Vous êtes sur la page 1sur 11

Indonesian Journal of Physics

Vol 16 No. 4, October 2005




115
Study of Deep Sounding Time-Domain Electromagnetic (TDEM) Method
Using Horizontal Electric Dipole to Infer Subsurface Resistivity Structure
Wahyu Srigutomo
1)
, Tsuneomi Kagiyama
2)
, Wataru Kanda
3)
, and Hisashi Utada
4)
1)
Earth and Computational Physics Research Group,
Department of Physics, Institut Teknologi Bandung,
Jl. Ganesa 10 Bandung 40132, Indonesia
2)
Institute for Geothermal Sciences,
Graduate School of Science, Kyoto University, Minamiaso, Kumamoto 869-1404, Japan
3)
Disaster Prevention Research Institute,
Kyoto University, Sakurajima, Kagoshima 891-1419, Japan
4)
Earthquake Research Institute, University of Tokyo, Japan
e-mail: wahyu@fi.itb.ac.id
Abstract
Electromagnetic methods to infer subsurface resistivity structure particularly in volcanic and hydrothermal regions are
considered highly effective to apply. Time domain electromagnetic (TDEM) method as one of them, is a robust method
utilizing controlled-galvanic source which has higher ability to penetrate deeper structure. This paper tries to elaborate the
theoretical formulation of TDEM with a long-grounded current wire for a layered-earth problem, the mathematical
inversion algorithm to find the suitable model for interpretation as well as its sensitivity analysis, and the application to
synthetic and real measured data.
Keywords: resistivity structure; time domain electromagnetic method.

1. Introduction
Investigations on complete structure of volcanic
regions in upper crust play important roles in
understanding various volcanic processes such as growth
of volcano, history and mechanism of eruptions,
formation of hydrothermal system as well as magma
supply system and magma degassing process associated
with it. Information on magma supply system and
degassing process will also provide alternative suggestive
answers to question arisen from volcano monitoring and
scientific views such as presence of current subsurface
activity, type of eruption, location, depth and dimension
of main magma chamber, interconnection between
magma chamber and deeper magma source, geometrical
form of magma propagation to the surface, and possible
interaction between volcanic gas and groundwater.
Commonly, volcanic regions are characterized by
the presence of high porosity rocks, extremely dry or
saturated with ionic hydrothermal fluids and the presence
of wet, hot magma reservoir, which exhibit also
temperature contrasts in the region
1,2)
. Among other
geophysical parameters, electrical resistivity (or its
inverse, conductivity, = 1/) is one of the most sensitive
to the above composition and temperature, particularly to
the presence of fluids in the crust or upper mantle in the
form of ion-bearing aqueous phase, or as partial melt
distributed along the pores and cracks in the host rock
matrix
3)
. Generally, the bulk conductivity of the host
medium follows Arrhenius relation:
kT E
e
/
0

= (1)
where
0
is a constant that depends on the number and
mobility of charge carrier, E is an activation energy, k is
the Maxwell-Boltzman constant, and T is the temperature.
At T less than several hundred degrees C, most of dry
rocks are essentially resistive insulator, however small
amounts of fluids in pores can significantly increase the
bulk conductivity according to empirical Archies Law
for brine-saturated porous materials, expressed in
resistivity:
m
fluid bulk
a

= , (2)
where
bulk
is the electrical resistivity of the porous
medium,
fluid
is the electrical resistivity of the saturating
fluid, a is the coefficient of saturation, m is the
cementation factor, and is fractional porosity. The ratio
fluid bulk
: is called the formation factor F. For the
typical sandstones it was found that the coefficient a was
approximately 1 and the exponent m was close to 2. Study
with other rocks and sediments shows that this power law
is generally valid but with varying coefficients and
exponents. Table 1.1 shows brief summary of the
coefficients for different materials
4)
.





116 IJP Vol. 16 No. 4, 2005


Tabel 1.1 Fractional porosity, coefficient of saturation and cementation factor

Rock a m
Weakly cemented, detrital (Tertiary) 0.25 0.45 0.88 1.37
Moderately well cemented (Mesozoic) 0.22 0.35 0.62 1.72
Well cemented (Paleozoic) 0.05 0.25 0.62 1.95
Dense, igneous, metamorphic < 0.05 1.4 1.6
High porosity volcanic 0.2 0.8 3.5 1.4

fluid
is inversely proportional to the mobility of ion
contained in the fluid which is proportional to the its
radius and viscosity. Increasing temperature will decrease
viscosity, hence increasing in mobility which causes
decreasing in the total resistivity (or increasing the
conductivity). Resistivity also depends on clay content,
for materials with high clay contents, when fluid and clay
are mixed, concentration of ions is increased in the
solution in the vicinity of the clay surface and (2) is
modified into
5)

+
=
s
fluid
bulk
F
2 1

(.3)
where is a length parameter, which is a weighted
volume-to-surface-area ratio and
s
is the surface
conductivity.
Geophysical methods to infer electrical resistivity
structure down to about 10 km depth or more of active
volcanic regions mainly can be classified into two groups:
natural source electromagnetic methods such as
geomagnetic depth sounding (GDS) method (cf. Gough
6)
)
and the widely used magneto-telluric (MT) method
7)
.
Although MT method is more practical because no source
instruments needed, it still suffers from several possible
drawbacks such as galvanic distortion due near surface
heterogeneity, its dependency on the electrodes
performance and expected low quality data in areas with
highly artificial noise due to power lines or trains. The
other group is controlled-source EM methods, which
utilize either inductive source such as current loop or
galvanic source such as horizontal grounded wire. The
inductive types use secondary magnetic field as a source,
which decay rapidly in the earth whereas the galvanic
types use injected current as a source and hence has
higher ability to penetrate into deeper structure. The time
domain electromagnetic (TDEM) method is more
appropriate than its counterpart, namely frequency
domain electromagnetic (FDEM) method. The former
only measures secondary signal reflecting the earth
responses by switching off the current, while the later
measures total field composed by secondary field and
primary field from the harmonically oscillating current.
The secondary fields of the deeper part are usually several
order of magnitude smaller than the primary fields
causing separation between them is highly difficult and
inaccurate. TDEM method, also called transient
electromagnetic (TEM) is good choice for inferring
resistivity structure since it is robust against noise, having
best lateral and vertical resolution with regard to highly
conductive targets, less influenced by near-surface
heterogeneity and less ambiguous compared to other
methods.
2. Theoretical background
In this section, the term TDEM will refer to the
method in which a time-dependent magnetic field is
generated by passing bipolar square waves of current
through a grounded wire source. The components of the
magnetic fields from the wire are detected at a receiver
site using a sensor.
Electromagnetic field is generated by passing a
step of current along a cable lying on the surface of the
earth. Both ends of the cable are grounded to the earth.
The magnetic field from this current circuit can be
separated into two parts, one from the current flowing
through the cable and the other from the return currents
flowing through the earth. The magnetic field from the
current in the cable establishes a magnetic field more or
less instantaneously within the insulating half-space
above the earth. Energy from this magnetic field is
refracted nearly vertically into the earth and is scattered
back to the surface from any conductive regions within
the Earth. For a laterally uniform earth which is the of the
main interest in 1-D structure, only a vertical component
of magnetic field from the source exists at any
observation point on the earths surface. If the Earth is not
laterally uniform, a horizontal component of a magnetic
field can be contributed by scattering. The magnetic field
from return current in the earth spreads very slowly
compared to the rate at which the magnetic field above
the earth establishes itself.
When current is transmitted, stationary magnetic
field is generated due to Amperes law. If the current is
shut off at t = 0, the eddy current is induced at the area
near the dipole and the earth surface to preserve the
existing magnetic fields (Faradays law). This induced
current itself generates the secondary magnetic field, and
this process is repeated downward and outward to cancel
the changes in magnetic field. The magnitude of the
induced current decreases with the increasing time.
Because of this induction process, the magnetic fields also
show diffusive nature. The magnetic field does not
disappear soon in the Earth after shutting off the source
current but it took a time to diffusive completely. Time
constant of the diffusion can be estimated
2
0
r = ,
using the typical earths conductivity, the magnetic
permeability of free space
0
and the distance between
IJP Vol. 16 No. 4, 2005 117


the dipole and the receiver r. If the time after shutting off
the current is small enough (t<<), the magnetic field
shows transient and distorted. This diffusion process can
be measured as a function of time at the surface. The rate
of the diffusion process depends on the Earths
conductivity structure, so that the observed decay curves
show different behaviors. In conductive earth, the
transient decays slower, whereas in resistive earth, the
transients decay faster.





















Figure 1. Basic principle of TDEM. The current source is
abruptly stopped at t = 0, and diffusion process is
observed on the surface.
2
0
r = is the diffusion time
constant,
0
is the magnetic permeability in free space,
is conductivity of the medium and r is the source-receiver
distance.
3. One-dimensional (1D) forward problem
Comprehensive theoretical background of time
domain electromagnetic problems had been discussed in
several literatures. The derivation of electromagnetic
fields in this section will be mainly based on formulations
derived by Kaufman and Keller
8)
; Ward and Hohmann
9)
;
and also based on the TDEM study previously conducted
by Kanda
10)
and Srigutomo
11)
. The behavior time domain
electromagnetic problem is completely specified by the
Maxwells equations incorporating the impressed source
distribution. In this study, the electromagnetic fields due
to excitation of injected current will be observed. The
fields generated by a horizontal electric dipole (HED)
source distribution J
s
can be expressed as
( ) ( ) ( ) ( ) t r J t r r t r
s
, , , + = (4)
( )
( )
t
t r
t r


=
,
,
0
(5)
where ( ) r is the electrical conductivity which varies in
all direction. For simplicity, is assumed the Earth is
isotropic and the magnetic permeability is everywhere
equal to its free space off
0
(=4 x 10
-7
H/m). Magnetic
induction B can be represented using the constitutive
relation, namely =
0
. The expression of source
distribution J
s
generated by a HED located at r = r
0
whose
is cut-off at time t = t
0
can be given as
( ) ( ) ( ) [ ]
0
1 , t t r r t r J
o s
= (6)
where I is the current intensity and the H (t) is the
Heaviside function.
In transient problems, it is often more convenient
to use the Laplace transform rather than Fourier
transform, we substitute the expression of i by s, the
Laplace domain variable. This transformation is intended
to avoid the source singularity in time. The expression for
equation (4) and (5) then become:
s
J + = , (7)
= s , (8)
where s denotes the Laplace variable.
For the sake of the simplicity of formulation, the
conductivity is assumed to vary only in vertical z
direction The earth consists of layers with N number of
layers and the conductivity of each layer is constant, the
configuration is shown in Figure 2. The HED source with
the current moment I dx is located at z = -h, where I and
dx denote the current intensity and the infinitesimal dipole
length, respectively. In the i-th layer (z
i-1
< z < z
i
), which
is the source free region, equations (7) and (8) became:
i i i
= , (9)
i i
s = , (10)



Figure 2. Coordinate system for the horizontally layered
conductivity model of the Earth used in this study. z
j
is
the depth to the bottom of the jth layer and
j
is the
conductivity of the jth layer. HED represents the source
at a height h above the surface, and r indicates the
measurement location.

x
y
z
z =-h
z =0
z1
z2
zj-1
zj
zN-2
zN-1
0

1 N

HE
r
r
HED
receiver 0 >t
0 <t <
,

2
r
t =0
t =0
I
H

time
time
118 IJP Vol. 16 No. 4, 2005


Kaufman & Keller
8)
make use of the vector
potential A, such that
i i
= , (11)
substituting (11) to (10) yields:
( ) 0
0
= +
i i
s (12)
Electric fields can be expressed by introducing the
arbitrary scalar potential U as:
i i i
U s =
0
(13)
from (6):
i i i i
i
i
i i
U s = =
=

0
2
(14)
Defining a gauge condition as:
i i i
U = (15)
and apply it to the last equations of (15), then:
( ) 0
2 2
= +
i i
k (16)
where,
i i
s k
0
2
= . Without loss of generality, we can
define that the y component of the vector potential is
equal to zero, that is:
( )
i i i
z x = , 0 , (17)
Then, we have:
( ) 0
2 2
= +
i i
x k (18)
( ) 0
2 2
= +
i i
z k (19)
We have to solve equations (18) and (19) under the
appropriate boundary conditions, which continuity of
tangential components of the electric and magnetic fields
at the layer interface. Making use the definition for vector
potential (equations (12) and (13)), we have:


=
y
x
x
z
z
x
y
z
i i i i
i
, , (20)

=
z
U
z s
y
U
x
U
x s
i
i o
i i
i o i
, , (21)
where
i i
U = . Continuity of tangential EM fields
component is assured, if
i
x , , ,
i
i
z
z
x


and U
i
, are
continuous. For the boundary at z = z
i
, it requires that:
1 +
=
i i
x x (22)
1 +
=
i i
z z (23)
z
Ax
z
Ax
i i

+1
(24)

+
+
z
Az
x
Ax
z
Az
x
Ax
i i
i
i i
i
1
1
1 1

(25)
Consequently, we solve equations (18) and (19) under the
condition of (22) - (25) in the source free region (z > 0).
In the upper half-space (z < 0), labeled the zero-th
layer, the vector potential component can be written as
sum:
0 0 0
x x x
s p
+ = (26)
0 0 0
z z z
s p
+ = (27)
where
0
x
p
and
0
z
p
are the components of vector
potential for an electric dipole source in a uniform full-
space, and
0
x
s
and
0
z
s
are the vector potential terms
representing the effect of the secondary fields. For
convenience, we assume for the time being that the upper
half-space is also the conducting full-space has only a
single component, which is perpendicular to the direction
of the dipole, and has cylindrical symmetry. Therefore,
the vector potential has only one component,
0
x
p
in the
present case, and we have 0
0
= z
p
. Kaufman and
Keller
8)
give appropriate expressions for
0
x
p
at the
point (r,0,z) for the dipole situated at (0,0,-h):
( )


d r J e
m
Idx
R
e dx
x
h z m
R ik
p
0
0
0
0
0
0
4 4
+

= (28)
where r is the distance from the dipole source, J
0
is the
Bessel function of the first kind of order O, and;
( ) ( ) ,
2 2
h z r R + =
0 0
2 2
0
2
0
s k m + = = (29)
0
x
p
is only the function of the coordinate r and z, so
that we can represent the vector potential for the
secondary and total fields as function of yet to be
determined from the boundary condition of the equation
(25), the expression for
0
z
s
can be written as:
( ) ( )

d r J e C
m
Idx
x
z m s
0 0
0
0
0
0
4

= (30)
where C
0
() is unknown function of () yet to be
determined from the boundary condition. Taking into
account that we have to calculate div A for the boundary
condition of the equation (25), the expression for A
s
z
0
can
be written:
( ) ( )

d r J e D
x
Idx
z
z m s
0
0
0 0
0
4

= (31)
where D
0
() is also to be determined. From the equation
(26) and (27), the vector potential in the upper half-space
can be written as:
( ) [ ] ( )

d r J e C e
m
Idx
x
z m h z m
0
0
0
0
0
0 0
4

+
+ = (32)
( ) ( )

d r J e D
x
Idx
z
z m
1
0
0 0
0
4

= (33)
Similarly solutions of the equations (18) and (19)
can be written by using the coordinate r and z only.
Taking into account for similarity to the equations (32)
and (33), and for the boundary conditions, we can write
them for the i-th layer ( N i 1 ) as:
IJP Vol. 16 No. 4, 2005 119


( )
( )
( )
( )
[ ] ( )

d r J e c e C
Idx
x
i i i i
z z m
i
z z m
i i 0
0
1 1
4


+ =
(34)
( )
( )
( )
( )
[ ] ( )

d r J e d e D
x
Idx
z
i i i i
z z m
i
z z m
i i 0
0
1 1
4

=

(35)
where C
i
(), c
i
(), D
i
(), d
i
(), are unknown to be
determined, and:
i i i
s k m
0
2 2 2
+ = = (36)
Since EM fields must vanish with z , in the lowest
(N-th) layer we have:
( ) ( ) 0 = =
N N
D C (37)
We solved the sequential equations (32) and (34)
under the boundary condition (22) and (24). From the
straight forward algebraic calculations, we have following
regressive expressions for C
i
()and c
i
():
( )
( )
i
i i i i
i i i
i
i i
i
n
i
m
h m m
hi m m
m
C c
Ci c
tanh
tanh
1
1
+
+
+
+
=
+

,
( ) 1 1 N i (38)
N N
m = (39)
where h
i
is the thickness of the i-th layer, that is:
1
=
i i i
z z h (40)
At the surface of the Earth (z=0) the same boundary
conditions are applied, we have:
( )
h m h m
o
e
m
Y m
e R C

= =
1 0
1 0
0
0
(41)
If the dipole is located on the surface ( ) 0 h , we have:
[ ] ( ) ( )

d r J e
m
Idx
d r J e R
m
Idx
x
z m z m
0
0
1 0
0
0
0
0
0 0
2
4
1
4

+
= + =
(42)
We have to solve equations (33) and (36) under the
boundary conditions (23) and (25) to seek for the
expressions for A
z
. This time equations to be solved are a
bit complicated because of the boundary conditions (25).
However, by replacing D
i
() and d
i
() with the following
regressive expressions, they result in the similar equations
which we solved to determine C
i
() and c
i
(), that is:
( ) ( ) ( )


i
i
i i
c
m
F d
2
+ = (43)
( ) ( ) ( )


i
i
i i
c
m
G D
2
= (44)
Then we have following expressions for F
i
() and G
i
():
( )
( )
i
i i i i
i i i i
i
i i
i i
i
n
h m Z n
h m n Z
n
G F
G F
Z
tanh
tanh
1
1
+
+
+
+
=
+

= ,
( ) 1 1 N i (42)
,
n
n
N N
m
n Z

= =
i
i
i
m
n

= (46)
For the boundary condition at the surface:
( ) ( ) [ ]
h m h m
e R R e
Z n
Z n
C D
0 0
||
1 0
1 0
0 0
1 1

+ =

+ =

(47)
In accord with (41) and with 0 h , we have:
[ ] ( )

d J e R R
x
Idx
Az
z m

=
0
0 || 0
0
1
4
(48)
With the obtained expression of vector potential,
we can write expressions for the magnetic field
components at the surface (z = 0). Taking into account for
the symmetry, we drive them in cylindrical coordinates (r,
, z):
[ ] ( ) ( ) ( )

+ + =

0
0
0
1 ||
1
1 sin
4

d r J R r d r J R R
r
Idx
H
r

(49)

[ ] ( ) ( ) ( )

+ + + =

0
0 ||
0
1 ||
1
cos
4

d r J R r d r J R R
r
Idx
H

(50)
( ) ( )

d r J R
m
Idx
H
z 1
0
0
1 sin
4

+ = (51)
At this moment, we recognize that the upper half-space is
an insulator, namely 0
0
. Then with 1
||
R and

0
m , we have:
( ) ( ) ( ) ( )

+ =


0
0 1
0
0
, , d r J s Q r d r J s R r H H
r r
(52)
( ) ( )

d r J s R r H H
1
0
0
,

= (53)
( ) ( ) ( ) d r J s R r H s h H H
z z z z 1
0
2
0 0
,

= = (54)
where J
k
(k = 0 or 1) is the Bessel function of the first
kind of order k and:
( ), sin
4
2
0 0

r
Idx
H H
r z
= =
( )

cos
4
2
0
r
Idx
H = (55)
( ) ,
2
,
1
+
=


s R ( )
1
1
2
,
+


s Q (56)
In Cartesian, the magnetic field intensity on the
surface and in the earth can be expressed as
120 IJP Vol. 16 No. 4, 2005


( ) ( ) ( ) =

d r J e D
r
xy Idx
Hx
z m
1
0
0
3
0
0
2
4
( ) ( ) ( )

d r J e D
r
xy Idx
z m
0
2
0
0
2
0
4

(57a)
( )
( )
( )
( )
[ ] ( ) + =

d r J e d e D
r
xy Idx
Hx
i i i i
z z m
i
z z m
i i 1
0
1 1
3
2
4

( )
( )
( )
( )
[ ] ( )

d r J e d e D
r
xy Idx
i i i i
z z m
i
z z m
i 0
2
0
1 1
2
4


+ (57b)
( ) [ ] ( ) + + =

d r J e C m e m
m
Idx
Hy
z m h z m
0
0
0 0 0
0
0
0 0
4
( ) ( ) +

d r J e D
r
x
r
Idx
z m
1
0
0
3
2
0
2 1
4

( ) ( )

d r J e D
r
x Idx
z m
0
2
0
0
2
2
0
4

(58a)
( )
( )
( )
( )
[ ] ( ) + =

d r J e c m e C m
Idx
Hy
i i i i
z z m
i i
z z m
i i i 0
0
1 1
4
( )
( )
( )
( )
[ ] ( ) + +

d r J e d e D
r
x
r
Idx
i i i i
z z m
i
z z m
i 1
0
1 1
3
2
2 1
4

( )
( )
( )
( )
[ ] ( )

d r J e d e D
r
x Idx
i i i i
z z m
i
z z m
i 0
2
0
1 1
2
2
4


+ (58b)
( ) [ ] ( )

d r J e C e
m r
y Idx
Hz
z m h z m
0
0
0
0
0
0 0
4

+
+ = (59a)
( )
( )
( )
( )
[ ] ( )

d r J e c e C
r
y Idx
Hz
i i i i
z z m
i
z z m
i i 1
0
1 1
4


+ =
.(59b)
Similar expression for vertical and horizontal electric
fields can be obtained by using (21).
To calculate the related Hankel transform integral,
digital linear filtering based on the works of Chave
12)
and
Anderson
13-15)
algorithm is used. To obtain the response
value in the time domain, the inverse Laplace transform is
applied to all components of the desired data at any given
location. The field to be inversed is the total value and it
has the form of
( ) ( ) dt s s F
i
t
st
c
c

=
2
1
h . (60)
The numerical integration of the above equation is carried
out using the Gaver-Stehfest approximate inversion
formula
16,17)

=
=
L
l
l l
s F K
t
t h
1
) (
2 ln
) ( ,
t
l s
l
2 ln
= (61)
and
( )
( )
( )( )( )

+
=
+


=
2
, min
2
1
2
2
2 1 ! !
2
! 2
1
L
l
l
k
L
L
l
l
l k k l l k k
L
k k
K (62)
where L is the number of the coefficient which depends
on the computer used to calculate the inverse. Since we
are interested in the transient response after the current is
switched-off at t = 0, the magnetic field measured as a
function of time can be expressed as:
( ) ( ) 0 =

t t h h t h
DC
, (63)
where h
DC
is the induced magnetic field at t or
0 due to HED source, calculated in the same manner as
h(t) for the stratified Earth except that the value of the
Laplace variable s is set to zero.
4. Sensitivity calculation
Actual geophysical problems mainly are inverse
problems, in which one has to develop such a
mathematical process by which data are used to generate
a model that is consistent with the data. Inverse problem
requires calculation of sensitivity, which correlates
changed in a proposed model to the resulting changes in
the forward modeled data. Comparative study of several
methods for calculating the sensitivity of non-linear
problems involving perturbation or brute force approach,
sensitivity equation approach, and adjoint equation
approach was discussed by McGillivray and Oldenburg
18)

and Furquharson and Oldenburg
19)
. The layered
conductivity structure in Figure 1 can be described in
terms of linear combination e.g.,
20,21)
:
( ) ( ) z z
N
j
j j
=
=
1
, (64)
where
j
is the basis function that equals to unity within
the jth layer zero everywhere else. The coefficient
j

denotes conductivity of the jth layer. If this coefficient is
changed by a small amount, the resulting value of h
z
value
measured at the surface can be expressed by a Taylor
series expansion about the original conductivity structure:
[ ] [ ]
[ ]
2
1

=
+

+ = +
N
j
j
j
O
h
h h

, (65)
where ( )
T
N
, ,
1
= is a vector of original
conductivity structure, whereas
( )
T
N
, ,
1
= represents the changes in the
coefficients, and the first order partial derivatives are the
sensitivities, and represents the l
2
norm. Based on
the principle of linearity, the sensitivity calculation is
started with the expression of A. For vertical magnetic
field h
z
, the equation (18) can be written in a form of
HEDsource A i u
dz
d
x
N
j
j j
=



=0
0
2
0
2
2

(66)
which is valid for - < z < .
Differentiation with respect to
j
and realizing that
the source is independent of
j
will yield
IJP Vol. 16 No. 4, 2005 121


x j
j
x
N
j
j j
A i
A
i u
dz
d


0
0
0
2
0
2
2
'
' ' =



=
(67)
which is the inhomogeneous ordinary differential
equation for the sensitivity.
The boundary conditions are

z
A
j
x
, 0

. This boundary value problem


can be solved using the adjoint Greens function
method
22)
:
( ) ( ) ( ) ( )dz z z G z A z i z
A
x
x
x j
j
x
' ; '
0

, (68)
where the adjoint Greens function ( )
'
; z z G

satisfies
( ) ( ) ' ' ;
0
0
2
0
2
2
'
' ' z z z z G i u
dz
d
N
j
j j
=

(69)
and
( ) 0 ' ,

z z G as z . (70)
Two linear independent solutions of eq. (69) are
necessary to construct the adjoint Greens function. One
is for boundary condition for z - and the other is for
z . In the region - < z < z = 0, where the
conductivity is zero, the adjoint Greens function has the
form exp(u
0
z). In the region z < z = 0 < , equation (69)
is the complex conjugate of equation (66), and since
0

x
A as + z , the adjoint Greens function is
proportional to

x
A in this region.
( )
( )

=
=
=
+

0 ' ,
0 ' ,
' ,
0
z z z F c
z z e c
z z G
z u
(68)
In regard to the boundary condition, at z = z = 0,
( ) ' , z z G

must be continuous and its derivative with


respect to z must be discontinuous by an amount equal to
1 according to the Greens function theory. At ' z z = ,
( ) ' , z z G

= ( ) ' , z z G

+
(71.a)
( ) ( )
1
; ;
' '
+



+
z
z z G
z
z z G
. (71.b)
At z = 0 (surface) the coefficients are found to be:
( )
( )
( ) 0
0
0
0

=
A u
z
A
A
c
x
x
(72)
( )
( ) 0
0
1
0

=
A u
z
A
c
x
.
Using the above results and applying the property
of
j
that is unity in the jth layer and zero elsewhere,
equation (68) can be expressed by
( )
( )
( ) [ ]

= =

j
j
z
z z
x
x
z
j
x
dz z A
A u
z
A
i A
1
2
0
0
0 ' 0
0

. (73)
Eq. (73) can be simplified by substitution of equation (32)
into eq. (73) and by replacing i into Laplace variable s:
( ) [ ]

= =
=

j
j
z
z z z
j
dz z A
Idx
s A
1
2
0
0 '

. (74)
Comparison between the sensitivities calculated
with the adjoint method and the perturbation or brute
force method is shown in Figure 3. The brute force
method was conducted using the central finite difference
scheme. The sensitivities are calculated at the surface of
the Earth, at a distance about 5.3 km from the HED
source due to a change in conductivity at depth z = 4 km.
The 1-D layered structure is the same as that discussed in
the next chapter 4, referred as the background structure
and the values are normalized by the maximum value of
the adjoint Greens function results for a certain range of
time (0.0078 s to 5.0 s). The pattern of the sensitivity
curves is similar. However, observable differences are
found within the time interval. These differences are
probably due to the perturbation approach whose
sensitivities depend on the conductivity perturbation used
in the calculation. The ratio of perturbation to the original
conductivity used in the comparison,

, is 1.135%.


Figure 3. Comparison of vertical magnetic field
sensitivities calculated with the adjoint Greens function
approach (solid line) and perturbation or brute force
approach (open circle). The sensitivities are due to change
in conductivity at depth 4 km, measured at the surface
with distance from the HED source approximately 5.3
km.
5. One-dimensional TDEM inverse with smoothness
constraint
Smoothness constrained inverse problems
concerning transient EM with horizontal electric dipole
source had for layered earth appeared for example in
Furquharson and Oldenburg
20,23)
, whereas inversion of
TDEM data using Bayesian statistic approach known as
ABIC inversion was introduced by Kanda
10)
. In this study,
122 IJP Vol. 16 No. 4, 2005


we follow the idea of Occams inversion
24,25)
. The use of
Occam's inversion produces a smooth model that fits a
data set within certain tolerances, although a smooth
model might not be the best fit to the data. In addition to
surface data, borehole data inversion from large magnetic
loop as inductive source was conducted by Zhang and
Xiou
27)
, whereas in this study we use long grounded wire
as galvanic source, which formulation has been describe
in the previous section.
The TDEM non-linear inverse problem is
formulated as
min
Lm
and
( ) WG m - Wd
(75)
where L is a finite difference approximation of a first or
second derivative and
1 1 1
, , ,
1 2
diag
N

=





W L
,
a diagonal matrix consisting the uncertainty
i
in the i-th
datum. We assume that the measurement errors in d are
independent and normally distributed. The goodness of fit
of model predictions to the actual values is measured by
the misfit criterion,
( ) ( )
2
2
2
1
N d G
i i
i
i

=
=
m
.
Given a model m, we use Taylors theorem to obtain the
approximation
( ) G(m+ m) G(m) + J m m, (76)
where J(m) is the Jacobian matrix (see the previous
section of sensitivity)
( )
( ) ( )
( ) ( )
1 1
1
N N
1
m m
M
m m
M


=










G m G m
J m
G m G m
L
L L L
L
(77)
N is the number of data points (value of magnetic field for
each time step) and M is the number of model parameters
(resistivity value assigned to each layer).
Under this linear approximation, the damped least
squares problem in equation (75) becomes
min ( ) ( ) ( ) ( )
2 2
2
+ W G m + J m m- d L m+ m
(78)
where the variable is m and m is a constant. In practice,
it is easier to formulate this as a problem in which the
variable is actually m m + which yields
min
( )( ) ( ) ( ) ( ) ( )
2 2
2
+ WJ m m+m - Wd- WG m +WJ m m L m+m

(79)
The solution is given by
24,26)

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

T T
T


+ m+ m = WJ m WJ m L L WJ m Wd m
,
(80)
by letting ( ) ( )

d m = d - G m + J(m)m and assuming that


J(m) and

d are constants.
This method is similar to the GaussNewton
method applied to the damped least squares problem.
Convergence could be improved by using line search to
pick the best solution along the line from the current
model to the new model given by (80). This iteration can
be used to solve (75) for a number of values of , and
then pick the largest value of for which the
corresponding model satisfies the data constraint
( ) WG m - Wd
. Alternative approach can be use to solve
(74) much faster in practice. At each iteration of the
algorithm, we pick the largest value of which keeps the
data
2
value of the solution within its bound. If this can
not be obtained, we pick the value of that minimizes the

2
. The above algorithm can be summarized by beginning
with a solution m
0
using the formula
( ) ( ) ( ) ( )
1
1 2

T T
k k k k k


+
= +



T
m WJ m WJ m L L WJ m Wd m
,
and then picking the largest value of such that
( )
2 1 2

+ k
m . If the value does not exist, then pick the
value of which minimizes
( )
2
2 1 k
m
+
.
6. Inversion results and discussion
The inversion scheme was tested with synthetic
data. The model resistivity structure is shown in Figure 4.
The total depth of the model is 10 km and is divided into
40 layers. The resistivity varies between 10 m and 500
m with rather sharp boundary in the deeper part. The
scheme is also intended to invert resistivity structure from
borehole data. In addition to individual inversion of
surface data and borehole data, joint inversion is also
performed. The surface data have a higher S/N at early
time channels whereas the borehole data have higher S/N
ratio at late time channels. Consequently the surface data
can be inverted to better resolve shallow structures, and
the borehole data can better detect deep structures. One of
the purposes of this scheme is to explore the merit of joint
inversion in TDEM 1-D problem. 3 % Gaussian noise is
added to all magnetic field data. The longest time channel
involved is 2.123 s.
Figure 3 shows the surface synthetic and
calculated data for the model as an example. Acceptable
agreement can be seen, however resolution decreases in
the late time. Figure 4 shows the model and the
corresponding inverted model from the surface data,
borehole data, and lastly from the joint inversion of both
the surface and borehole data. It can be seen that at
borehole depth of 200 m, the discrepancy among the
inverted models from individual inversions and joint is
not so obvious. The discrepancy between the true model
and the inverted ones is rather large in the shallow
structure, this is because the synthetic data is too short in
early times. It is clearly shows that the structure is better
recovered by the joint inversion of surface data and
borehole at 600 m. The results suggest that joint inversion
IJP Vol. 16 No. 4, 2005 123


will give better resolution to recover the resistivity
structure. The rate of convergence can be seen in the
Figure 5. At the early number of iteration the rms-misfit is
large for joint inversion of surface data and the deeper
borehole data and appears constantly beyond 14-th
iteration.
The inversion scheme was applied to the actual
data taken from the eastern part of Unzen volcanic area,
Simabara peninsula, Kyushu, Japan in 2001 and 2002
11)
.
An approximately 1.4 km long grounded-wire transmitter
was deployed to generate EM signal and the earths
transient responses were recorded using fluxgate
magnetometer with sampling rate of 32 and 128 Hz.
Robust stacking and deconvolution procedures were
applied to enhance the S/N ratio as well as removing the
effects of measurement system to the data. The vertical
magnetic data to be inverted then were normalized by
their DC values as discussed in the previous section. The
longest time channel was 5 seconds for this measurement,
hopefully longer enough to recover the structure of the
deeper part. . Example of the observed and calculated data
from four stations is shown in Figure 6. The inverted
model of the complete resistivity structure in the area is
shown in Figure 7. The existence of low resistive (or
conductive) layer at shallow depth over dominantly high
resistive layer at depth was revealed This conductive
layer corresponds with the existence of water saturated
layer which also characterizes the eruption mechanism
and the hydrothermal system in the area
28)
.

Figure 4. Comparison of synthetic data (dots) from model
in Figure 4 and the calculated data from inversion
(continuous line).

The above examples clearly show the usefulness
and effectiveness of TDEM method particularly in
volcanic and hydrothermal area as well as other fields
such as deep groundwater study and hydrocarbon. As for
Indonesia, a country whose, according to DVGHM
29)

website has about 129 volcanoes and 241 geothermal
areas, the application of transient EM methods still few.
To name one of it, only Merapi volcano that has been
more ore less studied by using the long-offset transient
electromagnetic (LOTEM) method from surveys in 1998,
2000 and 2001
e.g.,30)
. By this condition the scientific
challenge is still widely open, and TDEM study, together
with other methods will also enrich the knowledge toward
a better understanding of volcano related system
particularly in Indonesia.
1
10
100
1000
10000
1 10 100 1000 10000
D epth (m )
R
e
s
i
s
t
i
v
i
t
y

(
o
h
m
.
m
)
model
surface
bore200
joint200
joint400
joint600

Figure 5. The inverted model for surface data; borehole at
200 m depth; joint inversion of data from surface and 200
m, 400 m and 600 m depth respectively.


Figure 6. Rms misfit vs. iteration number.

Figure 7. Comparison of the observed and inverted data
for vertical magnetic fields data recorded at four different
stations.
124 IJP Vol. 16 No. 4, 2005




Figure 8. Resistivity structure revealed from TDEM data
in the western part of Shimabara peninsula, Japan.

Acknowledgement
We wish to express our thanks and high
appreciation to all members of Earth and Computational
Physics Research Group and the Physics Department
executives of ITB and also members of Earthquake
Research Institute, Tokyo University for all the support
and encouragement for this study.
References
1. Muller M., Hordt, A., Neubauer, F.M.,
Electromagnetic Technique Success at Vesuvius
points to Use in Forecasting Eruptions. EOS, 80.35
(1999).
2. Hrdt, A., and Mller, M., Understanding LOTEM
data from mountainous terrain, Geophysics, 65,
1113-1123 (2000).
3. Hermance, J. F., Electrical conductivity models of the
crust and mantle, Global Earth Physics, a Handbook
of Physical constants, AGU, pp. 90-205, (1995).
4. Keller, G. V., Rock and mineral properties, in :
Electromagnetic methods in applied geophysics, vol.
1 theory, edited by M. N. Nabigian, SEG,
Investigation in Geophysics, 3, 13-51 (1987).
5. Johnson, D.L., J. Koplik, and L.M. Schwartz, New
poresize parameter characterizing transport in
porous media, Phys. Rev. Lett, 57, 2564-2567 (1986).
6. Gough, D. I., Magnetometer array studies, Earth
structures and tectonic process, Rev. Geophys., 27,
141157, (1989).
7. Cagniard, L., Basic theory of the magnetotelluric
method of geophysical prospecting, Geophysics, 55,
605-645 (1953).
8. Kaufman, A.A. and Keller, G.V., Frequency and
transient sounding, p. 685, Elsevier, 1983.
9. Ward, S. H., and Hohman, G. W., Electromagneic
theory for geophycial applications, in Nabighian, M.
N., Ed., Electromagnetic method in applied
geophysics, Vol. 1 - Theory: Soc. Expl. Geophys
(1988).
10. Kanda, W., On a deep transient electromagnetic
soundingmeasurement and modeling, Ph.D Thesis,
The University of Tokyo, Japan, 1997.
11. Srigutomo, W., Resistivity structure of Unzen
Volcano from time domain electromagnetic (TDEM)
data and its implication to volatile-groundwater
interaction, Ph.D Thesis, The University of Tokyo,
Japan, 2002
12. Chave, D., Numerical integration of related Hankel
transforms by quadrature and continued fraction
expansion, Geophysics 48, 1671-1686 (1983).
13. Anderson, W.L., Numerical integration of related
Hankel transforms of orders 0 and 1 by adaptive
digital filtering, Geophysics, 44, 1287-1305 (1979).
14. Anderson, W. L., Algorithm 588: Fast Hankel
Transforms Using Related and Lagged Convolutions,
ACM Transactions on Mathematical Software
(TOMS), v.8 n.4, p.369-370 (1982).
15. Anderson, W.L, A hybrid fast Hankel transform
algorithm for electromagnetic modeling, Geophysics,
v. 54, no. 2, p. 263-266 (1989).
16. Knight, J.H. and Raiche, A.P., Transient
electromagnetic calculations using the Garver-
Sthefest inverse Laplace transform method,
Geophysics, 47, 47-50 (1982).
17. Villinger, H., Solving cylindrical geothermal
problems using Gaver-Stehfest inverse Laplace
transform, Geophysics, 50:10, 1581-1587 (1985).
18. McGillivray, P. R. and Oldenburg, D. W., 1990.
Methods for calculating Frchet derivatives and
sensitivities for the non-linear inverse problem, A
comparative study, Geophys. Prosp., 38, 499524.
19. Farquharson, C.G., and D.W. Oldenburg,
Approximate sensitivities for the electromagnetic
inverse problem, Geophysical Journal International,
126, pp235-252 (1996).
20. Farquharson, C.G., and D.W. Oldenburg,. Inversion
of time-domain electromagnetic data for a
horizontally layered Earth, Geophysical Journal
International, 114, 433-442 (1993).
21. McGillivray P.R., Oldenburg D.W., Ellis R.G., and
Habashy T.M., Calculation of sensitivities for the
frequency-domain electromagnetic problem,
Geophysical Journal International, 116 , 1-4 (1994).
22. Lanczos, C., Linear Differential Operators, Van
Nostrand, p 393, 1961.
23. Farquharson, C.G., D.W. Oldenburg and Y. Li, An
Approximate Inversion Algorithm for Time-Domain
Electromagnetic Surveys, Journal of Applied
Geophysics, 42, 71-80 (1999)
24. Constable, S.C., R.L. Parker, and C.G. Constable,
Occam's Inversion: a practical algorithm for
generating smooth models from EM sounding data,
Geophysics, 52, 289-300 (1987).
25. deGroot-Hedlin, C. and S.C. Constable, Occam's
inversion to generate smooth, two-dimensional
models from magnetotelluric data, Geophysics, 55,
1613-1624 (1990).
26. Aster, R. C., Borchers, B., and Thurber, C.,
Parameter estimation and inverse problems, Elsevier
Academic Press, 2004.
27. Zhang, Z. and Xiao, J., Inversions of surface and
borehole data from large-loop transient
IJP Vol. 16 No. 4, 2005 125


electromagnetic system over a 1-D earth,
Geophysics, 66, 1090-1096 (2001).
28. Kagiyama, T., Utada, H. and Yamamoto, T., Magma
ascent beneath Unzen Volcano, SW Japan, deduced
from the electrical resistivity structure, J. Volcanol.
Geotherm. Res., 89, 35-42 (1999).
29. http://www.vsi.esdm.go.id/volcanoes/
30. Commer, M., S. L. Helwig, A. Hrdt, and B. Tezkan,
Interpretation of long-offset transient
electromagnetic data from Mount Merapi, Indonesia,
using a three-dimensional optimization approach, J.
Geophys. Res., 110, B03207 (2005).

Vous aimerez peut-être aussi