You are on page 1of 12

170

Three-dimensional magnetotellurics modeling using edge-


based nite-element unstructured meshes*
Liu Changsheng
12
, Ren Zhengyong
13
, Tang Jingtian
1
, and Yan Yan
4
Abstract: Three-dimensional forward modeling is a challenge for geometrically complex
magnetotellurics (MT) problems. We present a new edge-based finite-element algorithm
using an unstructured mesh for accurately and efciently simulating 3D MT responses. The
electric field curl-curl equation in the frequency domain was used to deduce the H (curl)
variation weak form of the MT forward problem, the Galerkin rule was used to derive a linear
finite-element equation on the linear-edge tetrahedroid space, and, finally, a BI-CGSTAB
solver was used to estimate the unknown electric elds. A local mesh renement technique
in the neighbor of the measuring MT stations was used to greatly improve the accuracies
of the numerical solutions. Four synthetic models validated the powerful performance of
our algorithms. We believe that our method will effectively contribute to processing more
complex MT studies.
Keywords: Magnetotelluric modeling, edge-based nite-element, unstructured mesh, local
mesh renement
Manuscript received by the Editor June 2, 2008; revised manuscript received August 13, 2008.
*This work was nancially supported by National High Technology Research and Development Program (863 Program) (No.
2006AA06Z105, 2007AA06Z134).
1.School of Info-physics and Geometrics Engineering, Central South University, Changsha 410083, China.
2.Changsha Aeronautical Vocational and Technical College, Changsha 410124, China
3.Institute of Geophysics, ETH-Swiss Federal Institute of Technology, Zurich 8092, Switzerland.
4.Institute of Higher Education, Central South University, Changsha 410083, China.
APPLIED GEOPHYSICS, Vol.5, No.3 (September 2008), P. 170 - 180, 10 Figures.
DOI:10.1007/s11770-008-0024-4
Introduction
The MT technique is a distinct geophysical method
widely used in deep geological structural prospecting
for ground water sources and deep natural gas. The two
principal procedures involved in the MT method are
forward and inversion modeling. Forward modeling is
used to estimate the EM eld distribution under a given
geo-electrical structure and inversion is used to reduce
errors between the estimated results from forward
modeling and the real measured results from MT stations
by iteration. Accurate and efcient forward modeling is
critical to accelerating the inversion process. Forward
modeling calculations include analytical methods,
physical experiments, and numerical simulations.
Analytical methods are only suitable for a few simple
cases due to the need for rigorous models. Physical
experiments have uncertainty due to interference
from the experiment process and are sometimes time-
consuming. However, numerical simulations dont have
the disadvantages of the other methods. Currently, there
are mainly three numerical simulation methods (Avdeev,
2005; Ogawa, 2003): the popular integral equation method
(Chen et al., 1990; Lu et al., 2003), the finite-different
method (Tan et al., 2003), and the less used nite-element
method (FEM) (Liu et al., 2007; Shi and Xu, 1997; Tan et
al., 2006). The rst two methods are computationally fast
and simple in theory. However, if complex geo-electric
models are considered, the FEM is preferable.
171
Liu et al.
Since Coggon (1971) successfully used the finite
element method to solve the geo-electric eld problem for
the rst time, many researchers have given their attention
to extending the application of the FEM in various EM
elds. FEM arithmetic for the geo-electric eld is mainly
based on scalar-vector potential and Helmholtz or curl-
curl equations. The potential equations (Badea et al., 2001;
Mitsuhata and Uchida, 2004) guarantee the continuity
of the solved electric or magnetic potentials which is
a precondition for node-based FEM but brings extra
unknowns at each node. Moreover, if node-based FEM
is used for solving the Helmholtz or curl-curl numerical
solutions (Jin, 2002) there may be spurious solutions and
the discontinuity condition of the electric or magnetic eld
normal component will be unsatisfied. These problems
will lead to loss of accuracy for the numerical solutions.
Fortunately, an edge-based nite-element method (Sugeng,
1998) for solving the Helmholtz or curl-curl equations can
efciently avoid these problems.
In this paper, we introduce a new edge-based FEM
using three-dimensional unstructured meshes rather than
structured meshes (Nelson., 1998; Nam et al., 2007;
Sugeng, 1998; Yoshimura and Oshiman, 2002) to solve
3D MT forward problems. 3D unstructured meshes are
a set of tetrahedron elements owning irregular triangles
faces and edges which can efficiently approximate the
surfaces or curves of geological models and decrease
the errors from geometrical discretization. There are
two principal characteristics inherent in an unstructured
mesh. The first characteristic is the irregular triangle
faces which can unlimitedly approximate curved faces.
As a result, the unstructured mesh is more suitable
for geometrically complex geological models. 2D
unstructured mesh research cases can be seen in the
references (Key and Weiss, 2006; Yuguo and Kerry,
2007). The second characteristic is that no element
size is required for unstructured meshes so that the free
divergence condition of electric field can be satisfied.
Therefore, the numerical solutions quickly converge.
In contrast, structured meshes must avoid large sized
structural elements which complicate the structured
mesh generation.
Our unstructured mesh is generated based on the
Delaunay refinement algorithm (Si, 2007) which can
form a good sharp gradient mesh with nodes and
elements clustered in interesting areas. The advantage
is to locally refine the mesh in the interesting areas
while the mesh density is sparse in uninteresting areas.
After mesh generation, using the curl-curl equation of
the electric eld, we project the electric eld function
space into the H (curl) conforming Hilbert space, and
then scaling the H (curl) space into a linear Nedelec
edge-based nite element space, we can obtain the nal
sparse, complex, linear equation.
Following that, based on the electric field solution,
the magnetic field at the central node was recovered
by averaging a cluster of constant values surrounding
the node. As the densities of elements and mesh nodes
are rather high, the accuracy of the estimated magnetic
field is satisfactory. Finally, we obtain a high-quality
numerical solution with a fast computation time due
to the good shape of the generated meshes, optimum
gradients, and moderate nodes for each mesh in the
interesting areas. To validate our arithmetic in this paper,
a homogeneous half-space earth model, a horizontal
four-layer model, and two COMMEMI 3D models were
tested.
H (Curl) variation formulation
MT responses are described by the Maxwells
equations. Frequency domain MT modeling with time
factor e
-it
and neglecting the displacement current can
be written as (Harrington, 1961):
i



E H
H E
,

i



E H
H E, (1)

where E is the electric field, H is the magnetic field,
i is the imaginary unit, is the free air magnetic
permeability, is the angular frequency, and is the
conductivity distribution of the MT model.
After taking the curl operation on the second
component of equation(1) and substituting the result into
the rst component of equation(1), we get the curl-curl
equation for electrical eld E,
0 i E E . (2)
To uniformly define the electrical field in a bounded
solution domain, we must supply its tangent component
on the outer boundary (Harrington, 1961):

0
n E n, (3)
where n is the unit normal of the outer boundary, E
0
is a
known electric eld distribution on the outer boundary.
Generally, the known electric eld E
0
can be evaluated
from a 1D layer-earth model, without considering the
effects of terrain and anomalous conductive bodies. This
assumption is proven to be efcient when the modeling
area boundary is far enough from the anomaly (Nam et
al., 2007). Here we present an analytic solution for a 1D
homogeneous earth model.
172
Three-dimensional magnetotellurics modeling

1
1
1 1
0
( 1)
1
1 1
( 1)
, air
( 1)
earth
( 1)
z h
i
i C
C z
i h
C
e
i h

E (4)
where the positive z-axis of the Cartesian coordinate
system points vertically down, C is a given constant of
the top of the air layer (z = 0), h
1
is the thickness of the
air layer, and
1
is the skin depth of the homogeneous
earth layer.
Equations (2) and (3) are the boundary value problems
for MT forward modeling. By terms of the vector
identify formula B (

A)=A (

B)+(AB) and the


Greens integral equation (Sugeng, 1998; Nam et al.,
2007), its variation expression can be easily deduced as,

( , ) ( ), ( ) b f H curl E V V V
( , ) ( ), ( ) b f H curl E V V V , (5)
where
2
0
( ) { || ( ), n n } H curl L V V V V , L
2
() is
the second-order derivative continuous function space,
n is the unit normal vector on outer boundary, and b is a
bilinear formula:

( ) b i d eo
O
= V V O
}
E V E V
. (6)
If we regard the Dirichlet boundary condition by
Equation(3) as a source term, f is expressed as
0
f d

V E
. (7)
Edge-based nite element method
To solve the electric field expressed by equation(5)
using the vector finite-element method, we divide
the entire computing domain into a set of irregular
tetrahedron elements. Unlike the node-based finite-
element process, where three components of the
unknown electric field are defined at each node, a
tangential component of the electric eld is assigned to
each edge of the tetrahedral mesh which has powerful
capabilities to approximate geometrically complex
boundaries. Given a tetrahedron element designated by e,
the electric eld E can be approximated as

5
0
e e e
i i
i
E

E
(8)
where E
e
i
and
e
i
denote the tangential electric field
and the vector shape function on the ith edge of the ith
element, respectively. We designate the ith edge vector
shape function using the expressions (see Figure 1),

1 2 2 1
( )
e e e e e e
i i i i i i
l L L L L , (9)
where l
e
i
is the length of ith edge. L
e
i1
and L
e
i2
are the origin
(a) An irregular tetrahedron (b) A standard element tetrahedron
Fig.1 The edge vector tetrahedron element and its projections.
3
E5
E4
2 1
E6
E2
(x, y, z)
E3
E1
4 (0, 0, 1)
E1
E3
E5
E6
E4
(0, 1, 0)
3
2
1
(0, 0, 0) ( , , )
E2
node shape functions defined at each node of the ith
edge, which can be expressed by a local coordinate
system as

1,2 1,2
( , , )
e e
i i
L g , (10)
where the explicit form of function g
e
i1,2
is well-known (Xu,
1994; Zienkiewiez and Taylor, 2000).
As the divergence of
e
i
is zero, the requirements for
free divergence on the current density are easily satised
and since
e
i
is only continuous along the tangential
component of edges, the continuity of the tangential
component of the electric field is satisfied, but the
electric eld normal discontinuity remains. Additionally,
based on the vector-shape function, the tetrahedral
element can also efficiently suppress the spurious
physical solutions. In contrast, an extra penalty term
must be added to the node-based nite-element method
to avoid the spurious solutions which results in increased
computation complexity (Jin, 2002).
173
Liu et al.
After substituting equations(8) and(9) into equation(5),
we get a complex linear equation
AU = B, (11)
where U is the unknown vector at the tetrahedron
element edge. Matrixes A and B can are

0
e
e
e
ij i j i j
e
i i
L
i d
d
eo
O
c
= V V O
}
= I
}
A
B E

,


(12)

0
e
e
e
ij i j i j
e
i i
L
i d
d
eo
O
c
= V V O
}
= I
}
A
B E
,
where B
i
is nonzero just at the edges of the outer
Dirichlet boundary. Generally, we try to calculate
equation (12) by numerical integrals. The integral
equation (12) can be transformed from the irregular
(x, y, z) coordinates (see Figure 1a) into a local coordinate
system (, , ) (see Figure 1b), resulting in

( , , )
0
( ) | | d d d
( )
p p p
ij i j i j
p
i j i j
i
i
i
c q
eo c q
eo
A
=
= V V
}
= A V V
A J

( , , )
0
( ) | | d d d
( )
p p p
ij i j i j
p
i j i j
i
i
i
c q
eo c q
eo
A
=
= V V
}
= A V V
A J

,
(13)
where is the volume of the irregular tetrahedron
element (0 + + 1), J is the Jacobian matrix for
projecting coordinates from the irregular element
e
to
the regular element
e
, and (
p
,
p
,
p
) are the Gaussian
integral coordinates. Similarly, we get

0 ( , , )
0
( )
q q q
q
e e
i i i
i
L

B E , (14)
where L
e
i
is the length of ith edge of the eth element,
(
q
,
q
,
q
) is the Gaussian integral coordinates on the
edge. Here, the three-point Gaussian integral equation for
tetrahedra should be adopted to calculate the term
i

j

in the equation (13) accurately.
Using the right expression of

j
and
i
(see
Jin, 2002) and repeating element by element, we
can set up the global linear equation (11). Then, we
can use the Kylov subspace BICGSTAB solver with
LU preconditioner (Ruecker et al., 2006) to solve
the equation (11). After obtaining tangential electric
elds at each edge, the electric eld at each node (MT
station) is also calculated by local element projection.
Meanwhile, magnetic field H at each node can be
calculated by


1
i
H E
(15)
It is obvious that the magnetic eld is a constant for
each element. To improve accuracy, we use the average
magnetic eld H at adjacent node elements. Once both
the electric eld E and magnetic eld H are calculated,
we can estimate the apparent resistivities corresponding
to each measured surface point in X Y mode as


2 2
1 1
| |
x
y
E
Zxy
H


(16)
20 km
20km
10 km
0 ohm-m
x
y
z air
2000 m
earth
10 km
0 ohm-m
2000m 2000m
(a) The spatial node distribution.
(b) Mesh-section at z = -10 km. (c) Mesh-section at y = 0
Fig. 2 The nite-element mesh used for the homogenous earth model (ParaviewTeam, 2008) with 9407 nodes and 58741
elements. Red dots are the nodes.
174
Three-dimensional magnetotellurics modeling
and the corresponding phase variable


Im( )
arctan( )
Re( )
Zxy
Zxy
. (17)
Numerical examples
To test our method, we developed a 3D MT forward
modeling program based on the open source PHG
library (Zhang, 2008). In this program, an air-layer
with a small threshold conductivity is assumed and the
boundary conditions can be set up based on the depths,
resistivities, and number of earth layers. To efficiently
decrease storages requirements, the compressed row
storage (CRS) format is used for sparse matrixes; the
number of nonzero values for each row is in the range of
25-100. The resulting complex linear equation is solved
by translation into a double real linear equation. The
matrix is no long symmetric. To solve this problem, we
use the BICGSTAB solver with LU preconditioner. If
the residual norm is less than 1e-10, the computational
process is terminated. Furthermore, the magnetic
field values from the equation (15) may cause large
numerical errors. However, it is reasonable to average
a large quantity of finite element constant values
around the central node. Based on the unstructured
meshes generated by the TetGen library (Si, 2007)
and Rueckers results (Ruecker et al., 2006), we keep
a balance between accuracy and number of nodes by
enforcing 1.2 radius-edge ratio requirement for each
element.
Homogenous earth model
First, a homogenous half space earth model with
1 ohm-m resistivity and 10
10
ohm-m air layer is
considered. Our MT measuring stations are located
over an x-axis range of -1000 m to +1000 m at a 10
m interval. The electric structure is shown in Figure 2
with the corresponding nite-element mesh.
From Figure 2 we can see that the nodes and
elements are dense close to the measuring stations and
sparse far from the measuring region. These sharply
different gradient meshes suggest that numerical
solutions with higher accuracy can be obtained in the
vicinity of measuring stations and numerical results
accuracy are lower in the areas far from the measuring
stations. Figures 3a and 3b show the apparent
resistivity and phase curves at 1.0 Hz frequency. In the
outer boundary with |x| > 1000 m, the relative errors
at measuring stations are 1.0% - 2.4% and that are
0.01% - 1.0% in the inner range of |x| 1000 m. The
relative phase errors for |x| > 1000 m are 0.5%-1.5%
(a) Apparent resistivities at 1.0 Hz. (b) Phases at 1.0 Hz.
(c) Apparent resisitivities at 0.1 Hz. (d) Phases at 0.1 Hz.
Fig. 3 Comparison of analytical solutions of apparent resistivities and phases for the homogenous earth model in XY mode.
x (m)
A
p
p
a
r
e
n
t

R
e
s
i
s
t
i
v
i
t
y

(
o
h
m
-
m
)
-1000 -500 0 500 1000
1.0 Hz
1.3
1.2
1.1
1.0
.9
.8
.7
60
55
50
45
40
35
P
h
a
s
e

(
d
e
g
r
e
e
)
-1000 -500 0 500 1000
1.0 Hz
x (m)
x (m)
A
p
p
a
r
e
n
t

R
e
s
i
s
t
i
v
i
t
y

(
o
h
m
-
m
)
-1000 -500 0 500 1000
0.1 Hz
1.3
1.2
1.1
1.0
.9
.8
.7
60
55
50
45
40
35
P
h
a
s
e

(
d
e
g
r
e
e
)
-1000 -500 0 500 1000
0.1 Hz
x (m)
175
Liu et al.
air
1
2
3
4
20 km 20 km
100 ohm-m, 20 km
10 ohm-m, 10 km
100 ohm-m,
1 km
100 ohm-m,
250 m
10 ohm-m, 10 km
(a) Spatial node distribution and electric structure. (b) 3 D Finite-element mesh.
(c) Mesh-section at z = -10 km.
Fig. 4 The nite-element mesh used for the four-layer model
which has 5165 nodes and 29286 elements.
but that are 0.05%-0.5% in the inner range of |x| 1000
m. The same results were obtained at a low frequency
of 0.1 Hz (see Figures 3c and 3d). The number of
iterations of BICGSTAB with LU preconditioner with
a 1e-10 threshold for 1.0 and 0.1 Hz are 348 and 422,
respectively. The computation times are 4.5 and 5.8
minutes on a P4 2.8 G CPU with 512 MB RAM PC with
Fedora 8.0 operating system. This single earth model
not only validates our code but also shows the powerful
performance of the local refinement techniques at
measuring stations in FEM modeling.
Four-layer model
The second tested model is a four horizontal layers
model. Its spatial dimension is 20 km 20 km 31.25
km with each layer depth and conductive structure
shown in Figure 4a. The boundary conditions are taken
from the four-layer model analytical solutions (Xu,
2001). As the first layer is rather thin, the nodes and
elements are clustered on the air-earth interface and in
the rst layer which makes sure we get highly accurate
MT results in a wide frequency range of 0.001Hz to
1000 Hz. There are 121 measuring stations in a line
with a length of 3000 m on the x-axis. The model node
distribution and finite-element meshes are shown in
Figures 4a and 4b. The numerical apparent resistivity
and phase curves in the range of 0.001 Hz -1000 Hz
are shown in Figure 5. The sounding point is at the
coordinate origin. Our results agree with the 2D node-
based nite-element program (Shi and Xu, 1997) as well
as with the analytical solution.
3D Test model
COMMEMI 3D-1 model This is an internatio nal
universal test model, COMMEMI 3D-1 model, in which
a hexahedral anomalous low resistivity body is buried.
The resistivitiy of the anomalous body is 0.5 ohm-m and
the background resistivity is 100 ohm-m, so the contrast
ratio is 1:2000. High contrasting models are usually
difficult for common numerical modeling programs to
process (Mitsuhata and Uchida, 2004). The model size is
100 km along the x and y axes and 50 km for the air layer
and 50 km for the earth layer in the z-direction. The MT
measurement lines are 3 km in both x and y directions.
We know from model 1 that a highly accurate numerical
solution can be obtained by locally refining nodes and
elements near MT lines. We selected 121 MT measuring
points for generating dense nodes and meshes in their
176
Three-dimensional magnetotellurics modeling
10
-3
10
-2
10
-1
10
0
10
1
10
2
10
3
1
10
100
Analytical solution
2D fem
Tet 4
T (s)
A
p
p
a
r
e
n
t

r
e
s
i
s
t
i
v
i
t
y

(
o
h
m
-
m
)
(a) Apparent resistivitiy (b) Phase
(a) Mesh-section at z = -50 km (b) Mesh-section at x = 0
Fig. 5 Comparison of our algorithm (TET4) with analytical and 2D node-based nite-element algorithm solutions. 2DFEM denotes
Shi and Xus algorithm (Shi and Xu, 1997) and TET4 stands for our 3D edge-based nite-element algorithm.
vicinity using our mesh generation algorithm. Further
from the measuring points, less dense elements and
meshes are generated. The sharp gradient meshes from the
three axial symmetry planes are shown in Figure 6. With
regard to the boundary conditions, we use a horizontal
layer numerical solution (equation) with the 100 ohm-m
background resistivity of the homogenous layer.
10
-3
10
-2
10
-1
10
0
10
1
10
2
10
3
0
40
T (s)
P
h
a
s
e

(
d
e
g
r
e
e
)
80
60
20
Analytical solution
2D fem
Tet 4
(c) Mesh-section at y = 0
Fig. 6 The nite-element mesh used for the COMMEMI 3D-1
model (ParaviewTeam, 2008) containing 10553 nodes and
65896 elements.
To validate our algorithm, we compared our results
with those of the COMMEMI project (Zhdanov et
al., 1997), the staggered-grid finite-difference method
(Sasaki, 2001), and the vector-scalar potentials-based
T- algorithm (Mitsuhata and Uchida, 2004). We uses an
irregular unstructured mesh with 10553 nodes and 65896
elements for the XY mode and used a working frequency
of 0.1 Hz. After 541 iterations, the Bi-CGSTAB solver
terminated the iterations. The residual error is less than
1e-10. In Figure 7, we show the apparent resistivitiy
and phase curves measured on the lines along the x and
y-axis. Obviously, our numerical solutions agree well
with the results of COMMEMI, finite-difference, and
T- algorithms. However, we found some unstable
solutions exist at the both ends of the MT lines. From
Figure 6 we can see that the element and node density
at both ends of the MT line are sparser than along the
lines. The sparser elements and mesh must cause the
inaccuracy. This shows the importance of the local mesh
renement techniques.
177
Liu et al.
Sasaki (SFD)
Commemi
To3dmt
Tet 4
100
10
1
A
p
p
a
r
e
n
t

r
e
s
i
s
t
i
v
i
t
y

(
o
h
m
-
m
)
0 500 1000 1500 2000 2500 3000
x (m)
To3dmt
Sasaki (sfd)
Tet 4
P
h
a
s
e

(
d
e
g
r
e
e
)
0 500 1000 1500 2000 2500
3000
70
60
50
40
30
x (m)
-20 0 (km)
X
Z
(km)
30
10
10 ohm-m 1 ohm-m 10 ohm-m
100 ohm-m
0.1 ohm-m
20
100 ohm-m
X
20
-20
(km)
10 ohm-m
0
1 ohm-m
10 ohm-m
20
(km)
Y
100 ohm-m
-20
(a) Apparent resistivitiy along the x-axis line.
(b) Phase along the x-axis line.
(c) Apparent resistivity along the y-axis line.
(d) Phase along the y-axis line.
(a) Section view at y = 0. (b) Section view at z = 0.
Fig.8 The COMMEMI 3D-2 model
Fig. 7 Comparison of the apparent resistivity and phase curves obtained by our algorithm with other numerical algorithms at 0.1Hz
frequency in XY mode.
To3dmt
Commemi
Sasaki (SFD)
Tet 4
100
10
1
A
p
p
a
r
e
n
t

r
e
s
i
s
t
i
v
i
t
y

(
o
h
m
-
m
)
0 500 1000 1500 2000 2500 3000
x (m)
To3dmt
Sasaki (sfd)
Tet 4
P
h
a
s
e

(
d
e
g
r
e
e
)
0 500 1000 1500 2000 2500 3000
70
60
50
40
30
x (m)
3D Complex model
COMMEMI 3D-2 model Finally, we consider a
complex model, the COMMEMI 3D-2 model (Zhdanov
et al., 1997) with its geometrical conductivity shown in
Figure 8. The solution domain is 200 km 200 km 150
km in the x, y, and z directions, respectively. The depth
of the air layer is 50 km with a 10
10
ohm-m resistivity.
Our MT stations are located along the x axis in the range
of -55 km to 55 km. The analytical solutions of the three-
layer model are adopted as the boundary conditions with
10 ohm-m resistivity in the rst layer. The nite element
mesh is shown in Figure 9. In Figure 10, we compare our
selected results with the solution obtained by the vector-
scalar T- potentials finite-element method (Mitsuhata
and Uchida, 2004) at 0.001 Hz. In the XY mode, with
20355 nodes and 102938 elements, our results agreed
well with Mitsuhata and Uchidas.
178
Three-dimensional magnetotellurics modeling
x (km)
-80 -60 0 20
.01
1
100
To3dmt
Tet 4
A
p
p
a
r
e
n
t

r
e
s
i
s
t
i
v
i
t
y

(
o
h
m
-
m
)
.1
10
-40 -20 40 60 80
(a) Mesh-section at y = 0. (b) Mesh-section at z = 0 km.
Fig. 9 The nite-element mesh used for the COMMEMI 3D-2 model with 20355 nodes and 102938 elements.
(a) Apparent resistivity (b) Phases
Fig.10 The comparison of our results with the T- nite-element method. The responses were calculated along the x-axis at
0.001Hz in the XY mode.
P
h
a
s
e

(
d
e
g
r
e
e
)
100
90
80
70
50
x (m)
60
-80 -60 0 20 -40 -20 40 60 80
To3dmt
Tet 4
Conclusions
In this study, we investigated the tetrahedron edge-
based nite element method using unstructured meshes
for modeling 3D MT models. Highly accurate numerical
solutions can be achieved by locally rening the meshes
in the neighborhood of the interesting measuring
stations. Keeping sparse meshes in other regions, the
solution accuracy is improved and the computational
time is reduced, making fast MT forward modeling
possible. Using three simple examples, we show that
combining the edge-based FEM with unstructured
meshes has proven to provide efcient forward modeling
for MT problems. However, this work is only an initial
start. In the future, more complex cases including
topography will be studied. Additionally, we have also
recognized that if we want to decrease dependency on
artistic local renement mesh techniques, an automatic
adaptive 3D mesh refinement algorithm in cooperation
with a sophisticated goal-oriented error estimator (Oden
and Prudhomme, 2001) should be developed for the 3D
electromagnetic curl-curl equation.
Acknowledgements
The authors thank Dr. Hang Si and Prof. Linbo Zhang
for making their state-of-the-art open source libraries,
TetGen and PHG, freely available. Also, the authors
thank the Paraview teams for developing an excellent
mesh data visualization tool to show the nite-element
meshes.
References
Avdeev, D.B., 2005, Three-dimensional electromagnetic
modelling and inversion from theory to application:
Surveys in Geophysics, 26(6), 767 799.
Badea, E.A., Everett, M.E., Newman, G.A., and Biro,
179
Liu et al.
O., 2001, Finite-element analysis of controlled-source
electromagnetic induction using Coulomb-gauged
potentials: Geophysics, 66(3), 786 799.
Chen, J. P. , Chen, L. S. , and Wang, G. E. , 1990,
Magnetotelluric modelling of three-dimensional bodies
in layered earths: Chinese Journal of Geophysics (in
Chinese), 33(4), 480 488.
Coggon, J.H., 1971, Electromagnetic and electrical
modeling by the finite element method: Geophysics,
36(1), 132 155.
Harrington, R.F., 2001, Time-harmonic electromagnetic
elds: IEEE&Wiley, USA, 106 107.
J i n, J . , 2002, The f i ni t e el ement met hod i n
electromagnetics: John Wiley & Sons, USA, 223 224.
Key, K. , and Wei ss, C. , 2006, Adapt i ve fi ni t e-
element modeling using unstructured grids: The 2D
magnetotelluric example: Geophysics, 71(6), G291
G299.
Liu, X.J., Wang, J.L., and Yu, P., 2007, Secondary eld-
based 2D magnetotelluric numerical modeling by
finite element method: Journal of TongJi University
(Natural Science Edition) (in Chinese), 35(8), 1113
1117.
Lu, L.Y., Zhang, B.X., and Bao, G.S., 2003, Modeling
of three-dimensional magnetotellurics response for
a linear earth: Chinese Journal of Geophysics (in
Chinese), 46(4), 568 575.
Mitsuhata, Y., and Uchida, T., 2004, 3D magnetotelluric
modeling using the T-Omega finite-element method:
Geophysics, 69(1), 108 119.
Nelson, E.M., 1998, Advances in 3D electromagnetic
finite element modeling: Proceedings of the IEEE
1997 Particle Accelerator Conference, Vancouver,
Canada, 1837 1840.
Nam, M.J., Kim, H.J., Song, Y., Lee, T.J., Son, J.S.,
and Suh, J.H., 2007, 3D magnetotelluric modelling
i ncl udi ng sur f ace t opogr aphy: Geophysi cal
Prospecting, 55(2), 277 287.
Oden, J.T., and Prudhomme, S., 2001, Goal-oriented
error estimation and adaptivity for the nite element
method: Computers & Mathematics with Applications,
41(5 6), 735 756.
Ogawa, Y., 2003, On two-dimensional modeling of
magnetotelluric eld data: Surveys in Geophysics, 23
(2 3), 251 273.
ParaviewTeam, 2008, Paraview: an open-source multi-
platform parallel visualization application: http://
www.paraview.org/New/index.html, Accessed on Feb
10, 2008.
Ruecker, C., Guenther, T., and Spitzer, K., 2006,
3D modeling and inversion of dc resistivity data
incorporating topography-I-Modeling: Geophysical
Journal International, 166(4), 495 505.
Sasaki, Y., 2001, Full 3D inversion of electromagnetic
data on PC: Journal of Applied Geophysics, 46(1), 45
54.
Shi, M.J., and Xu, S.Z., 1997, Finite element method
using quadratic element in MT forward modeling:
Chinese Journal of Geophysics (in Chinese), 40(3),
421 430.
Si, H., 2007, TetGen: A quality tetrahedral meshes
generator and three-dimensional delaunay triangulator:
http://tetgen.berlios.de, Accessed on Oct 10, 2007.
Sugeng, F., 1998, Modeling the 3D TDEM response
using the 3D full-domain finite-element method
based on the hexahedral edge-element technique:
Exploration Geophysics, 29(4), 615 619.
Tan, H.D., Tong, T. and Lin, C.H., 2006, The parallel 3D
magnetotelluric forward modeling algorithm: Applied
Geophysics, 3(4), 197 202.
Tan, H.D., Yu, Q.F., John, B., and Wei, W.B., 2003,
Magnetotelluric 3D modeling using the staggered-
grid finite difference method: Chinese Journal of
Geophysics (in Chinese), 46(5), 705 711.
Xu, S.Z., 2001, The boundary element method in
geophysics: Society of Exploration Geophysicists,
USA, 97 129.
Yoshi mura, R. , and Oshi man, N. , 2002, Edge-
based finite element approach to the simulation
of geoelectromagnetic induction in a 3-D sphere:
Geophysical Research Letters, 29(3), 1039 1045.
Li, Y.G., and Kerry, K., 2007, 2D marine controlled-
source electromagnetic modeling: Part 1-an adaptive
nite-element algorithm: Geophysics, 72(3), WA51
WA62.
Zhang, L.B., 2008, PHG: a toolbox for developing
parallel adaptive nite element programs: http://lsec.
cc.ac.cn/phg/index_en.htm, Accessed on Feb 10, 2008.
Zhdanov, M.S., Varentsov, I.M., Weaver, J.T., Golubev,
N.G., and Krylov, V.A., 1997, Methods for modelling
electromagnetic elds Results from COMMEMI--the
international project on the comparison of modelling
methods for electromagnetic induction: Journal of
Applied Geophysics, 37(3 4),133 271.
Zienkiewicz, O.C., and Taylor, R.L., 2000, The finite
element method: Elsevier, USA, 79 145.
180
Three-dimensional magnetotellurics modeling
Tang Jingtian graduated with
a PhD f r om t he I ns t i t ut e of
Appl i ed Geophysi cs, Cent ral
South University, in 1992. He
was a research Professor at the
U.S. Lawrence Berkeley National
Laboratory in 1998-1999. He is
currently a full Professor at the
Institute of Applied Geophysics,
Central South University.
Yan Yan graduated with a Masters
degr ee f r om t he I nst i t ut e of
Higher Education, Central South
University, in 2008. Her research
i nt er est i s nat ur al r esour ces
evaluation.
Liu Changsheng graduated with a
Masters degree from the Institute
of Computer and Technologies,
Cent ral Sout h Uni versi t y. He
is now a Doctoral student for
EM modeling at the Institute of
Applied Geophysics, Central South
University.
Ren Zhengyong (Corresponding
author), graduated with a Masters
degr ee f r om t he I nst i t ut e of
Applied Geophysics, Central South
University, in 2007. He is now
a PhD candidate at the Institute
of Geophysics, Department of
Earth Science, ETH Zurich-Swiss
Federal Institute of Technology.
His research interest is mainly MT
inversion. Email: renzhengyong@gmail.com
Reproducedwith permission of thecopyright owner. Further reproductionprohibited without permission.