips are
decoupled (K
NV
=K
NH
=K
V H
=0) and the matrix (18) is di-
agonal. An alternative model of cracks with microcorrugated
faces (see Figure 4) is suggested by Schoenberg and Douma
(1988). Since the stresses in either the normal (x
1
) or vertical
FIG. 4. Cross-section of a vertical fracture with microcorru-
gated faces. This microstructure leads to the coupling between
the normal and tangential slips (after Schoenberg and Douma,
1988).
1824 Bakulin et al.
(x
3
) direction applied to such a fracture produce slips in both
x
1
- and x
3
-directions, the compliance component K
NV
must be
nonzero. In contrast, the component K
V H
can always be set
to zero by the appropriate rotation of the coordinate system
(Berg et al., 1991). Below, we showthat fractures characterized
by the compliances
K
NH
= K
V H
= 0 and K
NV
,= 0 (19)
cause inll-dependent shear-wave splitting.
Effective model of fractured media
If a fracture systemwith the compliances described by equa-
tions (18) and (19) is embedded in an isotropic rock, the effec-
tive stiffness matrix has the form [see equation (1)]
c
1
= s
b
s
f
, (20)
where the background compliance matrix s
b
is given by equa-
tion (2). Evaluating the components of the matrix c yields
c =
_
_
_
_
_
_
_
_
_
_
_
c
11
c
12
c
12
0 c
15
0
c
12
c
33
c
23
0 c
35
0
c
12
c
23
c
33
0 c
35
0
0 0 0 c
44
0 0
c
15
c
35
c
35
0 c
55
0
0 0 0 0 0 c
66
_
_
_
_
_
_
_
_
_
_
_
. (21)
Equation (21) describes a monoclinic mediumwith the vertical
symmetry plane [x
1
, x
3
]. The stiffness elements are given by
c
11
= ( 2)
1 E
V
D
, c
12
=
1 E
V
D
,
c
15
=
_
( 2)
E
NV
D
,
c
33
= ( 2)
(1 E
V
)(1 E
N
) E
2
NV
D
,
(22)
c
23
=
(1 2gE
N
)(1 E
V
) 2gE
2
NV
D
,
c
35
=
gE
NV
D
,
c
44
= , c
55
=
1 E
N
D
, and c
66
=

1 E
H
,
where
D = (1 E
N
)(1 E
V
) E
2
NV
, =
4( )
( 2)
2
,
g =

2
,
and and are the Lam e constants of the host rock. The
dimensionless compliances E
N
, E
V
, E
H
, and E
NV
are dened
as follows (Hsu and Schoenberg, 1993):
E
N
= ( 2)K
N
, (23)
E
V
= K
V
, (24)
E
H
= K
H
, (25)
E
NV
=
_
( 2)K
NV
. (26)
The stability condition requires that matrix (18) be nonneg-
ative denite. In our case, this condition implies that all dimen-
sionless compliances E
N
, E
V
, and E
H
are nonnegative and
E
N
E
V
E
2
NV
0. (27)
Shear-wave splitting coefcient
The velocities of the split shear waves traveling inthe vertical
direction in the monoclinic mediumdescribed by equation (21)
are given by
V
2
S1
=

(28)
and
V
2
S2
=
2
c
33
c
55
c
2
35
c
33
c
55
_
(c
33
c
55
)
2
4c
2
35
, (29)
where is the density. While the velocity V
S1
of the fast shear
wave is simply equal to the background S-wave velocity, the
slower velocity V
S2
depends on the fracture compliances E
N
,
E
T
, and E
NV
.
Assuming weak anisotropy (E
N
_1, E
T
_1, and E
NV
_1)
and keeping the linear and quadratic terms in the dimension-
less compliances, we obtain the shear-wave splitting coefcient
[equation (17)] as
(S)
E
V
2

E
2
NV
g(3 4g)
2(1 g)
. (30)
Equation (30) shows that the coupling between the normal and
tangential slips (i.e., E
NV
,=0) always reduces the value of
(S)
.
To analyze the inuence of uid content on shear-wave split-
ting for fractures with microcorrugated faces, we generalize the
criterion given by Schoenberg and Sayers (1995) for isolated
penny-shaped cracks. They point out that the ratio K
N
/K
V
may
serve as an indicator of uid saturation because it vanishes for
uid-lled cracks and is equal to unity if the cracks are dry.
The result of Schoenberg and Sayers (1995), however, is for-
mulatedfor K
NV
=0andmust bemodiedfor microcorrugated
fractures.
Let us consider the 2 2 submatrix
s
f
=
_
K
N
K
NV
K
NV
K
V
_
(31)
of the compliance matrix (18). Note that the Schoenberg-
Sayers criterion for K
NV
=0 is equivalent to the statement that
the fractures are uid lled if one eigenvalue of s
f
is equal to
zero, whereas the fractures are dry if the matrix s
f
has two
equal eigenvalues.
We assume that the same relationship between uid satu-
ration and the eigenvalues of s
f
holds for the more general
case with K
NV
,=0. To justify this assumption, we recall that
the excess compliance matrix s
f
relates the stress applied to
the fracture faces to the slip or the jump of the displacement
across the fractures (Schoenberg, 1980). The eigenvalues of s
f
or s
f
are the coefcients that relate the magnitudes of the slip
and the stress vectors in the principal directions. Intuitively, we
expect dry fractures to be equally compliant in all directions
(there is no material inside to stiffen them), so the eigenvalues
of the fracture compliance matrix are supposed to be equal. In
contrast, uid-lled fractures are noticeably stiffer in a particu-
lar direction where the applied stress tends to squeeze the uid.
Thus, the eigenvalue corresponding to this direction should be
signicantly smaller than the other eigenvalues.
The submatrix (31) has equal eigenvalues if and only if
K
N
=K
V
and K
NV
=0, which means [see equation (26)] that
E
NV
= 0. (32)
For one of the eigenvalues of s
f
to go to zero, it is required that
K
2
NV
=K
N
K
V
or [equations (23), (24), and (26)]
E
2
NV
= E
N
E
V
. (33)
Since E
2
NV
E
N
E
V
[inequality (27)], equations (32) and (33)
impose strict bounds on possible absolute values of E
NV
.
Substituting equations (32) and(33) intoequation(30) yields
the shear-wave splitting coefcients for dry and uid-lled
cracks:
(S)
dry

E
V
2
(34)
and
(S)
wet

E
V
2
_
1
E
N
g(3 4g)
1 g
_
. (35)
Hence, for microcorrugatedfractures thesplittingcoefcient
is always higher for dry than for uid-lled fractures, which is
consistent with the observations of Guest et al. (1998). Figure 5
illustrates the inuence of the dimensionless compliance E
NV
on the shear-wave splitting coefcient. In this particular exam-
ple,
(S)
decreases by about 30%as the uid saturation changes
from zero (E
NV
=0) to 100%. Also note that the approxima-
tion (30) gives a reasonably accurate qualitative description of
(S)
.
FIG. 5. Shear-wave splitting coefcient
(S)
computed fromthe
exact equations (17), (28), and (29) (solid) and the approxima-
tion (30) (dashed) as a function of the dimensionless compli-
ance E
NV
. The model parameters are g =0.16, E
N
=1.3, and
E
V
=0.25.
The objective of this series of papers was to analyze the de-
pendence of seismic signatures on the physical parameters of
fracture networks and to develop fracture characterization al-
gorithms operating with surface seismic data. In part I we con-
sidered the simplest model of a single vertical fracture system
in a purely isotropic host rock (HTI medium) and showed that
the fracture compliances, or the dimensionless fracture weak-
nesses, are the only quantities that can be unambiguously esti-
mated from seismic data. The microstructure of the fractured
formation (e.g., the shape of fractures, their possible interac-
tion, the presence of equant porosity, etc.), however, cannot
be evaluated without additional information. For example, if
the fractures are known to be penny-shaped and isolated from
pore space, the weaknesses give a direct estimate of the crack
density and uid content of the fracture system.
By deriving the relationships between the weaknesses and
Thomsen-type anisotropic coefcients, we were able to study
the behavior of surface seismic signatures in fracture-induced
HTI (part I) and orthorhombic (part II) media. The analytic
results for HTI media provided the basis for inversion algo-
rithms designed to estimate the orientation and weaknesses of
a vertical fracture set using P-wave reections alone or a com-
bination of P and converted (PS) data. In part II we extended
our parameter estimation methodology to orthorhombic me-
diaformedeither byasinglefractureset inananisotropic (VTI)
host rock or by two orthogonal fracture sets in an isotropic
background.
This paper has investigated the inverse problem for an ef-
fective monoclinic medium caused by two nonorthogonal sets
of rotationally invariant fractures. The weaknesses and az-
imuths of both fracture systems, along with the velocities in
the isotropic background, can be obtained using the vertical
velocities and NMOellipses (fromhorizontal interfaces) of the
P-wave and two split S-waves. In principle, pure S reections
can be replaced in the inversion procedure by the converted
(PS) waves. Using the weak anisotropy approximation, we ob-
tain simple, linearized expressions for the fracture parameters,
which can serve as a good initial guess for the nonlinear inver-
sion algorithm. Numerical analysis shows that the tangential
compliances are generally estimated with a higher accuracy
than the normal ones, but the difference between the normal
compliances of dry and uid-lled cracks can still be detected
in the presence of noise in the data.
Parameter estimation is particularly convenient to carry out
inthenatural coordinateframeassociatedwiththepolarization
directions of the vertically propagating S-waves. These direc-
tions are controlled only by the weaknesses
Ti
tangential to
the fracture faces and are independent from the normal weak-
nesses
Ni
. In the special case of equal tangential weaknesses
T1
=
T2
, shear-wave polarizationdirections bisect the angles
between the fracture systems, and all fracture weaknesses can
be obtained just from the shear-wave splitting coefcient
(S)
and the P-wave NMO ellipse. For different tangential compli-
ances, the polarization vector of the fast shear wave deviates
toward the strike of the more compliant fracture system.
It is important to note that the axes of the P-wave NMO
ellipse generally do not coincide with either the polarization
directions of the vertically traveling S-waves or the fracture
strike (the same is true for the S-wave NMO ellipses). This
1826 Bakulin et al.
result may explainthe observations of P erez et al. (1999) who
nd that the predominant fracture orientation obtained us-
ing P-wave azimuthal moveout analysis does not agree with
that inferred from shear-wave polarizations.
We alsoexamine another monoclinic model that contains a
single systemof vertical fractures withmicrocorrugatedfaces
in an isotropic host rock. An important feature of this model
is the coupling between the slips (jumps in displacement) in
the directions normal and tangential to the fractures. This
coupling leads to the dependence of the splitting coefcient
(S)
of the vertically traveling S-waves (determined by the
tangential slip or tangential weakness) on the uid content of
the fractures (which inuences the normal slip). For typical
fracture parameters,
(S)
may noticeably decrease with uid
saturation, which is consistent with the case study of Guest
et al. (1998).
ACKNOWLEDGMENTS
This work was performed during Andrey Bakulins visit to
the Center for Wave Phenomena (CWP), ColoradoSchool of
Mines, in 1998. We thank Sean Guest and Cees van der Kolk
of Shell for their provocative discussion of the shear-wave
splitting coefcient. We are also grateful to members of the
A(nisotropy)-Team of CWP for helpful discussions and to
Andreas R uger (Landmark) for his reviewof the manuscript.
The support for this work was provided by the members
of the Consortium Project on Seismic Inverse Methods for
Complex Structures at CWP and by the U.S. Department of
Energy (award #DE-FG03-98ER14908).
REFERENCES
Bakulin, A. V., Grechka, V., andTsvankin, I., 2000a, Estimationof frac-
ture parameters fromreectionseismic dataPart I: HTI model due
to a single fracture set: Geophysics, 65, 17881802, this issue.
2000b, Estimation of fracture parameters due to reection seis-
mic dataPart II: Fractured models with orthorhombic symmetry:
Geophysics, 65, 18031817, this issue.
Bakulin, A. V., and Molotkov, L. A., 1998, Effective models of frac-
tured and porous media: St. Petersburg Univ. Press (in Russian).
Berg, E., Hood, J. A., and Fryer, G. J., 1991, Reduction of the general
fracture compliance matrix Z to only ve independent elements:
Geophys. J. Internat., 107, 703707.
1999, 3-D moveout velocity analysis and parameter estimation
for orthorhombic media: Geophysics, 64, 820837.
Grechka, V., Contreras, P., and Tsvankin, I., 2000, Inversion of normal
moveout for monoclinic media: Geophys. Prosp., 48, 577602.
Guest, S., van der Kolk, C., and Potters, H., 1998, The effect of fracture
lling uids on shear-wave propagation: 68th Ann. Internat. Mtg.,
Soc. Expl. Geophys., Expanded Abstracts, 948951.
Helbig, K., 1994, Foundations of anisotropy for exploration seismics,
in Helbig, K., and Treitel, S., Eds., Handbook of geophysical explo-
ration 22: Pergamon Press.
Hsu, C.-J., and Schoenberg, M., 1993, Elastic waves through a simu-
lated fractured medium: Geophysics, 58, 964977.
Camb. Phil. Soc., 88, 371384.
Liu, E., Crampin, S., Queen, J. H., and Rizer, W. D., 1993, Behavior
of shear waves in rocks with two sets of parallel cracks: Geophys. J.
Internat., 113, 509517.
Martin, M. A., and Davis, T. L., 1987, Shear-wave birefringence: Anew
tool for evaluating fractured reservoirs: The Leading Edge, 6, 2228.
Mensch, T., and Rasolofosaon, P., 1997, Elastic-wave velocities in
anisotropic media of arbitrary symmetrygeneralization of Thom-
sens parameters , , and : Geophys. J. Internat., 128, 4364.
Molotkov, L. A., and Bakulin, A.V., 1997, An effective model of a
P erez, M. A., Grechka, V., and Michelena, R. J., 1999, Fracture de-
tection in a carbonate reservoir using a variety of seismic methods:
R uger, A., 1997, P-wave reection coefcients for transversely
isotropic models with vertical and horizontal axis of symmetry: Geo-
physics, 62, 713722.
Sayers, C., 1998, Misalignment of the orientation of fractures and the
principal axes for P and S waves in rocks containing multiple non-
orthogonal fracture sets: Geophys. J. Internat., 133, 459466.
1983, Reection of elastic waves from periodically stratied
media with interfacial slip: Geophys. Prosp., 31, 265292.
Schoenberg, M., and Douma, J., 1988, Elastic wave propagation in me-
dia with parallel fractures and aligned cracks: Geophys. Prosp., 36,
571590.
Schoenberg, M., Dean, S., and Sayers, C., 1999, Azimuth-dependent
tuning of seismic waves reected from fractured reservoirs: Geo-
physics, 64, 11601171.
1966.
Tsvankin, I., 1996, P-wave signatures and notation for transversely
isotropic media: An overview: Geophysics, 61, 467483.
1997a, Reection moveout and parameter estimation for hori-
zontal transverse isotropy: Geophysics, 62, 614629.
1997b, Anisotropic parameters and P-wave velocity for or-
thorhombic media: Geophysics, 62, 12921309.
Winterstein, D. F., 1990, Velocity anisotropy terminology for geophysi-
cists: Geophysics, 55, 10701088.
Winterstein, D. F., and Meadows, M. A., 1991, Shear-wave polariza-
tions and subsurface stress directions at Lost Hills eld: Geophysics,
56, 13311348.
APPENDIX A
COMPLIANCE OF AN ARBITRARY ORIENTED VERTICAL FRACTURE SET
The compliance matrix of a fracture system orthogonal to
the x
1
-axis is given in equation (3). Applying Bond transfor-
mation (4) [see Winterstein (1990)] to matrix (3), Schoenberg
et al. (1999) obtain the compliance s
f
(for brevity, the number i
of the fracture set is omitted) of a fracture system with the nor-
mal n =[cos , sin , 0] (Figure A-1). The nonzero elements of
the matrix s
f
have the following form:
s
11 f
=
3K
N
K
T
8

K
N
2
cos 2
K
N
K
T
8
cos 4,
(A-1)
s
12 f
=
K
N
K
T
8
(1 cos 4), (A-2)
s
16 f
=
K
N
2
sin 2
K
N
K
T
4
sin 4, (A-3)
s
22 f
=
3K
N
K
T
8

K
N
2
cos 2
K
N
K
T
8
cos 4,
(A-4)
FIG. A-1. Set of vertical fractures with the normal making the
angle with the x
1
-axis. The azimuth is positive in the coun-
terclockwise direction.
APPENDIX B
ANISOTROPIC PARAMETERS OF MONOCLINIC MEDIA
Using the coordinate frame inwhichthe x
1
- and x
2
-axes coin-
cide with the polarization directions of the vertically traveling
shear waves, Grechka et al. (2000) dene anisotropic param-
eters for monoclinic media by generalizing Thomsens (1986)
and Tsvankins (1997b) notations for VTI and orthorhombic
symmetry systems. Expressions for these parameters in terms
of the stiffness coefcients and density are given below.
V
P0
P-wave vertical velocity:
V
P0

_
c
33
. (B-1)
V
S0
velocity of the vertically traveling S
1
-wave, which is
polarized in the x
1
-direction:
V
S0

_
c
55
. (B-2)
Dimensionless anisotropic parameters:
(1)
c
22
c
33
2c
33
, (B-3)
(1)
(c
23
c
44
)
2
(c
33
c
44
)
2
2c
33
(c
33
c
44
)
, (B-4)
(1)
c
66
c
55
2c
55
, (B-5)
(2)
c
11
c
33
2c
33
, (B-6)
(2)
(c
13
c
55
)
2
(c
33
c
55
)
2
2c
33
(c
33
c
55
)
, (B-7)
(2)
c
66
c
44
2c
44
, (B-8)
(3)
(c
12
c
66
)
2
(c
11
c
66
)
2
2c
11
(c
11
c
66
)
. (B-9)
(1)
the parameter responsible for the rotation of S
1
-wave
NMO ellipse:
(1)
c
16
c
36
2c
33
. (B-10)
(2)
the parameter responsible for the rotation of S
2
-wave
NMO ellipse:
(2)
c
26
c
36
2c
33
. (B-11)
(3)
the parameter responsible for the rotation of P-wave
NMO ellipse:
(3)
c
36
c
33
. (B-12)
Parameters (B-1)(B-9) are identical to those dened by
Tsvankin (1997b) for the higher-symmetry orthorhombic me-
dia. The additional anisotropic coefcients
(1,2,3)
are responsi-
ble for the rotation of the NMO ellipses of the waves P (
(3)
),
S
1
(
(1)
), and S
2
(
(2)
) with respect to the coordinate axes. The
coefcient
(3)
is analogous to the parameter
z
introduced by
Mensch and Rasolofosaon (1997).
APPENDIX C
WEAK-ANISOTROPY APPROXIMATION FOR MEDIA WITH TWO VERTICAL FRACTURE SYSTEMS
Here, we use equations (1)(4) and the expressions for the
compliance matrix from Appendix A to obtain linearized ap-
proximations for the anisotropic parameters dened in Ap-
pendix B. We assume that the weaknesses of both frac-
ture systems are small (
Ni
_1 and
Ti
_1) so that the
anisotropy of the effective monoclinic medium is weak. Keep-
ing only terms linear in
Ni
and
Ti
leads to the expres-
sions listed below. The parameter
(3)
is not given here since
it has no inuence on the NMO velocities of horizontal
events.
s
26 f
=
K
N
2
sin 2
K
N
K
T
4
sin 4, (A-4)
s
44 f
= K
T
1 cos 2
2
, (A-5)
s
45 f
= K
T
sin 2
2
, (A-6)
s
55 f
= K
T
1 cos 2
2
, (A-7)
s
66 f
=
K
N
K
T
2

K
N
K
T
2
cos 4. (A-8)
1828 Bakulin et al.
V
P0
= V
P
_
1
(1 2g)
2
2
(
N1
N2
)
_
. (C-1)
V
S0
= V
S
_
1

T1
4
(1 cos 2
1
)

T2
4
(1 cos 2
2
)
_
.
(C-2)
(1)
= 2g
__
(1 g)
N1
(
T1
g
N1
) cos
2
1
_
sin
2
_
(1 g)
N2
(
T2
g
N2
) cos
2
2
_
sin
2
2
_
.
(C-3)
(2)
= 2g
__
(1 g)
N1
(
T1
g
N1
)sin
2
1
_
cos
2
_
(1 g)
N2
(
T2
g
N2
)sin
2
2
_
cos
2
2
_
.
(C-4)
(1)
= 2g
_
[(1 2g)
N1
T1
]sin
2
1
[(1 2g)
N2
T2
]sin
2
2
_
. (C-5)
(2)
= 2g
__
(1 2g)
N1
T1
_
cos
2
1
[(1 2g)
N2
T2
]cos
2
2
_
. (C-6)
(1)
=
_
2(
T1
g
N1
)cos
2

T1
2
_
sin
2
_
2(
T2
g
N2
)cos
2

T2
2
_
sin
2
2
. (C-7)
(2)
=
_
2(
T1
g
N1
)sin
2

T1
2
_
cos
2
_
2(
T2
g
N2
)sin
2

T2
2
_
cos
2
2
. (C-8)
(1)
=
g
4
[2g(
N1
sin 2
1
N2
sin 2
2
)
(
T1
g
N1
)sin 4
1
(
T2
g
N2
)sin 4
2
].
(C-9)
(2)
=
g
4
[2g(
N1
sin 2
1
N2
sin 2
2
)
(
T1
g
N1
)sin 4
1
(
T2
g
N2
)sin 4
2
].
(C-10)
(3)
= g(1 2g)(
N1
sin 2
1
N2
sin 2
2
). (C-11)
The values V
P
and V
S
are the P- and S-wave velocities in the
isotropic background, and
g
V
2
S
V
2
P
. (C-12)
Since the parameters of the effective monoclinic model de-
pend on only eight parameters of the fractures and background
medium, not all of the anisotropic coefcients are independent.
Combining equations (C-9)(C-11) yields
(3)
(1)
(2)
=
1
g
2. (C-13)
It can be proved that this result remains valid for any strength
of the anisotropy. Another relation between the effective co-
efcients, which follows from equations (C-3)(C-8), has the
form
(1)
(2)
= 4g
_
(1)
(2)
_
1 2g
1 g
_
(1)
(2)
_
.
(C-14)
Equation(C-14) coincides withthe constraint derivedinpart II
for the effective orthorhombic medium caused by one system
of fractures in a VTI background.
APPENDIX D
ESTIMATION OF FRACTURE PARAMETERS IN THE WEAK-ANISOTROPY LIMIT
Thelinearizedapproximations fromAppendixCcanbeused
to invert the anisotropic parameters of monoclinic media for
the orientation and compliances of both fracture sets. A con-
venient way of performing this inversion procedure is outlined
here.
Three -coefcients combined in the form[equation (C-13)]
(3)
(1)
(2)
=
1
g
2 (D-1)
give anestimate of g or the V
S
/V
P
ratio[see equation(C-12)] in
the background. To avoid errors stemming from dividing two
small numbers, g can be approximately found from the ratio of
vertical velocities: g V
2
S0
/V
2
P0
[see equations (C-1) and (C-2)].
Two combinations,
(1)
(2)
= 2g[(1 2g)(
N1
N2
) (
T1
T2
)]
(D-2)
and
(1)
(2)
g(
(1)
(2)
) =
g
2
[4(1 g)(
N1
N2
)
(
T1
T2
)], (D-3)
can be used to nd the sums of the normal and tangential
compliances:
N1
N2
=
A B
3 2g
S
N
(D-4)
and
T1
T2
=
4 (g 1)A (1 2g)B
3 2g
S
T
, (D-5)
where
A =

(1)
(2)
2g
and B =
2
g
_
(1)
(2)
g
_
(1)
(2)
__
.
Equations (C-7) and (C-8) can be rewritten in the form
T1
cos 2
1
T2
cos 2
2
=2
_
(1)
(2)
_
D
. (D-6)
Combined with the linearized equation (9),
T1
sin 2
1
T2
sin 2
2
= 0, (D-7)
equation (D-6) can be solved for
T1
and
T2
:
T1
=
D
sin 2
2
sin 2(
2
1
)
,
T2
=
D
sin 2
1
sin 2(
2
1
)
. (D-8)
Substituting equation (D-8) into equation (D-5) yields a rela-
tionship between the fracture azimuths:
D
cos(
2
1
) = S
T
cos(
2
1
). (D-9)
Note that the denominator in equations (D-8) does not vanish
for
T1
=
T2
because the azimuths
1
and
2
in this case have
opposite signs (
1
=
2
).
Similarly, equations (C-3), (C-4), (C-9), and (C-10) can be
combined to obtain
N1
cos 2
1
N2
cos 2
2
=

(1)
(2)
2g(1 g)
D
(D-10)
and
N1
sin 2
1
N2
sin 2
2
=

(2)
(1)
g
2
S
. (D-11)
Equations (D-10) and (D-11) give the following expressions
for the normal weaknesses:
N1
=
D
sin 2
2
S
cos 2
2
sin 2(
2
1
)
,
(D-12)
N2
=
D
sin 2
1
S
cos 2
1
sin 2(
2
1
)
.
Substituting equation (D-12) into equation (D-4), we nd a
second relation between the azimuths:
D
cos(
2
1
) S
sin(
2
1
) = S
N
cos(
2
1
).
(D-13)
Equations (D-9) and (D-13) can be solved for
2

1
and
1
:
1
= tan
1
_
D
N
D
T
S
T
S
_
. (D-14)
Using equation (D-9), we nd
1
= cos
1
_
D
T
cos(
2
1
)
_
. (D-15)
Equations (D-14) and (D-15) make it possible to determine
the fracture azimuths
1
and
1
.
After recovering the fracture orientation, we can compute
the four weaknesses
Ti
and
Ni
from equations (D-8) and
(D-12). Finally, equations (C-1) and (C-2) yield the P- and
S-wave velocities in the background.
APPENDIX E
SPECIAL CASE OF EQUAL TANGENTIAL WEAKNESSES
Suppose the tangential weaknesses of the crack systems are
equal, so that
T1
=
T2

T
. (E-1)
Then it follows from equation (9) that
1
=
2
. (E-2)
Substituting equations (E-1) and (E-2) into the weak-
anisotropy approximations (C-1)(C-11) yields
V
P0
=V
P
_
1
(1 2g)
2
2
(
N1
N2
)
_
, (E-3)
V
S0
=V
S
_
1

T
2
(1 cos 2)
_
, (E-4)
(1)
=2g
_
(1 g)(
N1
N2
)
[2
T
g(
N1
N2
)]cos
2
_
sin
2
, (E-5)
(2)
=2g
_
(1 g)(
N1
N2
)
[2
T
g(
N1
N2
)]sin
2
_
cos
2
, (E-6)
(1)
=2g{(1 2g)(
N1
N2
) 2
T
]sin
2
, (E-7)
(2)
=2g{(1 2g)(
N1
N2
) 2
T
]cos
2
, (E-8)
(1)
=
_
2[2
T
g(
N1
N2
)]cos
2

T
_
sin
2
,
(E-9)
(2)
=
_
2[2
T
g(
N1
N2
)]sin
2

T
_
cos
2
,
(E-10)
(1)
=2g
2
(
N1
N2
)sin cos
3
, (E-11)
(2)
=2g
2
(
N1
N2
)sin
3
cos , (E-12)
and
(3)
= g(1 2g)(
N1
N2
)sin 2. (E-13)
Equations (E-3)(E-13) are much easier to invert for the
fracture parameters than the more general expressions of Ap-
pendix C. Having estimated g from either equation (D-1) or
the V
S0
/V
P0
ratio, we obtain the fracture azimuth as
(1)
(2)
=

(2)
(1)
= tan
2
. (E-14)
Note that the angle between the two systems of cracks is equal
to 2. The tangential weakness
T
can be found from the
shear-wave splitting coefcient
(S)
which, in the limit of weak
1830 Bakulin et al.
anisotropy, reduces to the difference between
(2)
and
(1)
:
(S)
=
(2)
(1)
=
T
cos 2. (E-15)
Then any of equations (E-11)(E-13) for the coefcients give
the difference between the normal weaknesses
N1
N2
. The
sum
N1
N2
can be determined, for instance, from
(1)
and
(2)
[equations (E-7) and (E-8)]:
N1
N2
=
1
2g 1
_
(1)
(2)
2g
2
T
_
. (E-16)
Combined with the difference
N1
N2
obtained earlier,
equation (E-16) yields the individual values of the normal
weaknesses
N1
and
N2
.
GEOPHYSICS, VOL. 65, NO. 6 (NOVEMBER-DECEMBER 2000); P. 18311836.
Deconvolving the ghost effect of the water
surface in marine seismics
Santi Kumar Ghosh
ABSTRACT
The ghost lters arising from the effect of the water
surface onbothsource andreceiver sides have a common
time domain representation that consists of a unit im-
pulse followed by its ghost, which is a delayed, negative
unit impulse. The origin of the difculties of deghosting
lies in the zeroes in the spectrumof the ghost lter, which
render incorrect any deghosting through least-squares
inverse ltering in the time domain. Another shortcom-
ing of the time domain approach is that the digital de-
scription of the ghost lter is inexact when a sampling
instant does not coincide with the instant of the onset
of the ghost impulse. A frequency domain approach, on
the other hand, is straightforward and accurate because
it can avoid the zeroes of the lter either by explicitly
choosing a recording band that excludes the zeroes or
by recording at two depths. These two depths should be
selectedaccording tothe criterionthat their highest com-
mon measure is small enough to prevent zeroes at a com-
mon frequency of the two recordings. As the source-side
andthereceiver-sideghost lters havethesameform, the
criterion derived for the selection of the depths of the re-
ceivers would also hold for the selection of the depths of
two sources whose aggregate signature is desired to have
no zeroes in the spectrum, within the operative band. An
important ramication of the analysis consists of the dis-
proof of a prevalent conjecture that the zeroes in the
spectrum of a wavelet make its autocorrelation matrix
singular; actually, the zeroes cause an inexact and unac-
ceptable least-squares inverse, although the matrix itself
is well conditioned.
INTRODUCTION
Technical reasons require both the source and the receiver
to be placed at a certain depth underwater in marine reection
seismics. The water surface, however, is responsible for distort-
Manuscript received by the Editor April 24, 1996; revised manuscript received March 27, 2000.
National Geophysical Research Institute, Uppal Road, Hyderabad 500 007, India. E-mail: postmast@csngri.ren.nic.in.
ing both the source waveform as well as the data recorded in
the receiver in a manner to be described.
As the wave is recorded after it has traveled a long path
through the seawater and the layers within the earth and as
the source-receiver offset is small in comparison, one can ap-
proximately treat this as a normal incidence propagation in
which the spherical divergence effects are nearly the same for
the source (or receiver) and its nearby image. The similarity
between the two spherical divergence effects is a consequence
of a long path of travel from the source to the receiver.
The contact between the water surface and air acts like a
perfect reector whose reection coefcient is very nearly 1.
As a result, the effective source includes not only the actual
source but also its ghost or negative image, which is situated
above the water surface at a height equal to the depth of the
source. A similar effect arises on the receiver side as well. A
hydrophone detects not only an upcoming reection but also
its negative image reected from the water-air contact. Both
effects arising in the source and receiver sides can be treated
in terms of two lters of identical forms.
The ghost effect of the water surface is a well-knownobstacle
in exploration and has been mentioned by Ziolkowski (1971),
Giles and Johnston (1973), Brandsaeter et al. (1979), and
Waters (1987), among many others. Lindsey (1960) discusses a
deghosting solution when the reection coefcient, unlike the
present case, has a magnitude smaller than unity. In describing
the ill effects of ghosting, Schneider et al. (1964) note how an
overlap between the desired data and its ghost can compound
the subsurface interpretation and how they can interfere de-
structively, masking subsurface information.
Both ghost lters, arising on source and receiver sides, have
zeroes in their frequency response at many frequencies includ-
ing the d.c. According to Ziolkowski (1971), a number of air
guns at different depths can eliminate zeroes in the amplitude
spectrum of the source, making possible an eventual decon-
volution of the source signature. The present work presents
a comprehensive remedy to eliminate the zeroes in the spec-
trum of the data caused by the receiver side ghost, an issue
not addressed by Ziolkowski (1971). The remedy consists of
recording at two depths chosen according to the criterion that
1831
1832 Ghosh
the respective ghost lters do not have common zeroes in
their spectra. A similiar criterion applies to the selection of
the depths of the sources, a crucial aspect not in Ziolkowski
(1971).
Numerical examples will demonstratethat althoughthelter
is completely known, the conventional time domain method of
constructing a least-squares inverse is poor in deconvolving the
effect of the ghost. The analysis has important ramications in
that it yields interesting insights into the relation between the
zeroes inthespectrumof awavelet andits least-squares inverse.
A conjecture by Ziolkowski (1971) that zeroes in the spectrum
of a wavelet make its autocorrelation matrix singular turns out
to be false. Nevertheless, the inexactness of the least-squares
inverse of a wavelet is shown to have its origin in the zeroes in
the spectrumof the wavelet. This suggests that a frontal assault
on the problem may be more appropriate in the frequency
domain. Indeed, analysis corroborates this conjecture.
DEFINITION OF THE PROBLEM
Ziolkowski (1971) explains the nature of the distortion in
source and receiver waveforms caused by the water surface.
The discussion here pertains to the distortion on the receiver
side; that onthe source side is obtainable inanidentical manner
with an appropriate substitution of the source depth for the
receiver depth.
Apressure-sensitive hydrophone is kept at certain depth un-
der the sea surface. The fundamental reason for this is that the
free surface of water is always pressure free and excess pres-
sure cannot be measured there. The practical reason is to avoid
noise generated on the surface due to sea waves. According to
Pieuchot (1984), the hydrophone must be operated at a depth
between10 and20 m, whereas Waters (1987) prescribes a depth
of 1213 m. In contrast, operational depths for marine seismic-
energy sources are commonly between6 and9 m(Johnstonand
Cain, 1982). While keeping a hydrophone at depth is essential,
it causes distortion of the signal in the following way.
As stated earlier, the upcoming reected waveeld observed
at the receiver can be approximated by normally incident
plane-wavetrains. I designate this waveeld by s(t ). The ghost
arising from the water-air contact is also recorded at the re-
ceiver as a negative image of s(t ) with a time delay equal to the
traveltime from the sensor at depth to the water surface and
back.
In other words, instead of s(t ), the receiver senses f (t ), given
by
f (t ) = s(t ) s(t t
o
), (1)
where the delay time t
o
=2D/V
w
, D=depth of sensor, and
V
w
=velocity of sound in water =1500 m/s, approximately. The
effect can also be described by a lter, namely,
f (t ) = s(t ) h(t ), (2)
where
h(t ) = (t ) (t t
o
), (3)
where (t ) is the familiar Dirac delta function and denotes a
convolution.
It is instructive to investigate the characteristics of the lter
in the frequency domain. The Fourier transformof h(t ) is given
by
H() = 1 exp(i t
o
), (4)
where is the angular frequency and is according to the con-
vention that Fourier transform G() of a function g(t ) is given
by
G() =
g(t ) exp(i t ) dt. (5)

Equation (4) can be rewritten as
H() = 2 sin(t
o
/2) exp[i (/2 t
o
/2)], (6)
in which the exponential term stands for the phase spectrum
and the rest for the amplitude spectrum.
In the minimum delay deconvolution model, all the l-
ters involved in the process of propagation, beginning with
the source pulse, must have a minimum phase characteristics
(Robinson, 1967). The ghost lters, both at the source and at
the receiver, are evidently not minimumphase lters. This is be-
cause a minimum phase lter must have a stable one-sided fre-
quency inverse (Robinson and Treitel, 1980), and the ghost l-
ters do not possess a frequency inverse owing to having notches
in the spectrum at many frequencies (including the d.c. fre-
quency), as evident from equation (6). Therefore, the effect of
these lters cannot be removedandthe lost frequencies cannot
be restored by conventional deconvolution, even in principle.
This is in contrast with other earth lters such as the reverber-
ation lter, which is known to have a minimum phase with no
notches in the spectrum and which has a standard deconvolu-
tion operator. One also notes that no linear lter can restore
the frequencies lost due to the ghost effect. This would neces-
sitate design of a deconvolution scheme specic to the present
problem.
A comprehensive solution must also reconcile two conict-
ing requirements. While a hydrophone depth between 10 and
20 m would cause the rst non d.c. zero of the ghost spec-
trum between 37.5 and 75 Hz, the current emphasis on high
resolution requires a band width of 250 Hz (Pieuchot, 1984).
Thus on one hand, the requirement of high frequencies for
improving resolution and the consequent need to push the ze-
roes of the ghost lter beyond the Nyquist range enforce a
small depth for the hydrophone. On the other, the minimiza-
tion of sea wavegenerated noise requires a reasonably large
depth for the hydrophone. These conicting demands would
be reconciled satisfactorily if one could, by some means, retain
a large depth for the hydrophone without compromising the
requirement of a large bandwidth.
TIME DOMAIN DECONVOLUTION
After ascertaining the description of the ghost lter from a
knowledge about the depth of the hydrophone, one can make
a rst attempt at deghosting in the time domain by an inverse
lter whichis optimal inthe least-squares sense. Inother words,
given the ghost lter h(t ), one designs an inverse lter x(t )
such that its convolution with h(t ) is as close to (t ) as possible.
Recalling equation (2), it is seen that then the desired data
s(t ) is, in principle, recoverable by the convolution of f (t ), the
recorded data, with x(t ).
It is worthwhile to begin with a realistic example involving
a hydrophone at a depth of 12 m under the sea surface and a
Surface Deghosting in Marine Seismics 1833
sampling interval of 4 ms. This yields a two-way time t
o
of 16 ms
for the travel from the sensor to the water surface and back.
Under these circumstances, the ghost lter h(t ) has a discrete
time representation
h
t
= (1, 0, 0, 0, 1). (7)
One can construct an inverse of h
t
by the standard method of
nding the least-squares inverse (Robinson and Treitel, 1980).
Here the input is h
t
, and the output is d
t
=(1, 0, 0, 0 . . .), the
discrete representation of (t ). What is sought is x
t
, the optimal
inverse of h
t
. The autocorrelation of h
t
, symmetric about zero
time lag is given by
hh
() = (2, 0, 0, 0, 1) (8)
for positive time lag. The corresponding autocorrelationmatrix
hh
turns out to be
hh
=
2 0 0 0 1
0 2 0 0 0
0 0 2 0 0
0 0 0 2 0
1 0 0 0 2
. (9)
The inverse of this matrix is given by
1
hh
=
2/3 0 0 0 1/3
0 1/2 0 0 0
0 0 1/2 0 0
0 0 0 1/2 0
1/3 0 0 0 2/3
. (10)
If the successive sample values of x
t
constitute a column vector,
X =
x
0
x
1
x
2
x
3
x
4
,
then the solution of the normal equations yield
X =
1
hh
dh
, (11)
where
dh
is the crosscorrelation vector between d
t
and h
t
. This
crosscorrelation turns out to be d
t
itself for nonegative time
lags. Since d
t
=(1, 0, 0 . . .), x
t
=(2/3, 0, 0, 0, 1/3), i.e., merely
the successive entries in the rst column of the inverse of the
autocorrelation matrix in equation (10).
The performance of the lter x
t
is readily evaluated by com-
puting the actual output a
t
obtainable through the convolution
of h
t
and x
t
:
a
t
= (2/3, 0, 0, 0, 1/3, 0, 0, 0, 1/3). (12)
The poor spiking entails a normalized mean-square error of
1/3, a reduced amplitude at zero time, and two negative ghosts
of appreciable magnitude. Investigations reveal that neither in-
creasing the length of the inverse lter nor delaying the desired
output improves the result.
The autocorrelation matrix has an inverse and is far from
being singular. Indeed, it is well conditioned. The condition
number, dened as the ratio of the moduli of the largest and
the smallest eigenvalues turns out to be 3. This goes against the
conjecture of Ziolkowski (1971) that the zeroes inthe spectrum
of a wavelet lead to a singular autocorrelation matrix.
The poor quality of the inverse lter indeed is related to the
zeroes in the spectrum of the wavelet. The aforesaid ghost l-
ter (1, 0, 0, 0, 1) with a sampling interval of 4 ms had zeroes
at 62.5 Hz (half of the Nyquist frequency) and at 125 Hz (the
Nyquist frequency itself). This fact is responsible for the poor
spiking. To demonstrate this claim, consider a hypothetical ex-
ample with a hydrophone depth of 3 m, a rather unrealistic
choice in noisy sea conditions. It is easy to verify that with the
same sampling interval of 4 ms the ghost lter nowassumes the
form
h
t
= (1, 1), (13)
whichhas one zeroat the d.c. frequency andwhose other zeroes
are well beyond the Nyquist frequency. It will now be shown
that this lter has a least-squares inverse that is exact in an
asymptotic sense. Following a standard exercise of spiking, one
candesignaninverse of any length. By progressively increasing
the length of the lter, one discovers a pattern in the structure
of the inverse lter. The pattern reveals that an inverse lter
of length n has the representation
x
t
= [n/(n +1), (n 1)/(n +1), . . . 2/(n +1), 1/(n +1)].
(14)
The actual output a
t
, obtainable by a convolution of h
t
and x
t
,
turns out to be
a
t
= [n/(n +1), 1/(n +1), 1/(n +1), . . . 1/(n +1)].
(15)
The normalized error energy becomes
E = 1/(n +1). (16)
Evidently, as n tends to innity, a
t
approaches a spike of unit
amplitude whose tail assumes zero values everywhere. One
can argue that it is the absence of zeroes in the Nyquist range
in the frequency response of the lter which lends it an exact
inverse and makes it akin to a minimum phase lter. For a
reasonable performance, the lter length must be long, as seen
from equation (16).
In addition to the fact that the depth of the hydrophone is
unrealistically shallow, there is an implementational problem
with this inverse lter. The undesired tail in the actual output
in equation (15) is nothing but a negative and scaled rectangle
function, rich in d.c. and near-d.c. frequencies. The area un-
der the actual output wavelet is zero, indicating the lack of d.c.
component in it. The negative tail has the effect of reducing the
amplitudes of the very low frequencies. This is because when
convolved with the data, the tail has a low-pass effect because
of its smoothing property, and a negative sign in effect sub-
tracts the low-frequency amplitudes from the principal spike
at zero time, which contains all frequencies in the band. The
result is a suppression of very low frequencies whose phase
difference with the main spike is small enough to permit arith-
metic subtraction of Fourier amplitudes in place of vectorial
subtraction. Higher frequencies will undergo phase distortion
in addition to the amplitude distortion. In short, the effect of
1834 Ghosh
the tail compromises the very purpose of deconvolution, which
aims at an equal emphasis and a zero phase for all frequencies.
There is another difculty with deghosting in the time do-
main. As seen from equation (3), the representation of the
ghost lter h(t ) involves delta functions in continuous time. A
discrete representation implies band limiting. A delta function
band limited between the frequencies
m
and +
m
assumes
the form I
b
(t ) given by
I
b
(t ) = (1/) sin(2
m
t )/t. (17)
At t =0, I
b
(t ) is maximumand equals 2
m
. If one wishes to sam-
ple this function, the sampling interval t can be legitimately
chosen as 1/2
m
. The zeroes of I
b
(t ) occur at
t = nt, n = 1, 2, . . . . (18)
Thus, except for t =0, all other sample values of I
b
(t ) are iden-
tically zero. That is the reason that the delta function has an
exact discrete representation (1, 0, 0 . . .) when one ignores a
scale factor of 2
m
. As long as the sampling instant coincides
with the instant of onset of the impulse, the delta function will
continue to have such nite representation. When that is not
so, the exact representation will contain an innity of nonzero
sample values extending in both positive and negative time
direction. For actual computation, such a sequence must be
truncated at the cost of error. If one chooses 10 m as the depth
of the hydrophone and a sampling interval of 4 ms, then the
value of t
o
in the representation of h(t ) in equation (3) would
be 13.3 ms, which will not coincide with a sampling instant.
This will require a truncated representation of h(t ), involving
complication and error.
All of the three preceding examples argue against carrying
out the deghosting in the time domain. The next section shows
that the frequency domain offers better prospects.
FREQUENCY DOMAIN DECONVOLUTION
Since the difculties of deghosting originate from the zeroes
in the response of the ghost lter, a frontal assault on the prob-
lem is more apt in the frequency domain. A frequency domain
solution is readily possible if the recording is done in a fre-
quency band where the lowcut point is a little above zero (4 or
5 Hz) and the highcut limit is below 1/t
o
. This band contains
no zeroes in the lter response, and the effect of the lter can
be removed. From the frequency domain equivalent of equa-
tion (2), one can write
S() = F()/H(), (19)
where H() is given by equation (6). Operation (19) is sta-
ble because H() is nonzero within the band. Next, s(t ), the
desired data, can be recovered by taking an inverse Fourier
transform of S(). This, however, has limited applications, as
in the case below.
In marine seismics intended for the study of gross geologic
features, oneoperates inalower frequencyrange, typicallywith
a lowcut at 4 Hz and a highcut at 64 Hz. In such a case, a depth
of 10 m would produce a second zero in the lter response at
75 Hz. This is tolerable for the objective. Also, a depth of 10 m
is often adequate for avoiding surface noise.
If, however, the restriction of a smaller bandwidth is not
tolerable, a comprehensive solution to the problem is possible
by making the following modication in the data acquisition
procedure. Theonlyadditional requirement is torecordnot just
at one depth, but at two successive depths. Let these depths
be D
1
and D
2
with D
2
> D
1
. Now the upcoming signal s(t )
will reach D
2
and, after some delay, D
1
. If one xes D
2
as the
reference, then the upcoming signal at D
2
will be s(t ) and that
at D
1
will be s[t (D
2
D
1
)/V
w
]. In addition, each sensor will
record the respective delayed images of the signal. Therefore,
the recorded trace at D
2
would be
f
2
(t ) = s(t ) [(t ) (t t
2
)], (20)
and that at D
1
would be
f
1
(t ) = s[t (D
2
D
1
)/V
w
] [(t ) (t t
1
)], (21)
where t
1
=2D
1
/V
w
and t
2
=2D
2
/V
w
.
After transformationintothe frequency domain, the last two
equations would assume the following forms:
F
1
() = exp[i (D
2
D
1
)/V
w
]S()[1 exp(i t
1
)]
(22)
and
F
2
() = S()[1 exp(i t
2
)]. (23)
A rearrangement of equation (22) followed by its addition to
equation (23) yields
F
1
() exp[i (D
2
D
1
)/V
w
] + F
2
()
= S()[2 exp(i t
1
) exp(i t
2
)]. (24)
From equation (24), one obtains
S() = {F
1
() exp[i (D
2
D
1
)/V
w
] + F
2
()}/
[2 exp(i t
1
) exp(i t
2
)]. (25)
The process of computing S() is nowstable except at =0.
This is unavoidable but is inconsequential in practice, as the
zero frequency is never recorded. Except for this pathologic
case, the denominator in the last equation can always be made
nonvanishing in principle by a proper choice of D
1
and D
2
. As
shown in the Appendix, the proper choice turns out to be the
criterionthat D
1
and D
2
shouldhave nocommonmeasure, (i.e.,
D
1
/D
2
should be irrational). In practice, however, D
1
and D
2
will have common measures and the second zero will occur at
a nite frequency given by
=V
w
/(2the highest common measure of D
1
and D
2
),
(26)
as established in the Appendix.
For example, if D
2
=13 m and D
1
=10 m, their highest com-
mon measure is 1 m. This yields =750 Hz, which is well be-
yond any seismic frequency recorded in practice.
Ziolkowski (1971) suggests putting sources at multiple
depths as a means of eliminating the zeroes of the ghost l-
ter on the source side, though he does not provide a criterion
for choosing the depths. Analysis done here for deghosting on
the receiver side obtains this criterion that applies to the source
side as well because both ghost lters have identical forms. It is
evident that two sources at depths D
1
and D
2
are sufcient to
avoid zeroes in the frequency response of the aggregate ghost
lter, provided that D
1
and D
2
are chosen so that their high-
est common measure is as low as possible in order to push the
Surface Deghosting in Marine Seismics 1835
zeroes of the aggregate ghost lter beyond a frequency band
whose width is given by equation (26). An aggregate signature
of the sources and their ghosts recorded at large enough depth
can then be used for an effective signature deconvolution.
The ghost lters arising from the effect of the water surface
on both source and receiver sides are quite similar to each
other and dissimilar to the lters of the convolutional model.
Whereas the earth lters are known to have a minimum phase
character, the ghost lters do not have minimumphase charac-
teristics when the lter response contains zeroes in the Nyquist
range.
The ghost lters both on the source and the receiver sides
have zeroes in their frequency response. It is shown that the
standard time-domain method of inverse lters cannot effec-
tively get rid of the ghost. For realistic hydrophone depths and
for a reasonably large bandwidth, the ghost lter must con-
tain zeroes in the signal spectrum in addition to the manda-
tory zero at the d.c. frequency. These additional zeroes are
responsible for an imperfect deghosting, which consists of a
split of the ghost into two smaller ghosts while the main spike
is only two-thirds of its desired amplitude. Both increasing the
length of the inverse lter and delaying the temporal position
of the desired spike turn out to be fruitless exercises in this
case.
Though unrealistic, one can conceive of a shallow hy-
drophone depth which would restrict the ghost lters length to
two samples, thus avoiding zeroes in the lter spectrum within
the Nyquist range except for the essential zero at the d.c. fre-
quency. In this case, the inverse lters performance improves,
as its length increases, approaching that of the exact asymp-
totically. The inverse lters problem, however, lies in actual
implementation, when the lter length must be nite. Then,
the actual output turns out to be a spike followed by a negative
tail of uniform magnitude and whose length equals the length
of the lter. When convolved with the actual data, the tail has
a detrimental effect: the lower frequencies suffer attenuation
and even their phases are altered, compromising the very aim
of deconvolution.
The last two instances also make a point regarding the in-
verse of the ghost lter. When the ghost spectrum contains
zeroes, the inverse lter performs poorly, giving rise to rather
unsatisfactory deghosting. On the other hand, when the spec-
trum has no zeroes in the Nyquist range, the inverse lter be-
comes exact as its length tends to innity. The conjecture in
Ziolkowski (1971) that zeroes in the spectrum of a wavelet
make its autocorrelation matrix singular is belied in the treat-
ment of the rst instance. The autocorrelation matrix turns
out to be well conditioned. The link, however, lies in the poor
spiking ability of the autocorrelation matrix when the wavelet
spectrum has zeroes.
Another problem associated with the least-squares deghost-
ing in the time domain is the error and the complication which
may arise in the digital description of the ghost lter in which
theonset of theghost impulsedoes not coincidewithasampling
instant. In this case, the lters exact digital representation is
not nite, and any truncation for computational purposes leads
to error and complication.
In viewof the above, a frequency domain deghosting is more
appropriate. When the recording band has a relatively small
width and does not include the frequencies at which the ghost
lter has zeroes, a frequency domain deghosting is readily pos-
sible. When, on the other hand, one aims at a wide recording
band for greater resolution, one must record the seismic eld
at two depths. The depths are chosen such that the respective
ghost lters do not have common zeroes in their spectrum. A
criterion which yields this condition is that the depths D
1
and
D
2
should have a highest common measure that is as low as
possible with the width of the recording band being given by
equation (26).
Equation (25) is based on the assumption of a reected
plane wavefront traveling vertically upward. When the wave-
front normal (the ray) is not vertical but away from it by an
angle , the effective delay time of the ghost can be shown
to be t
o
cos , where t
o
[one recalls from equation (1)] is the
delay time for normal incidence. Whereas for normal inci-
dence the delay of the ghost corresponds to the two-way ver-
tical distance from the hydrophone to the water surface, for
oblique propagation the delay is produced along an oblique
ray which is initially parallel to the ray carrying from below
the true signal to the hydrophone and which after reection
at the free surface comes down to the hydrophone to pro-
duce the ghost. It is easy to verify that the decrease in de-
lay time is by about 3% for an emergence angle of 15
and
around 6%for an angle of 20
with the vertical. One, of course,

bears in mind that an emergence angle of 20
in water would
correspond by Snellss law for a horizontally stratied earth
to near-critical angles of reection in the deeper sediments
whose compressional velocity would often be more than two-
and-a-half times that of water. This would impose tolerable
limits on the error caused on equation (25) by nonvertical
propagation.
To implement this method of deghosting one must have two
parallel recording arrangements. This requires operating two
streamers at two preset depth levels constantly monitored by
depth transducers incorporated in the streamers. Also, an ac-
curate knowledge of the depths of the hydrophones is essential
for deghosting. The two recording systems must also be iden-
tical in their electromechanical response. All this will surely
increase the cost of operation, but it offers the benet of a
superior deghosting result.
In suggesting multiple sources at different depths,
Ziolkowski (1971) was aiming at avoiding common zeroes in
the source spectrum. He, however, provided no criterion for
selecting the depths. Analysis here points to the requirement
of only two sources at two depths chosen according to the crite-
rion, as in the case of receiver depths, that the highest common
measure of the two source depths should be as small as possible
in order to yield a source spectrum as broad as required.
ACKNOWLEDGMENTS
It is a pleasure to thank Dr. H. K. Gupta, Director, National
Geophysical ResearchInstitute, for his kindpermissiontopub-
lish this paper and for his constant encouragement.
This workwas originally submittedto The Leading Edge. Dr.
Sven Treitel, the then SEG Editor, was considerate enough to
think it appropriate for GEOPHYSICS and to put it in the review
process.
The suggestions by P. G. Kelamis, Associate Editor, and
C. G. Macrides, the reviewer, have rendered the manuscript
more readable.
1836 Ghosh
Thanks are due to Mr. V. Subrahmanyamfor typing the orig-
inal manuscript. I am grateful to Mr. G. Ramakrishna Rao for
keying the revised manuscript into a word processing system.
REFERENCES
Brandsaeter, H., Farestveit, A., and Ursin, B., 1979, A new high-
resolution or deep penetration airgun array: Geophysics, 44, 865
879.
Giles, B. F., and Johnston, R. C., 1973, Systemapproach to air-gun array
design : Geophys. Prosp., 21, 77101.
Johnston, R. C., and Cain, B., 1982, Marine seismic energy sources:
Acoustic performance comparison: Presented at the 14th Offshore
Tech. Conf.
Lindsey, J. P., 1960, Elimination of seismic ghost reections by means
of a linear lter: Geophysics, 25, 130140.
Pieuchot, M., 1984, Seismic instrumentation: Geophysical Press.
Robinson, E. A., 1967, Multichannel time series analysis with digital
computer programmes: Holden-Day.
Robinson, E. A., and Treitel, S., 1980, Geophysical signal analysis:
Prentice-Hall, Inc.
Schneider, A. W., Larner, K. L., Burg, J. P., and Backus, M. M.,
1964, A new data-processing technique for the elimination of
ghost arrivals on reection seismograms: Geophysics, 29, 783
805.
Waters, K. H., 1987, Reection seismology: John Wiley & Sons, Inc.
Ziolkowski, A., 1971, Design of a marine seismic reection prole sys-
tem using airguns as a sound source: Geophys. J. Roy. Astr. Soc., 23,
499530.
APPENDIX
DERIVATION OF THE CRITERION FOR CHOOSING THE HYDROPHONE DEPTHS
The computation of S() is unstable whenever the denomi-
nator in equation (25) is zero, i.e.
2 exp(i t
1
) exp(i t
2
) = 0. (A-1)
For nonzero frequencies, this would imply that
t
1
= 2k (A-2)
and
t
2
= 2n, (A-3)
where both k and n are integers. In terms of frequency , equa-
tions (A-2) and (A-3) can be combined to
= k/t
1
= n/t
2
. (A-4)
Recalling that t
1
=2D
1
/V
w
and t
2
=2D
2
/V
w
, equation (A-4)
can be readily rewritten as
D
1
/D
2
= k/n. (A-5)
The last equation provides the condition for the denominator
in equation (25) to become zero.
If D
1
and D
2
have no common measure (i.e., if D
1
/D
2
is ir-
rational), then condition (A-5) would never be satised. This,
however, may not be realistic for an actual marine seismic ex-
periment, and one allows D
1
and D
2
to have common measures
in a tolerable manner. Thus, one permits D
1
=ak and D
2
=an,
such that
D
1
/D
2
= (ak)/(an) = k/n, (A-6)
where k and n are two integers with no common factor between
them. Consequently, a is the highest commonmeasure between
D
1
and D
2
. One can then nd the lowest frequency at which
the denominator would become zero. Using equations (A-4)
and (A-6), one can write
= V
w
/(2D
1
/k) = V
w
/2a. (A-7)
Equations (A-7) and (26) make identical statements.
An approach to upscaling for seismic waves in statistically
isotropic heterogeneous elastic media
N. Gold
, S. A. Shapiro
, S. Bojinski
, and T. M. M uller
ABSTRACT
For complex heterogeneous models, smoothing is of-
ten required for imaging or forward modeling, e.g., ray
tracing. Common smoothing methods average the slow-
ness or squared slowness. However, these methods are
unable to account for the difference between scatter-
ing caused by uctuations of the different elastic moduli
and density. Here, we derive a new smoothing method
that properly accounts for all of the parameters of an
isotropic elastic medium. We treat the geophysical prob-
lemof optimumsmoothing of heterogeneous elastic me-
dia as a problem of moving-window upscaling of elastic
media. In seismology, upscaling a volume of a hetero-
geneous medium means replacing it with a volume of a
homogeneous medium. In the low-frequency limit, this
replacement should leave the propagating seismic wave-
eld approximately unchanged. A rigorous approach to
upscaling is given by homogenization theory. For ran-
domly heterogeneous models, it is possible to reduce the
problem of homogenization to calculating the coherent
waveeld (mean eld) in the low-frequency limit. After
deriving analytical expressions for the coherent wave-
eld in weakly heterogeneous and statistically isotropic
randommedia, we obtain a smoothing algorithm. We ap-
ply this algorithm to random media and to deterministic
models. The smoothing algorithm is frequency depen-
dent, i.e., for different dominant frequencies, different
smooth versions of the same medium should be consid-
ered. Several numerical examples using nite differences
demonstrate the advantages of our approach over com-
mon smoothing schemes. In addition, using a numerical
eikonal equation solver we showthat, in the case of com-
plex heterogeneous media, appropriate initial smooth-
ing is important for high-frequency modeling.
INTRODUCTION
The modeling of wave propagation in heterogeneous media
is increasingly important in seismic exploration. For example,
from vertical and horizontal log-based geostatistical modeling
of reservoirs, extremely complex models of media can be ob-
tained. Upscaling of heterogeneous media is necessary to sim-
plify models of the earth. In seismology, upscaling a volume
of a heterogeneous medium usually means replacing it with
a volume of a homogeneous medium. This substitution should
not change the propagating seismic waveeld. Atypical upscal-
ing problem for seismic waves is the smoothing of heteroge-
neous models. In seismic processing it is often useful to smooth
models for later application of imaging or ray-tracing methods
(Grubb and Walden, 1995). Since ray-tracing methods can be
correctly applied only to media with spatial uctuations larger
than the wavelength (

Cerveny et al., 1977), smoothing is nec-
Published on Geophysics Online May 30, 2000. Manuscript received by the Editor June 23, 1997; revised manuscript received April 7, 2000.
Karlsruhe University, Geophysical Institute, Hertzstrasse 16, 76187 Karlsruhe, Germany. E-mail: ngold@geophysik.uni-karlsruhe.de; sbojinsk@
geo.unitz.cr.
Freie Universit at Berlin, Fachrichtung Geophysik, Malteserstrasse 74-100, Haus D, 12249 Berlin, Germany. E-mail: shapiro@geophysik.fu-berlin.
de; tmueller@geophysik.fu-berlin.de.
essary for media containing small-scale inhomogeneities. Usu-
ally, smoothing methods average the slowness or the squared
slowness. However, since uctuations of the Lam e parameters
and the density cause different scattering effects, averaging
methods using only the velocities cannot smooth the medium
in a way that properly accounts for the effects of stiffness and
density uctuations.
A rigorous procedure for smoothing must correctly account
for scatteringeffects of heterogeneities of thesmall spatial scale
(a scale much smaller than the dominant wavelength). This
means the difference between the waveeld propagating in the
actual medium and the waveeld propagating in a smoothed
version of this medium must be minimized.
Arigorous approachtosmoothinginheterogeneous mediais
homogenization theory. The mathematical background of this
theory was summarized and further developed by Bensoussan
et al. (1978) andSanchez-Palencia (1980). The homogenization
1837
1838 Gold et al.
approach is briey explained as follows. In heterogeneous me-
dia the elastic parameters and the density (let us denote them
as functions f
i
) depend on the location r, f
i
= f
i
(r). To char-
acterize the nonsmooth behavior of f
i
(r) we must introduce a
lengthcharacterizingour observationsystem, i.e., thedominant
wavelength . Additionally, we introduce a small parameter .
With respect to our observation system, the functions f
i
will
be nonsmooth if they signicantly uctuate on the scale .
Apart from the uctuations on the small scale, the functions f
i
can uctuate independently on a larger scale. Therefore, they
are functions of two arguments, f
i
=g
i
(r/, r), where the rst
argument describes the uctuating character of f
i
on the small
scale.
The functions g
i
dene the linear differential operator
L
g
(r/, r) of the dynamic equation (i.e., Lam es equation):
L
g
_
r
, r
_
u
_
r
, r
_
= 0, (1)
where u(r/, r) is the displacement. In the homogenization ap-
proach a solution of this equation is sought in the form of the
following series:
u
_
r
, r
_
= u
0
(r) +u
1
_
r
, r
_
+
2
u
2
_
r
, r
_
+ . (2)
Under some conditions (which are applicable to usual
geological heterogeneous structures) homogenization theory
shows that the rst term, u
0
(r), in series (2) satises the follow-
ing new Lam e equation:
L
0
(, r)u
0
(r) = 0, (3)
where L
0
(, r) is the corresponding linear differential Lam e
operator. This equation describes the main part of the wave-
eld[onthe order of 1, while other terms of series (2) are onthe
order of and higher], which propagates in an elastic medium
without small-scale uctuations. This medium is a smoothed
counterpart of the original medium. Its density and elastic pa-
rameters [whichnowdenethenewLam eoperator L
0
(, r)] do
not depend on the small-scale variable r/. One can show that
if 0, then u u
0
and L
g
L
0
. Therefore, homogenization
theory provides a rigorous smoothing algorithm.
Moreover, in nonsmooth heterogeneous media equation (3)
is subject to further approximations such as geometrical op-
tics or ray tracing. However, the operator L
0
depends on
the small parameter which is the ratio between the char-
acteristic size of the small spatial-scale uctuations and the
wavelength. Therefore, L
0
is wavelength dependent. In other
words, for different frequencies one must work with differ-
ently smoothed versions of the same nonsmooth medium. This
has sound physical consequences. For instance, the ray-tracing
approximation must be wavelength dependent because ray
sets are different in different smooth versions of the same
medium.
In the general case of an arbitrary heterogeneity, the ho-
mogenization approach does not provide any closed analytical
recipe for smoothing. Therefore, to arrive at a general and sim-
ple smoothing rule, we must introduce some simplications.
Let us consider equation (3) in a spatial domain which is still
larger than the characteristic length of the small-scale uctua-
tions but which is so small that elastic moduli and density are
independent of the large spatial-scale variable r. In this case
equation (3) has the form
L
0
()u
0
(r) = 0. (4)
This equation implies that the smooth version of the medium
provided by the homogenization approach is a homogeneous
medium. Because the density and the elastic parameters of
this homogeneous medium are found in the long-wavelength
limit, it is often called the effective medium. Therefore, in this
limit the problem of smoothing is reduced to the problem of
upscaling elastic properties in heterogeneous media. Thus, if
we carefully scale up the elastic properties of the medium in a
very small spatial window and then move this window in space
and replace the elastic parameters in the center of the window
by the corresponding upscaled parameters, we arrive at a well-
substantiated smoothing procedure.
Generally, even in the case of equation (4) the homoge-
nization approach does not provide a simple analytic solution.
However, for randomly heterogeneous media equation (1) can
be solved for the averaged (coherent) waveeld u, where av-
eraging stands for statistical or spatial averaging on the small
spatial scale . Moreover, Keller (1977) shows that u =u
0
if
0. Thus, instead of solving the homogenization problem,
we can look for the coherent waveeld in the low-frequency
range. Assuming random weak heterogeneities, an analytical
recipe for upscaling arbitrary isotropic structures can be ob-
tained. This is the strategy of our paper.
For weakly heterogeneous acoustic random media, analyti-
cal solutions for the coherent part of the waveeldalready exist
(Rytov et al., 1987; Shapiro et al., 1996). In general, the effec-
tive medium, i.e., a ctitious homogeneous medium where the
waveeld equals the coherent eld, is dispersive and absorb-
ing. However, in the long-wavelength limit the result for the
effective medium is neither absorbing nor dispersive. In this
paper we derive the coherent waveeld in weakly uctuating
elastic random media by means of the elastic Bourret approx-
imation (see Appendix A). Using the low-frequency limit of
this theory, we nd a frequency-dependent smoothing method
where the smoothing area is small comparedtothe wavelength.
To demonstrate the usefulness of our smoothing method, we
show results of nite-difference modeling in original and dif-
ferently smoothed media. Finally, using a nite-difference so-
lution of the eikonal equation, we show that for complex het-
erogeneous media appropriate initial smoothing is important
for high-frequency modeling.
EFFECTIVE SLOWNESSES
The elastic Bourret approximation (Appendix A) provides
the effective Greens function for waveelds in an elastic
isotropic random medium, assuming small uctuations of the
mediumparameters andlowfrequencies. As a result, we obtain
the effective wavenumbers of the P- and S-waves, depending
on frequency, the spatial correlation functions of the medium
parameters, and their crosscorrelation functions.
Tocalculatethewavenumbers, theparameters of therandom
medium are assumed to be the sum of the parameters of a
homogeneous background medium, where density and Lam e
parameters are spatially averagedmeanvalues (
o
,
o
,
o
), and
of the uctuations of the medium parameters (
):
Upscaling for Seismic Waves 1839
(r) =
o
(1 +
(r));
= 0,
(r) =
o
(1 +
(r));
= 0, (5)
(r) =
o
(1 +
(r));
= 0.
As a precondition, the statistical average of the uctuations
vanishes.
We nd the following results for the slownesses of the co-
herent P-wave and S-wave in a 3-D random medium in the
low-frequency and weak uctuation limits:
S
p
e
= S
p
o
_
1 +
1
2
2
o
(
o
+2
o
)
2
+
2
3
o
(
o
+2
o
)
2
+
2
5
2
o
(
o
+2
o
)
2
+
4
15
o
(
o
+2
o
)
_
, (6)
S
s
e
= S
s
o
_
1 +
1
5
+
2
15
o
+2
o
_
. (7)
For 2-D media we nd
S
p
e
= S
p
o
_
1 +
1
2
2
o
(
o
+2
o
)
2
+

o
o
(
o
+2
o
)
2
+
3
4
2
o
(
o
+2
o
)
2
+
1
4
o
(
o
+2
o
)
_
, (8)
S
s
e
= S
s
o
_
1 +
1
4
+
1
4
o
+2
o
_
. (9)
In these equations
2
xz
denotes the normalized covariance of
uctuations of arbitrary parameters x and z,
2
xz
=
x
z
, and
S
p
o
and S
s
o
are the slownesses of P-waves and S-waves in the
homogeneous background medium, respectively:
S
p
o
=
_

o
o
+2
o
_
1/2
, (10)
S
s
o
=
_
o
_
1/2
. (11)
The result depends only on the background medium parame-
ters, the variances, and the covariances of the Lam e parame-
ters. Density uctuations have no effect on the properties of
the effective medium in the case of low frequencies because
variances of (r) do not occur in equations (6)(9). Again,
we emphasize those results have been obtained in the low-
frequency and small-contrast (i.e., weak-uctuation) approxi-
mations. Therefore, all terms with third and higher powers of
the medium-parameter uctuations have been neglected. We
denote such terms as O(
3
).
The medium parameterization given by equation (5) is not
the only one possible. In the case of another parameteriza-
tion, equations (5)(11) must be changed. For instance, let us
describe the random medium using the compliances instead
of the Lam e parameters. Now the parameters of the random
medium are the sum of the parameters of a homogeneous
background medium, where the density and the compliances
m =( +2)
1
and =
1
are spatially averaged mean val-
ues (m
o
,
o
,
o
) and of the uctuations of the medium para-
meters (
m
,
):
m(r) = m
o
(1 +
m
(r));
m
= 0,
(r) =
o
(1 +
(r));
= 0, (12)
(r) =
o
(1 +
(r));
= 0.
To obtain the new forms of equations (6)(9), we must express
the Lam e parameters in terms of the new parameterization:
=

o
2m
o
m
o
o
_
1 +
2m
o
_
o
_
m

2
m
_
o
2m
o
_
, (13)
=
1
o
_
1
+
2
_
. (14)
In these two equations we neglected all terms of order O(
3
)
and higher. The statistical averaging of these equations pro-
vides us with the properties of the old background medium:
o
=

o
2m
o
m
o
o
_
1 +
2
mm
2m
o
o
2m
o
_
, (15)
o
=
1
o
_
1 +
2
_
. (16)
Now equations (13)(16) can be used directly to express the
variances of the Lam e parameters in terms of the variances of
the compliances:
2
o
2
mm
+4m
2
o
4m
o
2
m
(
o
2m
o
)
2
, (17)
=
2
, (18)
2
m
2m
2
o
o
2m
o
. (19)
In the next step, to express the effective slownesses of P- and
S-waves in terms of the new parameterization, we must substi-
tute results (15)(19) into equations (6)(11). This procedure
will provide us witha newformfor equations (6)(9). The phys-
ical meaning and the values of slowness will not change [with
accuracy up to the terms on the order of O(
3
) and higher].
Any other parameterizationof the mediumcanbe chosen. In
the following we use our parameterization and equations (5)
(11), respectively. This parameterization is convenient in the
sense of its direct applicability to the Bourret approximation.
UPSCALING AND SMOOTHING
In this section we establish the smoothing method for an
arbitrary point O in an isotropic elastic random medium. First
we dene a circular area aroundthis point. Using the gridpoints
in this area, we determine the mean values, the variances, and
the covariances in equations (6)(9):
x
o
=
1
a
_
a
x(r) d
2
r, (20)
z
o
=
1
a
_
a
z(r) d
2
r, (21)
2
xz
=
1
x
o
z
o
a
_
a
(x(r) x
o
)(z(r) z
o
) d
2
r, (22)
1840 Gold et al.
where
a
denotes the circular area (spherical volume) of radius
a around point O,
a
=
a
,
and x and z are arbitrary properties of the medium. Numeri-
cally, a summation is performed in these equations for all grid
points in the area
a
. Using equations (6)(9), we calculate the
effective slownesses for point O. Using the averaged density
and the rule
e
=
o
, we can replace the real medium parame-
ters by the upscaled medium parameters:
e
=

o
_
S
s
e
_
2
, (23)
e
=

o
_
S
p
e
_
2
2
e
. (24)
The relation of the radius of the averaging area to the wave-
length is a subject of heuristic choice. However, the theo-
retical approach is based on the low-frequency assumption.
Therefore, the radius can be considered small compared to the
wavelength. In the following we discuss this issue using sev-
eral numerical examples. Finally, by moving the circular area
(spherical volume in three dimensions)
a
and repeating the
FIG. 1. A realization of the random medium with an expo-
nential correlation function and a correlation length of 10 m.
The area of the medium is 500 500 m
2
; the grid spacing is
1 m. Bright spots denote areas with large parameter values
compared with the mean value; in dark areas, parameters are
smaller than the mean value.
FIG. 2. Parameters of medium 1 at a depth of 200 m. The standard deviation of the stiffness tensor elements is 20%; M = +2
and are perfectly correlated.
above procedure for all points of a model, the smoothing will
be performed.
NUMERICAL RESULTS
To demonstrate our smoothing method and to estimate the
smoothing volume around a point, we performed several nu-
merical tests using an explicit nite-difference program. As a
criterion for the accuracy of our smoothing algorithm com-
pared to other methods, we considered the traveltimes in
the original and smoothed media. An appropriate smoothing
method should preserve both the amplitude as well as the trav-
eltimes of the propagating modes. However, we restricted our
attention to the traveltimes of P-waves because they are of
major importance for ray-tracing or imaging problems.
For all numerical simulations we used the same background
medium with a P-wave velocity of 4000 m/s, an S-wave ve-
locity of 2300 m/s, and density of 2.5 g/cm
3
and added uctu-
ations of the elastic parameters. As a source we used a plane
wave (Ricker wavelet) propagating fromthe surface intodepth
through the medium. Signicant frequencies were in the range
of 35 to 150 Hz with a maximum in the amplitude spectrum at
75 Hz. This corresponded to a wavelength of about 50 m for
the P-wave. The theoretical result for our smoothing algorithm
was valid in the case of a smoothing radius that was small com-
pared to the wavelength, so an averaging radius of 30 m was
slightly beyond the validity range. However, even in this case
we found good results using our upscaling method. Figure 1
shows a realization of the random medium with an exponen-
tial correlation function and a correlation length of 10 m.
Medium 1: Fluctuations of and
The rst medium we used had uctuations of the Lam e pa-
rameters and but no density uctuations. The uctuations
of and were perfectly correlated. The perfect correlation of
uctuations provided a most serious test of equations (6)(9)
because of the complete contributions of terms with crossvari-
ances. Figure 2 shows the medium parameters at a depth of
200 m. Because of the spatial uctuations of the medium pa-
rameters, the initial plane wave becomes distorted when prop-
agating through the medium. Figure 3 shows snapshots of the
waveeld at different times. Dark and light areas indicate neg-
ative and positive displacement values in the vertical direction.
Applying the 2-D smoothing algorithm to the medium shown
in Figure 2, we obtained the smoothed medium parameters
shown in Figure 4. Note the change in axis scale in comparison
with Figure 2. From the propagation of the elastic wave in the
smoothed medium, as shown in Figure 5, it is obvious that
the scattered part of the waveeld is much smalleran effect
causedby the reducedscattering at small-scale uctuations. We
can also see the numerical effect of the absorbing boundaries,
which is unimportant for further considerations.
Now we had to nd out whether the waveeld in the
smoothed medium gives better results for traveltimes than a
mediumsmoothedbyanalternativemethod. For this reasonwe
applied several other smoothing methods and performed the
same nite-difference modeling as shown in Figures 3 and 5. To
compare the seismic records, we stacked them along the hor-
izontal axis, resulting in a single seismogram. Figure 6 shows
the stacked seismic sections obtained for the original and the
smoothed media.
We see very good agreement between the traveltimes of
the unsmoothed medium and the medium smoothed by using
equations (8) and (9). In the case of medium 1, the smooth-
FIG. 3. Waveeld at times of 25 ms (left), 75 ms (center), and 125 ms (right), showing increasing waveeld distortion.
FIG. 4. Parameters of medium 1 after smoothing using the algorithm according to equations (8) and (9). Note the signicant
reduction of uctuations.
FIG. 5. Waveeldat times of 25 ms (left), 75 ms (center), and125 ms (right) inthe smoothedmedium. The initial waveeldparameters
are identical to those in Figure 3.
ing method using equations (8) and (9) resulted in a medium
where the wave propagates slightly faster than in the medium
obtained by averaging the slowness. This can be veried in Fig-
ure 6.
Medium 2: Density uctuations
Adifferent case is that of density uctuations (Lam e param-
eters are constant), as shown in Figure 7. The upscaling method
is then equivalent to the squared slowness averaging, as seen
in Figure 8.
Medium 3: Correlated density and stiffness uctuations
The previous examples showed the advantages of our ap-
proach. To demonstrate its limitations, we developed the fol-
lowing model. The medium uctuations were such that the ve-
locities were constant in the whole medium(4000 and 2300 m/s
for P- and S-wave velocity). Since density and stiffness values
1842 Gold et al.
uctuate, the effective medium had a slightly different velocity
than the velocities at every point.
Using the same smoothing radius as before, we found no
good agreement of the waveeld in the original and smoothed
media (see Figure 9). The reasonfor this effect lies inthe higher
order terms that were neglected in the zero-frequency limit of
the Bourret approximation [see equations (A-19A-20) and
(A-28A-29)]. Since those terms depended on the crosscor-
relation functions between density and stiffness parameters,
they were irrelevant for the previous models. For the modeling
results depicted in Figure 11, we used a smoothing radius of
8 m, being about 1/2 of the wavelength. In geological me-
dia we normally observe some correlation between stiffness
and density uctuations; fortunately, in most cases the density
uctuations are small compared with the stiffness uctuations.
FIG. 6. (left) Stacked sections of the waveeld transmitted in the original (dashed line) and smoothed by the upscaling method
(solid line) media. Good agreement can be observed for the traveltimes. The difference in amplitude is caused by scattering from
small spatial-scale inhomogeneities. (right) Results of different smoothing methods: (1) the stacked section for averaging of the
inverse squared velocity compared with the stacked section of the unsmoothed medium (dashed line), (2) the stacked section for
inverse velocity averaging, (3) upscaling method, (4) velocity averaging, and (5) squared velocity averaging, which corresponds to
stiffness averaging.
FIG. 7. A medium with density uctuations. The standard deviation of the density is 40%. This corresponds to velocity uctuations
of about 20%.
FIG. 8. The same as Figure 6, but for medium 2.
Moreover, the correlation between these parameters is never
perfect. Thus, we expected the upscaling method to yield better
results than other methods for a smoothing radius smaller than
1/2 of the wavelength. For decreasing smoothing radius, the
results improved signicantly, as shown in Figures 10 and 11.
The favorable effect of a small averaging radius compensated
for neglecting higher order terms with densitystiffness corre-
lations.
Medium 4: Uncorrelated density and stiffness uctuations
To support the hypothesis that the differences in model 3 be-
tween wave propagation in smoothed and unsmoothed media
are because of perfect correlation between density and stiff-
ness parameters, we used the same uctuations of medium
parameters as in model 3 but with uncorrelated density and
stiffness uctuations. Figure 12 shows the medium parameters.
For the waveeld we obtained very good agreement between
the original and smoothed media using the upscaling method
(see Figure 13). This suggestedthat the upscaling methodcould
apply for smoothing areas smaller than half of the wavelength,
provided that density and stiffness uctuations were weakly
correlated. If uctuations were highly correlated, the upscal-
ing method had to be applied using smaller smoothing areas,
as seen for model 3.
Marmousi model
We then applied the upscaling method to the 2-D Marmousi
model. In the following sections the smoothing effect on the
FIG. 9. The results for the correlated density uctuation model with 40% standard deviation of the medium parameters. (left)
Comparison between the stacked sections of the original medium and the medium smoothed by the upscaling method (dashed and
solid line, respectively). Smoothing was performed in an area with a radius of 30 m. (right) Stacked sections for the original medium
(dashed), (1) a homogeneous medium where the P-wave velocity is 4000 m/s (smoothed squared slowness), (2) upscaling method,
and (3) the section for a medium where the velocity was smoothed (all solid lines). In the latter case because of constant velocity,
the velocity averaging yields a homogeneous medium with V
p
=4000 m/s.
FIG. 10. The results for the correlated density uctuation model with a smoothing radius of 15 m. The same delineation of curves
as in Figure 9 is chosen. The difference between the sections of the smoothed and original media becomes signicantly smaller.
FIG. 11. The results for the correlated density uctuation model with a smoothing radius of 8 m. The same delineation of curves as
in Figure 9 is chosen. Compared with the results of velocity smoothing, the upscaling method now yields the best result.
waveeld is observed on individual traces rather than on the
mean eld (i.e., stacked traces). For the modeling procedure,
we used the central part of the model shown in Figure 14.
The sketch of the stiffness parameters and the density on the
right side shows that these parameters are highly correlated,
but stiffness uctuations are much larger than density vari-
ations. Therefore, with a smoothing length of wavelength/3,
we expected the upscaling method to yield good results. The
seismograms for the three subsurface locations indicated in
Figure 14 are displayed in Figures 1517.
The outcome of the experiment with respect to the best
smoothing method strongly depends on the recording position.
This is because the assumption of a random medium with an
isotropic correlation function is not valid for the Marmousi
1844 Gold et al.
model. If we consider a wave propagating from the source to
the receiver at position 1, areas with constant medium param-
eters are approximately oriented along the propagation
direction, so we expect a behavior of the waveeld tending
toward isostrain rather than toward an isotropic random
medium. On the other hand we expect an isostress behavior
for receiver position 3 since the medium changes signicantly
in the propagation direction. These two effects are seen in
FIG. 12. A medium with uncorrelated density and stiffness uctuations.
FIG. 13. Stacked seismic sections for a medium with uncorrelated density and stiffness uctuations. (left) Comparison between the
stacked sections of the original medium and the medium smoothed by the upscaling method (dashed and solid lines, respectively).
(right) As a result of the smoothing method, (1) the stacked section for averaging the inverse squared velocity compared with
the stacked section of the unsmoothed medium (dashed line), (2) the stacked section for inverse velocity averaging, (3) upscaling
method, (4) velocity averaging, and (5) squared velocity averaging.
FIG. 14. (left) Central part of the Marmousi model, showing the stiffness tensor element M. The crosses indicate the source position
(S) and receiver positions 1, 2, and 3. (right) A vertical section through the model at 600 m. Note the different axes labels.
(1) Figure 15, where the velocity in the medium smoothed by
our algorithm is smaller than the velocity in the unsmoothed
medium, and (2) Figure 17, where we see a tendency to the
opposite effect. This comparison of smoothing methods shows
that for some positions one can nd different methods which
produce the optimum results. However, our approach always
produces acceptable results and never shows the maximum
error compared to other methods.
SMOOTHING FOR HIGH-FREQUENCY MODELING
In this section we show the performance of the upscaling
method after applying a subsequent high-frequency wavefront
construction method. As an example, we initially chose two
laterally homogeneous blocks with overall dimensions of 500
by 500 m. In the upper block, the P-wave velocity increased
from 3600 m/s at the surface up to 4000 m/s at a depth of 400 m
with a constant velocity gradient. The S-wave velocity equally
increased from 2052 to 2280 m/s. At the same time, and ,
which were perfectly correlated, increased linearly with depth.
FIG. 15. Comparison between seismic records for original (dashed line) and smoothed medium(solid line) at a depth of 600 m. From
left to right we see the seismic records in the medium smoothed with the upscaling method, inverse squared velocity averaging, and
inverse velocity averaging. The seismograms are recorded at position 1.
FIG. 16. Equivalent to Figure 15, but at location 2.
FIG. 17. Equivalent to Figure 15, but at location 3.
In the lower block of 100 m thickness, velocities remained con-
stant at 4000 and 2280 m/s, respectively. Random uctuations
of and with a standard deviation of 20% were superim-
posed on the entire model, resulting in a standard deviation of
the P-wave velocity of about 10% (see Figure 18). The density
remained constant everywhere in the medium at 2.5 g/cm
3
.
We centered a point source at the surface of the model where
the model was smoothed by different methods with a smooth-
ing radius of 20 m. To observe wave propagation of the rst
arrivals, we used both nite-difference modeling to solve the
elastodynamic wave equation and a stabilized nite-difference
1846 Gold et al.
eikonal solver (Pasasa et al., 1998; Zhao et al., 1998). Fig-
ure 19 displays a windowed snapshot of the rst arrivals, cal-
culated by the nite-difference eikonal algorithm for a variety
of smoothed media as well as for the original medium.
The rst arrivals calculated by the nite-difference eikonal
solver in the original medium are far earlier than in the
FIG. 18. Vertical section (well log) of the medium with a veloc-
ity gradient superposed by statistical variations. The gradient
in the upper block and the constant velocity in the lower part
are delineated by straight lines.
FIG. 19. Windowed snapshot of rst-arrival isochrones in me-
dia with different smoothing schemes, calculated by a nite-
difference eikonal solver, after 120 ms. The propagation di-
rection is from top to bottom, originating from a point source
located at a depth of 3 m and centered horizontal position
(250 m). The wavefronts are delineated according to different
smoothing algorithms: averaged inverse squared velocity (light
blue), inverse velocity (white), upscaling method (solid black),
averagedvelocity (green), andaveragedsquaredvelocity (dark
blue). Note the result for the unsmoothed medium (dashed
black) which differs signicantly from the other isochrones.
The background shows the unsmoothed P-wave velocity.
smoothed media. This can be explained by the properties of the
modeling algorithm: similar to ray tracing, the eikonal solver
seeks the quickest path for the wave to propagate. In the orig-
inal medium, maximum values that exist from statistical vari-
ation of the elastic parameters can be exploited fully by the
algorithm. These maximum values can be seen in Figure 18
and vanish during any smoothing process with a reasonably
large smoothing radius. However, the question arises: which
rst-arrival curve is most adequate? Toanswer this questionwe
compare these results with the outcome of the nite-difference
experiment that solves the elastodynamic wave equation in
the original and smoothed media. The source is located at the
same position as before, and the frequency content of the
wavelet is equal to that in previous examples. Figure 20
shows snapshots of the wavefronts after 120 ms in differ-
ent smoothed media. We again observe results similar to
those for medium 1. The upscaling method yielded a satis-
factory representation of the original case compared to other
smoothing methods in terms of traveltimes as well as am-
plitudes. The overall results were also conrmed by other
traces of the same section. Further, the waveelds in the orig-
inal and smoothed media (at least in the smoothed version
obtained by our upscaling method) were very close to each
other. Therefore, their rst arrivals must also be close to each
other. However, if we compare the rst occurrence of energy
from the nite-difference synthetics with the rst arrivals from
the nite-difference eikonal calculations, we see reasonably
good agreement, except for the nite-difference eikonal re-
sults in the original medium. Thus, the eikonal solver gave
a good estimate of the rst-arrival time when it was applied
FIG. 20. Comparison between the results obtained from ordi-
nary nite-difference modeling and from Figure 19 for a single
trace cut out at 250 m horizontal position. Dashed lines indi-
cate amplitudes in the smoothed media; the solid line repre-
sents the nonsmooth case. The smoothing methods 2, 3, and 4
best approximate the original traces. (1) The inverse squared
velocity is averaged; (2) inverse velocity; (3) result from the
upscaling method; (4) averaged velocity; (5) averaged squared
velocity. The positions of rst arrivals (right) are calculated by
a nite-difference eikonal solver applied to the corresponding
media.
tothecorrectlysmoothedmodel rather thantotheoriginal one.
The corresponding rst-arrival isochrone is the most adequate
one.
This comparison shows that, in principle, a high-frequency
approximation is incapable of correctly constructing the rst-
arrival isochrone for complex unsmoothed heterogeneous
media. An appropriate (frequency-dependent) smoothing
procedure makes it possible. Therefore, appropriate initial
smoothing is of high importance for the performance of high-
frequency modeling.
CONCLUSIONS
The analytical description of wave propagation in random
media, obtained from the Bourret approximation, provides a
quick and physically justied smoothing algorithm for elastic
isotropic media. Compared to methods that average the slow-
ness or the squared slowness, the upscaling method proposed
here is sensitive to the nature of medium uctuations. Since
the smoothing radius depends onthe wavelength, the upscaling
method is also frequency dependent. Using synthetic data cal-
culated by nite-difference modeling, our approach gave very
good results if the smoothing radius was small enough and the
medium uctuations were isotropic. Over the whole range of
calculations, we found that this approach proved superior to
other smoothing methods, even if the assumption of isotropic
uctuations and low frequencies were not valid. Moreover,
because of its well-understood physical background, the up-
scaling method should be preferred to other intuitive meth-
ods, even in the case of results of equal quality. We also
showed that smoothing signicantly improves results of high-
frequency modeling. Finally, the upscaling formulas derived
here [equations (6)(9)] can be applied in more general situa-
tions thanjust for smoothing. For example, they couldbe useful
for blocking 2-D or 3-D statistically isotropic heterogeneous
models.
ACKNOWLEDGMENTS
Critical and constructive comments of Jerry Schuster, Mike
Schoenberg, and two anonymous reviewers greatly improved
the manuscript. We thank Linus Pasasa for providing us with
the nite-difference eikonal solver code and Stefan Buske
for proofreading. Being a part of the collaborative research
project SFB 381, we gratefully acknowledge the support of the
Deutsche Forschungs-gemeinschaft (DFG). In part, this work
has beenalsosupportedbythesponsors of theWIT-consortium
project.
REFERENCES
Bensoussan, A., Lions, J. L., and Papanicolaou, G., 1978, Asymptotic
analysis for periodic structures: North Holland Publ. Co.
Cerveny, V., Molotkov, I. A., and Psencik, I., 1977, Ray method in
seismology: Univ. Karlova.
Christensen, R. M., 1979, Mechanics of composite materials: John
Wiley & Sons, Inc.
Grubb, H. J., and Walden, A. T., 1995, Smoothing seismically derived
velocities: Geophys. Prosp., 43, 10611082.
Gubernatis, J. E., Domany, E., and Krumhansl, J. A., 1977, Formal as-
pects of the theory of the scattering of ultrasound by aws in elastic
materials: J. Appl. Phys., 48, 28042811.
Katsube, N., 1995, Estimation of effective elastic moduli for compos-
ites: Internat. J. Solids Structures, 32, 7988.
Keller, J. B., 1977, Effective behaviour of heterogeneous media, in
Landman, B., Ed., Statistical mechanics and statistical methods in
theory and application: Plenum Press, 631644.
Pasasa, L., Wenzel, F., and Zhao, P., 1998, Prestack high-resolution
imaging with non-Fermat traveltimes: 68th Ann. Internat. Mtg., Soc.
Rytov, S. M., Kravtsov, Y. A., and Tatarskii, V. J., 1987, Principles of
statistical radiophysics: Springer-Verlag New York, Inc.
Sanchez-Palencia, E., 1980, Non-homogeneous media and vibration
theory: Springer-Verlag New York, Inc.
Shapiro, S. A., Schwarz, R., and Gold, N., 1996, The effect of random
isotropic inhomogeneities on the phase velocity of seismic waves:
Geophys. J. Internat., 127, 783794.
Zhao, P., Uren, N. F., Wenzel, F., and Hatherly, P. J., 1998,
Kirchhoff diffraction mapping in media with large velocity contrasts:
APPENDIX A
EFFECTIVE WAVENUMBERS
The following calculation is analogous to the treatment by
Rytov et al. (1987), which is done for the case of acoustic ran-
dom media. Instead of the Helmholtz equation, we use the
elastic wave equation
(r)
2
u
i
(r) +(C
i j kl
(r)u
k,l
(r))
, j
= 0. (A-1)
Throughout this paper we use standard tensor index conven-
tion. Indices that occur twice imply summation; the subscript
,i denotes differentiation with respect to the i th coordinate.
In the following calculation we assume that the medium is
isotropic and that the uctuations of the medium parameters
are small compared with their mean values.
Equation (A-1) can be written in the following operator
form:
L
i k
(r)u
k
(r) = 0, (A-2)
with
L
i k
(r) (r)
2
i k
+
j
C
i j kl
(r)
l
,
=
o
i k
+
j
C
o
i j kl
l
_
=L
o
i k
_
+(r)
2
i k
+
j
C
i j kl
(r)
l
( =L
i k
(r)).
Here, we have separated the operator into a mean part, con-
taining the mean values of the elastic parameters C
i j kl
and ,
and a uctuating part.
A Greens function approach to the waveeld gives the fol-
lowing equation:
u
i
(r) = u
o
i
(r) +
_
d
3
(r
)G
o
i j
(r r
)L
j k
(r
)u
k
(r
), (A-3)
according to Gubernatis et al. (1977). Here, G
o
stands for the
Greens function of a homogeneous medium, where C
i j kl
=0
and =0.
1848 Gold et al.
Substituting G
o
for u
o
in equation (A-3) and expanding this
equation, we nd the following Born expansion of the Greens
function in the medium with uctuating parameters:
G
i m
(r, r
o
) = G
o
i m
(r, r
o
)
+
_
d
3
r
G
o
i j
(r, r
)L
j k
(r
)G
o
km
(r
, r
o
)
+
_ _
d
3
r
d
3
r
G
o
i j
(r, r
)L
j k
(r
)G
o
kl
(r
, r
)L
ln
(r
)G
o
nm
(r
, r
o
) +
_ _ _
.
(A-4)
To nd the effective (i.e., averaged) Greens function, we aver-
age equation (A-4). If we take into account that the mean value
of L
i k
equals 0andif weassumethat therandommediumis sta-
tistically homogeneous, then we nd for the effective Greens
function
G
i m
(r r
o
) = G
o
i m
(r r
o
) +
_ _
d
3
r
d
3
r
G
o
i j
(r r
)
L
j k
(r
)G
o
kl
(r
)L
ln
(r
)
G
o
nm
(r
r
o
) +
_ _ _
. (A-5)
Rytov et al. (1987) apply the Feynmantechnique toregroupthe
various scattering terms. The correspondence between the rep-
resentation in diagrams and integral expressions is exactly one
to one, so the result of reordering the Feynman diagrams can
be directly translated back to analytic formulas. In other words,
this treatment is totally independent of the exact formulas for
the Greens function and the mediumuctuation terms. So it is
possible to use the diagram result from Rytov et al. for elastic
random media. Translated back to integral equations, we have
the following expression for the effective Greens function:
G
i m
(r r
o
) = G
o
i m
(r r
o
) +
_ _
d
3
r
d
3
r
G
o
i j
(r r
)Q
j k
(r
)

G
km
(r
r
o
). (A-6)
This equation is called the Dyson equation. The quantity Q
j k
is called the kernel of mass operator:
Q
i k
(r
) =
_
L
i j
(r
)G
o
jl
(r
)L
j k
(r
)
+
_
d
3
r
L
i j
(r
)G
o
jl
(r
)L
lm
(r
)G
o
mn
(r
)L
mk
(r
)
+
_ _

_
. (A-7)
The problem is now to nd a solution of equation (A-6). Since
this equation contains two convolutions, it is easier to solve its
Fourier transform. We obtain
g
i m
() = g
o
i m
() +(8
3
)
2
g
o
i j
()q
j k
() g
km
(). (A-8)
Here, g
o
, g, and q are the Fourier transforms of G
o
,

G, and Q,
respectively. Therefore, we have
g
i m
() = W
1
i k
g
o
km
(), (A-9)
where W
1
i k
is the i kth element of the matrix W
1
, and
W
i k
=
_
i k
(8
3
)
2
g
o
il
()q
lk
()
_
. (A-10)
The Greens function of a homogeneous isotropic medium is
known:
G
o
i k
(r r
) =
1
4
2
_
_
i k
+
i
k
_
e
ir
r

i
k
r
i r
r
_
.
(A-11)
Here, and are the wavenumbers of the compression and
shear waves in the background medium. The Fourier transform
of this function is
g
o
i k
=
1
2
1
8
3
_
i k

i
2
+

i
2
_
. (A-12)
Therefore, it seems to be obvious to use the ansatz
g
i k
=
1
2
1
8
3
_
2
e
i k

i
2
e

2
+

i
2
e

2
_
(A-13)
for the Fourier transform of the effective Greens function.
Here,
e
and
e
denote the wavenumbers of the effective
Greens function and
e
is the effective density (recall that
e
=
o
). Inserting equations (A-12) and (A-13) into equation
(A-9) yields a lengthy equation containing the unknown quan-
tities
e
and
e
. This equation should be valid for all values
of if the Fourier transform of the effective Greens function
has exactly the form as in equation (A-13). Since this is not
necessarily the case, we solve this equation for those values
of where the Fourier-transformed effective and background-
medium Greens functions have the largest amplitudes or sin-
gularities, respectively. In this way, using = and = in
equation (A-9), we nd solutions for the effective wavenum-
bers of P- and S-waves. With chosen in the z-direction, which
is reasonable for a homogeneous isotropic medium, they read
e
=
_
1 +
8
3
2
q
33
()
_
1/2
,

_
1
4
3
2
q
33
()
_
; (A-14)
e
=
_
1 +
4
3
2
(q
11
() +q
22
())
_
1/2
,

_
1
2
3
2
(q
11
() +q
22
())
_
. (A-15)
This approximation is valid even in the low-frequency limit
becausethedependenceof equations (A-14) and(A-15) on
2
vanishes as a consequence of equation (A-17) and frequency
dependencies of terms in equations (A-19) and (A-20).
The Bourret approximation
Equation (A-7) shows an expansion of the quantity Q
i j
in
terms of the medium uctuations L
i j
and the Greens func-
tionof thehomogeneous backgroundmedium. Assumingsmall
uctuations of the medium parameters, the higher order (in
respect to L
i j
) terms of equation (A-7) can be neglected.
Therefore, we obtain
Q
i k
(r
) =
_
L
i j
(r
)G
o
jl
(r
)L
j k
(r
)
_
. (A-16a)
Operator L
i j
(r
) is shown in equation (A-2). Assuming an

isotropic elastic medium, we obtain
C
i j kl
=
i j
kl
+
i k
jl
+
il
j k
. (A-16b)
Thus, combining equation (A-16b) with equation (A-2), we
obtain
Q
i k
(r
) =
__
(r
)
2
i j
+
i
(r
)
j
+
j
(r
)
i
+
m
i j
(r
)
m
_
G
o
jl
(r
_
(r
)
2
lk
+
l
(r
)
k
+
k
(r
)
l
+
n
lk
(r
)
n
__
, (A-17)
where =
o
, =
o
, and =
o
. We can see that Q

i k
consists of a sum of 16 terms. The uctuations of the medium
parameters at positions r
and r
appear in the form of spatial

correlation functions and crosscorrelation functions. We use
the following denition of a spatial correlation function:
B
xz
(r)
1
2
(x(r +r)z(r) +z(r +r)x(r)),
(A-18)
where x and z are arbitrary medium parameters.
Using equations (A-14) and (A-15) combined with the spa-
tial Fourier transform of equation (A-17), we obtain the fol-
lowing result:
e
= +

2
o
2
_
d
3
r e
i z
4
B
(r)G
o
33
(r)
2
B
(r)G
o
jl, jl
(r) 4
2
B
(r)G
o
33,33
(r)
+4i
2
B
(r)G
o
33,3
(r) +2i
2
B
(r)G
o
m3,m
(r)
4
2
B
G
o
3m,3m
(r)
_
, (A-19)
e
= +

4
o
2
_
d
3
r e
iz
4
B
(r)
_
G
o
mm
(r) G
o
33
(r)
_
4
2
B
(r)
_
G
o
mm,33
(r) G
o
33,33
(r)
_
+4i
2
B
(r)
_
G
o
mm,3
(r) G
o
33,3
(r)
__
. (A-20)
Weusedpartial integrations toshift thespatial derivations from
medium parameters to Greens functions. We also assumed a
statistical isotropic medium. Therefore, the correlation func-
tions depend only on r =|r|.
Low-frequency limit
To nd the low-frequency limit of the effective wavenum-
bers, we calculated the corresponding limit of expressions like
G
o
33,33
using the spatial Fourier transform of the Greens func-
tion:
g
o
i k
=
1
2
1
8
3
_
i k

i
2
+

i
2
_
. (A-21)
To compute the Fourier transform of spatial derivatives of the
Greens function, we used the well-known property
A(r) a(),
i
A(r) i
i
a().
As an example let us consider the term related to density uc-
tuations. The following equation will be helpful for this:
g
o
33
=
1
2
1
8
3
_
2
3
2
+

2
3
2
_
,
=
1
2
1
8
3
_
2
+
2
_
,
=
1
2
1
8
3
_
(1

2
)

2
2
+

2

2
2
_
.
(A-22)
Here,

denotes the cosine of the angle between the direction
of the z-axes and the direction of the vector , cos(). In the
last step, we used
2
=

2
+

2
2
. (A-23)
To calculate G
o
33
and to get the lowest order terms in the low-
frequency limit, we can make an estimation of the order of G
o
33
in terms of wavenumbers and by neglecting the angular
behavior of g
o
33
. Then we obtain
G
o
33
= O
_

2
4
o
2
e
i r
r
_
+ O
_

2
4
o
2
e
ir
r
_
.
An explicit computation of G
o
33
using the inverse Fourier
transform leads to the same result. From the equation above
and equation (A-19), we see that the contribution of density
uctuations to =
e
behaves like
3
for low frequen-
cies. Consequently, in the low-frequency limit this term has no
inuence on the effective wavenumber, which is on the order
of .
For the other derivatives of the Greens function in equa-
tion (A-19), a treatment analogous to the calculation in equa-
tion (A-22) can be performed. We nd, among other terms,
g
o
33,33
=
1
8
3
2
_
2
(
4
)

2

4
2
+
2
4

4
2
_
, (A-24)
g
o
jl, jl
=
1
8
3
2
_
2
_
, (A-25)
1850 Gold et al.
g
o
3 j, j 3
=
1
8
3
2
_
2

4
2
_
. (A-26)
In these equations the underlined terms provide the lowest
order contributions to the effective wavenumber. In respect
to , they show only angular dependency. We expect that by
the inverse Fourier transformthey behave like -functions (see
the explanation below). For these terms is proportional to
(i.e., to ). Therefore, their contributions are signicant. It
is not necessary to calculate the Fourier transform of

n
ex-
plicitly. Indeed, we only have to nd the low-frequency limits
of the integrals in equations (A-19) and (A-20). The follow-
ing calculation shows how to do this. First, we assume that
n is an arbitrary even number. We then consider an arbi-
trary integral term in equation (A-19). Taking into account
the form of the lowest order (in respect to ) contributions
in equations (A-24)(A-26) and that in the same approxima-
tion the exponential factor under the integral (A-19) is equal
to 1, one can see that such an integral term can be written as
follows:
_
d
3
rX(r)
_
d
3
e
i r
n
=
_
d
3
n
_
d
3
rX(r) e
i r
,
= 8
3
_
d
3
X()
_
d

n
_
d

,
=
8
3
n +1
_
d
3

X() e
i 0
,
=
8
3
n +1
X(0), (A-27)
where X(r) is a function of the absolute value of the radius vec-
tor r. The third line in this derivation was obtained by chang-
ing the coordinates of the spatial inverse Fourier transform to
spherical ones and by taking into account that the spectrum
X() is independent of

. We see that the Fourier transform
of

n
(n is even) behaves like a -function when applied to a
function without angular dependence. Performing such calcu-
lations for all terms in equations (A-19) and (A-20), we nd
the following nal equations for the effective wavenumbers:
e
=
_
1 +
1
2
2
o
(
o
+2
o
)
2
+
2
3
o
(
o
+2
o
)
2
+
2
5
2
o
(
o
+2
o
)
2
+
4
15
_
, (A-28)
e
=
_
1 +
1
5
+
2
15
_
. (A-29)
In the low-frequency limit, since the change in wavenum-
ber has no imaginary part, the effect of the medium uctu-
ations results only in a frequency-independent velocity shift.
The results above coincide with equations (3.17) and (2.23)
from Christensen (1979) and equations (40) and (43) from
Katsube (1995), calculated for the case of a low concentration
of randomly distributed spherical inclusions in a background
medium.
An analogous calculation can be made for the 2-Dcase. One
can also calculate the 2-D case from the 3-D case under the
assumption that the waveeld and the medium parameters are
constant along one axis. Under this assumption we nd
e
=
_
1 +
1
2
2
o
(
o
+2
o
)
2
+

o
o
(
o
+2
o
)
2
+
3
4
2
o
(
o
+2
o
)
2
+
1
4
_
, (A-30)
e
=
_
1 +
1
4
+
1
4
_
. (A-31)
Therefore, by applying both the weak uctuations and the low-
frequency approximations, we arrive at formulas for effective
wavenumbers of P- and S-waves in statistically isotropic media
with heterogeneities of an arbitrary geometry. In our deriva-
tions, by interchanging the order of these two small-parameter
expansions, we have implicitly assumed they both are homo-
geneously converging.
GEOPHYSICS, VOL. 65, NO. 6 (NOVEMBER-DECEMBER 2000); P. 1851
Special Section on mining geophysicsIntroduction
Michael Asten
, Guest Editor
The Exploration97 Conference heldinTorontoin1997 was a
decennial event whichprovideda forumfor reviewof advances
ingeophysics appliedtomineral explorationandresource eval-
uation. This Special Section of GEOPHYSICS publishes a set of
16 papers, originally presented at that conference, which
demonstrate recent advances in mineral exploration and mine
geophysics.
One of the most dramatic changes in our profession, illus-
tratedbyveof thesepapers, is thesuccessful cross-fertilization
between petroleum and mining geophysics, whereby advances
in seismic methods, in particular 3-D data acquisition and
processing, have been brought into routine application in
the structurally-complex environments associated with min-
eral provinces and ore-deposit geometry. A further three pa-
pers describe developments in underground or crosshole seis-
mic, radar, and radio-wave imaging methods. Just as these
examples have extended traditional petroleum technolo-
gies into mineral-related problem solving, it is likely that the
high-resolution techniques and the integration of electromag-
netic methods with seismic methods will prove of benet to
petroleum geophysicists.
Papers in this Special Section are reprinted from Geophysics and Geochemistry at the Millennium, Proceedings of Exploration 97, the Fourth
Dicennial International Conference on Mineral Exploration. Used with permission.
Monash University, Department of Earth Sciences, Clayton, Victoria 3168, Australia. E-mail: masten@mail.earth.monash.edu.au.
Four papers reviewthe state of the art in acquiring and imag-
ing airborne electromagnetic and spectrometric gamma-ray
data. The quality of the data and images now achievable show
why these techniques are becoming essential tools for geologic
mapping and reconnaissance, as well as for their traditional
role of direct ore-body target location.
One paper on 3-D inversion of induced-polarization data
introduces a newadvance (from2-Dinto 3-Dgeometry) in one
of the most signicant breakthroughs of the last generation in
surface mineral geophysics: the stable inversion of resistivity-
IP data.
Finally, three papers discuss the application of geophysics
and borehole logging to mine planning and ore-body evalua-
tion. Acceptance of geophysical technology in these areas has
been slow, and it is notable that these case histories are able to
demonstrate how geophysics can bring signicant savings on
the order of millions of dollars tothe eldof mineral extraction.
I thank former GEOPHYSICS Editor Sven Treitel for initiating
this Special Section; Laurie Reed, Peter Annan, Ian MacLeod,
and the Technical Committee of Exploration 97 for their co-
operation; and the 60 authors involved for their hard work and
patience in bringing the set of papers together.
1851
Seismic reection imaging of mineral systems: Three case histories
Barry J. Drummond
, Bruce R. Goleby
, A. J. Owen
, A. N. Yeates
,
C. Swager
, Y. Zhang
, and J. K. Jackson
ABSTRACT
Mineral deposits can be described in terms of their
mineral systems, i.e., uid source, migration pathway,
and trap. Source regions are difcult to recognize in seis-
mic images. Many orebodies lie on or adjacent to major
fault systems, suggesting that the faults acted as uid mi-
gration pathways through the crust. Large faults often
have broad internal zones of deformation fabric, which
is anisotropic. This, coupled with the metasomatic ef-
fects of uids moving along faults while they are active,
can make the faults seismically reective. For exam-
ple, major gold deposits in the Archaean Eastern Gold-
elds province of Western Australia lie in the hanging-
wall block of regional-scale faults that differ from other
nearby faults by being highly reective and penetrating
to greater depths in the lower crust. Coupled thermal,
mechanical, and uid-ow modeling supports the the-
ory that these faults were uid migration pathways from
the lower to the upper crust. Strong reections are also
recorded from two deeply penetrating faults in the Pro-
terozoic Mt. Isa province innortheasternAustralia. Both
are closely related spatially to copper and coppergold
deposits. One, the Adelheid fault, is also adjacent to the
large Mt. Isa silverleadzinc deposit. In contrast, other
deeply penetrating faults that are not intrinsically reec-
tive but are mapped in the seismic section on the basis
of truncating reections have no known mineralization.
Regional seismic proles can therefore be applied in the
precompetitive area selection stage of exploration. Ap-
plying seismic techniques at the orebody scale can be
difcult. Orebodies often have complex shapes and re-
ecting surfaces that are small compared to the diame-
ter of the Fresnel zone for practical seismic frequencies.
However, if the structures and alteration haloes around
the orebodies are targeted rather than the orebodies
themselves, seismic techniques may be more successful.
Strong bedding-parallel reections were observed from
the region of alteration around the Mt. Isa silverlead
zinc orebodies using high-resolution proling. In addi-
tion, a prole in Tasmania imaged an internally nonre-
ective bulge within the Que Hellyer volcanics, suggest-
ing a good location to explore for a volcanic hosted mas-
sive sulde deposit. These case studies provide a pointer
to how seismic techniques could be applied during min-
eral exploration, especially at depths greater than those
being explored with other techniques.
INTRODUCTION
Deep seismic reection programs around the world are
mostly directed toward understanding the tectonic evolution
of the regions studied and therefore have often led only indi-
rectly to an improved understanding of their mineral potential.
In contrast, a seismic transect of the Mt. Isa inlier of northeast-
ern Australia, sponsored by the Australian Geodynamics Co-
operative Research Centre, was deliberately designed to place
Manuscript received by the Editor March 22, 1999; revised manuscript received June 4, 2000.
Australian Geological Survey Org., P.O. Box 378, Canberra, A.C.T. 2601, Australia. E-mail: Bruce.Goleby@agso.gov.au; Barry.Drummond@agso.
gov.au.
Formerly Geological Survey of Western Australia, 100 Plain Street, Perth, Western Australia 6000, Australia; presently North Ltd., P.O. Box 231,
Cloverdale, Western Australia 6105, Australia.
CSIRO, Div. of Exploration and Mining, Nedlands, Western Australia 6009, Australia.
Formerly Mount Isa Mines Exploration, GPO Box 1042, Brisbane, Queensland 4001, Australia; presently Sons of Gwalia, 16 Parliament Place,
West Perth, Western Australia 6005, Australia.
major orebodies in the inlier into their regional geodynamic
framework (Drummond et al., 1997).
The results from the Mt. Isa transect, together with the
ndings of Drummond and Goleby (1993) from the Eastern
Goldelds province of Western Australia, suggest that the seis-
mic proling technique could be imaging uid migration path-
ways within the crust. Higher resolution studies (e.g., Milkereit
et al., 1996; Yeates et al., 1997; and Goleby et al., 1997) suggest
that seismic techniques can also be successful at the orebody
1852
Seismic Imaging of Mineral Systems 1853
scale, especially if they are used to image the structures and
alteration zones around the orebody rather than the orebody
itself.
In this paper, we place the seismic results from three unre-
lated and widely separated terranes of different ages, i.e., the
Eastern Goldelds province (Drummond and Goleby, 1993),
the Mt. Isa Inlier (Drummond et al., 1997) and Tasmania
(Yeates et al., 1997), into a mineral system framework. Al-
though the mineral deposits studied differ in many ways, the
approach of simplifying mineral systems provides pointers on
how seismic techniques can be effective in the precompetitive
area selection stage and in direct exploration.
MINERAL SYSTEMS
Mineral systems are analogous to petroleum systems, which
describe the genetic relationship between a petroleum source
rock and an accumulation (Magoon and Dow, 1991). Using this
denition, a mineral system can be described in its simplest
form as a uid source, a migration pathway, and a trapping
mechanism that scavenges the minerals from the uids. The
uid source could be basin brines in the case of strata-bound
deposits, or it could be more deep-seated lower crustal or even
upper mantle hydrated rocks in the case of other deposits.
Migration pathways are needed to focus the uids fromtheir
source into the trap. Many mineral deposits lie on, or adjacent
to, major fault zones, suggesting a causal relationship. Fault sys-
tems provide fracture porosity as well as a focusing mechanism.
In the case of basin brines, the general distribution of perme-
able and impermeable rocks of the basin strata also allows uid
ow and inuences its form.
Trapping mechanisms take a variety of forms and require
the superposition of physical barriers to uid ow, e.g., local
structure, stratigraphy, and permeability, whether intrinsic or
fracture induced, with the appropriate chemical, thermal, and
probably palaeogeographic settings for the minerals to be de-
posited. Hence, studies that describe mineral deposits rather
than mineral systems and that focus mainly on the trapping
mechanisms tend to describe the unique and often complex
combinations of elements in the trapping mechanism for each
deposit but do not see the underlying unifying elements of the
mineral system.
Mineral systems are usually triggered by a thermal pulse,
which in turn can often be related to intraplate tectonics re-
sulting from interplate activity (Loutit et al., 1994). Whereas
petroleum systems are usually characterized according to the
age andtype of source rock(the uidsource) (Bradshaw, 1993),
the ages of mineral deposits are often less certain. Mineral sys-
tems may be characterized according to the age of the host
rocks or theageof thethermal event that triggeredthem. Just as
a sedimentary basin can have several superimposed petroleum
systems reecting the maturing through time of a number of
stacked source rocks and their associated uid pathways and
traps, so also can a mineral province be host to several mineral
systems.
Identifying uid source regions in seismic images may be dif-
cult. The dehydration of a large area of crust to create miner-
alizing uids will not necessarily leave an observable physical
imprint on the rocks that distinguishes that region from any
other region, especially in metamorphic rocks. This is because
the physical effects of dehydration may be similar to those of
metamorphism (higher densities and seismic velocities).
Large volumes of rock can be effectively dehydrated over
time by relatively low uid ux rates. But if the uids are
concentrated into fracture-induced permeability zones along
faults, higher ux rates will occur along the faults. This can
lead to wide alteration haloes along faults and metasomatism
within the fault zone. Where the fault zone is the focus of high
strain, mylonite zones develop. They characteristically have a
well-developed anisotropic fabric (e.g., Jones and Nur, 1984;
Siegesmund and Kern, 1990). Mylonite zones can be good re-
ectors (Jones and Nur, 1984; Goodwin and Thompson, 1988).
The seismic reectivity results from the constructive interfer-
ence of reections from the bands of altered and strained
anisotropic rock within the mylonite zones. The case histo-
ries presented in this paper describe the seismic effect of the
mineral system in terms of migration pathway and trapping
mechanism.
In the rst case history, the crustal structure of the East-
ern Goldelds is interpreted in terms of linked uid pathways.
Drummond and Goleby (1993) interpret some elements of
crustal reectivityintheArchaeanEasternGoldelds province
of Western Australia in terms of uid pathways through the
crust based on the geometry of the fault zones and their spatial
relationship to known mineralization. They make no attempt
to link this regional study with local studies at the orebody
scale.
The second case history, at Mt. Isa, tries to link from the
regional scale into the orebody scale. Salisbury et al. (1996)
demonstrat that many of the sulde minerals that typically
make upthe bulkof mineralizationconstituting orebodies have
seismic velocities similar to felsic and mac igneous rocks, and
also some sedimentary rocks, but much higher densities. Pyrite
has both higher density and higher seismic velocity. Therefore,
in many cases the orebody should have a signicant impedance
contrast with country rock of most compositions. However,
orebodies can have very complex shapes and often lie in com-
plexly folded or deformed host rocks. Orebody reections may
be lost among the reections and interference signals from the
surrounding host material. Orebodies are often very small in
size compared to the wavelength of seismic energy returned
from the earth. Therefore, to maximize the chance of success
in using seismic methods at the trap or orebody part of the
mineral system, we recommend targeting the controlling struc-
tures aroundthetrapandperhaps thebroader alterationhaloes
around the orebodies.
In the Mt. Isa study, the strategy was not to try to image the
orebody itself but rather to target a known larger alteration
zone characteristic of the environment where mineralization
might occurthe uid migration pathways adjacent to the ore-
body and the structure of the trap. This is analogous to the ap-
proach used in petroleum exploration, where the structure of
the reservoir would be the seismic target rather than the pool
of oil it might contain.
The third case history, in Tasmania, studies a totally different
style of ore environment and demonstrates that by targeting
the mineral system rather than the orebody, seismic methods
can be applied successfully in a range of environments.
SEISMIC METHODOLOGY CONSIDERATIONS
Typically, the geology and structure of the three mineralized
regions described in these case histories is complex, with a
range of lithologies subjected to at least three deformational
1854 Drummond et al.
and/or metamorphic events prior to the mineralizing event.
However, in all cases the available geological control is good,
with information from mining in the region, deep drillholes,
anddetailedsurface geological mapping. Two-dimensional and
some low-fold 3-D seismic reection as well as 2-D and 3-D
crustal-scale refraction techniques were used. However, only
seismic reection results are presented here.
The case studies usedregional transects that were focusedon
structurewithinthemiddletoupper crust andhigher resolution
seismic surveys of mine-scale structures.
Explosive charges in deep drillholes provided the energy
sources. The seismic data were collected with 96 or 120 chan-
nels, and quality control was primarily through eld monitors
and in-eld data processing to at least brute stack stage, espe-
cially for the high-resolution data. Typically, the station spac-
ing was 40 m for the regional surveys and 1020 m for the
higher resolution surveys. Shot-hole spacings were variable,
but a nominal stacking fold of between 12 and 24 was achieved.
Symmetrical split-spreadgeometries were used, whichresulted
in a maximum shotreceiver offset of 2400 m for the regional
surveys.
In this type of project, the main data processing problems
result from difcult static corrections and large velocity vari-
ations. Detailed refraction static analysis is required to adjust
for the effects of a highly variable regolith in most parts of
Australia, especially withthe higher frequencies neededinhigh
resolution studies. Near-surface velocity variations are high,
ranging fromaround 1000 m/s within parts of the regolith up to
7000 m/s in metamorphosed ultramac bedrock in the Eastern
Goldelds province.
The rocks are mostly highly deformed, so reector continu-
ity, although variable, is usually far shorter that that encoun-
tered within sedimentary basins. The amplitudes of reections
are often excellent, but they may not be primary reections.
We adopted a strategy of identifying regions of similar reec-
tor coherency and dip. We correlated those regions with the
surface geology or with seismic velocities from crustal-scale
refraction or tomography studies to assign rock type. Then we
use, the geometry and spatial relationships of regions of similar
reection character to infer tectonic processes. Interpretations
are usually conrmed with both qualitative and quantitative
interpretation of gravity and magnetic data, supplemented by
available geological evidence.
CASE HISTORY 1: THE EASTERN GOLDFIELDS
OF WESTERN AUSTRALIA
TheYilgarncratoninWesternAustralia(Figure1) consists of
several geological provinces. Gneissic granitoid with granitoid
plutons and greenstone supracrustal rocks are common in all
provinces. Each province can be divided into a number of ter-
ranes, each dened by the distinct stratigraphy of its volcanic
and sedimentary supracrustal rocks. The Eastern Goldelds
province is host to much of the regions known gold deposits,
most of which occur in the west of the province. The Ida fault
separates it from the Southern Cross province to the west.
A regional-scale, 213-km-long seismic reection traverse
was positioned eastwest across the regional strike (Figure 1)
(Goleby et al., 1993). The interpretation of the shallow part of
the seismic data is given by Swager et al. (1997). The green-
stone supracrustal rocks lie above a subhorizontal detachment
between 1.5 and 2.5 s (4.5 and 7.5 km) and therefore have
a tectonic boundary with the underlying, presumably felsic
gneissic basement (Figure 2). Many of the faults in the green-
stones, e.g., the Zuleika shear (Figure 2), are not reective and
are interpreted by their truncations of greenstone stratigra-
phy. These faults can be mapped laterally over considerable
distances within the greenstones, but they are not deeply pen-
etrating and sole on the detachment surface.
However, several faults penetrate the detachment. These
faults are often reective. Within the seismic section, the Ida
fault and the Bardoc shear are the prominent examples (Fig-
ure 2). The Ida fault dips approximately 30
to the east and

extends to 2530 km depth. The Bardoc shear dips west, pen-
etrates the detachment surface, and truncates against the Ida
fault at about 15 km depth. Bottomhole cuttings from the shot
holes along the traverse were chemically analyzed. Those from
near the Ida fault andBardoc shear have comparable alteration
FIG. 1. Major structural subdivisions of the southern Eastern
Goldelds province, Yilgarnblock, WesternAustralia. Position
of 1991 seismic transect is also shown (modied from Swager
and Grifn, 1990).
patterns, indicating that similar or the same uids moved along
both of the faults in the past. This supports the seismic obser-
vation that the faults are probably linked at depth (Goleby
et al., 1993).
Many of the faults in the region, including those that do not
penetrate the detachment, can be associated spatially with gold
deposits. However, theBardoc shear andits southernextension
(Boorara shear near Kalgoorlie, Lefroy fault near Kambalda)
FIG. 2. Portion of the 1991 deep seismic transect recorded within the Archaean Yilgarn block. Arrows indicate uid-ow directions
predicted by Drummond and Goleby (1993). D=diffraction, S=sill. Detailed images of the interpretation, particularly the top 2 s,
are found in Goleby et al. (1993) and Swager et al. (1997).
are associated spatially with major gold districts, including the
GoldenMile at Kalgoorlie andthe KamdaldaSt. Ives deposits.
Many of the gold deposits lie to the west of the shear, i.e., in
the hanging-wall block.
Based on near-surface uid ow patterns suggested by
Goleby et al. (1993), Drummond and Goleby (1993) proposed
that mineralizinguids migratingfromthelower crust tohigher
levels in the greenstones followed a pathrst into and along
east-dipping shear zones within the lower crust, then into
the Bardoc shear. From there they were able to percolate
into the hanging-wall block (arrows, Figure 2). Some leaked
along the detachment and into other faults within the green-
stone sequence that splay from the detachment (e.g., Zuleika
shear; Figure 2), but most concentrated in the greenstones
immediately above and to the west of the Bardoc shear.
Numerical modeling of uid ow within the Eastern
Goldelds province supports this linked uid-pathway model
(Figure 3) (Upton et al., 1997). The numerical modeling used
the crustal structure dened by the seismic proling; other
physical properties of the crust were assumed. The order of
deformational and thermal events was derived from the geo-
logical record. The modeling predicted an initial, single crustal-
scale convectioncell inwhichuids were drivenupthe Ida fault
from depth and down the Ida fault from the surface. These
mixed and owed up the Bardoc shear, resulting in a high
degree of chemical alteration and hence mineral deposition
within the upper crust, particularly within the greenstones. As
temperatures dropped, the single convection cell broke down
into smaller cells within the upper crust. These concentrated
the mineral species in the upper crust. Coupled deformation-
uid ow modeling predicted the occurrance of east-dipping
faults within the crust. These were seen in the seismic data. The
model also predicted focused uid ow into the east-dipping
shear zones.
FIG. 3. Fluid-ow modeling of the Yilgarn block. The fault geometry is derived from the seismic data. Crustal layering is implied
from reectivity patterns along the prole. The topmost layer was added to account for crust removed by erosion since the ore
deposits were formed. Arrows represent uid-ow vectors. East-dipping zones of longer vectors in the greenstones and base-
ment are predicted by the modeling and coincide with weak reections in the seismic data. LC=lower crust, UC=upper crust,
G=greenstones, C=crust removed by erosion.
CASE HISTORY 2: MT. ISA
The Proterozoic Mt. Isa inlier of northern Australia is recog-
nized for its world-class silverleadzinc and coppergold ore
deposits. The inlier consists of an Eastern fold belt, a Western
fold belt, and the central Kalkadoon block (Figure 4). A ma-
jor eastwest deep seismic traverse 255 km long was recorded
just to the south of Mt. Isa (Figure 4). The seismic reection
section shows a marked difference in the structure of the top
510 km between the Eastern fold belt and the Western fold
belt (Drummond et al., 1997; MacCready et al., 1999).
Seismic data from the Marimo region (Figure 5) combined
with detailed structural mapping show that the sediments of
the Eastern fold belt were emplaced by thin-skinned thrusting
from the east along several stacked and probably contempora-
neous subhorizontal thrusts. Further shortening then occurred
along steeper east-dipping reverse faults. They are mostly rec-
ognized in the seismic data by the offset they created on the re-
ective lowermost thrust detachment (MacCready et al., 1999)
and extend to depths of 15 to 18 km where they intersect a re-
gion of high seismic velocities, probably mac in composition
(Drummond et al., 1997).
The Marimo fault (M, Figure 5) is intrinsically reective in
the upper 6 km (2 s two-way time). Prior to the seismic survey,
no fault had been mapped in this region. Structural mapping
undertaken to support the interpretation of the seismic data
FIG. 4. Tectonic provinces of the Mt. Isa Inlier and the location of the seismic transect.
FIG. 5. Portion of migrated seismic data from the Eastern fold belt, Mt. Isa inlier, showing the earlier shallow detachment cut by
later steeper faults imaged within the Marimo region. Contours (in kilometers per second) show position of high-velocity body in
midcrust. R=reections from high-velocity body; LVL=low-velocity layer; M=Marimo fault (from Drummond et al., 1997).
found an east-dipping, 200-m-wide zone of hydrothermal alter-
ation. The reectivity of the fault is believed to result from the
broad zone of alteration. The spatial relationship of this fault
to mineral deposits in the region is seen as further evidence
that this fault acted as a uid pathway. The Mount McNamara
coppergold mine lies near the transect, and the Hampden,
Mount Dore, Selwyn, and Osborne coppergold mines lie on
a linear trend to the south (Figure 4).
Farther west, the seismic transect imaged a sequence of
folded and faulted reectors representing sequences within the
Leichhardt River fault trough of the Western fold belt. Those
of the Eastern Creek volcanics (ECV) are marked in Figure 6
(MacCready et al., 1997). The Mt. Isa, Adelheid, and Sybella
faults and a number of other minor faults form part of an anas-
tomosing fault systemthat extends for many tens of kilometers
along the western side of the Leichhardt River fault trough.
To the north of the transect, four major leadzinc(silver)
and copper deposits lie close to the Mt. Isa fault. The fault
dips west at 70
and extends into the upper to middle crust

(Figure 6). However, it is not reective. Its surface outcrop is
<1 kilometer to the east of the Adelheid fault. The data in
Figure 6 are not migrated and show that the Adelheid fault
not only has strong P-wave reections, which are interpreted
in the gure, but also S-wave reections which are unmarked
and lie between the Adelheid and Mt. Isa faults. The strongest
reections inthis pseudotrue-amplitude sectionare seenonthe
fault below the word vortex at the top of the gure. Heinrich
et al. (1989) propose that the leadzinc deposits in the region
formed from brines circulating within the Leichhardt River
fault trough. The seismic data indicate that the Adelheid fault
probably acted to focus these brines into the anastomosing
fault set near the present-day surface.
FIG. 6. Portion of seismic line from the Western fold belt, Mt. Isa inlier, showing the interpretation of MacCready et al. (1999). The
Adelheid fault is highly reective, but other faults are not. Reections between the Adelheid and Mt. Isa fault are interpreted as
S-wave reections from the Adelheid fault. ECV=Eastern Creek volcanics (from Drummond et al., 1997).
Both 2-D and 3-D high-resolution seismic data were re-
corded between the Mt. Isa and Hilton mines (12 km north
of Mt. Isa, Figure 4). They were designed to test predicted
cross-sections just north of the copper and leadzinc orebodies
(Figure 7) (Neudert and Russell, 1981). Locally, leadzinc min-
eralization tends to be in steeply west-dipping lenses parallel
to bedding within the Urquhart Shale. Copper mineralization
lies deeper. Both the leadzinc and copper mineralization lie
above the Paroo fault.
The Paroo fault and alteration haloes above the copper
deposits were the targets for the high-resolution survey; the
Urquhart Shale was expected to have strong impedance con-
trast with the underlying basement of Eastern Creek volcanics.
A 2-D data section is shown in Figure 8. Reections were
recorded from the subhorizontal part of the Paroo fault in
the east of the section; farther west it is mapped using trun-
cations of the reections from the Urquhart Shale. The data
also show unexpectedly strong, west-dipping reections from
within the Urquhart Shale; they are parallel to bedding and
correspond to predicted zones of alteration and sulde mineral
enrichment.
CASE HISTORY 3: TASMANIA
The Dundas trough and its constituent Mount Read vol-
canics in Tasmania (Figure 9) are host to a number of world-
class mineral deposits (Large, 1992). A seismic reection sur-
vey across the region in 1995 included both regional deep
and shallow high-resolution seismic proles. Drummond et al.
(1996) summarizetheresults of theregional proles. Theyshow
the Paleozoic section of the Dundas trough and Mount Read
volcanics to be a highly folded and faulted succession with a
total thickness of at least 4 km. Individual stratigraphic units
within the Dundas Group and Mount Read volcanics are gen-
erally not differentiated in the regional proles, probably be-
cause of their highly deformed nature and the low impedance
contrasts between the units.
The high-resolution data were expected to overcome these
problems. Theywereinterpretedusingdrillholecontrol (J. Silic,
A. McNeill, and S. Richardson, Aberfoyle Pty. Ltd., personal
communication, 1996) (Figure 10 and Yeates et al., 1997). Re-
ectors at about 900 and 1150 m below shotpoint 1055 are in-
terpreted as the top and base of the Que River Shale. This unit
overlies the Que-Hellyer volcanics. The base of the volcanics is
interpreted as the reector at about 1500 m. The Que-Hellyer
volcanics therefore have a noticeable bulge, similar in geome-
try to the mound-type zincleadcopper (silver, gold) Hellyer
deposit (Large, 1992) within the Que-Hellyer volcanics sev-
eral kilometers to the south. From the known lithologies in the
area, the strong reections above this zone, at 1150 m, infer
the presence of carbonates, dolerite, or massive suldes. Re-
ections within the bulge are weak, inferring a zone of strong
alteration that produced homogeneity within the volcanics in
the bulge.
DISCUSSION
The relation of fault reectivity to anisotropy within fault
zones and alteration caused by uids is both observationally
based and supported by modeling. In the Eastern Goldelds
FIG. 7. Schematic section across the Mt. Isa Valley leadzinc
and copper mineral eld, showing the geological structure
around and within the orebody (after Fallon et al., 1997).
province, some faults are reective and others are not. Those
that are reective penetrate to greater depths. The surface out-
crop of the Bardoc shear is hundreds of meters wide and shows
high strain (Swager and Grifn, 1990). It has several gold de-
posits in its hanging wall in the region of the transect. The geo-
chemical alteration signatures from shot-hole samples support
the seismic interpretation that it links in the crust with the Ida
fault. The surface outcrop of the Marimo fault in the Eastern
fold belt of the Mt. Isa Inlier has extensive hydrothermal alter-
ation and lies along strike from operating coppergold mines.
In northwestern Tasmania, the loss of reectors in the bulge of
the Que-Hellyer volcanics would infer alteration.
The physical model of shear zones of Jones and Nur (1984)
suggests that anisotropy is the primary cause of the reectivity
fromtwo crustal-scale shear zones. The model is based on mea-
surements of the physical properties of samples collected from
surface outcrops of mylonite zones. It consists of anisotropic
rock with low impedance normal to the shear zone, interlay-
ered with isotropic rock with impedance similar to the pro-
tolith on either side of the shear zone. The model required
constructive interference to create amplitudes comparable to
those observed. To do this, the layers had to be 110150 m
thick. Goodwin and Thompson (1988) also discuss reectivity
attributable to mylonite zones. Their physical models are based
on logs from a conveniently located well, and their layers are
much thinner (about 30 m). They also note tuning of the layer
thickness is important to achieve the observed amplitudes, but
that lateral variations in layer velocity and/or thickness over
distances of about 100 m are also important.
FIG. 8. Migrated high-resolution seismic data from the Mt. Isa
Valley, with interpreted positions of the Mt. Isa and Paroo
faults.
Neither of these studies compares the relative effects of
anisotropy resulting fromstrain and alteration caused by meta-
somatism. Siegesmund and Kern (1990) nd that reectivity as
a result of anisotropy cannot account for the amplitudes of
all observed lower crustal reections. Klemperer and BIRPS
(1986) study the Outer Isles thrust in Britain, and nd that
metasomatism is more important than anisotropy in causing
impedance contrasts. Anisotropy, although present, is disorga-
nized at scales of 100 m. Whereas Jones and Nur (1984) nd
that the bulk physical properties of anisotropic rock are near
those of the protolith, Jones (1986) nds that both velocity and
density increase from protolith through mylonite to ultramy-
lonite, accompanying a reduction in silicon and an increase in
iron, calcium, and magnesium.
The details of the physical properties on which these models
are baseddiffer fromstudy tostudy andfromfault tofault, indi-
cating that local factors are important. However, the principles
established by these studies are that on a regional, deep crustal
scale, fault-zone reectivity can be related to strain-induced
anisotropy and changes in bulk physical properties resulting
from alteration. These principles also apply to structures and
alteration at the mine scale.
FIG. 9. Geology of Tasmania and the locations of two deep
seismic transects. The high-resolution seismic line in Figure 10
is coincident with and just to the west of the bend in line 2 (after
Drummond et al., 1996).
However, the alteration products may be different because
they are formed at the end of the uid migration pathway. For
example, the alteration haloes around the copper orebody at
Mt. Isa contain signicant amounts of pyrite, whereas the lead
zinc orebodies at Mt. Isa lie within an altered zone containing
less pyrite. Salisbury et al. (1996) nd that massive sulde de-
posits can have impedances much higher than most common
host rocks, and Milkereit et al. (1986) conrm that massive sul-
de orebodies, if thick enough, can be good reectors. Pyrite,
in particular, has a higher density than the host rocks of the
Mt. Isa orebodies. If present in sufcient quantities in the al-
teration haloes, it will result in an impedance contrast between
the host rocks and the alteration zone. This is probably what
caused the high-amplitude reections in Figure 8 and possibly
also over the bulge in the data from northwestern Tasmania
(Figure 10).
Not all strongreections intheseismic sections arefromfault
zones. Fault zone identication often depends on whether the
zones link with surface outcrop and, in the case of reectors
which do not reach the surface, whether the structure mapped
at depth in the seismic section makes structural sense if the
reectors are or are not interpreted as faults.
We believe that reectivity of faults is one parameter that
could be used to indicate where uids have uxed from the
deeper crust. In our studies, the seismic methodology was used
to image the main structures that either control the orebody
FIG. 10. Portion of the 1995 high-resolution seismic section
recorded within the northern Mount Read VHMS district.
The data show a bulge in the Que Hellyer volcanics (QHV)
and weaker reections in the bulge, suggesting alteration.
High-amplitude reectors above the bulge could represent
dense rocksperhaps carbonate, dolerite, or massive suldes.
QRS =Que River Shale; BA=Bouguer anomaly gravity val-
ues in micrometers per second squared (after Yeates et al.,
1997).
itself or bound the extent of mineralization. In our experience,
in fold-belt terrains the strongest and most laterally continuous
reection zones often can be interpreted as faults and shear
zones. In the few occasions where we applied this principle at
the mine scale, our results were encouraging.
Many tens of thousands of kilometers of high-quality, deep
seismic proles now exist for a wide range of tectonic envi-
ronments around the world. Most are in the public domain. In
almost all cases, the data have not been examined extensively
for pointers to areas of enhanced prospectivity.
ACKNOWLEDGMENTS
This paper was originally presented in an expanded form
at the Exploration97 Conference in Toronto in 1997 (Goleby
et al., 1997). We thank the organizers of that conference
for recommending its republication in GEOPHYSICS. George
Thompson and Simon Klemperer provided valuable point-
ers to published and unpublished material on reectivity of
the lower crust and fault zones. We thank two reviewers
for their suggestions. We also thank Joe Mifsud for drafting
the gures. The Mt. Isa seismic data are published with the
permission of the Director, Australian Geodynamics Cooper-
ative Research Centre. A. Owen, B. Goleby, A. Yeates, and
B. Drummond have permission to publish from the Executive
Director, Australian, Geological Survey Org.
REFERENCES
Bradshaw, M., 1993, Australian petroleumsystems: PESAJ., 21, 4353.
Drummond, B. J., and Goleby, B. R., 1993, Seismic reection images of
major ore-controlling structures in the Eastern Goldelds, Western
Australia: Expl. Geophys., 24, 473478.
Drummond, B. J., Goleby, B. R., Goncharov, A. G., Wyborn, L. A. I.,
Collins, C. D. N., and MacCready, T., 1997, Crustal-scale structures
in the Proterozoic Mount Isa inlier of north Australia: Their seismic
response and inuence on mineralisation: Tectonophysics, 288, 43
56.
Drummond, B. J., Korsch, R. J., Barton, T. J., and Yeates, A. V., 1996,
Crustal architure in northwest Tasmania revealed by deep seismic
reection proling: Austr. Geol. Surv. Org. Res. Newslett., 25, 17
19.
Fallon, G. N., Andrews, P., Bartrop, S., and Jackson, J., 1997, Drillhole
electromagnetic surveying inthe mine environment: Expl. Geophys.,
27, 6775.
Goleby, B. R., Drummond, B. J., Owen, A. J., Yeates, A. N., Jackson, J.,
Swager, C., and Upton, P., 1997, Structurally controlled mineralisa-
tion in AustraliaHowseismic proling helps nd minerals: Recent
case histories: in A. G. Gubins, Ed., Proceedings, of Exploration
97: 4th Decennial Internat. Conf. on Min. Expl., 409420.
Goleby, B. R., Rattenbury, M. S., Swager, C. P., Drummond, B. J.,
Williams, P. R., Sheraton, J. W., and Heinrich, C. A., 1993, Archaean
crustal structure from seismic reection proling, Eastern Gold-
elds, Western Australia: Austral. Geol. Surv. Org. record 1993/15.
Goodwin, E. B., and Thompson, G. A., 1988, The seismically reective
crust beneath highly extended terranes: Evidence for its origin in
extension: GSA Bull., 100, 16161626.
Heinrich, C. A., Henley, R. W., and Seward, T. M., 1989, Hydrothermal
systems: Austral. Min. Found.
Jones, R., 1986, Seismic reections from major faults: Ph.D. thesis,
Cambridge Univ.
Jones, T., and Nur, A., 1984, The nature of seismic reections from
deep crustal fault zones: J. Geophys. Res., 89, 31533171.
Klemperer, S., and the BIRPS Group, 1986, Progress in understanding
the origin of crustal reections: 2nd Internat. Symp. on Deep Seis.
Proling of Cont. Lithosphere, Abstracts, 19.
Large, R. R., 1992, Australian VHMS deposits: Features, styles, genetic
models: Econ. Geol., 87, 471510.
Loutit, T. S., Wyborn, L. A. I., Hinman, M. C., and Idnurm, M., 1994,
Palaeomagnetic, tectonic, magmatic andmineralisationevents inthe
Proterozoic of northern Australia: Ann. Conf., Australasian Inst.
Min. Metal., Proceedings, 123128.
MacCready, T., Goleby, B. R., Goncharov, A., Drummond, B. J., and
Lister, G. S., 1999, A framework of overprinting orogens based on
interpretation of the Mount Isa deep seismic transect: Econ. Geol.,
93, 14221434.
MacCready, T., Goleby, B. R., Goncharov, A., Lister, G. S., and
Drummond, B. J., 1997, An evolutionary framework for the Isan
orogeny Proterozoic terranes: Geodynamics and Ore Deposits
Conf., Australian Geodynamics Cooperative Research Centre, Ab-
stracts, 4245.
Magoon, L. B., and Dow, W. G., 1991, The petroleum systemFrom
source to trap: AAPG Bull., 75, 627.
Milkereit, B., Eaton, D., Wu, J., Perron, G., andSalisbury, M., 1996, Seis-
mic imaging of massive sulde deposits: Part IIReection seismic
proling: Econ. Geol., 91, 829834.
Neudert, M. K., andRussell, R. E., 1981, Shallowwater andhypersaline
features from the Middle Proterozoic Mount Isa sequence: Nature,
293, 284286.
Salisbury, M. H., Milkereit, B., and Bleeker, W., 1996, Seismic imaging
of massive sulde deposits: Part IRock properties: Econ. Geol.,
91, 821828.
Siegesmund, S., and Kern, H., 1990, Velocity anisotropy and shear-
wave splitting in rocks fromthe mylonite belt along the Insubric line
(Ivrea zone, Italy): Earth and Planet. Sci. Lett., 99, 2947.
Swager, C., and Grifn, T., 1990, Geology of the Archaean Kalgoorlie
terrane, northernandsouthernsheets, 1:250 000: Geol. Surv. Western
Austral.
Swager, C. P., Goleby, B. R., Drummond, B. J., Rattenbury, M.S.,
and Williams, P. R., 1997, Crustal structure of granite-greenstone
terranes in the Eastern Goldelds, Yilgarn craton, as revealed
by seismic reection proling: Precambrian Res., 83, No. 13,
4356.
Upton, P., Hobbs, B., Ord, A., Zhang, Y., Drummond, B., and
Archibald, N., 1997, Thermal and deformation modelling of the Yil-
garn deep seismic transect: Geodynamics and Ore Deposits Conf.,
Australian Geodynamics Cooperative Research Centre, Ballarat,
Abstract, 2225.
Yeates, A., McNeill, A., Richardson, S., Barton, T. J., Drummond, B. J.,
and Richardson, R. G., 1997, High-resolution reection seismic in
the Hellyer ore environment: Newdevelopments in research for ore
deposit exploration: 3rd Nat. Conf., Geol. Soc. Austr., Abstracts,
78.
Application of 3-D seismics to mine planning at Vaal Reefs
gold mine, number 10 shaft, Republic of South Africa
C. C. Pretorius
, W. F. Trewick
, A. Fourie
, and C. Irons
ABSTRACT
During 1994, a 3-D seismic reection survey was un-
dertaken at Vaal Reefs No. 10 shaft with the objective
of mapping the detailed structure of the Ventersdorp
contact reef gold orebody. This would provide vital in-
put into future mine planning and development. The
survey benetted from 10 years of 2-D seismic expe-
rience and one previous 3-D mine survey, conducted in
the Witwatersrand Basin.
The seismic survey at No. 10 shaft accurately and spec-
tacularlydelineatedthe3-Dstructureof theVentersdorp
contact reef at depths ranging from 1000 to 3500 m,
imaging faults with throws in the 20- to 1200-m range.
The resultant structure plans were satisfactorily vali-
dated by subsequent surface drilling and underground
mapping mining operations during the period 1994 to
1996. These plans have been merged with drillhole, un-
derground, and sampling data into an integrated mine
modeling, gold reserve estimation, and mine scheduling
package.
The geology department now manages the planning
function at No. 10 shaft, and 3-D seismics has played
a signicant role in placing this important responsibil-
ity rmly within the geologists domain. Building on the
success of the No. 10 shaft survey, two other 3-D seis-
mic surveys were concluded over mines during 1996 and
1997.
INTRODUCTION
The Gold Division of the Anglo American Corp. of South
Africa (AAC) has successfully used seismic reection tech-
niques in its Witwatersrand basin analysis programs since 1983.
The emphasis during the rst ten years was on reconnaissance
2-D seismic surveys for subsurface structural mapping of the
Witwatersrand Triad rocks, particularly the auriferous Central
Rand Group within the main Witwatersrand Basin (Figure 1).
Building on the success of these 2-D surveys, AACs rst 3-D
seismic survey for mine planning and development took place
at Western Deep Levels gold mine in 1993, with the second
following one year later at Vaal Reefs No. 10 shaft (Figure1).
In many respects the No. 10 shaft survey represents the
mature application of seismics to detailed structural mapping
in a deep AAC gold mine. Seismic structural maps of the
Ventersdorp contact reef gold orebody at No. 10 shaft were
satisfactorily validated by subsequent drilling and mining op-
erations between 1994 and 1996. These positive results encour-
aged two more AAC mines to conduct 3-D seismic surveys in
199697, with the latest survey, at Western Ultra Deep Levels,
eight times the size of the No. 10 shaft survey and probably one
Manuscript received by the Editor February 10, 1999; revised manuscript received June 2, 2000.
Anglo American Corp. of South Africa, Ltd., Marshalltown 2107, South Africa. E-mail: cpretorius@isdial.com; wtrewick@anglogold.com.
Irons Geophysical Consulting, Henley on Thames, Oxfordshire, United Kingdom.
of the largest mine geophysical surveys undertaken to date in
the mineral industry.
This paper summarizes the 3-Dseismic data acquisition, pro-
cessing, and interpretation methodologies developed at No. 10
shaft and illustrates howthis geophysical technique can make a
powerful contribution to optimizing orebody extraction in the
dynamic Witwatersrand mining environment.
GEOLOGICAL SETTING AND ITS RELATIONSHIP
TO SEISMIC STRATIGRAPHY
Descriptions of Witwatersrand geology and associated seis-
mic stratigraphy appear in Pretorius et al. (1987, 1994), de Wet
andHall (1994), andWeder (1994). Most of thegoldintheMain
Witwatersrand basin occurs in thin auriferous conglomerates,
anomalously termed reefs, within the predominantly arena-
ceous Central Rand Group. The principal reef mined at No. 10
shaft is the Ventersdorp contact reef. The Ventersdorp contact
reef is found at the base of the Klipriviersberg Group, which
is the basal group of the Ventersdorp Supergroup (Figure 2).
The Ventersdorp contact reef is conformable with the overly-
ing lavas but unconformable on the underlying quartzites of
1862
3-D Seismics for Mine Planning 1863
the Witwatersrand Supergroup. Figure 3 is a generalized geo-
logical section through the No. 10 shaft lease area, showing the
unconformable relationship of the various lithostratigraphic
units. This presents more stratigraphic detail within the Cen-
tral Rand Group than could be displayed on the generalized
stratigraphic column in Figure 2.
The Ventersdorp contact reef is a highly channelized reef
with thicknesses varying from 20 to 400 cm. It can be divided
intotwomajor reef typesplateaureef andchannel reefwith
the channel reef being further subdivided into three subfacies.
There is a very strong correlation between facies type and gold
grade, as described by Trewick (1994).
Figure 4 shows part of a previous northwestsoutheast 2-D
seismic sectionpassingclosetotheNo. 10shaft 3-Dsurveyarea.
It clearly illustrates the seismic stratigraphy of the Ventersdorp
and Witwatersrand Supergroups at this locality. Note espe-
cially the strong angular unconformity (event 2) where the
Platberg Group sediments are draped over underlying, tilted
fault blocks of Klipriviersberg Group lavas. The drop in both
P-wave velocity (from 6300 to 5800 m/s) and density (from 2.9
FIG. 1. Regional location, surface, and subsurface geology of the Witwatersrand basin [after Pretorius (1986) with modication].
to 2.67 g/cm
3
) as seismic waves pass from the Klipriviersberg

lavas into the underlying Central Rand Group quartzites pro-
duces a strong reection coefcient, which fortuitously coin-
cides with the Ventersdorp contact reef. Reections at this
lava contact (event 3) can therefore be used to map the
Ventersdorp contact reef orebody. The Central Rand Group
between events 3 and 4 maintains a seismically transparent
character in this portion of the Witwatersrand basin. The nu-
merous positive and negative reections in the Jeppestown
Subgroup of the West Rand Group below event 4 are
caused by alternating shales and quartzites. This stripy re-
ectivity characterizes most of the West Rand Group ex-
cept for the Bonanza Formation quartzites of the Govern-
ment Subgroup, between events 5 and 6. These understandably
have a similar seismic signature to the Central Rand Group
quartzites.
Note how a large, pre-Platberg normal fault such as f
1
f

1
(2000 m throw) on Figure 4 is imaged not only where it dis-
places the Ventersdorp contact reef but also where it traverses
the reective West Rand Group. However, it is not easy to
1864 Pretorius et al.
pick the exact position of this fault throughout the section
since certain West Rand Group reectors still appear to pass
through it undisturbed. This phenomenon is because out-of-
plane events are imaged on this section; it illustrates a major
drawback of 2-D seismics, which prevents its use as a detailed
structural mapping tool. Another excellent example of such
sideswipe is the steeply dipping fault plane reector, labeled
fpr, just above the interpreted position of f
1
f

1
. This reector
appears to originate from a strike extension of f
1
f

1
, several
hundred meters north of the 2-D section line.
Tomapstructure at the minimumresolution(20 m) expected
by the mining clients and potentially available from the seis-
mic data at the Ventersdorp contact reef level, it is crucial to
record 3-D seismic data, followed by full 3-D depth migration
at the processing phase. This will restore reectors such as fpr
to their true subsurface position in 3-D space. Further discus-
sions on the need for 3-D seismic imaging in a structurally
complex hard-rock mineral exploration environment appear
in Milkereit and Eaton (1996).
SEISMIC DATA ACQUISITION
Three-dimensional seismic survey design criteria must ad-
dress issues such as the subsurface area to be imaged, the re-
quired spatial resolution, bin dimensions, fold of cover and the
required source and receiver congurations to achieve this,
FIG. 2. Lithostratigraphic columns in the Witwatersrand basin.
migration apertures, and static control. Ashton et al. (1994)
have published an interesting article on 3-D seismic survey
design, including oil industry examples. Notes on appropriate
modications toseismic survey designcriteria whenaddressing
the hard-rock Witwatersrand environment appear in Pretorius
et al. (1987).
We knew from previous 2-D seismic surveys undertaken
at No. 10 shaft that the peak frequencies achievable at the
Ventersdorp contact reef would be between 60 and 75 Hz. As-
suming anaverage velocity of 6000 m/s andquarter-wavelength
vertical resolution criteria limits, we decided that 20 m vertical
resolution was possible. We hoped this resolution could be fur-
ther improved with the benet of 3-D migration and graphic
workstation interpretation techniques such as 3-D visualiza-
tion of horizon picks and related seismic attribute data.
The 3-D survey design was oriented toward maximizing the
structural resolutionontheVentersdorpcontact reef withinthe
5-km
subsurface core area shown on Figure 5. This area cov-

ers proposed underground development plans until the year
2012 on a complex Ventersdorp contact reef at depths ranging
between 1000 and 3500 m. The seismic acquisition parameters
required to achieve the imaging are summarized in Table 1, and
the required surface coverage is displayed in Figure 5. The data
acquisitionwas undertakenby Geoseis (Pty.) Ltd. betweenDe-
cember 1993 and February 1994.
While a full discussion of the acquisition methodology is be-
yond the scope of this paper, it is pertinent to stress certain
important aspects of the design.
Target bin size.The target bin size of 20 m was chosen
to ensure that spatial aliasing would not occur at the typical
stratigraphic dips (<30
) and maximum frequencies (<75 Hz)

in the survey area. The spatial sampling parameters will, in
theory, allow 3-D migration of dips up to 56
at frequencies up
to 90 Hz.
Table 1. Vaal Reefs No. 10 shaft 3-D seismic survey data
acquisition parameters.
Field crew Geoseis
Instrument type SN368/CS2502
No. of channels 240
Record length 3 s
Sample rate 2 ms
Fold 2000%
Nominal Vibroseis source parameters
Vibrator type Pelton Mk II
Pattern 4 Vibs in-line 62
W of N
VP interval 40 m
Array length 30 m
Sweep length 16 s
Sweep frequency 1090 Hz
Gain 6 dB/octave boost
Taper 0.3 s
Nominal receiver array parameters
Geophone type GCR
Geophone frequency 10 Hz
Station interval 40 m
Pattern in-line
Spread 4 lines of 60 receivers
Migration aperture.To migrate dips of up to 30
in the in-
line (northwesterly) direction at depths of 3500 m, a migration
aperture of 2000 m was selected for the core area shown in
Figure 5. In contrast to this, the main concern in the cross-line
direction was to ensure that diffraction hyperboloids originat-
ing at faulted contacts were sufciently sampled to allow them
to be satisfactorily collapsed during migration and to provide
adequate spatial resolution of the diffraction points. Exper-
imental migration of partially sampled, computer-simulated
diffractions was used to estimate a satisfactory strike migra-
tion aperture of 700 m.
The surface survey area shown in Figure 5 provides full fold
coverage of the core area, and it includes the strike and dip
migration apertures and the stack-on zone for both in-line and
cross-line fold build-up. The total surface coverage required to
satisfactorily image the 5 km core area is 17 km.
Fold.Fold decimation exercises were undertaken on the
original 48-fold 2-D seismic sections, and the results were used
to determine that nominal 20-fold coverage would probably be
adequate for the 3-D survey.
Recording time.A recording time of 3 s was considered
adequate to allow for full 3-D migration of seismic data down
to the Jeppestown subgroup of the West Rand Group. This
was necessary to ensure that faults affecting the Ventersdorp
contact reef could be mapped at depths where they displace
the upper West Rand Group reectors (Figure 4).
Recording geometry.Figure 6 displays the acquisition ge-
ometry used at No. 10 shaft. The staggered brick wall tech-
FIG. 3. (a) Generalized geological section No. 10 shaft. (b) Idealized westeast section across the No. 10 shaft lease area, showing
unconformable relationships (Trewick, 1994).
nique generated a satisfactory range of offsets (01100 m) and
azimuths within the bins to facilitate maximum fold at target
reection times with adequate NMO, and it accommodated
the requirements set for an optimum refraction statics solu-
tion. The staggered brick pattern provided better coverage at
short offsets than a rectangular checkerboard layout, but it was
logistically more difcult to implement and thus incurred cost
penalties. Ashton et al. (1994) summarize the merits of various
shooting patterns. Note that the area highlighted in yellow on
Figure 6 represents the recording template and not the core
area displayed on Figure 5. Also note the omissions and com-
pensation shots, which were necessary around the mine dump
in the northeast corner.
Statics.The importance of good static control in the near
surface to improve structural resolution at great depths was
appreciated at an early stage in AACs 3-D seismic program.
To this end a multiline receiver template, restricted only by
the number of traces available and acquisition efciency, was
adopted. This provided the capability of producing a more
equal distribution of the total fold between the in-line fold and
the cross-line fold, a factor important for an accurate solution
of refractor-based static algorithms.
Additionally, a low-velocity-layer (LVL) seismic refraction
program on a 1-km grid over the survey area was conducted
together with eight uphole surveys. The data obtained from
these surveys were used to optimize the nal static solution
derived from the generalized linear inversion (GLI) refraction
static algorithm.
FIG. 4. A portion of 2-D seismic line OG-54 in the vicinity
of No. 10 shaft.
FIG. 5. 3-D seismic coverage at Vaal Reefs No. 10 shaft.
SEISMIC DATA PROCESSING
Data processing operations at No. 10 shaft were divided into
two phases:
1) eld quality control processing conducted on site by
Schlumberger/Geco-Prakla (SGP) using its Voyager sys-
tem and
2) full 3-Dprocessing undertakenby SGPat its Buckingham
Gate (Gatwick) Processing Center in the United
Kingdom.
The emphasis in the eld center was on producing 2-DBrute
stacks for quality control purposes, within a day of completing
each swath. This procedure ensured that potential data acqui-
sition problems (e.g., geometry, statics) could be quickly de-
tected and rectied in the eld if necessary. GLI statics were
also derived in the eld from the Vibroseis records, ensuring a
nal check on the crucially important aspect of static control
before the seismic, LVL, and survey crews were demobilized.
The nal processing route in the U.K. is summarized in
Table 2. Most of the processing methods and parameters
are normally accepted techniques adopted for most modern
land-acquired 3-D seismic surveys. The refraction and residual
static processes, together with DMO, proved to be the critical
factors in improving the S/N ratio. The main problem reported
by the processing team was the elimination of air-blast and
near-surface noise trains without disturbing the relative
amplitudes of the shallow data. This was achieved using
Table 2. Vaal Reefs No. 10 shaft 3-Dseismic survey nal data
processing route.
Factor Description
Demultiplex SEGD 6250 bpi 9-track reels produced
on Geoseis eld crew
Geometry SEGD reels copied to Phoenix vector
format 3480 cartridges. Line geometry
and eld statics appended to trace
headers. Traces dropped outside
351064 m offset.
Statics Hampson-Russell GLI statics applied.
Airwave ProMAX air blast attenuation 331 m/s.
Gain T**1.0 spherical divergence correction.
f -k filter Full off 05200 m/s fan taper 52009360 m/s.
500 ms recoverable AGC.
D.B.S. Predictive 180-ms operator, 2 ms gap.
Equalization Window: near # 3001100 ms, far
# 4001200 ms.
Sort To CMP order, 20 m bin width.
Autostatics Surface-consistent residual statics. Derived
fromrst-round NMO-corrected gathers.
NMO Velocities picked from constant-velocity
stacks after residual statics and DMO.
Mute 120% stretch mute.
Equalization 500-ms sliding window, 250-ms move-up.
DMO 8-ms/trace dip limits.
Stack Square root compensation, 20-foldnominal.
Phase Zero-phase conversion.
Migration One-pass depth migration.
Filter 10-20-80-90 Hz Band-pass lter
(time domain).
FIG. 6. Acquisition geometry used at Vaal Reefs No. 10 shaft.
Air-Blast attenuation software available within the Promax
processing system.
The means by which the velocity model for depth migration
was constructed was, however, fundamentally different from
normal accepted practices. With the restricted far offsets and
high interval velocities associated with the hard-rock stratig-
raphy, stacking velocities are highly inaccurate when used as
an approximation of the true P-wave velocity eld of the
subsurface. An alternative method was therefore conceived
and the model constructed in a three-phase manner.
First, based on borehole geophysical logs and prior 2-D seis-
mic knowledge, an initial velocity approximation of 6000 m/s
for the entire section was used to migrate 2-D dip lines ex-
tracted from the 3-D data set.
Second, the major velocity boundaries were interpreted on
these migrated sections, and time-based structure maps were
constructed. Local average velocities for the main geologi-
cal formations were then extracted from borehole geophysical
logs, anda 3-Dvelocity andtime model was constructed. Fortu-
nately, only two layers and one horizon (Ventersdorp contact
reef) needed construction for the No. 10 shaft area, making
this operation relatively simple. This provided the nal input
model for a 3-D time migration of the seismic data volume.
Finally, the mine geologists andmanagers requiredtheir nal
data set in depth. Adepth conversion was therefore carried out
by an experienced interpretation geophysicist, familiar with
the Witwatersrand stratigraphy, and integrated into the nal
migration phase of the processing.
The output data set from the 3-D time migration was rst
loadedontoaninterpretationworkstationandthe major veloc-
ity boundary at the Ventersdorp contact reef was interpreted.
Borehole depth control together with extracted two-way trav-
eltime values from this interpretation were then used to calcu-
late the average velocity down to the Ventersdorp contact reef
at each borehole position. These values were contoured to pro-
duce a velocity map for the Klipriversberg lava which, together
with the time interpretation of the Ventersdorp contact reef,
was used as an input model to the depth migration of the data
set. The quartzites andshales of the WitwatersrandSupergroup
underlying the lava were given a constant velocity of 5800 m/s.
The nal depth-migrated data set was output in depth for
detailed interpretation on the mine. The accuracy of this nal
velocity model for migration is borne out by the accuracy of
the predicted depth and position of the faults intersected to
date on the mine, as discussed below.
INTERPRETATION
Interpretation of the No. 10 shaft seismic data com-
menced on the depth-migrated data cube using Schlumberger-
Geoquests IESX software with the Geoviz geovisualization
module. On the basis of this interpretation, the mine design
has been completed for the entire life of the mine.
One must appreciate that the data cube will still contain
some processing artifacts and depth mismatches. The interpre-
tation is therefore expected to require minor updates through-
out the life of the mine. The important principle is to apply
these updates dynamically to maintain the required predictive
capability ahead of development ends (200300 m).
Figure 7 is aninterpretedin-line (dipline) throughthe center
of the data cube, showing the Ventersdorp contact reef horizon
and associated faulting. This is a direct screen dump from the
workstation. The stratigraphy is not described in detail because
this has already been done for the 2-D type section (0G-54) in
Figure 4. (Although 0G-54 does not fall within the 3-D survey
area, the seismic stratigraphies are broadly similar.)
FIG. 7. Portion of an interpreted dip line through the center
of the 3-D depth-migrated seismic data cube.
Clearly, the Ventersdorp contact reef is well imaged on Fig-
ure 7 and there are no indications of residual sideswipe, such as
feature fpr in Figure 4. Figure 8 is a 3-D representation of the
Ventersdorp contact reef and fault surfaces after computer-
assisted tracking of the horizon and faults has taken place
through approximately half of the data volume. For illustrative
purposes, the seismic cube has been peeled back to the cur-
rent in-line section, revealing the orebody surface, displayed
in yellow, with fault planes shown in blue. One of the surface
drillholes (G 40) to which the interpretation has been tied is
displayed in the northwest. Note how grid shading of the to-
pographic surface in Geoviz, using a shallow sun angle from
the north, helps highlight smaller faults such as f 5 (throw of
approximately 20 m) by casting a signicant shadow across the
discrete change in topography.
In Figure 9 the seismic data cube has been completely rolled
backtoreveal theinterpretedVentersdorpcontact reef surface.
As shown by the reference wireframe, the Ventersdorp contact
reef depths range from1000 mto about 3000 mwithin this area.
The survey has accurately and spectacularly delineated faults
withthrows inthe 20 m( f 2) to1200 m( f 3) range, witha lateral
positioning accuracy of better than 40 m. Note that the surface
drillhole control has done little to dene the fairly complex
FIG. 8. Interpreted VCR surface approximately midway
through the 3-D cube.
FIG. 9. Grid-shaded3-Dperspectivedisplayof theVCRat Vaal
Reefs No. 10 shaft.
faulting in the subsurface, particularly the smaller faults with
throws in the range of 20 to 100 m (e.g., 100-m fault f 4). Few if
any of these faults could be interpreted from surface drilling,
and most will have a bearing on short- to medium-term mine
planning. The 3-D imaging of structures is very satisfying. As
should be expected, the fault planes are generally nonlinear.
Fault throws vary along strike, andthere is evidence for transfer
structures displacing faults such as f 4.
An unexpected result of this survey was the denition of
the large upthrown block of Ventersdorp contact reef to the
northwest of the 1200-m Mariendal fault, f 3. This structure
has since been tested by drillhole G40, west of current mine
workings and sited to intersect the block within an extrapo-
lated Ventersdorp contact reef channel facies with good grade
potential. This hole intersected the Ventersdorp contact reef
within 20 m of the interpreted depth based on the seismic
cube.
Several of the smaller faults were intersected by mining op-
erations between 1994 and 1996, generally conrming the pre-
dicted structure. One example is illustrated in the enlarged dip
section displayed in Figure 10. On this section the Ventersdorp
contact reef pick is displaced by a 45-m fault ( f 6). Subsequent
mining operations have conrmed the presence and throw of
the fault, as shown by the underground survey pegs (p). Re-
cent underground survey control indicates that fault throws as
small as 15 mcan be resolved at shallower Ventersdorp contact
reef depths within the 3-D data cube. These results are a re-
markable conrmation of the seismic method and the velocity
model when one considers a regional dip of up to 30
resulting
in Ventersdorp contact reef event migration of over 500 m.
DYNAMIC MINE PLANNING AND DEVELOPMENT
Development of a reliable facies plan has greatly assisted
with the evaluation of Vaal Reefs No. 10 shaft. Together with
the accurate structural information provided by the 3-D seis-
mic survey, this has enabled shaft geologists to develop a ro-
bust geological model. Availability of the geological model, to-
FIG. 10. Mining conrmation of VCR seismic structure plan at
Vaal Reefs No. 10 shaft.
gether with the necessary computer technology in the form of
Geoquest, Microstation, and CADS-Mine, enabled the merg-
ingof seismic, drillhole, undergroundsurvey, andsamplingdata
intoanintegratedmine modeling, reserve estimation, andmine
scheduling package. The new system is being used to generate
an upgraded, accurate, and comprehensive planning database
which is sufciently exible to facilitate dynamic replanning in
response to new information.
The planning database is currently being used to optimize
the positioning of the subshaft system, which will allow access
to the deeper portions of the Ventersdorp contact reef orebody
in the west and may extend the life of the mine to the year 2012.
This subshaft must be optimally sited within Ventersdorp con-
tact reef fault loss to minimize the sterilization of resources
while avoiding complex fault intersections that could compro-
mise safety. The structure plan displayed in Figure 9 will greatly
help planners achieve this objective.
CONCLUSIONS
The 3-D seismic method has proven to be a very cost-
effective technique for mapping structure on the Ventersdorp
contact reef at No. 10 shaft, delivering adequate resolution for
short-, medium-, and long-term mine planning and develop-
ment. Figure 11 illustrates schematically just howcost effective
FIG. 11. Information economics at Vaal Reefs No. 10 shaft.
the survey has been, compared to surface drillholes, as a means
of providing structural information on a reective geological
horizon. If each binned seismic trace is considered to be the
equivalent of a structural drillhole, the survey has arguably de-
livered spatial information equivalent to 12 000 surface drill-
holes, on a 20 20 m grid, extending to Ventersdorp contact
reef depths and beyond within the core area. Cross-sections
within the seismic cube can be viewed and interpreted in any
orientation, including horizontal time slices, as schematically
displayedinFigure11. Thetotal surveycost of US$1.07million,
in 1994 money, would fund only one 3000-m-deep drillhole, in-
cluding deections.
The ability of 3-D seismics to predict structure well ahead
of the current stope faces and development ends has been ad-
mirably demonstrated over the last three years. The seismic
data set is expected to play a key role in the modeling and
design phases of mining operations as well as in the auditing
phase, where compliance with the extraction plan is audited.
In essence, the Mine Geology Dept. now manages the plan-
ning function at No. 10 shaft, and 3-D seismics has played a
signicant role in placing this responsibility rmly within the
geologists domain.
ACKNOWLEDGMENTS
The authors are grateful to the Anglo American Corp.
and Vaal Reefs GM for their permission to publish this pa-
per. We also extend our gratitude to our colleagues at AAC
and Vaal Reefs and our consultants at Schlumberger/Geco-
Prakla, Geoseis, Hydrosearch, Armstrong de Klerk, and Irons
Geophysical for their suggestions and comments during the
preparation of the manuscript. A nal special thank you to
Desiree Hendriques for her patient andprofessional assistance
in preparing the diagrams, under enormous time pressure.
REFERENCES
Ashton, C. P., Bacon, B., D eplant e, C., Ireson, D., and Redekop, G.,
1994, 3-D seismic survey design: Schlumberger Oileld Review, 6,
No. 2, 1932.
de Wet, J. A. J., and Hall, D. A., 1994, Interpretation of the Oryx 3-D
seismic survey: 15th Congress, Council for Mining and Metallurgy
Institution, Geology 3, 259270.
Milkereit, B., and Eaton, W. D., 1996, Towards 3-D seismic explo-
ration technology for the crystalline crust: in Lawton, S. E., Ed.,
Trends, technologies and case histories for the modern exploration-
est: Prospectors and Developers Assn. of Canada short course notes,
1736.
Pretorius, C. C., Jamison, A. A., andIrons, C., 1987, Seismic exploration
in the Witwatersrand basin, Republic of South Africa: Exploration
87, Proceedings, 3, 241253.
Pretorius, C. C., Steenkamp, W. H., and Smith, R. G., 1994, Devel-
opments in data acquisition, processing and interpretation over ten
years of deep vibroseismic surveying in South Africa: 15th Congress,
Council for Mining and Metallurgy Institution, Geology 3, 249258.
Pretorius, D., 1986, Compilation of the geological map of the
Witwatersrand basin, in Anhaueusser, C. R., and Maske, S., Mineral
deposits of southern Africa: Geol. Soc. South Africa, 1, 10191020.
Trewick, W. S. F., 1994, Exploration history and mining strategy for the
Ventersdorp contact reef at Vaal Reefs No. 10 shaftKlerksdorp
goldeld: 15th Congress, Council for Mining and Metallurgy Insti-
tution, Geology 3, 6770.
Weder, E. E. W., 1994, Structure of the area south of the central
Rand gold mines as derived fromgravity and Vibroseis surveys: 15th
Congress, Council for Mining and Metallurgy Institution, Geology
3, 271282.
Mineral exploration in the Thompson nickel belt, Manitoba, Canada,
using seismic and controlled-source EM methods
Don White
, David Boerner
, Jianjun Wu
, Steve Lucas
,
Eberhard Berrer
, Jorma Hannila
, and Rick Somerville
ABSTRACT
Seismic reection and electromagnetic (EM) data
were acquired near Thompson, Manitoba, Canada, to
map the subsurface extent of the Paleoproterozoic,
nickel ore-bearing Ospwagan Group. These data are
supplemented by surface and borehole geology and
by laboratory measurements of density, seismic ve-
locity, and electrical conductivity, which indicate that
Ospwagan Group rocks are generally more seismically
reective and electrically conductive than the Archean
basement rocks that envelop them. The combined seis-
mic/EM interpretation suggests that the Thompson
Nappe (cored by Ospwagan Group rocks) lies blind
beneath the Archean basement gneisses, to the east
of the subvertical Burntwood lineament, in a series of
late recumbent folds and/or southeast-dipping reverse
faults. The EM data require that the shallowest of these
fold/fault structures occur within the basement gneisses
or perhaps less conductive Ospwagan Group rocks. The
results of this study demonstrate how seismic and deep
sounding EM methods might be utilized as regional ex-
ploration tools in the Thompson nickel belt.
INTRODUCTION
The Thompson nickel belt (TNB) in northern Manitoba,
Canada (Figure 1) has been an important producer of nickel
since the early 1960s. Known ore deposits in the belt oc-
cur without exception along distinct stratigraphic horizons
within the Paleoproterozoic supracrustal Ospwagan Group.
The supracrustal rocks and the ore deposits they host have
been subjected to medium- to high-grade metamorphism and
intense multiepisode deformation, resulting in a complex in-
terference pattern of early recumbent folds overprinted by
tight, upright, doubly plunging folds. In spite of the extensive
Manuscript received by the Editor February 1, 1999; revised manuscript received May 2, 2000.
Geological Survey of Canada, 615 Booth St., Ottawa, Ontario K1A0E9, Canada. E-mail: white@cg.NRCan.gc.ca; dboerner@NRCan.gc.ca; slucas@
NRCan.gc.ca.
Inco Limited, Exploration, Copper Cliff, Ontario P0M 1N0, Canada.
postmineralizationstructural andmetamorphic history, thesul-
de ore bodies have generally remained within their original
host units. Thus, subsurface mapping of the structurally com-
plex supracrustal unit is essential to future exploration within
the TNB.
In an attempt to map the subsurface extent of the Ospwagan
Group in this complex geological environment, 19 km of seis-
mic reection data were acquired in 1991 (Figure 1). In addi-
tion, 20 km of deep-probing EM sounding data were acquired
to help constrain the interpretation of subsurface geological
structures. Coincident controlled-source EM data and seismic
reection data have proven to be complementary in previous
studies (e.g., Boerner et al., 1993, 1994). Both data sets were
acquired with joint funding from Inco Limited, Exploration;
Lithoprobe; and the Geological Survey of Canada.
Seismic reection methods have been applied within a va-
riety of geological environments in various mining camps
in Canada, e.g., low-grade greenstone belts of the Archean
Abitibi subprovince (Selbaie area, Perron et al., 1997;
Matagami area, Milkereit et al., 1992a; Ansil mine, Perron
and Calvert, 1997), the Sudbury igneous complex (Milkereit
et al., 1992b), andtheBuchans mineinNewfoundland(Spencer
et al., 1993). The TNB geological environment is distinct from
these previous cases, representing a highly strained, moder-
ately to steeply dipping, medium- to high-grade Proterozoic
collisional belt (see Weber, 1990; Bleeker, 1990b; and citations
therein). We present an interpretation of seismic reection and
EMsounding data fromthe immediate vicinity of the Birchtree
mine (line 1A, Figure 1). Seismic data from a second prole to
the north (line 1B, Figure 1) are presented elsewhere (White
et al., 1999). The seismic and EM interpretation is constrained
by laboratory measurements of density, seismic velocity, and
electrical resistivity for representative TNB lithologies. The
model is also constrained by surface and borehole geology.
We conclude by summarizing how these geophysical mapping
methods might be successfully used as exploration tools in the
TNB.
1871
1872 White et al.
GEOLOGICAL SETTING
The TNB (Figure 1) lies at the western margin of the
Archean Superior province, where it is in fault contact
with the Kisseynew domain, a part of the Paleoproterozoic
Reindeer zone of the Trans-Hudson orogen (Hoffman, 1988),
that is dominated by 1.861.84 Ga high-grade metaturbidites
(Machado et al., 1999). Near Thompson, Manitoba (Figure 1),
the TNBcomprises variably reworked Archean gneisses inter-
leaved and infolded with the remnants of a thin Paleoprotero-
zoic cover sequence, the OspwaganGroup(Scoates et al., 1977;
Bleeker, 1990b). All ore deposits in the vicinity occur within
the Ospwagan Group and are associated with ultramac sills
that intrude the Pipe Formation of the Ospwagan Group and
have been variably dismembered (Bleeker, 1990a).
Bleeker (1990a) interprets the TNB geology in terms of a
refolded nappe structure (the Thompson Nappe) as depicted
in Figure 2. The overall geological history for the TNB, based
on Bleeker (1990a), is as follows:
1) deposition of Proterozoic Ospwagan sediments on Supe-
rior Archeanbasement ina continental rift marginsetting
(ca. 21002000 Ma?);
FIG. 1. Map of the study area near Thompson, Manitoba,
Canada. The Thompson nickel belt comprises all of the area
to the east of the Superior boundary fault. The seismic and
EM proles are indicated, as are the mapped occurrences of
the Ospwagan Group. The town of Thompson is indicated by
the T in the shaded rectangle. The base map is from Bleeker
(1990a). The upper-left inset shows the approximate location
of Thompson (indicated by the star) near the boundary of the
Superior province and the Trans-Hudson orogen (THO).
2) formation of the Thompson Nappe during an early (F1)
deformation event (>1880 Ma);
3) F2 recumbent folding coeval with peak metamorphism
(ca. 1820 Ma);
4) refolding of the nappe structure into nearly upright, dou-
bly plunging F3 folds;
5) late east-side-up ductile reverse faulting; and
6) localized brittle-ductile strike-slip faulting along the
boundary with the Kisseynew domain.
Figure 2a shows the locations of mines around Thompson in
relation to the regional Thompson Nappe structure proposed
by Bleeker (1990a). The Birchtree mine resides on the lower
limb of the nappe structure, whereas the antiform that hosts
the Thompson mine and the Owl antiform are F3 structures
associated with the overturned limb of the nappe. This model
allows for the possibility of Ospwagan metasediments in the
core of the nappe lying blind beneath the overturned basement
gneisses.
SEISMIC IMAGING IN A COMPLEX ENVIRONMENT
The seismic reection method is primarily suited to imaging
structures with shallow/moderate attitudes. Thus, the geolog-
ical setting in the Thompson area, which comprises generally
moderately/steeply oriented rock fabric with layering as well
as fold and fault structures, presents a challenging environment
in which to obtain useful seismic images. This limitation must
be respected during seismic interpretation. For example, con-
sider the steep fold geometry from the Thompson area shown
in Figure 3a (Bleeker, 1990a). The outline of a single lithologic
horizon is reproduced in Figure 3b along with the calculated
seismic response before (Figure 3c), and after (Figure 3d) data
migration. The migrated seismic image fails to reveal the steep
upright limbs of the folds, but the fold hinges and moderately
dipping limbs are well represented. The absence of the steep
fold limbs in the seismic image makes interpretation of the
image nonunique. The moderately dipping fold limb that is im-
aged might be interpreted as a fault plane. In this situation, sur-
face and borehole constraints and the structural style assumed
for the area are critical in interpreting the seismic images.
EM IMAGING IN A COMPLEX ENVIRONMENT
Diffusive EM methods provide an image of the volume-
averaged electrical resistivity of the subsurface. In contrast to
seismic reection methods, EM methods are ideally suited to
imaging near-vertical structures. The EM data provide lower
spatial resolution than seismic reection images because resis-
tivity anomalies must have lateral dimensions roughly equiva-
lent to their depth to be imaged properly. Intrinsically limited
resolution can be a benet in a complicated environment such
as the TNB because minor parasitic folding or limited offset
faulting should not disturb the EM image signicantly. Fur-
thermore, EM methods provide a means of mapping specic
lithologic units in the subsurface. The major conductive rock
units in the Thompson area are generally limited to certain
formations of the Ospwagan Group supracrustals, whereas the
Archean gneisses are resistive (see following).
From limited rock property measurements (J. Katsube, per-
sonal communication, 1993), it appears that thePipeFormation
is electrically conductive. Thus, a surface EMsurvey across the
structure illustrated in Figure 3 would provide evidence for
Exploration in the Thompson Nickel Belt 1873
steeply dipping elongated conductors wherever the Pipe For-
mation crops out. In contrast, the same structure buried under
1 kmof resistive gneiss would appear as a single broad conduc-
tive zone.
THE CAUSE OF REFLECTIVITY
Rock property measurements
Compressional-wave velocities (Vp) and density measure-
ments have been made on a suite of rock samples from the
Thompson area to assess the reectivity of the various litho-
logic units. Seismic reectivity is controlled by variations in
seismic impedance (the product of velocity and density). Seis-
mic waves are reected from contacts between rock units with
contrasting impedances when the thicknesses of the adjacent
FIG. 2. (a) A 3-D sketch of the northern segment of the refolded Thompson Nappe structure as seen from the southeast (from
Bleeker, 1990a). Mine locations are indicatedby the small boxes andcorrespondtothose showninFigure 1. (b) Schematic transverse
section through the Thompson Nappe structure (fromBleeker, 1990a), showing the structural positions of the mines fromFigure 1.
rock units are at least 1/8 of a seismic wavelength (Widess,
1973). For example, this corresponds to a thickness of 7.5 m
for a signal with a 100-Hz dominant frequency in rocks having
Vp =6000 m/s. Case studies from other areas indicate that re-
ections can be caused by lithologic contacts (e.g., White et al.,
1994), alterationzones, brittle fault zones, andshear (mylonite)
zones. Bothbrittleandductileshear zones arewell documented
in the Thompson area (e.g., Fueten and Robin, 1989).
The Vp density measurements for a suite of rock samples
from the Thompson area are shown in Figure 4. The samples
include velocities measured under an applied uniaxial stress
of 17 to 25 MPa or under a hydrostatic conning pressure
of 30 MPa. Relatively high impedance (Z) values (Z =18.5
10
6
21.5 10
6
kg/m
2
s) are observed for amphibolite, iron
formation, pyroxenite, and sulde ore, as compared with the
1874 White et al.
other lithologies (Z =13.0 10
6
16.5 10
6
kg/m
2
s). Re-
ection coefcients for juxtaposed high- and low-impedance
lithologies range from 0.06 to 0.25, compared with 0.0 to 0.11
for juxtaposedlow-impedance lithologies. Thus, the largest am-
plitude reections will likely be associated with amphibolite,
iron formation, pyroxenite, and sulde ores embedded within
any of the low- tomedium-impedance units. The geology logs in
Figure 5 reveal that all of these high-impedance lithologies
with the exception of sulde ore, which was not intersected
in any signicant thickness in this boreholeare observed in
contact with various low-impedance lithologies.
BOREHOLE LOGS
Knowledge of the spatial distribution of lithologies (i.e.,
thickness, juxtaposition sequence, lateral extent) is required to
characterize the reectivity of the subsurface. This is best done
using borehole velocity and density logs in conjunction with
vertical seismic proling (VSP). However, in the absence of
downhole geophysical logging, the spatial distribution of seis-
FIG. 3. (a) Prole through the Thompson Nappe structure at the Thompson mine (from Bleeker, 1990a). (b) Trace of one of the
layers from(a). (c) Simulated seismic vertical incidence response (unmigrated) of the structure in (a). Response was calculated using
a forward modeling algorithm based on wave equation datuming (Berryhill, 1979). (d) Vertical incidence response of the structure
in (a) after constant-velocity migration of the seismic data using an algorithm based on the frequency wavenumber phase-shift
method (Gazdag, 1978). The Pipe Formation is represented by the pelitic schist unit.
mic impedance down the borehole can be estimated by using
the laboratory-measured velocities and densities in conjunc-
tion with the borehole geological logs. The estimated variation
of seismic impedance can then be used to calculate the syn-
thetic normal incidence seismic response (Wuenschel, 1960).
The reectivity estimated in this manner is strictly applicable
for a layered medium where individual lithologic units are lat-
erally continuous. To apply these results in the TNBgeological
environment, weassumethat theinterfacebetweencontrasting
lithologic units is laterally continuous over a minimum lateral
distance on the order of a Fresnel zone (Sheriff and Geldart,
1982). For example, this corresponds to a distance of 170 m
at 1000 m depth and 300 m at 3000 m depth for a signal with
a 100-Hz dominant frequency in rocks having Vp =6000 m/s.
The seismic response estimated in this manner for two bore-
holes in the vicinity of the Birchtree mine are shown in Fig-
ure 5. Signicant reections are predicted for (1) the amphi-
bolite bodies both within the basement gneisses (9001500 m
in borehole 86 250) and the Ospwagan supracrustals (700 and
850 m in borehole 38 682-1), (2) iron formation within the
Ospwagan Group (1600 and 18002000 m in borehole 86 250),
(3) pyroxenite (1050 and 1150 m in borehole 38 682-1), (4)
serpentinite when in contact with some lithologies (1570 m in
borehole 86 250; 1030 and 1150 m in borehole 38 682-1), (5)
schist in contact with quartzite (<500 m in borehole 38 682-1),
and (6) an ore zone (1720 m in borehole 86 250). A hypotheti-
cal 10-m-thick massive sulde zone placed within a schist host
rock produces a large-amplitude reection (1300 and 2100 m
in boreholes 38 682-1 and 86 250, respectively).
Based on these results, we conclude the following:
1) the contact of the basement gneisses and the Ospwagan
supracrustals should not be a signicant reector, region-
ally, because the orthogneisses and paragneisses have
similar impedances;
2) the reectivity of the Ospwagan Group should allow
discrimination from the more homogeneous basement
gneisses;
3) where amphibolite is prevalent within the basement
gneisses, they will be highly reective, making them dif-
cult to distinguish from the Ospwagan supracrustals;
4) massive sulde units will be reective when they are
>10 m thick; and
5) the reectivity of serpentinite may provide a more use-
ful marker for the location of massive suldes because
of the larger dimensions of the serpentinite bodies and
the genetic relation between the nickel ore deposits and
ultramac rocks in the TNB.
FIG. 4. MeanVpanddensities measuredfor asuite of rocksam-
ples fromthe Thompson area. The solid curves indicate lines of
constant acoustic impedance (product of Vp and density with
units of kg/m
2
s). Measurements were made on dry samples
under hydrostatic pressure (30 MPa) or under an applied uni-
axial stress (1725 MPa). In the absence of measurements on
massive suldes from the Thompson area, the Vp and density
values for massive sulde are for a pyrhotite-rich sample from
Sudbury measured at a conning pressure of 200 Mpa.
THE CAUSE OF CONDUCTIVITY
Surface observations of induced subsurface electrical cur-
rents cannot distinguish between the two dominant upper
crustal conduction mechanisms: metallic conduction or ionic
transport in uids. However, knowledge of local geological
structure lets us infer the predominant conduction mode. For
example, uids tend to produce laterally continuous, subhor-
izontal conductivity anomalies unless trapped by a sealing
mechanismthat prevents themfrommigrating. Thus, the pres-
ence of uids may be a reasonable interpretation where a per-
vasive, subhorizontal conductive layer is observed. In contrast,
solid semiconductors or conductors can attain any attitude or
orientation; thus, local discrete conductors of variable attitude
and orientation are likely metallic conductors.
Electrical rock property measurements
Resistivity and porosity measurements made on samples of
various lithologies fromthe Thompson area are plotted in Fig-
ure 6. The resistivity measurements were made using methods
described by Katsube et al. (1991) with samples saturated with
distilled water. The serpentinites and peridotites are the most
conductive and porous. Moderately conductive rocks (i.e., be-
tween100 and1000 ohm-m) include the dolomite andquartzite
of the OspwaganGroup. Amphibolite, biotite schist, schist, and
orthogneiss are all more resistive and show little porosity. Fig-
ure 6 illustrates that any known conductive units are intimately
related to the Ospwagan Group.
A useful means of characterizing electrical conductivity as
determined by ionic transport is through a simplied form of
Archies law:
observed
=
uid
m
,
where m is the saturation exponent, is the porosity,
observed
is the measured in-situ resistivity, and
uid
is the resistivity of
the pore uid. The straightline plotted on Figure 6 is a best t
of Archies law to all of the rock property data, regardless of
rock type. The exponent m =2.33 is within the range of experi-
mentally observedvalues (typically 1.33.0). The lowporosities
suggest that most rocks in the Thompson area are electrically
resistive. Aporosity of 3%would be required to generate a re-
sistivity of 10 ohm-m. This porosity is at least anorder of magni-
tude greater than that observed for the Ospwagan Group and
Archean gneisses, with the exception of some highly porous
ultramac rocks.
These data indicate that, in general, Ospwagan Group
supracrustal rocks and the Archean gneiss are characterized
by low porosity and that the dolomites and quartzites are
marginally conductive whereas the ultramac rocks are highly
conductive. The resistive host rocks afford deep penetration
of the EM elds by virtue of limited attenuation. This obser-
vation, coupled with the association of the conductive bodies
with the Ospwagan Group, means that EM mapping methods
should be capable of delineating the subsurface extent of the
supracrustal rocksa key objective of this study.
SEISMIC AND EM SURVEY DESIGN
Figure 7 shows the location of the coincident seismic and EM
proles inrelationtothe IncoBirchtree mine andthe outline of
the Ospwagan Group supracrustals. We refer to this combined
1876 White et al.
F
I
G
.
5
.
S
y
n
t
h
e
t
i
c
V
p
,
d
e
n
s
i
t
y
,
s
e
i
s
m
i
c
i
m
p
e
d
a
n
c
e
,
a
n
d
s
e
i
s
m
i
c
r
e
s
p
o
n
s
e
d
e
t
e
r
m
i
n
e
d
f
o
r
t
h
e
g
e
o
l
o
g
i
c
l
o
g
f
r
o
m
(
a
)
b
o
r
e
h
o
l
e
3
8
6
8
2
-
1
a
n
d
(
b
)
b
o
r
e
h
o
l
e
8
6
2
5
0
,
u
s
i
n
g
t
h
e
r
o
c
k
-
p
r
o
p
e
r
t
y
m
e
a
s
u
r
e
m
e
n
t
s
f
r
o
m
F
i
g
u
r
e
4
.
T
h
e
s
e
i
s
m
i
c
r
e
s
p
o
n
s
e
i
s
d
e
t
e
r
m
i
n
e
d
f
o
r
t
h
e
f
r
e
q
u
e
n
c
y
r
a
n
g
e
o
f
3
5
t
o
1
3
0
H
z
t
y
p
i
c
a
l
l
y
u
s
e
d
i
n
h
i
g
h
-
f
r
e
q
u
e
n
c
y
s
e
i
s
m
i
c
s
u
r
v
e
y
s
.
T
h
e
l
o
g
r
e
s
p
o
n
s
e
f
o
r
a
1
0
-
m
-
t
h
i
c
k
m
a
s
s
i
v
e
s
u
l
d
e
l
a
y
e
r
e
m
b
e
d
d
e
d
w
i
t
h
i
n
a
s
c
h
i
s
t
h
o
s
t
r
o
c
k
h
a
s
b
e
e
n
a
d
d
e
d
t
o
t
h
e
l
o
g
s
(
b
e
l
o
w
t
h
e
d
a
s
h
e
d
l
i
n
e
)
f
o
r
c
o
m
p
a
r
i
s
o
n
.
prole as line 1A. Also shown is the location of an orthogonal
EM prole that is referred to in the text but is not shown. The
seismic/EM prole was intended to provide a complete cross-
ing of the Ospwagan Group using existing roads. Starting to the
northwest of the Birchtree mine, the prole crosses the F1 syn-
form of the Thompson Nappe and continues southeast across
F3 antiforms (Owl and basement antiforms) which lie along
strike from the antiform that hosts the Thompson mine to the
northeast (see Figure 2). The seismic data were acquired using
a 240-channel telemetry acquisition system with in-eld stack-
ing, noise-rejection, and correlation capabilities. An asymmet-
ric (80/160 channel) split-spreadgeophone array was usedwith
a 20-m geophone group interval, 10-m source group interval,
and geophones with a resonant frequency of 30 Hz. The seis-
mic source consisted of two Heaviquip Hemi 50 vibrators, po-
sitioned in-line with one directly behind the other, sweeping
twice at each shot station and using a 12-s linear upsweep from
30 to 130 Hz. Data processing applied to these data included
crooked-line binning, deconvolution, time-variant band-pass
ltering, weathering static corrections, detailed velocity anal-
ysis, trim static corrections, and constant-velocity (6000 m/s)
f -k migration. Ageneral description of these processes can be
found in Yilmaz (1987). A hybrid of constant-velocity stacks
was more effective in imaging the structures with highly vari-
able dip than standard dip-moveout processing.
The EM survey consisted of two proles, parallel and per-
pendicular to strike of the Ospwagan Group, designed to con-
strain the true structural dip. Source loop dimensions of ap-
proximately 1000 1000 m (1% precision) were designed
to obtain reliable information about the diffusing current sys-
tem to depths of approximately 4000 m. Data were collected
with the UTEM system (West et al., 1984) in 20-channel mode
with a base frequency of 31 Hz. Receiver proles of at least
3000 m in length and centered about the loop allowed a three-
fold redundancy while allowing uniform resolution in the near
surface along the entire prole. The EM data were processed
using the depth image processing method (e.g., Macnae and
FIG. 6. Rock-property results froma suite of samples provided
by Inco Limited, Exploration. Plotted is the log of resistivity
versus the log of porosity to facilitate the interpretation of con-
ductivity in terms of ionic transport (uids) versus electronic
transport (metals and semiconductors).
Lamontagne, 1987) to convert step response data into esti-
mates of conductivity versus depth. Discrete, steeply dipping
conductors disrupt this processing, particularly if they are suf-
ciently conductive that the secondary EM eld induced within
them is characterized by a long time constant decay. The lo-
cations of such bodies are indicated with black arrows on Fig-
ure 7; the imaged conductivity section is not reliable in these
regions.
INTERPRETATION OF LINE 1A PROFILES
The unmigrated seismic data for line 1A are shown in Fig-
ure 8, and the combined migrated seismic data and EM resis-
tivity image are shown in Figure 9. Large, steeply dipping and
highlyconductivebodies at thewest endof thelinenear abroad
unit of exposed iron formation make the EMimage west of the
Burntwood lineament unreliable. Further east, there is little
disturbance from near-surface conductors. An interpretation
of the combined seismic/EMimage is marked on Figure 9, and
a perspective view of the interpretation is shown in Figure 10.
Seismic interpretation
1) Based on rock property studies, zones of high reectiv-
ity are interpreted as being associated primarily with the
Ospwagan Group. The highly reective regions outline
the F1 nappe folded by the upright F3 structures. The
F1 structure in general dips to the southeast, as do most
of the reections. Since the reection data are biased to-
ward shallowdips (<60
), the most prominent part of the

overall F1 structure imaged is the crest of the F3 antiform
(Owl antiform), which is approximately along strike
from the F3 antiformal culmination at the Thompson
mine.
2) The depth extent of the Ospwagan Group within the
F1 fold structure is difcult to determine from the seis-
mic data alone. In areas of favorable structural attitudes
(e.g., the core of the F3 Owl antiform), the seismic im-
age is dominated by shallowly southeast-dipping reec-
tions to at least 5000 m depth. Using the thickness of the
Ospwagan Group rocks in the vicinity of the Birchtree
mine (1800 m, Figure 7) as a guide, the lower limb (F1)
basement-cover contact must occur within this generally
reective zone (dotted line).
3) The pattern of reections observed to the southeast of
the Owl antiform is interpreted as showing the over-
turned basement-cover contact imbricated along steeply
southeast-dipping reverse faults that necessarily post-
date F3. This interpretation implies that the overturned
Ospwagan Group stratigraphy (i.e., upper limb of the
nappe) should be within 1000 m of the surface near the
southeast endof theline. Amappedantiformwithinbase-
ment rocks southeast of the Grass River lineament (Fig-
ure 7) appears to have been imaged within the hanging
wall of the easternmost of the three major faults. The
faults in this model can be projected to about 2000 to
4000 mdepth and appear to become listric, dipping more
shallowly to the southeast. These structures may accom-
modate northeast-side-up displacement and thus are ex-
amples of thelate, ductilereversefaults foundthroughout
the TNB (see Fueten and Robin, 1989).
1878 White et al.
4) The Burntwood lineament (mylonite zone) was not di-
rectly imaged on the seismic proles but is represented
by a zone of truncated reectivity west of the Owl an-
tiform. The seismic signature changes dramatically across
the Burntwood lineament from highly reective in the
east to nonreective in the west as a result of the west-
ward change across the structure to steep dips.
5) West of the Burntwood lineament, the geologic sections
(from borehole information) show all of the lithologic
contacts dipping steeply to the northwest in contrast to
the seismic data, which are dominated by moderately
southeast-dipping reections. These southeast-dipping
reectors are interpreted as being associated with known
southeast-dipping faults. The northwest-dipping litho-
logic contacts were not imaged for two principal reasons:
the steep dips are difcult to image with conventional
CMP processing and insufcient maximum offsets in
the downdip direction (northwest) were recorded during
acquisition.
FIG. 7. Location of seismic/EM prole in the vicinity of the Birchtree mine. The numbers annotated along the seismic line denote
vibration-point stations (every 10 m) that increment by 5 and spatially coincident geophone group stations (every 20 m) that
increment by 10. The dark arrows indicate the locations of strong conductors detected near the surface by the EM survey.
Interpretation of the EM data
The conductivity structure southeast of the Burntwood lin-
earment is characterized by a resistive (10
3
10
5
ohm-m) near-
surface layer that overlies a southeastward-dipping layer of
high conductivity (<100 ohm-m). The depth to the conduc-
tive layer increases from approximately 1000 m near the
Burntwood lineament to 2500 m at the southeast end of the
line. Decay curves fromadjacent loops on the southeast end of
the line demonstrate a gradual northwestward slowing of the
secondary magnetic eld diffusion along the prole, diagnostic
of a conductive layer that shallows to the northwest. The EM
data from the perpendicular prole (not shown) suggest that
the conductive layer also dips gently to the northeast, indicat-
ing a north true dip. The other discrete conductors identied
with the surface EM survey (Figure 7) all parallel the strike of
the Ospwagan supracrustal group and dip vertically with little
or no discernible plunge, although plunge is difcult to resolve
with these data.
The spatial association of surcial EM anomalies with the
metasediments andtheresults of thelaboratoryresistivitymea-
surements both suggest the most plausible source of the deep
conductive layer is Ospwagan Group rocks. While these rocks
are more porous thanthe basement gneiss, the conductive layer
FIG. 8. Unmigrated seismic data for line 1A. This stack was obtained by using low stacking velocities (6000 m/s). Vibration-point
stations are annotated along the top of the section. The location of geological features (from Figure 7) is also shown.
FIG. 9. Migrated seismic data for line 1Aoverlying the color resistivity versus depth image for the coincident EMprole. Vibration-
point stations are annotated along the top of the section. The location of geological features (from Figure 7) is also shown.
dips markedly to the north, making uids an unlikely conduc-
tion mechanism. The observed anomaly is laterally continu-
ous and thus may represent a conductive formation of the
Ospwagan Group. Alternatively, it may represent the cumu-
lative response of several large, discrete conductors such as
1880 White et al.
peridotite/serpentinite bodies embedded within a mildly con-
ductive matrix. The graphitic/sultic shale member of the Pipe
Formation probably contributes to the overall conductivity of
the formation (Zablocki, 1966).
COMBINED INTERPRETATION
A 3-D perspective view of the surface geology and subsur-
face interpretation is shown in Figure 10. The subsurface inter-
pretationis basedonthe seismic/EMimage inconjunctionwith
surface and borehole/mine geology. The Burntwood lineament
appears tobeasubvertical structural discontinuitythat disrupts
the inferred continuity of the Thompson Nappe structure (see
Figure 3) between the Birchtree mine and the Owl antiform.
The signicance of the Burntwood lineament is suggested by
the discontinuity in the seismic signature across the lineament,
the disparity in conductivity associated with Ospwagan out-
crops on either side of the lineament (i.e., Birchtree versus Owl
antiform), and an increase in metamorphic grade from north-
west tosoutheast across thefault, suggestingrelativesoutheast-
side-up movement.
Details of the interpretation model west of the Burntwood
lineament are based primarily on borehole and mine geology.
The seismic and EM surveys failed to provide robust subsur-
face images in this region because of the subvertical attitude of
the Ospwagan Group and the presence of large, strong, near-
surface conductors in this region (black arrows, Figure 7).
The Owl antiform containing Ospwagan supracrustals
plunges northeastward beneath the prole as suggested by sur-
facegeological observations. This is supportedbyspatial coinci-
dence of (1) the antiformimaged seismically and its position on
rudimentary downplunge projections of the surface fold traced
into the seismic imaging plane and (2) the top of the conduc-
tive zone with the crest of the antiform. This interpretation
is consistent with the observation that the Ospwagan Group
is generally conductive, but it is in apparent contradiction to
the resistivity image from the perpendicular EM prole (not
shown), which indicates that Ospwagan supracrustals within
the Owl antiform have a lower conductivity than those west
of the Burntwood lineament and a shallow rather than steep
plunge of this conductive layer to the northeast. This question
remains unanswered.
FIG. 10. Perspective view (looking from the northeast) of the
combined seismic/EM interpretation.
Southeast of the Owl antiform, the seismic interpretation
suggests that the overturned basement-cover contact is imbri-
cated along a series of late steeply southeast-dipping reverse
faults carrying Ospwagan supracrustals to the shallow subsur-
face. Although this structural interpretation is plausible, the
high resistivities observed for this same zone require that the
folds/faults actually lie within reective basement gneisses or
resistive supracrustal rocks. Assuming the latter, it is possible
that the reective rocks represent Ospwagan metasedimentary
rocks fromthe overturned limb of the Thompson Nappe (gen-
erally resistive) and that the deep conductive layer which dips
to the north actually represents metasedimentary rocks from
the basal (upright) limb of the nappe. If this is the case, the two
limbs of supracrustals may not be physically connected and are
perhaps separated by the Burntwood lineament.
CONCLUSIONS
The results of this combined seismic and EM study suggest
that these methods are effective in determining the subsurface
conguration of the nickel ore-bearing Ospwagan supracrustal
group. This conclusionis basedonrock-property measurments,
borehole logs, and correlation of the geophysical images with
mapped surface geology. The Ospwagan supracrustals should
generally be more seismically reective and electrically con-
ductive thanthe Archeangneisses that surroundtheminF1F3
fold interference structures in the Thompson area.
Further work is required to substantiate this hypothesis.
In particular, more rigorous ground-truthing of geophysical
images is essential. The only area along the existing prole
having boreholes or mineworks to depths sufcient to cor-
relate directly with the geophysical images is in the vicinity
of the Birchtree mine. Unfortunately, this area is where the
seismic and EMimages are poorly constrained because of sub-
vertical lithology and strong near-surface conductors, respec-
tively. The in-situ reectivity of the subsurface can be deter-
mined by acquiring downhole sonic and density logs and VSPs
to test the conclusions based on the existing rock-property
measurements.
The structural complexity of the Thompson Nappe dictates
that no single method of subsurface imaging will be effective
everywhere. The existing mines in the Thompson area are lo-
cated where steeply dipping nappe limbs breach the surface. In
these areas, vertical seismic proling (e.g., Eaton et al., 1996)
may be a more effective means of subsurface imaging since
surface seismic methods are hampered by the steep structural
attitudes. However, if the nappe model is valid, much of the
Ospwagan Group may lie blind beneath the Arohean gneisses
withstructural attitudes that areamenabletoimagingusingsur-
face seismics (e.g., east of the Burntwood lineament). In these
areas, surface seismics and EM methods may be strategically
exploited. Careful survey design will be required to account for
3-Dstructure; ultimately, 3-Dsurveys will be needed to obtain
accurate structural information.
ACKNOWLEDGMENTS
The seismic data were acquired by Enertec Geophysical
Ltd. with original data processing by Western Geophysical.
The UTEMdata were acquired and processed by Lamontagne
Geophysics. An early version of the manuscript was reviewed
by B. Roberts, W. Bleeker, D. Eaton, and an anonymous
reviewer. The authors thank Inco Limited, Exploration, for
providing access andallowing publicationof the geological logs
and rock-property measurement. This is Geological Survey of
Canada paper 2000021 and Lithoprobe contribution 1169.
REFERENCES
Berryhill, J. R., 1979, Wave equation datuming: Geophysics, 44, 1329
1344.
Bleeker, W., 1990a, Evolution of the Thompson nickel belt and its
nickel deposits, Manitoba, Canada: Ph.D. thesis, Univ. of New
Brunswick.
1990b, New structural-metamorphic constraints on early Pro-
terozoic obliquecollisionalong theThompsonnickel belt, Manitoba,
Canada, in Lewry, J. F., and Stauffer, M. R., Eds., The early Protero-
zoic Trans-Hudson orogen of North America: Geol. Assn. Canada,
Special Paper 37, 5773.
Boerner, D. E., Wright, J. A., Thurlow, J. G., and Reed, L. E., 1993, Ten-
sor CSAMT studies at the Buchans Mine in central Newfoundland:
Boerner, D. E., Kellett, R., and Mareschal, M., 1994, Inductive source
EMsounding of the Sudbury structure: Geophys. Res. Lett., 21, 943
946.
Eaton, D., Guest, S., Milkereit, B., Bleeker, W., Crick, D., Schmitt, D.,
and Salisbury, M., 1996, Seismic imaging of massive sulde deposits,
part III: Borehole seismic imaging of near-vertical structures: Econ.
Geol., 91, 835840.
Fueten, F., and Robin, P. F., 1989, Structural petrology along a tran-
sect across the Thompson belt, manitoba: Dip slip at the western
ChurchillSuperior boundary: Canadian J. Earth Science, 26, 1976
1989.
Gazdag, J., 1978, Waveequationmigrationwiththephase-shift method:
Hoffman, P. F., 1988, United plates of America, the birth of a cra-
tion: Early Proterozoic assembly andgrowthof Laurentia: Ann. Rev.
Earth Planet. Sci., 16, 543603.
Katsube, T. J., Best, M. E., and Mudford, B. S., 1991, Petrophysical
charatcteristics of shales from the Scotian shelf: Geophysics, 56,
16811689.
Machado, N., Zwanzig, H. V., and Parent, M., 1999, From basin to
continent: The chronology of plutonism, sedimentation and meta-
morphism of the southen Kisseynew metasedimentary belt, Trans-
Hudsonorgen(Manitoba, Canada): Can. J. EarthSciences, 36, 1829
1842.
Macnae, J. C., and Lamontagne, Y., 1987, Imaging quasi-layered con-
ductive structures by simple processing of transient electromagnetic
data: Geophysics, 52, 545554.
Milkereit, B., Adam, E., Barnes, A., Beaudry, C., Pineault, R., and
Cinq-mars, A., 1992a, Application of reection seismology to min-
eral explorationinthe Matagami area, Abitibi belt, Quebec: Current
Research, Geol. Surv. Canada Paper 21-1C, 1318.
Mikereit, B., Green, A., and the Sudbury Working Group, 1992b, Ge-
ometry of the Sudbury structure fromhigh-resolution seismic reec-
tion proling: Geology, 20, 807811.
Perron, G., and Calvert, A. J., 1997, Shallow, high-resolution seismic
imaging of the Ansil mining camp in the Abitibi greenstone belt:
Perron, G., Milkereit, B., Reed, L. E., Salisbury, M., Adam, E., and
Wu, J., 1997, Integrated seismic reection and borehole geophysical
studies at Les Mines Selbaie, Quebec: CIM Bull., 90, 7582.
Scoates, R. F. J., Macek, J. J., and Russell, J. K., 1977, Thompson nickel
belt project: Report of eld activities, Manitoba, Min. Res. Div., 47
53.
Sheriff, R. E., and Geldart, L. P., 1982, Exploration seismology vol. 1:
History, theory and data acquisition: Cambridge Univ. Press.
Spencer, C., Thurlow, G., Wright, J., White, D. J., Carroll, P., Milkereit,
B., and Reed, L., 1993, A Vibroseis reection seismic survey at the
Buchans Mine in central Newfoundland: Geophysics, 58, 154166.
Weber, W., 1990, The ChurchillSuperior boundary zone, southeast
margin of the Trans-Hudson orogen: A review, in Lewry, J. F.,
and Stauffer, M. R., Eds., The Early Proterozoic Trans-Hudson
orogen of North America: Geol. Assn. Canada, Special Paper 37,
4255.
West, G. F., Macnae, J. C., and Lamontagne, Y., 1984, A time-domain
EM system measuring the step response of the ground: Geophysics,
49, 10101026.
White, D. J., Milkereit, B., Wu, J., Salisbury, M. H., Berrer, E. K.,
Mwenifumbo, J., Moon, W., and Lodha, G., 1994, Seismic reectivity
of the Sudbury structure North Range fromborehold logs: Geophys.
Res. Lett., 21, 935938.
White, D. J., Jones, A. G., Lucas, S. B., and Hajnal, Z., 1999, Tectonic
evolution of the Superior boundary zone from coincident seismic
reection and magnetotelluric proles: Tectonics, 18, 430451.
Widess, M. B., 1973, Howthinis a thinbed?: Geophysics, 38, 11761180.
Wuenschel, P. C., 1960, Seismogram synthesis including multiples and
transmission coefcients: Geophysics, 25, 106129.
Yilmaz, O., 1987, Seismic data processing: Soc. Expl. Geophys.
Zablocki, C. J., 1966, Electrical properties of some iron formations
and adjacent rocks in the Lake Superior region, in Hansen, D. A.,
Heinrichs, W. A., Holmer, R. C., MacDougall, R. E., Rogers, G. R.,
Sumner, J. S., and Ward, S. H., Eds., Mining geophysics, Vol. I, Case
histories: Soc. Expl. Geophys., 465492.
Physical properties and seismic imaging of massive suldes
Matthew H. Salisbury
, Bernd Milkereit
, Graham Ascough
, Robin Adair
,
Larry Matthews
, Douglas R. Schmitt
, Jonathan Mwenifumbo
,
David W. Eaton
, and Jianjun Wu
ABSTRACT
Laboratory studies showthat the acoustic impedances
of massive suldes can be predicted from the physical
properties (V
p
, density) and modal abundances of com-
mon sulde minerals using simple mixing relations. Most
suldes have signicantly higher impedances than sili-
cate rocks, implying that seismic reection techniques
can be used directly for base metals exploration, pro-
vided the deposits meet the geometric constraints re-
quired for detection. To test this concept, a series of 1-,
2-, and 3-D seismic experiments were conducted to im-
age known ore bodies in central and eastern Canada. In
one recent test, conducted at the Halfmile Lake copper-
nickel deposit inthe Bathurst camp, laboratory measure-
ments on representative samples of ore and country rock
demonstratedthat theores shouldmakestrongreectors
at the site, while velocity and density logging conrmed
that these reectors should persist at formation scales.
These predictions have been conrmed by the detection
of strong reections from the deposit using vertical seis-
mic proling and2-Dmultichannel seismic imaging tech-
niques.
INTRODUCTION
In the early 1900s, the petroleum industry relied on surface
mapping, potential eld techniques, and wildcat drilling for ex-
ploration purposes, much as the mining industry still does to-
Manuscript received by the Editor January 27, 1999; revised manuscript received June 20, 2000.
Geological Survey of Canada, Bedford Inst. of Oceanography, 1 Challenger Drive, Dartmouth, Nova Scotia B2Y4A2, Canada. E-mail: matts@agc.
bio.ns.ca.
Institut f ur Geophysik, Universitat zu Kiel, Olshausenstrasse 40, D-24098, Kiel, Germany. E-mail: bernd@geophysik.uni-kiel.de.
Noranda Mining andExploration, Inc., 960 Alloy Drive, Thunder Bay, Ontario, P7B6A4, Canada. E-mail: ascoughg@normin.com; adairr@normin.
com.
Noranda and Mining Exploration, Inc., 605 5th Avenue SW, Calgary, Alberta, T2P 3H5, Canada. E-mail: larry.matthews@chel.com.
University of Alberta, Department of Geology, Edmonton, Alberta, 2G6 2E3, Canada. E-mail: doug@phys.ualberta.ca.
Geological Survey of Canada, Mineral Resources Division, 601 Booth St., Ottawa, Ontario K1A 0E8, Canada. E-mail: jarako@nrcan.gc.ca.
University of Western Ontario, Department of Earth Sciences, London, Ontario N6A 5B7, Canada. E-mail: deaton@julian.uw.ca.
Pulsonic Corporation, 301, 400-3rd Avenue SW, Calgary, Alberta, T2P 4H2, Canada. E-mail: jj@pulsonic.com.
day. Following the initial tests of the seismic reection method
by Karcher more than 75 years ago, however, and a string
of oil discoveries during the rst commercial reection sur-
veys several years later, the exploration methods employed by
the two industries rapidly diverged. Within a decade, reec-
tion seismology had become the principal exploration tool of
the petroleum industry, a position maintained to the present
day through spectacular advances in acquisition, processing,
and imaging technology (see Weatherby, 1940, and Enachescu,
1993, for review).
Despite the remarkable success of the seismic reection
method, the mining industry has been reluctant to embrace this
technology because, until recently, its needs could be met by
traditional methods. With the known shallow reserves of cop-
per and zinc declining, however, it has become obvious that
new deep exploration tools must be developed if the industry
is to remain viable in the future (Debicki, 1996). Given the
overall similarity of the exploration problems faced by the two
industries, it is appropriate to ask whether or not this sophis-
ticated technology might also be used for the direct detection
of massive suldes.
Although no single relationship can be used to predict the
resolution of seismic reection techniques, it should be possi-
ble to image a sulde deposit using seismic reection if three
general conditions are met:
1) The difference in acoustic impedance between the ore
and the country rock must be sufciently large to pro-
duce strong reections. If Z
o
and Z
c
are the acoustic
impedances, or velocity-density products, of the ore and
country rockrespectively, the reectioncoefcient R (the
1882
Properties and Imaging of Suldes 1883
ratio of reected to incident energy) can be calculated for
the case of vertical incidence from the relation,
R =
Z
o
Z
c
Z
o
+ Z
c
(1)
In practice, an impedance difference of 2.5 10
5
g/cm
2
s
(the contrast between mac and felsic rocks) gives
rise to a value of R =0.06, the minimum value re-
quired to produce a strong reection in most geologic
settings.
2) To be imaged as a planar reecting surface, the body must
have a diameter whichis greater thanthe widthof the rst
Fresnel zone, d
F
, dened from the relation
d
F
= (2zv/f )
1/2
, (2)
where z is the depth of the deposit, v is the average for-
mation velocity, and f is the dominant frequency used in
the survey. Smaller deposits with a diameter equal to one
wavelength can be detected as point sources, but not im-
aged. Since amplitudes are severely attenuated for bod-
ies which are less than one wavelength across (Berryhill,
1977), the diameter of the smallest body which can be
detected in practice, d
min
, is
d
min
= v/f. (3)
3) The minimum thickness, t
min
, that can be resolved us-
ing seismic reection can be estimated from the quarter
wavelength criterion,
t
min
= v/(4 f ). (4)
Thinner deposits can be detected, but their thickness can-
not be determined, and the reection amplitudes will be
decreased by destructive interference (Widess, 1973).
While it can be shown from these equations that many sul-
de deposits meet or exceed the size requirements for both
detection and imaging (for example, a 500 m diameter 15 m
thick deposit could easily be imaged at a depth of 2 km, as-
suming a peak frequency of 100 Hz and a formation velocity of
6.0 km/s), early tests in mining camps were inconclusive (e.g.,
Dahle et al., 1985; Reed, L., 1993, Mineral Industry Technol-
ogy Council of Canada report), in part because the acoustic
properties of the sulde minerals themselves were poorly un-
derstood, but also because signicant differences in target size,
structure, and acoustics between hard and soft rock environ-
ments had not been fully taken into account. In particular, the
signal-to-noise (S/N) ratio is anomalously low in hard rock ter-
rains, the targets are typically point sources or scatterers rather
than continuous reectors, and they are often steeply dipping
(Eaton, 1999).
To assess and solve these problems, the Geological Survey
of Canada (GSC) recently embarked on a major collaborative
research program with INCO, Falconbridge, and Noranda in-
volving integrated laboratory, logging, modeling, and seismic
tests at six mining camps in Canada, including Sudbury, Kidd
Creek, and Manitouwadge in Ontario, Matagami and Selbaie
in Quebec, and Bathurst in New Brunswick. The objectives of
the laboratory studies were rst to determine the basic acoustic
properties of the major sulde minerals frommeasurements of
their compressional wave velocities (V
p
) and densities at ele-
vated pressures, and then to determine for each camp the pairs
of lithologies which might be expected to produce strong re-
ections. Following the laboratory studies, velocity and density
logs were then run in selected boreholes in each camp to de-
termine if the laboratory results persisted at formation scales.
If the results of these tests were promising, seismic modeling
based on the laboratory and logging results plus the known
geology in each camp was performed to guide the subsequent
acquisition and interpretation of seismic reection data. Verti-
cal seismic proling (VSP) and/ or 2-D and 3-D multichannel
seismic (MCS) surveys were then conducted over known ore
deposits and marker horizons in each camp using state-of-the-
art technology to map the geology at depth and determine
if the deposits themselves could actually be detected and im-
aged. While the surveys were based on oil-eld techniques,
many changes were made to accomodate the unusual condi-
tions encountered in hard rock terrains. For example, dynamite
was used as the source because of its high-frequency content,
and particular care was taken to ensure high-fold coverage
and good shot and receiver coupling to bedrock to compen-
sate for low S/N ratios. In addition, large shot-receiver offsets
or VSP techniques were used where the targets were steeply
dipping. Finally, since the targets were small, unconventional
processing sequences based on Born scattering were often
used to process the data (Eaton et al., 1997; Milkereit et al.,
1997).
The results of these studies have been spectacular. The labo-
ratory studies have provided a quantitative basis for predicting
the reectivity of an ore body of any composition in any setting
(Salisbury et al., 1996), while the seismic studies successfully
detected all of the known deposits and identied several new
targets to be tested by drilling. Preliminary seismic results have
already been published for several of these studies (Milkereit
et al., 1992, 1996; Eaton et al., 1996) and the results of many of
themorerecent surveys conductedduringthis program, includ-
ing 2-, and 3-Dsurveys in Bathurst, Manitouwadge, Matagami,
and Sudbury, were presented at Exploration 97 (Adam et al.,
1997; Eaton et al., 1997; Milkereit et al., 1997; Roberts et al.,
1997; Salisbury et al., 1997). The purpose of the present paper is
to outline the basic acoustic properties of suldes and to show,
through a case study involving laboratory, logging, and seis-
mic imaging tests at the Halfmile Lake deposit in the Bathurst
camp, how these properties govern the reectivity of massive
suldes.
ACOUSTIC PROPERTIES OF SULFIDES
While the acoustic properties of silicate rocks are well
known from decades of laboratory studies (e.g., Birch, 1960;
Christensen, 1982), until very recently the properties of suldes
were so poorly known that it was difcult to predict whether
or not they should be reectors. The principal difculty was
that while the velocities and densities of some sulde miner-
als, such as pyrite (py) and sphalerite (sph) were well known
(Simmons and Wang, 1971), the properties of other volumet-
rically and thus acoustically important minerals such as chal-
copyrite (cpy) and pyrrhotite (po) had never been measured,
making it difcult to estimate the properties of mixed sulde
deposits.
To answer this question, we measured the densities and com-
pressional wave velocities of a large suite of ore and host rock
samples of known composition in the laboratory at conning
1884 Salisbury et al.
pressures ranging from 0 to 600 MPa using the pulse transmis-
sion technique of Birch (1960). In addition to ores of mixed
composition, samples of pure py, sph, po, and cpy were mea-
sured to establish the theoretical limits of the velocity-density
eld for common ores. The results, summarized in Figure 1
from Salisbury et al. (1996) for data at a standard conning
pressure of 200 MPa (the crack closure pressure), show sev-
eral important trends:
1) As predicted from earlier studies, the velocities of the
host rocks increase with density along the Nafe-Drake
curve for silicate rocks (Ludwig et al., 1971).
2) The suldes, however, lie far to the right of the Nafe-
Drake curve in a large velocity-density eld controlled by
the properties of pyrite, which is fast and dense (8.0 km/s,
5.0 g/cm
3
), pyrrhotite, which is very slow and dense
(4.7 km/s, 4.6 g/cm
3
), and sphalerite and chalcopyrite
which have intermediate and very similar velocities and
densities (5.5 km/s, 4.1 g/cm
3
).
3) The properties of mixed and disseminated suldes lie
along simple mixing lines connecting the properties of
end-member suldes and felsic or mac gangue. Thus, for
example, velocities increase dramatically with increasing
pyrite content, but they actually decrease with increas-
ing sphalerite, chalcopyrite, or pyrrhotite content along
trends which can be calculated using the time-average
relationship of Wyllie et al. (1958).
4) If lines of constant acoustic impedance(Z) corresponding
to the average impedances of mac and felsic rocks are
superimposed on the sulde elds as in Figure 1, it is clear
that most sulde ores have higher impedances than their
felsic or mac hosts. Since a reection coefcient of 0.06
FIG. 1. Velocity (V
p
)-density elds for common sulde ores
and silicate host rocks at 200 MPa. Ores: py =pyrite,
cpy =chalcopyrite, sph =sphalerite, po =pyrrhotite. Silicate
rocks along Nafe-Drake curve: SED=sediments, SERP= ser-
pentinite, F=felsic, M=mac, UM=ultramac, g =gangue,
c =carbonate. Dashed lines represent lines of constant acous-
tic impedance (Z) for felsic and mac rocks. Bar shows mini-
mum impedance contrast required to give a strong reection
(R =0.061).
is sufcient to give a strong reection, then in principle,
massive suldes with the appropriate geometry should be
strong reectors in many common geologic settings. For
example, massive pyrrhotite should be readily detectable
in felsic settings, and any combination of sphalerite, chal-
copyrite, and pyrite should be a strong to brilliant reec-
tor in most mac and felsic settings, depending on the
pyrite content.
CASE STUDY: SEISMIC REFLECTIONS FROM THE
HALFMILE LAKE DEPOSIT, BATHURST
While the laboratory results presented above showthat mas-
sive suldes should often be strong reectors, it is also clear
that the reectivity of any given deposit will be strongly inu-
enced by local conditions, such as the size and conguration
of the deposit, its actual mineralogy, and the composition and
metamorphic grade of the country rock. Thus to evaluate the
method, it was necessary to study the actual seismic response
of several deposits at different scales of investigation.
The Bathurst mining camp in New Brunswick was selected
for one of these studies because it provides an excellent oppor-
tunity to examine the seismic response of volcanogenic mas-
sive sulde deposits in low-grade metamorphic settings. The
results, which were obtained over the Halfmile Lake deposit,
the largest undeveloped deposit in the camp, provide a par-
ticularly clear example, not only of reections from a massive
sulde deposit, but of the factors which must be taken into
account in imaging these bodies.
Geology of the Halfmile Lake deposit
The Halfmile Lake deposit is an extensive massive sulde
sheet which ranges from 1 to 45 m in thickness and contains
26 million tons of total suldes. Mineralization is characterized
by py-po-rich layered suldes and po-rich breccia-matrix sul-
des withvariableamounts of sphalerite, galena, andarsenopy-
rite. The deposit is hosted by a thick sequence of turbidites,
felsic volcanic and epiclastic rocks, argillites, and intermediate
volcanic rocks of Cambro-Ordovician age (Adair, 1992). Four
periods of fold deformation are documented in the deposit
area and are accompanied by faulting. The deposit occurs on
the overturned south limb of a large antiform (Figure 2). The
sulde sheet extends 3 km along strike (Figure 2) and has been
drilled to a depth of 1.2 km. It is structurally overlain (strati-
graphic footwall) by a stringer zone containing between 5 and
30% stringer po and cpy (Figure 3). The overall dip of the
stratigraphy is 45
to the north-northwest, but structural inu-

ences locally steepen dips to near vertical. Metamorphism is
greenschist facies, and the sheet displays 100 m of topography.
Hosted within the sulde sheet are two signicant build-ups
of massive suldes in the Upper and Lower Zones (Figure 3),
where the thickness of the suldes ranges between 5 and 45 m.
The most important and largest undeveloped deposit in the
camp is the Lower Zone, which contains 6.1 million tons of
9.7% Zn, 3.34% Pb, 0.1% Cu, and 43 g/ton Ag. The Halfmile
Lake deposit and specically the Lower Zone were selected
for this study because it had been carefully mapped and de-
lineated by drilling, core was available for laboratory stud-
ies, drillholes were still open for logging and VSP tests, the
size and dip of the deposit seemed appropriate for detection
using 2-D surface seismic techniques and, since it had never
been mined, there would be no spurious reections from mine
workings.
Laboratory impedance measurements
To determine which lithologies were potential reectors at
Halfmile Lake, velocities and densities were measured in the
laboratory at elevated pressures on minicores cut from 28 sur-
face anddrill core samples representing all of the major ore and
host rock lithologies along the seismic line. Since sedimentary
and metamorphic rocks are often anisotropic, velocities were
measured parallel and perpendicular to bedding, banding, or
foliation in many samples, bringing the total number of mea-
surements to 53. The results, presented at a conning pressure
of 200 MPa in Figure 4, show that the host rocks all have very
similar average velocities and densities (about 6.0 km/s and
2.75 g/cm
3
). This initially surprising result is due to the fact that
the felsic igneous rocks (rhyolite, quartz porphyry, tuff) and the
metasediments all have very similar compositions, while the
mac rocks are actually basaltic andesites of intermediate com-
position in which the velocities have been depressed by the
alteration of mac minerals to chlorite by greenschist facies
metamorphism. The Halfmile Lake ores, on the other hand,
display a wide range of velocities (5.17.3 km/s) due to vary-
FIG. 2. Geologic map of Halfmile Lake deposit showing location of 2-D seismic line presented in Figure 7. Geologic cross-section
shown in Figure 3 extends from station 1 to station 166 (101660 m along line). Geophysical logging was conducted in holes HN
94-63 (Figure 5) and HN94-65. Offset VSP shown in Figure 6 was conducted in hole HN92-30. Dashed arrowshows axis and plunge
of F1 antiform. Inset shows regional setting of Halfmile Lake deposit.
ing proportions of po and py. Interestingly, the iron formation
sample plots in the velocity-density eld for the suldes due to
the high intrinsic velocity and density of magnetite (7.4 km/s,
5.2g/cm
3
). As aconsequence, theimpedancecontrasts between
the various country rock lithologies at Halfmile Lake should
be small, while the contrast between any of the ores and the
country rock will be very large. Since an impedance contrast of
2.5 is sufcient to give a strong reection, this implies that at
least in terms of their acoustic properties, the ores at Halfmile
Lake should be strong reectors in a virtually transparent host
rock setting.
Geophysical logging
Once the laboratory measurements had been completed,
two boreholes through the deposit (holes HN 94-63 and HN
94-65 in Figure 3) were logged by the GSC Mineral Resources
Division using slim-hole sonic velocity and density tools to de-
termine if the impedance differences measured in the labora-
tory persist at formation scales. The results, presented in Fig-
ure 5 for the deeper of the two holes, show that the densities
range narrowly from 2.6 to 2.75 g/cm
3
throughout the silicate
rocks, but increase erratically to values as high as 2.95 g/cm
3
in
the stringer zone, then dramatically to about 3.4 g/cm
3
in the
massive suldes, while negative excursions correspondtofaults
markedby thinintervals of fault gouge or breccia. Similarly, the
velocity log varies from 5.5 to 6.0 km/s throughout most of the
hole, withnegativespikes correspondingagaintonarrowfaults.
Signicantly, the velocity-density product, or impedance log,
shows very little variation about a mean of about 15, except for
the ores, which reach values of about 20, and the faults, which
reach values as low as 11. While the absolute values of the ve-
locity and impedance logs are lower than the laboratory results
due todifferences inpressure, the impedance contrasts are sim-
ilar to those predicted from laboratory studies, implying that
the massive suldes will be strong reectors, while the country
rock will be transparent except possibly where cut by faults.
Vertical seismic proling
While the laboratory and logging results were promising,
they were not denitive because they were obtained at much
higher frequencies than seismic surveys (1 MHz and 50 KHz
versus 10200 Hz) and the propagation paths were much
shorter (25 cm and 12 m versus a few kilometers). VSP pro-
vides a more convincing test because the method can be used
to determine if reections are actually generated at seismic fre-
quencies and if they are sufciently strong to be detected over
propagation paths of 1 km or more. To this end, an offset VSP
survey was conducted in borehole HN 92-30 by the University
of Alberta using 350 g Pentolite charges in a series of shallow
FIG. 3. Simplied geologic cross-section through Halfmile Lake deposit based on drilling results projected onto seismic line between
stations 1 and 166. UZ=Upper Zone, LZ=Lower Zone. VSP survey (Figure 6) was conducted in hole HN 92-30 (shots at X) and
logging was conducted in holes 94-63 and 94-65. No vertical exaggeration.
FIG. 4. Average compressional wave velocity (V
p
) at 200 MPa
versus density for ore and host rock samples fromthe Halfmile
Lake deposit superimposed on velocity-density elds for sul-
des and silicate rocks shown in Figure 1. Also shown are
lines of constant acoustic impedance for mac rocks (Z =20)
and felsic rocks (Z =17.5). Impedances of Halfmile Lake ores
(ellipse) are much greater than those of their silicate hosts.
(36 m) holes drilled to the top of basement about 100 m north
of hole HN 92-30 (X in Figure 3) and a 3-component borehole
seismometer clamped at 5 m intervals as the tool was raised
from a depth of 575 m to the surface.
Analysis of rst arrival traveltimes shows that the crust has
an average P-wave velocity (5.55 km/s) consistent with the
logging data, while the reection results, presented in Figure 6
after routine processing (upgoing waveeld separation, high-
pass 50290 Hz lter, deconvolution, scaling, rst break mute)
and transformation to geometric coordinates using the CDP
transform method (Wyatt and Wyatt, 1984; Hardage, 1985;
Kohler and Koenig, 1986), show a prominent, north-dipping
reector between 300 and 400 m depth which corresponds to
the massive sulde deposit south of the borehole (Figure 3). In-
terestingly, a deeper sulde horizonwas alsoimagedat the base
of the hole. As predicted fromthe laboratory and logging stud-
ies, no other contacts in the immediate vicinity have sufciently
large impedance contrasts to produce reections, and the faults
are toothintodetect. Fromthe results of this test, it is thus clear
that the Halfmile Lake deposit generates strong reections at
seismic frequencies, these reections propagate for signicant
distances in basement, and the deposit can be readily imaged
using borehole VSP techniques and conventional sources.
FIG. 5. Density, velocity (V
p
) and calculated impedance versus
depth in hole HN 94-63 from geophysical logging. Fault gouge
in core indicated by dashed lines. Note high impedance of mas-
sive suldes.
Multichannel seismic proling
While obviously successful, the borehole VSP survey con-
ducted at Halfmile Lake did not prove that 2-, or 3-D surveys
conductedat the surface wouldbe successful. Despite the many
similarities between the two techniques, signicant differences
still remain: The source to receiver paths are shorter for VSP
surveys, the S/N ratio is improved by clamping the receiver in
basement rather than placing it at the surface, and the over-
burden path is eliminated.
The denitive test of the seismic reection method is thus to
conduct a 2-, or 3-Dsurvey over the deposit fromthe surface as
if no boreholes were available. To this end, a 2-D multichannel
survey was conducted over the deposit along a 5.85 km line
which intersects the deposit at its southern end and extends
downdip for a considerable distance to the north (Figure 2),
withdownholecontrol providedbytheVSPandloggingresults.
Thesurveywas conductedbyEnertec Geophysical Services un-
der contract to Noranda using 340 g Pentolite charges in holes
drilled to basement every 40 malong the line; a portable, state-
of-the-art, 480-channel, 24-bit recording system; and 14 Hz re-
ceivers laid out along three parallel lines spaced 50 m apart:
a center line with groups every 10 m, 9 phones/station, giving
30 fold coverage, and a line to either side with a 20 m group
spacing, 9 phones/station, giving 15 fold coverage.
Although shooting conditions were difcult and the eld
records were very noisy, careful processing involving static cor-
rections, scaling, the application of a high-pass lter, deconvo-
lution, common-midpoint binning, stacking velocity analysis,
noise suppression, and poststack scaling gave a clear image of
the ore body (Figure 7) which coincides with the known lo-
cation of the deposit after migration. As in the VSP survey,
FIG. 6. CDP transform of VSP survey in borehole HN 92-30
showing reection from Halfmile Lake deposit (arrow). X in
Figure 3 shows location of shots. Reections are also observed
from a thin sulde layer between 550 and 600 m (see Figure 3).
No vertical exaggeration.
the country rock was weakly reective to virtually transparent
along the rest of the line. The results at Halfmile Lake thus
demonstrate not only that massive suldes can be detected us-
ing surface seismic reection techniques but that nothing else
in the immediate vicinity of the line causes strong reections.
Thus at Halfmile Lake, seismic reection provided relatively
little new information about deep structure, but the method
appears to be almost ideal for exploration.
CONCLUSIONS
From the results of this study, it is clear that massive suldes
can make strong reections in hard rock settings and that they
can be detected using state-of-the-art seismic reection tech-
niques. This has major implications, both for exploration in the
Bathurst camp itself and for the mining industry as a whole.
Sincetherst major discoveryinBathurst morethan40years
ago, more than 100 base metal deposits, 35 with dened ton-
nage, have been identied in the upper few hundred meters
FIG. 7. Unmigrated 2-D multichannel seismic image of the Halfmile Lake deposit. TWT=two-way traveltime.
No vertical exaggeration.
of basement in the camp using surface mapping and potential
eld techniques. As in many camps, however, the known shal-
low reserves are declining at Bathurst, forcing exploration to
deeper levels. Statistically, there should be just as many de-
posits per unit volume at greater depths in the camp, but these
have proven difcult to nd because the surface geology is too
complex to project to depth and potential eld methods lose
their resolution at depths greater than a few hundred meters.
Seismic reection, however, appears tobeanideal tool for deep
exploration at Bathurst and elsewhere because it has high res-
olution to the full depth limits of mining (3 km) and because
it can scan large volumes of crust at a cost which is low com-
pared to deep wildcat or delineation drilling (Pretorius et al.,
1997). While conditions at Bathurst appear to be particularly
suitable for seismic exploration, recent experimental surveys
over deposits inquite different settings suchas Matagami, Kidd
Creek, and Sudbury (Adam et al., 1997; Eaton et al., 1997;
Milkereit et al., 1997) are equally encouraging, suggesting that
high-resolution seismic reection techniques can be modied
to meet the deep exploration needs of the mining industry in
the new millennium.
ACKNOWLEDGMENTS
We thankR. Iuliucci fromthe Dalhousie/GSCHighPressure
Laboratory for his assistance in measuring the acoustic proper-
ties of the Bathurst samples, the members of the GSC Mineral
Resources Division logging crew for obtaining the borehole
geophysical logs presented in this study, and the members of
the University of Alberta borehole geophysics eld crew for
their assistance in conducting the VSP survey. Special thanks
are due to Kevin Coin from GSC Atlantic for his assistance
in the eld during the VSP survey. Finally, the MCS survey at
Bathurst would not have been possible without the consider-
able assistance of the Enertec Geophysical Services eld team
and Noranda staff from Bathurst.
REFERENCES
Adair, R. N., 1992, Stratigraphy, structure and geochemistry of the
HalfmileLakemassivesuldedeposit, NewBrunswick: Explor. Min.
Geol., 1, 151166.
Adam, E., Arnold, G., Beaudry, C., Matthews, L., Milkereit, B., Perron,
G., and Pineault, R., 1997, Seismic exploration for VMS deposits,
Matagami, Quebec, in Gubins, A., Ed., Proc. Exploration 97: 4th
Dec. Internat. Conf. on Mineral Expl., 433438.
Berryhill, J. R., 1977, Diffraction response for non-zero separation of
source and receiver: Geophysics, 38, 11761180.
Birch, F., 1960, The velocity of compressional waves in rocks to 10 kilo-
bars, 1: J. Geophys. Res., 65, 10831102.
Christensen, N. I., 1982. Seismic velocities, in Carmichael, R. S., Ed.,
Handbook of physical properties in rocks, vol. II: CRC Press,
1228.
Dahle, A., Gjoystdal, H., Grammeltdveldt, G., and Hansen, T., 1985,
Applicationof seismic reectionmethods for oreprospectingincrys-
talline rock: First Break, 3, 916.
Debicki, E., 1996, MITECs ExplorationTechnology Division: Helping
reverse the trend of declining mineral reserves in Canada: Canadian
Inst. Min. Metal. Bull., 89, 5359.
Eaton, D., 1999, Weak elastic wave scattering from massive sulde ore
bodies: Geophysics, 64, 111.
Eaton, D., Guest, S., Milkereit, B., Bleeker, W., Crick, D., Schmitt, D.,
and Salisbury, M., 1996, Seismic imaging of massive sulde deposits:
Part III. Borehole seismic imaging of near-vertical structures: Econ.
Geol., 91, 835840.
Eaton, D., Milkereit, B., and Adam, E., 1997, 3-D seismic exploration,
in Gubins, A., Ed., Proc. Exploration 97: 4th Dec. Internat. Conf. on
Mineral Expl., 6578.
Enachescu, M., 1993, Amplitude interpretation of 3-D reection data:
The Leading Edge, 12, 678683.
Hardage, R. A., 1985, Vertical seismic proling, part A: Principles:
Geophysical Press.
Kohler, K., and Koenig, M., 1986, Reconstruction of reecting struc-
tures fromvertical seismic proles withamovingsource: Geophysics,
51, 19231938.
Ludwig, J., Nafe, J., andDrake, C., 1971, Seismic refraction, inMaxwell,
A. E., Ed., The sea, v. 4: Wiley, 5384.
Milkereit, B., Berrer, E. K., Watts, A., and Roberts, B., 1997, Devel-
opment of 3-D seismic exploration technology for Ni-Cu deposits,
Sudbury basin, in Gubins, A., Ed., Proc., Exploration 97: 4th Dec.
Internat. Conf. on Mineral Expl., 439448.
Milkereit, B., Eaton, D., Wu, J., Perron, G., Salisbury, M., Berrer,
E., and Morrison, G., 1996, Seismic imaging of massive sulde de-
posits: Part II. Reection seismic proling: Econ. Geol., 91, 829
834.
Milkereit, B., Reed, L., and Cinq-Mars, A., 1992, High frequency re-
ection proling at Les Mines Selbaie, Quebec: Can. Geol. Surv.
Curr. Res., 92-1E, 217224.
Pretorius, C. C., Trewick, W. F., and Irons, C., 1997, Application of 3D
seismics to mine planning at Vaal Reefs gold mine, number 10 shaft,
Republic of South Africa, in Gubins, A., Ed., Proc. Exploration 97:
4th Dec. Internat. Conf. on Mineral Expl., 339408.
Roberts, B., Zaleski, E., Adam, E., Perron, G., Petrie, L., Darch, L.,
Salisbury, M., Eaton, D., and Milkereit, B., 1997, Seismic exploration
of the Manitouwadge greenstone belt, Ontario, in Gubins, A., Ed.,
Proc. Exploration 97: 4th Dec. Internat. Conf. on Mineral Expl.,
451454.
Salisbury, M. H., Milkereit, B., Ascough, G. L., Adair, R., Schmitt, D.,
and Matthews, L., 1997, Physical properties and seismic imaging of
massive sulphides, in Gubins, A., Ed., Proc. Exploration 97: 4th Dec.
of massive sulde deposits: Part I. Rock properties: Econ. Geol., 91,
821828.
Simmons, G., and Wang, H., 1971, Single crystal elastic constants and
calculated aggregate properties: M.I.T. Press.
Weatherby, B., 1940, The history and development of seismic proling:
Widess, M., 1973, How thin is a thin bed?: Geophysics, 38, 11761180.
Wyatt, K. B., and Wyatt, S. B., 1984, Determining subsurface struc-
ture using the vertical seismic prole, in Toksoz, M. N., and Stewart,
R. R., Eds., Vertical seismic proling, part B: Advanced concepts:
Geophysical Press, 148176.
Wyllie, M., Gregory, A., andGardner, G., 1958, Anexperimental inves-
tigation of factors affecting elastic wave velocities in porous media:
Development of 3-D seismic exploration technology for deep
nickel-copper depositsA case history
from the Sudbury basin, Canada
Bernd Milkereit
, E. K. Berrer
, Alan R. King
, Anthony H. Watts
,
B. Roberts
, Erick Adam
, David W. Eaton
, Jianjun Wu
,
and Matthew H. Salisbury
ABSTRACT
Following extensive petrophysical studies and presite
surveys, the Trill area of the Sudbury basin was selected
for conductingtherst 3-Dseismic surveyfor mineral ex-
plorationinNorthAmerica. The 3-Dseismic experiment
conrms that in a geological setting such as the Sudbury
Igneous Complex, massive sulde bodies cause a char-
acteristic seismic scattering response. This provides an
excellent basis for the direct detection of massive sul-
des by seismic methods. The feasibility study suggests
that high-resolution seismic methods offer a large de-
tection radius in the order of hundreds to thousands of
meters, together with accurate depth estimates.
INTRODUCTION
The Sudbury structure (Figure 1) is located at the erosional
boundary between the Archean Superior province and the
overlying sequence of early Proterozoic continental margin
deposits. The structure consists of the Sudbury Igneous Com-
plex (SIC), a differentiated sequence of norite, gabbro, and
granophyre overlain by breccias and metasedimenary rocks.
The Sudbury basin is thought to be associated with a large
impact event (e.g., Boerner et al., 1994). The Sudbury basin is
the richest nickel-producing area in the world, with signicant
by-products in copper and precious metals. The basin hosts
Kiel University, Department of Geophysics, 24118 Kiel, Germany. E-mail: bernd@geophysik.uni-kiel.de.

Inco Ltd., Exploration, Copper Cliff, Ontario P0M 1N0, Canada. E-mail: berrer@isys.ca; kinga@inco.com.
Falconbridge Ltd., Exploration, Toronto, Canada. E-mail: anthonywatts@compuserve.com.

Geological Survey of Canada, Ottawa, Canada. E-mail: broberts@cg.nrcan.gc.ca; matts@agc.bio.ns.ca; eadam@cg.nrcan.gc.ca.
Formerly Geological Survey of Canada, Ottawa, Canada; presently University of Western Ontario, Department of Earth Sciences, London,
Ontario, N6A 5B7, Canada. E-mail: deaton@julian.uwo.ca.
Formerly Geological Survey of Canada, Ottawa, Canada; presently Veritas, 715 5th Avenue S.W., Calgary, Alberta T2P 5AZ, Canada. E-mail:
jj wn@veritasdgc.com.
numerous mines along the outer rim of the SIC. In 1995, the
mines produced 170,000 metric tons of nickel and 160,000
metric tons of copper. This industry is a signicant economic
resource for Canada. The discovery of new deposits is vital for
the continueduse of the infrastructure of this important mining
district. In the early part of the twentieth century, exploration
was basedonsurface prospecting andgeology. Soonthereafter,
explorationreceivedassistancefrommagnetic surveys employ-
ing dip-needle instruments. Starting in the late 1940s, ground
electromagnetic methods were used to locate conductive
sulde deposits which contained nickel ore. With these instru-
ments, the highly-conductive sulde bodies containing mainly
pyrrhotite could be detected to a depth of about 100 m. Like
the oil and gas industry, the mining industry is looking progres-
sively deeper for newore deposits to replace declining reserves
at existing mines. This requirement for deepexplorationis driv-
ing the search for new technology to reduce costs and increase
drilling success rates. With modern large-loop electromagnetic
systems, the depth of exploration is extended up to 400 m
depending on the size of the target. It is highly likely that more
sulde nickel deposits will exist at greater depths, near the SIC
footwall contact. Therefore, there is great interest in locating
these deposits at depths up to about 2500 m, the depth limit
to which modern mining methods are capable of extracting
the ore.
Adapting seismic reection methods to hard-rock geology
may extend the depth range of exploration (e.g., Milkereit
et al., 1996a). Although recognized as having both the
depth penetration and resolving power required for deeper
1890
3-D Seismic Exploration Technology 1891
exploration, seismic prolinghas oftenbeenviewedbythemin-
ing industry as cost-ineffective, in part because of the paucity
of strong, unambiguous reections that typify a sedimentary
basin and in part because of a general lack of understanding of
howobserved signals correlate with subsurface features. In this
paper, we present a case history for the evaluation of 3-D seis-
mic exploration technology. We review the physical rock prop-
erties of massive suldes, present borehole geophysical data,
and discuss the seismic reection response of deep-seated ore
deposits. Examples illustrate various stages of the feasibility
study from survey design to heliportable data acqusition, and
from special processing considerations to nal interpretation
of the 3-D seismic data volume.
Geological setting
The Sudbury Ni-Cu deposits occur at the moderately to
steeply dipping lower contact of the SIC. Figure 1b shows a
simplied model of the crater oor derived from drilling and
FIG. 1. (a) Location map of Sudbury structure, Ontario,
Canada, with borehole locations and 3-D seismic survey area.
Norite, gabbro, and granophyre of the Sudbury Igneous Com-
plex (SIC) indicated by gray pattern. Stars indicate borehole
locations. (b) Geological model of the crater oor from drill
and mining operations with massive sulde deposits located
beneath the lower contact of the SIC [cross-section modied
after Morrison (1984)] 1 = norite, 2 = sublayer, suldes, and
breccia, 3 = footwall complex.
mining operations. At the base of the SIC, the sublayer is a
host for the ore. The sublayer consists of a mass of basic to ul-
trabasic inclusions of varying size and frequency of occurrence
in a matrix of norite and suldes. When the suldes are suf-
ciently concentrated, this zone constitutes the ore (Morrison,
1984). Although results from 2-D reconnaissance seismic pro-
les (Milkereit et al., 1992, 1996a) acquired across the Sudbury
basin have provided important information on gross structure
and regional geological setting, there have been problems in-
tegrating this new mapping technique into normal exploration
procedures. For example, borehole geophysical logs (i.e., den-
sity and sonic velocity), important for the interpretation and
calibration of reection seismic data, have not been routinely
gathered in existing slim diamond-drill holes.
Over the past ve years, an extensive database of geolog-
ical mapping information, geophysical logs of existing deep
drill holes, and core samples for physical rock property studies
has been assembled to support interpretation of the early 2-D
seismic data across the Sudbury structure. In-situ logging and
borehole seismic experiments have been conducted at multiple
locations within the SIC using full-waveform sonic and density
logging tools. Small-diameter holes (NQ, BQ) were drilled to
depths of about 2000 m, providing superb control on the ori-
gin of reections and the basic velocity structure of the SIC.
Figure 2 shows density and velocity logs aligned along the SIC-
footwall contact. A narrow range of velocities characterize the
SICnorite, ranging from6200 to6400 m/s, whereas the sublayer
and footwall complex display more extreme velocity variations
(60006700 m/s). Density logs follow the same trend as the
P-wave velocities: relatively uniform densities between 2.75
and 2.8 g/cm
3
are associated with the norite, while sublayer and
footwall complex densities scatter between 2.75 and 3.0 g/cm
3
.
There appears to be no variation of the physical properties lat-
erally within the SIC. In terms of seismic exploration, the con-
tact between the transparent SIC norite and the reective
footwall complex represents a clear regional marker, making
it possible to map the bottom of the SIC.
Physical rock properties of massive suldes
Sedimentary basins, where seismic techniques have been
usedmost extensively, differ fromthe crystalline crust ina num-
ber of characteristics that are signicant for seismic acquisition
and processing (Milkereit and Eaton, 1998). Unlike most sed-
imentary basins, the crystalline crust often lacks pronounced
horizontal continuity of prominent seismic reectors (reec-
tion coefcients rarely exceed 0.1). In addition, the velocity
structure of young sedimentary basins is characterized by low
compressional-wave velocities and relatively strong velocity
gradients. On the other hand, crystalline crust is characterized
by uniformly high compressional velocities, especially below
200 m where fracturing is less prevalent. For the analysis of
seismic reectivity, acoustic impedance becomes the most
important parameter. In-situ studies in deep boreholes con-
rm that signicant impedance contrasts exist at the contacts
between major lithological units of the SIC and the footwall
complex.
Reected and scattered seismic events occur at the contact
between rocks with contrasting elastic properties (e.g., com-
pressional wave velocity, Poissons ratio and, density). Fig-
ure 3 shows the velocity-density eld for common silicate rocks
ranging fromunconsolidated sediments to dense, high-velocity
1892 Milkereit et al.
ultramacs. The velocity-density eld for suldes (includ-
ing the pyrrhotite, chalcopyrite, millerite, niccolite, and pent-
landite assemblages typical for ores in Sudbury) is distinctly
different from that for common silicate rocks (such as norite in
the hanging wall and gneiss in the footwall). Composite ma-
terial, containing various suldes and silicate rocks, is gov-
erned by simple linear mixing rules (Salisbury et al., 1996).
Thus, the presence of sulde minerals will tend to dene a
unique velocity-density eld. Depending on grade and type,
theSudburymassive-suldedeposits becomestrong seismic re-
ectors based on impedance contrast with unmineralized host
rocks. For regional seismic exploration programs, the contact
between the reective footwall complex and the transparent
norite unit of the SIC represents the most prominent regional
seismic marker. For mineral exploration, high-density massive
suldes located at or close to embayment structures at the
norite-footwall contact will produce a strong seismic reec-
tion or scattering response (Milkereit et al., 1996a). Density
measurements from core samples conrm that sulde samples
from the 3-D seismic study area occupy the distinct density
eld (Milkereit et al., 1997).
Seismic modeling of deep massive sulde deposits
High-density mineralization introduces new problems for
modeling, acquisition, processing, and interpretation of seis-
mic data. In seismic modeling, ray tracing is most effective in
discrete layers that are separated by smooth interfaces. Ray
theory breaks down on highly irregular interfaces, as might be
the case for high-impedance bodies and other geological fea-
tures in the shallow crystalline crust. Other techniques, such as
a direct solution of the elastic wave equation by nite differ-
ences, are currently impractical for routine 3-D seismic mod-
eling. In this study, we apply the 3-D Born-approximation
method of Eaton (1996). The Born approximation method is
designed to handle geological situations where short wave-
length impedance anomalies (such as ore bodies) are super-
FIG. 2. Summary sketch of velocity and density logs from the Sudbury structure aligned at the contact between norite and footwall
complex. Borehole locations are shown in Figure 1.
imposed onto a smoothly varying background medium. Con-
sider the case of a dipping-lens model consisting of a 200 m
wide, disk-shaped high-impedance unit at 1650 m depth. The
background velocity is 6000 m/s. Figure 4 shows the zero-
offset scattering response from such a feature, inclined at
45
. The unmigrated data shows the region over which the

diffracted energy is visible is considerably larger than the
actual size of the scattering body. In the unmigrated data,
scattered energy is concentrated primarily in the downdip
direction.
Incontrast toreections fromcontinuous interfaces, thescat-
tering response from local heterogeneities remains stationary
(i.e., the traveltime response is symmetric withrespect tothe lo-
cation of the scatterer in the 3-Dcube) (Milkereit et al., 1996b).
The symmetric scattering response is best detected in a series
of time slices (Figures 4b, c). It is worth noting that for dipping
FIG. 3. Velocity-density elds for silicate rocks and sul-
des/oxides with lines of constant seismic impedance (modi-
ed after Salisbury et al., 1996). SED = sediments, F = felsic,
M= mac, UM= ultramac.
FIG. 4. 3-D modeling of scattering event caused by a local
high-impedance contrast (a dipping lens at 1650 m depth
and a background velocity of 6000 m/s). (a) In-line section.
Note the asymmetric zero-offset response. (b, c) Time slices
of zero-offset response. Note enhanced amplitudes in the
downdip direction (M). A= apex.
impedance contrasts, the strongest reection/scattering devel-
ops at large offsets in the dip direction, away from the deposit.
The directional characteristics of the scattering response are
important for the acquisition, processing, and interpretation of
3-D seismic data for massive suldes. Because of the inherent
limitations of conventional seismic modeling methods, some
of the strike- and dip-related characteristics of small seismic
energy scatterers may not have been fully appreciated in
the past (most conventional 2-D seismic reconnaissance
proles may have overlooked or underestimated the signif-
icance of local bright seismic diffractions caused by density
anomalies).
Special problems posed by the hard-rock environment
In the past, many tests of high-resolution seismic imaging
methods using off-the-shelf technology from hydrocarbon
exploration have had limited success in hard-rock environ-
ments. High levels of source-generated noise in the 1030 Hz
range canoverwhelmthe relatively weaksignals (Milkereit and
Eaton, 1998). The presence of the low-frequency noise window
and the need to resolve small features at depth demand the use
of a source spectrum that is shifted more towards higher fre-
quencies than those commonly used for seismic proling in
sedimentary basins. This fundamental difference must be con-
sidered in the design of hard-rock seismic surveys in order to
achieve success.
Until recently, the vast majority of seismic surveys for min-
eral exploration have used 2-D proling, where source and re-
ceiver locations are colinear for the basic acquisition geometry.
This approach, which essentially yields vertical cross-sectional
images of the subsurface, is most effective where major struc-
tural elements have a well-dened strike direction and the
2-D seismic proles can be oriented perpendicular to struc-
tural trends. In cases where subsurface structures do not have
a well-dened strike and dip direction, out-of-plane reections
(sideswipe) can seriously contaminate 2-D reection images,
producing false structural images. Why should 3-D reection
seismic surveys be considered for mineral exploration? The
spatial dimensions of economically viable targets embedded
in complex crustal structures make them less likely to be de-
tected in the course of reconnaissance 2-D seismic work. In
practice, spatial aliasing and the number of recording channels
pose some limitations on the design of 3-D seismic surveys.
For example, the use of high seismic frequencies requires small
separation of sensors to avoid spatial aliasing, the low signal-
to-noise ratio in the crystalline environment requires digital
recording equipment with large dynamic range and high stack-
ing fold, andthe needfor large source-receiver offsets demands
simultaneous recording of hundreds of sensors. Although high-
frequency seismic data promise improved resolution, there are
implicit complications that must be overcome in the hard-rock
environment (Milkereit and Eaton, 1998). Special attention
must be given to data acquisition and processing if the band-
width and resolution of such data are to be preserved. Unfa-
vorable near-surface conditions at the source or receiver may
attenuate the high frequencies. Thus, high frequencies of about
100 Hz must be generated by the seismic sources, recorded by
the receiver grid, and be preserved throughout the processing
sequence.
THE TRILL SITE 3-D SEISMIC EXPERIMENT
Results from 2-D proling, physical rock property studies,
and forward modeling provided strong support for the hypoth-
esis that high-density mineralization (for example, a massive
sulde deposit) is detectable by its characteristic scattering re-
sponse in a 3-Dseismic data set. In the spring of 1995, it was de-
cided to conduct the rst 3-D seismic survey for deep mineral
exploration in Canada. The relatively low-noise background
levels (from man-made infrastructures) led to the selection of
the Trill area (Figure 1a) for the 3-D seismic feasibility study
to detect and delineate deep massive suldes. With the avail-
ability of a well-dened target at about 1800 m depth and a
substantial borehole data base, enough information is avail-
able to evaluate the performance of this new exploration tech-
nology. In addition, the 3-D seismic data could provide new
information about the deep geology and associated mineral
deposits of the Trill structure. The objectives of the 3-D sur-
vey at Trill were threefold: (1) to image the steeply dipping
contact between norite and footwall complex, (2) to detect a
seismic response of known massive suldes at depth, and (3)
to integrate physical rock property, seismic and drill hole data.
Survey design and eld operations
Early 2-D seismic proles were acquired with vibroseis
sources along existing roads or trails (Milkereit et al., 1992;
1996a). The Trill area is characterized by steep hills, extensive
swamps, and limited access. For 3-D seismic surveys in such a
remote area, it is not possible to rely on vibroseis sources. Thus,
the3-Dseismic surveyhadtousesmall explosivecharges. Bore-
hole geophysical logging, vertical seismic proling (VSP), and
seismic source strength evaluations were conducted at the site
(Milkereit et al., 1997). Extensive tests were required to opti-
mize source parameters (i.e., bandwidth and dynamic range)
for off-road 3-D seismic exploration programs. Small explo-
sive charges (from 0.05 to 1.0 kg) in shallow boreholes were
used as the sources for VSP and noise spread surveys. The
seismic data recorded by stationary surface geophone spreads
and downhole three-component seismometers were analyzed
for bandwidth, dynamic range, and geometrical spreading of
the signal. Source depth tests were required to evaluate wave-
form consistency and to minimize source-generated noise (in
particular shear-wave energy).
The important contact between the norite of the SIC and
the footwall complex dips 30
60
. From surface and borehole

data, it is difcult to dene a single strike direction for the
study area. Massive suldes are located in an embayment in
the footwall. In map view, the top of the deep mineralization is
located at about 1800 m depth in the center of the survey area.
A detailed 3-D forward modeling study was conducted to ad-
dress the technological challenges of detecting and delineating
ore in a complex geological setting between 1 and 3 km depth.
At the same time, a geographic information system (GIS) was
built for the study areas. Special emphasis was placed on the
availability of digital topography, surface geological maps, and
borehole data (depth to footwall contact) to conduct a detailed
simulation of the proposed 3-D seismic experiment. In paral-
lel, environmental assesssment/impact studies were conducted,
and line cutting, drilling, and seismic data acquisition contrac-
tors were selected.
The nal 3-D survey design was based on detailed 3-D for-
wardmodelingof theTrill areaandoperational constraints. The
GIS data base supported the presurvey 3-D modeling of the
Trill area through a simple 3-D geological subsurface model.
We used the 3-D Born modeling approach to accommodate
the complex shape of the contact between SIC and footwall
complex, known mineralization at 1800 m depth, and source-
receiver geometries usedfor theactual seismic experiment. The
resulting synthetic data were processed and migrated to obtain
a realistic approximation to the anticipated seismic processing
sequence. Synthetic models indicated that steep dips and deep
targets required additional shot points at the southeastern cor-
ner of the study area. In addition, modeling of the scattering
response of deepmassive suldes predicteda unique amplitude
response (Figure 4). As a result, the design of the seismic data
acquisition program was modied. Additional shot points had
to be placed to compensate for irregular source and receiver
line spacing due to swamps, lakes, and the rough topography.
The Trill study area with 12 receiver lines and 14 source lines is
shown in Figure 5a. The receiver lines are oriented east-west,
perpendicular to the grid of shot lines. Details about the ac-
quisition parameters are given in Table 1.The recording crew
began the project in late October 1995. Due to heavy rain-
fall, many of the receiver lines became difcult to walk, and
water levels in the swamps and creeks were rising. The sur-
veying crews were forced to use hip waders, canoes, and boats
to a much greater extend then anticipated. During start-up
tests, temperatures dropped, ground froze, and lakes became
ice covered.
Receiver spacings of 30 m were used to give a nominal fold
of 30, and marsh phones were used to guarantee good cou-
pling under difcult near-surface conditions. Deployment and
pickup of the recording equipment was made possible through
the use of a helicopter. The data were collected using a 2000-
channel telemetry acquisition system. It is worth noting that it
took nine days to deploy the recording spread. Production (i.e.,
recording 1050 shot points), however, was completed within
30 hours. The rst snow storm hit the Sudbury area just hours
before completion of the equipment pickup, causing frozen ca-
bles and geophones.
Table 1. Data acquisition parameters.
Original survey parameters
Survey area approximate by 30 km
2
15 25 m sub bin size
12 receiver lines
30 m receiver station spacing
300 m receiver line spacing
14 source lines
50 m source point spacing
600 m source line spacing
1050 shot points (0.250.5 kg in 510 m deep holes)
Instrumentation
2000-channel, 24-bit telemetry system I/O System II
High cut lter: 270/188 Hz
Pre-amp gain: 36 dB
Geophone type: Mark L-210 (marsh phone)
Geophone array: bunched
Three-component downhole geophone SIE T42 locked at
1070 m depth
All channels live
Helicopter assisted deployment and pick-up of equipment
After collating the data from all of the sources (1050), the
data were sorted into bins with dimensions of 15 30 m, den-
ing the ultimate resolution limits of the nal migrated image.
Note that the bin size was changed to 20 20 m during data
processing. The nal fold diagram is shown in Figure 5b.
In the center of the study area, a three-component seis-
mometer was placed in a deep borehole (at 1070 m) and
recorded all shots. The three-component borehole seismome-
ter proved to be a valuable tool for quality control and data
editing (Milkereit et al., 1997). For example, small seismic
sources in thick, dry overburden do not produce sufcient en-
ergy for reection seismic studies but small charges in bedrock
do. During data acquisition, it was very difcult to observe
clear, high-quality reections. In the hard-rock environment,
we were dealing with source-generated noise (strong shear
waves) and low-frequency noise from owing water. In data
processing, bandpass ltering and high stacking fold were re-
quired to attenuate these types of noise.
Data processing
The data processing sequence hadtotake intoaccount highly
variable overburden conditions (deep swamps, thick eskers,
basement outcrop, and rough topography) as well as steeply-
dipping geological structures. First results were blurred,
and it appeared that conventional 3-D seismic data process-
ing strategies were not well suited to handle weak seismic
signals (from individual scatterers) in the crystalline crust.
A time slice through the data cube is shown in Figure 6a.
Various data processing options were evaluated and a ro-
bust processing sequence was developed, which consisted of
computation of weathering static corrections, deconvolution,
time-variant bandpass ltering, and stacking velocity analy-
sis. Alternative 3-D data binning and stacking schemes were
tested. Once strike and dip were determined, stacking with re-
strictedazimuthal coverage was usedfor stacking velocity anal-
ysis. The evaluation of various 3-D stacked sections revealed
that a limited azimuth stack provided the most robust results
FIG. 5. (a) Trill area geology with survey grid. (b) Fold diagram based on actual source and receiver locations.
(sufcient bandwidth and dynamic range). The nal processed
3-Ddata cube covers an area of 24 km
2
and consists of 200 east-
west lines and 301 north-south lines; the lines are separated by
20 m. The processing sequence is summarized in the Table 2.
A time slice from the nal stacked volume is shown in Figure
6b. The stacked and depth-migrated 3-D data volume cubes of
the study area were interpreted using workstations.
3-D seismic data interpretation
The Trill 3-D seismic survey provides new insights into
the deep structure of the Sudbury North Range. Local high-
impedance contrasts such as massive suldes cause scattering
of seismic energy. Directly above the known mineralization,
prominent circular and semicircular scattering events are ob-
served in the 3-D seismic data. The strongest events occur at
about 0.65 s two-way reection time (about 1800 m depth). An
in-line section through the 3-D cube shows the asymmetric
Table 2. Processing and interpretation.
Brute stack and parameter tests
Geometry
Edit shots and receivers
Compute static corrections (datum: 350 m above sea level)
Select deconvolution, ltering, and scaling
3-D binning (15 25 m bins)
Stack
Revise strike and dip estimates
Evaluate 3-D dip moveout (DMO) processing
Final processing sequence
Revise geometry
3-D binning (20 20 m bins)
High-pass lter: 30 Hz
Evaluate azimuthal coverage
Stacking velocity analysis
Stack
Poststack deconvolution
One-pass 3-D phase-shift migration
diffraction-like response of the known mineralization (Fig-
ure 7a). The high-impedance contrast between the massive
suldes and the steeply dipping norite causes the character-
istic high amplitudes of the reection response to be shifted
towards larger offsets in the downdip direction. The strong
diffraction response of the known mineralization covers a dis-
tance of about 1000 m, a distance which is considerably larger
than the actual size of the ore body. The prominent circular
and semicircular scattering events are best observed in a se-
quence of time slices through the 3-D stacked volume. The
strongest events occur at about 0.6 s two-way reection time
(about 1800 m depth). This sequence of events in Figure 7 is
centered above the known mineralization and can be traced
from 590 ms to about 700 ms. The observed seismic data con-
rms the forward modeling results shown in Figure 4. The in-
terpretation of the scattered energy is complicated by the fact
that structures located northeast and southwest of the known
mineralisation interfere at later times. In addition, phase and
amplitude of the scatteredwavefronts are not constant, causing
severe amplitude-versus-azimuth variations.
Important for the interpretation of the surface seismic data
(and its time-to-depth conversion) is the observation that the
average velocity of the hanging wall (above the target struc-
ture) is about 6300 m/s. Migration of the 3-D stacked data vol-
ume with its steeply dipping reections provided robust results.
Interpretation of the depth-migrated seismic data must honor
the available borehole data and geophysical logs. For the Trill
area, the data base consists of borehole locations and associ-
ated depth-to-footwall-contacts. Migrated sections must be
used for the interpretation of dipping lithological contacts. As
predicted by the borehole geophysical logs shown in Figure 2,
the lower contact of the SIC and associated footwall topogra-
phy are well imagedbetween1000 and2500 mdepth(Figure 8).
Strike and dip of the footwall contact change laterally within
the study area. Toward the east, the footwall contact forms a
local embayment structure at about 2000 and 2500 m depth.
FIG. 6. Scattering response as seen by a 3-D seismic time slice from the Trill area. The slice is from 790 ms into the data volume:
(a) brute stack, (b) nal stack.
In the east, the interpreted sublayer thickness reaches its max-
imum. The main results of the Trill 3-D feasibility study are
summarized in Figure 9. The perspective view (from the east)
shows the steeply dipping footwall contact, the location of the
known mineralization, and the associated scattering response
( a time slice through the stacked volume at 612 ms) caused by
the known mineralization.
FOLLOW-UP STUDIES
The 3-D seismic experiment conrmed that in a geologi-
cal setting such as the Sudbury North Range massive sulde
bodies cause a characteristic seismic scattering response. This
provides an excellent basis for the direct detection of massive
suldes by seismic methods. In addition, the seismic survey
provided new information about strike and dip of the footwall
contact. The eastward extension of the footwall embayment
structure led to a major revision of the geological exploration
model. Toward the southeast, the thickness of the sublayer de-
creases, and reectivity in the footwall of the SIC increases.
These deeper reections within the footwall complex could
not be explained by the available geological and geophysi-
cal logs, and required further investigation. In order to bet-
ter evaluate the performance of the 3-D seismic proling in
the hard-rock environment, it was decided to initate a com-
prehensive calibration and ground-truthing programto further
guide the development of this newexploration technology. The
main objective of the follow-up project was to evaluate the
resolving power (for example, accuracy of depth and dip es-
timates in an area without borehole control) of 3-D seismic
images.
A prominent target from the eastern part of the study area
is shown in Figure 10. The in-line section through the stacked
cube shows prominent scattering events A and B (Figure 10a).
The depth of the scatterers is well dened by the apex of the
diffraction events (assuming an average velocity of 6300 m/s
for the hanging wall). The horizontal slice at 2040 m through
the depth-migrated volume highlights the location of the fo-
cused scatterer A. The predrilling interpretation of the mi-
grated volume is shown in Figure 11. Depth and dip of the
granophyre (G), quartz-gabbro (QG), and norite (N) units of
the Sudbury Igneous Complex, the sublayer (SL), and footwall
complex (FW) are indicated. In the migrated data, events A
and B are characterized by local bright amplitudes at about
2050 and 2400 m depth. The proposed borehole location is in-
dicated. Drilling began in 1997 and veried the dip and depth
estimates for the SIC and footwall contact. In addition, two
local density anomalies were encountered at approximately
FIG. 7. Seismic response of a known mineralization located at about 1800 m depth. Cross-line section (a) and
sequence of time slices through the stacked 3-D data cube (bd). Top of the known mineralization is located
at about 1800 m depth (about 590 ms, reection time assuming an average velocity of 6300 m/s). At 590 ms,
we observe a weak circular reection event. The event expands and, at 640 ms, a high-amplitude semicircular
reection develop towards the east. Despite interference from structure and other events, the semicircular event
continues to grow (time slice at 670 ms). Arrows indicate diffraction. Compare with the synthetic response in
Figure 4.
2050 and 2400 m depth. The density log (derived from core
samples) is shown in Figure 11; magnite-rich units with high
densities are responsible for the scattering response shown in
Figure 10. As indicated by the velocity-density relationship in
Figure 3, magnetite-rich units are characterized by high seismic
impedance values. These will tend to produce seismic scatter-
ing/reection anomalies at least as bright as massive sulde ore
bodies.
In summary, postseismic drilling veried the depth and dip
estimates of the Trill 3-D seismic data interpretation. How-
ever, the interpretation of local seismic scatterering events
is not free of pitfalls. The 3-D seismic data delineated local
scattering events and provided accurate depth estimates. Scat-
tering shown in Figure 10 was caused by high densities. At
present, we cannot distinguish between scattering caused by
magnetite-rich and sulde-rich units.
CONCLUSIONS
The 3-D seismic survey in Sudbury demonstrates for the
rst time that large massive suldes generate a characteris-
tic seismic reection response. By adjusting acquisition and
processing parameters, high-frequency seismic reection pro-
ling techniques can be tailored to (1) image important litho-
FIG. 8. Depth-migrated section with interpretation overlay
(sublayer and footwall complex). The Sudbury Igneous Com-
plex above the sublayer reection is seismically transparent
(compare with borehole logs in Figure 2).
FIG. 9. Composite perspective view of interpreted footwall
contact (green) derived from migrated data, known mineral-
ization at 1800 m depth (red), and scattering event (indicated
by arrow) caused by the mineralization (time slice at 612 ms).
logical contacts and key geological structures such as embay-
ments, and (2) identify and delineate deeply buried, large mas-
sive sulde deposits in the crystalline crust. The effective use
of this new exploration technique requires an integrated ap-
proach incorporating detailed knowledge of the geological
setting, comprehensive physical rock-property studies, state-
of-the-art forward modeling techniques, and high-resolution
seismic data sets. The 3-D reection seismic method can sup-
port deeper exploration projects in existing and prospective
sites as well as prolonging the life of established mines in the
Sudbury basin. In addition, this new methodology could reju-
venate deep exploration projects in mature base-metal mining
camps.
ACKNOWLEDGMENTS
Inco Ltd., Falconbridge Ltd., and the Geological Survey of
Canada funded the 3-D feasibility study. The 3-D surface seis-
mic data were acquired by Solid State Geophysical of Calgary
and processed in-house.
FIG. 10. (a) In-line seismic section with scattering events A
and B. The apex of event A is located at 0.64 s two-way time
(approximately2050mdepthat 6300m/s). (b) Migratedsection
(depth slice at 2040 m) with reection from footwall contact
(FWC) and local amplitude anomaly A.
FIG. 11. Depth-migrated section with borehole location and density log. Magnetite-rich units (A, B) with high
densities are responsible for the scattering response shown in Figure 10. G= granophyre, QG= quartz-gabbro,
N = norite, SL = sublayer, and FW= footwall complex.
REFERENCES
Boerner, D. E., Milkereit, B., and Naldrett, A. J., 1994, Introduction
to the Special Section on the Lithoprobe Sudbury: Geophys. Res.
Lett., 21, 919922.
Eaton, D. W., 1996, BMOD3D: A program for three-dimesional seis-
mic modelling using the Born approximation: Geol. Surv. Can. Open
File Report 3357.
Milkereit, B., Berrer, E. K., Watts, A., and Roberts, B., 1997, Devel-
opment of 3-D seismic exploration technology for Ni-Cu deposits,
Sudbury basin, in Grubbins, A., Ed., Proceedings of Exploration 97,
439448.
Milkereit, B., andEaton, D., 1998, Imaging andinterpreting theshallow
crystalline crust: Tectonophysics, 286, 518.
Milkereit, B., Eaton, D., and Berrer, E., 1996b, Towards 3-D seismic
exploration technology for the crystalline crust: 66th Ann. Internat.
Mtg., Soc. Expl. Geophys., Expanded Abstracts, 638641.
Milkereit, B., Eaton, D., Wu, J., Perron, G., Salisbury, M., Berrer, E.,
andMorrison, G., 1996a, Seismic imaging of massive sulde deposits,
Part 2: Reection seismic proling: Econ. Geol., 91, 829834.
Milkereit, B., Green, A., and Sudbury Working Group, 1992, Deep
geometry of the Sudbury structure from seismic reection proling:
Geology, 20, 807811.
Morrison, G. G., 1984, Morphological features of the Sudbury struc-
ture in relation to an impact origin, in Pye, E. G., Naldrett, A. J.,
and Giblin, P. E., Eds., Geology and ore deposits of the Sudbury
structure: Ontario Geol. Surv., Special Volume 1, 513520.
of massive sulde deposits, Part 1: Rock properties, Econ. Geol., 91,
821828.
Crosshole seismic imaging for sulde orebody
delineation near Sudbury, Ontario, Canada
Joe Wong
ABSTRACT
Crosshole seismic instrumentation based on a piezo-
electric source and hydrophone detectors were used
to gather seismograms between boreholes at the
McConnell orebody near Sudbury, Ontario. High-
frequency seismograms were recorded across rock sec-
tions 50 to 100 m wide containing a continuous zone
of massive sulde ore. First-arrival traveltimes obtained
from a detailed scan were used to create a P-wave ve-
locity tomogram that clearly delineated the ore zone.
Refraction ray tracing on a discrete layer model con-
rmed the main features of the tomogram. The survey
demonstrated that it is possible to conduct cost-effective,
high-resolution crosshole seismic surveys to delineate
ore bodies on a scale useful for planning mining oper-
ations.
INTRODUCTION
There is growing interest inusing geophysics toassist inplan-
ning mining operations. In particular, borehole-to-borehole
methods are being recognized as having exceptional potential
in this regard. By using crosshole geophysics to provide greater
geometrical detail thanis availablefromtraditional delineation
via pattern drilling, mining of an economic deposit can be de-
signed or modied to minimize costs (Williams, 1996). This
paper presents a test survey in which high-resolution cross-
hole seismic tomography is demonstrated to be effective for
orebody delineation. Previous applications of seismic tomog-
raphy used sparkers (Greenhalgh and Mason, 1997) and mi-
croexplosions (Duncan et al., 1989) as seismic sources. I de-
scribe a nondestructive piezoceramic vibrator source driven
by a pseudorandom waveform.
The McConnell deposit, owned by INCO Ltd., is located
near Sudbury, Ontario, Canada. It has been used for testing
various downhole and crosshole geophysical techniques. Meng
and McGaughey (1996) applied crosshole seismic reection to
Manuscript received by the Editor January 26, 1999; revised manuscript received April 18, 2000.
JODEX Applied Geoscience Ltd., 9715 Horton Road S.W., Unit 101, Calgary, Alberta T2V 2X5, Canada. E-mail: jjodex @telusplanet.net.
image the deposit. They used blasting caps as downhole seismic
sources and recorded seismic frequencies in the 1- to 5-kHz
range. Livelybrooks et al. (1996) and Fullagar et al. (1996)
describe crosshole electromagnetic surveying to map the ore
body. The technique, sometimes referred to as the radio imag-
ing method (RIM), uses electric dipoles operating at frequen-
cies of 500 kHz and 5 MHz.
The target at the McConnell site is a near-surface, steeply
dipping massive sulde orebody residing in a host rock con-
sisting chiey of amphibolite and metasediments. Anumber of
BQ-sized (60 mmID) boreholes have been drilled through the
ore zone (Figure 1). Mueller et al. (1997) give more details con-
cerning the geological units intersected by the boreholes. Bore-
hole velocity logging has indicated the P-wave velocities of the
host rock range from 5.8 to 6.5 km/s. In contrast, the velocities
within the massive sulde zone are at least 20% lowerin the
range of 4.2 to 4.6 km/s. The situation is almost ideal for eval-
uating crosshole seismic methods for possible use to support
planning mining activities.
METHODS
Instrumentation
The CORRSEIS borehole seismic system, shown schemat-
ically on Figure 2, is designed for rapid and efcient acquisi-
tion of high-frequency seismograms between boreholes. The
system includes a controlled piezoelectric vibrator and mul-
tiple (8 or 16) high-sensitivity hydrophone detectors. Similar
equipment has been described by Wong et al. (1983, 1987) and
Harris et al. (1995). In the present case, the source was about
53 mm in diameter and 75 cm in length. Each hydrophone was
about 50 mm in diameter and 10 cm in length. Because both
the source and the detectors are coupled to the rock by wa-
ter in the boreholes, the recorded seismic energy is primarily
P-waves. The dominant seismic frequencies generated by the
source are about 3.3 kHz, resulting in wavelengths of about
2.0 m in the host rock with P-wave velocities of 6.5 km/s and
about 1.4 m in the massive suldes with P-wave velocities of
about 4.2 km/s. These wavelengths indicate the resolution that
1900
Crosshole Seismic Imaging 1901
can be expected from data acquired with the CORRSEIS sys-
tem. The output power of the vibrator is adjustable and is nom-
inally on the order of 10 to 30 W. The total source energies asso-
ciated with eld seismograms are typically in the range of 0.1 to
2.0 kJ.
There are several practical advantages inusing a controllable
piezoelectric vibrator as the source for crosshole seismic scan-
ning. Once deployed with electrical power in a borehole, the
vibrator can be operated for long periods of time without need-
ing to be returned to the surface. The piezoelectric source is
much more repeatable and safer than microexplosives or blast-
ing caps, and there is very little risk of damage to boreholes.
If 16 or more hydrophones are used in the detector array and
the equipment is congured with automatic winches, acquisi-
tion rates >2500 crosshole seismic traces per hour are possible
under low-noise conditions in low-loss rock.
Crosscorrelation used for data acquisition
The vibrator in the CORRSEIS system is driven by a pseu-
dorandom binary sequence (PRBS) in continuous cycles, and
impulsive seismograms are obtained from the detected vi-
brations by crosscorrelation with the PRBS pilot signal. The
length of the PRBS used in acquisition is 2047, but the ba-
sic concepts of crosscorrelation can be illustrated using the
PRBS of length 63 shown on Figure 3a. When the vibrator
is driven by a high-voltage replica of the sequence, its sys-
tem response modies the original sequence by convolving it
with a time-domain lter equal to, for example, the seismic
source wavelet shown on Figure 3b. The resulting vibration
FIG. 1. Cross-section of the McConnell ore body with ex-
ploratory boreholes.
detected at some distance fromthe source is the ltered PRBS
delayed by the seismic traveltime between the source and the
receiver. The detected signal, shown on Figure 3c, is digitized,
stacked, and then crosscorrelated with the original PRBS. The
result is the trace in Figure 3d, which is the source wavelet
(with some high-cut ltering) delayed by the seismic travel-
time. This is the equivalent of a seismic trace froman impulsive
source.
The dening characteristic of a pseudorandom or pseudo-
noise sequence is that its autocorrelation simulates a pure delta
function in time domain. The autocorrelation of the PRBS of
length 63, shown in Figure 3e, has a triangular shape with no
side lobes and is a good approximation to the delta function.
The height of the triangle depends on the sequence length N,
being proportional to the sum of N individual pulses of unit
amplitude. The absence of side lobes is important because it re-
sults in cleaner seismograms after crosscorrelation. The cross-
correlation method enhances S/N ratios by a factor equal to
the square root of the sequence length N. Additional signal
enhancement is achieved by synchronously summing or stack-
ing 20500 repeated waveforms. In many cases, using long se-
quences in conjunction with waveformstacking makes it possi-
ble to record good-quality seismograms even in the presence of
drilling noise. Hurley (1983) gives a more detailed discussion
on PRBS.
FIG. 2. Schematic of the CORRSEIS borehole seismic system.
1902 Wong
RESULTS
Field seismograms
Seismograms wererecordedwithdetectors inborehole(BH)
78 929 and the source in BH-78 930 and BH-80 555. Sample
seismograms between BH-78 929 and BH-80 555 are shown on
Figure 4 as a common detector gather. Each recorded seis-
mogram consists of 512 or 1024 sixteen-bit samples with a
sampling interval of 50 s. The columns of numbers to the
right of the gure under the headings TX and RX are the
source and detectors depths in meters. The S/N ratios of
the traces are very good, and the dominant frequencies are
in the 3- to 4-kHz range. First-arrival times fall in the range of
10 to 17 s. These data indicate that piezoceramic-based equip-
ment produces excellent crosshole seismograms for source-
detector distances of over 100 min the rocks at the McConnell
site.
For tomographic imaging, a complete seismic scan of the
rock panel between BH-78 929 and BH-78 930 was done. The
detectors were located in BH-78 929 at depths from 30 to
FIG. 3. (a) PRBS of length 63. (b) Time-domain lter, i.e.,
seismic wavelet. (c) Convolution of delayed PRBS and seis-
mic wavelet. (d) Cross-correlation of trace (c) with the PRBS,
recovering a delayed ltered version of the seismic wavelet.
(e) Autocorrelation of PRBS simulating a delta function.
140 m on 2-m intervals. The source was located in BH-78 930
at depths from 30 to 170 m on 0.62-m intervals. The scanning
pattern is depicted schematically on Figure 5, where seismic
raypaths joining source and detector positions are represented
by straight lines (true raypaths would curve or bend accord-
ing to Snells Law). The entire scan consisted of over 4200
raypaths, but only every fth raypath has been drawn on Fig-
ure 5. Examples of seismograms from this scan are shown on
Figure 8.
Tomographic imaging
First-arrival times for the scan of Figure 5 were picked us-
ing an automatic routine and were checked and edited visually
on a computer display. These rst-arrival times were used to
produce a P-wave velocity tomogram via a straight-ray back-
projection algorithm (Peterson et al., 1985). In essence, the
algorithm assigns seismic velocity values to an array of rectan-
gular pixels overlying the rock section between the source and
detector boreholes. Seismic traveltimes for direct P-wave ar-
rivals are calculated and compared with the observed times for
all source and detector positions in the scan. The velocity val-
ues are then automatically adjusted to reduce the differences
or residuals between observed and calculated arrival times.
This is repeated until the residuals are reduced to an accept-
able level, at which point we assume that the array of velocity
values within the pixels is a good representation of the true
velocity distribution in the scanned rock section between the
boreholes.
For the panel between BH-78 929 and BH-78 930 at the
McConnell site, therms valueof the4202observedarrival times
was about 9.44 ms. Pixel dimensions used for tomographic
imaging were 1.5 m high by 2.5 m wide. After 10 iterations
of the imaging algorithm, the rms value of the residual arrival
times was reduced to about 0.38 ms. The resulting P-wave ve-
locity tomogram is shown on Figure 6. The ore zone is clearly
outlined by the low-velocity (4.04.5 km/s) zone residing in the
higher velocity (5.96.5 km/s) host rock.
Tomographic analysis basedonthestraight-raypathassump-
tion provides a rapid interpretation of the rst-arrival times in
a seismic scan like that shown in Figure 4. However, the ac-
tual raypaths in a geological medium with large variations in
seismic velocities will inevitably curve or bend fromrefraction.
Therefore, while the gross aspects of the tomogramin Figure 6
are likely to be reliable, the details may be considerably differ-
ent from the actual geology. Also, velocity tomograms can be
degraded by the presence of false features or artifacts. These
arise during image reconstruction because of the large number
of pixels and degrees of freedom if the process is not carefully
constrained. In light of these shortcomings in the tomographic
analysis, the structure represented by the tomogramshould be
checked against other, independent models.
Refraction ray tracing
One way to accomplish this is to construct a discrete layer
model using the tomogram (and other information such as
borehole velocity logs) as a guide. The velocities are constant
within each layer. The layers may vary in thickness and dip,
and boundaries between layers may be curved. Refraction
FIG. 4. Crosshole seismograms between boreholes 78 929 and 80 555. First-arrival times are between 10 and 17 s; source-detector
separations are approximately 60100 m. The columns of numbers to the right of the gathers are, respectively, the source and
detector depths in meters.
according to Snells Law at the seismic boundaries is used to
trace seismic raypaths from one borehole to another.
On Figure 7, raypaths traced across one such model are
shown for two common detector gathers with the detector in
the left-hand borehole. The ore zone is represented by the yel-
low layer. Many raypaths are severely refracted at the bound-
ary between the high-velocity host rock and the low-velocity
sulde ore body. In some cases, the rays are reversed in the
1904 Wong
direction normal to the rockore interface by postcritical (to-
tal) reection. Total reection may result in shadow zones in
which the seismograms for particular sourcedetector loca-
tions have very weak rst-arrival amplitudes.
As part of the ray tracing, the arrival time for each ray can be
calculated. By adjusting the seismic velocities and the bound-
ary positions and dips in the model, the calculated arrival times
can be made to closely t observed arrival times. On Figure 8,
two common detector gathers are displayed on an expanded
time scale. Arrival times calculated for corresponding raypaths
traced across the velocity model on Figure 7 are plotted as red
bars. The agreement between the observed and calculated rst-
arrival times is good, indicating that the velocity model is a
reasonable representation of the eld data and the underlying
geology. It is relatively easy to t rst-arrival times for individ-
ual gathers separately. The difculty lies in obtaining a single
model that closely match both (or more) gathers equally well.
As more gathers of eld seismograms are included for compar-
ison, more layers may need to be added to obtain a good global
t. In this event, the process must be automated in a nonlinear
optimization scheme (i.e., inversion).
Analternative tostraight-ray imaging followedby refraction
modeling to conrm the tomogram is tomographic imaging
using curved raypaths. For example, Bishop et al. (1985) and
Bregman et al. (1989) describe imaging algorithms that ana-
lytically trace refracted raypaths across triangular cells within
which velocities are linear. Jackson and Tweeton (1994) have
published Migratom software, which creates tomograms using
FIG. 5. Schematic showing the raypaths of a detailed seismic
scan between boreholes 78 929 and 78 930. In the complete
scan, there are over 4200 raypaths; only every fth ray is shown
here.
refracted raypaths obtained by Huygens wavefront expansion.
Tomographic image reconstruction using refracted raypaths is
a nonlinear process. As such, it is slower thanstraight-ray imag-
ing andis more prone toproblems of nonuniqueness andimage
artifacts.
The successful imaging of the McConnell ore body using
rst-arrival times to create a P-wave velocity tomogram is at-
tributable to the large, massive nature of the deposit and its rel-
atively simple geometry. The velocity contrast between the ore
andthe host rockis high, andthe ore zone is a single layer that is
relatively thick compared to both the seismic wavelengths used
and the hole-to-hole distances. If the ore zone had been in the
form of thin stringers, rst-arrival times would have been less
diagnostic of the low-velocity suldes since the rst-arriving
raypaths would reside mostly in the high-velocity host rock.
Also, if the velocity distribution had been more complex, ei-
ther with much folding and faulting in the plane of the panel or
with signicant out-of-plane 3-D structure, the velocity image
might not have represented the true geology as well.
Nevertheless, in many situations, crosshole seismic imag-
ing using high-resolution seismic data collected with a
FIG. 6. P-wave velocity tomogram for the scan of Figure 5.
The tomogramwas produced fromobserved rst-arrival times
using a straight-ray iterative back-projection algorithm.
FIG. 7. Refraction ray tracing between boreholes 78 929 and 78 930 across a discrete layer model. Common detector gather with
detector (a) below the orebody and (b) above the orebody.
piezoceramic source and hydrophone detectors appears to be
a practical method for delineating ore bodies. Seismic imaging
complements the information gained from pattern drilling by
providing more details about the geometric aspects of an eco-
nomic deposit. The additional information gained by seismic
imaging may assist in optimizing mining procedures to reduce
costs.
The most critical factor in determining whether crosshole
seismic imaging will become widely acceptedby people respon-
sible for planning mining operations is its cost effectiveness.
In actual practice and for routine application, costs associated
with eld time for data acquisition are likely to be a major con-
sideration in deciding whether the crosshole seismic method
is used. It is therefore important to have a technology that al-
lows for the rapid recording of high-quality, high-resolution
crosshole seismograms. The experience at the McConnell site
supports the contention that piezoceramic-based equipment
similar to that described in this paper should have the neces-
sary resolution, range, and operational efciency to meet the
practical requirements.
ACKNOWLEDGMENTS
I thank INCO Technical Services Ltd. for providing access
to the boreholes at the McConnell site and especially Glenn
McDowell of INCO Ltd. for logistical support during the eld
work. I also thank Greg Turner and Lawrence Gochioco for
their constructive review of this paper.
REFERENCES
Bishop, T. N., Bube, K. P., Cutler, R. T., Langan, R. T., Love, P. L.,
Resnik, J. R., Shuey, R. T., Spindler, D. A., and Wyld, H. W., 1985,
Tomographic determination of velocity and depth in laterally vary-
ing media: Geophysics, 50, 903923.
Bregman, N. D., Bailey, R. C., and Chapman, C. H., 1989, Crosshole
seismic tomography: Geophysics, 54, 200215.
Duncan, G., Downey, M., Leung, L., and Harman, P., 1989, The de-
velopment of crosshole seismic techniques and case studies: Expl.
Geophys., 20, 127130.
Fullagar, P. K., Zhang, P., Yu, Y., and Bertrand, M. J., 1996, Applica-
tion of radio-frequency tomography to delineation of nickel sulde
deposits in the Sudbury basin: 66th Ann. Internat. Mtg., Soc. Expl.
Geophys., Expanded Abstracts, 20652068.
Greenhalgh, S., and Mason, I., 1997, Seismic imaging with application
to mine layout and development: 4th Decennial Internat. Conf. on
Min. Expl., Exploration 97, Proceedings, 585598.
Harris, J. M., Hoeksma, R. C., Langan, R. T., VanSchaak, M., Lazaratos,
S. K., and Rector, J. W., 1995, High-resolution crosswell imaging
of a West Texas carbonate reservoir: Part IProject summary and
interpretation: Geophysics, 60, 667681.
Hurley, P. M., 1983, The development and evaluation of a crosshole
seismic system for crystalline rock environments: M.S. thesis, Univ.
of Toronto.
Jackson, M. J., and Tweeton, D. R., 1994, MIGRATOMGeophysical
tomography using wavefront migration and fuzzy constraints U.S.
Dept of Interior Bureau of Mines Report RI 9497.
Livelybrooks, D., Chouteau, M., Zhang, P., Stevens, K., andFullagar, P.,
1996, Borehole radar and radio imaging to delineate the McConnell
ore body near Sudbury, Ontario: 66th Ann. Internat. Mtg., Soc. Expl.
Meng, F., and McGaughey, J., 1996, Ore body hunting with crosswell
imaging methods: 66th Ann. Internat. Mtg., Soc. Expl. Geophys.,
Mueller, E. L., Morris, W. A., Killeen, P. G., and Balch, S., 1997,
Combined 3-D interpretations of airborne, surface, and borehole
1906 Wong
FIG. 8. Arrival times calculated from the rays in Figure 7 plotted on the corresponding eld seismograms. The calculated arrival
times compare very well with the rst-arrival times on the recorded traces.
vector magnetics at the McConnell nickel deposit: 4th Decennial
Internat. Conf. on Min. Expl., Exploration 97, Proceedings, 657
665.
Peterson, J. E., Paulsson, B. N. P., and McEvilly, T. V., 1985, Appli-
cation of algebraic reconstruction techniques to crosshole seismic
Williams, P., 1996, Using geophysics in underground hard rock mining:
A question of value and vision: 66th Ann. Internat. Mtg., Soc. Expl.
Wong, J., Bregman, N., Hurley, P., and West, G. F., 1987, Crosshole
seismic scanning and tomography: The Leading Edge, 6, 3641.
Wong, J., Hurley, P., and West, G. F., 1983, Crosshole seismology and
seismic imaging in crystalline rocks: Geophys. Res. Lett., 10, 686
689.
In-mine seismic delineation of mineralization and rock structure
Stewart A. Greenhalgh
, Iain M. Mason
, and Cvetan Sinadinovski
ABSTRACT
Signicant progress has been made towards the goal
of generating detailed seismic images as an aid to mine
planning and exploration at the Kambalda nickel mines
of Western Australia. Crosshole and vertical-seismic-
proling instrumentation, including a slimline multi-
element hydrophone array, three-component geophone
sensors, and a multishot detonator sound source, have
been developed along with special seismic imaging soft-
ware to map rock structure.
Seismic trials at the Hunt underground mine estab-
lished that high frequency (>1 kHz) signals can be prop-
agated over distances of tens of meters. Tomographic
as well as novel 3-D multicomponent reection imaging
procedures have been applied to the data to produce
useful pictures of the ore-stope geometry and host rock.
Tomogram interpretation remains problematic because
velocity changes not only relate to differing rock types
and/or the presence of mineralisation, but can also be
caused by alteration/weathering and other rock condi-
tion variations. Ultrasonic measurements on rock core
samples help in assigning velocity values to lithology,
but geological assessment of tomograms remains am-
biguous. Reection imaging is complicated by the pres-
ence of strong tube-wave tobody-wave mode conversion
events present in the records, which obscure the weak
reection signatures. Three-dimensional reection data
processing, especially three-component analysis, is time
consuming and difcult to perform. Notwithstanding the
difculties, the seismic migrations at the Hunt mine show
a striking correlationwiththe knowngeology. Combined
seismic and radar surveying fromavailable underground
boreholes and mine drivages is probably needed in the
future to more condently delineate mineralisation.
INTRODUCTION
There are potentially great economic benets to be obtained
from an ability to seismically probe ahead of a mine face,
or to image with a reasonable clarity a volume of rock sur-
rounding an exploration or underground development bore-
hole. The sort of geological information that a high-frequency
(2002000 Hz) seismic technique can, in principle, provide in-
cludes existence of ore, changes of rock type, offset of mineral-
ization, location of structures such as faults and troughs, extent
of shear zones, etc. The seismic technique offers good resolving
power (a few meters) coupled with a sufcient degree of pen-
etration (probing distance tens to hundreds of meters), com-
pared to other geophysical techniques like ground-penetrating
radar, which suffers from severe attenuation.
Until recently, relatively few successful hard-rock seismic
surveys had been carried out in metalliferous mining areas.
Surface reection surveys for mapping mineralization have
been reported by Nelson (1984); Galperin (1984); Dahle et al.
Manuscript received by the Editor April 20, 1999; revised manuscript received May 15, 2000.
The University of Adelaide, Dept. of Geology and Geophysics, Adelaide, South Australia 5005, Australia. E-mail: sgreenhalgh@geology.
adelaide.edu.au.
University of Sydney, School of Geoscience, Sydney, New South Wales 2006, Australia. E-mail: geoimm@es.su.oz.au.
Australian Geological Survey Organisation, GPO Box 378, Canberra, ACT 0206, Australia. E-mail: cvetan.sinadinovski@agso.gov.au.
(1985), Spencer et al. (1993); Wright et al. (1994), Adam et al.
(1996), and Pretorius et al. (1997). Underground reection and
transmission measurements in hard rock mines and crosshole
tomography surveys in crystalline rock have been carried out
in Germany (Schmidt, 1959; Wachsmuth and Schmidt, 1962),
southern Africa (Reid et al., 1979; Mutyorauta, 1987; Carneiro
andGendzwill, 1996), the U.S.A. (Ruskey, 1981; Petersonet al.,
1985; Freidel et al., 1995), Canada (Wong et al., 1984; Young
et al., 1989; Gendzwill and Brehm, 1993), Sweden (Gustavson
et al., 1984), and Australia (Harman et al., 1987; Duncan et al.,
1989; Sinadinovski et al., 1995; Cao and Greenhalgh, 1997;
Greenhalgh and Mason, 1997).
In the 1990s, a major research program was initiated by
Western Mining Corporation at Kambalda, Western Australia,
to develop high-resolution seismic and radar techniques to
aid mining of and exploration for nickel sulde orebodies.
The economic justication is improved cost-effectiveness (and
safety) of undergrounddevelopment, orebodydelineation, and
1908
In-Mine Seismic Delineation 1909
surface exploration. This paper is concerned with the seismic
workdoneandimagingresults obtainedat just oneminesiteus-
ing both seismic tomography and seismic reection techniques.
The purpose was to assess the usefulness of underground (in-
mine) methods in mapping stope geometry and hard rock ge-
ological structure. Instrumentation developments and results
from other sites are described by Cao and Greenhalgh (1997),
Bierbaum and Greenhalgh (1997), Greenhalgh and Mason
(1997), and Greenhalgh and Bierbaum (2000).
GEOLOGICAL AND GEOPHYSICAL TARGETS
The Kambalda nickel sulde environment
The Kambalda region is situated within the south-central
part of the Norseman-Wiluna greestone belt, in the Eastern
Goldelds province of the Achaean Yilgarn craton. The town
of Kambalda is located 560 km east of Perth.
The volcanic sedimentary sequence at Kambalda, with its
enclosed magmatic nickel sulde deposits, has undergone de-
formation, metamorphism, intrusive activity, and late stage
gold mineralization (Cowden and Roberts, 1990). Many rocks
in the region have undergone upper greenschist to lower
amphibolite-facies metamorphism. There has been a series of
intrusions of mac to felsic stocks, dikes, and sills.
The nickel ores at Kambalda occur in 24 shoot complexes,
each containing multiple ore surfaces (pods) caused by struc-
tural dislocation of a few, grouped, north-northwest trending
ribbonlike orebodies. Most ore occurs at the base of the low-
est ultramac ow, in elongate, shallow (100 1500 10 m)
troughs, which are thought to have been original volcanic de-
pressions at the base of the komatiite lava. The ore is ap-
proached fromthe rear through stopes which lead off drivages
driven through the Lunnon basalt. The contact zone undulates.
The morphology of the ore zone is highly variable with massive
sulde layers at the base being irregular in thickness (generally
less than 5 m). Matrix and disseminated ore layers are more
regular in distribution and generally comprise 6080% of the
total thickness of the sulde zone.
Velocities and densities of ores and host rocks
Refraction surveying and downhole velocity measurements
were carried out at Kambalda in the 1970s to determine
whether major rock boundaries (which control mineralization)
could be mapped seismically (C. Porter, personal communica-
tion, 1978). Five refraction spreads established that the P-wave
velocityof theultramacs (56006500m/s) is signicantlylower
than that of the hanging-wall basalt (60006500 m/s) and that
of the footwall basalt (6800 m/s). Four well-shoot experiments
gave slightly lower velocities (ultramacs, 46005400 m/s; foot-
wall basalt, 63006800 m/s; porphyry intrusion, 4600 m/s). Note
that the ultramacs (comprising talc carbonate, talc chlorite,
and amphibole-chlorite rocks) have been serpentinised and,
therefore, exhibit lower velocities that one would expect for
such rocks (which in an unaltered state show higher velocity
than mac rocks).
Sonic logging carried out in just one hole in the late 1980s
gave an average velocity for basalt of 6500 m/s, which dropped
to 5500 m/s within a 2-m brecciated zone. The corresponding
densities were 2.9 and 2.65 g/cm
3
.
At the beginning of the new seismic research program, ad-
ditional elastic property data were obtained. In-mine refrac-
tion measurements at the Foster mine yielded high seismic
wavespeeds for basalt: 6700 m/s for P-waves and 4000 m/s
for S-waves. Laboratory determinations of ultrasonic velocity
and density were made for several hundred rock core sam-
ples collected from various mine sites. An analysis is given by
Greenhalgh and Mason (1997), who plotted velocity versus
density for the various rock and ore samples. Compressional
wave velocity shows a clear inverse relationship with ore grade
or density. The ultramacs gave an average P-wave velocity
of 5500 m/s and an average density of 2.87 g/cm
3
. The mac
samples showed slightly higher velocity (6050 m/s) and den-
sity (2.93 g/cm
3
). Massive ore was characterised by low veloc-
ity (4900 m/s) and high density (4.3 g/cm
3
). For matrix ore,
the velocity was somewhat higher (5200 m/s) and the density
lower (3.52 g/cm
3
). Plane-wave reection coefcients calcu-
lated from these physical property data are marginal (<0.05)
for disseminated-matrix ore/rock boundaries, but detectable
for the ultramac/mac boundary and the massive ore/mac
and massive ore/ultramac boundaries (0.080.15).
UNDERGROUND SEISMIC EXPERIMENTS
The seismic experiment was carried out at the Hunt mine,
D zone decline, 16 level, below the B05 ore surface. A com-
panion borehole radar survey was conducted at the same site.
The northern portion of the site had also been the test area for
a previous radar proling experiment in which the bottom of
the ore surface was partially imaged from a single scan along
the mine decline. This experiment suffered from electromag-
netic (EM) reverberation in the tunnel (noise which obscured
reected signals) and lack of dimensionality. With only two di-
mensions (time and horizontal distance along the scan line),
only target range and horizontal position can be recovered.
Angular elevation to the target requires that radar data be
acquired along at least two directionsthe decline as well as
one or more in-mine boreholes. Similar considerations apply
to seismic imaging.
The only previous seismic work attempted at Kambalda had
been at the Foster mine. In that work triaxial geophones were
mountedalong the decline wall, andshots redinshallowholes
drilled into the tunnel wall. The survey clearly showed that it
was possible to generate high frequency (>1 kHz) signals and
that they propagated over signicant distance (>100 m), but
attempts at imaging the geology were frustrated by receiver-
mount resonance problems, insufcient receiver array aper-
ture, spatial aliasing, poor angular coverage, and the lack of
any well-dened target (the orebody had been mined out). The
Hunt mine seismic experiment was thus designed to overcome
many of the Foster mine seismic survey limitations by operat-
ing primarily in boreholes at a close shot and receiver spac-
ing, using both crosshole and vertical-seismic-proling (VSP)
recording geometries. The anchors to the triaxial geophones
used along the decline were permanently grouted into the rock
to ensure good coupling.
The Hunt seismic experiment was the rst underground
test of the new borehole hydrophone streamer and the new
explosive borehole seismic source (Greenhalgh and Mason,
1997). Two transportation/navigation systems for borehole
transducer deployment, (a pushrod system and a hydraulic
1910 Greenhalgh et al.
injection system) were also tested during this experiment. In
this paper, we document the technical aspects of the work, as
well as present as results on the images obtained.
The test site
Figure 1 is a plan view of the test site. It shows the locations
of seismic boreholes DDH 16-18 and DDH 16-20 in relation
to the D zone decline at the 16 level, and the overlying stopes.
The stations G1G8 are the collar positions of the triaxial geo-
phones along the tunnel as well as being hammer impact points
(see next section). The boreholes 16-18 and 16-20 were drilled
downwards at an angle of 30
fromthe horizontal (so as to hold

water) to a depth of 45 m. The distance between borehole col-
lars is approximately 55 m. The volume of rock to be imaged
was largely boundedby the decline, the twoboreholes, andhor-
FIG. 1. Seismic test site at the Hunt mine showing source-receiver boreholes, geophone array along decline, and
stope geometry 20 m above decline.
izontal planes located 50 m above and below the level of the
decline. The stopes of the B05 ore surface are a height of about
20 m above the borehole collars. This can be seen in Figure 2
which is a 3-D perspective view of the site. The additional pair
of upward-inclined boreholes shown, DDH 16-19 and DDH
16-08, were used for the radar experiment. Most of the ore in
the vicinity of the seismic/radar test site has been mined out.
The 16 level represents the deepest development of the mine.
Extraction is continuing to the immediate south of the survey
area. In fact, mining activity dictated that the geophysical work
be undertaken on the afternoon and night shifts, when vehicles
were not operating in the decline.
Figure 3 gives two geological cross-sections in the vertical
planes of boreholes DDH 16-18 and DDH 16-20. The original
ore shoot (premined) is about 2 mthick and 2030 mwide, and
occupies the base of several troughs within the contact surface
between the footwall basalt and the hanging wall ultramacs.
There is remnant ore in the stopes where the surface has thick-
ened in the formof local jags. The ore has about 1.5:1 density
contrast with the enclosing rock, and possibly presents a slight
FIG. 2. 3-D perspective view of the Hunt D zone decline and
B05 stopes at the seismic test site. Boreholes used in the exper-
iment are shown.
FIG. 3. Cross-section at boreholes H16-18 and H16-20 showing
simplied stope geometry and ore troughs.
velocity decrease (say, 5000 m/s versus 6000 m/s). The com-
puted reection coefcient for the idealised case of a plane, lat-
erally continuous mineralised surface is about 0.11. In prac-
tice, the nite width of the ore, its discontinuous nature, and the
roughness of the surface would greatly diminish its reectivity,
especially for wavelengths of comparable size to the structure.
The ore wouldbehave more as a diffuse diffractor thanas a mir-
ror. Furthermore, the target will only reect sound back to
the receiver array if illuminated fromthe right range of angles.
The stope cavities themselves, rather than the small remnant
mineralization within them, are the more likely reectors in
this survey. The air/rock interface presents a perfect reecting
surface (reection coefcient of 1), but once again the re-
ceived signal will be downgraded by the nite extent (length
and width) of the tunnels and the nite wavelength. For a 1000-
Hz seismic signal, the wavelength is about 6 m (not unlike the
radar wavelength), which is of the same order as the width of
each stope.
The host medium is hardly homogeneous. Even within the
footwall basalt there is considerable variability in composition
and alteration, and hence elastic properties, as revealed by the
two drill cores. There are several faults, fractures, and quartz
veins, as well as a porphyry sheet (see Figure 3), which disturb
the host rock.
Data acquisition geometry
Three separate experiments (crosshole survey, hammer-
source VSP survey, and explosive-source reversed VSP) were
conducted at the site shown in Figure 1.
Crosshole survey.The crosshole survey entailed shots in
hole H16-20 at 1-m depth spacing, from 3.2 to 42.3 m, and
recording on a hydrophone array in hole H16-19, at 1-m spac-
ing, over the depth range 0 to 41 m. Since the hydrophone array
has its detectors at 2-m intervals and is limited to just 24 chan-
nels, the 1-m receiver array was synthesised by shooting twice
at each depth level in hole H16-20, once for hydrophones at
positions 0 m, 2 m, . . . , 40 m, and again after displacing the eel
by 1 m in the hole such that the elements were as positions
1 m, 3 m, . . . , 41 m. It was not possible to push the hydrophone
streamer into the hole any deeper than 43 m. The number of
useable recording channels was 21 per shot, yielding a total of
40 2 21 =1680 seismic traces.
It would have been desirable to have collected the seismic
data at a closer spacing (shot and receiver interval of 0.5 m
or less), but this would have greatly increased the acquisition
time. We were limited largely by the time to load and place
each shot.
Hammer VSP experiment.The hammer VSP experiment
was the rst phase of the work undertaken. It involved plac-
ing the hydrophone array in hole H16-18, with detectors at
2-m intervals over the depth range 141 m, and impacting the
tunnel wall with a 5-kg sledge hammer along the decline at
one-quarter station positions from G1 to G8.5 (1.25-m inter-
vals). This is a multilevel, walkaway VSP recording geometry.
The number of seismograms collected was 31 21 =651.
Triaxial VSP experiment.This experiment was a reversed
multilevel walkaway reversed VSP. It entailed having shots at
1-m spacing in both holes H16-18 (depth range 3.242.3 m)
and H16-20 (depth range 3.642.6 m) and recording on eight
three-component geophones along the decline wall (G1G8;
see Figure 1) at a nominal spacing of 5 m. The sensors were
placed in 2-m deep horizontal holes drilled perpendicular to
the tunnel axis, oriented into position, and then grouted per-
manently into each hole. The orientation system was such that
the X component was parallel to the decline, with its positive
terminal pointing northwest (up the decline), the Y component
pointing along and into the axis of the hole (northeast), and
the Z component vertically downwards. The purpose of using
three-component receivers was to achieve directional sensitiv-
ity in the response, which is of potential benet in 3-Dimaging
with a limited number of sensors arranged in a fairly short lin-
ear array.
The number of traces acquired in the triaxial VSP experi-
ment was 40 2 8 3 =1920.
Instrumentation
A block diagram of the recording instrumentation is shown
in Figure 4.
Seismic source.The mechanical impact source was a stan-
dard 5-kg sledgehammer, swung to strike the turned wall at
about waist height andat 90
tothe decline axis. Apiezoelectric

crystal sensor on the base of the hammer senses the moment of
impact and transmits a trigger pulse back to the seismograph
via a hammer cable.
The explosive source uses a single seismic detonator red in
a water-lled hole. Various detonator holding bars, including a
multishot gun, were tested. The ring line was attached to
a special control unit at the head of the tool and connected
at its other end to a Nissan capacitor discharge blaster that is
controlled by a synchronous timing unit. The blaster delivers
FIG. 4. Schematic illustration of underground seismic instrumentation.
over 10 A of current to ensure quick burning of the detonator
bridge wire. The detonator gun was deployed by means of a set
of 14 pushrods, each 3-m long and tted with special couplers
to allowquick joining and unjoining of sections. The rods were
distance marked to facilitate depth recovery to within a few
centimeters.
Receivers.A hydrophone array, or borehole eel, was de-
signed specically for this and similar experiments and was
constructed by Australian Sonar Systems. It is a 24-channel ar-
ray of 46-m aperture, connected to a 150-m leader cable. At a
diameter of 30 mm, the eel was sufciently narrowand rigid to
push into a horizontal or declined borehole of 45-mm diame-
ter to a depth of almost 200 m. The hydrophones are equipped
with preampliers and have a frequency range of 5020 00 Hz
and a dynamic range of 80 dB.
The array is provided with a front-end conditioner (FEC)
which provides power to the array preamplies, buffers the
array channels, and enables the level of the output to match
the input of the recording device.
The three-component geophones, assembled at Flinders
University, use high frequency (100-Hz) moving-coil elements.
The three pairs of leads, X, Y, and Z, connect to the takeouts of
a standard geophone cable. Each detector assembly is screwed
onto a PVC anchor which is grouted into the borehole. A long
wrenchenables rotationof the package toalign Z inthe vertical
direction.
Seismograph.ABison 7024 digital IFP 24-channel seismo-
graph, coupled to a Toshiba 1600 laptop computer, was used
for data acquistion. Both are powered by 12-V batteries. A
synchronous clock was used for instrument triggering with the
explosions.
All seismic records were obtained using a sample rate of
0.1 ms, which is the maximum possible with the Bison. The
record length was 100 ms, more than adequate to capture any
reectedwaves originatingfromwithinseveral hundredmeters
of the boreholes. The analog band-pass lter settings used on
the Bison were 32 Hz low cut and 2000 Hz high cut (antialias).
SIGNAL CHARACTERISTICS AND NOISE
Figure 5 is a sample hammer blow record from an impact
position at station G8. The record shows the direct P-, direct
FIG. 5. Sample hammer blow recording on hydrophone array,
from impact at G8. Trace separation is 2 m.
FIG. 6. Set of shot-gather crosswell hydrophone records for shots in hole H16-20 and receivers in hole H16-18. Shot-depth spacing
is 1 m, receiver spacing is 2 m.
S-, and tube-wave arrivals. The waves arrive rst at the closest
hydrophone station, and move out in time in accordance with
the source-receiver separation. The corresponding velocities
are 6, 3.5, and 1.2 km/s. The signals are quite strong and display
fairly high frequency (up to 1000 Hz).
Figure 6 is a sample set of shot-gather records fromthe cross-
hole experiment. Each panel represents a separate shot (depth
22.2 m, 23.2 m, etc.) recorded on 21 hydrophone receivers at
depths of 1 m, 3 m, . . . , 41 m. The rst arrival is the direct
P-wave. The record is full of other events which seemto move
out in both directions with the approximate speed of sound in
water (1400 m/s). This is best seen on receiver gathers (Fig-
ure 7), which reveal the origin of these events as the top, bot-
tom, and middle of the shothole. These secondary events are
tube wave to P and S conversions at each end of the hole,
and also from an impedance boundary (fracture zone) mid-
way along the hole. The top and bottom of the hole present
impedance barriers to the shock front set up from the actual
source in the hole. They behave as secondary sources of
body waves, which then radiate out into the medium and are
recorded on the receiver array, superimposed on the direct and
reected arrivals from the actual shot.
Thereis afamilyof tubewaves intheshothole(velocities 1.2
1.5 km/s), each of which undergoes mode conversion. The body
waves (fromtrue and secondary sources) can then excite tube-
wave arrivals inthe receiver hole frombothends as well as from
any fractures or other discontinuities in the hole. This makes
for a very cluttered record and represents a severe form of
coherent noise contamination. It is difcult to remove. Velocity
ltering is partially effective (due to low water-wave velocity
in the receiver-gather domain), but it can only be done if the
waves are not velocity aliased, which means a very close shot
spacing (say, 0.25 m). Unfortunately this was not the case with
the Hunt experiment. The tube-wave problem arises largely
from shooting in a water-lled hole. Explosions in dry holes
dont generate tube waves, but they dont generate much body-
wave energy either, because of the very poor coupling.
The detonator source produced high-amplitude, high-fre-
quency (up to 3000-Hz) signals. The high frequencies are
important for getting reections from small objects like ore
shoots, and for having the requisite resolution.
The 1-m shot and receiver interval used on the crosshole
survey, the 5-mgeophone spacing used on the triaxial VSP, and
the 2-m hydrophone interval/1.25-m impact interval used on
the hammer VSP were inadequate to combat the low-velocity
coherent noise, which became aliased at the high (kilohertz)
frequencies involved.
The seismic work at the Hunt mine was conducted mainly at
night. Background noise was plentiful, and included vibrations
from ventilation fans, pumps, surface drilling above, inductive
pickup, and vehicles operating nearby. On some occasions, the
noise was so high that the recording had to be discontinued.
Apart from the noise, the seismic survey was interrupted by
several instrument malfunctions.
TOMOGRAPHIC IMAGING
Procedure
Tomography is a 2-D image reconstruction procedure. The
image plane in the present context was formed by the two dip-
ping boreholes and the section of decline joining the two hole
collars. These three observation lines are not actually straight
nor perfectly coplanar because of the slight irregularity of the
tunnel and the drillholes. But as a starting point for tomo-
graphic inversion, we computed the best-t plane to the three
lines and projected the actual shot-receiver positions onto this
plane.
The input data comprised the rst arrival P-wave times,
which were picked interactively for every shot gather using
Seismic Unix software. Appropriate static corrections were
applied. Only the hydrophone data (crosshole and hammer
VSP records) were used in the reconstruction. More than 2300
FIG. 7. Set of receiver gathers for crosswell experiment, showing direct P- and S-waves plus tube-wave generated events at top,
bottom, and middle of shothole. Trace spacing is 1 m.
raypaths crisscross the image plane to yield good angular and
spatial coverage. The source and receiver positions along the
three arrays used to formthe U geometry are at roughly 1-m
spacing. The medium was divided up into cells of dimension
0.25 0.25 m.
The tomography procedure begins with a uniform(average)
velocity model and then iterates through using a multistage,
subspace method (Cao and Greenhalgh, 1995) in which the
long spatial wavelength velocity variations are solved for rst,
progressively adding inthe shorter wavelengthfeatures. Acon-
jugate gradienttype algorithmis at the heart of the model up-
date procedure, details of whichare givenby Zhouet al. (1992).
Ray tracing is accomplishedusing the minimumtraveltime tree
algorithm (Cao and Greenhalgh, 1993).
Modeling results
Figure 8 shows a simplied model, containing a faulted ore-
body sitting between two horizontal underground boreholes
60 m apart. The ore has low velocity compared to the rock. A
set of 24 shots and 24 geophones are placed in the boreholes
to simulate a seismic tomography experiment. The dimensions
andlayout were not chosentomatchthe actual experiment, but
merely to illustrate certain features. Shot and detector spacing
is 5 m. Raypaths are shown for rst arrivals. The synthetic ar-
rival times were then inverted by a DMLS tomography algo-
rithm(Zhou et al., 1992) to produce the tomogramof Figure 9.
The ore zones have been recovered, albeit with some blurring
and distortion due to the limited angular coverage. This model
clearly demonstrates the possibilities of tomography in cross-
well exploration.
Hunt Mine tomogram features
Figure 10 shows the nal tomogram obtained at the Hunt
mine. The color code at the side of the map denes the various
velocity values. The boreholes have been superimposed on the
tomogram to give the proper perspective (the crosses + de-
note collar positions in the decline).
The velocities range from about 5300 to 6400 m/s. From the
geological log of the holes we knowthat the rocktype is entirely
basalt, but the velocity eld is anything but homogeneous. The
inhomogeneity is presumably due to changes in alteration and
geochemistry of the basalt. The presence of quartz veins and
fractures will also affect the measured velocity values.
Ultrasonic velocity measurements were carried out on eight
core samples taken from the two holes to aid tomogram inter-
pretation. Thesamples includedthemajor changeinbasalt type
andcondition. Velocityresults aregiveninTable1. Themassive
basalt has a high velocity of 5900 m/s. Abundant disseminated
magnetite yields a lower velocity, as expected, through the ef-
fect of increased density. The ankerite-albite altered basalt has
a much lower velocity of just 2700 m/s. The massive quartz vein
has a velocity of 4700 m/s, similar to that of the amphibole-rich
basalt. These velocities are all lower than those obtained in situ
from the seismic tomography experiment. This can be largely
explained by the fact that the rock in situ is under considerable
lithostatic pressure. It is well known that pressure increases ve-
locity due to the closing of fractures and reduction of porosity.
FIG. 8. Synthetic 2-D crosshole seismic experiment showing
the orebody model and raypaths.
FIG. 9. Velocity tomogram reconstructed for the orebody
model of Figure 8.
FIG. 10. Tomographic reconstruction of the rst arrival VSP
and crosshole data at the Hunt mine.
If the relative velocity variations rather than the absolute
variations are considered along borehole H16-18, then the to-
mogram velocity variation is well matched by the ultrasonic
velocities. The high-, medium-, and low-velocity zones in the
tomogram seem to correlate well with the mineralogy and al-
teration of the basalt. This is an encouraging result, especially
given the fact that only single sample determinations for each
rock type were made. Besides random errors on single sam-
ples, which can be smoothed out by multiple sample measure-
ments, the laboratory sample may not be representative of the
rock behavior in situ because of the absence of macrofeatures
such as fractures. Alternatively, macrofeatures (such as 1-m
wide quartz veins) may have little effect on 6-m-long seismic
wavelengths in the eld, but multiple millimeter-wide fractures
in a laboratory sample may dramatically affect the velocity ob-
tained using a 1-cm-long ultrasonic wave.
It thus appears that seismic tomography at this location of-
fers a useful lithologic mapping tool. It is unfortunate that no
mineralization lies within the image plane. The boreholes dip
downwards, whereas the ore is in the stopes above the decline.
3-D REFECTOR IMAGING
Kirchhoff modeling of stope diffractions
The principal targets of the crosswell/VSP reection experi-
ments were the stope tunnels and residual ore, lying about 20 m
above the decline, as explained earlier (see Figures 2 and 3).
In order to assess the likely diffracting/reecting response of
sucha structure, tohelpidentify any reections inthe elddata,
and as a test of our 3-Dmigration algorithm, we computed the
acoustic response for the idealised model shown in Figure 11.
The model has very similar geometry to the Hunt B05 site.
Figure 11 shows the seismic observation plane dened by the
two boreholes and the decline. The stope tunnels have been
approximated by three bounded rectangles of width 4 or 5 m
and of xed length of 60 m. They have been assigned refection
coefcients of 1.
The modeling was done using a modied Kirchhoff sur-
face integral formulation (Sinadinovski, 1994). For noncoin-
cident source-receiver geometry, it yields the complete reec-
tion/diffraction response of the structure:specular returns, as
well as corner and edge diffractions.
Figure 12 show a portion of a set of shot gathers for shots
at 2-m intervals in borehole H16-20, starting at 2-m depth,
and detectors every 2-m in hole H16-18 over the depth range
Table 1. Ultrasonic velocities of core samples from Hunt
H16-18.
Longitudinal
Sample no. Rock description velocity (m/s)
1 Amphibolite rich basalt 4640
2 Massive basalt 5930
3 Massive basalt with abundant
disseminated magnetite 3990
4 Massive quartz vein 4690
5 Awaiting X-ray diffraction
results 5710
6 Epidote altered basalt 6130
7 Biotite-rich basalt and
disseminate pyrite 5240
8 Alteration zone: ankerite-
albite altered basalt 2680
240 m. The plotted quantity is
sgn(Z)
X
2
+Y
2
+ Z
2
,
where X, Y, Z are the three components of particle displace-
ment, in order to match the pressure waveforms of the cross-
hole hydrophone data. The synthetic seismograms, computed
for a source center frequency of 2 kHz, show the complex
diffraction response arising from the different parts of the
structure. The seismograms do not include the direct-wave ar-
rivals or the mode conversions. The P-wave reections exhibit
arrival times in the range 1320 ms. The amplitudes vary with
receiver position, and interference of the different diffraction
signals can also be observed.
Figure 13 is a comparisonof the model results withthe exper-
imental data for a source depth of 30 m. The reection signals
are visible in the observed data, but they are obscured by shear
waves and by tube-wave generated secondary arrivals.
Migration procedure
The reection imaging procedure is described by
Sinadinovski (1994). In simple terms, it is a prestack,
vector Huygens-Kirchhoff migration method.
A reector may be visualized as an ensemble of closely
spacedpoint scatterers inthe subsurface. The impulse response
of a point scatterer is a hyperbola, whose curvature is a func-
tion of the medium velocity. Therefore, summation (or inte-
gration) over the numerous hyperbolas appropriate for each
time sample of the input data set will add in-phase all diffrac-
tion signals and image all reectors. The summation opera-
tor must be properly weighted to honor the wave equation.
FIG. 11. Simplied model of the Hunt reection experiment,
showing the source-receiver observation plane (black) and the
overlying stope diffractors (approximated by rectangular scat-
terers).
FIG. 12. Set of synthetic shot gathers for the model of Figure 11 obtained by the Kirchhoff
integral method. Shots at 2-mintervals in hole H16-20; receivers at 2-mspacing in hole H16-18.
Note the complicated diffraction/reection response.
FIG. 13. Comparison of the synthetic diffraction response for model of Figure 11 (with a shot
at depth of 30 m) and the real crosswell seismograms for approximately the same recording
geometry.
The procedure involves breaking up the volume into small
cells and calculating the probability that a diffractor/reector
exists at each grid point by summing amplitudes over all shots
and geophones. The raypaths are rst traced in three dimen-
sions fromevery shot to every voxel and back to every receiver.
Thecorrespondingtraveltimes (for theassumedwavetype: PP,
PS, SP, SS) dene the search trajectory through the traces. For
three-component receiver data, the azimuth and inclination of
each computed ray permit the additional projection of the sig-
nal vector into the plane of the wavefront (S) or parallel to
the ray (P). Constructive interference will occur only for those
voxels at which seismic scattering take place.
We modied the procedure for the Hunt experiment in two
ways. First, we useda semblance-basedmeasure of trace energy
alignment by forming all of the possible cross products of sam-
ple amplitude over a 1020 point sliding time window for each
gather. Second, we performed a partially coherent migration
by adding only the absolute values for each shot gather, which
were coherently summed individually over time, receiver po-
sition, and vector component. The reason for doing a partially
rather than a fully coherent migration is to minimize the effect
of shot static-type timing errors and other randomuctuations.
The cell size used in the volume interrogation procedure
was a 1-mcube. The grid size was 60 60 15, or 54000 voxels.
Larger cells were also explored, but computer memory limita-
tion precluded an increased 3-D grid.
The data to be migrated must undergo some pretreatment,
such as three-component rotation, muting of direct P- and
S-wave arrivals (1020 sample suppression window), and gain
function application to boost the later arrivals. The input data
comprise the velocity eld, the source receiver coordinates, and
the eld seismograms. The process is repeated for the various
modes (PP, SS, etc.), if desired.
The output fromthe migration is a set of positive numbers at
each point in the 3-D grid, which represent the scatterer prob-
abilities. These numbers are color coded according to some
appropriate nonlinear scale (e.g., logarithmic), and the image
planes (horizontal slices through the volume) are then dis-
played at various depths.
Results
The data subsets used for the migration were the statically
corrected and pre-processed records from the triaxial VSP ex-
periment (80 shots, 8 receivers) and the crosswell experiment
(40 shots, 2 21 receivers). The individual data sets were mi-
grated separately, as well as being combined into a single mi-
gration. The attraction of a combined geometry is that the re-
ectors are illuminated from a greater range of angles.
Figure14is ahorizontal slicethroughthenal P-wavemigra-
tion 3-D volume at the 234 m elevation level. This elevation
corresponds to the approximate average position of the stope
oor across the survey area. The southwest stope tunnel actu-
ally drops in depth by about 4 m over a horizontal distance of
55 m between the two boreholes, from an elevation of 231 m
at the northern end to an elevation of 235 m in the south.
The northeastern stope oor maintains an elevation of 234 m
over its northern extent.
The warm colors (brown, red, yellow) in Figure 14, indicate
the strongest scatterer positions, whereas the cold colors (blue,
purple) indicate the absence of any reectors. Geographical
coordinates, as well as the projected position of the D zone
decline (20 m below the elevation shown), are given on the
map to aid location and identication of features. The migra-
tion result show a striking correlation with the known mine
workings (see Figure 1). The two green elongated zones of
moderate reectivity, which subparallel the decline, delineate
the stopes. The southwestern ridge bends in close to the decline
at its northern end, following the actual stope cavity. The re-
gion of strong reectivity, colored red, seems to coincide with
the far boundary of the northeastern tunnel. The reectivity
may well be enhanced by the curvature of the ore trough and
FIG. 14. P-wave migration result for VSP/crosshole data. The
map is a horizontal slice of the 3-Dvolume at the 234 mlevel.
The scatterers parallel the stope geometry.
FIG. 15. Migration map for horizontal slice of 230 m level.
The response is less intense than Figure 14 as the level rises
above the stopes.
the sharp lip of the structure, as seen in Figure 3. This would
cause a focussing of back-scattered energy, as observed in the
decline radar prole. The presence of residual, unmined ore on
this side of the stope (see Figure 3) may also be contributing
to the strong reection signal.
It is interesting to observe on the computer how the re-
ection (migration) pattern changes with elevation. Migration
slices at the 230 m (Figure 15) and the 240 m levels (not
shown), which are just above and below the stope oor and
roof elevations, respectively, yield reduced reectivity from
the maximum which occurs within the 237 to 236 m depth
range.
The nite time window used in forming the semblance, as
well as the nite wavelength of the signal (say, 6 m) means that
resolution is limited, and features get averaged, or smoothed in
the migration, over a spatial distance of at least several meters.
Migration images obtained at elevations of 226 and 246 m
(not shown) exhibit almost no structure, conrming that the
reectors are conned in depth to the vicinity of the stopes.
CONCLUSIONS
A high-resolution seismic hardware-software system has
been developed for crosshole/VSP imaging of orebodies and
associated hardrock geological structure, either within mines
or from the earths surface using available boreholes.
Although the seismic signals are of high frequency and high
quality, records are strongly contaminated by tube-wave and
mode-conversion noise. Dense arrays of sources and/or re-
ceivers are needed to enable proper preprocessing of the data
prior to imaging.
An example of an underground survey at the Hunt mine,
Kambalda, Western Australia, demonstrates the possibility of
both tomographic imaging to map velocity inhomogeneities,
andreectionimaging todelineate stope geometry. The seismic
maps show a striking correlation with the known geology.
The seismic images (especially reection) are difcult to ob-
tain and not easy to relate to mineralization. Companion elec-
trical images such as radar, radiowave imaging method (RIM),
and applied potential, are needed to aid interpretation, and are
likely to see increased use in mine development over the next
fewyears. The seismic method offers great promise in prospect
exploration. We are working on improvements to borehole
geophysical instrumentation and at developing more versatile
image reconstruction procedures.
ACKNOWLEDGMENTS
We acknowledge the support and cooperation of Peter
Fullagar, Peter Williams, and Greg Turner from Western Min-
ing Corporation in the conduct of this work.
REFERENCES
Adam, E., Milkereit, B., Arnold, G., and Pineault, R., 1996, Seis-
mic response of the Bell Allard orebody, Matogami, Quebec : 66th
Ann. Internal. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 634
637.
Bierbaum, S., and Greenhalgh, S. A., 1998, A high frequency down-
hole sparker sound source for crosswell seismic surveying : Expl.
Geophys., 29, 280283.
Cao, S., and Greenhalgh, S. A., 1993, Calculation of the seismic rst-
break time eld and its ray path distribution using a minimum trav-
eltime tree algorithm: Geophys. J. Internat., 114, 593600.
1995, Relative-error based non-linear inversion: Application to
seismic travel time tomography: Geophys. J. Internat., 121, 684694.
1997, Cross-well seismic tomographic delineation of mineral-
ization in a hard-rock environment: Geophys. Prosp. 45, 449460.
Carneiro, D., and Gendzwill, D., 1996, High resolution seismic reec-
tion survey in a deep gold mine: 66th Ann. Internat. Mtg., Soc. Expl.
Cowden, A., and Roberts, D., 1990, Komatiite hosted nickel sulphide
deposits Kambalda, in Hughes, F. E., Ed., Geology and Mineral De-
posits of Australia and Papua New Guinea: Aust. Inst. Min. Met.,
567581.
Dahle, A., Gjoystdal, H., Grammeltvedt, G., and Soyland Hansen, T.,
1985, Application of seismic reection methods for ore prospecting
in crystalline rock: First Break, 3, 916.
Duncan, G., Downey, M., Leung, L., and Harman, P., 1989, The de-
velopment of crosshole seismic techniques and case studies: Expl.
Geophys., 20, 127130.
Friedel, M. J., Jackson, M. J., Scott, D. E., Williams, T. J., and Olson,
M. S., 1995, 3-D tomographic imaging of anomalous conditions in a
deep silver mine: J Appl. Geophys. 34, 122.
Galperin, E. I., 1984, The polarisation method of seismic exploration:
D. Reidel Publ. Co.
Gendzwill, D. J., and Brehm, R., 1993, High-resolution seismic reec-
tions in a potash mine: Geophysics, 58, 741748.
Greenhalgh, S. A., and Bierbaum, S., 2000, Undergroung seismic re-
ection experiment in a gold mine: Expl. Geophys., 31, 321327.
Greenhalgh, S. A., and Mason, I. M., 1997. Seismic imaging with ap-
plication to mine layout and development, in Gubins, A., Ed., Geo-
physics and geochemistry at the millenium: Proc. Exploration 97
Fourth Decennial Internat. Conf. on Mineral Expl., 585598.
Gustavsson, M., Israelson, H., Ivansson, S., Moren, P., and Pihl, J., 1984,
Experiment with the seismic crosshole method in an iron mine: The
Leading Edge, 3, no. 11, 143145.
Harman, P., Leung, L., and Downey, M., 1987, Cross-hole seismic sur-
vey for mineral exploration in the West Kimberely area, Western
Australia: Expl. Geophys. 18, 8083.
Mutyorata, J. J., 1987, High resolution seismic reection an exploration
tool in an underground environmentExample from Zimbabwe: J.
African Earth Sci. 6, 109115.
Nelson, R. G., 1984, Seismic reection and mineral prospecting: Expl.
Geophys. 15, 229250.
Peterson, J. E., Jr., Paulsson, B. N. P., and McEvilly, T. V., 1985, Appli-
cations of algebraic reconstruction techniques to crosshole seismic
data: Geophysics 50, 15561580.
Pretorius, C. C., Trewick, W. F., and Irons, C., 1997, Applications of
3-D seismics to mine planning at Vaal Reefs gold mine, number
10 shaft, Republic of South Africa, in Gubins, A., Ed., Geophysics
and geochemistry at the millenium: Proc. Expl. 97 Fourth Decennial
Reid, A. B., Polome, L. G., and Greene, B. W., 1979. Ultra-high reso-
lution reection in chromite detection: Presented at the 49th Ann.
Internat. Mtg., Soc. Explor. Geophys.
Ruskey, F., 1981, High-resolution seismic methods for hard rock min-
ing, in Premining investigation for hardrock mines: Proc. U.S. Bur.
Mines Technology Transfer Seminar, 428.
Schmidt, G., 1959, Results of underground seismic reection investi-
gations in the siderite district of the Siegerland: Geophys. Prosp. 7,
287290.
Sinadinovski, C., 1994, Seismic imaging with application to mine layout
and development: Ph.D. thesis, Flinders University of South Aus-
tralia.
Sinadinovski, C., Greenhalgh, S. A., and Mason, I. N., 1995, Three-
dimensional reector imaging of in-mine high frequency crosshole
seismic data: Expl. Geophys. 26, 325330.
Spencer, C., Thurlow, G., Wright, J., White, D., Caroll, P., Milkereit,
B., and Reed, L., 1993, A Vibroseis reection seismic survey at the
Buchans Mine in central Newfoundland: Geophysics 58, 154166.
Wachsmuth, C., and Schmidt, E., 1962, Experience with the reection
seismograph in a mine within a slat plug: Geophys. Prosp. 10, 491
496.
Wong, J., Hurley, P., and West, G. F., 1984, Crosshole audio-frequency
seismology in granitic rocks using piezoelectric transducers as
sources and detectors: Geoexplor. 22, 261280.
Wright, C., Wright, J. A., and Hall, J. 1994, Seismic reection tech-
niques for base metal exploration in eastern Canada: Examples from
Buchans, Newfoundland: J. Appl. Geophys., 32, 105116.
Young, R. P., Hutchins, D. A., McGaughey, J., Towers, J., Jansen, D.,
and Bostock, M., 1989, Geotomographic imaging in the study of
mine induced seismicity: Pure Appl. Geophys., 129, 571596.
Zhou, B., Greenhalgh, S. A., and Sinadinovski, C., 1992, Iterative in-
verse techniques for DMNLS in seismic tomography, in Asakura,
N., Ed., Geotomography, II: Soc. Explor. Geophys. Japan, 111128.
Radio tomography and borehole radar delineation of the McConnell
nickel sulde deposit, Sudbury, Ontario, Canada
Peter K. Fullagar
, Dean W. Livelybrooks
, Ping Zhang
, Andrew J. Calvert
,
and Yiren Wu
ABSTRACT
In an effort to reduce costs and increase revenues
at mines, there is a strong incentive to develop high-
resolution techniques both for near-mine exploration
and for delineation of known orebodies. To investi-
gate the potential of high-frequency EM techniques
for exploration and delineation of massive sulde ore-
bodies, radio frequency electromagnetic (RFEM) and
ground-penetrating radar (GPR) surveys were con-
ducted in boreholes through the McConnell massive
nickelcopper sulde body near Sudbury, Ontario, from
19931996.
CrossholeRFEMdatawereacquiredwithaJW-4elec-
tric dipole system between two boreholes on section
2720W. Ten frequencies between 0.5 and 5.0 MHz were
recorded. Radiosignals propagatedthroughtheSudbury
Breccia over ranges of at least 150 m at all frequen-
cies. The resulting radio absorption tomogram clearly
imaged the McConnell deposit over 110 mdowndip. Sig-
nal was extinguished when either antenna entered the
sulde body. However, the expected radio shadow did
not eventuate when transmitter and receiver were on
opposite sides of the deposit. Two-dimensional model-
ing suggested that diffraction around the edges of the
sulde body could not account for the observed eld
amplitudes. It was concluded at the time that the sulde
body is discontinuous; according to modeling, a gap as
small as 5 mcould have explained the observations. Sub-
sequent investigations by INCOestablished that pick-up
in the metal-cored downhole cables was actually respon-
sible for the elevated signal levels.
Bothsingle-hole reectionproles andcrosshole mea-
surements were acquired using RAMACborehole radar
systems, operating at 60 MHz. Detection of radar reec-
tions from the sulde contact was problematic. One co-
herent reection was observed from the hanging-wall
contact in single-hole reection mode. This reection
could be traced about 25 m uphole from the contact.
In addition to unfavorable survey geometry, factors
which may have suppressed reections included host
rock heterogeneity, disseminated suldes, and contact
irregularity.
Velocity andabsorptiontomograms were generatedin
the Sudbury Breccia host rock from the crosshole radar.
Radar velocity was variable, averaging 125 m/s, while
absorption was typically 0.8 dB/mat 60 MHz. Kirchhoff-
style 2-D migration of later arrivals in the crosshole
radargrams dened reective zones that roughly parallel
the inferred edge of the sulde body.
The McConnell high-frequency EM surveys es-
tablished that radio tomography and simple radio
shadowing are potentially valuable for near- and in-mine
exploration and orebody delineation in the Sudbury
Breccia. The effectiveness of borehole radar in this par-
ticular environment is less certain.
INTRODUCTION
Borehole EMmethods are being applied increasingly in and
near metalliferous mines for both exploration and orebody
Manuscript received by the Editor May 10, 1999; revised manuscript received June 16, 2000.
Fullagar Geophysics Pty. Ltd., Level 1, 1 Swann Road, Taringa, Queensland 4068, Australia. E-mail: p.fullagar@mailbox.uq.edu.au.
University of Oregon 1274, Department of Physics, Eugene, Oregon 97403-1274. E-mail: dlivelyb@hendrix.uoregon.edu.
Electromagnetics Instruments, Inc., P.O. Box 463, El Cerrito, California 94530-0463.

Simon Fraser University, Department of Earth Sciences, 8888 University Drive, Burnaby, British Columbia V5A 1S6, Canada.
Beijing Xin Yi Hightech Research Institute 70 Bei Lishi Road, Beijing 100037, China.
delineation. In the context of near-mine exploration for base
metals, downhole transient EM methods such as UTEM have
proved effective for detecting deep orebodies at ranges of hun-
dreds of meters from boreholes (e.g., King, 1996). At the same
1920
Radio and Radar Delineation 1921
time the drive for more efcient mine production techniques
to reduce costs and increase revenues has stimulated inter-
est in high-frequency borehole EM techniques. Radio imaging
and ground-penetrating radar (GPR) methods showparticular
promise for orebody delineation.
The use of radio waves to dene geological features be-
tween boreholes can be traced back to a 1910 German patent
(Thomson and Hinde, 1993). Radio tomography (RT) per se
was pioneered by Lager and Lytle (1977) and has since been
applied extensively to detect faults and other disruptions in
the continuity of seams of coal or potash (McGaughey and
Stolarczyk, 1991; Vozoff et al., 1993). The utility of radio imag-
ing in metalliferous exploration and mining is currently un-
der investigation. Radio frequency methods in the 10 kHz to
1 MHz band have already enjoyed some success in this context,
e.g., Nickel and Cerny (1989), Anderson and Logan (1992),
Thomson et al. (1992), Wedepohl (1993), Zhou et al. (1998),
and Stevens and Redko (2000). RT can deliver higher resolu-
tion than traditional borehole EM systems, both by virtue of
its higher frequencies and because the radio transmitter can be
lowered down a borehole, closer to the target.
GPR detects changes in permittivity and conductivity us-
ing high-frequency electromagnetic pulses (10 MHz1 GHz).
Conventional GPR has found application in underground coal
mines (Coon et al., 1981; Yelf et al., 1990) and is used routinely
to dene auriferous zones in the Witwatersrand (Campbell,
1994) and at the Sixteen to One mine in California (Raadsma,
1994). Other mining applications include mapping conductive
salt structures (Stewart and Unterberger, 1976), exploring for
placer deposits (Davis et al., 1985), detecting geotechnical haz-
ards (Fullagar and Livelybrooks, 1994a), and delineating lat-
eritic deposits. Borehole radar is well established in salt and
potash mines (Mundry et al., 1983; Eisenburger et al., 1993;
Thierbach, 1994), but application in nonevaporite mines is still
relatively uncommon, notwithstanding the strong commercial
incentive to accurately dene ore boundaries and structures.
Encouraging experimental applications of borehole reection
radar have been reported from coal mines (Murray et al.,
1998), Witwatersrand gold mines (Wedepohl et al., 1998), and
base metal sulde mines (Liu et al., 1998; Zhou and Fullagar,
2000).
This article summarizes activities undertaken at the
McConnell nickel sulde deposit from 19931996 by a re-
search group formed under the NSERC/TVX Gold/Golden
Knight Chair in Borehole Geophysics for Mineral Exploration
at Ecole Polytechnique, Montreal. Borehole radar trials were
performed with a 22-MHz RAMAC I system in December
1993 (Fullagar and Livelybrooks, 1994b; Stevens and Lodha,
1994) and with a 60-MHz RAMAC LI system in June 1996
(Calvert and Livelybrooks, 1997). Both single-hole reection
and crosshole data were recorded. A radio imaging survey was
undertaken in April 1994 with a JW-4 electric dipole system
(Fullagar et al., 1996). Data were recorded at ten frequencies
between 0.5 and 5.0 MHz.
The McConnell deposit, owned by INCO Ltd., is located
near Garson mine on the southeastern rim of the Sudbury
basin, Ontario, Canada. It is a tabular body of massive sul-
des (pentlanditepyrrhotite), about 200 malong strike, 300 m
downdip, and averaging 15 m true thickness (Figure 1). The
mineralization is hosted by a quartz diorite dyke, intruded into
heterogeneous Sudbury Breccia country rock, comprised of
blocks of metavolcanics and metasediments. The dyke strikes
85
east and dips 70
south.
The McConnell sulde body is shallow, extensively drilled,
and close to an operating mine, rendering it an excel-
lent geophysical test site. Prior to the work described here,
the Geological Survey of Canada (GSC) had recorded
a comprehensive suite of downhole geophysical logs at
McConnell (Mwenifumbo et al., 1993; Killeen et al., 1996). The
dc resistivities varied over four orders of magnitude, fromnear
zero in the massive suldes to over 30 000 ohm-m within the
breccia. The high resistivities of the host rocks rendered them
favorable for propagation of EM waves, while the high con-
ductivity of the massive sulde body was expected to produce
FIG. 1. McConnell deposit shown (a) in plan and (b) on section
2720W. Top of the sulde body has been projected to the sur-
face in (a). Transmitter locations for the RTsurvey are marked
with dots in (b); thick black lines mark massive sulde inter-
sections. Westings in (a) are in feet; depths in (b) are in meters.
1922 Fullagar et al.
strong radar reections and to create radio shadow zones. The
McConnell site therefore represented an ideal location to as-
sess the efcacy of radio tomography and borehole radar for
delineating a known massive sulde body.
THEORY
For EMwaves propagating at angular frequency through a
homogeneous isotropic medium with conductivity , real per-
mittivity
r
, and permeability , the real and imaginary parts,
and , of the complex wavenumber k can be written as (Cook,
1975)
=
r
2
(
1 + Q
2
+1)
1/2
(1)
and
=
r
2
(
1 + Q
2
1)
1/2
. (2)
The value Q is the ratio of displacement current to conduction
current,
Q =

r
e
, (3)
where
e
is the effective conductivity (Turner and Siggins,
1994),
e
= +
i
, (4)
with
i
denoting the imaginary part of permittivity.
The scalar wavenumber determines the wavelength, while
the absorption coefcient controls attenuation. For both ra-
dio imaging and GPR surveys, the effective range increases
with decreasing frequency, but at the expense of resolution.
Thus, any survey involves a trade-off between resolution and
range.
Radio imaging involves propagation of monofrequency EM
signals inthe 1 kHz to10 MHz bandbetweenboreholes or mine
accessways. In a typical crosshole RT survey, the transmitter is
xed in one hole while signal amplitude (and perhaps phase)
is recorded at successive receiver stations in one or more other
holes. Signals which are weak and retarded in phase are indica-
tive of more conductive material between the transmitter and
receiver. Tomographic reconstruction of the signal amplitudes
and phases at all receiver stations provides an image of the
conductivity distribution between the boreholes.
GPR involves emission, propagation, and detection of EM
pulses, with center frequencies usually between 10 MHz and
1 GHz. GPR can be deployed in transmission mode, like radio
imaging, but it is more commonly applied in reection mode.
In the latter case it is closely analogous to the seismic reection
method. GPRcanmapcontrasts inelectrical properties or uid
content which usually coincide with geological contacts and
structures. Radar reectioncoefcients dependonpermittivity,
conductivity, and permeability contrasts.
Most radar surveys are carried out under low-loss conditions
(Q 1), for which the propagation involves polarization of the
medium. Typical rocks are characterized by relative dielectric
constants between 3 and 30. The presence of water, with its
large relative dielectric constant (80), serves to dramatically
retard propagating waves.
The demarkation between GPR and RFEM applications is
not sharp, but generally radio imaging is used at frequencies
between 1 kHz and 10 MHz, often in lossy (low Q) environ-
ments. RFEM equipment can operate effectively in the dif-
fusion domain Q 1, where conduction currents are at least
comparable to, and often much larger than, displacement cur-
rents. In the limit of low Q, and assume the same functional
form:
= =
e
2
. (5)
McCONNELL RADIO FREQUENCY EM SURVEY
RFEM data acquisition
Tomographic radio frequency surveys were undertaken be-
tween boreholes 80 578 and 78 930 in April 1994 using a JW-4
borehole radio imaging systemdeveloped by the Chinese Min-
istry of Geology and Mineral Resources (MGMR). The JW-4 is
an electric eld system, capable of recording axial component
amplitude over a programmable sweep of frequencies between
0.5 and 32 MHz (Qu et al., 1991). At McConnell, a suite of ten
frequencies was read between 0.5 and 5 MHz at 0.5-MHz in-
tervals.
Tomographic coverage was 90170 min hole 78 930 and 130
310 m in hole 80 578, i.e., in both the hanging wall and footwall
of the sulde deposit (Figure 1b). The transmitter sites were
10 m apart, and the receiver station interval was 2 m.
A center-fed half-wave dipole transmitter with 18-m arms
was deployed in hole 78 930 on the rst day, but a monopole
antenna with a 7-m arm was used in hole 80 578 on the second
day. Water inltration into the transmitter lter pod prompted
the change. Signal strength was satisfactory at all frequencies at
ranges up to 150 m. The receiver was a monopole, with a 7-m
arm, throughout. Analog to digital conversion was affected
downhole in the receiver electronics pod, which is 1.2 m long
and has an outside diameter of 40 mm.
A limited repeatability test was conducted during ascent of
hole 80 578 at the McConnell site, with the transmitter at 30 m
depth in hole 78 930 (Figure 1b). The greatest difference in
repeat readings was only 0.6 dB at 0.5 MHz. In conjunction
with a similar repeatability test performed at other sites, these
comparisons engendered condence in data precision.
The transmitter was operated from each hole in turn to per-
mit a reciprocity test. However, given the change in transmit-
ter and in view of the gradual inltration of water into the
transmitter on the rst day, there is no reason to expect reci-
procity to apply exactly. The amplitude differences when the
positions of the transmitter and receiver were interchanged
were indeed substantial, typically between 4 and +4 dB at
5 MHz.
RFEM data reduction and image construction
In RFEMsurveys, the amplitude of a single component of ei-
ther electric or magnetic eldis measuredat the receiver. Toto-
mographically reconstruct the conductivity structure between
boreholes, far-eld conditions are assumed. For an electric
dipole in a homogeneous and isotropic medium, the received
electric eld strength in the far-eld is given by (Ward and
Hohmann, 1988, p. 173)
|E
f
(r)| = A
0
e
sin
t
sin
r
r
, (6)
where r is the transmitter-receiver separation, A
0
is the source
strength, and
t
and
r
are the polar angles of the ray with
respect to the transmitter and receiver axes (Figure 2). Treating
absorption as an analog for slowness, the apparent attenuation
or traveltime,
n
, for the nth ray is dened by
n
=
C
n
(x, y, z) dl, (7)
where integration is along the raypath C
n
. Taking logarithms
of equation (6), it follows from equation (7) that can be ex-
pressed as
=
a
r = 20 log
10
|E
f
| +20 log
10
A
0
+20 log
10
sin
t
sin
r
r
, (8)
where
a
is the apparent absorption coefcient in decibels per
meter.
The three main sources of error for were (1) observational
error, (2) uncertainty in A
0
, and (3) near-eld effects. From
our limited repeatability tests, described above, we concluded
that observational error is small (1 dB). This established data
precision but not accuracy.
Generally, A
0
is estimated via analysis of data collected in
homogeneous host rocks. After accounting for geometrical ef-
fects, A
0
is the intercept on the amplitude versus distance plot.
Transmitter sites at depths 130, 140, and 150 m in hole 80 578,
far from the massive sulde contact, were selected for estima-
tion of A
0
. After minimum l
1
-norm regression (Fullagar et al.,
1994), A
0
values of 0.38, 0.38, and 0.40 were obtained for the
three transmitter gathers. A starting estimate for the apparent
absorption coefcient was computed assuming
e
was 10
4
S/m
and
r
was 6.5
0
, in keeping with the best available informa-
tion for Sudbury resistivities (Mwenifumbo et al., 1993) and
permittivities (Fullagar and Livelybrooks, 1994a). The details
FIG. 2. Crosshole survey schematic, where
t
and
r
are the
polar ray angles at the transmitter and receiver, respectively.
of the l
1
-norm minimization are given by Fullagar et al. (1994).
An A
0
value of 0.38 was ultimately adopted.
RFEM transmitter performance is strongly affected in or
near sulde mineralization. Since the transmitter strength is
not identical at all locations, its output ideally should be contin-
uously monitored. The JW-4 usually monitors transmitter input
current, but this facility was inoperable for the McConnell sur-
vey because no suitable three-conductor cables were available
on site.
If the source strength is assumed constant, a static shift error
is introduced in the tau values at each transmitter site if source
strength varies. McGaughey (1990) suggests a simple approach
for estimation of such static shifts by enforcing reciprocity. An
implementation of this approach is described by Fullagar et al.
(1994). It was not applicable at McConnell because reciprocity
did not apply. More recently, Cao et al. (1998) proposed fre-
quency differencing as a means to suppress variations in source
strength caused by large changes in conductivity near the
boreholes. This approach may be suitable for the McConnell
data.
Near-eld effects can become important as frequency de-
creases, resistivity increases, path length decreases, or trans-
mitter polar angle decreases (Pears, 1997). To assess the va-
lidity of the far-eld assumption for the McConnell data, the
differences between the full analytic solution and the far-eld
approximation for an electric dipole in a whole space were
compared for transmitters and receivers distributed in space
as at McConnell. In a medium with effective of conductivity
of 10
4
S/m and dielectric constant of 6.5
0
, the wavelength at
0.5 MHz is 235 m, larger thanthe longest raypathat McConnell.
Therefore, far-eld conditions did not apply at the lower fre-
quencies.
The relative error between far-eld and exact amplitudes
was computedfor anelectric dipoleinahomogeneous medium,
where
=
|E| |E
f
|
|E|
. (9)
For the McConnell survey geometry, the maximum relative er-
ror in homogeneous host was estimated to be 5% at 500 kHz.
This corresponds to an error of less than 1 dB. The McConnell
tomogram presented here corresponds to a frequency of
5 MHz, for which far-eld conditions certainly apply.
To invert equation (7), the image plane is discretized and
the line integral on the rightside is represented as a summa-
tion (e.g., Stewart, 1991). Tomographic reconstruction then re-
duces to the solution of a system of linear equations for the
unknown absorption coefcients within each of the grid cells.
Migratom, a SIRT algorithm developed by the U.S. Bureau
of Mines (Jackson and Tweeton, 1994), has been used here,
assuming straight rays.
To perform a ray-based tomographic reconstruction, it is as-
sumed that the amplitude at the receiver is governed solely
by absorption of the signal and that reection, scattering, and
diffractioneffects canbe ignored. Thus, there is animplicit low-
contrast assumption. If a highly conductive body intervenes
between the transmitter and receiver, there will be no direct
transmission and the inferred absorption coefcient after to-
mographic reconstruction will not be an accurate indication of
the actual absorption coefcient of the conductor.
RFEM interpretation
The absorption tomogram generated from the 5-MHz
McConnell data using a modied form of Migratom is pre-
sented as Figure 3. The starting model was homogeneous, and
clamping weights (Pears and Fullagar, 1998) were applied. De-
spite the fact that reciprocity is assumed implicitly during to-
mographic reconstruction, the tomogrambasedonthe full data
set was superior (insofar as it provided better denition of the
sulde body) than that obtained when the data acquired with
the symmetric transmitter in hole 78 930 were disregarded.
The tomographic reconstruction clearly denes the contin-
uation of the sulde body over a distance of 110 m between
intersections in the two boreholes. However, the inferred ab-
sorption is not uniform along the conductor, and irregularities
in the contacts appear to be related to the discretization of the
image plane.
When the transmitter and receiver are on opposite sides of
the sulde body, the signal amplitude is low but measurable.
Propagation of radio signals through a 15-m-thick massive sul-
de body is impossible at megahertz frequencies. Nevertheless,
the expected radio shadow at McConnell did not eventuate. It
was concluded at the time that the signals received on the other
side of the conductor either travelled around the edges of the
deposit or passed through a gap in the suldes.
Diffraction around the edges of the sulde body is a plausi-
ble explanation for the relatively high signals since the dimen-
sions of the McConnell deposit (approximately 300 200 m)
are comparable to the estimated radio wavelength (230 m) at
500 kHz in the Sudbury Breccia and much larger than the es-
timated wavelength (22 m) at 5 MHz. To determine whether
FIG. 3. Radio absorption tomogram at 5 MHz between holes
78 930 and 80 578. Transmitter locations for the RT survey are
marked with dots.
diffraction alone could explain the observations, 2.5-D nite-
difference modeling was performed using the EMSUN pro-
gram (Smith et al., 1990). The model used to characterize
diffraction around the bottom of the slab is depicted in Fig-
ure4. Themodel conductor is twodimensional (inniteinstrike
length), but the source is nite (vertical electric dipole). Except
for the 2-D limitation, the cross-sectional geometry of the Mc-
Connell experiment was replicated as closely as possible for
two transmitter positions, T1 and T2, at depths 170 and 190 m
in hole 80 578 and two receiver positions, R1 and R2, at depths
108 and 168 m in hole 78 930.
The orebody model was assigned a conductivity of 100 S/m
at its core, enclosed within an inner layer of conductivity 1 S/m
and an outer layer of conductivity 0.01 S/m. The transitional
layers were introduced for numerical stability, since a contrast
of 1:100 was regarded as the greatest which could be handled
accurately. Geologically, such a transition zone often exists in
the form of a disseminated sulde halo; indeed, at McConnell
sulde is known to occur as blebs in the quartz diorite and
as disseminations in the breccia immediately adjacent to the
quartz diorite (Grant and Bite, 1984).
The observed and modeled ratios of electric eld amplitude
at the two receiver sites for the two transmitter positions at
500 kHz are recorded in decibels on Figure 4. For a contin-
uous conductor (no gap), modeling results indicate that after
diffraction around the bottom of the slab the signal amplitude
is 61 dB lower at R2 on the far side of the conductor than at
R1 on the near side for the transmitter at T1 and 54 dB lower
FIG. 4. A 2-D sectional model adopted to investigate diffrac-
tion around the bottom of the McConnell deposit.
when the transmitter is at position T2. Even if the signal at
R2 is increased by a factor of three to allow for diffraction lat-
erally around the ends of the tabular body, the signal at R2
would still be 51 or 44 dB down relative to that at R1 (note
the 15 dB or 16 dB differences observed). Therefore, diffrac-
tion around the edges of the deposit cannot fully explain the
relatively high eld strengths recorded when the transmitter
and receiver are on opposite sides of the conductor. Numeri-
cal instability precluded 5-MHz modeling; however, at higher
frequencies diffraction would be a less effective mechanism.
In light of the modeling, a break (or hole) in the sulde body
was proposed as the explanation for the absence of a shadow
(Fullagar et al., 1996). Geologically, there is a hint of a break in
continuity of the sulde in hole 80 555, which passed through a
small sulde interval (188190 m) and then back into Sudbury
Breccia (190197 m) before passing through the main sulde
interval (Figure 1b). By introducing a 5-m-wide break in the
2-D model in approximately this position, the predicted ra-
tios of near-side to far-side signal amplitudes increase to levels
comparable with those observed, even though the gap was a
small fraction (2%) of a wavelength in the breccia at 500 kHz
(Figure 4).
More recently, INCO geophysicists have demonstrated that
the radio signal strength drops substantially if the metal-cored
receiver cable is replaced with reinforced ber-optic cable
(McDowell and Verlaan, 1997). Thus, we now conclude that
the JW-4 in-line lters were inadequate to prevent cable pick-
up, i.e., that the receiver and transmitter cables were acting as
parasitic antennae during the 1994 survey. It is therefore not
necessary to invoke a gap or hole in the sulde body to explain
the anomalously high amplitudes in the RT data.
McCONNELL BOREHOLE RADAR
Radar data acquisition and processing
The borehole radar data described here were recorded
in 60-mm-diameter boreholes in June 1996 with the Ecole
Polytechnique omnidirectional 60-MHz RAMAC LI system.
Both single-hole and crosshole congurations were used. The
transmitter was positioned automatically using a computer-
controlled electric winch.
The RAMAC system is comprised of a control unit con-
nected via optical bers to a transmitter pulser, amplier, and
antenna and to a receiver antenna, amplier, trigger, and A/D
converter (Olsson et al., 1992). The transmitter antenna emits
a short pulse of approximately three half-cycles duration. The
received wavetrain for each transmitterreceiver pair is con-
structed over a window, with only one time sample recorded
each time the transmitter res. For each of 512 time lags, 128
measurements are stacked.
Although the radar data are single fold, the processing has
nevertheless been adapted from seismic processing techniques
drawn from a number of sources, including Yilmaz (1987),
Fisher et al. (1992), Liner and Liner (1995), and Young et al.
(1995). Somecaremust beexercisedwhenborrowingfromseis-
mic technology since radar signals exhibit greater attenuation
and dispersion than seismic waves.
A dc drift correction, entailing subtraction of the average
value for each record prior to the rst arrival, was performed
on all traces. Subsequent processing steps included spreading
and exponential compensatory (SEC) gaining (Annan, 1993),
frequency-domain band-pass ltering, and spectral equaliza-
tion. Processed crosshole reection radargrams were migrated
in an effort to delineate radar reectors.
Radar tomography
Crosshole 60-MHz radar data were acquired between bore-
holes 78 930 and 78 929 in 1996. For each of 18 transmitter posi-
tions in hole 78 929, data were recorded at 25 receiver depths in
hole 78 930. Transmitter and receiver spacings were both 2.5 m,
comparable to the expected radar wavelength. The crosshole
survey was originally designed to image the radar velocity and
attenuation above the ore deposit, primarily to gauge the ab-
sorption and velocity heterogeneity of the Sudbury Breccia.
The scanning pattern is not suitable for imaging the ore bound-
aries.
Straight-ray tomographic inversion of rst-arrival times was
performedusingMigratom. Thevelocitytomogram(Figure5a)
reveals a fairly high degree of heterogeneity within the breccia.
Alobe of lowvelocity extending down fromhole 78 929 toward
hole 78 930 may correspond to the low-resistivity zones dened
in the borehole logs: 8895 min 78 929 and 114118 min 78 930
(Figure 6). The low velocity and low resistivity might indicate
water-lled fractures.
FIG. 5. (upper) Hanging-wall radar velocity tomogram be-
tween boreholes 78 929 and 78 930. Text gives details regarding
the tomographic inversion process. Radar velocities are in me-
ters per microsecond. (lower) Radar absorption tomogram for
the same region. Absorption coefcient in decibles per meter.
First-arrival amplitudes were corrected for spherical spread-
ing and for a dipolar radiation pattern, as per equation (8),
prior to SIRT reconstruction. The resulting absorption to-
mogram is shown in Figure 5b. The radar absorption co-
efcient in the breccia is typically 1 dB/m at 60 MHz. In
general, velocity (/) and absorption are anticorrelated, as
expected from equations (1) and (2), when attenuation is
appreciable.
Single-hole reection GPR imaging
The 1996 single-hole reection surveys were undertaken in
holes 78 929 and 78 930. The transmitter and receiver were sep-
arated by 4.7 m. Coverage was from 75 to 130 m in hole 78 929
and from 75 to 170 m in hole 78 930 at a station spacing of
0.25 m. The center frequency of the received radar signals was
approximately 48 MHz.
An SEC gain was applied to the radargrams. A velocity of
125 m/s (the average fromthe velocity tomogram) and an ab-
sorptioncoefcient of 0.75 dB/m(basedonthe attenuationrate
of trace envelopes) were adopted to specify the gain function.
A band-pass lter with corner frequencies of 30 and 75 MHz
was applied to all traces to suppress long-period uctuations
and late-time drift. Spectral equalization (Young et al., 1995),
which attens the spectrumin a manner akin to deconvolution,
was also applied.
Reection data were recorded above, within, and below
the massive sulde mineralization, and data amplitudes vary
markedly. The resulting radargrams for boreholes 78 929 and
78 930arepresentedinrawandprocessedforminFigure6, with
geology and resistivity logs (Mwenifumbo et al., 1993). The re-
sistivity logs are clipped at around 30 000 ohm-m because of
loss of sensitivity (negligible current).
A clear radar reection from the massive sulde hanging-
wall contact was observed in hole 78 929 (Figure 6a). How-
ever, there is no obvious reection from the footwall contact
in78 929 nor fromeither contact inhole 78 930 (Figure 6b). This
was a surprising and, from a practical viewpoint, disappointing
observation. The main factors responsible for the absence of
reliable reections are probably (1) unfavorable endre ori-
entation of the antennae with respect to the contact, (2) het-
erogeneity of the breccia, (3) local irregularity of the sulde
contact, and (4) minor concentrations of disseminated sulde
in and immediately adjacent to the quartz diorite.
There is total extinction of signal when either antenna is
within the sulde body. In addition, there are highly absorptive
intervals and zones of low resistivity in the hanging wall in
both holes. The correlation between low resistivity and low
radar amplitude is by no means perfect as plotted in Figure 6,
but a depth shift of approximately 5 m would achieve a close
correspondence. A systematic depth error in one or the other
data set is suspected.
The combination of SEC gain and spectral equalization re-
covered interpretable signal in the low-amplitude intervals.
The intervals of high absorption within the breccia are not
correlated with lithology and may signify uid-lled fracture
zones. Weak reections, coherent for tens of meters, can be
discerned elsewhere in the breccia. These are presumably from
fractures or lithological boundaries.
Crosshole radar reection
The crosshole rst arrivals carry little or no information on
the deposit geometry. However, an attempt has been made
to focus crosshole reections from the sulde contact into an
image via migration (Calvert and Livelybrooks, 1997).
The survey geometry is unfavorable for reception of reec-
tions from the sulde, given that the dipole antennae are in
almost endre orientation with respect to the hanging-wall
contact (Figure 1b). Reection amplitudes would be further
reduced by attenuation when one or both of the antennae are
far from the sulde contact. When both transmitter and re-
ceiver antennae are close to the suldes, the reection from
the contact would be difcult to distinguish from the direct
arrival. Survey geometry also dictates that any crosshole re-
ections from the contact will originate from a relatively small
area straddling the point where hole 78 929 pierces the sul-
de body (Belleeur and Chouteau, 1998). From a practical
viewpoint, therefore, theanalysis of crosshole reections is des-
tinedtoprovide limitednewinformationabout the shape of the
contact.
After suppression of the direct arrivals and application of
geometric spreading correction, the data were migrated using a
2-DKirchhoff-style algorithm. Aconstant velocity of 105 m/s
was assumed. The migrated data are depicted in Figure 7.
Reective zones between the boreholes have been identi-
ed. Those in midpanel, near true depth 60 m, could be from
lithological contrasts within the breccia. Of greater interest is
the inferred reectivity trend paralleling the sulde contact.
Given the assumptions underlying the migration (2-D geome-
try, constant velocity, spherical radiation pattern), this result
is reasonably encouraging. The recording window was only
623 ns; some improvement could be expected if data were
recorded over a longer interval.
Belleeur and Chouteau (1998) have achieved an improved
result migrating these same data by (1) restricting reector dip
to a particular range, (2) suppressing down-going waves, and
(3) adopting the velocity tomogram as the migration velocity
model. In their migrated image, the strongest reection is co-
incident with the sulde contact.
DISCUSSION
Investigation of the efcacy of radio tomography and bore-
hole radar for delineation of massive sulde bodies com-
menced at McConnell in 1993. The impetus for the research
was the need for accurate geophysical denition of ore con-
tacts, both during near- and in-mine exploration and for ore-
body delineation.
Radiofrequency tomography witha JW-4 electric dipole sys-
tem successfully imaged the McConnell sulde deposit over a
down-dip distance of 110 m. Data were recorded at ten fre-
quencies between 0.5 and 5.0 MHz. Signals propagated over
150 m through Sudbury Breccia host rock, even at 5 MHz. The
sulde deposit shape was well dened on the resulting tomo-
gram as a zone of high absorption. Thus, the rst RT survey
at McConnell demonstrated the potential of radio imaging to
map sulde bodies in highly resistive host rocks. The results
were sufciently encouraging to prompt further RT investiga-
tions in the Sudbury area by INCO (McDowell and Verlaan,
1997) and, more recently, by Falconbridge (Stevens andRedko,
2000).
There was complete extinction of radio signal when either
antenna was located within the massive suldes. However, sig-
nal amplitudes were considerably stronger thanexpectedwhen
the transmitter and receiver were on opposite sides of the sul-
debody. The2.5-Dmodelingindicatedthat diffractionaround
the edges of the conductor was inadequate to explain the ob-
served signal levels, and the conclusion at the time pointed to
a break in continuity of the suldes. According to the model-
ing, a 5-m-wide gap is compatible with the observed data. A
break or hole in the mineralization had not been previously
FIG. 6. Two 60-MHz single-hole reection radargrams, without and with AGC, for boreholes (a) 78 929 and (b) 78 930 (after Calvert
and Livelybrooks, 1997). Resistivity logs (Mwenifumbo et al., 1993) shaded according to lithology, shown at left.
interpreted geologically, though two mineralized intervals had
been intersected in hole 80 555 and interpreted as a reentrant
(Figure 1b).
The transmitter and receiver cables were copper cored (two-
conductor) and could therefore have served as parasitic anten-
nae. Although the JW-4 system included a lter to isolate the
receiver, investigations by INCOsince the original experiment
have conrmed that the 1994 data were affected by cross-cable
pick-up. The potential of radio imaging to detect narrow gaps
in conductors remains an interesting possibility, but there is no
longer any reason to invoke the existence of a gap to explain
the 1994 McConnell RT data.
GPR, especially in reection mode, offers the potential for
high-resolution mapping of geological contacts and structures.
Thesingle-holeboreholeradar reectionsurveys at McConnell
in 1993 and 1996 demonstrated that the sulde contacts only
sometimes produce strong coherent reections. The main fac-
tors responsible for the erratic occurrence of clear reections
are probably (1) unfavorable endre geometry, which reduces
the S/N ratio, (2) heterogeneity of the host breccia, (3) mi-
nor concentrations of disseminated sulde in or immediately
adjacent to the quartz diorite, and (4) local irregularity of the
sulde contact. Of these, only the rst can be altered. Borehole
radar would be far more likely to succeed in holes drilled par-
allel to the sulde contact, permitting broadside illumination
of the target. This exposes the contradistinction between the
geophysical view of drillholes as accessways for instruments
and the geological view of drillholes, as voids created during
sampling. Purpose-drilled holes for borehole radar represent
an additional survey cost; on the basis of the results obtained
to date, it is unclear whether the additional expense would be
justied at McConnell.
The crosshole GPR survey between boreholes 78 929 and
78 930 was suitable for tomographically imaging velocity and
absorption coefcient in the breccia above the orebody. The
velocity within the breccia exhibited signicant variation, rang-
ing between 95 and 145 m/s and averaging about 125 m/s.
The absorption coefcient ranged up to 2.5 dB/m, averaging
1.2 dB/m. The crosshole tomography thus conrmed that the
breccia is appreciably heterogeneous and hence less favorable
FIG. 7. Kirchhoff-style migration of crosshole radar data for
the region between boreholes 78 929 and 78 930. Text gives de-
tails regarding data processing and migration. The geologically
inferred location of the McConnell orebody is also outlined for
reference (after Calvert and Livelybrooks, 1997).
for radar propagation than its high resistivity might suggest.
For =1.2 dB/m =0.14 neper/m and =125 m/s, it follows
that Q 11 in the Sudbury Breccia. Although derived from
60-MHz data, this Q value could be reasonably characteris-
tic of the breccia at all radar frequencies since conductivity is
approximately proportional to frequency for many materials
(Johnscher, 1977), while the real part of permittivity is often
only weakly dependent on frequency (Collett and Katsube,
1973).
Migration of crosshole reection data has dened a reec-
tivity feature which lies above and parallel to the contact.
While a worthwhile technical achievement, the migration of
McConnell data was destined never to greatly advance the
knowledge of the contact geometry, given that reections orig-
inate near where hole 78 929 pierces the contact. However, in
another situation, between parallel boreholes drilled orthogo-
nal to a contact, migration of crosshole reection could yield
valuable information.
CONCLUSIONS
Radio imaging is a potentially effective means for imaging
massive sulde orebodies in resistive environments. Speci-
cally,
1) RT surveys depict orebodies as zones of enhanced atten-
uation;
2) radio imaging uses lower frequencies than GPR, thereby
achieving greater range but lower resolution;
3) RFEM is less sensitive to the presence of disseminated
sulde;
4) the absence of a physical link between transmitter and
receiver is a signicant logistical advantage in mining en-
vironments; and
5) typically, RFEM data can be collected more rapidly than
GPR, since ner spatial sampling is required for GPR.
Recording times for both RFEM and GPR could be
slashed using multichannel acquisition systems.
The advantages and disadvantages of single-hole reection
radar can be summarized as follows:
1) radar reection surveys offer greater resolution but
smaller range than either radio imaging or crosshole
radar;
2) both transmitter and receiver are in the same borehole,
so data acquisition is uncomplicated and efcient;
3) radar reection data are amenable to seismic processing,
a highly developed technology;
4) reection radar is sensitive to heterogeneity in the host
rock and to the presence of even minor concentrations
of disseminated sulde; and
5) unless dip and strike are known a priori, or unless direc-
tional antennae are deployed, surveys in more than one
hole are required to overcome rotational ambiguity.
Crosshole GPR surveying shows some promise for imaging
ore boundaries. Its strengths and weaknesses are
1) times and amplitudes of rst arrivals can be used to image
interhole velocity and absorption;
2) direct arrivals can be distinguished from reections and
diffractions;
3) later reections can potentially be migrated to dene
strong reectors;
4) larger data volumes and more elaborate processing add
to the cost of crosshole GPR; and
5) the need for a physical connection between the trans-
mitter and receiver limits applicability, especially under-
ground.
ACKNOWLEDGMENTS
This work was completed under the auspices of the
NSERC/TVX Gold/Golden Knight Chair in Borehole Geo-
physics for Mineral Exploration, Department of Mineral En-
gineering, Ecole Polytechnique, Montreal, Canada.
The McConnell surveys were jointly funded by INCO Ltd.
and Noranda; their support is gratefully acknowledged. Data
acquisitionwas expeditedby the efforts of many INCOperson-
nel, while AlanKing andGordMorrisonassistedwithplanning
and interpretation. The data have been released for publica-
tion by the kind permission of Larry Cochrane of INCO Ltd.,
Ontario.
The Migratomtomographic software was provided by Daryl
Tweeton and Michael Jackson (U.S. Bureau of Mines, Min-
neapolis). The EMSUN modeling program was made freely
available by Keeva Vozoff (Harbourdom, Sydney, Australia).
Geoff Smith (Univ. of Technology, Sydney, Australia) assisted
with the EMSUN modeling.
The 1993 borehole radar data were recorded by an AECL
crew, led by Kevin Stevens. The 1994 radio imaging data
were recorded by Wu Yiren, Qu Xingchang, and Zhang Ziling
from the Chinese Ministry of Geology and Mineral Resources
(MGMR), Beijing. Maree-Josee Bertrand, Noranda, assisted
with interpretation of the radio imaging data.
The paper benetted from comments by Greg Turner and
an anonymous reviewer. Mike Asten is to be commended for
his editorial persistence and patience.
REFERENCES
Anderson, C. G., and Logan, K. J., 1992, The history and current status
of geophysical exploration at the Osborne Cu and Au deposit, Mt.
Isa: Expl. Geophys., 23, 18.
Annan, A. P., 1993, Practical processing of GPRdata: 2nd Govt. Work-
shop on GPR, Proceedings, 122.
Belleeur, G., and Chouteau, M., 1998, Borehole radar and delineation
of the McConnell massive sulde deposit, Sudbury, Ontario: 7th
Internat. Conf. on Ground Penetrating Radar Proceedings, 353358.
Calvert, A. J., and Livelybrooks, D., 1997, Borehole radar reection
imaging at the McConnell nickel deposit, Sudbury: Exploration 97,
4th Decennial Internat. Conf. on Min. Expl., Proceedings, 701704.
Campbell, G., 1994, Geophysical contributions to mine development
planningA risk reduction approach: 15th CMMI Congress, S.
African Inst. Min. Metall., 3, 283325.
Cao, J., Nie, Z., and Zhu, J., 1998, Dual frequency conductivity tomog-
raphy: 4th Soc. Expl. Geophys. Japan Internat. Symp., Proceedings,
227230.
Collett, L. S., and Katsube, T. J., 1973, Electrical parameters of rocks
in developing geophysical techniques: Geophysics, 38, 7691.
Cook, J. C., 1975, Radar transparencies of mine and tunnel rocks: Geo-
physics, 40, 865875.
Coon, J. B., Fowler, J. C., and Schafers, C. J., 1981, Experimental uses
of short pulse radar in coal seams: Geophysics, 46, 11631168.
Davis, J. L., Annan, A. P., and Vaughan, C. J., 1985, Placer exploration
using radar and seismic methods: Can. Inst. Min. Bull., 80, 6072.
Eisenburger, D., Sender, F., and Thierbach, R., 1993, Borehole radar
An efcient geophysical tool to aid in the planning of salt caverns
and mines: 7th Internat. Symp. on Salt, 1, 279284.
Fisher, E., McMechan, G. A., Annan, A. P., and Cosway, S. W.,
1992, Examples of reverse-time migration of single-channel ground-
penetrating radar proles: Geophysics, 57, 577586.
Fullagar, P. K., and Livelybrooks, D., 1994a, Trial of tunnel radar for
cavity and ore detection in the Sudbury mining camp, Ontario: 5th
Internat. Conf. on Ground Penetrating Radar, Proceedings, 3, 883
894.
1994b, NSERC/TVX Gold/Golden Knight Chair in Borehole
Geophysics for Mineral Exploration annual report: Ecole Polytech-
nique de Montreal, Department of Mineral Engineering.
Fullagar, P. K., Zhang, P., and Wu, Y., 1994, Trial of radio tomography
for exploration and delineation of massive sulde deposits in the
Sudbury Basin: Ecole Polytechnique de Montreal, Department of
Mineral Engineering.
Fullagar, P. K., Zhang, P., Wu, Y., andBertrand, M.-J., 1996, Application
of radio frequency tomography to delineation of nickel suldes in
the Sudbury basin: 66th Ann. Internat. Mtg., Soc. Expl. Geophys.,
Extended Abstracts, 20652068.
Grant, R. W., and Bite, A., 1984, Sudbury quartz diorite offset dikes,
in Pye, E. G., Naldrett, A. J., and Giblin, P. E., Eds., The geology and
ore deposits of the Sudbury structure: Ontario Ministry of Natural
Resources, 275300.
Jackson, M., and Tweeton, D., 1994, MIGRATOMGeophysical to-
mography using wavefront migration and fuzzy constraints: U.S. Bu-
reau of Mines Report of Investigations RI 9497.
Johnscher, A. K., 1977, The universal dielectric response: Nature, 267,
673679.
Killeen, P., Mwenifumbo, C., and Elliott, B., 1996, Mineral deposit sig-
natures by borehole geophysicsData from the borehole geophys-
ical test at the McConnell nickel deposit (Garson offset), Ontario:
Natural Resources Canada, Geological Survey of Canada, Open le
report 2811.
King, A., 1996, Deep drillhole electromagnetic surveys for nickel/ cop-
per suldes at Sudbury, Canada: Expl. Geophys., 27, 105118.
Lager, D. L., and Lytle, R. J., 1977, Determining a subsurface elec-
tromagnetic prole from high frequency measurements by applying
reconstruction technique algorithms: Radio Sci., 12, 249260.
Liner, C. L., and Liner, J. L., 1995, Ground-penetrating radar: A near-
face experience from Washington County, Arkansas: The Leading
Edge, 14, 1721.
Liu, Q., Osman, N., Manning, P., Hargreaves, J., Mason, I., and Turner,
G., 1998, Borehole radar reection characteristics of nickel sulde
shoots: Australian Mining Tech. Conf., Proceedings, 333346.
McDowell, G., andVerlaan, L., 1997, Radioimagingfor suldeorebody
delineation: High-Res. Geophys. Workshop, Dept. of Mining and
Geol. Eng., The University of Arizona, Proceedings.
McGaughey, W. J., 1990, Mining applications of crosshole seismic to-
mography: Ph.D. thesis, Queens Univ.
McGaughey, W. J., and Stolarczyck, L. G., 1991, Tomographic inver-
sion of EM seam-wave absorption in the Prairie Evaporite Forma-
tion, Saskatchewan: 61st Ann. Internat. Mtg., Soc. Expl. Geophys.,
Mundry, E., Thierbach, R., Sender, F., andWeichart, H., 1983, Borehole
radar probing insalt deposits: 6thInternat. Symp. onSalt, 1, 585599.
Murray, W., Williams, C., Lewis, C., and Hatherly, P., 1998, Develop-
ment of geophysical logging and imaging tools for use in in-seam
drilling: Australian Mining Tech. Conf., Proceedings, 327332.
Mwenifumbo, C. J., Killeen, P. G., Elliott, B. E., and Pug, K. A., 1993,
The borehole geophysical signature of the McConnell nickel deposit,
Sudbury area: 5th Internat. Symp., Min. and Geotech. Logging Soc.,
Proceedings.
Nickel, H., and Cerny, I., 1989, More effective underground explo-
ration for ores using radio waves: Expl. Geophys., 20, 371377.
Olsson, O., Falk, L., Forslund, O., Lundmark, L., and Sandberg, E.,
1990, Crosshole investigationsResults from borehole radar inves-
tigations (revised1990, original report 1987): Stripa Project technical
report 87-11.
Pears, G. A., 1997, Tomography-based AIM inversion of radio fre-
quency electromagnetic data: M.S. thesis, Univ. of Queensland.
Pears, G. A., and Fullagar, P. K., 1998, Weighted tomographic imaging
of radio frequency data: Expl. Geophys., 29, 554559.
Qu, X., Gao, W., and Zhou, H., 1991, JW-4 subsurface electromagnetic
meter: Presented at Conf. on Appl. of Computing Tech. in Geosci.
Raadsma, J. M., 1994, Ground penetrating radar applications on high
grade gold deposits at the Sixteen to One Mine, California: 5th Inter-
nat. Conf. on Ground Penetrating Radar, Proceedings, 3, 925940.
Smith, G. H., Williamson, P. R., and Vozoff, K., 1990, The application of
nested dissection to the solution of a 2.5Delectromagnetic problem:
Appl. Computat. Electromag., 5, 87106.
Stevens, K. M., and Lodha, G. S., 1994, Borehole radar surveys at the
McConnell depositGarson offset, Sudbury, Ontario: Appl. Geo-
phys. Branch, Atomic Energy of Canada, Ltd., Pinawa, Manitoba,
AECL Reference #008263-002.
Stevens, K., and Redko, G., 2000, In-mine applications of the radio
wave method in the Sudbury igneous complex: 70th Ann. Internat.
Mtg., Soc. Expl. Geophys., Expanded Abstracts, (in press).
Stewart, R. D., and Unterberger, R. R., 1976, Seeing through rock salt
with radar : Geophysics, 41, 123132.
Stewart, R. R., 1991, Exploration seismic tomographyfundamentals,
in Domingo, S. N., Ed., Course notes series, 3, Soc. Expl. Geophys.
Thierbach, R., 1994, Twenty years of ground-probing radar in salt and
potash mines: 5th Internat. Conf. on Ground Penetrating Radar,
Proceedings, 3, 957980.
Thomson, S., and Hinde, S., 1993, Bringing geophysics into the mine:
Radio attenuation imaging and mine geology: Expl. Geophys., 24,
805810.
Thomson, S., Young, J., and Sheard, N., 1992, Base metal applications
of the radio imaging methodcurrent status and case studies: Expl.
Geophys., 23, 367372.
Turner, G., and Siggins, A. F., 1994, Constant Q attenuation of subsur-
face radar pulses: Geophysics, 59, 11921200.
Vozoff, K., Smith, G. H., Hatherly, P. J., and Thomson, S., 1993, An
overview of the radio imaging method in Australian coal mining:
First Break, 11, 1321.
Ward, S. H., and Hohmann, G. W., 1988, Electromagnetic theory for
geophysical applications, in Nabighian, M. N., Ed., Electromagnetic
methods in applied geophysics 1, Theory: Soc. Expl. Geophys., In-
vestigations in Geophysics 3, 131311.
Wedepohl, E., 1993, Radio wave tomography-imaging ore bodies us-
ing radio waves: 3rd Tech. Mtg., South African Geophys. Assn., Ex-
panded Abstracts, 8588.
Wedepohl, E., Trickett, J., van Schoor, M., Grodner, M., and
Schweitzer, J., 1998, A geophysical toolbox for deep level gold
mining, South Africa: Australian Mining Tech. Conf., Proceedings,
269291.
Yelf, R., Turner, G., Hatherly, P., and Hagen, D., 1990, Appraisal of
ground penetrating radar in underground coal mines: Final Report
for NERDDC Project 1210, Australian Coal Industry Research
Laboratories.
Yilmaz, O., 1987, Seismic data processing: Soc. Expl. Geophys.
Young, R., Deng, Z., and Sun, J., 1995, Interactive processing of GPR
data: The Leading Edge, 14, 275280.
Zhou, B., and Fullagar, P. K., 2000, Borehole radar conductivity and
permittivity tomography: 8th Internat. Conf. on Ground Penetrating
Radar, Proceedings (in press).
Zhou, B., Fullagar, P. K., and Fallon, G. N., 1998, Radio frequency
tomography trial at Mt. Isa mine: Expl. Geophys., 29, 675
679.
3-D inversion of induced polarization data
Yaoguo Li
and Douglas W. Oldenburg
ABSTRACT
We present an algorithm for inverting induced polar-
ization (IP) data acquired in a 3-D environment. The
algorithm is based upon the linearized equation for the
IP response, and the inverse problem is solved by mini-
mizing an objective function of the chargeability model
subject to data and bound constraints. The minimization
is carried out using an interior-point method in which the
bounds are incorporated by using a logarithmic barrier
and the solution of the linear equations is accelerated
using wavelet transforms. Inversion of IP data requires
knowledge of the background conductivity. We study
the effect of different approximations to the background
conductivity by comparing IP inversions performed us-
ing different conductivity models, including a uniform
half-space and conductivities recovered from one-pass
3-D inversions, composite 2-D inversions, limited AIM
updates, and full 3-Dnonlinear inversions of the dc resis-
tivity data. We demonstrate that, when the background
conductivity is simple, reasonable IP results are obtain-
able without using the best conductivity estimate derived
from full 3-D inversion of the dc resistivity data. As a -
nal area of investigation, we study the joint use of surface
and borehole data to improve the resolution of the re-
covered chargeability models. We demonstrate that the
joint inversion of surface and crosshole data produces
chargeability models superior to those obtained from
inversions of individual data sets.
INTRODUCTION
In recent years, there has been much progress in rigorous in-
version of induced polarization (IP) data assuming a 2-Dearth
structure. Published work on 2-Dinversions has demonstrated
that inversion can help extract information that is otherwise
unavailable from direct interpretation of the pseudosections.
Manuscript received by the Editor February 23, 1999; revised manuscript received June 2, 2000.
Formerly University of British Columbia, Department of Earth and Ocean Sciences; presently Colorado School of Mines, Department of Geo-
physics,1500 Illinois St., Golden, Colorado 80401. E-mail: ygli@mines.edu.
University of British Columbia, Department of Earth and Ocean Sciences, 2219 Main Mall, Vancouver, B.C. V6T1Z4, Canada. E-mail: doug@eos.
ubc.ca.
In application, the technique has matured sufciently that it is
now routinely applied to data sets acquired in mineral explo-
ration projects and in environmental problems.
The 2-D IP data are commonly inverted using a linearized
approach (LaBrecque, 1991; Oldenburg and Li, 1994), in which
the chargeability is assumed to be relatively small and the ap-
parent chargeability data are expressedas a linear functional of
the intrinsic chargeability. Alinear inverse problemis solved to
obtainthe chargeability model. Inadditiontothe linearizedap-
proach, Oldenburg and Li (1994) also propose two other meth-
ods. The second obtains the chargeability by performing two
separate dc resistivity inversions and then taking the relative
difference of the recovered conductivities. The third method
makes no assumption about the magnitude of the chargeability
and performs a full nonlinear inversion to construct its distri-
bution.
The effectiveness of IP inversions has been documented in
several case histories (e.g., Oldenburg et al., 1997; Kowalczyk
et al., 1997; Mutton, 1997). When the data set is acquired in a
truly 2-Denvironment, the inversion algorithmhas performed
well. However, 2-D inversions face difculties when the basic
2-Dassumptionis violatedbecauseof theuseof 3-Dacquisition
geometry or the presence of a 3-Dgeoelectrical structure such
as severe 3-Dtopography or 3-Dvariation of conductivity and
chargeability. Under these circumstances, a 3-D algorithm is
required.
The methods developed in 2-Dare general and applicable to
3-D problems. For instance, recovering chargeability by com-
puting the difference between two conductivity inversions is
demonstrated by Ellis and Oldenburg (1994) using polepole
data. The implementation of the linearized approach is also
straightforward in principle; however, numerical and compu-
tational challenges require specic treatment. The foremost
challenge is the computational complexity related to generat-
ing background conductivity in three dimensions and the so-
lution of the large-scale constrained minimization problem to
construct the 3-Dchargeability model. This paper concentrates
on these associated computational issues. We assume that the
1931
1932 Li and Oldenburg
chargeability is small and that the data are not affected by EM
coupling effect. Therefore, we adopt the linearized representa-
tion of the IP response and develop the inversion methodology
applicable for general electrode congurations, including sur-
face arrays, downhole arrays, and crosshole electrode congu-
rations. We present a detailed algorithmthat solves large-scale
problems.
Our paper begins with a summary of the basics of IP in-
version and the formulation of the inverse solution. The use
of approximate conductivity models in the 3-D IP inversion
is discussed next to demonstrate how an efcient IP solution
can be obtained in practice. We then study the joint inversion
of surface and crosshole data and its improvement in model
resolution. We conclude with an application to a eld data set
and a discussion.
BACKGROUND
The commonly used electrode congurations in most explo-
ration work include the polepole, poledipole, dipoledipole,
and gradient arrays. These arrays are usually arranged in a co-
linear conguration, and the source and potential electrodes
are generally aligned parallel to the traverse direction. How-
ever, to image a 3-Dstructure, truly 3-Ddata are often needed.
This requires that off-line or cross-line data be acquired and
that the orientationof the current electrodes be varied. Inaddi-
tion, high-resolution surveys carried out in ore delineation and
geotechnical investigations often acquire surface-to-borehole
and crosshole data in three dimensions. Thus, a generally ap-
plicable inversion algorithm must be able to work with arbi-
trary electrode congurations. Inthis paper, we assume that the
time-domain IP measurements are acquired using an arbitrary
electrode geometry over a 3-D structure. The current source
can be a single pole, dipole, or widely separated bipole either
on the earths surface or in boreholes. The resulting potential
or potential difference can be measured as data anywhere on
the surface or in the borehole. The commonly used polepole,
poledipole, and dipoledipole arrays on the surface or in the
borehole constitute only a small number of possible congura-
tions.
Let (r) be the conductivity as a function of position in three
dimensions beneath the earths surface and (r) be the charge-
ability as dened by Seigel (1959). The dc potential produced
by a current of unit strength placed at r
s
is governed by the
partial differential equation
(
) = (r r
s
), (1)
where
denotes the potential in the absence of IP effect.

When the chargeability is nonzero, it effectively decreases the
electrical conductivity of the media by a factor of (1)(Seigel,
1959). The corresponding total potential
is given by
((1 )
) = (r r
s
). (2)
Thus, the secondary potential measured in an IPsurvey is given
by the difference
s
=
, (3)
while the apparent chargeability is dened
a
=

. (4)
The apparent chargeability is the preferred form of IP data,
and it is well dened in some surface and downhole surveys.
However, in crosshole experiments using dipole sources or
receivers, the electric eld often reverses direction along the
borehole, and the measured total potential differences can ap-
proach zero in the vicinity of the zero crossing. The zero cross-
ing canalsooccur withnoncolinear arrays onthe surface. These
near-zero potentials cause the apparent chargeability to be un-
dened. It is therefore necessary to use the secondary potential
as data when these conditions occur.
When the magnitude of the chargeability is moderate, the
secondary potential
s
measured in an IP experiment is well
approximated by a linear relationship with the intrinsic charge-
ability. Applying a Taylor expansion to equation (3), neglect-
ing higher order terms, and discretizing the earth into cells
of constant conductivity
j
and chargeability
j
results in
the following equation (e.g., Seigel, 1959; Oldenburg and Li,
1994):
si
=
M
j =1
j

i
ln
j

M
j =1
j
J
i j
, (5)
where J
i j
is the sensitivity of the secondary potential
si
and
i
is the corresponding total potential. If the total potentials
do not approach zero, the linearized equation for apparent
chargeability
a
is given by
ai
=
M
j =1
j
ln
i
ln
j

M
j =1
j
J
i j
, (6)
where J
i j
is the corresponding sensitivity. Note that J
i j
is un-
dened when
i
approaches zero.
Given a set of measured IP data, inversion of either equa-
tion (5) or (6) allows the recovery of the intrinsic chargeabil-
ity model. Since the true conductivity structure is unknown in
practical applications, an approximation to it is substituted in
calculating the sensitivities. This approximation is usually ob-
tained by inverting the accompanying dc potential data. Thus,
the IP inverse problemis a two-stage process. In the rst stage,
an inverse problem is solved to recover a background conduc-
tivity fromthe dc resistivity data. This conductivity is then used
to generate the sensitivity for the IP inversion, and a linear in-
verse problem is solved to obtain the chargeability.
FORMULATING THE INVERSION
Assume we have a set of N IP data, which can be apparent
chargeabilities or secondary potentials. Further assume that a
dc resistivity inversion has been performed (see next section)
to obtain a reasonable approximation to the true conductivity;
the IP sensitivity is calculated from it. To invert these IP data
for a 3-D model of chargeability, we rst use the same mesh as
in the dc resistivity inversion to divide the model region into
M cells and assume a constant chargeability value in each cell.
The data are formally related to the chargeabilities in the cells
by the relation in equations (5) and (6),
d = J
, (7)
where the data vector d = (d
1
, . . . , d
N
)
T
and the model vector
= (
1
, . . . ,
M
)
T
. J is the sensitivity matrix corresponding to
the data, whose elements J
i j
are calculated from the assumed
approximation to the background conductivity by using an
adjoint equationapproach(McGillivaryandOldenburg, 1990).
For this calculation and all the numerical simulations in this
3-D Inversion of IP Data 1933
paper, we use the nite-volume method (Dey and Morrison,
1979) to solve equation (1) to obtain the electrical potentials.
The number of model cells is generally far greater than the
number of data available; thus, an underdetermined problem
is solved. To obtain a particular solution, we minimize a model
objective function, subject to the data constraints in equa-
tion (7). We used a model objective function that is similar
to that for the 2-Dcase but that has an extra derivative termin
the third dimension. Let m = generically denote the model.
The objective function is given by
m
=
s
_
V
(m m
0
)
2
dv +
x
_
V
_
(m m
0
)
x
_
2
dv
+
y
_
V
_
(m m
0
)
y
_
2
dv +
z
_
V
_
(m m
0
)
z
_
2
dv,
(8)
where m
0
is a reference model. The positive scalars
s
,
x
,
y
,
and
z
are coefcients that affect the relative importance of the
different components. We usually choose
s
to be much smaller
than the other three coefcients, so the recovered model be-
comes smoother as the ratios
x
/
s
,
y
/
s
, and
z
/
s
increase.
For numerical solutions, equation (8) is discretized using the
nite-difference approximation. The resulting matrix equation
has the following form;
m
= (mm
0
)
T
_
s
W
T
s
W
s
+
x
W
T
x
W
x
+
y
W
T
y
W
y
+
z
W
T
z
W
z
__
mm
0
)
W
m
(mm
0
)
2
. (9)
The data constraints are satised by requiring that the total
mist between the observed and predicted data be equal to a
target value. We measure the data mist using the function
d
=
_
_
W
d
_
d
pre
d
obs
__
_
2
, (10)
where d
pre
and d
obs
are, respectively, predicted and observed
data and where W
d
is a diagonal matrix whose elements are
the inverse of the standard deviation of the estimated error of
each datum: W
d
= diag{1/
1
, . . . , 1/
N
}. If we assume that the
contaminating noise is independent Gaussian noise with zero
mean, then
d
has X
2
distribution with N degrees of freedom,
and its expected value is equal to N. Thus, a reasonable target
value is
d
= N.
In addition to the data constraints, we also need to impose a
lower and an upper bound on the recovered chargeability. The
bounds are required because the chargeability is dened in the
range [0,1). The bound constraints ensure that the recovered
model is physically plausible. For numerical implementation,
the lower bound must be zero since the chargeability of the
general background is zero. The upper bound, denoted by u,
can take on the theoretical value of unity or can be smaller if
a better estimate of the upper bound is known.
Having dened the model objective function, the data mist
and its expected value, and the appropriate bounds, we now
solve the inverse problemof constructing the 3-Dchargeability
model by the Tikhonov regularization method (Tikhonov and
Arsenin, 1977) with additional bound constraints:
minimize =
d
+
m
subject to 0 m < u, (11)
where is the regularizationparameter that controls the trade-
off between the model norm and mist. Ultimately, we want
to choose such that the data mist function is equal to a pre-
scribedtarget value
d
. The minimizationis solvedwhena min-
imizer m
is found whose elements are all within the bounds.

This is a quadratic programming problem, and the main dif-
culties arise from the presence of the bound constraints. We
use an interior-point method to performthe minimization. The
original problem in equation (11) is solved by a sequence of
nonlinear minimizations in which the bound constraints are
implemented by including a logarithmic barrier term in the
objective function (e.g., Gill et al., 1991; Saunders, 1995):
B(m, ) =
d
+
m
2
_
M
j =1
ln
_
m
j
u
_
+
M
j =1
ln
_
1
m
j
u
_
_
, (12)
where is the barrier parameter and the regularization pa-
rameter is xed during the minimization. The minimization
starts with a large and an initial model whose elements are
well within the lower and upper bounds. It then iterates to the
nal solution as is decreased toward zero. As approaches
zero, the sequence of solutions approaches the model that min-
imizes the original total objective function in equation (11).
Since we are only interested in the nal solution, we do not
carry out the minimization completely for each value of in
the decreasing sequence. Instead, we take only one Newton
step and limit the step length during the model update to keep
the model within the bounds throughout the minimization. The
steps of the algorithm are as follows:
1) Set the initial model mand the , and calculate the start-
ing value of the barrier parameter by
=

d
+
m
2
M
j =1
_
ln
_
m
j
u
_
+ln
_
1
m
j
u
__
. (13)
2) Take one Newton step for each value of by solving the
following equation for a model perturbation m:
_
J
T
W
T
d
W
d
J +W
T
m
W
m
+X
2
+Y
2
_
m
= J
T
W
T
d
W
d
(d d
obs
) W
T
m
W
m
(mm
0
)
+(X
1
Y
1
)e, (14)
where X = diag{m
1
, . . . , m
M
}, Y = uI X, and
e = (1, . . . , 1)
T
.
3) Determine the maximumsteplengthof the model update
that satises the bounds:
= min
m
j
<0
m
j
|m
j
|
,
+
= min
m
j
>0
u m
j
m
j
,
= min(
,
+
). (15)
4) Update the model and barrier parameter by the limited
step length:
m m+m,
[1 min(, )]. (16)
5) Return to step 2 and iterate until convergence according
to the criteria that
d
+
m
has reached a plateau and
the barrier term is much smaller than this quantity. The
parameter is usually prescribed to be a value close to
unity. Its role is to prevent model elements from reach-
ing the bounds exactly so that the logarithmic barrier
iteration can continue. Values between (0.99, 0.999) have
been commonly used in literature (e.g., Gill et al., 1991).
Our experience with IP inversions suggests that a slightly
smaller value works better, so we have typically used
0.925 in our algorithm.
The central task of the algorithm is solving the linear sys-
tem in equation (14). We obtain the solution by using the con-
jugate gradient (CG) technique. The reasons for using a CG
solver are twofold. First, any practical application will require a
large number of cells (at least on the order of 10
4
) to represent
the geology reasonably. As a result, the linear system in equa-
tion (14) is large, and explicit formation of J
T
W
T
d
W
d
J is im-
practical. This precludes the use of any direct solver. The CG
technique is the obvious choice for an iterative solver since the
matrix (J
T
W
T
d
W
d
J + W
T
m
W
m
) is symmetric. Also, each sub-
problem at a given value of barrier parameter only generates
one step among a sequence that leads to the nal solution. It is
unnecessary to solve equation (14) precisely. Instead, it is com-
mon to solve the central equation approximately to produce
a partial solution. The resulting update is called a truncated
Newton step. This is designed to reduce the required amount
of computation without compromising the quality of the nal
solution. The CGtechnique can be terminated at an early stage
by supplying it with a relaxed stopping criterion. We have typ-
ically used a criterion that the ratio of the normof the residual
and the norm of the right-hand side in equation (14) be less
than 10
2
. This has led to large computational savings.
CGiterations require the repeated multiplication of the ma-
trix J
T
W
T
d
W
d
J and W
T
m
W
m
to vectors. The matrix W
T
m
W
m
is
extremely sparse, and the matrixvector multiplication is eas-
ily obtained. However, applying J
T
W
T
d
W
d
J is computationally
intensive since J is dense. We perform a fast matrixvector
multiplication by using wavelet transforms (Li and Oldenburg,
1999a), in which a sparse representation of the dense matrix is
formed in the wavelet domain and matrixvector multiplica-
tion is carried out by sparse multiplications.
The remaining issue is how to determine the optimal value
so that the data constraint
d
=
d
is satised. There are
two situations that require different treatments. In the rst,
a reliable estimate is available for the standard deviation of
the errors that have contaminated the data and, therefore, the
value of
d
is known. We then need to nd the value of that
yields this target mist. This is achieved by an efcient line-
search technique that uses a number of approximate solutions
to the minimization problem. In the second case, the standard
deviations of errors are unknown; hence, the optimal value
of must be estimated independently. We achieve this with
the generalized cross-validation technique (Golub et al., 1979;
Wahba, 1990; Haber and Oldenburg, 2000). The use of these
techniques in large-scale 3-D inversions is detailed in Li and
Oldenburg (1999a).
We now illustrate our algorithm using a test model com-
posed of ve anomalous rectangular prisms embedded in a uni-
formhalf-space (Figure 1). Three surface prisms simulate near-
surface distortions, and two buried prisms simulate deeper tar-
gets. The conductivity and chargeability of the prisms are listed
in Table 1. The dc resistivity and IP data fromboth surface and
crosshole experiments have been computed.
The surface experiment is carried out using a poledipole
array with a =50 m and n =1 to 6. There are seven traverses
spaced 100 m apart in both eastwest and northsouth direc-
tions. There are 1384 observations, and these have been con-
taminated with uncorrelated Gaussian noise whose standard
deviation is equal to 2% of the datum value. Figures 2 and 3
show apparent conductivity pseudosections and apparent
chargeability pseudosections at three selected eastwest tra-
verses. The pseudosections are dominated by the responses to
the near-surface prisms, and there are only subtle indications
of the buried conductive prism.
Werst invertedthedc resistivitydatausingaGauss-Newton
approach that constructs a minimum structure model using a
model objective function similar to that in equation (8) but
applied to the logarithmic conductivity as the model. We set
the coefcients to
s
=0.0001 and
x
=
y
=
z
=1 and used a
reference conductivity model of 1 mS/m. The recovered con-
ductivity model is shown by two plan sections and one cross
section in Figure 4. It is a reasonably good representation of
the true conductivity model. All three surface prisms and the
buried conductive prism are clearly imaged, and there is indi-
cation of a resistive prism at depth. This conductivity model is
then used to calculate the sensitivity for the subsequent IP in-
version. The inverted chargeability model is shown in Figure 5
in the same plan and cross-sections. The surface prisms are
clearly imaged, and the chargeability at depth is concentrated
at the locationof the twoburiedtargets. The separationof these
bodies is not clearly dened, but this decrease of anomaly def-
inition with increasing depth is expected when surface data
are inverted. Overall, the model is a good representation of
the true anomalous chargeability zone. The contrast between
the pseudosections shown in Figure 3 and the cross-section of
the recovered model in Figure 5 illustrates the improvement
gained by performing the 3-D inversion.
CONSTRUCTION OF APPROXIMATE CONDUCTIVITIES
As discussed in the preceding section, the inversion of IP
data requires a background conductivity model for calculating
the sensitivity. IP inversion is therefore a two-stage process,
and its success depends upon the availability of a conductivity
model that is a reasonable approximation to the true conduc-
tivity. The usual approach to generating such a conductivity
model is to invert the dc resistivity data that accompany the
IP data. Numerous papers have been published on 3-D dc re-
sistivity inversions, and different approaches have been pro-
posed. For example, Park and Van (1991), Ellis and Oldenburg
(1994a), Sasaki (1994), Zhang et al. (1995), and LaBrecque and
Morelli (1996) all perform linearized inversion to construct a
conductivity model from the dc resistivity data, although de-
tails of their algorithms and implementations may vary greatly.
Li and Oldenburg (1994), on the other hand, apply approxi-
mate inverse mapping (AIM) formalism to construct a model
that reproduces the data. For the current study, we implement
a regularized inversion and use Gauss-Newton minimization
to accomplish this (Li and Oldenburg, 1999b). The basics of
that algorithm are summarized here.
Let m = ln dene the model used in the conductivity in-
version, and let d
obs
be the dc potential data. As in the IP inver-
FIG. 1. Perspective viewof the ve-prismmodel. Seven surface
traverses in the eastwest direction and four boreholes are also
shown. For clarity, seven traverses in the northsouth direction
are not shown. The physical property values of the prisms are
listed in Table 1.
FIG. 2. Examples of the apparent conductivity pseudosections
at three eastwest traverses. The data are simulated for a
poledipole array, and they have been contaminated by in-
dependent Gaussian noise with a standard deviation equal to
2% of the accurate datum magnitude. The pseudosections are
dominated by the surface responses. The grayscale shows the
apparent conductivity in mS/m.
sion, we minimize the model objective function in equation (8)
subject to tting the data to the degree determined by the es-
timated error. Thus, the desired conductivity model solves the
following minimization problem:
minimize =
d
+
m
subject to
d
=
d
, (17)
where
d
and
m
are the same as those dened in equations (8)
and (10), is the regularization parameter, and
d
is the target
mist value for the dc resistivity problem.
The potential data depend nonlinearly upon the conductiv-
ity; hence, minimization (17) must be solved iteratively. Let m
be the current model, d its predicted data, and ma model per-
turbation. Performing a rst-order Taylor series expansion of
the predicteddata as a functional of the newmodel m+mand
substituting into the total objective function in equation (17),
we obtain
(m +m)
_
_
W
d
(d +Jmd
obs
)
_
_
2
+W
m
(m+mm
0
)
2
. (18)
Table 1. Conductivity and chargeability of the prisms. The
half-space has a conductivity of 1 mS/mand zero chargeability.
Prism Conductivity (mS/m) Chargeability (%)
S1 10 5
S2 5 5
S3 0.5 5
B1 0.5 15
B2 10 15
FIG. 3. Examples of the apparent chargeability pseudosections
along three eastwest traverses. The data have been contami-
natedby independent Gaussiannoise witha standarddeviation
equal to 2%of the accurate datummagnitude plus a minimum
of 0.001. The same masking effect of near-surface prisms ob-
served in apparent-conductivity pseudosections is also present
here. The grayscale shows the apparent chargeability multi-
plied by 100.
The value J is the sensitivity matrix of the potential data. Its
elements are given by
J
i j
=

i
ln
j
(19)
and are evaluated at the current model. Differentiating with
respect to m and setting the derivative to zero yields the
equation for the model perturbation:
_
J
T
W
T
d
W
d
J +W
m
W
m
_
m
= J
T
W
T
d
W
d
(d d
obs
) W
T
m
W
m
(mm
0
). (20)
FIG. 4. The conductivity model recovered from inversion of
surface data using a Gauss-Newton method. The model is
shown in one cross-section and two plan sections. The posi-
tions of the true prisms are indicated by the white lines.
This is the basic equation solved for a Gauss-Newton step.
The new model is then formed by updating the current model:
m m + m. This process is repeated iteratively until the
minimization converges and an optimal value of is found to
produce the desired data mist in equation (17).
The nonlinear inversion of 3-D dc resistivity data provides
the best approximation to the actual conductivity distribution,
but it is a costly undertaking. One may not always want to
expend that amount of computation, especially when the re-
covery of the conductivity model is but an intermediate step to-
ward the end goal of constructing a chargeability model. More
importantly, good IP inversion results are often obtained by
using less rigorous approximations to the conductivity. Our
FIG. 5. The chargeability model recovered from inversion of
surface data. The conductivity from full 3-D dc inversion is
used to calculate sensitivities. The positions of the true prisms
are indicated by the white lines.
experience with 2-D inversions (Oldenburg and Li, 1994) has
shownthat goodrst-order results concerningthechargeability
distribution can often be obtained by approximating the earth
using a homogeneous conductive half-space. This suggests that
a reasonable recovery of a 3-D chargeability model might be
achieved by using intermediate approximations between the
two end members corresponding to a uniform half-space and
the conductivity model recovered froma full nonlinear 3-Ddc
inversion.
To explore this, we compare ve options for generating a
conductivity model to be used in the IP inversion. The rst
four require much less computation than does the full 3-D in-
version:
1) Auniformhalf-space: This is the simplest approximation,
and no inversion of dc data is involved. When invert-
ing apparent chargeability data, the actual value of the
half-space conductivity is arbitrary since the sensitivity is
independent of it. However, when the secondary poten-
tials are inverted, the best tting half-space from the dc
resistivity data should be used.
2) One-pass approximate 3-D inversion: This conductivity
model is obtained from a linear inversion of the dc
data assuming that the actual conductivity consists of
weak perturbations of a uniform half-space (e.g., Li and
Oldenburg, 1994; M/ller et al., 1996). Such a model
captures the gross features in the conductivity structure
and demands the least amount of computation. We have
implemented the approximate inversion in the spatial
domain, in which the model objective function in equa-
tion (8) is minimized explicitly so that a minimum struc-
ture model is obtained. This is identical to performing the
rst iteration of the Gauss-Newton inversion with both
the initial and reference model being equal to the chosen
background, m
b
. The equation to be solved is
_
J
T
W
T
d
W
d
J +W
m
W
m
_
m
= J
T
W
T
d
W
d
(d d
obs
), (21)
where d is the predicted data. The approximate conduc-
tivity is given by m = m
b
+m.
3) Composite 2-D inversions: When surface data along par-
allel traverses are available, independent 2-D inversions
canbe carriedout along eachline sothat a 2-Dmodel that
reproduces the observations is generated (Oldenburg
et al., 1993; Loke and Barker, 1996). The series of 2-D
models are then combined to form a 3-D representation
of the true conductivity structure. Such a model should
perform well when there are strong 2-D features in the
data.
4) Limited 3-D AIM updates: Using the one-pass 3-Dinver-
sionas anAIM, wecaniterativelyupdatetheconductivity
model by the AIMalgorithm(Oldenburg and Ellis, 1991)
Table 2. List of the dc andIPmist for different approximations tothe backgroundconductivity. The dc mist is calculatedbetween
the observed dc resistivity data and the predicted data obtained from3-Dforward modeling of the approximate conductivities. The
IP mist is the value achieved by the IP inversion when an approximate conductivity is used to calculate the sensitivity.
Mist Half-space One-pass 3-D approximation 2-D composite Limited AIM updates Nonlinear inversion
DC 2.96 10
6
6.44 10
4
1.22 10
5
6.96 10
3
1.35 10
3
IP 2184 1720 1637 1634 1510
such that a nal model reproducing the 3-Dobservations
is constructed. The greatest mist reduction is achieved
within the rst two or three iterations (Li and Oldenburg,
1994). Thus, by performing only a limitednumber of AIM
updates, we can obtain a conductivity model for the IP
inversion. Let F
1
denote the one-pass approximate in-
version and m be the current model. The model pertur-
bation is dened by the difference between models gen-
erated by applying the approximate inverse mapping to
the observed and predicted data, respectively:
m m+F
1
[d
obs
] F
1
[d], (22)
where d is the predicted data from the current model.
The iteration starts with an initial model which can be
supplied by m = F
1
[d
obs
].
5) Full 3-D nonlinear inversion: We use the Gauss-Newton
inversion discussed at the beginning of this section. This
approach provides the best approximation to the conduc-
tivity, but it is the most computationally intensive. Each
iteration requires calculation of the sensitivity and sev-
eral additional forward modelings.
The relative merits of these ve methods will probably de-
pend upon the complexity of the actual conductivity distribu-
tion. A general statement may therefore be difcult to make,
but insight can be obtained from applications to specic data
sets. We have applied these ve methods to the inversion of
our synthetic test data set shown in Figures 2 and 3.
We rst focus upon generating the approximate conductiv-
ities. The composite 2-D conductivity was obtained by invert-
ing the data from the seven eastwest lines using a 2-D algo-
rithmand stitching together the resulting 2-Dconductivities to
form a 3-D model. The one-pass approximate 3-D inversion
was carried out using a uniform background of 1 mS/m and a
model objective function with
s
=0.0001 and
x
=
y
=
z
=1;
we chose an optimal regularization parameter by the L-curve
criterion (Hansen, 1992) to account for both the linearization
error and the added randomerrors. The selected regularization
parameter, together with the objective function, also dened
the approximate inverse mapping. We performed two itera-
tions of AIM updates to produce the AIM approximation of
the conductivity. Last, we had the conductivity model from a
full 3-D inversion (Figure 4). For comparison, we have listed
in Table 2 the data mist between the observed dc potential
data and the predicted data obtained by applying 3-D forward
modeling to each of the ve approximate conductivity mod-
els. A comparison of these models with the true conductivity
model is shown in Figure 6. We selected the cross-section at
northing =475 m, which passes through four of the ve prisms
in the model. The four models from different inversions show
different levels of detail about the conductivity anomaly, and
they present a general progression toward better representa-
tions of the true model. However, the improvement diminishes
as the approximation approaches the best model that is ob-
tained from the full 3-D inversion. Although the inverted con-
ductivity models are similar, there is a substantial difference
between the true model and any one of these approximations.
Using these ve approximations to calculate the sensitivities,
we performed ve different inversions of the IP data. Since
some of the conductivity models are poorer approximations,
the corresponding IP inversions are not expected to achieve
the expected data mist. Instead, we chose an optimal regular-
izationparameter for eachinversionaccording toa generalized
cross-validation criterion. The result is that different inversions
mist the observed IP data by different amounts (Table 2).
The resulting chargeability models are compared with the true
model in Figure 7. Each panel in that gure is the cross-section
of the recovered chargeability model at northing 475 m. All
ve models recover the essential features of the true model,
and they present a general trend of improvement as the ap-
proximation to the background conductivity improves. How-
ever, the improvement in the recovered chargeability model is
not proportional to the increased computational cost involved
in constructing a better conductivity approximation. Less rig-
orous approximations of conductivity which require much less
computation have produced good representations of the true
chargeability model.
FIG. 6. Comparison between the ve approximate conductivity models with the true conductivity. All sections are at
northing =475 m, which passes through four of the ve prisms. The positions of the true prisms are outlined by the white boxes. As
the approximation improves, the inverted conductivity model is a better representation of the true model.
JOINT INVERSION OF SURFACE AND CROSSHOLE DATA
Crosshole data have been used to achieve higher resolution
image of the subsurface structure obtained from dc resistivity
and IP experiments (e.g., Spies and Ellis, 1995; LaBrecque and
Morelli, 1996). However, although crosshole data are sensitive
to the vertical variation of conductivity and chargeability, they
have rather poor sensitivity to the lateral variation because the
data have limited spatial distribution and the array separation
is restricted to a small range. Surface data, however, usually
have good areal coverage and therefore possess better resolv-
ing power for determining lateral variations in the subsurface
structure. Surface data can provide good complementary in-
formation to the crosshole data if the targets are within the
depth of penetration of the surface arrays. Joint inversion of
these two data sets was expected to improve the resolution of
the recovered chargeability model.
We placed four vertical boreholes around the anomalous re-
gion in the test model. The locations are shown in Figure 1. We
simulated crosshole data from a poledipole tomographic ex-
periment. Current sources were placed along the source hole
from 0 to 400 m depth at an interval of 25 m. For each current
location, potentials in another borehole (receiver hole) were
measured with a 50-m dipole at an interval of 25 m between
z =0 and 400 m. Figure 8 illustrates the electrode conguration
between two holes. Only one borehole in any pair of boreholes
was used as the source hole, and the reverse conguration of
switching the source and receiver holes was not used. This re-
sulted in six independent pairs of sourcereceiver holes.
Atotal of 1530 observations were generated for both dc and
IP experiments using this conguration. Because of the pres-
ence of zero crossings in the measured total potentials, we used
the secondary potential, instead of apparent chargeability, as
the IP data. The data were contaminated with independent
Gaussian noise. The standard deviation for dc potentials was
equal to 2% of each accurate datum; for secondary potentials
it was equal to 5% of each accurate datum plus a minimum
of 0.1 mV. (All the potentials were normalized to unit current
strength.) Figure 9 displays the crosshole dc data as the appar-
ent conductivity between two pairs of boreholes. The vertical
axis of the plot indicates the position of the current electrode in
the source hole, and the horizontal axis indicates the midpoint
of the potential dipole in the receiver hole. The data plots are
remarkably featureless, and identication of individual prisms
inthetruemodel is impossible. Figure10displays thesecondary
potentials in the same two pairs of boreholes. Again, there is
no distinct feature in the secondary potential plots. The lack
of distinct features in the borehole data is a direct indication
of the datas poor sensitivities to the lateral variations in the
conductivity and chargeability distributions.
FIG. 7. Comparison of chargeability models recovered fromthe 3-Dinversion of surface IP data using ve different approximations
to the background conductivity. The process by which each conductivity approximation is obtained is shown in each panel. The
lower-right panel is the true chargeability model.
Next, we compared the chargeability models obtained from
inverting the crosshole data alone with the model obtained
by jointly inverting the surface and crosshole data. For this
study, we invertedthe dc resistivity data using the full nonlinear
inversion so that the best conductivity approximation at our
disposal was used for the sensitivity calculation. In both dc and
IP inversions, we chose a model objective function by setting
the coefcients to
s
=0.0001 and
x
=
y
=
z
=1. A uniform
half-space of 1 mS/m was used as the reference model for dc
resistivity inversion, and a zero reference model was used for
theIPinversion. Intheinversions, theknownvalues of theerror
standard deviations were used, and the target mist value was
set to the number of data points being inverted. All inversions
converged to the expected mist value.
Figure 11 shows the conductivity model obtained from in-
verting the crosshole dc resistivity data, displayed in two plan
sections and one cross-section. This model is a crude represen-
tation of the true conductivity. Only the large surface conduc-
tor and the buried conductor are identied, and the recovered
anomaly amplitude is very low. We used this conductivity in the
inversion of crosshole IP data and the recovered chargeability
model (Figure 12). This model is a poor representation of the
true chargeability. Anomalies are recovered near the surface,
but they do not correspond to the locations of the true prisms.
The two deep prisms are marginally identied. The vertical
FIG. 8. Crosshole electrode conguration for collecting tomo-
graphic data. The current source A in hole B moves at an in-
terval of 25 m from z = 0 m to z = 400 m. For each current
location, the potential electrodes M and N, separated by 50 m,
measure the potential differences in hole D. The midpoint of
the potential dipole moves at an interval of 25 mfromz = 25 m
to z = 375 m. For a given pair of holes, no data are collected
by interchanging the current and potential holes.
FIG. 9. The crosshole plot of apparent-conductivity data. The vertical axis is the location of the current source in the source hole,
and the horizontal axis is the midpoint of the potential dipole in the receiver hole. The left panel is for data between holes C and
A, and the right panel is for data between holes B and D, as shown in Figure 1.
FIG. 10. The crosshole plot of secondary potential data in the same format as the crosshole apparent conductivity plots in Figure 8.
The potentials are normalized to unit current strength, and the grayscale indicates the value in millivolts.
extent is well imaged, as would be expected from borehole
data, but the orientations and horizontal boundaries of the
recovered anomalies differ from those of the true model. In
addition, there is excessive structure in the region immediately
surrounding the boreholes. This is typical when crosshole data
are inverted unless special weighting in the objective function
is included to counter it.
We next jointly inverted the surface data in Figures 2 and 3
and the crosshole data in Figures 8 and 9. Figure 13 displays
the recovered conductivity model from the joint inversion of
surface and borehole data. This model is dominated by the fea-
tures recovered in the surface data inversion. The minor im-
provements are the increased amplitude and the slightly better
denition of the depth extent of the conductivity prism. Using
this model we calculatedthe sensitivity andthenperformedthe
joint inversion of the two IPdata sets to recover the chargeabil-
ity model shown in Figure 14. It shows dramatic improvement
compared with the models from individual inversions in Fig-
ures 5 and 12. All ve prisms are well resolved, and artifacts
surrounding the boreholes are minimized. The most noticeable
improvement is the clear image of the two separate buried tar-
gets. The recovered amplitudes, positions, and orientations of
the two anomalies all correspond well with the true model.
FIELD EXAMPLE
As our last example, we illustrate the 3-D inversion algo-
rithm using a set of poledipole data from the Mt. Milligan
coppergold porphyry deposit in central British Columbia,
Canada. These data were rst analyzed by Oldenburg et al.
(1997) using a 2-D algorithm. We invert them using the 3-D
algorithm, which illustrates the 3-D inversion in a mineral ex-
ploration setting and provides a comparison with the result
from a series of 2-D inversions.
The Mt. Milligan deposit lies within the Early Mesozoic
Quesnel terrane, which hosts a number of Cu-Au porphyry de-
posits, and it occurs within porphyritic monzonite stocks and
adjacent volcanic rocks. The initial deposit model consists of a
vertical monzonitic stock, known as the MBX stock, intruded
FIG. 11. Conductivity model recoveredfromthe crosshole data
alone. This model shows an elongated conductor on the sur-
face and a broad conductor at depth. Both conductors are sur-
rounded by resistive halos, and the amplitude is small.
into volcanic host rocks. Dykes extend from the stock and
cut through the porous trachytic units in the host. Emplace-
ment of the monzonite intrusive is accompanied by intensive
hydrothermal alteration primarily near the boundaries of the
stock and in and around the porous trachytic units cross-cut
by monzonite dykes. Potassic alteration, which produced chal-
copyrite, occurs in a region surrounding the initial stock, and its
intensity decreases away from the boundary. Propylitic alter-
ation, which produces pyrite, exists outward from the potassic
alteration zone. Strong IP effects are produced by these alter-
ation products, and the IP survey is well suited for mapping
the alteration zones. The poledipole dc resistivity and IP sur-
veys over Mt. Milligan were carried out along eastwest lines
spaced 100 mapart. The dipole length was 50 m, and n-spacing
was from 1 to 4. This yielded 946 data points along 11 lines in
FIG. 12. Chargeability model recovered from the crosshole
data alone. This model shows little resolution near the sur-
face where the anomalies are conned to small volumes. The
deeper anomalies are identied but not well resolved. As ex-
pected, the depth extent of the buried chargeable bodies is well
delineated.
our study area of 1.2 1.0 km. This area, directly above the
MBXstoc, has a gentle surface topography, and the total relief
is about 100 m. Figure 15 displays the apparent chargeability
data in plan maps of constant n-spacings. For brevity, we have
not shownthe dc resistivity data here. The apparent chargeabil-
ity data showlarge anomalies toward the western and southern
regions. The north-central region of lowapparent chargeability
is related to the intrusive stock that has signicantly less sulde
from the alteration processes.
To invert these data, we used a mesh that consisted of cells
25 m wide in both horizontal directions and 12.5 m thick in
the region of interest. The mesh was extended horizontally and
downward by cells of increasing sizes. The total number of cells
in the inversion was about 72 000. We rst performed the full
nonlinear inversionof theDCresistivitydataandthenusedit to
FIG. 13. Conductivity model recoveredfromthe joint inversion
of surface and crosshole data. This model is similar to that
obtained from surface data alone, but it has a slightly higher
amplitude for the buriedconductive anomaly. The depthextent
of the anomaly is also slightly better dened.
carry out the IP inversion. The resulting model is shown in Fig-
ure 16. For comparison, we alsoplottedthe chargeability model
created by combining the 2-Dsections obtained frominverting
the 11 lines of data separately using a 2-D algorithm. The re-
covered 3-D chargeability models from these two approaches
were consistent, and they both imaged the large-scale anoma-
lies reasonably well. This was not surprising since the limited
array length meant there was little redundant information in
the data from adjacent lines. The model recovered from the 3-
Dinversion was somewhat smoother and showed less spurious
structure than the composite 2-Dmodel. It also showed a well-
dened central zone of low chargeability at depth. This was a
clearer image of the monzonite stock than what was imaged in
the 2-D inversions.
FIG. 14. Chargeability model recovered from the joint inver-
sion of surface and crosshole data. This model shows the great
improvement achieved by joint inversion of the two comple-
mentary data sets. All ve anomalies are well resolved. Es-
pecially noticeable is that the boundaries of the two buried
chargeable bodies are well delineated.
DISCUSSION
We developed a 3-D IP inversion algorithm that applies to
data acquired using arbitrary electrode congurations on a to-
pographically variable earth surface or in boreholes. We as-
sumed that the chargeability is small and formulated the in-
version as a two-step process. First, the dc resistivity data are
inverted to generate a background conductivity. That conduc-
tivity is used to generate the sensitivity matrix for the IP equa-
tions. The 3-Dchargeability model is then generated by solving
the systemof equations, subject to a restriction that the charge-
ability is everywhere positive andsmaller thananupper bound.
The analysis of large IP data sets often takes place in stages.
The rst goal is to obtain an image that reveals the major sub-
surface structures, answers questions about the existence of
buried targets, and supplies approximate details about size and
location. Major components affecting this image are the choice
of model objective function, the amount and type of errors on
the data, and the degree to which the data are t. For our two-
step process, we must also pay attention to how valid the lin-
earization process is and how close the recovered conductivity
is to the true conductivity.
The question of how close the recovered conductivity needs
to be to the true conductivity so that sensitivity J is a good esti-
mate of the true sensitivities is not addressed quantitatively in
this paper. We have, however, carried out an empirical test in
a single example. Four approximate conductivities were used
to generate the sensitivities. In general, higher quality dc in-
FIG. 15. The IP data from an area above the MBX stock of the Mt. Milligan coppergold porphyry deposit in central British
Columbia. The data were acquired using a poledipole array with a dipole length of 50 mand n-spacing from1 to 4. The four panels
are plan maps of the data corresponding to different n-spacings.
versions yielded better IP results, with the half-space conduc-
tivity, a one-pass linearized inversion, a few passes of an AIM
approach, and the Gauss-Newton inversion giving progressive
improvement. However, the differences in the nal IP inver-
sions fromthese various approximations were fairly subtle (see
Figure 7). In fact, these differences were smaller than changes
in the image obtained by adjusting the degree to which the data
are mist or by slightly altering the model objective function
being minimized. Yet the various approximations to the con-
ductivity can be produced with substantially fewer computa-
tions than the full Gauss-Newton solution. This allows the user
to carry out a number of rst-pass inversions with a data set
to achieve insight about the gross distribution of earth charge-
ability. If the conductivity structure is not overly complicated,
then this result may be satisfactory for nal interpretation. The
question of how well the conductivity must be known is a po-
tential area for further research.
Another approximate conductivity model is that generated
by combining results from 2-D inversions. The prevalence of
2-D inversion algorithms means that this information is gen-
erally available when data have been collected along parallel
lines. We knowthat off-line anomalies and 3-Dtopography will
cause distortions in the recovered 2-D conductivity models, so
some degree of caution is required. In the synthetic modeling
presented here and in the Mt. Milligan example, the 2-D anal-
ysis for conductivity worked satisfactorily. Further research is
required to provide more detailed rules about when 2-D is ap-
plicable.
An important aspect of our IP inversion is the incorporation
of upper and lower bounds on the chargeability. The lower
bound is physical since chargeability is positive. The upper
bound might be (1) assignable from a priori knowledge about
the nature of the mineralization or (2) assigned to generate a
model consistent with the linearized formulation of the equa-
tions. Linearization requires that the chargeability be small.
The positivity and upper bounds are implemented through a
primal logarithmic barrier method. This increases the complex-
ity of the algorithm, but the method is well established in the
literature and we provide an explicit algorithm for its imple-
mentation.
Another major component of our algorithm is the introduc-
tion of the wavelet transform to perform the matrixvector
multiplications. The sensitivity matrix can be compressed by a
factor of at least 10. This leads to substantial savings in both re-
FIG. 16. Comparison of the chargeability models obtained from 2-D and 3-D inversions of the data from Mt. Milligan shown in
Figure 15. The column on the left shows one cross-section and two plan sections of the 3-Dmodel obtained by combining eleven 2-D
sections recovered from 2-D inversions. The column on the right shows the model obtained by performing a single 3-D inversion
of all the data. The two results are generally consistent. However, less spurious structure is present in the model from the 3-D
inversion, and the central zone of low chargeability corresponding to the MBX stock is imaged better.
quired memory and CPUtime. This has made the algorithmat
least ten times faster than a direct approach and consequently
has allowed us to routinely handle problems that have a few
thousand data and a hundred thousand cells with relative ef-
ciency.
Last, the application of our algorithm to joint surface and
crosshole data has demonstratedthat the inversionof these two
complementary data sets can greatly improve the resolution
of the inverted chargeability model. The noticeable gains are
in the enhanced denition of both horizontal boundary and
vertical extent of buried chargeable zones.
ACKNOWLEDGMENTS
We thank Roman Shekhtman for his valuable assistance
in programming the code and in running numerical exam-
ples. This work has been supported by an NSERC IOR grant
and an industry consortium, 3-D Inversion of DC resistivity
and Induced Polarization Data (INDI). Participating compa-
nies are Placer Dome, BHP Minerals, Cominco Exploration,
Falconbridge, INCO Exploration & Technical Services, New-
mont Gold Company, and Rio Tinto Exploration.
REFERENCES
Dey, A., and Morrison, H. F., 1979, Resistivity modelling for arbitrarily
shaped three-dimensional structures: Geophysics, 44, 753780.
Ellis, R. G., and Oldenburg, D. W., 1994a, The pole-pole 3-D DC-
resistivity inverse problem: A conjugate-gradient approach: Geo-
phys. J. Internat., 119, 187194.
1994b, 3-Dinduced polarization inversion using conjugate gra-
dients: Presented at the John Sumner Memorial Internat. Workshop
on Induced Polarization (IP) in Mining and the Environment.
Gill, P. E., Murray, W., Ponceleon, D. B., andSaunders, M., 1991, Solving
reduced KKT systems in barrier methods for linear and quadratic
programming: Stanford Univ. Technical Report SOL 917.
Golub, G. H., Heath, M., and Wahba, G., 1979, Generalized cross-
validation as a method for choosing a good ridge parameter: Tech-
nometrics, 21, 215223.
Haber, E., and Oldenburg, D. W., 2000, A GCV-based method for
nonlinear ill-posed problems: Comp. Geosci., in press.
Hansen, P. C., 1992, Analysis of discrete ill-posed problems by means
of the L-curve: SIAM Review, 34, 561580.
Kowalczyk, P. L., Logan, K. J., and Bradshaw, P. M. D., 1997, New
methods in geophysics to visualize geology in tropical terrains: 4th
Decennial Internat. Conf. Min. Expl., Proceedings, 829834.
LaBrecque, D. J., 1991, IP tomography: 61st Ann. Internat. Mtg., Soc.
LaBrecque, D. J., and Morelli, G., 1996, 3-D electrical resistivity to-
mography for environmental monitoring: Symp. on Appl. Geophys.
to Engin. and Environ. Problems, Proceedings.
Li, Y., and Oldenburg, D. W., 1994, Inversion of 3-DDCresistivity data
using an approximate inverse mapping: Geophys. J. Internat., 116,
527537.
1999a, Fast inversion of large scale magnetic data using wavelet
transforms: Geophysical J. Internat., accepted for publication.
1999b, 3-D inversion of DC resistivity data using an L-curve
criterion: 69th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded
Abstracts, 251254.
Loke, M. H., and Barker, R. D., 1996, Rapid least-squares inversion of
apparent conductivity pseudosection using a quasi-Newton method:
McGillivary, P. R., andOldenburg, D. W., 1990, Methods for calculating
Frechet derivatives and sensitivities for nonlinear inverse problem:
A comparative study: Geophys. Prosp., 38, 499524.
Mller, I., Christensen, N. B., and Jacobsen, B. H., 1996, 2-D inversion
of resistivity prole data: Symp. on Appli. Geophys. to Engin. and
Environ. Problems, Proceedings.
Mutton, A. J., 1997, The application of geophysics during evaluation of
the Century zinc deposit: 4th Decennial Internat. Conf. Min. Expl.,
Proceedings, 599614.
Oldenburg, D. W., and Ellis, R. G., 1991, Inversion of geophysical data
using an approximate inverse mapping: Geophys. J. Internat., 105,
325353.
Oldenburg, D. W., and Li, Y., 1994, Inversion of induced polarization
Oldenburg, D. W., McGillivary, P. R., and Ellis, R. G., 1993, General-
ized subspace method for large scale inverse problems: Geophys. J.
Internat., 114, 1220.
Oldenburg, D. W., Li, Y., and Ellis, R. G., 1997, Inversion of geophys-
ical data over a copper-gold porphyry deposit: A case history for
Mt. Milligan: Geophysics, 62, 14191431.
Park, S. K., and Van, G. P., 1991, Inversion of pole-pole data for 3-D
resistivity structure beneath arrays of electrodes: Geophysics, 56,
951960.
Sasaki, Y., 1994, 3-D resistivity inversion using the nite-element
method: Geophysics, 59, 18391848.
Saunders, M., 1995, Cholesky-based methods for sparse least squares:
The benets of regularization: Stanford Univ. Technical Report SOL
951.
Seigel, H. O., 1959, Mathematical formulation and type curves for in-
duced polarization: Geophysics, 24, 547565.
Spies, B. R., and Ellis, R. G., 1995, Cross-borehole resistivity tomogra-
phy of a pilot-scale, in-situ vitrication test: Geophysics, 60, 886898.
Tikhonov, A. V., and Arsenin, V. Y., 1977, Solution of ill-posed prob-
lems, ed. J. Fritz: John Wiley & Sons.
Wahba, G., 1990, Spline models for observational data: Soc. Ind. Appl.
Math.
Zhang, J., MacKie, R. D., and Madden, T. R., 1995, 3-D resistivity
forward modelling and inversion using conjugate gradients: Geo-
physics, 60, 13131325.
The application of geophysics during evaluation
of the Century zinc deposit
Andrew J. Mutton
ABSTRACT
During the period 1990 to 1995, experimental pro-
grams using high-resolution geophysics at several
Australian operating mines and advanced evaluation
projects were undertaken. The primary aimof those pro-
grams was to investigate the application of geophysical
technology to improving the precision and economics of
the ore evaluationandextractionprocesses. Geophysical
methods used for this purpose include:
1) borehole geophysical logging to characterize ore
and rock properties more accurately for improved
correlations between drill holes, quantication of
resource quality, and geotechnical information
2) imaging techniques between drill holes to map
structure directly or to locate geotechnical prob-
lems ahead of mining
3) high-resolution surface methods to map ore con-
tacts and variations in ore quality, or for geotech-
nical requirements
In particular, the use of geophysics during evalua-
tion of the Century zinc deposit in northern Australia
demonstrated the potential value of these methods to
the problems of dening the lateral and vertical ex-
tent of ore, quantitative density determination, predic-
tion of structure between drill holes, and geotechnical
characterization of the deposit. An analysis of the po-
tential benet of using a combination of borehole geo-
physical logging and imaging suggested that a more
precise structural evaluation of the deposit could be
achieved at a cost of several million dollars less than
the conventional evaluation approach based on anal-
ysis from diamond drill-hole logging and interpolation
alone.
The use of geophysics for the Century evaluation also
provided substance to the possibility of using systematic
geophysical logging of blast holes as an integral part of
the ore extractionprocess. Preliminary tests indicate that
ore boundaries can be determined to a resolution of sev-
eral centimeters, and ore grade can be estimated directly
to a usable accuracy. Applying this approach routinely
to production blast holes would yield potential benets
of millions of dollars annually through improved time-
liness and accuracy of ore boundary and quality data,
decreased dilution, and improved mill performance.
Althoughthe indications of substantial benets result-
ing from the appropriate and timely use of geophysics at
Rio Tintos mining operations are positive, some chal-
lenges remain. These relate largely to the appropriate
integration of the technology with the mining process,
and acceptance by the mine operators of the economic
value of such work. Until the benets are demonstrated
clearly over time, the use of geophysics as a routine com-
ponent of evaluation and mining is likely to remain at a
low level.
INTRODUCTION
Geophysics has been used by the metalliferous mining in-
dustry primarily as a tool for exploration. By contrast, the
petroleum and coal industries have exploited geophysics to
a far greater extent in terms of using the data to quantify the
value, size, or production capabilities of their resource. The ap-
plication of geophysics to mineral-deposit evaluation and min-
ing is not developed well in general. Some possible reasons for
this are:
Rio Tinto Technical Services, P.O. Box 2207, Milton, Queensland 4064, Australia. E-mail: andrew.mutton@riotinto.com.
1) insufcient knowledge by mining managers, engineers,
and operators of the existence of geophysical method-
ologies, coupled with a high techhigh cost perception
of geophysics
2) perceived infrastructure and logistical difculties in using
geophysics in a working mine environment
3) lack of knowledge of the physical properties of the ore
and host rocks which may be exploitable in a mining
environment
1946
Geophysical Applications at Century Zinc Deposit 1947
4) limited research effort applied to standard geophysi-
cal methods to develop higher-resolution acquisition or
interpretation tools applicable to mining applications,
rather than tools applicable to exploration
5) scarcity of geophysicists with access to mine problems or
with sufcient knowledge of mining culture or opera-
tional requirements to champion the use of geophysics in
this area
During the period 1990 to 1995, CRA Limited (now
Rio Tinto) undertook experimental programs using high-
resolution geophysics at several Australian operating mines
and advanced evaluation projects, to investigate opportunities
to develop and test geophysical applications related directly to
resource denition and mining problems. Using some exam-
ples of results obtained during the evaluation by CRA of the
Century zinc deposit innorthernAustralia, this paper describes
the opportunities available and the rationale for applying geo-
physics beyond the exploration stage of mineral development.
WORK UNDERTAKEN
Arange of geophysical surveys, some conventional andsome
adapted for specic problems, has been tested in recent years
at several CRA evaluation projects and mining operations to
enhance ore-boundary denition and recoveries. The surveys
fall into several categories, including:
1) High-resolution surface surveys to map remotely the
lateral and vertical limits of ore or other geotechnical
parameters required for mine design. Successful applica-
tion of such methods potentially can reduce the amount
of closely spaced expensive drilling generally needed for
these requirements, and can enable improved reserves
calculation and mine design. Examples of such work
incorporate the use of standard methods such as mag-
netics, electromagnetics (EM), and IP/resistivity using
improved modeling and data enhancements, or new-
generation methods such as borehole EM imaging and
ground-penetrating radar (GPR).
2) Geophysical logging of evaluation drill holes to quan-
tify mineralogical, geotechnical, and structural parame-
ters of ore and host rocks in situ. Many of the standard
petroleum logging tools, such as density, sonic, and dip-
meter, have been adapted successfully to the hard-rock
environment, along with developments in new tool tech-
nology such as magnetic susceptibility and conductivity.
Improved condence and demonstrated success with this
approach is aimed mainly at providing signicant cost
savings by the future routine use of noncore rather than
core drilling.
3) Interborehole imaging methods using electromagnetic or
seismic sources. Examples of suchmethods include radio-
imaging (RIM) surveys between holes to map ore con-
tinuity and structure. The use of these methods is de-
signed to improve the knowledge and reliability of the
ore boundaries at an earlier stage of evaluation, which
should reduce the time and cost of the evaluation pro-
cess.
4) Geophysical logging of blast holes to dene ore-waste
contacts to a resolution of a fewcentimeters and, in some
cases, to estimate ore grade directly. Applying this ap-
proach routinely to production blast holes is designed to
yield substantial cost benets through improved timeli-
ness and accuracy of resource information, less dilution,
and improved mill performance.
Much of CRAs investigation of the use of geophysics for
mining and mine evaluation was carried out at the site of
the Century zinc deposit in northern Australia, discovered by
CRA Exploration in 1990. Since 1997, it has been owned and
operated by Pasminco Pty Limited. The discussion and exam-
ples presented belowwill focus on the work at Century and the
role geophysics played in the various phases of the evaluation
of the deposit prior to 1995.
RESOURCE DEFINITION GEOPHYSICS AT THE CENTURY
ZINC DEPOSIT
Geology
The Century Zn-Pb-Ag deposit is located about 250 km
north-northwest of Mount Isa, Queensland. Substantial
drilling (about 500 diamond core holes) from1990 to 1995 out-
lined a geologic resource containing about 120 million metric
tons of 10.5% zinc, 1.5% lead, and 35 g/ton silver.
The Century deposit occurs within sediments of the
Proterozoic Mount Isainlier andis locallyunconformablyover-
lain by Cambrian rocks comprising dolomitic limestone, chert,
and chert breccia (Figure 1). Dolomitic siltstones and carbona-
ceous shales host the deposit. The mineralized sequence is
FIG. 1. Century deposit location and geology.
1948 Mutton
about 40 m thick and consists of four laterally continuous sub-
divisions (units 1 to 4). The bulk of the mineralization occurs
as strata-bound, banded sphalerite, galena, and pyrite within
black carbonaceous shales of units 2 and 4. A barren dolomitic
siltstone bed within unit 3 represents a semicontinuous marker
bedthroughout muchof the deposit. The mineralizedsequence
has excellent grade continuity but is disrupted by late faulting.
A detailed description of the geology and mineralogy of the
deposit is included in Waltho et al., 1993.
Figure 2 shows a schematic north-south cross-section
through the deposit. The deposit consists of a smaller, shallow
southern block which subcrops in the southwestern margin of
the orebody, and a larger, deeper northern block completely
concealed beneath Cambrian limestone and recent alluvium.
Bounding surfaces to the deposit comprise either postmineral-
ization faults or Cambrian and younger erosional surfaces.
Physical characteristics of the ore and host
Comprehensive laboratory petrophysical measurements on
Century core were made on several samples as part of an
Australian Mineral Industry Research Association (AMIRA)
project (project P436) to investigate the application of geo-
physics in mine planning and mining (Fullagar et al., 1996a;
Fullagar and Fallon, 1997). These data have been evaluated in
conjunction with extensive geophysical log data collected from
a large proportionof the holes drilled. Atypical composite geo-
physical log through the Century ore sequence and host rocks
is presented in Figure 3.
The petrophysical results show that the Century orebody
is typical of many shale-hosted sulde base-metal deposits
FIG. 2. Century deposit cross-section 46800E.
throughout the world. In particular, the ore is characterized
by distinctive physical properties, including higher density, low
natural radioactivity, and low magnetic susceptibility.
The main difference from many other shale-hosted deposits
is that the Century ore, consisting mainly of sphalerite with
relatively low iron sulde, is not very conductive and in parts is
more resistive than barren host shales. However, petrophysi-
cal measurements on core and downhole induced-polarization
(IP) surveys show the ore has high electrical chargeability.
Figure 4a shows a typical downhole IP log through the deposit,
and Figure 4b is a statistical summary of the IP and resistivity
response recorded in several drill holes. These results clearly
indicate howchargeable the ore units are compared with other
lithologies.
The analysis of physical properties also indicates diagnos-
tic contrasts among some of the host lithologies. Although
this knowledge is not specically useful for ore evaluation,
it has helped to address other problems arising during the
project development. For example, the high resistivity of the
Cambrian limestone has been exploited to assist with a hydro-
logic problem associated with development of a large pit (see
below).
Role of geophysics
The discovery of Century was based largely on testing of
a zinc soil geochemical anomaly which was delineated on
regional gravity, ground magnetic, and soil sample traverses
(Thomas et al., 1992; Broadbent, 1996). No anomalous re-
sponse was detected on the gravity or magnetic data or on
other geophysical surveys attempted prior to the discovery.
Despite the lack of success of geophysical techniques in
the initial detection of ore, the recognition after the discov-
ery of contrasts in the physical properties of the ore and
host sediments provided some condence that geophysical
methods could have a role in further exploration and evalu-
ation of the deposit. Geophysical surveys subsequently carried
out over and within the Century deposit since its discovery
have included detailed ground and airborne surveys (gravity,
magnetics, electromagnetics, IP/resistivity, reection seismic),
borehole geophysical logging, and interborehole geophysical
imaging
The work resulted in some successes, in particular relating
to dening the overall lateral and vertical extent of mineraliza-
tion, mapping lithologic boundaries and structure within the
deposit, providing data for ore-reserve estimates and mine
planning, supporting geotechnical studies for pit design, and
aiding ongoing exploration in the area
Detailed surface surveys
Lateral and vertical extent of mineralization.After the ini-
tial discovery of zinc-rich, low-sulde mineralization and the
realisation that EMmethods did not detect the mineralization,
the IP method was suggested as a possible way of mapping the
extent of the mineralization beneath the Cambrian cover. A
FIG. 3. Century deposit composite geophysical log.
limited number of IP/resistivity traverses subsequently were
completed over the Century deposit and immediate environs,
indicatingthepresenceof agoodIPanomalyapproximatelyco-
incident withthe mineralizationas knownat that time (Thomas
et al., 1992). IP/resistivity logging of selected drill holes sub-
sequently conrmed the source of the surface IP anomalies.
Figure 4a presents an example of downhole IP results in one
hole, which clearly indicates the high chargeability response of
the mineralized zone.
The downhole IP/resistivity surveys thus provided the reas-
surance that IP anomalies represented good exploration tar-
gets. In fact, initial drilling into the deeper northern ore zone
(Figure 2), which is masked completely by hundreds of meters
of barren limestone and sediments, essentially was guided by
the presence of the IP anomaly in that area. Although only a
limited number of IP lines were completed over the deposit,
the IP results, in combination with drilling, helped to conrm
the lateral limits of mineralization far sooner in the evaluation
process than if IP surveying had not been attempted.
As well as mapping the lateral extent of the orebody, later
work by CRA Exploration demonstrated that it is possible to
transform the IP/resistivity data into an approximate depth
image of the mineralization (representing the vertical distri-
bution of chargeable material). This work used inversion algo-
rithms developed by the Geophysical Inversion Facility group
1950 Mutton
at the University of British Colombia (Oldenburg et al., 1997).
Figure 5 shows the image derived from unconstrained inver-
sion of surface IP/resistivity data adjacent to the cross-section
presented in Figure 2. The outline of the orebody derived from
subsequent drilling is superimposed.
This work clearly demonstrates the potential mapping
capabilitiesboth laterally and verticallyof a surface geo-
a)
FIG. 4. Century downhole IP/resistivity results: (a) log for DDH LH117; (b) statistical summary for all IP/resistivity logs from
Century deposit.
physical method for which the physical property contrasts have
a close association with the mineralization, dening the ap-
proximate extent of ore prior to extensive drilling. The image
of the orebody provides the explorer with improved capability
for successful drill targeting. However, it also has the poten-
tial, by the application of constraints from initial drilling, to
dene the ore boundaries and estimate reserves substantially
at a far earlier stage of the evaluation process than hitherto
possible.
Geotechnical applications.Although the predominantly
sphaleritic Century ore has little electrical conductivity con-
trast with the host sediments, surface electromagnetic methods
have been applied successfully, based on the conductivity con-
trast between the Proterozoic sediments and overlying lime-
stone (Figure 4), to assist geotechnical work associated with
pit design. Two phases of Controlled Source Audio-Frequency
Magnetotelluric (CSAMT) surveys were completed over the
deposit to (1) locate large blocks of detached Proterozoic shale
within the limestone (the shale presents a geotechnical hazard
for pit-slope stability during initial excavation of the limestone
portion of the proposed pit, and (2) determine the thickness
of the surrounding water-saturated limestone to estimate the
likely water ow into the open pit as it is excavated.
The CSAMT method was chosen in preference to conven-
tional TEM methods because of logistical considerations as-
sociated with the limestone topography, and the belief that
CSAMT is more sensitive than TEM methods to delineat-
ing contrasts in the more resistive lithologies present in the
Century area.
The rst work was only partially successful, in that only very
large blocks of shaletypically greater than 50 50 50 m
could be delineated. The survey results did not offer sufcient
resolution to locate smaller blocks condently, but improved
survey design may have provided better resolution.
FIG. 5. Century deposit Line 46800E IP pseudosection and inversion model compared with orebody location from drilling. IP data
are from 100-m dipole-dipole frequency, domain survey.
The second phase of work proved useful in mapping the base
of the limestone and hence assisting with the hydrologic study
for the pit design. Figure 6 (after Mayers and Bourne, 1994)
shows a plan of the resistivity model obtained from inversion
of the CSAMT data and collated at a vertical depth of 100 m.
The intense blue represents the presence of resistive limestone
at this depth, and the warmer colors indicate the presence of
less resistive shale and siltstone.
The depth to the base of the limestone is shown in selected
drill holes. In general, there is a good correlation between the
intersected thickness of limestone and depths predicted from
the CSAMT resistivity model. The results show that postde-
positional faulting has led to large variations in the limestone
thickness in the vicinity of Century and, importantly, the lime-
stone is necked to the north and east of the deposit. The
CSAMT result thus gives some condence that the total vol-
ume of water ow into a proposed pit will be signicantly less
from the thinning of the water-saturated limestone than if the
limestone had been uniformly thick to the north and east of
Century.
Borehole surveysgeophysical logging
Although geophysics did not feature strongly in the initial
discovery of ore, downhole geophysical characterization log-
ging was carriedout at Century inseveral campaigns commenc-
ing soon after the initial drill holes were completed. The work
was commissioned for several reasons:
1952 Mutton
1) to help geologists and geophysicists gain a better under-
standing of the in situ physical properties of the deposit
and its host rocks as an aid to interpretation of explo-
ration and geotechnical data
2) to provide data to assist with the correlation of lithologies
and denition of the structure of the ore zones between
drill holes
3) to provide quantitative in situ density information to as-
sist with estimation of the ore reserves
4) to provide rock-strength information to assist in the
geotechnical assessment of the deposit by mining
engineers
The logging programs used standard suites of electrical and
nuclear tools (resistivity, resistance, self potential [SP], natural
gamma, density, andneutron), as well as tools suchas dipmeter,
magnetic susceptibility, and sonic velocity, which had not been
used commonly in base-metal geophysical logging prior to that
time.
Lithologic correlations.Interpretation of the downhole
logs indicates that the major lithologic units and the mineral-
ized zones can be distinguished readily by their density, natural
FIG. 6. Century deposit region showing smooth-model inver-
sionof CSAMTresistivity at 100-mdepth, comparedwithlime-
stone depth from drilling. The more resistive areas (blue) rep-
resent limestone greater than 100 m thick.
Table 1. Summary of Qualitative Log Responses.
Prot. Prot. shale and Prot. shale
Cambrian sandstone siltstone Ore zone and siltstone
limestone (h/wall) (h/wall) (Units 14) (footwall)
Natural gamma (API cps) very low (25) high (160) moderate (125) low-mod (70150) high (200)
Magnetic Susceptibility very low (<10) low (050) low-moderate v. low (ore)-mod (waste) low-moderate
(SI 10
5
) (0100) (0150) (50150)
Resistivity (ohm-m) high (>1000) low (60) low (75) variable (50200) low (80)
Density (g/cm
3
) moderate (2.7) low (2.6) moderate (2.7) mod-high (2.83.0) moderate (2.62.7)
Neutron (API cps) high (1500) high (1600) moderate (1200) low (700-variable) moderate (1200)
Sonic velocity (m/s) very high moderate moderate (4000) mod-high (40005000) moderate (4000)
(50006000) (4500)
gamma, neutron, resistivity, sonic, and magnetic susceptibility
properties. Figure 3 shows a composite log of a typical drill
hole through the Century deposit. Table 1 summarizes the re-
sponses that characterize the main lithologic units for each of
the main log types.
The mineralized sequence (units 1 to 4) is differentiated well
by the density, natural gamma, andmagnetic susceptibility logs.
The density contrast is attributed to the presence of sulde and
an increase in iron carbonate (siderite) content. The internal
waste unit within the ore zone (unit 3.2) is characterized by
lower density and generally higher gamma, magnetic suscepti-
bility, and resistivity, compared with the mineralized intervals.
Quantitative analysis.The quantitative analysis of the log
data, which attempts to transform such data into usable physi-
cal or chemical quantities such as density, grade, rock strength,
etc, is dependent on the equivalence of data from the different
logging tools used at various times and by various contractors.
Differences arise from differences in tool specications and
also from environmental differences, including borehole di-
ameter, uid content and properties, temperature, weathering,
smoothness of the hole wall, etc.
Comparisons to a calibration standard, such as the API stan-
dard, are used by most logging contractors but do not account
for most of the environmental factors. For example, an inspec-
tion of different API-standard calibrated natural gamma logs
can show some marked offsets in absolute values recorded.
The most successful form of calibration achieved at Century
has been the use of a calibration drill hole, which was resur-
veyed at regular intervals during each logging program. Such
actionensuredthat the repeatability of logs over a knowninter-
val could be compared and quantitatively adjusted by means
of a correction factor, if necessary.
Quantitative analyses undertaken on the Century data have
includedautomatedlithologic andgrade-predictionstudies but
primarily have been aimed at determining in situ density and
rock-strength variations.
Automated lithologic prediction.Experimental work de-
signed to test the possibility of predicting lithology and grade
directly froma combination of geophysical logs was carried out
on selected geophysical logs from Century. Three approaches
were attempted: (1) articial neural networks (ANN), (2) lin-
ear regression and discriminant analysis, and (3) cluster analy-
sis (lithology prediction only).
Some encouragement was obtained from the initial predic-
tion tests, but it was clear that factors such as measurement
uncertainty (calibration differences, depth discrepancies in the
training sets, etc.) or the level of data conditioning (such as
ltering) can have a large inuence on the predictions. The
results produced from the ANN tests were not sufciently en-
couraging in either lithology or grade prediction for this to be
considered a viable method at present.
On the other hand, reasonable predictions of lithology were
obtained from the cluster analysis method, using the LogTrans
FIG. 7. Lithology-basedLogTrans geologic interpretationof Century drill hole LH643 derivedby using statistics fromnearby control
drill holes. The interpretation has predicted successfully a normal fault across which part of the orebody is missing.
software developed as part of the AMIRA P436 project
(Fullagar et al., 1996a). Figure 7 shows an example of predic-
tion of lithology at Century using this method. The prediction
for the hole presented (LH643) is based on a statistical analysis
of data from holes nearby, so in effect, the prediction is uncon-
strained by the log results and geology fromLH643. The results
compare favorably with the mapped lithologies, even though
part of theoresequenceinLH643is missingbecauseof faulting.
1954 Mutton
Improvements to and routine use of such software in the
future could save considerably on the cost of manual logging
of core, but the greatest saving would result from the use of
percussion rather than core drilling for a large proportion of
the drill holes. This approach would be applicable to a deposit
such as Century in which the need for closely spaced drilling is
related more to structural interpretation than to grade control.
Density.The most extensive quantitative analysis of the
Century log data has been done for the purpose of density
determinationfor ore-reserve estimation. Several independent
methods were considered and tested to determine a suitable
approach. The methods included:
1) Archimedes-type measurements on whole core at site
or by a laboratory on core fragments submitted for geo-
chemical analyses
2) laboratory measurement of density of pulverized and ho-
mogenized drill-core samples by acetone titration
3) physical measurement (dimensions and weight) of whole
drill core at site
4) stoichiometry, based on base metal and sulfur assay data
5) geophysical (gamma-gamma) logging of drill holes
Of these methods, geophysical logging is considered to pro-
vide the most consistent measure of bulk density, although
physical measurement and stoichiometric calculations based
on zinc, lead, iron, manganese, and sulfur assay data also
FIG. 8. Comparison of gamma-gamma log-derived density (showing differences in calibration standards) and density measurements
on core for Century drill hole LH483.
proved useful. The titration method proved to be the most
unreliable, yielding many spurious values probably caused by
small variations in the measurement method.
All methods have limitations. For example, the chemical
assaybased methods yield data restricted to the assay inter-
vals only, and the physical-measurement methods are more
subject to measurement error arising from slight variations in
the procedure. Geophysical logging appears to yield very con-
sistent results. However, the conversion of the measured count
rate to a density value is subject to error because of differences
betweenthe material containedwithinthe volume of rocksam-
pled by the logging tool and the smaller volume of rock in the
core used as the reference value.
Calibrationfactors suppliedby the logging contractor should
be used with caution in hard-rock mining environments, be-
cause most of the calibration standards are devised for logging
in lower-density sedimentary environments hosting petroleum
or coal deposits. Figure 8 shows the comparison between the
contractors derived density and the recalculated density log
using the calibration derived from the physical measurement
of core. The result of using the former is a large overestimation
of the ore density and therefore the contained tonnage of ore.
The major advantages of using geophysical logging for den-
sity determination are that (1) the data are available and con-
sistent through the length of the drill hole rather than just over
an assayed interval, and (2) data can be obtained for noncored
holes, reducing the requirement for cored holes. However, the
noncored holes must be drilled well so that the hole walls are
reasonably smooth. Caving andirregularities inthe wall are the
main sources of error with this method. Calliper data collected
in conjunction with the density logs can be used to correct or
exclude data affected in this manner.
The main limitation of density logging is the use of a strong
radioactive source and the possibility that such a source could
become stuck in a drill hole, thus requiring some effort and
expense to recover the source. For the work at Century, a pro-
cedure was implemented to use the density tool only after one
or more successful runs with other nonradioactive tools. In this
way, the safety riskwas consideredmanageable, anda highpro-
portion of density logs from the available holes was achieved.
Rock strength.Measurement of the strength and fractur-
ing characteristics of both the ore and the host rock (includ-
ing the overlying shale, sandstone, and carbonate) is essential
for mine planning at Century and has implications for pit de-
sign, blasting requirements, and milling of the ore. Initially, a
traditional approach to prediction and modeling of the rock-
strength variations within the deposit was taken, using stan-
dard procedures such as measurement of rock quality descrip-
tor (RQD) on all cores. These data then were compared with
test measurements of unconned compressive strength (UCS)
and other strength parameters on selected pieces of core.
To assist with this problem, sonic velocity data acquired from
a standard slimline sonic tool were investigated to determine
if such data could provide more uniform and extensive infor-
mation than reliance on RQD and limited core measurements.
FIG. 9. Comparison of geotechnical data derived from laboratory measurements and sonic logs for Century drill hole LH643.
This approach has been used widely in rock-strength analyses
in Australian coal-deposit evaluation for many years (Asten,
1983; McNally, 1990; Davies and McManus, 1990).
A study (Duplancic, 1995) was initiated as part of the
AMIRA P436 research project to address this requirement.
The focus of this specic study was to determine the relation-
ship between rock strength (measured as UCS) and sonic ve-
locity, measured both on core samples in the laboratory and in
situ in the borehole.
Figure 9 shows a portion of a sonic velocity log from
Century, with the laboratory velocity (P-wave measured at
1 MHz) and UCS measurements superimposed. The velocity
data show a reasonable correlation, but the correlation with
UCS is poor. The analysis found that the relationship between
velocity and rock strength is a function of the lithology and
the rock porosity, more than the intrinsic variation in strength.
Figure 10 demonstrates this relationship schematically, sug-
gesting that it would not be feasible to predict UCS from
the borehole velocity data without rst accounting for the
porosity and lithologic variation.
The study also noted that sample quality, laboratory mea-
surement problems, andinsufcient samples togaugestatistical
reliability couldaffect these results adversely. However, it must
be concluded fromthis work that at best, the prediction of UCS
from velocity can be regarded as indicative only.
Structural prediction from geophysical logs.One of the
earliest problems recognized at Century was the presence of
1956 Mutton
small-scale faulting within the deposit (Waltho et al., 1993).
Thesefaults aremostlysteeplydippingandtrendinseveral pre-
ferred directions. Because surface exposures of these features
are masked by the irregular structure of the Cambrian lime-
stone cover, drill-hole intersections provided the only indica-
tion of the presence of these faults initially. However, because
most of the drill holes are vertical, only limited information
about the structures is available from drilling.
Some thought was put into methods that may assist with
the prediction of the presence and geometry of such faults be-
tween drill holes. Seismic and electromagnetic imaging were
postulated and are discussed below. Because of lack of con-
dence and uncertainty of the availability of these techniques at
the time, an alternate method was sought that used individual
boreholes and made predictions from the information in the
borehole.
Measurement of the core orientation and the dip of the
strata in the core is the traditional approach to interpreting
structure between drill holes in stratiform deposits. However,
the problem at Century was that because most of the holes
were vertical, the reliability of the available core-orientation
data was questionable. To address this issue, tests using a com-
bined dipmeter/hole deviation tool developed by BPB Instru-
ments were carried out at Century. The results demonstrated
that good-quality oriented dip information could be obtained
on the more laminated units (shales and siltstones), includ-
ing the ore sequence. The amount and quality of the data ob-
tainedenabledthe estimationof structural offsets betweendrill
holes.
Blast-hole logging.The recognition that lithologic contacts
could be delineated readily fromborehole geophysical logs led
to the possibility that the blast holes planned for the bench
mining of the deposit could be logged geophysically to pro-
vide more accurate information about ore-waste contacts than
conventional geologic logging from blast cuttings. Such infor-
mation could provide a substantial benet to the mining eco-
nomics by optimizing the blast design and the mining of ore
and waste. Most of the benet would come in reduced dilution
of ore for grade-control purposes.
Figure 11 shows the result of logging a test blast hole within
a small pit excavated to obtain a bulk sample of the ore (Fig-
FIG. 10. Approximate relationships between sonic velocity,
rock strength (UCS), and porosity, Century deposit (after
Duplancic, 1995).
ure 1). Only natural gamma and magnetic susceptibility logs
were obtained for this test. The prediction of the ore unit con-
tacts was based on a comparison with curve shapes, using a
template overlay, from nearby reference core holes. The main
ore units, as mapped from the pit face and extrapolated to the
blast hole, are shown also. The comparison withthe interpreted
lithologies suggests that the ore contacts can be predicted to
a vertical precision of about 10 cm. This result was conrmed
fromthesubsequent geophysical loggingof about 50blast holes
drilled in the trial pit.
The possibility of measuring ore grade in blast holes has
been considered also and some testing has been conducted.
The key to grade estimation is the measurement of density, but
the use of a highly radioactive source in an active mining area
was discouraged strongly. A test program was conducted by
the Commonwealth Scientic and Industrial Research Organ-
isation Australia (CSIRO) to try a low-activity spectrometric
probe developed by the CSIRO (Charbucinski et al., 1997)
to measure zinc grade directly in blast holes. Such a probe
would be an alternative to laboratory chemical analysis for
grade control. Information from such a probe would be avail-
able in a much shorter time frame than chemical analyses, and
potentially should provide more accurate information on ver-
tical grade variations than would be possible from analysis of
samples taken from drill cuttings.
Initial results of such tests were promising. The tests suggest
that such a tool could achieve better than 2.5% Zn deter-
mination, with substantial opportunity for improvement (to a
calculated limit of about 1.5% Zn). Ore boundaries can be
located to a vertical resolution of 10 cm. Although the preci-
sion of this grade prediction is not as good as with a chemical
assay, it is unlikely that the result obtained from a bulk assay
of drill cuttings will reect the actual grade of the material to
any better accuracy than indicated by the probe. Further work
on the assessment and development of this technology is rec-
ommended.
Borehole surveysinterborehole imaging
Because the Century deposit is faulted locally, the delin-
eation of faults within the deposit and the impact these would
have on mine design and extraction of the ore have been sub-
ject to much investigation. To determine possible options to
address this problem, some analysis of the potential applica-
tion of seismic tomography, radio frequency electromagnetic
(RFEM) imaging, and borehole radar was undertaken to de-
termine if any of these methods can dene the structure of the
ore zone between drill holes at Century.
Seismic tomography.Although surface seismic has little
application at Century because of the rugged limestone to-
pography and weathering (Thomas et al., 1992), analysis of
some test data suggested that reectors are associated with
lithologies within the ore zone. It was postulated therefore that
borehole seismic tomography could map these units and pro-
vide a basis for structural interpretation between drill holes.
However, seismic tomography was not undertaken at Century,
mainly because of the limited support to service such work in
Australia.
Radio imaging.As an alternative to seismic, RFEM imag-
ing was assessed by CRA as having potential application at
several of its operations. These include delineation of the struc-
ture, including offsets caused by faulting, of coal seams or ore
units between drill holes or mine-development access ways,
detection and location of geologic or operational hazard zones
between drill holes, detection of unknown ore not intersected
by drill holes, assessment of quality of coal or ore between drill
holes, and detection and location of large voids (cavities, old
workings, etc.) between drill holes.
Unlike seismic tomography, several potentially suitable EM
tomography systems are available, such as the RIM system
(Stolarczyk, 1992), the Russian FARA system, the South
African RT system, and the Chinese JW-4 system (Fullagar
et al., 1996b). The RIM system has advantages in Australian
conditions because of its lower frequency capability (down to
FIG. 11. Lithologic prediction fromgeophysical log in blast hole, based on curve matching fromnearby reference
hole, Century bulk sample pit.
12.5 kHz). It was also the only commercially available system
in Australia when the work at Century was being considered.
A trial of the RIM system therefore was undertaken at
Century in 1992 to determine its suitability in delineating the
structure of the ore zone between drill holes. Atest was carried
out in the vicinity of an exploration shaft which had been sunk
into the orebody to obtain a bulk sample of ore and to provide
access for geotechnical investigations andmapping of structure
from drives within the orebody (Figure 12a, b). This mapping
provided direct evidence of a previously unknown fault with
throwabout 1015 madjacent to holes in which an RIMsurvey
was carried out.
Initial processed results from this survey did not provide
much encouragement that the method could offer any infor-
mation on the structure between holes to the resolution re-
quired for mine development. As a result, no further surveys
1958 Mutton
FIG. 12. Comparisonof mappedgeology inundergrounddrive andradio-imaging method(RIM) survey results, Century exploration
shaft area: (a) location plan; (b) geologic section through drive; (c) reprocessed RIMtomogram(frequency 520 kHz) fromdrill-hole
section oblique to drive.
have been undertaken. However, alternative data-processing
options were investigated subsequently, resulting in the im-
age shown in Figure 12c. This result was obtained by using
exactly the same data as the initial processing, but in this case
the tomographic inversion used software developed by VIRG-
Rudgeozika for its FARA RFEM system. The main differ-
ence with this software is that it uses both the amplitude and
phase information. The reprocessed image clearly indicates the
presence of the fault very close to its predicted position. It is
interesting tonote that nospecic a priori geologic information
available from the drill holes or the subsequent underground
development was used to constrain the resulting image.
The use of RIM or equivalent radio-imaging systems there-
fore could play an important part in evaluation of the structure
of an orebody such as Century in advance of inll drilling and
mine development. At Century, this has not happened so for,
probably because of unfavorable impressions gained by the
mine-evaluation staff from the initial processing of the data,
and the passage of time before better results were obtained.
However, it is estimated that if the radio-imaging technology
and processing had been sufciently developed, proved, and
accepted at the time of the Century discovery, the structural
evaluation of the deposit might have been achieved at a sig-
nicantly lower cost ($5 million to $10 million less) and in a
shorter time frame. The basis for this substantial cost benet is
in the reduction of the number and type of boreholes required
(i.e., noncore), but the main benet comes fromincreased con-
dence in the reserves and the subsequent mine design and
mining plan.
Borehole radar.A trial borehole radar survey was carried
out at Century as part of the AMIRA P436 research project.
The objectives of this work were to map faults within the Pro-
terozoic sediments and determine their geometry; and to de-
tect cavities, representing a future mining hazard, within the
limestone.
The use of borehole radar to map faults was a possible al-
ternate to radio imaging, on the basis that the higher frequen-
cies would give higher resolution of the structures and that
the radar could be carried out in a single-hole reection mode
requiring less logistical effort. The results of this work were dis-
appointing because the penetration distance of the radar signal
using a 60-MHz source frequency through the weakly conduc-
tive sediments was very small (<5m), effectively rendering the
technique of little value for this problem.
The detection of cavities in the limestone with borehole
radar was more promising. A reection range of about 25 m at
60-MHz frequency was obtained, and features that may be re-
lated to cavities were noted. No testing or proving of such fea-
tures was possible at that time. However, the results were con-
sidered sufciently encouraging to suggest that a radar method
(including surface radar) may have application to the location
of potentially hazardous voids during waste-removal activities
for open-pit development.
DISCUSSION
The use of geophysics in the evaluation of the Century de-
posit was an attempt to inuence, change, and improve the pro-
cess of resource denition in CRA and eventually integrate
such technology into routine mining practice. The work has
demonstrated some of the potential benets that could ow
from the routine use of geophysics during resource evaluation
or when integrated with mining. With regard to evaluation of a
potentially minable resource, areas in which successful use of
geophysics could have a signicant positive benet to a project
development include:
1) Locating the lateral and vertical extent of ore at a much
earlier stage of drill evaluation. The inuence this would
have on the time frame and accuracy of initial reserve
estimations could have a major impact on development
criteria.
2) Correlation of mineralized intercepts and prediction of
structural offsets between drill holes. The information
obtained on the structure and volume of ore should be of
great benet to mine planning and reserve estimation.
3) Substitution of cored with noncored holes. If grade varia-
tion is relatively uniform, the use of geophysically logged
noncored holes as a substitute for more expensive and
time-consuming cored holes could represent substantial
cost benets at an early stage of the project.
4) Direct measurement of physical parameters such as den-
sity or rock strength. Geophysical logging potentially
should provide the most consistent and reliable informa-
tion on the in situ rock properties.
5) Geotechnical evaluation of cavities, zones of weak-
ness, or incompetent rocks representing potential min-
ing hazards. Reliable remote detection of such hazards
has enormous implications for mine design and safety
management.
6) Evaluation of hydrologic conditions, and foundations for
mine infrastructure. High-resolution surface geophysics
integrated with information available from drilling and
water monitoring can provide potentially more denitive
solutions to some of these problems.
Knowledge obtained during the resource-denition phase
of the project also provides the evidence and condence for
successful use of geophysical technology in the mining process.
Areas in which application of geophysics has demonstrated
some potential for integration into mining operations include:
1) Ore-boundary delineation. Detailed information on ore-
waste contacts from geophysical logging of production
drill holes or high-resolution surface surveys can inu-
ence the optimization of the blast design or ore extrac-
tion, yielding a substantial benet to the mining eco-
nomics, largely because of the reduction in dilution of
ore.
2) Grade control. Where a relationship between grade and
one or more physical properties is present, direct grade
prediction in production drill holes is possible, yielding
substantial benets because of improved grade control,
more rapid availability of data, and reduced laboratory
costs.
3) Hazard detection. The remote detection of cavities, bro-
ken ground, or geologic impediments in advance of min-
ing can reduce personal and economic risk substantially.
Knowledge and understanding of contrasts in the physical
properties of the materials being mined are the fundamental
principles which determine if there is a role for geophysics in
1960 Mutton
the denition, evaluation, andmining of a mineral deposit. This
knowledge can be gained only from physical measurements
on core or geophysical logging of drill holes. Therefore, as a
prerequisite to any mine-evaluation program, it is necessary to
acquire a suite of information on a range of physical properties,
and then use either previous experience or indicative modeling
studies to determine what geophysical methods may achieve
the required goals of the resource denition.
CONCLUSIONS
Experience gained by CRA in the early 1990s during the
evaluation of the Century deposit and at other Australian
mining operations suggests that successful application of geo-
physics in mine evaluation or during mining is achievable in
certain circumstances. These circumstances are controlled by
the contrasts in the physical properties of the materials being
mined. To use geophysics successfully in mining, it is there-
fore necessary to acquire information on a range of physical
properties, and then use this information to determine what
geophysical methods may assist the evaluation process or min-
ing operation.
The overall economic benet of successful implementation
of geophysical technology into all phases of resource deni-
tion and mining at Rio Tintos operations is estimated at tens
of millions of dollars annually. This projected benet alone
should stimulate the need for investigating further technolog-
ical improvements and evaluation of available technologies at
existing operations.
Despite this positive assessment, some barriers inhibiting
the routine testing and implementing of such technology ex-
ist. These barriers relate largely to the appropriate integration
of the technology with the mining process, and acceptance by
mine operators of the economic value of such work, compared
with traditional approaches to resource denition. Some effort
by mine management will be required to address these issues
and support further evaluation of the technology before the
use of geophysics will be accepted as a routine component of
resource denition and mining.
Further anticipated technological developments, cost pres-
sures onproduction, andgradual acceptance by mine operators
eventually will ensure the use of geophysics in mining. The re-
ality may be that those mining operations that evaluate and
successfully exploit such developments will be the most ef-
cient and economically viable mines of the future.
ACKNOWLEDGMENTS
The assistance of the staff of CRA Exploration and Century
Zinc Limited with the work undertaken at Century from 1990
to 1996 is acknowledged gratefully. In particular, I acknowl-
edge the efforts of Barry Bourne and Theo Aravanis in ini-
tiating some of the innovative work presented here, and the
encouragement of John Main, Andrew Waltho, and the CZL
staff to undertake such work. I also thank the AMIRA P436
team for its input. This paper is published with the permission
of Rio Tinto Limited and Century Zinc Limited.
This paper was published 1997 in Proceedings of Exploration
97: Fourth Decennial International Conference on Mineral Ex-
ploration, edited by A. G. Gubins. Some updates have been
made to this version at the request of the Special Editor for
publication in GEOPHYSICS.
REFERENCES
Asten, M. W., 1983, Borehole log analysis using an interactive com-
puter: Bull. Aust. Soc. Expl. Geophys., 14, 310.
Broadbent, G. C., 1996, The Century discoveryIs exploration ever
complete?: in Mauk, J. L., and St. George, J. D., Eds., Proceedings
Pacrim 95 Congress: Aust. Inst. Min. Metall. Publ. 9/95, 8186.
Charbucinski, J., Borsaru, M., and Gladwin, M., 1997, Ultra-low ra-
diation intensity spectrometric probe for ore body delineation and
grade control of Pb-Zn ore: Gubins, A. G., Ed., Proceedings of Ex-
ploration97: FourthDecennial International ConferenceonMineral
Exploration, 631638.
Davies, A. L., and McManus, D. A., 1990, Geotechnical applications of
downhole sonic and neutron logging for surface coal mining: Expl.
Geophys., 21, 7382.
Duplancic, P., 1995, A comparison of mechanical properties of rock
with in situ velocity measurements at the Century deposit: B.E. the-
sis, Univ. Queensland.
Fullagar, P. K., and Fallon, G. N., 1997, Geophysics in metallifer-
ous mines for orebody delineation and rock mass characterization,
Gubins, A. G., Ed., Proceedings of Exploration97: FourthDecennial
International Conference on Mineral Exploration, 573584.
Fullagar, P. K., Fallon, G. N., Hatherly, P. J., and Emerson, D. W., 1996a,
Implementation of geophysics at metalliferous minesFinal report
AMIRA Project P436: Cooperative Research Centre for Mining
Tech. and Equipment Report MM1-96/11.
Fullagar, P. K., Zhang, P., Wu, Y., and Bertrand, M-J., 1996b, Applica-
tion of radio frequency tomography to delineation of nickel sulde
deposits in the Sudbury Basin: 66th Ann. Internat. Mtg., Soc. Expl.
Mayers, P. J., and Bourne, B. T., 1994, Geological and geophysical study
of the Lawn Hill limestone annulus: Century Zinc Ltd. Internal Re-
port 94/002.
McNally, G. H., 1990, The prediction of geotechnical rock properties
from sonic and neutron logs: Expl. Geophys., 21, 6572.
Oldenburg, D. W., Li, Y., and Farquharson, C. G., 1997, Geophysical in-
version: Fundamentals andapplications inmineral explorationprob-
lems: in Gubins, A. G., Ed., Proceedings of Exploration 97: Fourth
Decennial International Conference on Mineral Exploration, 545
548.
Stolarczyk, L. G., 1992, Denition imaging of an orebody with the
Radio Imaging Method (RIM): IEEE Trans. on Industry Applica-
tions 28, 11411147.
Thomas, G., Stolz, E. M., and Mutton, A. J., 1992, Geophysics of
the Century zinc-lead-silver deposit, northwest Queensland: Expl.
Geophys., 23, 361366.
Waltho, A. E., Allnutt, S. L., and Radojkovic, A. M., 1993, Geology
of the Century zinc deposit, northwest Queensland: Proc. Internat.
SymposiumWorld Zinc 93, 111130.
Exploration geophysics at the Pyh asalmi mine and grade control
work of the Outokumpu Group
Aimo Hattula
and Timo Rekola
ABSTRACT
Thepower of geophysics is oftenrealizedwhilesurvey-
ing barren exploration holes. Integrated interpretation
of borehole electromagnetic (EM) andlithogeochemical
data led to the discovery of a new volcanogenic massive
sulde (VMS) ore deposit at 500 m depth in the Pyh a-
salmi area, which belongs to the Main Sulde ore belt in
Finland. In the deep exploration program, wide-band
multifrequency EM ground surveys were successfully
used to detect both new ore lenses and geological struc-
tures. Mise-a-la-masse (MAM) borehole andgroundsur-
veys as well as borehole EM surveys were effectively
used to correlate intersections between drill holes and
to locate new orebodies. The latest modeling of MAM
data resulted in an exploration target at 700 m depth.
The use of geophysics for exploration has been ex-
tended to mine production at Outokumpu. Geophysical
logging detects ore-waste boundaries, reduces expen-
sive core drilling, and obtains physical property infor-
mation quickly on ore intersections. Depending on ore
type, geophysical borehole logging can also be applied
to classify mineralization, interpret lithology, and some-
times to transformphysical responses to metal grades in
ore. At the Pyh asalmi zinc-copper-sulfur mine, density
logging in percussion boreholes is used to locate mine-
able ore boundaries andtoclassify drillhole intersections
as massive or semimassive sulde ore types. Pyrrhotite-
bearing zones are separated from other suldes by in-
ductive conductivity logs. The use of geophysical logging
for grade estimation and control has been most effective
in the nickel mines at Enonkoski, Finland, and Namew
Lake, Canada (using conductivity logs), and in the Kemi
chromium mine, Finland (using gamma-gamma density
logs).
INTRODUCTION
Economically signicant sulde ore deposits exist on the
Main Sulde ore belt of Finland (Figure 1). The belt is
40150 km wide and at least 400 km long. It occupies an area
between a gravity lowrelated to a Proterozoic complex and an
Archaean basement complex. The Proterozoic portions of the
belt host most of the major sulde ore deposits and are charac-
terized by schists and granitoids. Pyrrhotite-bearing graphite
schists, or black schists, abound in the belt and are responsi-
ble for many magnetic and electromagnetic (EM) geophysical
and geochemical anomalies. The Archaean part consists of ex-
tensive greenstone belts and granite gneiss. Some 9597% of
the bedrock is covered by 69-m-thick Quaternary till, sand,
and clay. The deposits of the Main Sulde ore belt contain
more than 90% of the known mineral resources in Finland.
This conrms that conditions for sulde ore-forming processes
must have been favorable (Kahma, 1973).
Theimportanceof geophysical techniques insite-specic and
deposit-scale exploration is demonstrated by investigations on
Manuscript received by the Editor May 10, 1999; revised manuscript received June 9, 2000.
Outokumpu Mining Oy, P.O. Box 15, Kummunkatu 34, FIN-83501 Outokumpu, Finland. E-mail: aimo.hattula@outokumpu.com; timo.rekola@
outokumpu.com.
sulde ore deposits of the volcanogenic massive sulde (VMS)
type in the Pyh asalmi area (Figure 1). Applications are pre-
sented to illustrate the successful extension of geophysics into
mine production using borehole logging for grade control at
mines operated by the Outokumpu Group.
SITE-SPECIFIC EXPLORATION
Geology of Pyh asalmiMullikkor ame area
Bimodal volcanics, metapelites, tonalite gneisses, pyroxene
granites, gabbros, and granites characterize the geology of the
main Pyh asalmi deposit area (Figure 2). Altered volcanics
(cordierite, anthophyllite, sericite, chlorite, phlogopite, etc.)
are ubiquitous in the area. All of the known VMS-type zinc-
copper-sulfur-ore deposits and mineralized zones are located
in altered rocks (Puustjarvi, 1992).
The massive pyrite-rich zinc-copper-sulfur ore deposit at
Pyh asalmi is 650 m long and up to 75 m wide, becoming nar-
rower to the south and north and dipping almost vertically
1961
1962 Hattula and Rekola
down to a depth of 1450 m. Contacts between ore and waste
aresharp. Acidpyroclastic rocks andquartz porphyries host the
ore (Maki, 1986). The ore deposit is surroundedby a large alter-
ation zone. Sericitized quartzites occur closest to the ore, while
cordierite-mica schists and cordierite-anthophyllite rocks are
more distal. The isoclinally folded alteration zone is over 6 km
long and 100500 m wide, and it reaches depths in excess of
1450 m. A new deep extension of the ore deposit was discov-
ered by down-dip drilling at a depth of 1050 to 1450 m from
1996 to1998. Mine productiontodate since the mine start-upin
1962 has been 33.4 million tons (Mt). Current ore reserves are
19.7 (Mt), containing 1.16% copper, 2.25% zinc, 37.7% sulfur,
and 0.4 g/ton gold.
The Mullikkor ame deposit, a satellite mine of Pyh asalmi, is
located about 7 km to the northeast. The Mullikkor ame For-
mation is up to 1000 m wide and almost 3000 m long (Fig-
ure 3). The stratigraphy of the area from bottom is as follows:
mica gneiss (not encounteredwithinthe Mullikkor ame area)
felsic volcanicssulde ore depositsmac volcanics. Mac
volcanics with variable amounts of pyrrhotite dominate the
western part of the formation. To the east, zinc-copper-lead-
sulfur ore of the VMS type is commonly hosted by altered
felsic volcanics within a bimodal sequence (Puustjarvi, 1992).
Abundant sericitic and chloritic alteration is typical of the min-
eralized horizons. As part of the exploration effort, 148 holes
totaling 26.2 km were drilled into the Mullikkor ame Forma-
tion from 1987 to 1995. Subsequently, 0.27 Mt of ore contain-
ing 5.8%zinc, 0.2%copper, 0.3%lead, 22%sulfur, 22g/t silver,
and 1.3 g/ton gold were mined from surface in 1991 and 1992.
Another orebody, located at a depth of 400 to 600 m, is being
FIG. 1. The Main Sulde ore belt and ore deposits in Finland.
mined; it contains 0.7 Mt of 8.3%zinc, 0.4%copper, 1.3%lead,
17.8% sulfur, and 1.2 g/ton gold. Exploration in the surround-
ing areas has been initiated.
The Mullikkor ame deposit dips about 60
east at surface
but becomes more at lying with depth. The plunge of the ore
lenses is not well established. At greater depths it is almost
horizontal, trending between north and east as inferred from
mise-a-la-masse (MAM) data. The volcanics, surrounded by
granites to the east and west, are cut by uralite porphyrite
dikes, which are controlled by factors not yet fully under-
stood. A conductive mylonite layer almost 100 m thick at
FIG. 2. Geological map of the Pyh asalmi deposit area.
Mining Geophysics at Outokumpu Group 1963
the eastern contact between host rocks and granite hampers
the use of electrical methods because of its relatively low
resistivity.
Geophysical surveys in the Pyh asalmi
and Mullikkor ame deposits
The high-density (4.0 g/cm
3
) and conductive (10100 S/m)
Pyh asalmi ore deposit was delineated relatively easily using
horizontal loop electromagnetics (HLEM) and gravity meth-
ods (Figure 4). The deposit contains 24% pyrrhotite, which
greatly increases its conductivity compared with the pyrite.
Gravity data were interpreted using a prism model to simu-
late the orebody. In Figure 5 the cross-section models show
howthe orebody is near vertical in upper levels and inclined at
FIG. 3. Geological map of the Mullikkor ame area.
deeper levels where a strong D4 shear zone deforms the whole
orebody into a vertical propeller shape.
In the Mullikkor ame area, an airborne EM and magnetic
survey (at an altitude of 40 m above terrain) was carried
out in 1978. The rst ground follow-up was conducted using
HLEM, magnetic, and gravity surveys. Resulting anomalies
were checked with percussion drill sampling, and a shallow
orebody was discovered (Hattula and Rekola, 1997). The de-
posit was indicated by a weak 4% HLEM anomaly in both
in-phase and out-of-phase components (Figure 6). Stronger
EM responses result largely from the conductive pyrrhotite
mineralization (210%sulfur) in the west. The poorly conduc-
tive zinc-copper-sulfur orebody (0.10.5% sulfur) in the east
(body A) was weakly indicated only by the higher frequencies
of the out-of-phase component (Figure 7).
IP measurements outline the shallow orebody rather well
(Figure 8). The felsic volcanics, with variable amounts of pyrite,
are indicated by apparent resistivities in the range of 600
800 ohm-m compared with host rock values of unmineralized
mac volcanics, which usually exceed 1000 ohm-m.
The Mullikkor ame mac volcanics (2.9 g/cm
3
) are the source
of a 1.5-mGal gravity anomaly. These volcanics are interpreted
to continue eastward beneath granite (Figure 9). Magnetic and
IP surveys suggest the same interpretation. Although ore has
the highest density (3.6 g/cm
3
) among rocks in the area, gravity
surveys do not necessarily possess practical value in locating
orebodies because the targets are relatively small compared to
the whole volcanic formation (Rekola, 1992).
MAM surveys were also carried out during the exploration
program. Measurements on near-surface orebodies delineated
the extent of suldes fairly well (Figure 10); however, it was
difcult to distinguish between ore and noneconomic pyrite
conductors because of similar resistivities. MAM was one of
the main methods used to correlate drillhole ore intersections
and guide drilling during a deep exploration program. First,
this led to the intersection of a 45-m-thick pyrite-predominant
alteration zone containing 0.6% zinc and 0.12% lead in hole
MU-99 (Figure 11, top); later, an ore lens (with a surface pro-
jection as shown in Figure 11, top) was intersected by fur-
ther drilling, including extending hole 99. When the deep-
seated orebody (B) was discovered, a transmitter (current)
electrodewas placedinanintersectioninholeMU-108. Surface
and downhole MAM results indicated a potential extension
both north and south (Figure 11). Interpretation by VIRG-
Rudgeozika, a Russian geophysical institute, using analog-
digital modeling fromthe earlier and latest holes showed three
conductive blocks at 700900 m depth, under the known deep
orebody. Correlation with drill results on one prole demon-
strates good conrmation of this interpretation (Figure 11,
bottom).
In the deep exploration program at Mullikkor ame, wide-
band multifrequency EM surveys using the Genex 400S sys-
tem(Aittoniemi et al., 1987) were carried out with a broadside
coil conguration across the mineralized zone. Coil separation
varied between 200 and 800 m. Resistivity results interpreted
from the 800-m coil separation survey data are presented sta-
tion by station in Figure 12. A distinct indication of a deep
orebody (B) at a depth of 500 m was detected. A dipping low-
resistivity zone fromwest to east is related to pyrrhotites in ba-
sic volcanics; existence of this sulde conductor was indicated
by DHEM measurements in hole MU-120 (Figure 12). The
Genex 400S survey also suggested a low-resistivity structure
below a depth of 800 m, but this indication has not yet been
tested by drilling. Results of the Genex 400S measurements
contributed to the decision to drill into the deep extension of
the surface orebody.
Drillholes MU-98 and MU-99 intersected an alteration zone
at a depth of 200 to 300 m. Thereafter, all holes were logged
with a Geonics EM-37 borehole EM system. These DHEM
contributed to a decision to deepen an existing hole MU-108
from 450 m (Figure 13), resulting in the discovery of the rst
lens of the deep orebody B which in turn led to additional sys-
tematic drilling. Further DHEMand MAMsurveys around the
deep orebody followed; the DHEM measurements indicated
many additional ore lenses in the deeper parts of the volcanics
formation.
MINE GRADE CONTROL
Excessive dilution often results as waste is mined and pro-
cessed with ore (Elbrond, 1994). Geophysical borehole logging
is used in the mines operated by the Outokumpu Group for
FIG. 4. The Pyh asalmi deposit and geophysical results.
FIG. 5. Measured and calculated gravity eld over the
Pyh asalmi deposit.
quickly detecting ore boundaries and quality to reduce rock
dilution and subsequent grade reduction.
Outokumpu developed a borehole logging system, OMS-
logg, to directly measure differences in physical characteristics
of mineralization fromborehole walls (Figure 14). Often when
a mineral (or minerals) has a dominating property, the bound-
aries of mineralization can be determined. Grade estimation
can be done using an empirical calibration procedure. Probes
FIG. 6. HLEM out-of-phase results in the Mullikkor ame area.
FIG. 7. HLEM results with different frequencies in the
Mullikkor ame area.
are customizedaccording toindividual orebody characteristics.
The objective is to use percussion drilling or production holes
for in-hole geophysics sothat slowandexpensive diamondcore
drilling is no longer needed in detailed stope planning or grade
control. Depending on ore type, OMS-logg applications are
divided into the following categories: determination of litho-
logical boundaries, denition of ore zones, estimation of dry
bulk densities, and grade estimation of ore.
For the past 13 years, borehole logging has been in routine
use at mines of the Outokumpu Group worldwide. Density
conductivitysusceptibility logging is currently used at the
Pyh asalmi zinc-copper-sulfur and Mullikkor ame zinc-copper-
lead-sulfur mines in Finland and in the Forrestania nickel mine
in Australia. The Tara zinc-lead mine in Ireland and the Kemi
chromium mine in Finland utilize gamma-gamma density log-
ging. All of these are underground mines except for Kemi,
which is an open pit. The Enonkoski and Telkkala nickel mines
in Finland, the Namew Lake nickel mine in Canada, and the
Viscaria copper mine in Sweden used inductive conductivity
logging. In all of these mines, logging was mainly carried out in
percussion and/or production boreholes.
The latest model of the OMS-logg system provides a mode
in which measurements can be made without the necessity of
a logging cable. Probes are connected to drill rods and moved
in the holes by a drill machine. This permits logging long holes
in any direction.
FIG. 8. IP data in the Mullikkor ame area.
The boundaries of the Pyh asalmi pyrite-rich massive sul-
de ore are dened using gamma-gamma density logging
(Figure 15). Parts of the ore contain pyrrhotite that has almost
the same density as typical pyrite ore. For separating barren
pyrrhotite mineralizationfromeconomic ore tobe stoped, con-
ductivity logging combined with gamma-gamma density log-
ging is used. The latter methodis alsousedinthe Mullikkor ame
zinc-copper-lead-sulfur mine. High-density blocks accurately
correlate with economic ore boundaries (Figure 16).
The OMS-logg conductivity probe was designed to observe
responses from sulde nickel ores. A 10-cm coil length allows
the detectionof sharpcontacts andnarrowveins as well as mea-
sures close to the bottom of the holes. The Enonkoski nickel
mine is an example in which nickel grade was successfully es-
timated using conductivity logging calibrated in massive ore
(Hattula, 1992). In addition, reliable lithological information
was obtained on the distribution of disseminated mineraliza-
tion and waste. Figure 17 shows proles of assay and nickel-
calibrated logging data in (a) a diamond drill hole and (b) a
percussion borehole.
Lithology in the Kemi chromium mine (Figure 18) is classi-
ed using the OMS-logg density logging for product control.
Density is correlated with Cr
2
O
3
grades to decide which parts
FIG. 9. Gravity data in the Mullikkor ame area.
of the mined ore will be separated for upgraded lump ore.
Logging is carried out in production holes in the open pit and
in holes drilled for a new underground mine development. A
strict calibration procedure between the gamma-gamma log-
ging and density for various ore types is fundamental. It is
essential to classify ore types but also to detect boundaries for
internal or external waste blocks.
CONCLUSIONS
The Pyh asalmi and Mullikkor ame mine examples showhow
many anddifferent petrophysical elements are relatedtoVMS-
type deposits. The most complicated are EMand electrical pa-
rameters because both waste and ore suldes cause geophysi-
cal responses. Wide-band multifrequency EM surveys provide
means of detecting weakly conductive sulde ore bodies and
mineralization as well as highly conductive ore zones. Detailed
interpretation of MAM data assists in correlating drillhole in-
tersections but is often useful also in indicating/detecting off-
hole lenses. The use of different geophysical methods and their
FIG. 10. MAM data for the Mullikkor ame shallow orebody.
geologically focused modeling provides a tool to distinguish
waste and ore sulde lenses.
Borehole geophysics plays an important role in the nal tar-
geting and delineation of ore deposits. In many cases, geophys-
ical in-hole survey data may result in discoveries before ac-
tually intersecting ore. Borehole geophysics can also greatly
benet mine grade control in various types of ore deposits.
Depending on the physical properties of an orebody, borehole
logging provides fast determination of both ore and/or waste
boundaries, although geophysical logging seldom provides re-
sults sufciently accurate for grade estimation. Fallon et al.
(2000) present very similar ndings in their investigation of
interpretation of geophysical borehole logs.
FIG. 11. MAM data for the Mullikkor ame deep orebody.
FIG. 12. Resistivity cross-section of the wide-band, multifre-
quency EM survey using the Genex 400S system over the
Mullikkor ame deep orebody structure.
ACKNOWLEDGMENTS
The authors thank Outokumpu Mining Oy for permission
to publish this paper. Kaarlo Makela and Heikki Puustjarvi,
project managers for Outokumpu Mining Oy Exploration,
and J. P. Matthews, Manager Geophysical Services for Phelps
Dodge Exploration Corp., are thanked for critically reviewing
the manuscript.
REFERENCES
Aittoniemi, K., Rajala, J., and Sarvas, J., 1987, Interactive inversion
algorithm and apparent resistivity versus depth (ARD) plot in mul-
tifrequency depth soundings: Acta Polytech. Scandinavica, Applied
Physics Series No. 157.
Elbrond, J., 1994, Economic effects of ore losses and rock dilution:
CIM Bull., 87, 131134.
FIG. 13. DHEM data from boreholes MU-98 and MU-108 in
the Mullikkor ame deposit.
Fallon, G. N., Fullagar, P. K., and Zhou, B., 2000, Towards grade esti-
mation via automated interpretation of geophysical borehole logs:
Expl. Geophys., 31, 236242.
Hattula, A., 1992, Borehole logging systemat Outokumpu Enonkoski
Ni mine: Analys i borrh al, Swedish Mineral Industry Research Or-
ganisation report MITU 1992:14 T.
Hattula, A., and Rekola T., 1997, The power and role of geophysics
applied to regional and site-specic mineral exploration and mine
grade control in Outokumpu Base Metals Oy: 4th Decennial Inter-
nat. Conf. on Min. Expl. Proceedings, 617630.
Kahma, K., 1973, The main metallogenic features of Finland: Geol.
Surv. Finland Bull. 265.
Maki, T., 1986, Lithogeochemistry of the Pyh asalmi zinc-copper-pyrite
FIG. 14. Schematic of the OMS-logg borehole logging system.
FIG. 15. Conductivity logging combined with gamma-gamma
logging for separating pyrrhotite zones from ore to be stoped
at the Pyh asalmi mine.
deposit, Finland: Prospecting in areas of glaciated terrain 1986: In-
stitution of Mining and Metallurgy, 6982.
Puustjarvi, H., 1992, ExplorationinFinlandSearchfor VMS-deposits
in the Pyh asalmi district, central Finland: Irish Assn. Econ. Geol.
course notes.
Rekola, T., 1992, Recent geophysical surveys for massive suldes in
the Pyh asalmi area, central Finland: 54th Meeting and Technical
Exhibition, Eur. Assn. Eng. Geosci. Abstracts, 316317.
FIG. 16. Correlation of high-density blocks with boundaries of
ecomomic ore-grademineralizationintheMullikkor amemine.
FIG. 17. Comparison of (a) core sample assays froma diamond
drill hole and (b) sludge sample assays froma percussion bore-
hole, versus conductivity calibrated to nickel content at the
Enonkoski mine.
FIG. 18. Density logging prole to dene ore boundaries and
classify ore for product control at the Kemi chromium mine.
Ultralow-radiation-intensity spectrometric probe for orebody
delineation and grade control of zinc-lead ore
Jacek Charbucinski
, Mihai Borsaru
, and Michael Gladwin
ABSTRACT
A fully spectrometric gammagamma probe using an
ultralow-activity gamma-ray source (i.e., <2 MBq) has
been developed. The prototype low-radiation-intensity
probe was tested at a zinc-lead deposit in Queensland,
Australia. The probe showed an excellent capability for
orebody delineation. Lithological proles derived from
logging data showed sharp anomalies both in selected
spectral regions and in spectral ratios during probe tran-
sitionfromorebody tobarrenrock, or vice versa. Incom-
parison to standard geochemical analysis, the instrument
demonstrated good potential for quantitative determi-
nation of lead and zinc content in ore with rms error of
0.3%and2.4%, respectively. Delineationof nonlitholog-
ical boundaries, through an application of a cutoff-grade
algorithm, has been demonstrated as a practical stand-
alone mine control tool.
INTRODUCTION
The mineral exploration and mining industry needs safe,
portable, and maneuverable borehole logging equipment for
deposit delineation and grade control, especially during the -
nal stages of exploration, mine development, and ore produc-
tion. Geophysical logging has become an integral part of mod-
ern exploration and mine development activities. Among the
many different borehole logging techniques, nuclear borehole
logging is practically the only technique that provides on-line,
quantitative, in-situ grade control.
Most commercial logging companies use a nonspectrometric
gammagamma borehole logging-technique (gammagamma
density log). This technique is commonly applied for coal-seam
evaluation or orebody delineation. When ore-grade determi-
nations are required, it is assumed that ore grade and ore bulk
density are correlated. While ore grade and density may be
correlated quite well in some deposits, that correlation is often
too weak or the polymetallic nature prevents a reliable deter-
CSIRO Exploration and Mining, P.O. Box 883, Kenmore, Queensland 4069, Australia. E-mail: j.charbucinski@dem.csiro.au; m.borsaru@dem.
csiro.au; m.gladwin@dem.csiro.au.
mination of ore grade or chemical concentrations of key im-
purities. Spectrometric techniques have proven useful for both
identifying elemental constituents and quantitatively estimat-
ing ore. Spectrometric methods can differentially detect and
record different probe-response events, e.g., those from dif-
ferent gamma-ray energies or events having different arrival
times. This capability to use gamma-ray energies is important
when identifying and estimating some elemental constituents
of the ore, e.g., the prompt gamma neutron activation analysis
(PGNAA) technique.
Thecommercial gammagammadensityloggingtools usera-
dioactivesources withhighactivities (around20006000MBq).
These levels of radioactivity present difculties for man-
aging the safety and security of mine personnel and the
environmental impact if a probe is irretrievably stuck in a
borehole. CSIRO-developed Sirolog spectrometric borehole
logging probes have achieved signicant use and recognition
in the mining industry because of their superior quantita-
tive performance (Eisler et al., 1990; Charbucinski, 1993a;
Borsaru et al., 1994). However, many mining companies still
consider that the risk associated with the irretrievable loss
of even a standard Sirolog probe is too high (even though
its source activity of 40 MBq is about 1/100th of a con-
ventional commercial probe). These companies would pre-
fer to reduce the source intensity by another order of mag-
nitude. The states of Victoria, New South Wales, and Queens-
land in Australia dene 3.7 MBq as the minimum activity of
a gamma-ray source requiring a license for possession, use,
and transport of radioactive substances. Therefore, require-
ments for operations using logging equipment with a radioac-
tive source of activity lower than 3.7 MBq would be much
simpler and would minim the risk associated with radioactive
sources.
The Sirolog instrumentation goes one step further toward
introducing to the mining industry a spectrometric borehole
logging systembased on applying an ultralow-activity gamma-
ray source (<2 MBq). Tests have been reported for coal-
seam delineation and on-line quantitative determination of
ash in coal (Charbucinski, 1993b; Borsaru and Ceravolo 1994;
1970
Low-Radiation-Intensity Probe 1971
Charbucinski et al., 1996) and for iron orebody delineation
(Borsaru et al., 1995).
The objective of our work was to optimize probe design and
to demonstrate the delineation and quantitative capabilities
of the low-radiation-intensity spectrometric probe for in-situ
analysis of lead and zinc.
TECHNIQUE OVERVIEW
Two different low-activity probe congurations have been
developed for the coal mining industry. In the Low Activity
tool (Borsaru et al., 1994), a single
137
Cs gamma-ray source
of 1.8 MBq activity was positioned axially from the detector
behind a very short (30 mm) shield of tungsten and iron. In the
second conguration, the Zero Probe (Charbucinski, 1993b),
three
137
Cs microsources (each 0.37 MBq) were placed radially
around the detector at virtually zero distance from the crystal
(measured along the borehole axis). This ultimately short
distance provided the probe with the best possible vertical
resolution.
For mineral applications these ultralow activity probes re-
quired redesign of probe congurations and development of
new calibration and inversion algorithms. This was necessary
because the probes developed for the black coal industry were
designed to perform optimally in low-density, low-equivalent-
atomic-number environments. The zinc and lead ores have sig-
nicantly higher equivalent atomic numbers and much higher
density than the host lithology. The higher densities required
new optimization of sourcereceiver distances. Planned larger
diameter holes (>150 mm) allowed further optimization of
scintillator crystal volumes. However, the optimized geome-
try still featured a very short sourcereceiver distance, so the
probe operated in the preinversion zone of the universal cali-
bration curve (Artsybashev, 1972). In that zone, the count rate
(recorded in the density window) of gamma quanta scattered
in a high-density medium (ore) is higher than the count rate
recorded in a lower density scatterer (barren rock).
We needtoemphasize that the low-radiation-intensity probe
is the low-activity version of the Sirolog spectrometric gamma
gamma probe. As such, the probe is unable to measure the
concentration of zinc directly. The gammagamma probe re-
sponse is relatedtothe overall contributions givenby the major
components with a high atomic number present in the ore, e.g.,
lead, zinc, iron, manganese. Figure 1 shows spectra recorded
FIG. 1. Spectra recorded with a 76-mm diameter 76-mm de-
tector in large (200-liter) ore samples of various zinc, lead, and
iron content.
in ve large geophysical models. Four models contained zinc-
lead ore of various concentrations, and the fth model (W2)
contained waste. The prominent features of the spectra are the
180
backscatter peak around 210 keV and leads characteris-

tic X-ray peak around 87 keV. Not shown in the gure is the
photopeak of primary radiation from the probes
137
Cs source
(667 keV), which was utilized to stabilize the logging systems
hardware gain. The probe was designed to use single scattered
gamma rays (180
) to provide information about the bulk den-

sity of the logged formation, while multiscattered quanta of
lower energies convey information on the average chemical
composition of the ore. The lead X-ray peak helps determine
percent lead content (%Pb).
The main way the major ore components contribute to the
value of a spectral ratio used to correlate with the average
chemical composition can be expressed in percent zinc-metal
equivalent (%ZME) units. ZME units provide a useful mea-
sure of the response of the probe to the total composition of
the material. They are dened from equivalence factors (EF),
which are numerical values that translate contents of other
heavy elements (e.g., lead, iron, manganese) into equivalent
zinc content. The EF is a multiplier such that gamma radiation
scattered and/or absorbed by a compound of a multiheavy-
element medium (and measured through a given spectral pa-
rameter) results in the same value as that of the spectral param-
eter being measuredina monoheavy-metal ore (e.g., zinc only).
For percent zinc content (%Zn), the %ZME value is
%ZME = %Zn +%Pb
EF
Pb/Zn
+%(Fe +Mn)
EF
(Fe+Mn)/Zn
.
In other words, gamma-ray interaction with matter (above
the energy of the lead X-ray peak) in a multiheavy-element
mediumshould be the same as the interaction in a hypothetical
monoheavy-element medium described by the same %ZME.
EFs are usually established empirically and have differing val-
ues for different spectral ratios or windows.
The quantitative determination of zinc content in ore, af-
ter correcting for lead contents spectral contribution (X-ray
peak), is possible only if the iron and manganese content are
either relatively constant, at least within a given ore unit,
or can be measured by a complementary physical or chem-
ical technique. The iron component requires an alternative
measurement techniqueperhaps a magnetic susceptibility
probethat can be incorporated into the measurement sys-
tem to provide a single-pass logging operation. Pyrite, which
is fairly common in lead-zinc suldes, does not give a response
to susceptibility; hence, iron from pyrite cannot be estimated
this way.
INSTRUMENTATION
The large hole size (142 mm) permitted application of
a higher volume crystal than the standard detectors of the
60-mm Sirolog probes. A probe with an external diameter
of 100 mm was designed and manufactured. This probe uses
a 76 76-mm NaI crystal. The auxiliary equipment consists
of a portable single conductor cable and winch with a winch
controller, the surface electronics unit (Siromca), and a ded-
icated PC computer. Because of the very weak gamma-ray
source applied (<2 MBq), there was no need for a source
1972 Charbucinski et al.
container/transporter. The Siromca unit combines the func-
tions of pulse shaper, nuclear amplier, and digitally gain-
stabilized multichannel analyzer.
In the new generation of Sirolog instrumentation, the
Siromca surface electronics unit has been eliminated. All
Siromca functions and additional features have been incor-
porated directly into the logging probe, and all Sirolog spec-
trometric systems use essentially the same hardware. Recent
microprocessor developments have allowed the design of a
generic multichannel analyzer (MCA) system that can fulll
the needs of these low-count-rate and low-power-consumption
systems while coping with the higher count rates of logging
(100,000 cps). The system was designed to t inside a 60-mm
probehousinganduses serial communications totransmit com-
plete spectra up the cable in digital format to a PC.
FIELD TRIALS
To establish the bed resolution capability of the logging
probe, static measurements were performed in a two-layer
model simulating barren rock and zinc-lead ore. The measure-
ment points were established by moving the probe uphole in
1-cm increments. Figure 2 presents the change of count rate
recorded in the spectral window (112135 keV) for the ore-to-
rock transition. A value of 120 mm is a distance for which a
full transition from ore count rate to barren rock count rate
takes place. This value is the distance fromthe barren rock/ore
boundary, where the probe will record information reecting
the intrinsic value of the spectral parameter for the lithological
unit.
Two cored holes, reamed later to a diameter of 142 mm
(which is close to the anticipated diameter of future production
holes) were available for the tests. Both holes were water lled.
Cors obtained from those holes were subdivided according to
the observedlithological classication. Over 80 sections of core
were assayed, and the results of laboratory analysis were used
as reference points for assessing the probes potential to deter-
mine lead and zinc content.
Delineation of the orebody
The borehole tests were performed at logging speeds be-
tween1.5and2.0m/min, implyingintegrationtimes of 1.5s for a
sampling interval of 5 cm. The spectral parameters, which were
applied as delineation indicators, were the same as the best in-
FIG. 2. Ore-to-rock transition response prole (static measure-
ments, 10-mm increments).
dicators found during laboratory testsnamely, the spectral
window (112135 keV) and the spectral window ratio (190
235/112135 keV). Figure 3 shows a geophysical prole based
on this spectral ratio plotted alongside the zinc chemical assay
log. The spectral ratio changes signicantly with the transition
from barren rock to zinc ore, from around 1.41.5 to 2.32.5,
providing a goodmeans for ore delineation. Acomparisonwith
previously acquired commercial logs indicate the Sirolog low-
radiation-intensity probe offers better boundary denition.
Probe quantitative performance
The analysis of the spectrometric data was aimed at selecting
a set of spectral ratios and/or spectral windows to be used to
quantitatively analyze acquireddata. Initially, the windows and
ratios used for to analyze the laboratory data were applied.
Then, a number of newwindows and ratios were evaluated and
subjected to an analysis where step-by-step changes of Pb/Zn
and (Fe +Mn)/Zn EFs were investigated. The nal stage of
selecting the optimal spectral parameter for quantitative zinc
predictions (after tting the right EFs) was a ne-tuning of
the spectral boundaries of the best spectral ratio found in the
preceding stage. Figure 4 shows correlation between values of
%ZME derived from logging, applying the optimal spectral
parameter (based on the special ratio presented above) for a
linear calibration equation and the values of %ZME based on
chemical assays.
The rms deviation between laboratory assays and the tools
predictions was 2.4%ZME. Thestandarddeviationfor thepop-
ulation, s, was 10.4%ZME, and the number of assays in the re-
gression equation was 72. The correlation coefcient was 0.97.
Only 72 samples were considered in the regression equation.
Thereasonfor not includingall 80+samples was tohaveamore
uniform distribution of samples in the regression. Therefore, a
number of low-grade samples (ZME<9) were (randomly) not
included in the analysis. If all of the samples were included in
the equation, the data points would have been clustered in the
area of low ZME.
Determination of lead content
The 87-keV characteristic X-ray peak showing prominently
in the backscattered gamma-ray spectrum was used to deter-
mine lead content. The results of the laboratory investigations
of the potential of the Sirolog low-radiation-intensity probe
for lead determination has been reported by Almasoumi et al.
(1997). The spectral parameter basedona ratioof net leadpeak
area to gamma-ray intensity of energy just above the lead peak
provided a good measure of lead content in the ore. Figure 5 is
a crossplot of the predicted percent of lead content (obtained
from a two-variable calibration equation) against the labora-
tory assays. The rms deviation between the eld estimates and
the laboratory assays was 0.3%Pb, and the standard deviation
of the population was 1.7%Pb. The correlation coefcient for
the regressed data was 0.98.
Determination of zinc content
To assess the feasibility of zinc-grade predictions from
log-derived values of %ZME and lead, and also iron and
manganese contributions to %ZME, predictions need to be
estimated. Spectrometric logging data permit the application
of an intrinsic correction of the lead contribution by measur-
ing the net area of the characteristic lead X-ray peak. Because
the iron and manganese contributions to the total value of
%ZME cannot be measured from the spectrometric gamma
gamma log, the corrections for iron and manganese were based
on statistical information from the geological data base. Fig-
ure 6 shows a crossplot between zinc content derived from
%ZME and the chemical assays. The correlation coefcient
for the data set was equal to 0.90, and the rms deviation was
equal to 2.4%zinc. Only when the lead contribution was taken
into account were the quoted values 0.88 and 2.6%Zn, cor-
respondingly. However, a noticeable improvement may be
possible when percent iron (and percent manganese) con-
tent is measured directly by another physical or chemical
technique.
Reproducibility tests
Each hole was logged at least twice to assess how repro-
ducible the obtainedproles andquantitative predictions were.
Figure 7 shows logs of duplicate runs in one of the holes.
The logs show spectral intensity measured in energy window
112135 keV. For better clarity of comparison, run 2 has been
shifted vertically along the y-axis. Runs 1 and 2 are almost
identical, proving a high quality of delineation reproducibility.
Reproducibility of quantitative predictions was tested by
comparing values of %ZMEobtained fromrun 1 with the cor-
responding values obtained from run 2. The 83 logging splits,
corresponding to the core samples taken for chemical analysis,
FIG. 3. Chemical assay for the test hole and the logging prole based on the spectral ratio: 140235/110135 keV (T. Petrov, 1996,
unpublished ATD Rio TintoC.R.A. report).
were used to test reproducibility. Values of the spectral ratio
used for analyses were derived from both logging runs; subse-
quently, pairs of %ZME were calculated using the calibration
equation. Finally, the values fromrun 2 were regressed against
the values from run 1. The correlation coefcient for the data
set was equal to 0.99, and the standard deviation (rms) for the
regression was equal to 0.85% ZME, indicating high repro-
ducibility of the logging data.
Determination of cut-off boundary
The application of the Sirolog low-radiation-intensity probe
for predicting the cut-off boundaries of zinc ore was tested on
FIG. 4. Crossplot of log-derived %ZME versus %ZME
obtained from chemical assays.
1974 Charbucinski et al.
FIG. 5. Comparison between log-derived %Pb and %Pb from
chemical assays.
FIG. 6. Comparison between %Zn, log-derived (%Zn-log),
and %Zn from chemical assays (%Zn-chem). The log-derived
values of %Zn are Pb and Fe +Mn corrected.
one of the ore units. By denition, the cutoff boundary is the
depth where the integrated zinc value from the top of the ore
unit reaches the lowest economically acceptable value for the
minedunit. Theintegratedvalues of %Znwerecalculatedfrom
the top of the ore unit in 5-cm-depth increments. We assumed
that the concentrations of zinc, lead, iron, and manganese were
constant for each of the 5-cm splits that belonged to a chem-
ically assayed sample interval. This assumption might not al-
ways be accurate. Figure 8 shows proles of integrated (from
the top of the ore unit) values of % Zn derived from chemical
assays (%Zn-chem) and two proles of %Zn derived fromlogs
(%Zn-log), one corrected on only %Pb and the second addi-
tionally corrected by the average (for a particular unit) value of
%Fe +%Mn. The proles of the log-derived %Zn-log match
the proles of %Zn obtained from chemical assays. The %Zn
log proles show more structure than %Zn-chem for the top
section of the ore unit.
CONCLUSIONS
The spectrometric low radiation intensity probe is effec-
tive in the zinc-lead ore application. Nonspectrometric probes
would be incapable of equivalent performance standards. The
probe utilizes a gamma-ray source of activity <2 MBq and pro-
vides minimal risk to safety and environmental integrity.
The probe showed excellent delineation characteristics and
indicated considerable variation in the log. The logging data
showed sharp denition, both in selected spectral regions and
FIG. 7. Proles of duplicate logs in one of the test holes (prole
of run 2 has been shifted vertically on the count-rate axis).
FIG. 8. Proles of integrated values of %Zn from the top of
the ore unit for both chemical assays and log-derived values.
in spectral ratios during probe transition fromore body to bar-
ren rock, or vice versa. Placing the source a very short distance
from the detector provided the probe with the best possible
vertical resolution.
The quantitative performance of the probe was satisfacto-
rily demonstrated, providing determination of % Zn to within
2.4% rms and %Pb within 0.3% rms. Delineation of nonlitho-
logical boundaries by applying a cutoff grade boundary was
also demonstrated as a practical stand-alone mine control tool.
ACKNOWLEDGMENTS
The authors are grateful for the collaboration and support
given by RTZCRA. The authors thank A. Hugill and T. Petrov
for their active collaborationin particular, for providing for-
ward modeling of the ore body. Contributions fromA. Mutton,
A. Waltho, and L. Partridge are also acknowledged. The au-
thors are also grateful for the technical support provided by
R. Dixon and Z. Jecny of CSIRO.
REFERENCES
Almasoumi, A., Borsaru, M., and Charbucinski, J., 1997, The applica-
tion of the gammagamma technique for the determination of lead
concentration in boreholes in a Pb-Zn deposit using very low activ-
ity gamma-ray sources: Internat J. Appl. Rad. Isotop., 49, No. 12,
125131.
Artsybashev, V. A., 1972, Yaderno-geophizitshaya razvedka: Atomiz-
dat.
Borsaru, M., and Ceravolo, C., 1994, A low activity spectromet-
ric gammagamma borehole logging for the coal industry: Nucl.
Geophys., 9, 343350.
Borsaru, M, Charbucinski, J., and Eisler, P. L., 1994, Nuclear in-situ
analysis techniques for the mineral and energy resources mining
industries: Proc. 9th Pacic Basin Nuclear Conference, 1, 391398.
Borsaru, M., Ceravolo, T., and Tchen, T., 1995, The application of the
low activity borehole logging tool to the iron ore mining industry:
Nucl. Geophys., 9, 5562.
Charbucinski, J., 1993a, Comparison of spectrometric neutron-gamma
andgammagammatechniques for in-situassayingfor ironoregrade
in large diameter production holes: Nucl. Geophys., 7, 133141.
1993b, The ZERO PROBE low radioactivity borehole logging
tool: Nucl. Sci. Symp., 2, 855859.
Charbucinski, J., Ceravolo, C., and Tchen, T., 1996, Ultra-low activ-
ity spectrometric probe for the coal mining industry: J. Radiochem.
Nucl. Chem., 206, No. 2, 311319.
Eisler, P. L., Charbucinski, J., and Borsaru M., 1990, CSIRO borehole
logging research and development for the solid resource mining
industries: Proc. Internat Symp. Internat. Atomic Energy Agency,
IAEA-SM-308/59, 299318.
Case History
The Spectrem airborne electromagnetic systemFurther developments
Peter B. Leggatt
, Philip S. Klinkert
, and Teo B. Hage
ABSTRACT
The Spectrem airborne electromagnetic (AEM) sys-
tem has been in full production in Canada and
Africa since 1989. The prototype, built by A-cubed of
Mississauga, Canada, was subject to two major upgrades
by Anglo American Corp. staff to produce the Spec-
trem II instrument that located the Photo Lake and
Konuto Lake orebodies in Manitoba, Canada. Subse-
quent work in tropical Africa has made it desirable to
further increase the transmitter power and use greater
available computer power in the aircraft to substantially
improve the ability of the system to reject sferic noise,
although there appears to be a limiting sferic level above
which current ltering methods are ineffective.
Although designed to detect massive sulde bodies
deep below conductive overburden, techniques have
been developed to map the regolithin particular, to
compute resistivitydepthsections using anapproximate
but fast method. Acomparisonis presentedof sucha sec-
tion with the same section own and processed using a
frequency-domain helicopter AEM system.
The latest upgrade of the system, Spectrem 2000, with
a 50% increase in transmitter power, was completed
in October 1999 and is currently at work in northern
Canada.
INTRODUCTION
The Spectrem airborne electromagnetic (AEM) system has
been ying surveys in Canada and Africa since 1989. In that
year, the prototype, described by Annan (1986), was seven
years old and the original computer hardware then consider-
ably out of date. A review of operational performance in 1990
identied major inefciencies attributable to the poor avail-
ability of spare parts for the gasoline-powered engines of the
DC3 (built in 1940). As a result, Basler Turbine Conversions
Manuscript received by the Editor June 14, 1999; revised manuscript received June 9, 2000.
Spectrem Air Ltd., P.O. Box 457, Hanger 204, Lanseria Airport, Lanseria 1748, South Africa. E-mail: peterl@spectrem.co.za; teo@spectrem.co.za.
Inc. of Oshkosh, Wisconsin was hired to modify the airframe
and t modern PC16 turbine engines. While this work was in
progress, almost all of the electronic hardware was redesigned
and replaced by Anglo American Corp. staff, as was all of the
computer hardware.
The newAEMsystemwas named SpectremII (described by
Klinkert et al., 1997), and operations commenced in Manitoba,
Canada, in 1993. The rms transmitter dipole moment had been
increased by a factor of two to about 300 000 Am
2
. The princi-
pal result was the discovery of the Photo Lake orebody, whose
prole is shown in Figure 1. Subsequently, at least two new
conductors located by this survey have been followed up and
evaluated as massive sulde orebodies. These are now produc-
ing mines or are about to be developed.
The success of this effort is largely because of vigorous
ground follow-up of the conductive targets by the Flin Flon
mine exploration staff.
It is worth examining the results of this early success to iden-
tify the best method likely to improve the system.
SIGNAL PROCESSING
The AEM system described here differs from the pulse sys-
tems which evolved fromthe original Input time-domain AEM
system in that the transmitter waveform approximates a mag-
netic step rather than the magnetic impulse of the pulse sys-
tems. Having a step transmitter waveform means effectively
that the pulse width of the inducing signal can be almost 100%
of the time for a half-waveform. Liu (1998) gives a good ex-
position of the merits of different waveforms. He states, The
ratio of the target voltage response due to two pulses of differ-
ent shape is approximately equal to the ratio of the areas that
these pulses enclose with the time axis for a target of large time
constant.
The drawback of having the transmitter always active is that
one must effect the separationof the transmitter primary signal
from the ground secondary signal at the receiver by signal pro-
cessing. This is unlike pulse systems, which effectively measure
the secondary signal when the transmitter is off. The primary
1976
Spectrem AEM Developments 1977
and secondary signals are rather similar in shape. Unlike the
situation with ground EM systems with a static geometry, the
amplitude of the transmitter primary signal at the receiver can-
not be exactly computed from the static geometry. This is be-
cause the exact geometry at any moment is unknown because
of the unpredictable relative motion between the towed bird
and the aircraft. These primary signal variations are generally
orders of magnitude larger thanthe secondary-eldamplitudes
and, if not removed, will entirely swamp the latter.
In Figure 1 the X primary eld (bottomtrace) shows a varia-
tion of about 10%just to the right of the Photo Lake body over
a width much the same as the anomaly. In the more turbulent
conditions to be found in the tropics, these variations are even
larger. Nearly all this variation in primary signal is from rota-
tion of the transmitter and receiver coils with respect to each
other.
To deconvolve the received signals and compute the impulse
response, one must construct an operator which, when con-
volved with the received primary signal, will reduce the latter
FIG. 1. The Photo Lake orebody detected by the AEM system in 1993. Two groups of proles are shown. The upper group is the Z
channel (coil oriented vertically), while the lower is the X channel (coil oriented horizontally along the line of ight). Within each
group the early time channels are at the top. The bottom channel in each group is the primary eld. Vertical scales decrease with
later time. The scales are chosen so that residual low-frequency effects show the same amplitude on all channels to aid recognition.
to an impulse. It has long been known in the seismic indus-
try (Robinson, 1978) that this is normally possible only when
the primary signal is minimum phase. The transmitter primary
waveformhas accordingly been designed so the half-waveform
is minimum phase.
Once the impulse response has been determined, the step re-
sponse of both primary and secondary signals is calculated. For
the step response, the secondary signal resulting fromconned
conductors can be closely approximated by a sum of exponen-
tial decays (Kaufman, 1978), while the primary signal takes the
formof a dc level as designedthis level being the amplitude of
the step which that signal has become. In this form the primary
signal is easier to separate from the desired secondary signal.
Effectively, towithina scaling constant, one computes the mag-
netic eld that results from a transmitter whose current wave-
form is a perfect step. This has advantages in interpretation
and, of course, in the computation of conductivitydepth sec-
tions using a technique pioneered by Macnae and Lamontagne
(1987).
1978 Leggatt et al.
An example of the deconvolution is shown in Figures 2
and 3. Figure 2 shows the signal, stacked down to a single
half-waveform after removal of the low-frequency and sferic
noise. This plot of the rst 200 samples of the 256-sample half-
waveform of the combined primary and secondary signals is
almost indistinguishable from a stack devoid of any secondary
signal because the latter is so small. To make this point, the
lower of the two traces shown in Figure 2 is the combined
primary and secondary signal over Photo Lake superimposed
on the primary signal alone. (The latter is recorded at a great
height away from ground signals.)
A least-squares adjustment of the scale of the prerecorded
primary reference signal has been applied to equalize the am-
plitudes of the twosignals superimposedinthe lower trace. The
upper trace is the same scaled reference signal, shifted upward
20 mVtofacilitate comparison. One cansee, withsuitable mag-
nication, the added secondary signal as a slight thickening of
the lower traceparticularly at early time.
In Figure 3 one sees the same data converted to step re-
sponse with the primary signal removed. The nal 56 points
have been omitted to simplify the plots. The last 128 samples
of the step response are used as an initial estimate of the level
of the primary step, which has been subtracted in the plot. The
rst 128 points of the waveform are averaged down to seven
windows, shown as boxes in Figure 3. Usually only these seven
values and the value of the primary signal are saved for further
processing and interpretation from prole plots. A nal value
of the eighth secondary signal window (occurring in samples
129 to 256), the late time-decay constant, and the corrected
FIG. 2. Sample of the 200-ms X channel data ltered and
stacked to a single half-waveform at the peak of the secondary
signal over the Photo Lake orebody. The lower trace shows
the primary and secondary superimposed; the upper trace is
the primary alone. To aid comparison the upper trace has been
displaced upward by 20 mV. Only the rst 200 of the 256 points
of the waveform are plotted here. The secondary signal is so
small that it can only be identied by the slight thickening of
the lower trace.
primary signal can be computed from this data further down
the processing chain.
In the AEM system signals shown here, there is no specic
sferic removal applied in the stream data apart from removing
all frequencies that are not odd multiples of the fundamental
transmitter frequency. (The transmitter signal has half-wave
symmetry, so its spectrum contains only frequencies that are
odd multiples of the primary frequency.) Figures 2 and 3 illus-
trate howcareful one must be not todamage the tiny secondary
signal by data processing.
NEWAREAS, NEWPROBLEMS
In 1997 the aircraft returned from Canada to its home in
Africa and surveys were own in subtropical and later tropical
areas with large areas of conductive cover that presented a
very different prospecting environment than northern Canada.
More recently, work has been completed in West Africa.
All the Africansurveys, ownwithinabout 15
latitude of the
Equator, encountered an unprecedented level of sferic noise
compared to that experienced in higher latitudes.
The experience conrms the analysis of Buselli andCameron
(1996), who report sferic levels in northern Australia some 100
times worse than levels in the south. While we have tried many
of the robust methods of sferic reduction mentioned in their
article, we have had most success using the methods of Strack
(1990). This is a simple selective stack where only a proportion
of the data close tothe medianof the stackis used. However, we
had not anticipated that the recently encountered high sferic
levels could persist for days at a time.
The practice of simply waiting for unusually high levels of
sferics to pass, like a magnetic storm, leads to unacceptable loss
FIG. 3. The data of Figure 2 converted to a display of the step
response. The last 128 samples of the waveform (not all of
which are shown) have been used to estimate the dc value
representing the primary eld. The boxes are a display of the
seven values of secondary signal saved, the height of each box
being the average of all the samples between its ends.
of production in equatorial latitudes. James Macnae (personal
communication, 2000) advises that heavy sferics are inevitable
in all seasons in equatorial Africa (see thunder.msfc.nasa.gov).
A200-ms sample of data with moderate sferic noise is shown
in Figure 4, and an example of data from tropical Africa is
shown in Figure 5, both using 25-Hz data. Turbulent motion
FIG. 4. A 200-ms buffer of moderately noisy data recorded at
25 Hz with 7680 digitized values from the Y channel of the re-
ceiver. After removal of lowfrequencies, the result is displayed
1 V below the time axis. After removing sferics, the result is
displayed 1 V above the time axis.
FIG. 5. Data similar to Figure 4 with very large amounts of
sferic andturbulencenoise. Notethehugelow-frequencysignal
which comes from turbulent motion of the bird in the earths
magnetic eld.
of the towed receiver coils causes them to rotate with respect
to the transmitter. This has two undesirable effects: (1) the
amplitude of the primary signal changes because of the change
in coupling with the transmitter coil and (2) the rotation itself
induces low-frequency signals from coupling changes with the
earths magnetic eld. In Figure 5 the Y channel is displayed
(horizontal component perpendicular to the ight direction).
This component is nominally null coupled to the transmitter;
not only can the amplitude of the primary eld change, but also
its sign.
A two-stage ltering process is demonstrated. First, the low-
frequency (<25 Hz) signal is removed with a linear lter. The
result is displayed along with a 1.0-V displacement. Then
the ve negative half-waveforms of the ten in the stack are
inverted. For each of the 768 samples in the half-waveform,
we have ten specimens. The median of these ten specimens is
computed, and the six specimens with values furthest from the
medianare discarded. The other four are averagedtoproduce a
singlepoint of the768-point waveform. Thetenhalf-waveforms
shown displaced +1.0 V above zero are a replication of this
resultant single half-waveform.
The implied inference in this procedure is that one can ex-
pect no sferics for about half the time. Figure 5 shows this is a
fallacy in some parts of the world. Considering the small sec-
ondary signal of an orebody shown in Figure 2, the accurate
removal of sferics and other noise is very important. While
lower fundamental transmitter frequencies improve the depth
penetration, lowering the frequency from 75 to 25 Hz reduces
the number of half-waveforms to stack from 30 to 10, with a
consequent reduction in noise rejection capability.
NEWTRANSMITTER WAVEFORM
Early in October 1999 a new system, identied as Spectrem
2000, was completed. A new transmitter using quasi-resonant
techniques has been developed to attempt an rms dipole mo-
ment of 1 million Am
2
. For the time being the rms dipole mo-
ment has been increased to 450 000 Am
2
, since realizing full
power will involve rather costly modications to the loop wires,
their suspension, and mounting on the airframe.
The transmitter waveform is approximately a square wave
because of the desirable properties of that waveform as previ-
ously discussed (Liu, 1998). The new waveform is an improved
approximation to a square wave, with a faster rise time and a
atter top. The higher rms dipole moment increases power in
the lower frequency components, which facilitates deep pene-
tration, while the faster rise time permits better resolution of
shallow features.
The better resolution assumes greater importance when sur-
veying prospects having a complex conductivity structure, as
distinct from discrete conductors in resistive host rock. A mas-
sive sulde orebody in the latter category gives an anomaly
with a sharp and localized peak. But the complex conductiv-
ity structure of Africa, with deep and conductive weathering,
requires more emphasis on mapping the regional conductive
structure and interpreting it in combination with other data
such as magnetics and radiometrics integrated with mapped
geology.
In the Spectrem I system, data were stacked for a period
of 200 ms down to a single half-waveform of 128 points. Af-
ter deconvolution, the data volume was further reduced by
1980 Leggatt et al.
averaging (or binning) the 128 points into seven binary spaced
windows (sometimes called gates), each twice as wide as its
predecessor. Apart from a few test ights, the 128-point stack
was not recorded. In the current AEM receiver, the stacked
half-waveform (now 256 samples because of faster digitizers)
is routinely saved to tape for postight analysis. This has given
opportunities for new processing insights, the most important
being that there are categories of noise that, once blurred into a
common stack, cannot be removed effectively. In consequence
the latest receiver records all data fromall digitizers in an unin-
terrupted stream (prestack) during each active traverse. Data
storage requirements have increased from 800 bytes/s (binned
data) to 21 280 bytes/s (stacked half-waveform data) and most
recently 738 880 bytes/s (prestack data).
RESISTIVITY MAPPING
The mapping of regional conductivity variations has proved
rather useful in Africa. The volume of data collected during a
survey is too great to permit a full 3-D inversion of an entire
survey, even if this were desirable. An approximate method,
described by Macnae et al. (1991) and termed conductivity-
depth imaging, is used to compute a conductivity or resistivity
depth parasection.
The method strictly applies only to a horizontally layered
earth. The changing amplitude of the secondary EMsignal as a
function of time is used to compute a slowness (reciprocal ve-
locity) function at the position of an assumed receding image
source. In turn, conductivity as a function of depth is com-
puted from the slowness of the downward receding image. The
method, akin to resistivity mapping, explicitly assumes that the
earth model is a 1-D layered earth.
To honor the layered-earth model, one must assume that
the data reect a model that is at least approximately a layered
earth to within, say, twice the horizontal coil spacing, or about
250 m. The AEM system has a vertical transmitterreceiver
separation of 35 m and a horizontal spacing of 125 m, geome-
try achieved by use of a very high-drag receiver bird. This small
vertical separation results in the Z-component anomaly being
effectively symmetrical for proles across horizontal conduct-
ing slabs. The EM response at the edges of such slabs consists
of negative side lobes about 100 m wide when ying at 100 m
survey height. As these negative-amplitude features, together
with other narrow responses from culture, power lines, and
shallow discrete conductors, degrade the conductivity calcula-
tions, they are removed using a low-pass lter prior to com-
puting the conductivity parasection. While there is no doubt
that this reduces the horizontal resolution, it is a requirement
forced by limitations of the layered-earth model.
Figure 6 shows a resistivitydepth parasection computed
fromsynthetic data. The model is a 500 500 50-m-thickhor-
izontal conductive slab of conductivity 1 S/m embedded in a
half-space of 0.004 S/m. The top of the slab is on the ground sur-
face. The upper part of the gure shows the ltered Z-channel
secondary amplitudes, while the lower part shows the conduc-
tivity parasection on which is superimposed the traces of the
image depths computed from each channel. Lateral ltering
has removed the side lobes at the edges of the body, and the
minor asymmetry that remains is less noticeable in the image
than in the EM amplitude data. The resistivitydepth parasec-
tion represents the body as being more narrow than it really is
andhas roundedtheupper surfacequiteseverely. Thethickness
is represented as about twice the actual thickness. The repre-
sentation of the body is reasonably symmetrical so that one
would have difculty identifying whether the line was own
left to right or right to left. The most serious asymmetry is the
small displacement of the entire body to the right. This is easily
rectied in plan presentations by applying a lag to the entire
prole.
The numbers displayed along the ground surface are sample
(ducial) numbers; in this case, we have 200 samples across the
full 1 km of the section. The horizontal scale is about 10 times
ner than one would normally use to display a real section.
Acceptingthelimitations of theunrealistic earthmodel used,
conductivitydepth data are independent of aircraft terrain
clearance and ying direction. So the depth parasections can
be used to produce maps of several derived parameters, in-
cluding contours of depth to a given interface, the conductivity
or thickness of the overburden, or the conductivity below a
given depth. In practice the limitations of the model are less
restrictive thanthe synthetic example might suggest. Infact, for
features of lateral dimension larger than the aforementioned
250-m restriction, considerable vertical variations can be
displayed.
Figure 7 shows a resistivitydepth parasection own by both
xed-wing AEM and a frequency-domain helicopter electro-
magnetic (HEM) system (Dighem) for comparison. The heli-
copter data have been processed with two different algorithms
which give slightly different results. Details of xed-wing AEM
processing are given in Leggatt and Pendock (1993). HEM de-
tails are given in Huang and Fraser (1996) and Sengpiel (1988).
While both systems give parasections that are broadly similar,
FIG. 6. A resistivitydepth parasection computed from syn-
thetic data from the Emigma modeling package. The model
was a 500 500 50 m horizontal slab of 1 S/m conductivity
embedded in the top of a half-space of 0.004 S/m conductivity.
The numbers along the surface of the ground are sample num-
bers, and each represents 5 mof traverse. The entire traverse is
1 km long. The true position of the body is shown by the rect-
angular outline. The resistivitydepth parasection has rounded
the shoulders of the body signicantly and overestimated its
thickness. The major asymmetry in the resistivitydepth para-
section is the small displacement of the model to the right.
three boreholes were selected to examine the section at points
where the EM results differed most from each other.
At borehole 6750, xed-wing AEMshows a thin surface con-
ductor over a resistive layer about 90 mthick with another con-
ductive layer below it. The HEM data image the resistive layer
poorly and fail to detect the deeper conductive layer below it.
Unfortunately, the hole was not drilled deep enough to test
the thickness of the lower conductor dened by the xed-wing
system.
Borehole 6650 conrmed the presence of the deep conduc-
tor and showed the interface at its top and bottom at depths
predicted by xed-wing AEM. The HEMsurvey did not detect
the base of this conductor. Borehole 6500 conrmed both the
AEM conductivity and the Sengpiel inversion of HEM for the
upper part of the section, while both methods of processing
HEM failed to resolve the resistive layer lower down. One ex-
pects a helicopter EM system to be slightly better at resolving
shallowhorizons than a xed-wing AEMsystem, but this is not
evident in this example. The greater low-frequency component
of the AEM square-wave signal, designed to maximize pene-
tration, is shown to advantage in the resolution of structures
below 90 m.
FIG. 7. Comparison of resistivitydepth parasections computed for HEM (Dighem) and xed-wing AEM (Spectrem) systems. The
drillholes were located at points where the two systems differ. The xed-wing system running at a transmitter frequency of 75 Hz
seems to have shallow-layer vertical resolution as good as HEMand better depth penetration. Despite inherent limitations in lateral
resolution of the xed-wing system discussed in the text, the horizontal resolution seems no worse than for the HEM parasection.
Parasections of AEM data usually display conductivity, but
inthis example, resistivityhas beenplottedusingcolors approx-
imately the same as those used for HEM, for ease of compari-
son. Even in view of the lateral lter applied to the xed-wing
AEM data (15 ds), the lateral resolution of the xed-wing
system appears at least as good as the HEM system in this
particular example.
CONCLUSIONS
AEM survey work in Africa encounters more problems of
noise from geological sources than is the case in the Canadian
Shield. The tropical environment creates more turbulent air
conditions whenying over hot areas. The heat itself (occasion-
ally >50
C in the aircraft) makes operating conditions more

difcult and unpleasant and is harder on the equipment. In the
Intertropical Convergence Zone, the level of sferics is often
so high that good data acquisition can be very difcult. There
is a critical level of sferic noise which, when exceeded, makes
recovery impossible within the short (200 ms) time available
for an AEM reading. This results in data dropouts.
Conductivitydepth imaging is a useful tool for approximat-
ing the structure of an essentially horizontally layered earth,
1982 Leggatt et al.
which has useful applications in mapping the subsurface ge-
ology. Despite an inherent limitation in lateral resolution,
considerable detail can be deduced from a parasection that
is conrmed by drilling.
ACKNOWLEDGMENTS
The authors thank the Anglo American Corp. for permis-
sion to publish this work and Steve Lynch for calculating the
Emigma model. Emigma is an EM forward-modeling software
package from PetRos Eikon Inc. Thanks are also extended to
James Macnae and an anonymous reviewer for constructive
and appreciated criticism.
REFERENCES
Annan, A. P., 1986, Development of the prospect 1 airborne electro-
magnetic system, in Palacky, G. J., Ed., Airborne resistivity mapping:
Geol. Surv. Canada, 6370.
Buselli, G., and Cameron, M., 1996, Robust statistical methods for re-
ducing sferics noise contaminating transient electromagnetic mea-
surements: Geophysics, 61, 16331646.
Huang, H., and Fraser, D. C., 1996, The differential parameter method
for multifrequency airborne resistivity mapping: Geophysics, 61,
100109.
Kaufman, A., 1978, Frequency and transient response of electromag-
netic elds created by currents in conned conductors: Geophysics,
43, 10021010.
Klinkert, P. S., Leggatt, P. B., and Hage, T. B., 1997, The Spectrem
airborne electromagnetic systemlatest developments and eld ex-
amples: 4th Internat. Conf. on Min. Expl., GEO/FX, Proceedings,
557563.
Leggatt, P. B., and Pendock, N., 1993, Conductivity-depth imaging of
airborne electromagnetic step-response data using maximum en-
tropy, in Mohammad-Djafari, A., and Demoment, R. G., Eds., Max-
imum entropy and Bayesian methods, Kluver Acad. Publ.
Liu, G., 1998, Effect of transmitter current waveform on airborne tem
response: Expl. Geophys., 29, 3541.
Macnae, J. C., and Lamontagne, Y., 1987, Imaging quasi-layered con-
ductive structures by simple processing of transient electromagnetic
Macnae, J. C., Smith, R. S., Polzer, B, D., Lamontagne, Y., and Klinkert,
P. S., 1991, Conductivity-depth imaging of airborne electromagnetic
step-response data: Geophysics, 56, 102114,
Robinson, E. A., 1978, Multichannel time series analysis with digital
computer programs: Hobden Day.
Sengpiel, K. P., 1988, Approximate inversion of airborne EMdata from
a multi-layered earth: Geophys. Prosp., 36, 446459.
Strack, K. M., 1990, A practical review of deep transient time electro-
magnetic exploration: Harbour Dome Consulting.
Advanced inversion methods for airborne electromagnetic exploration
Klaus-Peter Sengpiel
and Bernhard Siemon
ABSTRACT
Airborne electromagnetic (AEM) surveys can con-
tribute substantially to geologic mapping and target
identication if good-quality multifrequency data are
produced, properly evaluated, and displayed. A set of
multifrequency EM data is transformed into a set of ap-
parent resistivity (
a
) and centroid depth (z
p
) values,
which then are plotted as a sounding curve. These
a
(z
p
)
curves commonly provide a smoothed picture of the ver-
tical resistivity distribution at the sounding site. We have
developedandcheckedmethods toenhance the sensitiv-
ity of sounding curves to vertical resistivity changes by
using new denitions for apparent resistivity and cen-
troid depth. One of these new sounding curves with en-
hanced sensitivity to vertical resistivity contrasts is plot-
ted from
NB
, z
s
values derived from differentiation of
the
a
( f ) curve with respect to the frequency f . This ap-
proachis similar tothe Niblett-Bosticktransformusedin
magnetotellurics. It not only enhances vertical changes
inresistivity but alsoincreases the depthof investigation.
Sounding curves can be calculated directly from EM
survey data and can be used to generate a resistivity-
depth parasection. Based on such a section, it can be de-
cided whether a Marquardt-type inversion of the AEM
data into a 1-D layered half-space model is adequate.
Each sounding curve can be transformed into an initial
step model of resistivity as required for the Marquardt
inversion. We have inverted data from sedimentary se-
quences with good results. For data from a dipping con-
ducting layer and a dipping plate, we have found that the
results depend on the right choice of the starting model,
in which the number of layers should be large rather
than too small. Complex resistivity structures, however,
oftenare representedbetter by using the sounding-curve
results than with the parameters of a layered half-space.
INTRODUCTION
The applicability of the airborne electromagnetic (AEM) in-
duction method for exploration depends on the tools available
for a detailed study of the ground resistivity distribution. Such
studies include the determination of the boundaries of con-
ducting bodies and zones in three dimensions, determination
of the thickness and type of sediment above the bedrock, trac-
ing of paleochannels that are lled with sediments of different
conductivity than the surroundings, and identication of fault
systems and the structure of the bedrock.
A detailed investigation of the resistivity pattern requires a
high-resolution measuring device. Maximum horizontal reso-
lution is achieved for coincident transmitter and receiver
coils (Duckworth et al., 1993), but in practice, it is suf-
cient if the coil spacing is smaller than the height of the sen-
sor above the ground. For frequency-domain electromagnetic
(FEM) systems, the vertical resolution depends on the num-
ber of frequencies used. Because FEM systems are housed in
Published on Geophysics Online July 19, 2000. Manuscript received by the Editor March 15, 1999; revised manuscript received May 12, 2000.
Federal Institute for Geosciences and Natural Resources (BGR), Stilleweg 2, D-30655 Hannover, Germany. E-mail: sengpiel@bgr.de.
Presently at Geophysik GGD, Hamburger Allee 12-16, D-30161 Hannover. E-mail: bernhard.siemon@t-online.de. Permanent address: Federal
Institute for Geosciences and Natural Resources (BGR), Stilleweg 2, D-30655 Hannover, Germany.
a bird, space and thus the number of the coil systems are
limited.
For helicopter electromagnetic (HEM) data, methods for
deriving apparent resistivity
a
from the measured secondary
eld data were developed by Fraser (1978) and by Mundry
(1984). The rst relation that assigns a depth to the apparent
resistivity using the centroid depth (z
) concept for dipole in-

duction was published by Sengpiel (1988). Since then, it has
been possible to transform HEM data into
a
(z
) sounding
curves. A color-coded representation of all
a
(z
) curves along
a ight line provides a resistivity-depth parasection (Sengpiel,
1990). Because apparent resistivity values are used in these
sections, only a smoothed image of the true resistivity pattern
is obtained.
Although 1-D inversion procedures based on a layered half-
space model have been used to interpret frequency-domain
HEM data for more than a decade (e.g., Holladay et al.,
1986; Paterson and Reford, 1986; Bergeron et al., 1987, 1989;
1983
1984 Sengpiel and Siemon
Best, 1990; Fraser, 1990; Hogg and Boustead, 1990; Kovacs
and Holladay, 1990; Won and Smith, 1991; Palacky et al.,
1992; Huang and Fraser, 1996; Qian et al., 1997; Fitterman
and Deszcz-Pan, 1998), examples of detailed resistivity cross-
sections rarely are presented in literature. BGR scientists
have developed a new 1-D inversion procedure (Fluche and
Sengpiel, 1997; Sengpiel and Siemon, 1998) that has been ap-
plied successfully to HEM survey data since 1993. Compared
with a
a
(z
) sounding curve, the 1-D inversion emphasizes

vertical resistivity contrasts. Particularly in groundwater explo-
ration surveys, this inversion procedure has become a very use-
ful tool for the interpretation of HEM data (compare Siemon
and Sengpiel, 1997).
In this paper, we present our recent inversion procedures
and show some inversion results based on synthetic and eld
data.
NEWTYPES OF MULTIFREQUENCY SOUNDING CURVES
Mundry (1984) has shown that the well-known integral de-
scribing the (normalized) secondary magnetic eld Z (Wait,
1982) measured with a horizontal coplanar coil system above
a homogeneous or layered half-space (z),
Z = r
3

0
R
o
( f, , (z))
2
e
2h
J
o
(r) d, (1)
can be simplied if the coil spacing r is smaller than the sen-
sor height h (r 0.3 h). In the Mundry integral, the Bessel
function J
o
is replaced by 1 and the coil spacing r is no longer
present in the integral (R
o
is the reection factor, f the fre-
quency, and the wavenumber). Physically, this is equivalent
to the superposed dipole condition, as postulated by Fraser
(1978).
For a uniform half-space, the Mundry integral depends only
on the ratio =h/p (Mundry, 1984), where p is the skin depth,
p =
2/
o
(2)
in a half-space, where =2 f and
o
=410
7
Vs/Am. This
leads to a straightforward inversion of measured data into the
half-space parameters apparent resistivity
a
and apparent dis-
tance D
a
of the coil system from the top of the conducting
half-space. The above-mentioned third half-space parameter,
namely the centroid depth z
, originally was dened as

z
= d
a
+ D
a
Re(C), (3)
using the real part of a complex transfer function C (Sengpiel,
1988) and the apparent depth d
a
= D
a
h (Fraser, 1978).
Siemon (1996) found that by inserting =
a
into equation (2),
the centroid depth can be calculated simply, as follows:
z
p
= d
a
+ p
a
/2, (4)
where p
a
is the apparent skin depth. (
a
, z
p
) denes a point
on a standard sounding curve
a
(z
p
) for each measured fre-
quency. The sounding curves
a
(z
) and
a
(z
p
) are compared
in Figure 1. Although the upper part of the model is approxi-
mated sufciently by both curves, only
a
(z
p
) shows the lower
conducting layer at the correct depth.
A number of sounding curves which are more sensitive to
vertical resistivity contrasts (called enhanced sounding curves
here) have been published by Siemon (1996). The basic ideas
were adopted from magnetotellurics and modied for dipole
induction. Two enhanced sounding curves,
a
(z
p
) and
NB
(z
s
),
are shown in Figure 1. The rst is derived from the ratio of the
quadrature and in-phase components of the HEMdata and the
true ight altitude h above the ground, leading to
a
=
a
(h/D
a
)
2
. (5)
The corresponding centroid depth is z
p
(equation 4). The sec-
ond curve is dened as
NB
=
a
1 +m
1 m
, m
= m
m +c
1 +c
,
m =
f
d
a
d f
(6)
where c =3 log(5) was obtained empirically from model calcu-
lations, and the corresponding centroid depth is
z
s
= d
a
+ p
a
/
2. (7)
Here, z
s
differs slightly from z
s
=
2z
p
, previously published
by Siemon (1996) and by Sengpiel and Siemon (1998). Huang
and Fraser (1996) presented a similar formulation, but they
differentiated a conductance curve with respect to an effective
depth z
ef f
(z
ef f
is a function of d
a
and p
a
), yielding a
(z
)
sounding curve (Figure 1). While their differentiation is based
on discrete
a
and z
ef f
values of two neighboring frequencies,
the differentiation in our case is conducted after a spline inter-
polation through the
a
( f ) values, leading to stable results.
The enhanced sounding curves are more sensitive to vertical
resistivity contrasts than the standard sounding curve
a
(z
p
).
The most sensitive one is
a
(z
p
), but this sounding curve has
two disadvantages. First, it overemphasizes the resistivity of
FIG. 1. Sounding curves for a layered half-space model and for
frequencies between 50 Hz and 300 000 Hz showing apparent
resistivities
a
versus centroid depths z
and z
p
, as well as three
enhanced sounding curves
a
(z
p
) and
NB
(z
s
), and, for com-
parison,
(z
), after Huang and Fraser (1996). The enhanced

sounding curves showa better t to the model resistivities than
the
a
sounding curves.
Inversion Methods for AEM Exploration 1985
very thick, homogeneous layers; i.e., the half-space below a
cover layer is not approximatedcorrectlyevenat extremelylow
frequencies. Second, it requires exact altitude measurements.
Sounding curves for a large number of frequencies are shown
in Figure 1. In practice, however, the number of available
frequencies is limited. Modern HEM systems operate at ve
frequencies, for example, the BGR system manufactured by
Geoterrex-Dighem. Are a few frequencies enough to outline
a conductor embedded in a resistive host rock?
FIG. 2. Apparent resistivity parasections derived from ve-frequency HEM data of 1-D models: (a) 51 models
with three layers (the thickness of the rst layer, d
1
, varies from 0 to 200 m), (b)
a
(z
p
) parasection, (c)
a
(z
p
)
parasection, and (d)
NB
(z
s
) parasection. The coil spacing for all systems is 6.7 m, and the sensor height is 30 m
throughout. The centroid depth values z
(z
p
or z
s
) for the frequencies f
1
f
5
are indicated by black dots.
Toanswer this question, wecalculatedtheHEMresponses of
1-D models with three layers at ve frequencies ( f
1
=375 Hz,
f
2
=1792 Hz, f
3
=8600 Hz, f
4
=41 000 Hz, f
5
=195 000 Hz).
In a second step, these data were inverted by using the
a
(z
p
),
a
(z
p
), and
NB
(z
s
) algorithms. In Figure 2a, 51 1-Dthree-layer
models are shown as color-coded resistivity columns. The mod-
els consist of a conducting layer (
2
=10 ohm-m, d
2
=20 m) in a
less-conducting host (
1
=
3
=100 ohm-m). The only parame-
ter varied in these 51 models is the thickness of the upper layer
(d
1
=0200 m), gradually shifting the conductor to greater
depth.
The resistivity columns of the
a
(z
p
),
a
(z
p
), and
NB
(z
s
) in-
versions are shown in Figure 2b through 2d. These columns are
derived from the apparent resistivity values
a
,
a
, and
NB
at
the corresponding centroid depths z
p
and z
s
(black dots) by in-
terpolation or extrapolation. It is evident from Figure 2b that
the three-layer model is approximated sufciently by
a
(z
p
)
only if d
1
is less than 60 m. For greater thicknesses of the
cover layer, a decrease in
a
with greater depth still is indi-
cated. This is valid even for extremely great thicknesses of the
cover layer, but there is no indication of an increase in resistiv-
ity below the conductor; i.e., the conductor seems to have no
FIG. 3. Apparent resistivitydepth parasections derived from ve-frequency HEM data of a dipping plate [500 m
(strike) 400 m (dip), conductance = 2 S, dip angle = 30
, depth of the upper edge = 1 m] in 100-ohm-m host:

(a) anomalous HEM data, (b)
a
(z
p
) parasection, (c)
a
(z
p
) parasection, and (d)
NB
(z
s
) parasection. The coil
spacing for all systems is 6.7 m, and the sensor height is 30 m throughout. The centroid depth values z
(z
p
or z
s
)
for the frequencies f
1
through f
5
are indicated by black dots.
lower boundary. The
a
(z
p
) inversion (Figure 2c) outlines the
conductor better than the standard
a
(z
p
) inversion. There is a
stronger decrease in resistivity, and the more resistive substra-
tum is indicated for d
1
<70 m. The best results are obtained
with the
NB
(z
s
) inversion. The conducting layer is outlined
clearly down to 80-m depth, and it is still visible at d
1
=200 m
because of the increase of z
s
with respect to z
p
[equations (4)
and (7)].
In Figure 3, the model is a dipping plate (500 m [strike]
400 m [dip], conductance = 2 S, dip angle = 30
, depth of the
upper edge = 1 m) in 100-ohm-m host rock. The correspond-
ing HEM responses (Figure 3a) are calculated by using a 3-D
thin-sheet program(Weidelt, 1981). In contrast to the 1-Dcase,
edge effects appear inthe 3-Dcase. Using a horizontal coplanar
coil system, a vertical plate or a horizontal cable causes sym-
metric double-peak HEM responses if the ight direction is
perpendicular to the strike of the body. For a dipping plate, the
HEM responses are asymmetric, with the greater peak above
the plate.
The
a
(z
p
),
a
(z
p
), and
NB
(z
s
) parasections are shown in
Figures 3b through 3d. The dip and depth of the conduct-
ing plate (black line) are approximated sufciently by all
inversions, except for great depths and in the vicinity of the
plate edges, where the 3-D effect dominates. Again, the best
results are achieved by using the
NB
(z
s
) algorithm, which out-
lines the limited thickness of the 3-Dbody down to 150-mplate
depth.
As shown above, buried conductors can be delineated in the
enhanced sounding curves better than in the
a
(z
p
) sounding
curves. These resistivity-depth parasections are calculated eas-
ily and rapidly. On the other hand, even the enhanced sounding
curves do smooth sharp resistivity contrasts, and they are not
very sensitive to resistive layers which are covered by a con-
ductor. It will be discussed in the following section whether
multilayer inversion can provide better results for the models
of Figures 2 and 3.
IMPROVING MULTILAYER INVERSION
A signicant improvement of HEM multifrequency sound-
ing was achieved by the development of our Marquardt-type
data inversion for a layered half-space model (Fluche and
Sengpiel, 1997; Sengpiel and Siemon, 1998). This procedure
requires an initial model with a given number of layers (N).
The layer parameters are changed gradually during the inver-
sion until an adequate t of the calculated data to the measured
HEMdata is obtained. Eachstarting model is derivedindividu-
allyfromasoundingcurvewhichis approximatedbyan N-layer
model in the depth range z
5
z
1
(centroid depth of the high-
est and lowest frequency). The necessary forward calculation is
very fast, because the Mundry integral (and its derivatives) can
be solved numerically by using a Laplace transform (Fluche,
1990), which is calculated more than 10 times faster than the
fast Hankel transform.
In our recent paper (Sengpiel and Siemon, 1998), we demon-
stratedthat the increase inthe number of frequencies used(ve
instead of three) for the 1-D inversion provides a better verti-
cal resolution of resistivity. In this paper, we compare standard
and enhanced sounding curves with 1-D inversion results, and
we discuss the inuence of the starting model on the inversion
result.
Our rst example is based on the 51 1-D models shown in
Figure 2a. The corresponding HEM data are inverted by using
different starting models derived from
a
(z
p
) sounding curves.
As shown in Figure 2b, the conducting layer is indicated down
to 60-m depth in the
a
(z
p
) parasection and thus in the start-
ing models for the 1-D inversion. Subdivision of the sounding
curve into a layered starting model is a very critical step in
the modeling process. An optimumstarting model should have
layer thicknesses on the order of those of the original model. In
practice, however, this information is not available, and the set-
ting of the layer boundaries is arbitrary. In our examples, we
subdivided the range between z
5
and z
1
into logarithmically
equidistant layers.
The results of a three-layer inversion are shown in Figure 4a.
The inversion is exact for d
1
<40 m. At depth d
1
40 m, the
conductive layer appears as the substratum with gradually in-
creasing resistivity toward depth. But the upper boundary of
the conducting layer is reproduced reasonably, even for depths
as great as 200 m.
If we increase the number of layers and thus decrease the
thicknesses of these layers in the starting model, the inversion
results are clearly better. In Figure 4b, in which a ve-layer
starting model was used, the conducting layer is reproduced
correctly down to 104 m.
Five-layer inversion results are shown in Figure 4c which
are based on starting models using
NB
(z
s
) sounding curves.
In Figure 4b through 4d, the indicated number of model layers
(ve or nine) is often not discernible in the parasections where
a three-layer model apparently prevails. The missing bound-
aries exist but do not appear because of very small resistivity
contrasts between adjacent model layers. Because the centroid
depth values z
s
are larger than z
p
[equations (4) and (7)], the
starting model is extended to greater depth and now yields
correct inversion results down to 172-m depth.
A nine-layer starting model, again based on a
NB
(z
s
)
sounding curve, provides good inversion results even down
to 184-m depth (Figure 4d). The 1-D inversion models now
appear smoother than the original 1-D models, but the con-
ducting layer is indicated clearly. In this case, the number of
model parameters (nine resistivities and eight thicknesses) is
greater than the number of measured values (ve frequencies
two EM components); i.e., underdetermined equation sys-
tems have to be solved. It is evident from Figure 4d that our
inversion procedure can handle this problem.
When synthetic HEM data are used, the relative rms error
of the inversion is negligible, except where the actual resistivity
distribution does not correspond to the 1-D model. Therefore,
the rms error is a useful measure of the applicability of the 1-D
inversion model.
In the following discussion, we examine the application of
1-D inversion to synthetic HEM data for the dipping plate
model in Figure 3. The results are shown in Figure 5. The
three-layer inversionusing starting models derivedfrom
a
(z
p
)
sounding curves (Figure 5b) is only a crude representation of
the original model. The rms error is on the order of 4%.
Again, better results are achievable if the number of layers
is increased and if the starting models are derived from
NB
(z
s
)
sounding curves (Figure 5c). The plate model, in this case, is
reproduced more adequately than before. The error q is re-
duced to less than 2%. A small error for a particular location,
however, is not an indicator that the true resistivity distribution
there is in fact 1-D (Ellis, 1998).
For such a 3-D model, the advantage of the multilayer in-
version results (Figure 5) over those of the resistivity-depth
parasections (Figure 3) is not as clear as it is for the 1-D mod-
els (Figures 2 and 4). For investigation of an unknown area,
it is best to calculate both. In spite of the remaining inaccura-
cies, the result can serve as a basis for an exploration drilling
program.
INVERSION OF SURVEY DATA
As mentioned above, several examples for 1-D inversion
results of survey data already have been presented by Sengpiel
and Siemon (1998). These inversion results are based on three-
frequency HEM responses.
Since 1998, BGR has used a ve-frequency HEM system
developedincooperationwithGeoterrex-Dighem. This system
uses only horizontal, coplanar coils. Internal calibration coils
provideabsolutecalibrationandphaseadjustment duringight
without interferencefromtheground. This modernequipment,
FIG. 4. 1-D inversion results derived from ve-frequency HEM data of 51 1-D models as indicated in Fig-
ure 2a: (a) three-layer inversions, starting models derived from
a
(z
p
) sounding curves, (b) ve-layer in-
versions, starting models derived from
a
(z
p
) sounding curves, (c) ve-layer inversions, starting models
derived from
NB
(z
s
) sounding curves, and (d) nine-layer inversions, starting models derived from
NB
(z
s
) sounding curves. The coil spacing for all systems is 6.7 m, and the sensor height is 30 m throughout.
combined with the described inversion procedures, provides a
wide range of new applications of airborne EM.
One of the rst inversionresults obtainedwiththe newHEM
system is shown in Figure 6. The HEM data (Figure 6a) for this
3-km-long north-southprole was collectedina landscape cha-
racterized by terminal moraines in northeastern Germany. The
a
(z
p
) and
NB
(z
s
) parasections are presented in Figures 6b
and 6c and the 1-D inversion results in Figure 6d. In all sec-
tions, the dotted line (bird in Figure 6) is the sensor elevation in
meters above mean sea level (m.s.l.) as derived from the baro-
metric altimeter measurements. The (apparent) ground eleva-
tion (solid line marked topo) is obtained from the difference
between the sensor elevation and the radar (or laser) altitude
of the bird. The ground elevation, however, may be erroneous
because of the presence of trees or buildings, for example.
In the apparent resistivitydepth parasections (Figures 6b
and 6c), the upper boundary of the colored area is dened by
FIG. 5. 1-D inversion results derived from ve-frequency HEM data of the dipping plate (compare Figure 3):
(a) anomalous HEM data, (b) three-layer inversions, starting models derived from
a
(z
p
) sounding curves, and
(c) six-layer inversions, starting models derived from
NB
(z
s
) sounding curves. The relative rms error q of the
inversion is plotted below the inversion results. The coil spacing for all systems is 6.7 m, and the sensor height is
30 m throughout.
the calculated apparent distance D
a
for the highest frequency
( f
5
=192, 424 Hz) below sensor elevation (line marked bird).
The lower boundary of the colored area is determined by ex-
trapolationbelowthe centroiddepth z
1
of the lowest frequency
( f
1
=383 Hz). All centroid depth values are marked by black
dots within the colored areas in Figures 6b and 6c.
The starting models for parasection6dwere derivedfromthe
NB
(z
s
) sounding curves at each of the 150 measuring points
of the prole, providing independently calculated inversion re-
sults. The layer thicknesses and color-coded resistivities of the
1-D inversion models (Figure 6d) are plotted downward from
the topo line. The rst (resistive) model layer includes trees
and resistive ground.
Because of the lack of boreholes along the prole, a denite
geologic interpretation of the resistivity-depth parasections is
not possible. However, boreholes a few kilometers north and
FIG. 6. Resistivity parasections and associated curves along a ight line in northeastern Germany: (a) measured
ve-frequency HEM data, (b)
a
(z
p
) parasection, (c)
NB
(z
s
) parasection, and (d) 1-D inversion results and
relative rms error q of the inversion. The coil spacing for all systems is about 6.7 m. The sensor height (bird) is
plotted (black dots) above each parasection. The color-coded inversion results are referred to the topography
(topo), including vegetation (forest). The centroid depth values z
(z
p
or z
s
) for the frequencies f
1
through f
5
are
indicated by black dots in the apparent resistivitydepth parasections b and c.
south of the prole show that the groundwater table lies at the
boundary between the rst and second layers (in Figure 6d).
Thus, the highly resistive cover layer (light blue) represents the
vegetation (forest) and dry sand and gravel. The groundwater
table does not appear in the apparent resistivitydepth para-
sections, but the fact that the apparent depth d
a
of the highest
frequency is positive indicates that the cover layer is more re-
sistive than the underlying one. The green and blue areas of
the resistivity-depth parasections can be interpreted as water-
saturated sand and gravel. It is supposed that thin silt layers
exist in that region, but location and depth are unknown. Al-
though there is no indication of thin conducting layers in the
a
(z
p
) parasection, the
NB
(z
s
) parasection and the 1-D inver-
sion results indicate such layers at 5060 m above m.s.l. There
are prominent layer boundaries at about 0 m and 60 m below
m.s.l. While the yellow layer is assumed to contain sand with a
relatively high silt content, the bottom layer (red) has an ele-
vated clay content. This strong decrease in resistivity is also
visible in the
NB
(z
s
) parasection, but it is not so clear in the
standard
a
(z
p
) parasection. In the central part of the prole,
the conductive layer appears to be more than 180 m below
ground level. The scattering of the parameter values for the
bottom layer probably is caused by its great depth.
CONCLUSIONS
The results of airborne EM surveys still are presented com-
monly in the form of apparent resistivity maps (for different
frequencies) and anomaly maps. The apparent resistivity map
is always useful, but it lacks lateral resolution (because of the
gridding), and it generally does not contain a depth reference.
Therefore, we overlay our apparent resistivity maps with a con-
tour plan of the centroid depth of the pertinent frequency.
Our main product, however, has become the resistivity-
depth parasection along a ight line containing the measured
EMdata, the birdaltitude, the topography, andthe color-coded
resistivity distribution. The bird altitude and the topography
should be related to sea level (measured by a barometric and
laser altimeter in the bird and a reference barometer at a base
station). The topography along the ight line is very useful
information for evaluation of the HEM results, especially for
groundwater investigation.
The standard sounding curve
a
(z
p
) and the enhanced
sounding curve
NB
(z
s
) do not need the (often distorted) sen-
sor altitude as aninput. This is anadvantage over the
a
(z
p
) and
(z
) sounding curves, which require the true sensor height

above ground.
Ingeneral, our resistivity-depthparasections result from1-D
multilayer inversion. This typeof inversionuses startingmodels
which are derived fromstandard or enhanced sounding curves.
It yields the best t to the true resistivity pattern if the ground is
not too complex or if a certain layer prevails. We have demon-
strated that a thin, dipping conductor can be traced better in
the ground if the number of layers (N) in the starting model
is increased, whereby N may be greater than the number of
induction frequencies.
A complex (2-D or 3-D) resistivity pattern often is repre-
sented better by the simpler inversion procedures based on
a uniform half-space model. In particular, vertical resistivity
boundaries are resolved much better in the approximate in-
version than in the 1-D inversion, which forces boundaries to
be horizontal.
The development of more sophisticated inversion proce-
dures (2-D, 3-D) should be accompanied by steady improve-
ment of survey data quality. This is because there are still
some error sources such as calibration errors, zero-level drift,
and mechanical instabilities of the coil arrangement which can
degrade the inversion results. With the new ve-frequency
EM system, which incorporates only coplanar coils and higher
dipole moments of the transmitters, as well as in-ight ampli-
tude calibration and phase adjustment, airborne EM explo-
ration has been brought forward a signicant step.
ACKNOWLEDGMENTS
The authors thank the BGR helicopter geophysics staff for
cooperation, and in particular, J. Pielawa for his assistance in
producing the gures. Furthermore, we thank the two review-
ers for their constructive comments, as well as the BGR trans-
lators R. C. NewcombandH. Toms for their thoroughchecking
of the English manuscript.
REFERENCES
BergeronJr., C. J., Ioup, J. W., andMichel II, G. A., 1987, Applicationof
MIM to inversion of synthetic AEM bathymetric data: Geophysics,
52, 794801.
1989, Interpretation of airborne electromagnetic data using the
modied image method: Geophysics, 54, 10231030.
Best, M. E., 1990, Synthetic modeling and airborne electromagnetic
interpretation: in Fitterman, D. V., Ed., Proceedings of the USGS
workshop on developments and applications on modern airborne
electromagnetic surveys, 1987, U.S. Geol. Surv. Bull., 1925, 2132.
Duckworth, K., Krebes, E. S., Juigalli, J., Rogozinski, A., and Calvert,
H. T., 1993: A coincident-coil frequency-domain electromagnetic
prospecting system: Can. J. Expl. Geophys., 29, 411418.
Ellis, R. G., 1998, Inversion of airborne electromagnetic data: Expl.
Geophys., 29, 121127.
Fitterman, D. V., and Deszcz-Pan, M., 1998, Helicopter EM mapping
of saltwater intrusion in Everglades National Park, Florida: Expl.
Geophys., 29, 240243.
Fluche, B., 1990, Verbesserte Verfahren zur L osung des direkten und
des inversen Problems in der Hubschrauber-Elektromagnetik: in
Haak, V., and Homilius, J., Eds., Protokoll Kolloquium Elektromag-
netische Tiefenforschung, 249266.
Fluche, B., and Sengpiel, K.-P., 1997, Grundlagen und Anwendungen
der Hubschrauber-Geophysik: in Beblo, M., Ed., Umweltgeophysik:
Ernst und Sohn, 363393.
Fraser, D. C., 1978, Resistivity mapping with an airborne multicoil
electromagnetic system: Geophysics, 43, 144172.
1990, Layered-earth resistivity mapping, in Fitterman, D. V.,
Ed., Developments and applications of modern airborne electro-
magnetic surveys: U. S. Geol. Surv. Bull., 1925, 3341.
Hogg, R. L. S., and Boustead, G. A., 1990, Estimation of overbur-
den thickness using helicopter electromagnetic data: in Fitterman,
D. V., Ed., Proceedings of the USGS workshop on developments
and applications on modern airborne electromagnetic surveys, 1987,
U. S. Geol. Surv. Bull., 1925, 103115.
Holladay, J. S., Valleau, N., and Morrison, E., 1986, Application of
multifrequency helicopter electromagnetic surveys to sea-ice thick-
ness and shallow-water bathymetry: in Airborne resistivity mapping,
Palacky, G. J., Ed., Geol. Surv. Can., Paper 86-22, 9198.
Huang, H., and Fraser, D. C., 1996, The differential parameter method
for multifrequency airborne resistivity mapping: Geophysics, 61,
100109.
Kovacs, A., andHolladay, J. S., 1990, Sea-icethickness measurement us-
ing a small airborne electromagnetic sounding system: Geophysics,
55, 13271337.
Mundry, E., 1984, On the interpretation of airborne electromagnetic
data for the two-layer case: Geophys. Prosp., 32, 336346.
Palacky, G. J., Holladay, J. S., and Walker, P., 1992, Inversion of
helicopter electromagnetic data along the Kapuskasing transect,
Ontario: in Current research, Part E, Geol. Surv. Can., Paper 92-
1E, 177184.
Paterson, N. R., and Reford, S. W., 1986, Inversion of airborne elec-
tromagnetic data for overburden mapping and groundwater explo-
ration: in Airborne resistivity mapping, Palacky, G. J., Ed., Geol.
Surv. Can., Paper 86-22, 3948.
Qian, W., Gamey, T. J., Holladay, J. S., Lewis, R., and Abernathy,
D., 1997, Inversion of airborne electromagnetic data using an Oc-
cam technique to resolve a variable number of multiple layers: in
Proceedings of high-resolution geophysics workshop, SAGEEP 97,
735744.
Sengpiel, K.-P., 1988, Approximateinversionof airborneEMdatafrom
a multi-layered ground: Geophys. Prosp., 36, 446459.
1990, Theoretical and practical aspects of ground-water ex-
ploration using airborne electromagnetic techniques: in Fitterman,
D. V., Ed., Proceedings of the USGS workshop on developments and
applications on modern airborne electromagnetic surveys, 1987, U.S.
Geol. Surv. Bull., 1925, 149154.
Sengpiel, K.-P., and Siemon, B., 1998, Examples of 1-D inversion of
multifrequency HEM data from 3-D resistivity distribution: Expl.
Geophys., 29, 133141.
Siemon, B., 1996, Neue Verfahren zur Berechnung von schein-
baren spezischen Widerst anden und Schwerpunktstiefen in
der Hubschrauberelektromagnetik: in Bahr, K., and Junge, A.,
Eds., Protokoll Kolloquium Elektromagnetische Tiefenforschung,
89100.
Siemon, B., and Sengpiel, K.-P., 1997, Helicopter-borne electromag-
netic groundwater exploration survey in the Kuiseb Dune area, Cen-
tral Namib Desert: in Proceedings of fth technical meeting of the
South African Geophysical Association, 137142.
Wait, J. R., 1982, Geo-Electromagnetism: Academic Press Inc.
Weidelt, P., 1981, Dipolinduktion in einer d unnen Platte mit leitf ahiger
Umgebung und Deckschicht: BGR report 89727.
Won, I. J., and Smith, K., 1991, Airborne electromagnetic bathymetry:
Geoexploration, 27, 297319.
The use of airborne gamma-ray spectrometryA case study
from the Mount Isa inlier, northwest Queensland, Australia
Prasantha Michael Jayawardhana
and S. N. Sheard
ABSTRACT
An airborne survey was undertaken on the Mount
Isa inlier in 19901992. During this survey, both air-
borne magnetic and gamma-ray spectrometric data were
recorded over 639 170 line-km. Because of perceived
value of the radiometric data, stringent calibration pro-
cedures, including the creation of a test range, were
adopted. In addition to the data from the newly-own
areas, 76 760 line-kmof existing data were acquired from
other companies, and were reprocessed and merged with
the Mount Isa survey. The total area covered by the
Mount Isa airborne survey was 151 300 km
2
.
Over the last ve years, several studies have been
undertaken that seek to exploit the Mount Isa region
gamma-ray database and maximise the use of radio-
metrics for mineral exploration. This paper highlights
the results of these studies by focussing on radiomet-
ric signatures of major mines in the Mount Isa Inlier,
radioelement contour maps, geomagnetic/radiometric
interpretation maps, lithological mapping, regolith map-
ping, geochemical sampling, and spatial modeling using
geographical information systems (GIS).
Due to the recent introduction of GIS technology and
better techniques for handling high quality digital data,
there has been a revived interest in making more use
of image data sets. The integration of raster and vector
data sets for both spectral and spatial modeling has max-
imized the potential of this approach.
INTRODUCTION
M.I.M Exploration Pty Ltd (MIMEX) has been using air-
borne radiometric techniques to assist in mineral exploration
since the late 1960s. The initial work was largely for uranium
exploration, but a gradual evolution to mapping using multi-
channel systems occurred. Unfortunately, the data quality of-
ten varied between surveys and companies, and thus the data
was compromised and too often not interpreted properly.
M.I.M. Exploration Pty Ltd, Level 2, Boundary Court, 55 Little Edward Street, Spring Hill, Brisbane, Queensland 4000, Australia. E-mail:
100253.132@ compuserve.com; snsheard@ibm.net.
In 1990, MIMEX embarked upon a major airborne geo-
physical program in northwest Queensland, Australia (Fig-
ure 1). A signicant component of this program was the
need to collect high-quality radiometric data which would as-
sist mapping and target generation in this highly prospective
region.
To maintain radiometric data quality throughout the survey
(which occurred over a 12-month span and used four different
aircraft), stringent survey specication and calibration proce-
dures were adopted (Figure 2). These specications were es-
tablished in conjunction with World Geoscience Corporation
Ltd (WGC), the contractors. The adherence to these standards
allowed the collection of 639 170 line-km of data converted to
elemental concentrations. This allowed accurate comparison
across areas using different acquisition aircraft and detector
systems. The paper gives a brief description of the collection
procedures and shows different processing and interpretation
techniques used to assist exploration in the region.
SURVEY PROCEDURES
The survey was largely owneast-west using 200-mline spac-
ing, a terrain clearance of 80 m, and a sampling interval of 60 m.
Areas where the interpreted depth to magnetic basement was
greater than 300400 m were own at 400-m line spacing. This
wider spacing was chosen due to the reduced prospectivity of
locating economic mineralisation at these depths.
The four spectrometers used were Geometrics Exploration
GR800B units with detector volumes of 33.56 liters. Two hun-
dred and fty-ve (255) channels of data were measured from
0.006 to 3.0 MeV, and one channel, the cosmic channel, from
3.0 to 6.0 MeV. All channels were recorded, but only the counts
in the energy ranges shown in Table 1 were used for data pro-
cessing and interpretation.
To maintain data quality the following routine quality con-
trol procedures were adopted.
1) Daily spectral plots were made using a standard cesium
source to monitor crystal tuning and drift.
2) Preight and postight ground calibration was done in
a preselected location. A thorium source, a uranium
1993
1994 Jayawardhana and Sheard
source, and background recordings were made. The tho-
rium and uranium source data were corrected for back-
ground and used to monitor spectral stability.
3) Weekly spectral drift and resolution checks of each
individual detector were made using cesiumand thorium
sources.
4) At the commencement of the survey at two-month
intervals thereafter, at times of crystal changes, and
at completion the following additional checks were
FIG. 1. Mount Isa inlier radiometrics survey location diagram.
FIG. 2. Sample calibration range prole (eld copy) showing ground/airborne and geochemical assaying along
the same ground prole.
Table 1. Spectrometer energy ranges used for interpretation.
Energy (MeV)
Name From To
Total count 0.402 3.000
Potassium-40 1.373 1.562
Uranium 1.668 1.858
Thorium 2.414 2.804
Cosmic 3.000 6.000
MIM Airborne Gamma Ray Spectrometry 1995
undertaken: pad calibration to determine stripping ratios
(Compton scattering), high-altitude stack to determine
background coefcients, and test range to determine
height attenuation coefcients and to determine sensi-
tivity coefcients. The pads used for calibration were
manufactured by Dr. R. L. Grasty of the Geological
Survey of Canada (Grasty et al., 1991). The test range
established for this survey allowed accurate conversion
of data to percent potassium, equivalent parts per million
uranium, and equivalent parts per million thorium. An
example of the results from an in-eld calibration run is
shown in Figure 2, where the concentrations computed
from surface geochemical sampling are plotted against
ground and airborne counts.
The rigorous implementation of the above procedures at
the time of survey 1990 has resulted in a data base unique
in Australia, which allows quantitative interpretation of the
worlds largest continuous commercial airborne survey at the
time. The collection of 256 channel data will allowreprocessing
of data if and when required.
DATA ENHANCEMENT AND INTERPRETATION
Radioelement contour maps
Radioelement contour maps over the survey area were gen-
erated by the contractors at 1:100 000 and 1:25 000 scales. The
radiometric data was corrected for system dead time, back-
ground radiation, channel interaction (i.e., stripping), height
attenuation, and system sensitivity, and then gridded using a
70-mcell size. These would have been different for each of the
four aircraft. System parallax was removed and microlevelling
was applied. Figure 3 shows part of the Quamby 1:100 000-scale
radioelement contour maps over the Dugald River deposit.
Contour maps were generated for total count (50 counts per
second contour interval), potassium (0.1% contour interval),
equivalent thorium (1.0 ppm contour interval) and equivalent
uranium (0.5 ppm contour interval). These maps are used to
aid in the visual interpretation of potential exploration areas.
Hard copies of these maps were produced at 1:100 000 and
1:25 000 scales for subsequent interpretation of the airborne
gamma-ray data.
Geomagnetic/radiometric interpretation maps
Fifty-three 1:100 000-scale geomagnetic/radiometric inter-
pretation maps were compiled fromradiometric and magnetic
contour maps. An example is shown in Figure 4. Geological
control was from Australian Geological Survey Organisation
(AGSO) maps published at 1:100 000 scale and localised maps.
The area shown in Figure 4 is approximately 20 km north-
west of Mount Isa and on the southern part of the Kennedy
Gap 1:100 000 scale map sheet. The large dome-shaped body
in the center is the Sybella batholith and can be distinguished
clearly in the radiometrics. The eastern and marginal parts of
the batholith are nonmagnetic and highly enriched in potas-
sium (K). The west and central part of the batholith shows
consistent weak, noisymagnetic patterns andvariableradioele-
ment response. Enrichment in uranium (U) and thorium (Th)
is alsocommon. These geomagnetic/radiometric interpretation
maps have been useful in rst-pass targeting.
Radiometric signatures for major deposits
in the Mount Isa inlier
Detailedvisual analysis of the airborne radiometrics was car-
ried out over 80 mineral deposits. An example fromthis compi-
lation is the Dugald River deposit situated 300 kmnorthwest of
Mount Isa on the Quamby 1:100 000-scale map sheet. Dugald
River is a zinc/lead/silver deposit with an estimated resource of
60 million tons at 10% zinc, 1% lead, and 30 g/ton silver. The
deposit type is a sheeted mineralisation of Proterozoic age.
The host rock is the Dugald River Shale Member containing
black, commonly carbonless shales. Mineralisation occurs in
ne-grained black slate with abundant suldes. The primary
mineralogy is pyrrhotite, sphalerite, pyrite, and galena. The
airborne radiometrics for this deposit (Figure 5) highlight
the presence of a 2-km north-south lineament centered about
the deposit and enriched in U and K. The large K high to the
west of the deposit is due to a Knapdale Quartzite ridge, which
is dominated by potassium feldspar. The magnetic data over
this deposit (Figure 6) shows a prominent north-south mag-
netic lineament, 6 kmlong andwitha peakamplitude of 660 nT,
associatedwitha pyrrhotite-enrichedstratigraphy surrounding
and including the mineralisation. This has been mapped as a
steep synform in the Corella Formation.
Lithological mapping
Radiometrics has been a useful tool in helping geologists
with the delineation of lithological boundaries and geologi-
cal mapping mismatches (Graham and Bonham-Carter, 1993;
Jaques et al., 1997). This interpretation procedure involves
overlaying of geological maps or traced geological boundaries
onto radiometrics and Landsat Thematic Mapper images to
highlight areas of variation and possible mismatch. While this
can be achieved using hard copies of maps and images, recent
improvements in computer hardware and software capabili-
ties and the introduction of geographical information system
(GIS) technology, has enabled the overlaying of data sets on
the computer screen. Figure 7 shows the published AGSO ge-
ological boundaries (white) overlain on a ternary image of the
gamma-ray data. The area shown in Figure 7 is approximately
70 kmnorthof Mount Isa andonthe northeasterncorner of the
Kennedy Gap 1:100 000 sheet. The dark areas (low K, Th, and
U) are part of the orthoquartzite component of the Leander
Quartzite. The bright area (high K, Th, and U) in the lower
left of Figure 7 differentiates the Gunpowder Creek Forma-
tion and, in particular, the micaceous siltstone and ferruginous
siltstone components.
In areas of good outcrop, the radiometric data is very good
at delineating the geology and is readily interpreted (Darnley
and Grasty, 1971). Interpretation of this data for lithological
mapping becomes more difcult in deep, transported soils. In
these areas, radiometrics have proven more useful for regolith
mapping and planning geochemical surveys.
Regolith mapping
The awareness of the value of regolith mapping in mineral
exploration is rapidly increasing (Wilford, 1995; Cook et al.,
1996; Wilford et al., 1997). By understanding the regolith, grea-
ter insight into the underlying geology can usually be attained
and, importantly, exploration techniques to explore below the
regolith can be devised. When the radiometric data are com-
bined with Landsat Thematic Mapper data, aerial photogra-
phy, and digital elevation models, a better understanding of
the regolith landforms can be attained. An example of the use
of the Mount Isa airborne radiometrics data set to identify
depositional areas is shown in Figure 8. The various regolith
boundaries are in gray and have been overlaid onto the ternary
image (K:Th:U=R:G:B). The lighter colored arcuate channel
in the top center of the gure highlights an area of alluvial clays
and silts within a depositional zone. By using the radiometrics,
these depositional zones are easily identied in similar regolith
settings elsewhere (Wilford et al., 1997).
FIG. 3. Radioelement contour maps over the Dugald River deposit (black circle). (a) Total count with a contour interval of 50
counts per second. (b) Potassium with a contour interval of 0.1%. (c) Thorium with a contour interval of 1.0 ppm. (d) Uranium
with a contour interval of 0.5 ppm.
Geochemical sampling
Radiometric data has been recognised (but too often ig-
nored) as useful in planning exploration geochemical surveys
and in interpreting previous ill-constrained surveys. By recog-
nising the context of regolith landformsdepositional, ero-
sional, or residualthe sample site signicance can be better
used in the interpretation (Dickson and Scott, 1997).
Spatial modeling using GIS
The introduction of GIS technology has enabled the devel-
opment of semiautomated spatial analysis models to aid in the
interpretation and analysis of large quantities of the airborne
radiometric data within a reasonable time frame. The tech-
niques described below were developed as part of the data
synthesis, regional interpretation, and targeting for the Mount
Isa block and involved MIMEXstaff and B. Dickson of CSIRO
(Exploration and Mining).
The aimwas to identify anomalous radiometric (enrichment
or depletion) responses within lithological units (B. Dickson,
personal communication, 1994). Figure9shows theradiometric
normalisation methodology adopted. The technique required
the development of a GIS-based model to analyse and nor-
malise the radiometric data withineachlithological unit. Litho-
logical boundaries were obtained from the AGSO 1:100 000-
scale published maps.
The GIS model generated normalised images for a given
1:100 000-scale map sheet. The radioelement statistics for each
lithological unit were also automatically generated. These in-
cluded gcode (unique number for each lithological unit), min-
imum, maximum, mean and standard deviation (std). Once
normalised images were created, standard processing was car-
ried out to stretch the images to highlight areas of enrichment
(3 std) and depletion (2 std). Different cutoffs were cho-
sen for enrichment and depletion as the normalised distribu-
tion showed a negative skew. The results were saved as a new
nine-band image: bands 13 contain K, Th, and U from the
original image; bands 46 show areas of enriched K, Th, and
U; and bands 79 showareas of depleted K, Th, and U. The re-
sulting nine-band image was taken back to the GIS for second
pass processing. In the second pass processing, the enrichment
and depletion areas from pass one were masked out of the ra-
diometrics and the normalisation rerun to highlight the subtle
highs and lows. The results from the second pass were com-
bined with the rst pass to create radiometric enrichment and
depletion maps (Figures 10 and 11). Figure 10 shows an ex-
ample of part of the Mount Isa 1:100 000-scale radioelement
enrichment map around the Mount Isa deposit.
FIG. 4. An extract from the southern part of the Kennedy Gap 1:100 000-scale geomag-
netic/radiometric interpretation map.
The Mount Isa deposit is a lead/zinc/silver/copper deposit in
the Mount Isa inlier mineral province. The estimated resource
is 150 million tons at 7% zinc, 56% lead, and 140 g/ton sil-
ver, and 255 million tons at 3.3%copper. The deposit types are
FIG. 5. The ternary gamma-ray spectrometric (K=red, Th =
green, U=blue) image (K:Th:U=R:G:B) over the Dugald
River deposit (white circle).
stratabound for lead/zinc and discrete tabular bodies for cop-
per. The host rock for the lead/zinc and copper orebodies is
the Urquhart Shale, which contains dolomitic and variable car-
bonaceous siltstone rich in ne-grained pyrite and numerous
tuff beds. The copper orebody has a silica dolomite alteration
halo. In Figure 10, the lithological boundaries are in gray and
the ternary image (K:Th:U=R:G:B) highlights areas of K, Th,
and U enrichment.
FIG. 6. Pseudocolor image of total magnetic intensity for the
Dugald River deposit.
FIG. 7. Radiometric colour image (K:Th:U=R:G:B) overlaid
with geological boundaries (white) from the northeastern cor-
ner of the Kennedy Gap 1:100 000-scale published mapping.
FIG. 8. Radiometrics ternary image (K:Th:U=R:G:B) with re-
golith boundaries overlaid in gray. The lighter colored arcuate
channel running across the top center of the gure is an area
of alluvial clays and silts within a depositional prole.
FIG. 9. Radiometric normalisation methodology.
An interesting nding from this technique is the large area
of K enrichment to the south of Mount Isa. This occurs on the
Native Bee Siltstone unit, which is stratigraphically above the
Urquhart Shales and composed of bedded dolomitic siltstone,
laminated siltstone, minor tuff, and chert. Field assessment us-
ing a handheld spectrometer conrmed this area as an area of
high K alteration. This could be interpreted as a possible uid
channel for the ore-bearing uids of the Mount Isa deposit.
The technique has also highlighted lithological boundaries.
For example, the area of Th enrichment on the western edge of
FIG. 10. Anextract fromtheMount Isa1:100 000-scaleradioelement enrichment map. Thecenter of theMount Isa
deposit is shownby a blackcircle. The lithological boundaries are ingray andthe ternary image (K:Th:U=R:G:B)
highlights areas of K, Th, and U enrichment.
FIG. 11. An extract fromthe Mount Isa 1:100 000-scale radioelement depletion map. The lithological boundaries
are in gray, roads in gold, and the ternary image (K:Th:U=R:G:B) highlights areas of K, Th, and U depletion.
Figure 10 was due to a mapped geological boundary mismatch
betweenthe Sybella Granite andCainozoic cover inthe AGSO
database.
Figure 11 shows an example of part of the Mount Isa
1:100 000-scale radiometric depletion map around the Mount
Isa deposit. The lithological boundaries are in gray, roads are in
gold, and the ternary image (K:Th:U=R:G:B) highlights areas
of K, Th, and Udepletion. The depletion results identify a large
area of Th depletion coincident with the surface projection of
the lead/zinc orebody. The Th depletion is mainly within the
Urquhart Shale and could not be due to geological boundary
effects. This area is covered by mining infrastructure, and the
source which could be geological or manmade has thus far not
been identied.
Another example of geological mismatch is the sliver of
K depletion to the west of Mount Isa. This resulted from a
lithological boundary mismatch between the Eastern Creek
Volcanics and the Mount Guide Quartzite.
This semiautomated spatial modeling technique has resulted
in dening areas of enrichment/depletion and thus prospective
for mineralisation. It has also assisted in more precise geologi-
cal mapping of areas considered previously to be well mapped.
CONCLUSION
Each of the products and techniques described have pro-
vided valuable information to assist mineral exploration in the
complex geological environment of the Mount Isa block. The
overall compilationof the radiometric signatures for the 80 ma-
jor mines inthe Mount Isa inlier was boundintoa three-volume
set for internal company use. Hard copies of the radioelement
contour maps at 1:100 000 and 1:25 000 scale forma convenient
supplement wheninterpreting the airborne radiometric survey.
The geomagnetic/radiometric interpretation maps have been
useful in rst-pass targeting, as have the Mount Isa airborne
survey data set in helping geologists with the delineation of
lithological boundaries and identication of geological map-
ping mismatches. When the radiometric data are combined
with Landsat Thematic Mapper, aerial photography, and dig-
ital elevation models, a better understanding of the regolith
landforms can be attained. In terms of geochemical sampling,
the radiometric data have been recognised (but too often ig-
nored) as useful in planning exploration geochemical surveys
and in interpreting previous ill-constrained surveys. GIS anal-
ysis has enabled the development of a semiautomated spatial
modeling technique for dening areas of enrichment/depletion
and hence for generation of new prospects for mineralisation.
It has also assisted in more precise geological mapping of areas
considered previously to be well mapped.
By adopting internationally established airborne radiomet-
ric survey standards, the large airborne radiometric survey is of
a consistent high quality and is timeless in its value for mineral
exploration.
REFERENCES
Cook, S. E., Corner, R. J., Groves, R. J. and Grealish, G., 1996, Applica-
tion of airborne gamma radiometric data for soil mapping: Austral.
J. Soil Research, 43, 183194.
Darnley, A. G., and Grasty, R. L., 1971, Mapping from the air by
gamma-ray spectrometry: Proc. 3rd Internat. Geochemical Symp.:
Canadian Inst. Mining and Metallurgy Special Volume 11, 485500.
Dickson, B. L., and Scott, K. M., 1997, Interpretation of aerial gamma-
ray surveysAdding the geochemical factors: AGSO J. Austral.
Geol. Geophys., 17, 187200.
Graham, D. F., and Bonham-Carter, G. F., 1993, Airborne radio-
metric data: a tool for reconnaissance geological mapping using a
GIS: Photogrammetric Engineering and Remote Sensing, 59, 1243
1249.
Grasty, R. L., Holman, P. B., and Blanchard, Y. B., 1991, Transportable
calibration pads for ground and airborne gamma-ray spectrometers:
Geological Survey of Canada Paper 90-23.
Jaques, A. L., Wellman, P., Whitaker, A., and Wyborn, D., 1997,
High-resolution geophysics in modern geological mapping: AGSO
J. Austral. Geol. Geophys., 17, 159173.
Wilford, J. R., 1995, Airborne gamma-ray spectrometry as a tool for
assessing relative landscape activity and weathering development of
regolith, including soils: AGSO Research Newsletter, 22, 1214.
Wilford, J. R., Bierwirth, P. N., andCraig, M. A. 1997, Applicationof air-
borne gamma-ray spectrometry in soil/regolith mapping and applied
geomorphology: AGSO J. Austral. Geol. Geophys., 17, 201216.
The detection of potassic alteration by gamma-ray
spectrometryRecognition of alteration related
to mineralization
Robert B. K. Shives
, B. W. Charbonneau
, and K. L. Ford
ABSTRACT
Canadian case histories document the use of airborne
and ground gamma-ray spectrometry to detect and map
potassium alteration associated with different styles
of mineralization. These include: volcanic-hosted mas-
sive suldes (Cu-Pb-Zn), Pilleys Island, Newfoundland;
polymetallic, magmatic-hydrothermal deposits (Au-Co-
Cu-Bi-W-As), LouLake, Northwest Territories; andpor-
phyry Cu-Au-(Mo) deposits at Mt. Milligan, British
Columbia and Casino, Yukon Territory. Mineralization
in two of these areas was discovered using airborne
gamma-ray spectrometry.
In each case history, alteration produces potas-
sium anomalies that can be distinguished from nor-
mal lithologic potassium variations by characteristic
lows in eTh/K ratios. Interpretations incorporating air-
borne and ground spectrometry, surcial and bedrock
geochemistry and petrology show that gamma-ray spec-
trometric patterns provide powerful guides to mineral-
ization. This information complements magnetic, elec-
tromagnetic, geological, and conventional geochemical
data commonly gathered during mineral exploration
programs.
INTRODUCTION
In Canada, many studies (see Shives et al., 1995, and refer-
ences therein) document application of gamma-ray spectrom-
etry to surcial (glacio-uvial deposits, tills, soils) and bedrock
mapping at regional to deposit scales, exploration for a wide
variety of commodities (rare, base, and precious metals, gra-
nophile elements, and industrial minerals) and environmental
studies. In Australia and elsewhere, where the effects of trop-
ical weathering and landform development may signicantly
Manuscript received by the Editor April 12, 1999; revised manuscript received June 19, 2000.
Geological Survey of Canada, Mineral Resources Division/Airborne Geophysics Section, 601 Booth Street, Ottawa, Ontario K1A 0E8, Canada.
E-mail: rshives@nrcan.gc.ca; bcharbo@nrcan.gc.ca; kford@nrcan.gc.ca.
modify bedrock radioactive element distribution (Dickson and
Scott, 1997), the patterns from airborne gamma-ray surveys
provide important information for soil, regolith, and geomor-
phology studies used for land management and mineral explo-
ration strategies (Wilford et al., 1997).
The use of gamma-ray spectrometry to determine concen-
trations of elemental potassium, regardless of the associated
potassium mineral species, enables alteration mapping in a
wide range of geological settings. For example, potassic al-
teration in the form of sericite is commonly associated with
many types of volcanic-associated massive-sulde base-metal
and gold deposits (Franklin, 1996; Poulsen and Hannington,
1996). Potassiumfeldspar alteration has been documented as a
regional alteration product at volcanic-associated base-metal
deposits in the Bergelagen district, Sweden (Lagerblad and
Gorbatschev, 1985), in the Mount Read volcanics, Tasmania,
Australia(Crawfordet al., 1992), andat theQueRiver massive-
sulde deposit, Tasmania, Australia (Ofer and Whitford,
1992). Potassium alteration is common in shear-hosted gold
deposits, such as those at Hemlo, Ontario, Canada (Kuhns,
1986) and Red Lake, Ontario, Canada, (Durocher, 1983), and
of many other deposit types (Hoover and Pierce, 1990).
Many alkaline and calc-alkaline porphyry Au-Cu (Mo)
deposits have extensive potassic hydrothermal alteration ha-
los (Davis and Guilbert, 1973; Schroeter, 1995) which vary
mineralogically, both laterally and vertically, with changes in
pressure, temperature, eH, and pH during magmatic, hypo-
gene, and subsequent supergene processes. A well-established
alteration-mineralization potassic zoning sequence (Lowell
and Guilbert, 1970) may be evident within a single de-
posit, ranging from a central, orthoclase and/or biotite core
( sericite as fracture controlled and pervasive replacements)
outwards through successive phyllic (sericitic), argillic, and
propylitic zones. Although phyllic zones may contain less bulk
potassium gain than potassic cores, their peripheral distribu-
tion commonly offers much larger targets for detection by
gamma-ray spectrometry.
2001
2002 Shives et al.
As thoriumenrichment generallydoes not accompanypotas-
sium during hydrothermal alteration processes, eTh/K ratios
provide excellent distinction between potassium associated
with alteration and anomalies related to normal lithological
variations (Galbraith and Saunders, 1983). This important cor-
relation of low eTh/K ratio with alteration is evident in count-
less studies worldwide, including the four Canadian examples
described below (Figure 1).
VOLCANIC-HOSTED MASSIVE SULFIDESPILLEYS
ISLAND, NEWFOUNDLAND
On the Island of Newfoundland, the Geological Survey
of Canada has collected more than 65 000 km of combined
gamma-ray spectrometric, total-eld magnetic, and very low
frequency electromagnetic (VLF-EM) airborne data at 1000-m
line spacing covering approximately 50% of the island.
Within the major volcanic belts, a number of potassium
anomalies are directly associated with felsic volcanic units
andpast-producingor prospectivevolcanic-hostedmassivesul-
des (VHMS) alteration systems. Figure 2b shows the potas-
sium distribution map for the western part of Notre Dame
Bay. Corresponding geology for the area is shown in Fig-
ure 3a (Colman-Sadd et al., 1990). High potassium concen-
trations west of Green Bay correspond to felsic intrusive units
within the Siluro-Devonian, post-tectonic Kings Point Com-
plex, which comprises units ranging fromgabbro to syenite and
peralkaline granite. The prominent area of low K values north
of Halls Bay accurately reects the distributionof primitive-arc
ophiolitic mac volcanic rocks. Southwest of Halls Bay, a broad
region of moderate-to-high potassium values corresponds to
felsic volcanic units of the Silurian Springdale Group (Coyle
and Strong, 1987).
The northern part of the Roberts Arm Group (RAG,
Figure 3a) contains lower tholeiitic and upper calc-alkaline
suites of mac volcanics with localised felsic volcanic centers
(Bostock, 1988). Figure 2b shows that small potassium anoma-
lies overlie these felsic centers. Ground follow-up studies on
southernPilleys Island(Ford, 1993), including groundgamma-
ray spectrometry, geological mapping, and litho-geochemical
and petrological investigations, have shown that two of the
strongest potassium anomalies (Figure 2c) are associated with
FIG. 1. Location of examples discussed in this paper.
hydrothermally altered felsic pyroclastics and massive dacitic
ows (Figure 3b; Tuach et al., 1991). Potassium anomalies at
Bumble Bee Bight, Manseld Showing, and the 3B deposit are
associated with narrow zones of hydrothermally altered pil-
low basalts, dacitic ows, and pyroclastics in the hanging walls.
The large range in K values (Figure 4d) reects sporadic dis-
tribution of potassium enrichment. Maximum potassium con-
centrations at Bumble Bee Bight represent a two-to three-fold
increase over the average potassium values (3.2% K 1.0%;
Figure 4a) for equivalent unalteredtoweakly-alteredfelsic vol-
canic sequences within the Roberts Arm Group. Potassium
enrichment at Spencers Dock (Figure 4b) and the Old Mine
area (Figure 4c) shows similar variability with maximum con-
centrations of approximately 12% K.
Lithogeochemical alteration indices (Figure 4f) dened
by [(K
2
O+MgO) 100]/(K
2
O+MgO+CaO+Na
2
O) gen-
erally range between 70 and 98 for samples containing greater
than 7%K, and between 30 and 60 for less altered samples con-
taining 26% K. Maximum potassium concentrations are not
associated with widespread sericitic alteration, but are asso-
ciated with an unusual ne-grained K-feldspar alteration that
Thurlow (1996) described as volumetrically equivalent to the
sericitic alteration. In the Spencers Dock area, Thurlow de-
scribed felsic volcanic units affected by this alteration as un-
remarkable, light grey-green, pyrite-free, hard, dacitic lavas
having a weakly silicied aspect. K-feldspar alteration in the
hangingwall of theOldMine, Spencers Dock, andBumbleBee
Bight areas is overprinted by sericitic microfractures, suggest-
ing that the K-feldspar alteration may precede the pervasive
sericitization and massive sulde mineralization (Santaguida
et al., 1992). Notably, no potassium enrichment is associated
with barren pyritic gossans hosted by pillow basalts on Sunday
Cove Island (Figure 4e).
Althoughthe areas of intense K-feldspar enrichment are less
directly associated with VHMS mineralization than sericitic
and chloritic alteration, the proximity of potassium anomalies
(Figure 2c) to mineralization provides evidence of signicant
uid/rock interaction associated with the mineralizing event.
The ability to map and quantify this potassium alteration from
theair andonthegroundhas important implications for VHMS
exploration and mapping.
POLYMETALLIC MAGMATIC-HYDROTHERMAL DEPOSITS
(Au-Co-Cu-Bi-W-As), LOU LAKE, NORTHWEST
TERRITORIES
In 1974, the magnetite breccia/rhyodacite ignimbritehosted
Sue-Dianne Cu-U-Au deposit (to date, 8 million tons grading
0.8% Cu, open to depth and along strike) was discovered as a
result of eU and eU/eTh anomalies on a gamma-ray spectro-
metric survey with 5-km line spacing (Richardson et al., 1974;
Charbonneau, 1988). The deposit is located 20 km north of the
Lou Lake area (Figure 1) in the southern part of the 18701840
Ma, Proterozoic volcano-plutonic Great Bear magmatic zone
(GBmz) in northern Canada. Subsequent metallogenic studies
(Gandhi, 1994) documented similarities between Sue-Dianne
and the giant Olympic Dam polymetallic deposit in Australia,
and promoted the exploration potential of the entire magmatic
zone.
In 1993, the Geological Survey of Canada (GSC), encour-
aged by the success of the Sue-Diane discovery, surveyed a
selected part of the southern GBmz using 500-m line spacing,
Detecting Ore Using GRS and K Alteration 2003
combining gamma-ray spectrometric, magnetic total eld, and
VLF-EM sensors in a Skyvan xed-wing aircraft. Results were
published in 1994 in digital format and as a bound booklet
containing twelve 1:100 000scale color interval maps (K, eU,
FIG. 2. Potassium maps compiled from airborne gamma-ray spectrometer surveys own using 1-km line spacing: (a) Island of
Newfoundland (LMIS = Lake Michael Intrusive Suite, DBG = Deadmans Bay Granite, AG = Ackley Granite, FG=Francois
Granite, NBGS = North Bay Granite Suite), (b) Notre Dame Bay area, (c) Pilleys Island.
eTh, eU/eTh, eU/K, eTh/K, total count, ternary radiometric,
magnetic total eld, calculated magnetic vertical gradient,
VLF-EM total eld, and quadrature), stacked proles,
and a color geology map with mineral occurrences (Hetu
2004 Shives et al.
et al., 1994). This more detailed survey provided a regional
framework for ongoing geological mapping and metallogenic
studies, and delineated several new and existing exploration
targets. The survey clearly showed the previously unrealized
FIG. 3. Simpliedgeology, correspondingtoairbornepotassiummaps inFigures 2b, c: (a) NotreDameBayarea, showingdistribution
of major Cambrian to Ordovician volcanic rocks, derived from Colman-Sadd et al. (1990); (b) general geology, southern Pilleys
Island, derived from Tuach et al. (1991).
signicance of several known, small, scattered mineral oc-
currences at Lou Lake (Figure 1), placing them within the
context of a large, potassium and iron-enriched, polymetallic
(Au-Co-Cu-Bi-W-As) hydrothermal system. Publicationof the
FIG. 4. Variations in equivalent thorium and potassium concentrations as measured by in-situ gamma ray spectrometry: (AE)
unaltered and variably K-altered mac and felsic volcanics of the Roberts Arm Group, (F) variation of lithogeochemical alteration
index with increasing potassium concentrations as measured by in-situ gamma ray spectrometry.
2006 Shives et al.
survey prompted extensive new exploration activity through-
out the GBmz.
Airborne radioactive element and magnetic signatures of
the Lou Lake area are shown in Figure 5 and in prole format
in Figure 6. The broad potassium anomaly (Figure 5a) covers
FIG. 5. Airborne geophysical patterns (AD) and geology (EF) for the Lou Lake area, Northwest Territories.
a 3 4 km area and contains contoured values which locally
exceed 7% K southeast of Lou Lake. Corresponding prole
data(Figure6) across thesameareaarenot subject toaveraging
inherent in the contouring process, resulting in values which
exceed 8% K.
The potassium anomaly is coincident with a high mag-
netic total-eld anomaly, which has a peak intensity exceed-
ing 2000 nT (Figure 5d). Polymetallic mineralization occurs
where coincident potassiumand magnetic intensities are great-
est. The southeast trending axis of the magnetic anomaly par-
allels a belt of lower Proterozoic metasedimentary rocks con-
taining synsedimentary, stratiform magnetite beds and lenses
and later, hydrothermal polymetallic magnetite veins and dis-
seminations.
The strong potassium anomaly is characterized by eTh/K ra-
tios of less than 2.5 10
4
(Figure 5b). In general, low eTh/K
ratios are excellent indicators of potassium alteration (Shives
et al., 1995). Unaltered lithologies typically reect the normal
ratio of crustal abundances of K and Th, of approximately
5 10
4
(Galbraith and Saunders, 1983). During the process
of potassium alteration, however, thorium does not usually ac-
company potassium. The resulting loweTh/Kratio, as observed
at Lou Lake, thus enables distinction of potassium anomalies
that have exploration signicance from those related solely to
lithological variations.
Uraniumenrichment, evident on the eU/eTh ratio map (Fig-
ure 5c), is peripheral to the potassium anomaly and relates
to numerous small pitchblende veins. This mineralization may
represent lateral movement of uranium away from the hy-
drothermal center of the system.
Ground spectrometry across the mineralised zones at Lou
Lake (N. Prasad, Geological Survey of Canada, personal com-
munication, 1995; Gandhi et al., 1996) measured potassium
concentrations as high as 15% K in K-feldspar altered rhy-
olitic units and up to 6% K in biotite-altered metasediments
(Figure 7). Unaltered equivalents of these rocks contain less
than 4% K.
In addition, company reports indicate a 3-mgal gravity
anomaly is coincident with the potassium and magnetic air-
borne anomalies, and with resistivity lows that outline the
mineralized zones. Currently, in-pit mineable reserves for the
Bowl zone include 42.145 million tons grading 0.1% Co,
0.5 grams/ton Au, and 0.12% Bi. The inferred geological
resource is more than 100 million tons (R. Goad, Fortune
Minerals Ltd., personal communication, 2000).
FIG. 6. Airborne geophysical stacked prole across Lou Lake
area, Northwest Territories (line 16, position indicated in
Figure 5).
Based on geological mapping and metallogenic studies, the
numerous polymetallic mineral occurrences at Lou Lake are
interpreted as hydrothermal, related to a deep-seated gran-
ite pluton (Figure 8). It was suggested (Gandhi et al., 1996)
that the mineralizing solutions moved upwards through metal-
rich argillaceous metasedimentary beds, scavenged the metals
and redepositing them at the unconformity with the overlying
Lou Lake volcanics. The discovery of the large potassium and
iron enrichment zone by the airborne spectrometric and mag-
netic survey further emphasizes the hydrothermal character
and large size of the mineralizing system. This example under-
scores the importance of combining geophysical/geochemical
informationprovidedbyregional airbornesurveys withground
studies, good geological control, and metallogenic models to
identify and delimit mineralization.
PORPHYRY Au-Cu-(Mo) DEPOSITSMT. MILLIGAN,
BRITISH COLUMBIA, AND CASINO, YUKON TERRITORY
In the Canadian Cordillera, exploration for porphyry cop-
per and molybdenum boomed in the 1960s and 1970s, respec-
tively, when several successful mines were discovered. Dur-
ing those periods, some researchers used ground gamma-ray
spectrometry to measure potassium enrichment in porphyry
deposits (Moxham et al., 1965; Davis and Guilbert, 1973; Port-
nov, 1987), and a few enlightened explorationists attempted to
use the technique to detect mineralizing intrusions or associ-
ated alteration. Unfortunately, despite the research successes,
some of these early exploration tests were seen as failures, per-
haps due to some combination of inferior instrumentation, in-
correct techniques, or inappropriate application to alkali-poor
(low-potassium) targets. Other geophysical techniques, such as
magnetic and induced polarization surveys, proved far more
successful and became well established, while spectrometry
was virtually forgotten.
During the late 1980s, when exploration turned towards
gold-bearing alkalic porphyry systems, several deposits were
FIG. 7. Potassium versus thorium concentrations determined
by in-situ gamma-ray spectrometry, Lou Lake area, Northwest
Territories.
2008 Shives et al.
reevaluated for their gold potential, and new discoveries were
made, including those at Mt. Milligan, British Columbia, and
Casino, Yukon Territory (Figure 1). Despite the obvious potas-
sium enrichment associated with these alkalic systems, ground
gamma-ray spectrometers generally remained dust covered in
company storage rooms. Field crews relied on qualitative stain-
ing of rock slabs and drill core, or sparse whole-rock geochem-
ical analyses to map potassic alteration.
In 1990, authors Shives and Ford conducted brief ground
spectrometric surveys over the Mt. Milligan and Mt. Polley
(250 km south of Mt. Milligan) alkalic porphyry copper-gold
deposits in British Columbia to reestablish the applicability
of the technique to porphyry deposit exploration. The re-
sults, which indicated strong potassium variations (Figure 9),
FIG. 8. Geological cross-section along line A-B (see Figure 5 for section location and geological legend). Note
K and Fe enrichment halo associated with hydrothermal Au-Co-Cu-Bi-W-As mineralization. Modied from
Gandhi et al. (1996).
FIG. 9. Potassium versus thorium concentrations determined by in-situ gamma-ray spectrometry over deposits at Mt. Milligan (a)
and Mount Polley (b) (approximately 250 km to southwest) in British Columbia. Note progressive K enrichment, with maximum
values associated with mineralization.
demonstrated the ability of the ground technique to detect
potassium alteration associated with the deposits and to dis-
tinguish various related intrusive and extrusive lithologies.
This work encouraged subsequent GSC xed-wing and con-
tracted helicopter- borne surveys over several porphyry sys-
tems throughout the Cordillera. Twoexamples are summarized
below.
Mt. Milligan deposit area
The large low-grade Mt. Milligan porphyry copper-gold de-
posits occur within the lower Mesozoic Quesnel terrane in
central British Columbia. The deposits are associated with
lower to middle Jurassic, alkalic, porphyritic monzonite stocks
which intrude latitic, andesitic to high-potassium basaltic and
trachytic volcanic rocks of the Witch Lake Formation (WL in
Figures 10a, b; Nelson et al., 1991).
Mineralization occurs in several zones (combined resource
299 million tons grading 0.45 grams/ton Au, 0.22% Cu) as
pyrite, chalcopyrite, magnetite, bornite, molybdenite, and gold.
A detailed description of the exploration history, geology, al-
FIG. 10. Airborne geophysical data compiled from surveys using 500-m line spacing over the Mt. Milligan deposit area, British
Columbia (AC) and Casino deposit area, Yukon Territory (DF). Mt. Milligan area bedrock geology after Nelson et al. (1991);
surcial geology after Kerr (1991).
teration, and mineral zoning of the deposits is provided by
Sketchley et al. (1995).
In 1991, the GSC conducted the rst public domain, high-
sensitivity gamma-ray spectrometric survey own in British
Columbia, combined with magnetic total-eld measurements,
over the Mt. Milligan deposit area (Geological Survey of
Canada, 1992). Magnetic total-eld and potassium maps for
2010 Shives et al.
part of this survey, own with 500-m line spacing, are shown
in Figure 10. A broad, regional magnetic high is associated
with exposed and buried portions of the Mt. Milligan Intrusive
Complex (Figure 10a). High potassium concentrations are as-
sociatedwithbedrockexposures of the complex at Mt. Milligan
(Figure 10b). To the south, discrete K anomalies associated
with the Mt. Milligan deposits provide smaller, better-focused
exploration targets relative to the regional magnetic signature.
Elsewhere, despite generally few outcrop exposures and ex-
tensive, often thick overburden (till, glaciouvial, and collu-
vial deposits locally exceeding 100 m), potassium anomalies
overlie several previously known and newmineral showings or
prospects, providing signicant exploration vectoring. Along
the eastern margin of the survey, coincident uranium, thorium,
and potassium concentrations characterize rocks within the
Wolverine Metamorphic Complex (WMC in Figure 10b).
Based on detailed ground spectrometry and correlation of
the airborne survey with regional surcial geological mapping
(Kerr, 1991), a threshold value of 1.2%potassiumtypically dis-
tinguishes higher concentrations associated with glaciouvial
deposits from lower values over glacial tills (Figure 10c). This
offers substantial aid to ongoing surcial mapping. Although
all potassium anomalies throughout the survey warrant care-
ful eld investigation, above-threshold values in tills overlying
bedrock mapped as andesitic may be considered rst-order
anomalies.
Twoof these anomalous-Ktill sites occur west of the Mt. Mil-
ligan deposits, in the Phillips Lakes area (labelled K5 and
K6 in Figure 10). Ground follow-up at these sites included
till, bedrock, and biogeochemical sampling, ground spectrom-
etry, and magnetic susceptibility measurements. Bleached,
K-altered andesitic volcanics outcropping in the K5 area con-
tain pyrite and chalcopyrite in quartz carbonate veins. Al-
though no outcrop occurs in the K6 area, many large, angular,
sulde-bearing, quartz-veined, K-feldspar-altered porphyritic
intrusive boulders dene a granite till, containing numerous
gold grains and high Cu and Au concentrations. These results
prompted exploration activity, and drilling has conrmed the
presence of blind, low-grade Au and Cu mineralization in a
K-altered porphyritic intrusion, extending into the enclosing
volcanic host rocks.
Casino deposit area
The Casino Au-Cu-Mo porphyry deposit is located in west-
central Yukon Territory (Figure 1) in deeply weathered,
unglaciated terrain (extremely rare in Canada). The deposit
is associated with the Upper Cretaceous Casino Intrusive
Complex, which intrudes middle-Cretaceous granodiorites of
the Dawson Range Batholith. A preserved, 70-m thick, gold-
bearing, leached cap overlies a Cu-rich supergene zone, with
the base of weathering extending to 300 m below the sur-
face. Geological reserves include 28 million tons grading 0.68
g/ton Au, 0.11% Cu, and 0.024% Mo in the leached cap and
supergene-oxide zone; 86 million tons grading 0.41 g/ton Au,
0.43% Cu, and 0.031% Mo in the supergene-sulde zone; and
445 million tons grading 0.27 g/ton Au, 0.23% Cu, and 0.024%
Mointhehypogenezone. Acentral, mineralized, potassic alter-
ation zone consisting of K-feldspar and biotite is surrounded
by barren phyllic (sericite) and propylitic alteration. Bower
et al. (1995) provide detailed description of the deposit history,
geology, alteration, and mineralization.
FIG. 11. Potassium versus thorium concentrations determined
by in-situ gamma-ray spectrometry over the Casino deposit
area, Yukon Territory.
Ground spectrometry over the deposit and surrounding area
(Figure 11) demonstrated that despite deep weathering, ra-
dioactive element concentrations differentiate host litholo-
gies and alteration. Maximum bedrock K concentrations (6
8% K) were measured in mineralized felsic microbreccias
within the leached cap. Maps of airborne potassiumconcentra-
tions and eTh/K ratios for part of a 1993 helicopter-borne sur-
vey (Geological Survey of Canada, 1994) using 500-mline spac-
ing are shown in Figure 10. A unique, low eTh/K ratio bullseye
anomaly (Figure 10e) clearly distinguishes high-potassiumval-
ues associated with altered, mineralised bedrock in the Casino
deposit from similar high-potassium patterns (Figure 10d) as-
sociated with felsic volcanic rocks in the older Yukon Meta-
morphic Terrane, which has normal eTh/K values. These ra-
dioactive element and magnetic features are best viewed, in
detail, on stacked proles (Figure 10f).
CONCLUSIONS
The ability of gamma-ray spectrometry to map potassium,
uranium, and thorium enrichment or depletion provides pow-
erful exploration guidance in a wide variety of geological set-
tings. The case histories presented highlight the use of gamma-
ray spectrometry to measure and map potassium enrichment
related to volcanic-hosted massive sulde, polymetallic, and
porphyry mineralization in Canada. Potassium enrichment in
these and many other geological settings is characterised by
anomalously low eTh/K ratios relative to normal lithological
signatures, thus providing signicant exploration vectors.
ACKNOWLEDGMENTS
For the examples presented, the authors appreciate the
cooperation of Fortune Minerals Limited, Imperial Met-
als Corporation, Continental Gold Corporation, Pacic Sen-
tinel Gold Corporation, and Archer, Cathro and Associates.
N. Prasad is thanked for eld data and digital images for the
Lou Lake area. F. Santaguida supplied valuable eld guidance
and petrological information regarding Pilleys Island.
REFERENCES
Bostock, H. H, 1988, Geology and petrochemistry of the Ordovician
volcano-plutonic Roberts ArmGroup, Notre Dame Bay, Newfound-
land: Geol. Sur. Canada Bull. 369.
Bower, B., Payne, J., DeLong, C., and Rebagliati, C. M., 1995,
The oxide-gold, supergene and hypogene zones at the Casino
gold-copper-molybdenumdeposit, west central Yukon, inSchroeter,
T. G., Ed., Porphyry deposits of the northwestern Cordillera of
North America: Can. Inst. Min. Metal. Petr. Special Volume 46, 352
366.
Charbonneau, B. W., 1988, Gamma spectrometric and magnetic
anomalies associated with Cu-U mineralization, Faber Lake vol-
canic belt, District of MacKenzie, N.W.T.: Geol. Surv. Can. Paper
81-1C, 255258.
Colman-Sadd, S. P., Hayes, J. P., and Knight, I., 1990, Geology of the
Island of Newfoundland: Newfoundland Dept. Mines and Energy
Map 9091.
Coyle, M., and Strong, D. F., 1987, Geology of the Springdale Group: A
newly recognizedSilurianepicontinental-type caldera inNewfound-
land: Can. J. Earth Sci., 24, 11351148.
Crawford, A. J., Corbett, K. D. and Everard, J. L, 1992: Geochemistry
of the Cambrian volcanic-hosted massive sulde-rich Mount Read
volcanics, Tasmania, and some tectonic implications: Econ. Geol.,
87, 597619.
Davis, J. D., and Guilbert, J. M., 1973, Distribution of radioelements
potassium, uranium and thorium in selected porphyry copper de-
posits: Econ. Geol., 68, 145160.
Dickson, B. L., and Scott, K. M., 1997, Interpretation of aerial gamma
ray surveysAdding the geochemical factors: AGSO J. Austral.
Geol. Geophys., 17, 187200.
Durocher, M., 1983, The nature of hydrothermal alteration associated
with the Madsen and Starrett-Olsen gold deposits, Red Lake area;
in Colvine, A. C., Ed., The geology of gold in Ontario: Ontario Geol.
Surv. Misc. Paper 110, 123140.
Ford, K. L., 1993, Application of gamma-ray spectrometry to explo-
ration for VMS deposits in the Roberts Arm Group and Tulks
Volcanic Belt, central Newfoundland: Newfoundland Geol. Surv.
Branch Report of Activites 1993, 1213.
Franklin, J. M., 1996, Volcanic-associatedmassive sulphide base metals,
in Eckstrand, O. R., Sinclair W. D., and Thorpe, R. I., Eds., Geology
of Canadian mineral deposit types: Geol. Surv. Can., Geology of
Canada, 8, 158183.
Galbraith, J. H. and Saunders, D. F., 1983, Rock classication by char-
acteristics of aerial gamma ray measurements: J. Geochem. Expl.,
18, 4973.
Gandhi, S. S., 1994, Geology andgenetic aspects of mineral occurrences
in the southern Great Bear magmatic zone, Northwest Territories,
in Sinclair, W. D., and Richardson, D. G., Eds., Studies of rare-metal
deposits in the Northwest Territories: Geol. Surv. Can. Bull., 475,
6396.
Gandhi, S. S., Prasad, N., and Charbonneau, B. W., 1996, Geological
and geophysical signatures of a large polymetallic exploration tar-
get at Lou Lake, southern great Bear magmatic zone, Northwest
Territories, in Geol. Surv. Can., 147158.
Geological Survey of Canada, 1992, Airborne geophysical survey,
Mount Milligan area, British Columbia (NTS 93 O/4W, N/1, N/2E):
Geol. Surv. Can. Open File 2535.
Geological Survey of Canada, 1994, Airborne geophysical survey,
Selwyn River, Yukon Territory, (NTS 115I/12W, J/9, J/10, and parts
of J/11, 12, 13): Geol. Surv. Can. Open File 2816.
Hetu, R. J., Holman, P. B., Charbonneau, B. W., Prasad, N., andGandhi,
S. S., 1994, Multiparameter airborne geophysical survey, Mazenod
Lake area, NWT, 1993 (NTS 85N/10 and parts of 85 N/11, 14, 15):
Geol. Surv. Canada, Open File 2806.
Hoover, D. B., and Pierce, A. A., 1990, Annotated bibliography of
gamma-ray methods applied to gold exploration: U.S. Geol. Surv.
Open-File Report 90-203.
Kerr, D. E., 1991, Surcial geology of the Mount Milligan area, NTS
93N/1E, 93O/4W: British Columbia Ministry of Energy, Mines and
Petroleum Resources, Open File 1991-7.
Kuhns, R. J., 1986, Alteration styles and trace element dispersion as-
sociated with the Golden Giant deposit, Hemlo, Ontario, Canada,
in MacDonald, A. J., Ed., Proceedings of Gold 86, an International
Symposium on the Geology of Gold: 340354.
Lagerblad, B., and Gorbatschev, R., 1985, Hydrothermal alteration as
a control of regional geochemistry and ore formation in the central
Baltic Shield: Geologische Lunschau, 74, 3349.
Lowell, J. D., and Guilbert, J. M., 1970, Lateral and vertical alteration-
mineralization zoning in porphyry ore deposits: Econ. Geol., 65,
373408.
Moxham, R. M., Foote, R. S., and Bunker, C. M., 1965, Gamma ray
spectrometer studies of hydrothermally altered rocks: Econ. Geol.,
60, 653671.
Nelson, J., Bellefontaine, K., Green, K., and Maclean, M., 1991, Re-
gional geological mapping near the Mount Milligan copper-gold
deposit (93N/1 and 93K/16): British Columbia Ministry of Energy,
Mines and Petroleum Resources, Geological Fieldwork 1990, Paper
1991-1.
Ofer, R., andWhitford, D. J., 1992, Wall-rockalterationandmetamor-
phism of a volcanic-hosted massive sulphide deposit at Que River,
TasmaniaPetrology and mineralogy: Econ. Geol., 87, 686705.
Portnov, A. M., 1987, Specialization of rocks towards potassium and
thorium in relation to mineralization: Internat. Geol. Rev., 29, 326
344.
Poulsen, K. H., and Hannington, M. D., 1996, Volcanic-associated mas-
sive sulphide gold, in Eckstrand, O. R., Sinclair, W. D., and Thorpe,
R. I., Geology of Canadian mineral deposit types: Geol. Surv. Can.,
Geology of Canada, 8, 183196.
Richardson, K. A., Holman, P. B., and Elliott, B., 1974, Airborne ra-
dioactivity map, District of Mackenzie, Northwest Territories, 85N,
O, P, 75M; Geol. Surv. Can. Open File 188.
Santaguida, F., Hannington, M. D., and Jowett, E. C., 1992, An al-
teration and sulphur isotope study of the Pilleys Island massive
sulphides, central Newfoundland, in Geol. Surv. Can. Paper 92-1D,
265274.
Schroeter, T. G., Ed., 1995, Porphyry deposits of the northwestern
cordillera of North America: Can. Inst. Min. Metal. Petr. Special
Volume 46.
Shives, R. B. K., Ford, K. L., and Charbonneau, B. W., 1995, Appli-
cations of gamma ray spectrometric/magnetic/VLF-EM surveys
Workshop manual: Geol. Surv. Can. Open File 3061.
Sketchley, D. A., Rebagliati, C. M., and DeLong, C., 1995, Geology,
alteration and zoning patterns of the Mt. Milligan copper-gold de-
posits, in Schroeter, T. G., Ed., Porphyry deposits of the northwest-
ernCordillera of NorthAmerica: Can. Inst. Min. Metal. Petr. Special
Volume 46, 650665.
Thurlow, J. G., 1996, Geology of a newly discovered cluster of blind
massive sulphide deposits, Pilleys Island, central Newfoundland, in
Newfoundland Dept. Geol. Surv. Report 96-1, 181189.
Tuach, J., Hewton, R. S., and Cavey, G., 1991, Exploration targets for
volcanogenic, base- metal sulphide deposits on Pilleys Island, New-
foundland, in Ore horizons, V. 1, Newfoundland Geol. Surv. Branch,
8998.
Wilford, J. R., Bierwirth, P. N., and Craig, M. A., 1997, Application
of airborne gamma ray spectrometry in soil/regolith mapping and
applied geomorphology: AGSO J. Austral. Geol. Geophys., 17, 201
216.
There are several corrections to the equations in Appen-
dix B. Please note that, once corrected, they still support
the claim that the scaled Poissons ratio change is equiva-
lent to the fluid factor, as stated in the paper.
Equation (B-5)
/ g V
P
/ V
P
Equation (B-9)
B =
5
4
A (1 ) 4
2 2
V
S
V
S
,
Equation (B-9a)
5
=
A (1 ) B
2
V
S
V
S
4
.
Equation (B-11)
F = [1.6 0.232 (1 ) / ] A + B.
0.29
2
Errata
To: Effective AVO crossplot modeling: A tutorial (C. P. Ross, GEOPHYSICS, 65, 700-711)
Equation (B-12)
F = A + B.
Equation (B-12a)
a = 1.6 0.232
2
[ [
(1 )
Equation (B-13)
F = 1.252A + 0.58B,
Equation (B-14)
F = k(0.50A + 0.23B),
Equation (B-15)
k
-1
F 0.50A + 0.23B.

Anatomy Inverse Problems

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Anatomy Inverse Problems

Uploaded by

Copyright:

Available Formats

GEOPHYSICS, VOL. 65, NO. 6 (NOVEMBER-DECEMBER 2000); P. 17081710, 4 FIGS.

The Anatomy of Inverse Problems

and Roel Snieder

A major task of geophysics is to make quantitative state-

and Tor Arne Johansen

. Actually, equation (7), conditioning the

, along with equa-

, and Alberto Malinverno

Baker Atlas, 10205 Westheimer, Houston, Texas 77042. E-mail: sheng.fang@waii.com.

(via LU factorization and back substitution, for instance).

was chosen as small as possible but without com-

changed and the iterations progressed in their way toward

, and G abor Hurs an

and Umesh C. Das

f (x, y)( x cos y sin ) dx dy,

and Anthony F. Gangi

C/km, and S may range between 0.05 and

are coefcients obtained by tting the

C/km, a sedimentation rate S =0.08 km/m.y., and a

C. After 75 m.y., the depth of burial is

C. On the other hand,

=6.48. The pore compressibility c

, and Ilya Tsvankin

) lies in the plane

-wave is polarized perpendicular to the

-wave requires a better azimuthal coverage

(in the plane perpendicular to the fractures) from P

-wave in the direction

-waves, respectively. If the vertical velocities and

-wave across the fractures and the PS

, and Ilya Tsvankin

with respect to planes [x

with respect to the symmetry

, and Ilya Tsvankin

from the strike of the rst fracture system. In the limit of

Estimated 0.5 0.20 0.04 40

and S-wave polarizations described by P erez et al. (1999). The

; the actual values are

g(t ) exp(i t ) dt. (5)

with the vertical. One, of course,

) is shown in equation (A-2). Assuming an

. We can see that Q

appear in the form of spatial

to the east and

and extends into the upper to middle crust

) as seismic waves pass from the Klipriviersberg

subsurface core area shown on Figure 5. This area cov-

) and maximum frequencies (<75 Hz)

, and Rick Somerville

), the most prominent part of the

to the north-northwest, but structural inu-

Kiel University, Department of Geophysics, 24118 Kiel, Germany. E-mail: bernd@geophysik.uni-kiel.de.

Falconbridge Ltd., Exploration, Toronto, Canada. E-mail: anthonywatts@compuserve.com.

. The unmigrated data shows the region over which the

. From surface and borehole

, and Cvetan Sinadinovski

fromthe horizontal (so as to hold

tothe decline axis. Apiezoelectric

Electromagnetics Instruments, Inc., P.O. Box 463, El Cerrito, California 94530-0463.

east and dips 70

and Douglas W. Oldenburg

denotes the potential in the absence of IP effect.

is found whose elements are all within the bounds.