Journal of Applied Geophysics 133 (2016) 82–91

Contents lists available at ScienceDirect

Journal of Applied Geophysics

journal homepage: www.elsevier.com/locate/jappgeo

Interpretation of very low frequency electromagnetic measurements in

terms of normalized current density over variable topography
Anand Singh ⁎, S.P. Sharma
Department of Geology and Geophysics, Indian Institute of Technology, Kharagpur, 721302, West Bengal, India

a r t i c l e i n f o a b s t r a c t

Article history: A 2D inversion approach is developed to interpret VLF electromagnetic measurement recorded over variable to-
Received 9 December 2015 pography. To depict the variable topography accurately, an octree mesh discretization is incorporated. Subsurface
Received in revised form 15 July 2016 structure is modeled in terms of apparent current density distribution and compared with the inversion results
Accepted 20 July 2016
for actual resistivity distribution obtained using numerical techniques. The study demonstrates that the results
Available online 25 July 2016
obtained using both approaches (current density and resistivity distribution) are comparable, but due to analyt-
Keywords:
ical expression, current density imaging is faster. The conjugate gradient method is used to reduce the computa-
VLF electromagnetic tion time and storage space when solving the matrix equations, resulting in feasible and practical imaging
Subsurface imaging inversion of VLF data. The preconditioned matrix, which is determined by the distances between the blocks
Current density and observation points, has an important function in improving the resolution. In case of flat earth,
Inversion preconditioned conjugate gradient inversion of data results in images that are comparable to those obtained
Topography using resistivity inversion. We also test whether topography variation in the order of skin depth is significant
to incorporate topography in the modeling. The example of a topographical field VLF data inversion shows the
efficacy of the presented approach to appraise the subsurface structure in terms of current density.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction
The Very Low Frequency (VLF) electromagnetic method is a popular
plane wave electromagnetic (EM) method to map shallow geological
conductors on large scale due to its low cost and fast survey speed.
The VLF method has been successfully used in solving a variety of geo-
logical problems in geothermal investigation (Baranwal and Sharma,
2006; Zlotnicki et al., 2006), geotechnical investigation (Sharma et al.,
2010; Sungkono et al., 2014b), groundwater contamination and waste
management studies (Poddar and Rathor, 1983; Monteiro Santos
et al., 2006), archeological investigations (Drahor, 2006), mineral,
groundwater and environmental investigations (Paterson and Ronka,
1971; Philips and Richards, 1975; Bernard and Valla, 1991; Benson
et al., 1997; Tezkan, 1999; Sharma and Baranwal, 2005; Mohanty
et al., 2011; Sundararajan et al., 2011; Sungkono et al., 2014a; Sungkono
et al., 2015; Biswas and Sharma, 2016). The method has been used for
land as well as airborne surveys (Arcone, 1978; Pedersen and Oskooi,
2004; Pedersen et al., 2009).
Different qualitative and quantitative approaches have been devel-
oped to interpret VLF electromagnetic data. Quantitative interpretation
⁎ Corresponding author.
E-mail addresses: anand@gg.iitkgp.ernet.in (A. Singh), spsharma@gg.iitkgp.ernet.in
(S.P. Sharma).
of VLF data has been developed to delineate 2D and 3D subsurface struc-
tures (Beamish, 1994, 1998, 2000; Sharma and Kaikkonen, 1998;
Kaikkonen and Sharma, 1998; Baranwal et al., 2011; Kaikkonen et al.,
2012; Kamm and Pedersen, 2014). The aforesaid interpretation tech-
niques need a large amount of memory for the computing as well as a
great deal of computation time. To speed up computation, analytical
techniques were developed for imaging of the subsurface conductors.
The filtering techniques developed by Fraser (1969) and Karous and
Hjelt (1983) are frequently used to interpret VLF data qualitatively.
Boukerbout et al. (2003) interpreted isolated conductors from VLF
data using the wavelet approach. A more advanced filtering technique
was discussed by Pedersen and Becken (2005) to construct correspond-
ing equivalent current density distribution images, which would also
provide quantitative information about the depth of the conductor. Re-
cently, Singh and Sharma (2015) showed better results than obtained
through the filtering techniques developed by Karous and Hjelt
(1983) and Pedersen and Becken (2005). They developed a subsurface
imaging technique using inversion strategy rather than an inverse filter.
Singh and Sharma (2015) also revealed that the results get closer to
those obtained by the aforementioned quantitative interpretation tech-
niques. All the aforementioned analytical approaches have been devel-
oped for flat earth models.
In plane wave electromagnetic methods, small-scale rugged obser-
vation surfaces have little or no effect. However, a large-scale rugged

http://dx.doi.org/10.1016/j.jappgeo.2016.07.018
0926-9851/© 2016 Elsevier B.V. All rights reserved.
 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91
observation surface can affect the responses even at frequencies below
10 Hz (Chouteau and Bouchard, 1988; Jiracek et al., 1989; Wannamaker
et al., 1986). Since the VLF transmitters operate in a high frequency
range (between 5–30 kHz), topographical variation in the order of
10 m will also affect the response depending on the resistivity of the
area and the frequency used (Arcone, 1978; Fischer, 1989; Karous,
1979; Liu and Becker, 1992). There are two regular approaches for the
inversion to deal with topographical surfaces. In the first approach,
topographical surfaces are considered a field distortion and corrections
are made accordingly (Baker and Myers, 1980; Chouteau and Bouchard,
1988; Jiracek et al., 1989). In the second approach, topography is incor-
porated explicitly during the modeling and inversion (Baba and Chave,
2005; Key et al., 2006; Franke et al., 2007; Baranwal et al., 2011). To deal
with the surface topography, rectangular discretization is commonly
used. As shown in Fig. 1a, the near surface and air block is divided into
two parts separated by the topographical surface. Such discretization
produces large errors in describing rugged terrains. An octree-mesh
generation divides the subsurface blocks into smaller blocks to depict
the surface topography accurately (Fig. 1b). This technique has been
used in electromagnetics (Ascher and Haber, 2001; Haber and
Heldmann, 2007; Baranwal et al., 2011), large-scale earthquake model-
ing (Bielak et al., 2005), and airborne magnetic data interpretations
(Davis and Li, 2013).
In the present study, an attempt is made to compute the vertical
component of the magnetic field for a given current density distribution
over topographic variations. Further, a 2D inversion approach incorpo-
rating the preconditioned conjugate gradient approach is developed
for inversion of the VLF data directly for models including topography.
Initially, the inversion approach was tested on a flat earth model
and the efficacy of the preconditioner (determined by the distances
between the blocks and the observation points) in improving the
quality of the imaging technique is studied. The developed imaging
technique is also compared with the actual resistivity distribution
obtained using a rigorous resistivity inversion (Baranwal et al.,
2011) on the flat earth model. Finally, the application of the approach
is demonstrated using data measured over undulating topography
(synthetic as well as real data) and compared with the results obtained
using resistivity inversion.
2. Computational approach
In the VLF method, the ratio of the vertical component of the mag-
netic field to the horizontal component of the total field is measured.
Further, this ratio is transformed into a real anomaly and expressed as
a percentage (Smith and Ward, 1974).


Hz
%Real anomaly ¼ 100  Re
Hx
where, Hx = HPx + HSx (term HPx denotes primary and HSx denotes the sec-
ondary horizontal magnetic field) and Hz denotes the vertical magnetic
field. Since the primary horizontal field is uniform in the study area, the
Fig. 1. Discretization approaches to deal with the surface topography. (a) Rectangular division and (b) octree-mesh division.
83
measured real anomaly is proportional to the vertical component of the
magnetic field. The vertical component of the magnetic field in the
study area revealed the presence of 2D and 3D lateral conductivity var-
iations as the homogeneous half-space and 1D variations in resistivity
do not produce the vertical secondary magnetic field. Since Hsx b HPx , it
is an acceptable approximation to neglect the anomalous horizontal
component of the magnetic field. Hx in Eq. (1), can be considered uni-
form in the study area. In the present work, Hx is considered as unity
in order to equalize the relationship between the real anomaly and
the vertical component of the magnetic field.
2.1. Computation of the vertical component of the magnetic field from 2D
current density distribution
Using Biot-Savart law, Karous and Hjelt (1983) and Singh and
Sharma (2015) computed the vertical component of the magnetic
field for a given two-dimensional current density distribution. Karous
and Hjelt (1983) assumed that a small block (dxdz) will have a uniform
current density and derived the vertical component of the magnetic
field. Further, they developed an efficient filter to calculate equivalent
current density from the observed magnetic field. However, Singh and
Sharma (2015) derived a frequency dependent analytical expression.
Since topographic data contains both positive and negative z values, it
is not acceptable to assume that current density decreases exponentially
with depth. So, a uniform current density in a particular small block was
considered (the same law of physics that has been considered in the
Karous and Hjelt (1983) filtering approach) and an analytical solution
was derived to compute the vertical component of the magnetic field
along the profile for a given 2D rectangular block (Appendix 0).
Fig. 2 presents an octree-mesh discretization where smaller ele-
ments are used to depict the near surface zone; as depth increases the
size of the elements increases accordingly. The observation points at
the Earth's surface were located between −250 m and 250 m with an
increment of 10 m. The plotted subsurface is divided into 466 cells.
The cells are numbered from left to right and top to bottom from 1 to
466 (Fig. 2a). Further, for the model including topography, the observa-
tion points along the topographic surface were located between
−500 m and 500 m with an increment of 25 m and similarly, the cells
are numbered from left to right and top to bottom, correspondingly,
from 1 to 249 (Fig. 2b) in the present study.
The vertical component of the magnetic field of rectangle ABCD at
the observation point (0, 0) is expressed as (Fig. 3)
! !3
2
x2kþ1 þ z2kþ1 x2kþ1 þ z2k
z
6 kþ1 ln −z k ln 7
6 x2k þ z2kþ1 x2 þ z2k 7
6 ! k 7
6 7
k jk 6 −1 xkþ1 ðzkþ1 −zk Þ 7
ΔHz ¼ 6 þ2xkþ1 tan 7 ð2Þ
ð1Þ 4π 6 xkþ1 þ zk zkþ1
2 7
6 ! 7
6 7
4 −1 x ðz
k kþ1 −z k Þ 5
−2xk tan
x2k þ zk zkþ1
where, ΔHkz is the vertical magnetic field due to the kth rectangular
84 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91

Fig. 2. Octree-mesh parameterization to compute the vertical component of the magnetic field for models not accounting for topography and including topography. (a) To compute the
response on the flat earth model, the total number of measuring points is 51 (black arrow). The plotted subsurface is divided into 466 cells. The cells are numbered from left to right and top
to bottom, correspondingly, from 1 to 466. (b) Similarly for models including topography the total number of measuring points is 51 (black arrow) and the cells are numbered from left to
right and top to bottom, correspondingly, from 1 to 253.

block, jk is the current density of the kth block, and (xi, zi) is the coordi- Let (xi0, zi0)(i = 1, 2, 3, ... , M) be the coordinates of the data point
nate of the kth rectangular block. Total vertical field due to all rectangu- where M is the number of data points along the profile. Then the vertical
lar blocks can be written as summation of the fields due to the component of the magnetic field at the ith data point is given by
individual blocks at the observation points

X
N

X
N di ¼ ai;k jk ð4Þ
Hz ¼ ΔHkz ð3Þ k¼1
k¼1

where, jk is the current density at top of the kth block and ai ,k is matrix
where, N is the number of the rectangular blocks. element representing the influence of the kth block on the ith data point.

Fig. 3. The vertical component of the magnetic field of rectangle ABCD at the observation point (0, 0).
 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91 85

Using Eq. (2), the exact expression for ai ,kis The Eq. (4) can be written in matrix notation

d ¼ G:m ð6Þ

2  2  2 ! 3
  x −xi þ zkþ1 −zi0 where, d is the observed vertical component of the magnetic field, m is
6 zkþ1 −zi0 ln kþ1 0  2 7 the model parameter vector (current density), G is the M × N dimen-
6 2 7
6 xk −xi0 þ zkþ1 −zi0 7
6     ! 7 sional sensitivity matrix of which the elements are the anomalies
6   2 2 7
6 x −x i
þ z −z i 7 caused by blocks with unit current density and M and N are the number
6− zk −zi0 ln kþ1 0 
k 0
 7
6 2 2 7
1 6
6
xk −x0 þ zk −z0
i i 7
7 of observed data and blocks, respectively.
aði; kÞ ¼        ! ð5Þ
4π 6
6   x −x i
z −z i
− z −z i
7
7
6 þ2 xkþ1 −x tan
i −1

kþ1 0
2 
kþ1 0

k 0
 7
6 0 7 3. Inverse modeling
6 xkþ1 −x0 þ zk −z0 zkþ1 −z0 7
i i i
6 ! 7
6          7
6 7
4 −2 x −xi tan−1 xk −x0 zkþ1 −z0 − zk −z0
i i i
5 The problem of imaging the subsurface conductor is attributed to
k 0   2   
xk −x0 þ zk −z0 zkþ1 −z0
i i i
the optimization of the current density distribution by solving the in-
verse problem of Eq. (6). The conjugate gradient method is one of the
most effective approaches to solve such equations. This approach has

Fig. 4. Current density sections for the vertical component of the magnetic field anomalies with different β values.
86 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91

been widely used in geophysical inversions, such as magnetotelluric Algorithm.

(Mackie and Madden, 1993; Newman and Alumbaugh, 2000; Rodi
and Mackie, 2001), electromagnetic (Ellis and Oldenburg, 1994), di-
rect current resistivity (Spitzer, 1995; Zhang et al., 1995), magnetic
data inversion (Pilkington, 1997; Liu et al., 2013; Liu et al., 2015),
and gravity inversion (Liu et al., 2015). The conjugate gradient meth-
od has several advantages. First, the iteration times are finite. Theo-
retically, the algorithm can converge after a finite number of
iterations. The objective function ϕ(m) of VLF data inversion is
given by

T
min ϕðmÞ ¼ ðd−GmÞ W d ðd−GmÞ ð7Þ
m

s:t:mimin b mi b mimax for some or all i ð8Þ


1 1 1
where, W d ¼ diag ; ; :::; and σi is the standard deviation
σ 21 σ 22 σ 2M
of the ith observation (Li and Oldenburg, 1996, 1998). mmin i and mmax
i
are the lower and upper geological constraints for the ith model pa-
∂ϕðmÞ
rameter. Minimizing the objective function ϕ(m), let ¼ 0 and
∂m
then obtain

GT W d GΔm ¼ GT W d Δd ð9Þ

where, Δd is the difference between the observed and modeled ver-

tical component of the magnetic field data and Δm is the update in
the model parameters. In the present inversion problem, the condi-
tion number of the coefficient matrix, i.e., GTWdG in Eq. (9) is large
and usually exceeds 10 15 , thereby seriously affecting the conver-
gence. Pilkington (1997) improved the condition number of the ma-
trix and the convergence rate of the conjugate gradient method by
multiplying the preconditioned matrix on both sides of the matrix
equation
PGT W d GΔm ¼ PGT W d Δd ð10Þ
where, P is the preconditioned matrix given by P = (GTWdG)− 1. The
preconditioned matrix P plays a vital role in improving the quality of
subsurface imaging. We solve Eqs. (9) and (10) to obtain the current
density distributions in the subsurface. Compared with the general
form of regularization inversion of Tikhonov and Arsenin (1977)
used by Singh and Sharma (2015), the objective function in the pres-
ent approach does not include the depth-weighting and roughness
constraints in the regularization term. We add the depth-weighting
in the form of a pre-conditioner. The flows of inverse modeling for
VLF data are as follows:
To begin with, we divide the subsurface by applying an octree-mesh
discretization. Then an initial model m0 is given and the vertical compo-
nent of the magnetic field (dpre) is computed using Eqs. 4–6. Subse-
quently, we calculate the update in the model parameters (i.e., Δm)
by solving Eq. (10) using the conjugate gradient algorithm. The algo-
rithm flowchart steps of VLF imaging using the preconditioned conju-
gate gradient method are as follows:
3.1. Preconditioned matrix
In the present study, the dimensions of matrices G and GTWdG are
51 × 466 and 466 × 466, respectively. The condition number of the co-
efficient matrix GTWdG is 1019 and this makes the solution of Eq. (9) un-
stable. To overcome this problem, Pilkington (1997) inverted surface
magnetic data with β equal to 3 for the magnetic method. Liu et al.
(2013) assumed that the diagonal elements of the coefficient matrix
GTWdG will be attenuated by β power with the distance between the ob-
servation points and blocks. Thus, the diagonal elements can be written as
 
diag GT W d G ¼ ðΔLÞ−β I ð11Þ
where, ΔL is the distance between the observation points and blocks. The
preconditioned matrix P is given by
P ¼ ðΔLÞβ I ð12Þ
where, β is a constant related to the rapid attenuation of the vertical com-
ponent of the magnetic field with distance between the observation
points and the blocks. The vertical component of the magnetic field due
to the small element dxdz is expressed as (Eq. (A2))
0 jy dxdz jy dxdz  x 
∂Hz ð0; 0Þ ¼  sinθ ¼  2 ð13Þ
2πr 2π r
Eq. (13) shows that when r N 0, the vertical component of the
magnetic field decays by the 2nd power of the distances between
the field source and the observation points, respectively. Therefore
 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91 87

Table 1 4. Results and discussion

Parameters of flat Earth model and model including topography.

Model parameters Flat Earth model Model including The results are presented in three sections. In the first section, inver-
topography sion results are presented for the flat earth model. Real VLF anomaly has
Body 1 Body 2 Body 3 Body 1 Body 2 been computed for the actual resistivity model (Model 1 of Table 1)
using a finite element code (Baranwal et al., 2011) for 19.8 kHz and an
Location (m) −85 0 85 −275 225
Width (m) 20 30 20 50 50 attempt has been made to retrieve subsurface structures in terms of ap-
Depth to the top (m) 36 18 11 16 35 parent current density by inverting real VLF anomaly using the present-
Depth Extent (m) 62 36 26 58 86 ed approach. ii) In the second section, results are presented for the
Target Resistivity (Ωm) 10 50 100 50 10 topographical data and the influence of the topography on the VLF re-
Host Resistivity (Ωm) 1000 1000
sponse has been examined. Finally, in Section 3, the real VLF anomaly
from a field example has been interpreted using the developed ap-
for the present study, β ≤ 4 is the theoretical range of value (Liu et al., proach and the results are compared with the model parameters obtain-
2013). ed using the rigorous resistivity inversion of the published data of
A theoretical example of the presented inversion is shown in Fig. 4 Baranwal et al. (2011).
for different β values using the vertical component of the magnetic In the presented preconditioned conjugate gradient inversion, β
field and the preconditioned conjugate gradient method. value and the maximum number of iterations have been taken as 4
The forward response, i.e., the vertical component of the magnetic and 200, respectively. On the testing the results, minimum and
field was generated for a 60 m × 45 m blocks and depth to the top at maximum value of current density was found to be constrained
36 m using Eq. (6). Current density in the block was considered as at − 0.005 A/m2 and 1 A/m2, respectively.
0.0025 A/m2 (Fig. 4a). The inversion results demonstrate that the appro-
priate β and preconditioned matrix can improve the resolution effi- 4.1. Synthetic example — flat earth model
ciently. When β = 0, the preconditioned matrix has no role. Thus,
current density concentrates around the earth surface (Fig. 4b). As β in- Model 1 depicts three conductors at different depths in a homoge-
creases, the target gradually shift towards actual depth such that resolu- neous half-space of 1000 Ωm resistivity (Fig. 5a). Resistivity of the target
tion increases gradually and the effect by which current density 1, 2 and 3 is 10, 50 and 100 Ωm, respectively. The depth to the top of
concentrates near the surface decreases significantly. For β ≥ 3, the cur- targets 1, 2, and 3 is 36 m, 18 m and 11 m, respectively. The horizontal
rent density concentrates near the actual target and the resolution is im- distance between the targets is 85 m. Fig. 5b reveals the synthetic VLF
proved significantly (Fig. 4d). When β ≥ 4, the change in β does not data (real and imaginary anomaly) computed using the finite element
affect the inversion results (Fig. 4e). Hence, the appropriate range is approach (Baranwal et al., 2011). We inverted the data (real and imag-
3 ≤ β ≤ 4 for the present approach. inary both) shown in Fig. 5b using the resistivity inversion technique of

Fig. 5. Inversion result on the flat earth model. (a) True Resistivity model. (b) Fittings between the observed (blue dots for real anomaly and blue circles for imaginary anomaly) and 2D
joint resistivity inversion modeled VLF data (red solid line for real anomaly and red dashed line for imaginary anomaly). (c) 2D resistivity image using joint inversion of the real and
imaginary parts of the VLF data (Baranwal et al., 2011). (d) Fittings between the observed (blue dots) and the presented approach modeled data (red solid line) of real anomaly.
(e) Inverted apparent current density section using the presented approach. (For interpretation of the references to color in this figure legend, the reader is referred to the web version
of this article.)
88 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91

Baranwal et al. (2011). We take a 1000 Ωm half-space as the initial and a
priori model. The resulting model of the resistivity distribution is shown
in Fig. 5c. Fitting between the observed and modeled VLF data is also
shown in the Fig. 5b. Next, we inverted only real anomaly data using
the presented approach (pre-conditioned conjugate gradient ap-
proach). We assumed zero current density as an initial model in all
the blocks. For independent gaussian noisy data (with zero mean and
1% standard deviation), fitting between the observed and modeled
data is shown in Fig. 5d. The resulting model of the apparent current
density distribution is shown in Fig. 5e. The horizontal locations are
especially well resolved. The vertical resolution is also good for
single-frequency VLF data. The level of the apparent current density
distribution does not match the difference in resistivity of the shal-
low (100 Ωm) and the deep (10 Ωm) targets. However, the depth
resolution is enough to conclude that the shallow conductor is in-
deed shallow and the deep conductor is indeed deep. The apparent
current density distribution and the resistivity distribution depict
similar results. The subsurface image obtained by the presented ap-
proach is highly focused, as is obtained using rigorous resistivity
inversion.
Model 2 of Table 1. We inverted the data using resistivity inversion by
Baranwal et al., 2011 (Fig. 6c) and the presented approach (Fig. 6e) gen-
erated for the above model. The comparison of the results between the
observed and modeled VLF data is shown in Fig. 6b. The results obtained
by the presented approach are closer to the actual resistivity distribu-
tion. For the noise-free synthetic data, fitting between the observed
and predicted data with the root mean square 0.1 is shown in Fig. 6d.
We also interpreted the above data by considering the flat earth
model and present the result in Fig. 6f. The position of the region with
the low resistivity is seen to have shifted downward from its actual po-
sition, which is due to the coupling of the response between the conduc-
tive bodies and topographical surface. Actual positions of the conductive
bodies are indicated by the white solid line in Fig. 6f. This example clear-
ly demonstrates that the rugged observation surfaces may become sig-
nificant when the scale length of topographic variations (≈20 m) is in
the order of the magnitude of the skin depth (100–200 m). This is
why it is important to consider the topographical surfaces to obtain ac-
curate results.
4.3. Field example

4.2. Synthetic example — model including topography
The synthetic model displayed in Fig. 6a consists of two anomalous
regions having resistivities of 50 Ωm and 10 Ωm, respectively, within
a 1000 Ωm half-space with a smooth, but pronounced topography.
The 40 observation sites are located at 25 m intervals from −490 m to
485 m. A detailed description of Body 1 and Body 2 was shown in
Finally, we tested the efficiency of the presented approach on field
data. The real VLF anomaly of field data was taken from the published
data of Baranwal et al. (2011). It was collected from an area with undu-
lating rugged surfaces in Bräunsdorf, Germany with 21.8 kHz frequency.
A pyrite mineralization of hydrothermal origin was geologically
mapped. Interpretation of the above-mentioned real anomaly data
was performed using the presented approach to obtain the apparent

Fig. 6. Inversion result on the model including topography. (a) True Resistivity model. (b) Fittings between the observed (blue dots for real anomaly and blue circles for imaginary
anomaly) and 2D joint resistivity inversion modeled VLF data (red solid line for real anomaly and red dashed line for imaginary anomaly). (c) 2D resistivity image using joint inversion
of the real and imaginary parts of the VLF data (Baranwal et al., 2011). (d) Fittings between the observed (blue dots) and the presented approach modeled data (red solid line) of real
anomaly. (e) Inverted apparent current density section using the presented approach. (e) Inverted apparent current density section using the presented approach considering flat
earth model. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91 89

current density. Fig. 7a depicts the apparent current density distribution
which resembles the single conductor obtained for the model including
topography located between x = 50 m and x = 100 m which is very
well correlated with Baranwal et al. (2011) (Fig. 7c). We have also
inverted the above data for the flat earth model. It yields a dipping
type conductor located between x = 50 m and x = 100 m. The flat
earth model also shows another conductive body at x = − 50 m
which seems to be due to topography alone (Fig. 7d). The inversion re-
sult of Baranwal et al. (2011) for the flat earth model comprises an ex-
panded split conductor at a larger depth located between x = 50 m
and x = 100 m. They also interpreted another conductive body at
x = −50 m (Fig. 7f). Fitting between the field data and modeled data
for the inverted model on the rugged observation surfaces is shown in
Fig. 7b and similar fitting was obtained by considering flat earth
model (Fig. 7e).
Analysis of the results from synthetic and field data reveals that var-
ious conductors are highlighted equally well in the apparent current
density image similar to resistivity inversion. However, we observed
that the horizontal and vertical dimensions of the conductors are re-
solved quite well, and that the relative magnitudes of current densities
for different conductors are almost similar. Therefore, apparent current
densities may not be able to reveal the differences between good and
very good conductors. Moreover, in resistivity inversion, we get resis-
tivity values according to the actual resistivity of the subsurface
structures.
It is important to highlight that VLF-EM data contains both high
frequency (generally low amplitude of real and imaginary anomaly)
as well as low frequency noises (generally high amplitude of real and
imaginary anomaly). These noises must be removed using empirical
mode decomposition techniques (Jeng et al., 2007; Sungkono et al.,
2014a) from the VLF-EM data before applying any filtering/imaging.
Karous and Hjelt (1983) filtering or inversion of raw data associated
with high frequency noises will result in multiple unwanted shallow
structures. However, apparent current density image using Karous
and Hjelt (1983) filtering remains unchanged even in the presence
of low frequency noise in the data. Moreover, inversion results are
affected by low frequency noise in the data. Therefore, it is always
better to remove the high and low frequency noises from the VLF
data.
5. Computation time
The computation was performed on a simple Core-2 duo PC. The
computation time to perform resistivity inversion is about 30 min,
while for the presented approach it is only a few seconds for 466 blocks.
Computation time for resistivity inversion will increase greatly with in-
crease in the number of block whereas for the presented approaches, it
will remain insignificant.
6. Conclusions
A subsurface imaging technique was developed for the interpre-
tation of very low frequency (VLF) electromagnetic data including
topography. Initially, an analytical expression was derived to com-
pute the vertical component of the magnetic field from the given
current density distribution in the subsurface. Even though the an-
alytical expression which we derived in the present work is not
necessarily an important result, the highlight of our work is the
manner in which we developed the approach to locate the conduc-
tors. An octree mesh discretization was incorporated to improve
the efficiency of the presented approach on the topography. A 2D
preconditioned conjugate gradient inversion was used to appraise
the subsurface conductors in terms of the current density distribu-
tion for the models including topography. The study concludes
that the images in terms of apparent current density are closer to
the results of resistivity inversion. However, the presented

Fig. 7. Inversion result on the topographical field data. (a) Inverted apparent current density section using the presented approach of Bräunsdorf data for a model including topography.
(b) Fitting between observed data and modeled data obtained after preconditioned conjugate gradient inversion of Bräunsdorf data starting with zero A/m2. Blue dots and red solid line
represent observed and computed data, respectively. (c) 2D resistivity image using joint inversion of the real and imaginary parts of the VLF data (Baranwal et al., 2011) of Bräunsdorf data
for a model including topography. (d) Inverted apparent current density section using the presented approach of Bräunsdorf data considering flat Earth model. (e) Fitting between
observed data and modeled data obtained after PCG inversion of Bräunsdorf data considering flat Earth model. (f) 2D resistivity image using joint inversion of the real and imaginary
parts of the VLF data (Baranwal et al., 2011) of Bräunsdorf data considering flat Earth model. (For interpretation of the references to color in this figure legend, the reader is referred to
the web version of this article.)
90 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91
2 3
Zz2 Zz2
approach is very fast due to the analytical expression used in com- 0 jy 6  2 2
  2 2
 7
parison to resistivity imaging using numerical techniques. Synthetic ΔH z ð0; 0Þ ¼ 4 ln x2 þ z dz− ln x1 þ z dz5 ðA6Þ
4π
examples with and without topography are presented to demon- z1 z1

strate the suitability of the approach. The example of the field VLF
data inversion shows the efficacy of the presented approach in ap- Since
praising the subsurface conductor in terms of apparent current den-
Z z
sity similar to the rigorous inversions to obtained structures in    
ln a2 þ z2 dz ¼ z ln a2 þ z2 −2z þ 2a: tan−1 ðA7Þ
terms of resistivity distribution in the subsurface. The study demon- a
strates that the rugged observation surfaces may become significant
when the scale length of topographic variations is in the order of By combining (A6) and (A7), the vertical component of the magnetic
the magnitude of the skin depth. This is why it is important to con- field at the point (0,0) due to a rectangular block extending from x1 to x2
sider the topographical surfaces for accurate results. in profile direction and z1 to z2 in vertical direction can be derived as
"

z2
Acknowledgments 0 jy   z
ΔH z ð0; 0Þ ¼ z ln x22 þ z2 −2z þ 2x2 : tan−1 ðA8Þ
4π x2 z1
We thank Prof. Jianghai Xia (Editor) and two anonymous reviewers

z2 #
  z
for their comments and suggestions which helped improve the quality − z ln x21 þ z2 −2z þ 2x1 : tan−1
of the manuscript. A.S. thanks the Indian Institute of Technology x1 z1

Kharagpur and the Ministry of Human Resources and Development, " ! !

India, for financial assistance in the form of a Research Fellowship. 0 jy x2 þ z22 x22 þ z21
ΔH z ð0; 0Þ ¼ z2 ln 22 −z 1 ln ðA9Þ
This work is a part of the doctoral dissertation of A.S. 4π x1 þ z22 x21 þ z21
! !#
x2 ðz2 −z1 Þ −1 x1 ðz2 −z1 Þ
Appendix A þ 2x2 tan−1 −2x 1 tan
x22 þ z1 z2 x21 þ z1 z2
Let us assume that subsurface consist of 2D rectangular blocks
whose strike is perpendicular to the plain of Fig. 3. Due to electro- Eq. (A9) is the expression of the vertical component of the mag-
magnetic induction, induced current is setup in the conducting fea- netic field at the point (0,0) due to a rectangular block extending
tures. Due to this current there will be a distribution of current from × 1 to × 2 (x2 N x 1 ) in the x direction and z1 to z2 (z 2 N z1) in
density in the subsurface. the vertical direction.
Let us consider that there is a uniform current density (j y) in a
small rectangular block. Magnetic field on the surface due to this References
small block can be computed by double integration of the field
Arcone, S.A., 1978. Investigation of a VLF airborne resistivity survey conducted in north-
due to small element (dxdz) over length and breadth of this block. ern Maine. Geophysics 43, 1399–1417.
The conduction current (I) in the small element will be j y dxdz. Ascher, U., Haber, E., 2001. Grid refinement and scaling for distributed parameter estima-
Therefore, according to Biot-Savart's Law, tangential magnetic tion problems. Inverse Prob. 17, 571–590.
Baba, K., Chave, A.D., 2005. Correction of seafloor magnetotelluric data for topographic
field due to current in an infinitesimal small element in the subsur- effects during inversion. J. Geophys. Res. 110, B12105.
face observed on the surface (0,0) will be Baker, H.A., Myers, J.O., 1980. A topographic correction for VLF-EM profiles based on
model studies. Geoexploration 18, 135–144.
Baranwal, V.C., Sharma, S.P., 2006. Integrated geophysical studies in the east-Indian geo-
0 I 1 jy dxdz thermal province. Pure Appl. Geophys. 163, 209–227.
ΔH r ¼  ¼ ðA1Þ
2π r 2πr Baranwal, V.C., Franke, A., Borner, R.U., Spitzer, K., 2011. Unstructured grid based 2-D in-
version of VLF data for models including topography. J. Appl. Geophys. 75, 363–372.
Beamish, D., 1994. Two-dimensional regularised inversion of VLF data. J. Appl. Geophys.
The vertical component of the magnetic field due to the small ele- 32, 357–374.
ment dxdz expressed as Beamish, D., 1998. Three-dimensional modeling of VLF data. J. Appl. Geophys 39, 63–76.
Beamish, D., 2000. Quantitative 2D VLF data interpretation. J. Appl. Geophys 45, 33–47.
Benson, A.K., Payne, K.L., Stubben, M.A., 1997. Mapping groundwater contamination using
0 jy dxdz jy dxdz x jy dxdz  x  DC resistivity and VLF geophysical method — a case study. Geophysics 62, 80–86.
ΔH z ð0; 0Þ ¼  sin θ ¼  ¼  2 ðA2Þ Bernard, J., Valla, P., 1991. Groundwater exploration in fissured media with electrical and
2πr 2πr r 2π r
VLF methods. Geoexploration 27, 81–91.
Bielak, J., Ghattas, O., Kim, E., 2005. Parallel octree-based finite element method for large-


0 jy dxdz x scale earthquake ground motion simulation. Comput. Model. Eng. Sci. 10, 99–112.
ΔH z ð0; 0Þ ¼  2 ðA3Þ Biswas, A., Sharma, S.P., 2016. Integrated geophysical studies to elicit the subsurface struc-
2π x þz 2
tures associated with uranium mineralization around South Purulia shear zone, India:
a review. Ore Geol. Rev. 72, 1307–1326.
Boukerbout, H., Gibert, D., Sailhac, P., 2003. Identification of sources of potential fields
where, x and z are the coordinate of the center of the small elementdxdz.
with the continuous wavelet transform: application to VLF data. Geophys. Res. Lett.
The vertical component of the magnetic field due to one block 30 (8) article 1427.
will be Chouteau, M., Bouchard, K., 1988. Two-dimensional terrain correction magnetotelluric
surveys. Geophysics 53, 854–862.
Davis, K., Li, Y., 2013. Efficient 3D inversion of magnetic data via octree-mesh
Zx2 Zz2
 discretization, space-filling curves, and wavelets. Geophysics 78, J61–J73.
0 0 jy x
ΔH z ð0; 0Þ ¼ ∬ ΔH z ¼ dxdz ðA4Þ Drahor, M.G., 2006. Integrated geophysical studies in the upper part of Sardis archaeolog-
xz 2π x2 þ z2 ical site, Turkey. J. Appl. Geophys 59, 205–223.
x1 z1
Ellis, R.G., Oldenburg, D.W., 1994. The pole-pole 3D DC resistivity inverse problem: conju-
gate gradient approach. Geophys. J. Int. 119, 187–194. http://dx.doi.org/10.1111/j.
1365-246X.1994.tb00921.x.
Let x2 + z2 = p ⇒ dp = 2xdx and x21 + z2 =p1, x22 + z2 =p2. Fischer, G.A., 1989. A strong topographic valley effect in AMT and VLF-R measurements.
Then the Eq. (A4) can be written as Geophys. J. Int. 96, 469–475.
Franke, A., Börner, R.U., Spitzer, K., 2007. Adaptive unstructured grid finite element simu-
lation of two-dimensional magnetotelluric fields for arbitrary surface and seafloor to-
Zp2 Zz2
 Zz2 pography. Geophys. J. Int. 171, 71–86.
0 jy 1 jy p
ΔH z ð0; 0Þ ¼ dpdz ¼ ½ ln ðpÞp12 dz ðA5Þ Fraser, D.C., 1969. Contouring of VLF-EM data. Geophysics 34, 958–967.
4π p 4π Haber, E., Heldmann, S., 2007. An octree multi grid method for quasi-static Maxwell's
p1 z1 z1
equations with highly discontinuous coefficients. J. Comput. Phys. 223, 783–796.
 A. Singh, S.P. Sharma / Journal of Applied Geophysics 133 (2016) 82–91
Jeng, Y., Lin, M.J., Chen, C.S., Wang, Y.H., 2007. Noise reduction and data recovery for a
VLF-EM survey using a nonlinear decomposition method. Geophysics 72 (5),
F223–F235. http://dx.doi.org/10.1190/1.2752561.
Jiracek, G.R., Reddig, R.P., Kojima, R.K., 1989. Application of the Rayleigh-FFT technique to
magnetotelluric modeling and correction. Phys. Earth Planet. Inter. 53, 365–375.
Kaikkonen, P., Sharma, S.P., 1998. 2D nonlinear joint inversion of VLF and VLF-R data
using simulated annealing. J. Appl. Geophys 39, 155–176.
Kaikkonen, P., Sharma, S.P., Mittal, S., 2012. 3D modeling and inversion of VLF and VLFR
electromagnetic data. Geophysics 77 (4), WB219–WB231.
Kamm, J., Pedersen, L.B., 2014. Inversion of airborne tensor VLF data using integral equa-
tions. Geophys. J. Int. 198, 775–794.
Karous, M., 1979. Effects of relief in EM methods with very distant source. Geoexploration
17, 33–42.
Karous, M., Hjelt, S.E., 1983. Linear filtering of VLF dip-angle measurements. Geophys.
Prospect. 31, 782–794.
Key, K., Constable, S., Weiss, C.J., 2006. Mapping 3D salt using the 2D marine
magnetotelluric method: case study from Gemini Prospect, Gulf of Mexico. Geophys-
ics 71, B17–B27.
Li, Y., Oldenburg, D.W., 1996. 3D inversion of magnetic data. Geophysics 61, 394–408.
Li, Y., Oldenburg, D.W., 1998. 3D inversion of gravity data. Geophysics 63, 109–119.
Liu, G., Becker, A., 1992. Evaluation of terrain in AEM surveys using the effects boundary
element method. Geophysics 57, 272–278.
Liu, S., Hu, X., Liu, T., Feng, J., Gao, W., Qiu, L., 2013. Magnetization vector imaging for bore-
hole magnetic data based on magnitude magnetic anomaly. Geophysics 78,
D429–D444.
Liu, S., Hu, X., Xi, Y., Liu, T., 2015. 2D inverse modeling for potential fields on rugged ob-
servation surface using constrained Delaunay triangulation. Comput. Geosci. 76,
18–30.
Mackie, R.L., Madden, T.R., 1993. Three-dimensional magnetotelluric inversion using con-
jugate gradients. Geophys. J. Int. 115, 215–229. http://dx.doi.org/10.1111/j.1365-
246X.1993.tb05600.x.
Mohanty, W.K., Mandal, A., Sharma, S.P., Gupta, S., Misra, S., 2011. Integrated geological
and geophysical studies for delineation of chromite deposits: a case study from
Tangarparha, Orissa, India. Geophysics 76, B173–B185.
Monteiro Santos, F.A., Mateus, A., Figueiras, J., Gonçalves, M.A., 2006. Mapping groundwa-
ter contamination around a landfill facility using the VLF-EM method — a case study.
J. Appl. Geophys. 60, 115–125.
Newman, G.A., Alumbaugh, D.L., 2000. Three-dimensional magnetotelluric inversion
using non-linear conjugate gradients. Geophys. J. Int. 140, 410–424. http://dx.doi.
org/10.1046/j.1365-246x.2000.00007.x.
Paterson, N.R., Ronka, V., 1971. Five years of surveying with very low frequency – electro-
magnetics method. Geoexploration 9, 7–26.
Pedersen, L.B., Becken, M., 2005. Equivalent images derived from very low frequency
(VLF) profile data. Geophysics 70 (3), G43–G50.
Pedersen, L.B., Oskooi, B., 2004. Airborne VLF measurements and variations of ground
conductivity: a tutorial. Surv. Geophys. 25, 151–181.
Pedersen, L.B., Persson, L., Bastani, M., 2009. Airborne VLF measurement and mapping of
ground conductivity in Sweden. J. Appl. Geophys. 67, 250–258.
Philips, W.J., Richards, W.E., 1975. A study of the effectiveness of the VLF method for the
location of narrow-mineralized fault zones. Geoexploration 13, 215–226.
91
Pilkington, M., 1997. 3D magnetic imaging using conjugate gradients. Geophysics 62,
1132–1142. http://dx.doi.org/10.1190/1.1444214.
Poddar, M., Rathor, B.S., 1983. VLF survey of the weathered layer in southern India.
Geophys. Prospect. 31, 524–537.
Rodi, W., Mackie, R.L., 2001. Nonlinear conjugate gradients algorithm for 2D
magnetotelluric inversion. Geophysics 66, 174–187.
Sharma, S.P., Baranwal, V.C., 2005. Delineation of groundwater-bearing fracture zones in a
hard rock area integrating very low frequency electromagnetic and resistivity data.
J. Appl. Geophys. 57, 155–166.
Sharma, S.P., Kaikkonen, P., 1998. Two-dimensional non-linear inversion of VLF-R data
using simulated annealing. Geophys. J. Int. 133, 649–668.
Sharma, S.P., Anbarasu, K., Gupta, S., Sengupta, A., 2010. Integrated very low-frequency
EM, electrical resistivity, and geological studies on the Lanta Khola landslide, North
Sikkim, India. Landslides 7, 43–53. http://dx.doi.org/10.1007/s10346-009-0187-y.
Singh, A., Sharma, S.P., 2015. Fast imaging of subsurface conductors using very low fre-
quency electromagnetic data. Geophys. Prospect. 63 (6), 1355–1370. http://dx.doi.
org/10.1111/1365-2478.12323.
Smith, B.D., Ward, S.H., 1974. On the computation of polarization ellipse parameters. Geo-
physics 39, 867–869.
Spitzer, K., 1995. A 3D finite difference algorithm for DC resistivity modeling using conju-
gate gradient methods. Geophys. J. Int. 123, 903–914. http://dx.doi.org/10.1111/j.
1365-246X.1995.tb06897.x.
Sundararajan, N., Ramesh Babu, V., Chaturvedi, A.K., 2011. Detection of basement frac-
tures favorable to uranium mineralization from VLF-EM signals. J. Geophys. Eng 8,
330–340.
Sungkono, Bahri, A.S., Warnana, D.D., Monteiro Santos, F.A., Santosa, B.J., 2014a. Fast, si-
multaneous and robust VLF-EM data denoising and reconstruction via multivariate
empirical mode decomposition. Comput. Geosci 67, 125–137. http://dx.doi.org/10.
1016/j.cageo.2014.03.007.
Sungkono, Husein, A., Prasetyo, H., Bahri, A.S., Monteiro Santos, F.A., Santosa, B.J., 2014b.
The VLF-EM imaging of potential collapse on the LUSI embankment. J. Appl. Geophys
109, 218–232. http://dx.doi.org/10.1016/j.jappgeo.2014.08.004.
Sungkono, Bahri, A.S., Santosa, B.J., 2015. Application of multivariate EMD to improve
quality VLF-EM data: a synthetic and fields data. Appl. Mech. Mater 771, 170–173.
Tezkan, B., 1999. A review of environmental applications of quasi-stationary electromag-
netic techniques. Surv. Geophys 20, 279–308.
Tikhonov, A.N., Arsenin, V.Y., 1977. Solution of Ill-Posed Problems. Wiley, New York,
p. 258.
Wannamaker, P.E., Stodt, J.A., Rijo, L., 1986. Two-dimensional topographic responses in
magnetotellurics modeled using finite elements. Geophysics 51, 2131–2144.
Zhang, J., Mackie, R.L., Madden, T.R., 1995. 3D resistivity forward modeling and inversion
using conjugate gradients. Geophysics 60, 1313–1325. http://dx.doi.org/10.1190/1.
1443868.
Zlotnicki, J., Vargemezis, G., Mille, A., Bruere, F., Hammouya, G., 2006. State of the hydro-
thermal activity of Soufriere of Guadeloupe volcano inferred by VLF surveys. J. Appl.
Geophys. 58, 265–279.