Professional Documents
Culture Documents
Advances in Modeling and Interpretation in Near Surface Geophysics PDF
Advances in Modeling and Interpretation in Near Surface Geophysics PDF
Arkoprovo Biswas
Shashi Prakash Sharma Editors
Advances in
Modeling and
Interpretation
in Near Surface
Geophysics
Springer Geophysics
The Springer Geophysics series seeks to publish a broad portfolio of scientific
books, aiming at researchers, students, and everyone interested in geophysics. The
series includes peer-reviewed monographs, edited volumes, textbooks, and confer-
ence proceedings. It covers the entire research area including, but not limited to,
applied geophysics, computational geophysics, electrical and electromagnetic
geophysics, geodesy, geodynamics, geomagnetism, gravity, lithosphere research,
paleomagnetism, planetology, tectonophysics, thermal geophysics, and seismology.
Editors
Advances in Modeling
and Interpretation in Near
Surface Geophysics
123
Editors
Arkoprovo Biswas Shashi Prakash Sharma
Department of Geology Department of Geology and Geophysics
Banaras Hindu University Indian Institute of Technology Kharagpur
Varanasi, Uttar Pradesh, India Kharagpur, West Bengal, India
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Historically, geophysics has been used to illustrate deep exploration targets, such as
economic mineralization, groundwater resources, and oil and gas deposits, as well
as environments that are relatively free of human impact. Together, geologist,
hydrogeologist, civil engineers, archaeologists, soil scientists, and others have
applied the customary geophysical methods with long-trusted but simple interpre-
tation schemes to detect, classify, and describe buried geological or anthropogenic
targets in the shallow subsurface which is just a few meters below the Earth’s
surface and is of great importance. In recent times, there is an urgent need for
subsurface resources has increased. There is a coercing need for exploration targets
have been completed by noteworthy advances in near-surface applied geophysics
techniques and interpretation theory that have caused existing textbooks and
monographs on the subject to become outdated.
Recent advances in geophysical methods and interpretation approaches have
made the integrated geophysical exploration as an important approach for exploring
natural resources as well as for tackling different geotechnical and environment
issues. Among the different possible fields of application, most common are
hydrological and hydro-geological characterization and monitoring, mineral
exploration, archaeological surveys, locating voids, soil classification, contaminant
assessment, etc. Moreover, modeling and interpretation of such subsurface features
are very important in identifying the actual zones of interest. For this motivation,
there is coercing in the scientific community toward the modeling and integrated
interpretation of several geophysical data sets, in order to identify and distinguish
the subsurface from the analysis of different physical properties.
This edited book provides a stimulating, theoretical advancement in near-surface
geophysics and its applications in various fields of investigations. Techniques
covered include gravity, magnetics, resistivity imaging, VLF-EM, TEM, MT,
ground-penetrating radar, and more. Some chapters on the data analysis and inverse
theory are provided and chapters are amply illustrated by case studies. This is an
important edited book for advanced undergraduate and graduate students in geo-
physics, and a treasured reference for enthusiastic geophysicists, geologists,
hydrologists, archaeologists, civil and geotechnical engineers, and others who use
v
vi Preface
geophysics and its application in their professional research and teaching. The book
will also serve as a valuable reference for geoscientists, engineers, and others
engaged in academic, government, or industrial pursuits that call for a near-surface
geophysical investigation. The accessible techniques are characterized by different
penetration and resolution capabilities. Sections on Theoretical advancements,
modeling, inversion, joint inversion, and new methods are amply illustrated.
Various case studies on the application of near-surface geophysics are also
illustrated.
The present work has developed from a protracted series of lively communication
with my colleagues, seniors, and juniors both in India and abroad, especially, during
the last few years, on “Advances in Modeling and Interpretation in Near Surface
Geophysics.” The present book will showcase some advancement in near-surface
geophysical methods, mathematical modeling, and its application in real-field data.
The present work also stresses the significance of exploration, contamination, and
environmental problems. The book has a broad literature survey, and all pains have
been taken to take care of proper citation at the requisite places. We would per-
sonally like to thank them on behalf of me and the authors of other chapters,
respectively. Any inadvertent error/omission in this regard is sincerely regretted. In
addition, we also thank all the “Authors” of the respective chapters, who have
contributed to the same with their valuable time, effort and expertise in the respective
area of research/study, as provided in each chapter. Many persons have provided
warm friendship, steady guidance, firm support and valuable guidance throughout
the years; the following are those who have particularly stood out as being instru-
mental to the successful completion of this book. Apart from all the contributors who
also helped us in reviewing the chapters of the present book, we would personally
like to thank Prof. N. Sundararajan, SQU, Oman; Prof. E. L. Ekinci, BEU, Turkey;
Prof. K. Essa, CU, Turkey; Prof. N. Bhatt, MSU, India; and Dr. T. Acharya, PU,
India. Special thanks to Dr. Anand Singh, IIT Bombay, India, and Mr. Reetam
Biswas, UT, Austin, USA, for providing useful suggestions and helping me in
reviewing many chapters. We would also like to thank our Institutions/University for
supporting us for the successful development of this book.
vii
Contents
ix
x Contents
xi
xii Editors and Contributors
Contributors
Abstract In the present study, joint inversion of Direct Current (DC) resistivity
and Audio-Magnetotelluric (AMT) sounding data has been attempted to overcome
the limitation of Schlumberger resistivity sounding data in depicting the concealed
aquifer layers at large depth. It is observed that Schlumberger resistivity sounding
data is unable to reflect the presence of multiple aquifer layers located at depth. The
study reveals that 4-layer, 6-layer or 8-layer subsurface structures in the study area
yield Schlumberger sounding data that reflects only 4-layer KH-type sounding
curves. Therefore, it is not possible to predict 6- and 8-layers from the
Schlumberger sounding data measured in the area. However, unlike Schlumberger
sounding data, theoretically it has been observed that AMT data over the same
4-layer, 6-layer or 8-layers subsurface structures yields different responses.
Therefore, it is possible to resolve the suppression ambiguity arising in
Schlumberger sounding data using joint interpretation of Schlumberger sounding
and AMT sounding data. It is interesting to highlight that even AMT sounding data
also does not reflect the exact number of layers from either apparent resistivity or
phase data, but systematic inversion/joint inversion is able to resolve the above
mentioned ambiguity and delineate the presence of aquifers at depth. The Very Fast
Simulated Annealing (VFSA) global optimization approach has been used to the
study the efficacy of joint interpretation of DC resistivity and AMT sounding in
solving the practical problem in the area. The approach is general and similar
approach can be used to solve practical problems associated with other geophysical
applications such as mineral investigation.
Keywords Resistivity sounding AMT sounding Suppression Joint inversion
Groundwater
1.1 Introduction
Electrical method of geophysical prospecting is the most reliable method for the
delineation of shallow subsurface structures (Telford et al. 1990). The method has
been widely used for different applications related with mineral, groundwater, civil
engineering applications as well as archaeological investigations (Auken et al. 2006;
Goldman and Neubauer 1994; Guerin and Benderitter 1995; Loke et al. 2013). The
electrical method possesses simple mathematical development, easy and well
developed interpretation approaches, cheap instrumentation, and simple data
acquisition. With the rapid development in instrumentation and data acquisition
approaches, these days it is possible to record multidimensional measurements
efficiently (Loke et al. 2013). Apart from its simplicity, electrical method also suffers
from different ambiguity in the data interpretation like all other geophysical methods
(Sharma and Kaikkonen 1999). The Equivalence and Suppression are the two major
ambiguities (Sharma and Verma 2011) associated with the interpretation of resis-
tivity sounding data. Equivalence is a condition where either product or ratio of the
thickness & resistivity of the sandwiched layer is interpreted from the resistivity
sounding data (Bhattacharya and Patra 1968). Further, Suppression deals with a
problem where the presence of a layer is not reflected in the resistivity sounding data.
Various joint inversion schemes have been proposed in the past to resolve these
two ambiguities by using different geophysical data sets (Vozoff and Jupp 1975;
Raiche et al. 1985; Dobroca et al. 1991; Verma and Sharma 1993; Gallardo and Meju
2003; Monteiro Santos et al. 2006). However, most of the studies deal with the
resolution of equivalence problem only. Suppression problem is trickier than
equivalence and it has remained unsolved. Further, there are different types of
suppression problems in the field measurement. In a continuously increasing sub-
surface resistivity distribution, it is not possible to identify the number of layers from
apparent resistivity data. For example, a continuously increasing apparent resistivity
field data could be fitted equally well with the model data from a two-layer or
multiple layers. Further, when a thin conducting layer is located at depth in a rather
resistive subsurface, then the presence of conducting layer is also not reflected in the
Schlumberger sounding data or in multidimensional resistivity imaging data. The
later scenario is more practical and it is related with groundwater and mineral
investigations. Sharma and Verma (2011) have presented an extensive study to
resolve both equivalence and suppression. However, their study concluded that only
equivalence could be solved but solution of suppression is ambiguous. Only apriori
information from other geophysical measurements or geological information could
help in reducing the suppression ambiguity to some extent.
In the present study, an attempt is made to solve suppression type ambiguity that
was observed in Schlumberger sounding data measured in a lateritic terrain of West
1 Resolving Suppression Ambiguity in Schlumberger … 3
Bengal, India. Schlumberger soundings carried out in the area exhibits only a
4-layer KH type sounding curve. A typical data measured using Schlumberger array
in the area is presented in Fig. 1.1. According to the interpretation of sounding data
shown in Fig. 1.1 there is no conducting layer below 30 m depth in the area as the
resistivity sounding curve shows continuously increasing apparent resistivity trend.
Information available from a few deep boreholes in the area reveals that there could
be two thin aquifer layers at depth and they are located around 100 and 200 m
depth ranges. However, these two layers may or may not be present at any par-
ticular location in the area. This uncertainty in the area results in failure of a large
number of deep tube-wells. Since, Schlumberger soundings performed in the area
are unable to reveal any information about these aquifers at depth, alternate geo-
physical approach is required that could solve this problem. This problem was
theoretically examined by Sharma et al. (2018) by employing different electrical
and EM methods. They pointed out that AMT method could be the best approach
that yields significantly different responses in the presence and absence of deeper
aquifers. Therefore, in the present study, a subsurface model is proposed from the
available borehole information in the area. Forward modeling is performed to
compute DC resistivity as well as AMT sounding data for the proposed models.
Since neither DC resistivity data nor AMT data reveals the exact number of layers,
systematic inversion/joint inversion will be carried out in this study which could
resolve the suppression problem and delineate deeper aquifers in the area.
Z1
qa ðsÞ ¼ s 2
TðkÞJ1 ðksÞkdk: ð1:1Þ
0
Ti þ 1 þ qi tanhðkhi Þ
Ti ¼ Ti þ 1 tanhðkhi Þ
ð1:2Þ
1þ qi
In Eq. (1.2), the resistivity and thickness of the ith layer are qi and hi, respec-
tively, TN = qN, and i = N − 1, …1. Ghosh (1971a) used the following change of
variables, x = ln(s) and y = ln(1/k), to transform Eq. (1.1) from Henkel integral to
convolution integral.
Z1
qa ðxÞ ¼ TðyÞfe2ðxyÞ J1 ðexy Þgdy ð1:3Þ
1
The convolution integral shown in Eq. (1.3) can be written in finite terms of
multiplication and summation and used to calculate theoretical Schlumberger
resistivity sounding data. In Eq. (1.3) the first term (resistivity transform) is the
input function and the second term is the filter function. Digital linear filtering
technique proposed by Ghosh (1971a, b) has made computation of a forward
response very simple. Ghosh (1971b) and Koefoed (1979) have developed the sets
of coefficients for this filter function to compute the Schlumberger apparent resis-
tivity. An optimized short length 19 point filter (Guptasarma 1982) is used to
compute the resistivity sounding data in the present study.
(b) AMT sounding
The apparent resistivity (qa) and phase (u) over a multi-layered model is obtained
from the surface impedance Z1 given by Vozoff (1991):
1
qa ðxÞ ¼ jZ j2 ; ð1:4Þ
xl 1
Zi þ 1 þ Ti
Zi ðxÞ ¼
1 þ Si Zi þ 1
pffiffiffiffiffi
Zn ¼ k qn ;
1 Resolving Suppression Ambiguity in Schlumberger … 5
Table 1.1 Resistivity model in the area depicting a 4-, 6- and 8-layer subsurface structures
Layer 4-Layer 6-layer 8-layer
number Resistivity Thickness Resistivity Thickness Resistivity Thickness
(Ωm) (m) (Ωm) (m) (Ωm) (m)
1 5.0 0.5 5.0 0.5 5.0 0.5
2 25 3.5 25 3.5 25 3.5
3 10 25 10 25 10 25
4 7500 7500 70 7500 70
5 10 10 10 10
6 7500 7500 70
7 10 10
8 7500
Apparent resistivity and phase are computed over a large frequency range to
depict the variation in responses from different models shown in Table 1.1.
A large number of global optimization methods are being used these days for
minimization of objective function (different norms between the observed and
model data). Simulated Annealing (SA), Genetic algorithm (GA), Particle Swarm
Optimization (PSO), Artificial Neural Network (ANN), Artificial Bee Colony
(ABC) optimization etc. are in used for multi-parametric optimization (Goldberg
and Deb 1991; Sen and Stoffa 1995; Storn and Price 1997; Sharma and Kaikkonen
1999; Singh et al. 2005; Juan et al. 2010). It has been observed that a variant of SA
which very fast simulated annealing (VFSA) is the most efficient for interpreting
resistivity sounding data (Sharma and Verma 2011). For complete description of the
VFSA method readers are referred to Sharma and Verma (2011) and Sharma and
6 S. P. Sharma et al.
ð1:6Þ
In above equation qO i;1 and qi;1 are the ith observed and computed (model)
C
apparent resistivity for Schlumberger array, qO i;2 and qi;2 are the ith observed and
C
computed (model) apparent resistivity for AMT data, u0i and uci are the ith observed
and computed (model) phases AMT data, respectively. In Eq. (1.6) NS is the
number of observation points (current electrode spacings) for Schlumberger
sounding data and NF are the numbers of frequencies for AMT sounding mea-
surements. All the responses (observed and computed) are transformed in loga-
rithmic domain and normalized with respect to the observed response in order to
avoid bias in the objective function towards a particular data set. The model
parameters (resistivities and thicknesses) and associated misfit error of the above
randomly selected model are kept in memory. Subsequently, each model parameter
is modified (updated) with respect to its previous value according to the Cauchy
probability distribution given in Eq. (1.7). The factor yi used to update ith parameter
is given by the expression:
1 Resolving Suppression Ambiguity in Schlumberger … 7
In Eq. 1.7, ui is a random variable drawn between 0 and 1 such that yi varies
between −1 and +1. The variable Ti denotes the temperature which could be the
same or different for various model parameters. Ti may be kept different for each
model parameters for faster convergence of the solution, however, in the present
study we kept is same for all model parameters as the computation time is very
small. A new model Pm+1 i is obtained with respect to its previous value Pm
i using
Eq. (1.8).
þ1
max
Pm
i ¼ Pm
i þ yi P i Pmin
i ð1:8Þ
Next, the misfit error is calculated for the new model obtained using Eq. (1.8)
and it is compared with the previous model. Now suppose the misfit error of the
new models is e2 and the previous model is e1 and De = e2 − e1, then there could be
two possible situations for the selection and rejection of the new model. (i) When
the misfit error of the new model (e2) is smaller than previous model (e1), the new
model is accepted with probability exp(−De/T), (ii) When the misfit error of the new
model (e2) is larger than that of the previous model (e1), a random number is drawn
between 0 and 1 and compared it with the estimated probability (exp(−De/T)). If the
probability is greater than the random number drawn, then also the new model is
accepted with the same probability, otherwise new model is rejected keeping the
previous model and its misfit error in the memory. Further, at a particular tem-
perature level, a predefined number of moves are made by accepting and rejecting
the models according to the selection criterion mentioned above. This makes an
iteration of the process at a particular temperature level (say, kth). Next, the tem-
perature level is lowered according to the following cooling schedule:
Ti ðk Þ ¼ T0i exp ci k1=M : ð1:9Þ
T0i is the initial temperature, ci is a constant and its value is used as unity in the
present study, k is the iteration number varying from 1 to 500, 1/M is known as
cooling schedule where M is number of parameters. In the present study T0i is used
as 0.01 and 1/M as 0.4 to reduce the computation time.
After reducing the temperature at subsequent lower levels and selecting better
models by performing required number of moves at each level, we finally reach to
the lowest temperature level and get the best fitted model with least misfit error.
This yields one solution. The same procedure is repeated several times by starting
from different randomly selected initial model and reaching to the final solution. In
the present study we obtain 10 different best fitted models and statistical mean
model and associated uncertainties are estimated. If NR is the number of runs and ek
8 S. P. Sharma et al.
i is
is the misfit error obtained at the kth run then the statistical mean model P
computed from Eq. (1.10) given below (Tarantola 1987).
PNR
i ¼ k¼1 Pi;k expðek Þ
P P NR ð1:10Þ
k¼1 expðek Þ
Since square root of the diagonal element of the covariance matrix is used to
estimate the uncertainty in the final solution, the covariance matrix is computed
from
PNR
i ÞðPj;k P
ðPi;k P j Þ expðek Þ
CovPði; jÞ ¼ k¼1
PNR ð1:11Þ
k¼1 expðek Þ
In above equation i and j vary from 1 to M. The uncertainties (dPi ) in the mean
model is given as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dPi ¼ CovPði; iÞ: ð1:12Þ
1.4 Results
that it is not possible to assess the number of layers more than 4 from the measured
data.
Sharma et al. (2018) had proposed that AMT sounding would be the best
approach to solve such suppression problems in DC resistivity measurement. Next,
theoretical AMT soundings data has been computed for the models shown in
Table 1.1 and shown in Fig. 1.3. AMT soundings clearly reveal different responses
for 4-, 6-, and 8-layer models. However, the dilemma is that one cannot assess the
number of layers from AMT soundings as well. Therefore, joint interpretation of
DC resistivity and AMT soundings is proposed for the solution of this problem.
Since, individual inversions cannot solve this problem, only joint inversion results
of DC resistivity and AMT data have been presented systematically assuming
various situations arising in the field.
Fig. 1.3 Apparent resistivity and phase data for 4-, 6- and 8-layer models (Table 1.1) depicting
different AMT sounding curves (unlike similar DC resistivity sounding curves)
10 S. P. Sharma et al.
Fig. 1.4 Fitting between the observed and model data for DC resistivity and AMT data over
4-layer structure
Table 1.2 Model parameter after joint inversion of DC resistivity and AMT data over 4-layer
structure
Model parameters Actual value Search range Interpreted models
q1 (Ωm) 5.0 1–10 5.29 ± 1.25
q2 (Ωm) 25 5–50 26.29 ± 4.02
q3 (Ωm) 10 1–20 10.15 ± 1.14
q4 (Ωm) 7500 350–10,000 7496 ± 25
h1 (m) 0.5 0.1–2 0.55 ± 0.15
h2 (m) 3.5 1–10 3.17 ± 1.75
h3 (m) 25 5–50 25.53 ± 3.64
Misfit error – – 2.00 10−5
1 Resolving Suppression Ambiguity in Schlumberger … 11
Table 1.3 Model parameters after joint inversion of DC resistivity and AMT data over 6-layer
structure using 4-layer and 6-layer models
Model parameters Search range Interpreted models using
4-layer model 6-layer model
q1 (Ωm) 5.0 1–10 3.28 ± 1.07 4.56 ± 1.90
q2 (Ωm) 25 5–50 37.34 ± 5.67 24.51 ± 7.70
q3 (Ωm) 10 1–20 13.23 ± 0.24 8.53 ± 1.92
q4 (Ωm) 7500 350–10,000 6911 ± 89 6550 ± 1635
q5 (Ωm) 10 1–20 – 11.32 – 4.20
q6 (Ωm) 7500 350–10,000 – 7481.24 – 22
h1 (m) 0.5 0.1–2 0.35 ± 0.12 0.45 ± 0.23
h2 (m) 3.5 1–10 1.21 ± 0.34 5.36 ± 2.44
h3 (m) 25 5–50 45.41 ± 0.98 20.42 ± 5.96
h4 (m) 70 10–100 – 70.32 – 3.22
h5 (m) 10 5–50 – 11.69 – 4.15
Misfit error 1.22 10−2 5.98410−4
Schlumberger sounding data nor AMT data give any indication that observed data
corresponds to more than 4 layers. Therefore, initially joint inversion was carried
out assuming a 4-layer model. We can see that optimization of a 4-layer model
gives erroneous model parameters (actual model is not within the estimated
uncertainties) as well as it shows poor fitting between the observed and model data
(Table 1.3; Fig. 1.5). Figure 1.5 reveals that AMT phase data shows the worst
fitting. This is because phase in EM methods is more sensitive than apparent
resistivity. Any deviation in the subsurface model parameters with respect to their
actual value will cause more deviation in phase response than apparent resistivity
response. The above discrepancy gives an indication that observed data does not
correspond to a 4-layer structure and 6-layer or 8-layer model optimization is
required.
Fig. 1.5 Fitting between the observed and model data for DC resistivity and AMT data when
4-layer model is optimized using data recorded over 6-layer structure
12 S. P. Sharma et al.
Fig. 1.6 Fitting between the observed and model data for DC resistivity and AMT data when
6-layer model is optimized using data recorded over 6-layer structure
Table 1.4 Model parameters after joint inversion of DC resistivity and AMT data over 8-layer
structure using 4-layer, 6-layer and 8-layer models
Fig. 1.7 Fitting between the observed and model data for DC resistivity and AMT data when
4-layer model is optimized using data recorded over 8-layer structure
is important to highlight that we have kept search ranges same for various models
for comparison.
After getting an unsatisfactory fitting between the observed and model data
while optimizing a 4-layer model to fit observed data over 8-layer structure, next we
moved to optimize this data using a 6-layer model. Surprisingly, the theoretical data
14 S. P. Sharma et al.
Fig. 1.8 Fitting between the observed and model data for DC resistivity and AMT data when
6-layer model is optimized using data recorded over 8-layer structure
generated for an 8-layer subsurface structure fits very well with the model data
obtained after optimizing a 6-layer structure only. The corresponding results have
been presented in Fig. 1.8 and Table 1.4. If we compare the interpreted model
parameters obtained after optimizing 6-layer structure shown in Table 1.4 with the
actual model parameters, then we can see that the interpreted thickness h4 and h5 are
more than their actual values while resistivity of these layers is resolved very well.
Since data observed over an 8-layer model fits well with model data optimized
using a 6-layer model only, therefore, how to identify the 3rd aquifer layer (7th
layer in Table 1.1) present at a larger depth? This is another challenging situation.
The solution of this problem requires additional a priori information in the study
area. If we know by some apriori information that third aquifer is present in the
subsurface or the interpreted thickness of the 2nd aquifer using 6-layer model is
more than the expected thickness of aquifer in the area, then we can always interpret
this data to optimize an 8-layer structure. This could be the possible solution to
identify and locate the third aquifer layer in the area.
While interpreting observed data over 8-layer structure using 6 layer model we
get a higher estimate of h5 which is incorrect according to aprori information from
the area. It is important to highlight that we restricted the higher bound of the search
range for thickness as 15 m and lower bound of its resistivity at 8 Ωm. If we keep
the restriction of the resistivity and thickness of the target layer then the optimized
model parameter reveals that resistivity lies at the lower boundary and thickness at
the higher boundary. This means optimization technique is trying to find the
equivalent physical model in order to fit the response. But this physical model does
not obey the geological information of the area. Thus restricting the search range
according to the local geology we can still find alternate way to detect the presence
of third aquifer hidden at depth in the resistivity and AMT sounding data.
Therefore, it can be concluded that we may get good fitting model in the area using
6-layer optimization, however, it is important to analyze whether the obtained
model is geologically reliable. If not, we can increase the number of layers to get
geologically meaningful results.
1 Resolving Suppression Ambiguity in Schlumberger … 15
Fig. 1.9 Fitting between the observed and model data for DC resistivity and AMT data when
8-layer model is optimized using data recorded over 8-layer structure
1.5 Discussions
Now, we know very well that recorded DC resistivity sounding data in the area will
show only a 4-layer structure whether the subsurface is 4-layer, 6-layer or 8-layer
structure. Therefore, first we should interpret only Schlumberger sounding data
using a 4-layer model. Next, we should vary the search ranges for q1, q2, q3, h1, h2,
h3 within the estimated uncertainty obtained after the individual inversion of DC
resistivity data and proceed systematically to perform joint inversion of DC resis-
tivity and AMT data mentioned in the results section above to get the best fitted and
geologically correct model. When all the observed data fits very well with 4-layer
model then the subsurface structure is 4-layer only. If not, we will perform 6-layer
inversion. Now responses will definitely fit. When the observed responses are fitted
well with the model data and also estimated model is geologically correct then the
subsurface structure is 6-layer. However, when the observed data and model data
are fitted well but interpreted models are not geologically correct, then we should
optimized 8-layer model. Table 1.5 presents the results obtained after such kind of
inversion and these are the most correct and reliable results.
16 S. P. Sharma et al.
Table 1.5 Model parameters after joint inversion of DC resistivity and AMT data over 4-layer,
6-layer and 8-layer models after restricting the search range for model parameters for top 3-layers
1.6 Conclusions
References
Auken E, Louise P, Christensen NB, Sorensen K (2006) A survey of current trends in near-surface
electrical and electromagnetic methods. Geophysics 71:249–260
Bhattacharya PK, Patra HP (1968) Direct current geoelectric sounding. Elsevier, Amsterdam
Dobroka M, Gyulai A, Ormos T, Csokas J, Dresen L (1991) Joint inversion of seismic and
geoelectric data recorded in an underground coal mine. Geophys Prospect 39:643–665
Gallardo LA, Meju MA (2003) Characterization of heterogeneous near-surface materials by joint
2D inversion of DC resistivity and seismic data. Geophys Res Lett 30:1658
Ghosh DP (1971a) The application of linear filter theory to the direct interpretation of geoelectrical
resistivity sounding measurements. Geophys Prospect 19:192–217
Ghosh DP (1971b) Inverse filter coefficient for the computation of apparent resistivity sounding
curves for a horizontally stratified Earth. Geophys Prospect 19:769–775
Goldberg, DE, Deb K (1991) A comparison of selection schemes used in genetic algorithms. In:
Rawlins GJE (ed) Foundations of genetic algorithms, pp 69–93
Goldman M, Neubauer FM (1994) Groundwater exploration using integrated geophysical
techniques. Surv Geophys 15:331–361
Guerin R, Benderitter Y (1995) Shallow karst exploration using MT-VLF and DC resistivity
methods. Geophys Prospect 43:635–654
Guptasarma D (1982) Optimization of short digital linear filters for increased accuracy. Geophys
Prospect 30:501–514
Juan LFM, Esperanza GG, José PFÀ, Heidi AK César OMP (2010) PSO: a powerful algorithm to
solve geophysical inverse problems: application to a 1D-DC resistivity case. J Appl Geophys
71:13–25
Koefoed O (1979) Geosounding principles 1: resistivity sounding measurements. Elsevier,
Amsterdam
Loke MH, Chambers JE, Rucker DF, Kuras O, Wilkinson PB (2013) Recent developments in the
direct-current geoelectrical imaging method. J Appl Geophys 95:135–156
Monteiro Santos FA, Sultan SA, Represas P, Sorady A L El (2006) Joint inversion of gravity and
geoelectrical data for groundwater and structural investigation: application to the northwestern
part of Sinai. Egypt Geophys J Int 165:705–718
Raiche AP, Jupp DLB, Rutter H, Vozoff K (1985) The joint use of coincident loop transient
electromagnetic and Schlumberger sounding to resolve layered structures. Geophysics
50:1618–1627
Sen M, Stoffa PL (1995) Global Optimization Methods in Geophysical Inversion. Elsevier
Publishing Company, 281
Sharma SP, Biswas A (2013) A practical solution in delineating thin conducting structure and
suppression problem in direct current resistivity sounding. J Earth Syst Sci 122:1065–1080
Sharma SP, Kaikkonen P (1999) Appraisal of equivalence and suppression problems in1-D EM
and DC measurements using global optimization and joint inversion. Geophys Prospect
47:219–249
Sharma SP, Verma SK (2011) Solutions of the inherent problem of the equivalence indirect current
resistivity and electromagnetic methods through global optimization and joint inversion by
successive refinement of model space. Geophys Prospect 59:760–776
Sharma SP, Panda KP, Jha MK, Mogi T (2018) Challenges for electrical and electromagnetic
measurements for depicting groundwater aquifers at greater depths in lateritic terrains of West
Bengal, India (submitted to Journal of Applied Geophysics)
Singh UK, Tiwari RK, Singh SB (2005) One-dimensional inversion of geo-electrical resistivity
sounding data using artificial neural networks—a case study. Comput Geosci 31:99–108
Storn R, Price K (1997) Differential evolution—a simple and efficient heuristic for global
optimization over continuous spaces. J Global Optim 11:341–359
Tarantola A (1987) Inverse problem theory. Elsevier Publishing Company, Methods of data Fitting
and Model Parameter Estimation
18 S. P. Sharma et al.
Telford WM, Geldart LP, Sheriff RE (1990) Applied geophysics, 2nd edn. Cambridge University
Press
Verma SK, Sharma SP (1993) Resolution of thin layers using joint inversion of electromagnetic
and direct current resistivity sounding data. J EM Waves Appl 7:443–479
Vozoff K (1991) The magnetotelluric method. In: Nabighian M (ed) Electromagnetic methods in
applied geophysics II. Society of Exploration Geophysicists, Tulsa Oklahoma
Vozoff K, Jupp DLB (1975) Joint inversion of geophysical data. Geophys J Roy Astron Soc
42:977–991
Chapter 2
GPR and ERT Investigations of Karst
Structures at the Buhui-Cuptoare
Cave System, Anina Karst Region
(Banat Mountains, Romania)
Abstract Romanian karst is representative for the temperate zone by its landforms,
by its diversity and also by the amplitude of the exo—and endokarst features. The
connection between surface and underground karst features is the main difficulty in
protecting karst environment. Due to this, geomorphological studies developed on
karst require increased attention on both exokarstic and endokarstic features. Buhui
Cave represents the biggest cave in Anina karst region and the 2nd important cave
in Banat Mountains. Through Buhui Cave flows the homonymous stream, which is
among the longest underground streams in Romania. Using GPR and ERT as
geophysical methods in studying two known caves, Cuptoare Cave and Buhui
Cave, we were able to identify rock stratification and orientation, sections of these
caves’ galleries and the high level of karstification in this karst massif. Our
approach is the first one of this kind in the analysis of karst terrains in Romania.
Due to this, we have to observe the pros of our study, but also to try to pay attention
to those possible minuses resulted from the GPR signal in clay layers or the
response of the ERT method in karstic regions. Because this approach was a
successful one in the study of the Buhui-Cuptoare Cave System, we intend to
develop other researches in the Anina karst region in the future in order to discover
unknown caves or unknown underground galleries. This region has a big potential
in new cave discoveries and having the functionalities provided by the GPR and
ERT methods, we have to capitalize these functionalities. To conclude, we would
like to highlight that this geophysical approach is a premier in the Romanian
geomorphology in karst terrains and karst cavities investigation. Similar approaches
may be used in caves protection, karst management plans and strategic plans
developed in karst areas. Besides, such an approach may be very useful in karst
groundwater studies.
2.1 Introduction
Karst topography represents a different terrain resulted from the dissolution of the
carbonate rock massifs. The results from this dissolution process are represented by
an efficient underground drainage (Waltham et al. 2005). Ford and Williams (2011)
state that karst terrains may be better understood only if the factors affecting dis-
solution processes and the drainage in karst terrains are deeply analysed.
In Romania limestone covers more than 2% of the country, representing about
4500 km2 (Orghidan 1972). The most extensive and massive area of calcareous
rocks is located in the southwestern part of Romania, more exactly in Banat
Mountains and is represented by the Reşiţa—Moldova Nouă Zone (Orăşeanu and
Iurkiewicz 2010).
Romanian karst is representative for the temperate zone with its landforms, by its
diversity and also by the amplitude of the exo—and endokarst features (Cocean
2001). The Mesozoic limestones occupy the largest surfaces covered by carbonate
rocks and this type of limestone presents the highest level of karstification processes
due to the large thickness of Mesozoic deposits (Trufaș and Sencu 1967).
Karst topography is a particular geological environment, both at surface and into
the underground. Surface and underground karst environments represent a con-
tinuous habitat. Karst environment still remains mostly unknown and many other
studies can be developed in order to discover more about karst terrains (De Waele
et al. 2009).
The rapid connection of surface with the underground karst features represents
the main difficulty in protecting karst environment. Any risk of contamination,
which occurs at the surface is subsequently transferred, with all the potential
dangers, into the underground. As a result, the consequences of degradation and
pollution of karst are very serious (Parise and Pascali 2003). Due to this, geo-
morphological studies developed on karst require increased attention to both surface
and underground karst features.
2 GPR and ERT Investigations of Karst Structures … 21
Anina karst region is part of the Anina Mountains, a subunit of the large geo-
graphical unit of Banat Mountains, Romania. The location is shown yellow shaded
in Fig. 2.1.
The study area has Jurassian relief and it is a faulted region, being defined by
numerous synclines and anticlines disposed parallel on a north-northeast—
south-southeast direction (Mateescu 1961). The main geomorphological charac-
teristic is represented by elongated ridges, disjointed by large karst plateaus and
very deep valleys (Bucur 1997). Onac (2000) includes this karst area into the
suspended karst plateaus category, with a very high level of karstification. The main
orientation of the faults was the element that imposed the main flow pathway and
these tectonic features serve as the limit of the major karstic structures in this area
(Iurkiewicz et al. 1996).
Anina karst region includes many karstic landforms, both at surface and into the
underground: karrens, karren fields, uvalas, sinkholes, dry valleys, blind valleys,
sinkhole valleys, gorges sectors, and more than 150 vertical shafts and caves, caves
which are included in 6 karstic basins (Goran 1982).
Buhui Cave represents the biggest cave in Anina karst region and the 2nd
important cave in Banat Mountains. Through Buhui Cave flows the homonymous
stream, which is among the longest underground streams in Romania.
Cuptoare Cave has a length of 135 m and in the past it was a fossil gallery of the
Buhui Cave. The connection between these two caves is represented by a pit which
nowadays is blocked entirely with large blocks of rock and it has not penetrable
ways for humans into Buhui Cave (Sencu 1963).
Buhui and Cuptoare represent a system of caves that have evolved on three
karstification levels: active, subfossil and fossil. We consider that there are many
voids or galleries that still unknown and undiscovered. Geophysics investigations
are able to provide a comprehensive picture for the underground and the cave
system.
Buhui-Cuptoare Cave System is located at Cuptoare Saddle, at about 1 km
distance from Anina town and we performed our geophysical investigations above
the known overlapped galleries of those two caves (Fig. 2.2).
2.3 Methodology
During the last years, GPR explorations have been utilized broadly to detect
underground voids and to portrait the underground structure of karst terrains. Sites
for GPR investigations have been chosen on those sections where caves may be
further investigated by direct inspection (Cardarelli et al. 2010).
2 GPR and ERT Investigations of Karst Structures … 23
Fig. 2.2 Location of the GPR and ERT measurements site above Buhui-Cuptoare Cave System
2 GPR and ERT Investigations of Karst Structures … 25
underground structure and finally reflected back to the surface. GPR device works
with high-frequency electromagnetic waves, ranging from 10 up to 1000 MHz.
The clay content of soils may hinder the penetration profoundness of the Ground
Penetrating Radar device (Anchuela et al. 2008). Consequently, the properties of
soil such as soil’s saturation, mineralogy and solidness, the frequency of antennas
utilized (Anchuela et al. 2009) and the dielectric properties of soil (Alsharahi et al.
2016) will determine the results gained using GPR method. Previous researches to
identify karst voids utilizing the GPR approach have been developed by
Chamberlain et al. (2000), Kadioglu and Ulugergerli (2012), Łyskowski et al.
(2014).
Our GPR explorations were performed utilizing a RAMAC GPR device.
This GPR device from MalåGeoSciences has been supplied with rough terrain
antenna (RTA) having the central frequency of 25 and 50 MHz (Fig. 2.3). These
geophysical studies are the first ones developed in the karst terrain of Banat
Mountains. Due to this, we choose to use radar’s settings from previous studies
developed in limestone areas, as for example Chamberlain et al. (2000), Kadioglu
and Ulugergerli (2012), Łyskowski et al. (2014). In our investigations, an aggregate
wave velocity of 0.12 m/ns for the depth transformation of the radar signal was
utilized, since RTA antennas are compact type, no Wide Angle Reflection and
Refraction (WARR) or Common Midpoint (CMP) (Apel and Dezelic 2005;
Kadioglu and Ulugergerli 2012).
Fig. 2.3 GPR field measurements using RAMAC GPR system from MalåGeoSciences. Source
Andras Nemedi (2015)
26 L. Artugyan et al.
Fig. 2.4 Location of BCCS-1 and BCCS-2: GPR and ERT profiles above the Buhui-Cuptoare
Cave System
2 GPR and ERT Investigations of Karst Structures … 27
Fig. 2.5 Location of BC profile: GPR profile above a section of Buhui Cave
The obtained GPR profiles were processed using Reflex2D Software (Sandmeier
Scientific Software) and we performed processes like filtering, noise removal and
smoothing, in order to make the interpretation easier.
28 L. Artugyan et al.
Fig. 2.6 GeoTom MK8E1000 system used for ERT measurements and study area images
(1-GeoTom MK8E1000; 2-forest road above cave system; 3-Cuptoare Cave entrance; 4-site of
ERT measurements). Source Laurenţiu Artugyan, Andras Nemedi (2015)
2 GPR and ERT Investigations of Karst Structures … 29
GPR measurements in karst terrains have some limitations due to numerous factors,
such as trees, bushes or even contact with bedrock being possible obstacles in
obtaining excellent radargrams. Another obstacle in obtaining very good results is
given by the ratio between depth range and resolution. For a higher depth range,
GPR measurements will provide lower resolution, while for higher resolution, the
depth range will be lower (Łyskowski et al. 2014).
Because the aim of this research was to observe the karst massif at higher depths
as possible, we choose to reach the highest depth that we were able by using GPR.
Due to this, we had to assume that the resolution obtained is lower.
In the interpretation of our GPR results, we compared our radargrams with
results from other studies that have used GPR in cave detection and carbonate
environment, such as El-Qady et al. (2005), Gosar (2012), Łyskowski et al. (2014),
Zhao et al. (2015) and Zieliński et al. (2016). Of course, our interpretation is based
on our familiarity with the geological and morphological characteristics of the study
area.
30 L. Artugyan et al.
Fig. 2.7 GPR profiles for BCCS-1 site using 25 MHz (a) and 50 MHz (b) frequency antennas
The first radargrams for BCCS-1 (Fig. 2.7) have been done using 25 and
50 MHz frequency antennas above the known gallery of Cuptoare Cave the transect
having N-S orientation.
Based on these radargrams, we are able to observe a lot of fissures in the karst
massif, but also we identified positive anomalies at depths of about 6–8 m. We
suggest that on both radargrams those anomalies indicate the air-filled cavity of
Cuptoare Cave. At 90 m from north of BBCS-1 site, we have identified also a
positive anomaly that we have attributed to an unknown clay-filled karstic void.
Besides those positive anomalies associated with cavities, on the same profile,
we identified also elements of stratification on these radargrams. Elements of
stratification were identified into the karst massif at depths of 20 m. These strati-
fication features have a decisive role in the cave formation and water discharge in
this karst massif.
Both radargrams reflect the same results, the difference being given by the depth
reached using the 25 MHz frequency antenna, which allows us to reach a depth
close to 40 m.
The second transect above Cuptoare Cave, BCCS-2, was also conducted using
25 and 50 MHz frequency antennas, having a NE-SW orientation. This profile was
done as a transversal section above the biggest hall in Cuptoare Cave, in order to
have less rock between GPR antenna and the air filling this hall.
BCCS-2 profile presents results showing the high complexity of Buhui-Cuptoare
Cave System. As it can be seen in Fig. 2.8, the GPR radargrams reflect stratification
features at depths between 20 and 40 m, compact rock of the karst terrain in the NE
at depths between 10 and 30 m, but also we could identify the bedrock at 40 m
depth for the 25 MHz transect and at 22–25 m depth on the transect obtained using
50 MHz antenna.
Again the 25 MHz frequency antenna gives a deeper penetration, getting close
to 45 m, while the 50 MHz antenna reaches only 30 m depth. Karst massif strat-
ification has the same orientation as the orientation obtained in the profile BCCS-1.
2 GPR and ERT Investigations of Karst Structures … 31
Fig. 2.8 GPR profiles for BCCS-2 site using 25 MHz (a) and 50 MHz (b) frequency antennas
Fig. 2.9 GPR profile from BC site using the 25 MHz frequency antenna
Gosar (2012), Łyskowski et al. (2014) or Zhao et al. (2015). Despite the fact that
reflectors are very similar, the positive hyperbolas of the reflectance are associated
with possible cavities in the shallow subsurface.
Previous studies such as El-Qady et al. (2005), Rădulescu et al. (2007), Sabo
(2008), McGraw (2010), Martínez-Pagán et al. (2013) and Nouioua et al. (2013)
have been used to interpret the transects obtained after ERT measurements.
The ERT results indicate several features that are important to mention. On the
profile BCCS-1, we could identify small resistivity values that we associated with
the clay layer, marked on the ERT transects at 3. At about 28 m from the starting
point, the clay layer was found between 2 and 4 m, and 64 m from the starting
point we could identify a bigger clay deposit between 0 and 6 m deep. On the
Dipole-dipole profile number 1 represents an air-filled cavity, due to the high
resistivity value (Fig. 2.10a). With the Schlumberger array we observed resistivity
values that we associate with water-filled cavity (number 2 on ERT transect), most
probably the creek that flows through Buhui Cave (Fig. 2.10b). The alternation of
high values with low values of the resistivity is showing that the karst massif is
highly fragmented, the limestone being fissured and some of the fissured zones are
filled with clay. On Wenner array, we identified only bedrock and clay, more
compact than on Schlumberger and Dipole-Dipole arrays (Fig. 2.10c).
For the second profile of ERT measurements, namely BCCS-2, we observe very
high resistivity values for the surfaces marked with 1 on all those three ERT arrays.
These high values we consider to be the last chamber of Cuptoare Cave, a very
large hall with the ceiling very close to surface. The high resistivity values are
attributed to air-filled cavity (Fig. 2.11a–c). Dipole-Dipole array returns a depth of
12 m only, but a better lateral resolution and this method reveals the massiveness of
the bedrock. In the NE, a very high resistivity is associated with an air-filled cavity
(Fig. 2.11a). With the Schlumberger array, the results reveal again the massiveness
and compactness of the bedrock below 8 m deep, but this array indicates also
2 GPR and ERT Investigations of Karst Structures … 33
Fig. 2.10 ERT profiles using dipole-dipole array (a) Schlumberger array (b) and the Wenner
array (c) for the BCCS-1 site. Legend 1—air-filled cavity; 2—water-filled cavity; 3—clay deposits;
4—bedrock
Fig. 2.11 ERT profiles using dipole-dipole array (a), Schlumberger array (b) and the Wenner
array (c) for the BCCS-2 site. Legend 1—air-filled/collapse cavity; 3—clay layers; 4—bedrock;
5—air-filled cavity/bedrock
34 L. Artugyan et al.
higher resistivity values between 30 and 40 m distance from the NE starting point
(Fig. 2.11b).
Because Wenner array reveals large surfaces having high values of resistivity
and based on precedent studies (Sabo 2008; McGraw 2010), we have attributed
those surfaces to bedrock. Even so, there is the possibility to have some unknown
air-filled cavities in those high values of resistivity. We could also remark the clay
layer in the first 2 m of this profile (marked with number 3) explained by the low
resistivity values (Fig. 2.11c).
2.4.3 Discussions
Consequently, this method may be used with success in identifying unknown caves
in this region and also in other karst regions in Romania.
On the other hand, the depth on the ERT transects is lower than the depth
obtained using GPR method. Due to this, beyond 20 m the ERT method is not
available to validate the results obtained using GPR method.
In addition, using the ERT method we found that the clay layer is thin in the
Colonovăţ Plateau, especially above Cuptoare and Buhui caves and the results
obtained using GPR provide a good quality. Due to its thin depth above these two
caves, the clay layer is not a factor that disturbs or obstructs the GPR signal.
Correlating both GPR and ERT methods, it can be observed that there are
matching aspects of both methods, but there are also some outputs those are dif-
ferent. These outputs those remained unconfirmed need more attention and further
research, maybe using other geophysical methods such as seismic refraction.
This approach is the first one of this kind in the analysis of karst terrains in
Romania. Due to this, the pros of this geophysical approach should be observed, but
also we must pay attention to those possible minuses resulted from the GPR signal
in clay layers or the response of the ERT method in karstic regions.
2.5 Conclusions
Geophysical methods are very useful in studies that are focused on the location and
geometry of karstic caves (Martinez-Moreno et al. 2013) and due to this our
approach is very valuable for the carbonate areas in Romania, being the first
geophysical study involving ERT and GPR methods.
Using the GPR method, we have succeeded in identifying structural elements
and a description of the underground to depths of approx. 30–40 m, and even the
identification of some unknown galleries of the Buhui-Cuptoare Cave System.
The results obtained on the basis of the Electric Resistivity Tomography method
showed that a number of possible karstic voids in the Cuptoare Cave area also
appear on the resistivity profiles, thus confirming the radargrams obtained with the
GPR method.
Because this approach was a successful one in the study of the Buhui-Cuptoare
Cave System, we intend to develop other researches in the Anina karst region in the
near future in order to discover unknown caves. This region has a big potential in
new cave discoveries and having the functionalities provided by the GPR and ERT
methods, we have to capitalize these functionalities.
To conclude, we would like to highlight that this geophysical approach, com-
bining GPR and ERT, is a premier in the Romanian geomorphology in karst terrains
and karst cavities investigation. Similar approaches may be used in caves protec-
tion, in karst management plans and strategic plans developed in karst areas.
Besides, such an approach may be very useful in karst groundwater studies.
36 L. Artugyan et al.
Acknowledgements Sincere thanks to Andras Nemedi and Alexandru Hegyi who helped us in
data field acquisition applications. Department of Geography from West University of Timisoara
has ensured the logistic support for the geophysical equipment and we are thankful for this
support.
References
Aitken JA, Stewart RR (2005) Signal processing enhancements of GPR data in a carbonate
environment. CREWES Res Rep 14
Alsharahi G, Driouach A, Faize A (2016) Performance of GPR influenced by electrical
conductivity and dielectric constant. Procedia Technol 22:570–575
Apel DB, Dezelic V (2005) Evaluation of high frequency ground penetrating radar (GPR) in
mapping strata of dolomite and limestone rocks for ripping technique. Int J Surf Min Reclam
Environ 19(4):260–275. https://doi.org/10.1080/13895260500275418
Anchuela ÓP, Casas-Sainz AM, Soriano MA, Pocoví-Juan A (2008) Mapping subsurface karst
features with GPR: results and limitations. Environ Geol 58(2):391–399. https://doi.org/10.
1007/s00254-008-1603-7
Anchuela ÓP, Soriano APJMA, Casas-Sainz AM (2009) Characterization of karst hazards from
the perspective of the doline triangle using GPR—examples from Central Ebro Basin (Spain).
Eng Geol 108(3–4):225–236. https://doi.org/10.1016/j.enggeo.2009.06.022
Bucur II (1997) Formaţiunile mezozoice din zona Reşiţa-Moldova Nouă (Munţii Aninei şi estul
Munţilor Locvei). Ed. Presa Universitară Clujeană, Cluj-Napoca 214p
Cardarelli E, Cercato M, Cerreto A, Di Filippo G (2010) Electrical resistivity and seismic
refraction tomography to detect buried cavities. Geophys Prospect 58:685–695
Chalikakis K, Plagnes V, Guerin R, Valois R, Bosch FP (2011) Contribution of geophysical
methods to karst-system exploration: an overview. Hydrogeol J 19:1169–1180. https://doi.org/
10.1007/s10040-011-0746-x
Chamberlain AT, Sellers W, Proctor C, Coard R (2000) Cave detection in limestone using ground
penetrating radar. J Archaeol Sci 27:957–964. https://doi.org/10.1006/jasc.1999.0525
Church M (2010) The trajectory of geomorphology. Prog Phys Geogr 34(3):265–286
Cocean P (2001) Environment threats in Romanian karst, Department of Geography, University of
Cluj. In: 13th International congress of speleology, 4th Speleological Congress of Latin
América and Caribbean, 26th Brazilian Congress of Speleology, pp 613–617
Coskun N (2012) The effectiveness of electrical resistivity imaging in sinkhole investigations. Int J
Phys Sci 7(15):2398–2405
De Waele J, Plan L, Audra P (2009) Recent developments in surface and subsurface karst
geomorphology: An introduction. Geomorphol 106:1–8
De Waele J, Gutiérrez F, Parise M, Plan L (2011) Geomorphology and natural hazards in karst
areas: a review. Geomorphol 134:1–8
El-Qady G, Hafez M, Abdalla MA, Ushijima K (2005) Imaging subsurface cavities using
geoelectric tomography and ground-penetrating radar. J Cave Karst Stud 67(3):174–181
Ercoli M, Pauselli C, Forte E, Di Matteo L, Mazzocca M, Frigeri A, Federico C (2012) A
multidisciplinary geological and geophysical approach to define structural and hydrogeological
implications of the Molinaccio spring (Spello, Italy). J Appl Geophys 77:72–82. https://doi.org/
10.1016/j.jappgeo.2011.11.011
Ford D, Williams P (2011) Geomorphology underground: The study of karst and karst processes.
In: Gregory KJ, Goudie AS (eds) The SAGE Handbook of Geomorphology, SAGE
Publications Ltd., 648p
Gambetta M, Armadillo E, Carmisciano C, Stefanelli P, Cocchi L, Tontini FC (2011) Determining
geophysical properties of a nearsurface cave through integrated microgravity vertical gradient
2 GPR and ERT Investigations of Karst Structures … 37
and electrical resistivity tomography measurements. J Cave Karst Stud 73(1):11–15. https://
doi.org/10.4311/jcks2009ex0091
Gibson PJ, Lyle P, George DM (2004) Application of resistivity and magnetometry geophysical
techniques for near-surface investigations in karstic terranes in Ireland. J Cave Karst Stud 66
(2):35–38
Gomez-Ortiz D, Martín-Crespo T (2012) Assessing the risk of subsidence of a sinkhole collapse
using ground penetrating radar and electrical resistivity tomography. Eng Geol 149–150:1–12
Goran C (1982) Catalogul sistematic al peşterilor din România. Speol., Fed. Rom.
Turism-Alpinism, Com. Centr. Speol. Sport., Bucureşti, Inst, p 496
Gosar A (2012) Analysis of the capabilities of low frequency ground penetrating radar for cavities
detection in rough terrain conditions: the case of Divača Cave. Slovenia Acta Carsologica 41
(1):77–88
Hamdan H, Economou N, Kritikakis G, Andronikidis N, Manoutsoglou E, Vafidis A, Pangratis P,
Apostolidou G (2012) 2D and 3D imaging of the metamorphic carbonates at Omalos plateau/
polje, Crete, Greece by employing independent and joint inversion on resistivity and seismic
data. Int J Speleol 41(2):199–209
Iurkiewicz A, Dragomir G, Rotaru A, Bădescu B (1996) Karst systems in Banat Mountains
(Reșița-Nera Zone). Theor Appl Karstol 9(1996):121–140
Kadioglu S, Ulugergerli EU (2012) Imaging karstic cavities in transparent 3D volume of the GPR
data set in Akkopru dam, Mugla, Turkey. Nondestruct Test Eval 27(3):263–271
LaBrecque DJ, Sharpe R, Wood T, Heath G (2004) Small-scale electrical resistivity tomography of
wet fractured rocks. Ground Water 42(1):111–118
Łyskowski M, Mazurek E, Zietek J (2014) Ground penerating radar investigation of limestone
karst at the Odstrzelona Cave in Kowala, Świetokrzyskie Mountains, Poland. J Cave Karst
Stud 76(3):184–190. https://doi.org/10.4311/2014EX0001
Mafteiu M (1991) Contributions to the investigation of the karst of Padiș area (Bihor Mountains,
Romania) by means of resistivity measurements. Theor Appl Karstology 4:65–76
Martinez A, Byrnes AP (2001). Modeling dielectric‐constant values of geologic materials: an aid
to ground—penetrating radar data collection and interpretation, Curr Res Earth Sci, Bull 247,
part 1:1–16
Martinez-López J, Rey J, Duenas J, Hidalgo C, Benavente J (2013) Electrical tomography applied
to the detection of subsurface cavities. J Cave Karst Stud 75(1):28–37. https://doi.org/10.4311/
2011ES0242
Martinez-Moreno FJ, Pedrera A, Ruano P, Galindo-Zaldivar J, Martos-Rosillo S,
Gonzalez-Castillo L, Sanchez-Ubeda JP, Marin-Lechado C (2013) Combined microgravity,
electrical resistivity tomography and induced polarization to detect deeply buried caves
Algaidilla Cave (Southern Spain). Eng Geol 162:67–78
Martínez-Pagán P, Gómez-Ortiz D, Martín-Crespo T, Manteca JI, Rosique M (2013) The electrical
resistivity tomography method in the detection of shallow mining cavities. A case study on the
Victoria Cave, Cartagena (SE Spain). Eng Geol 156:1–10
Mateescu F (1961) Influențe structurale în relieful Munților Carașului, Probleme de geografie.
Institutul de Geologie și Geografie, București 8:205–219
McGraw TJ (2010) Assessment of ground penetrating radar and comparison with resistivity for
detecting subsurface cavities within karst topography in North-Central Ohio. (Thesis) College
of bowling green, State University of Ohio
Mihevc A, Stepišnik U (2012) Electrical resistivity imaging of cave Divaska jama, Slovenia.
J Cave Karst Stud 74(3):235–242. https://doi.org/10.4311/2010ES0138R1
Mitrofan H, Povară I, Mafteiu M (2008) Geoelectrical investigations by means of resistivity
methods in karst areas in Romania. Environ Geol 55:405–413
Moldoveanu M, Stewart RR, Aitken JA (2002) Shallow imaging using ground-penetrating radar
(GPR) data in a carbonate environment: Belize, Central America, CREWES Res Rep, vol 14
Monteiro Santos FA, Andrade Afonso AR (2005) Detection and 2D modelling of cavities using
pole–dipole array. Environ Geol 48:108–116
38 L. Artugyan et al.
Nouioua I, Rouabhia AEK, Fehdi CH, Boukelloul ML, Gadri L, Chabou D, Mouici R (2013) The
application of GPR and electrical resistivity tomography as useful tools in detection of
sinkholes in the Cheria Basin (northeast of Algeria). Environ Earth Sci 68:1661–1672. https://
doi.org/10.1007/s12665-012-1859-9
Onac B (2000) Geologia regiunilor carstice (Geology of karst terrains). Universitatea
“Babes-Bolyai” Cluj-Napoca, Institutul de Speologie “Emil Racoviță” Cluj-Napoca 399p
Orășeanu I, Iurkiewicz A (2010) Karst hydrogeology of Romania. Editura Federația Română de
Speologie, Oradea, p 444
Orghidan T (1972) The fiftieth anniversary of the first speleological institute of the world. AAPG
Bull/Int J Speleol 4(1):1–7
Trufaș V, Sencu V (1967) Tipuri litologice de carst in Romania. Analele Universității București,
seria Ştiințele Naturii, Geologie-Geografie XVI(1):115–121
Parise M, Pascali V (2003) Surface and subsurface environmental degradation in the karst of
Apulia (southern Italy). Environ Geol 44:247–256
Putiška R, Kušnirák D, Dostál I, Lačný A, Mojzeš A, Hók J, Pašteka R, Krajňák M, Bošanský M
(2014) Integrated geophysical and geological investigations of karst structures in Komberek,
Slovakia. J Cave Karst Stud 76(3):155–163. https://doi.org/10.4311/2013ES0112
Rădulescu V, Rădulescu F, Diacopolos C, Popescu M (2007) Geoelectrical study for delineating
underground cavities in Karst Areas. GEO-ECO-MARINA. Coastal zone processes and
management. Environ Legis 13:89–95
Reynolds JM (1997) An introduction to applied and environmental geophysics, 1st edn, Wiley,
806pp
Sabo SH (2008) Evaluation of capacitively-coupled electrical resistivity for locating solution
cavities overlain by clay-rich soils. MS Thesis, Bowling Green State University, 43pp
Saribudak M (2015) The Million Dollar Question: Which Geophysical Methods Locate Caves
Best Over the Edwards Aquifer? A Potpourri of Case Studies from San Antonio and Austin,
Texax, USA. NCKRI SYMPOSIUM 5, Proceedings of the 14th multidisciplinary conference
on sinkholes and the engineering and environmental impacts of karst, pp 355–364
Sencu V (1963) Cercetǎri asupra carstului din jurul localităţii Anina (Banat)—Peşterile din
bazinele pâraielor Anina şi Buhui, Probl Geogr 10:156–177
Stepišnik U (2008) The application of electrical resistivity imaging in collapse doline floors:
Divača karst, Slovenia, Studia Geomorphologica Carpatho-Balcanica, Vol. XLII, PL ISSN
0081-6434:41–51
Stepišnik U, Mihevc A (2008) Investigation of structure of various surface Karst formations in
limestone and dolomite bedrock with application of the electrical resistivity imaging. Acta
Carsologica 37(1):133–140
Waltham T, Bell F, Culshaw M (2005) Sinkholes and subsidence, Springer, 416p
Youssef AM, El-Kaliouby HM, Zabramawi YA (2012) Integration of remote sensing and electrical
resistivity methods in sinkhole investigation in Saudi Arabia. J Appl Geophys 87:28–39
Zieliński A, Łyskowski M, Mazurkiewicz E (2016) Ground penetrating radar investigation of
limestone karst objects in the botanical garden in Kielce. Geol Geophys Environ 42(1):31–38
Zhao C, Li S, Wu J, Xu Z, Zhang Y, Zhou L (2015) Case study of ground penetrating radar for
geological prediction in Qiyueshan Tunnel. In: 2nd International conference on green materials
and environmental engineering (GMEE 2015), Atlantis Press, pp 100–103
Zhou W, Beck BF, Adams AL (2002) Effective electrode array in mapping karst hazards in
electrical resistivity tomography. Environ Geol 42:922–928
Zhu J, Currens JC, Dinger JD (2011) Challenges of using electrical resistivity method to locate
karst conduits—a field case in the Inner bluegrass region, kentucky. J Appl Geophys 75:523–
530. https://doi.org/10.1016/j.jappgeo 2011.08.009
Chapter 3
Integrated Geoelectrical
and Hydrochemical Investigation
of Shallow Aquifers in Konkan Coastal
Area, Maharashtra, India: Advanced
Artificial Neural Networks Based
Simulation Approach
S. Maiti (&)
Department of Applied Geophysics, IIT (ISM), Dhanbad, India
e-mail: saumen@iitism.ac.in
G. Gupta
Indian Institute of Geomagnetism, Mumbai, India
Keywords Geo-electrical Geophysical inversion Bayesian neural networks
Hybrid Monte Carlo Hydrogeology Konkan
3.1 Introduction
Groundwater is the chief source of water being used to cater the daily needs of
human being living in a semi-arid and hard rock region of India (Verma et al.
1980). Therefore, exploration of groundwater lenses and their appropriate charac-
terization in semi-arid hard rock setting is very important. Moreover, the coastal
regions of India have become vulnerable in the face of over-exploitation and
sea-level rise, causing mixing of fresh water and saline water (Todd 1959; Asfahani
2007; Song et al. 2007; CGWB 2009). While the sources of water supply are
limited, the fulfilment of water demand in semi-arid hard rock regions is associated
with significant uncertainty. This uncertainty is characterized by the scarcity of rain
fall and availability of limited aquifer. Groundwater usually occurs in the fractured
basement in hard rock setting (Verma et al. 1980). Fractured basement is defined by
a layer that is sandwiched between the overburden and the bedrock, where a sig-
nificant amount of water are expected to occur (Batte et al. 2008). Therefore,
delineation of geologically weak zones/fractured basement and characterization of
hydro-geological models may be of a great societal implication.
The present study area is located in Sagareshwar-Redi-Malewad region,
Sindhudurg district, Maharashtra where groundwater problem is typically associ-
ated with hard rock setting, limited water retention capacity of sub-surface geo-
logical medium and saline water intrusion (CGWB 2009). There is a lack of sand
and gravel aquifer (CGWB 2009) and thus it may be difficult to construct high
yielding wells. It may however, be noted that the delineation of fractured basement,
may be useful to develop some prospects of constructing high yielding bore wells in
the study area. Figure 3.1a depicts broad hydro-geological cross-section which has
been constructed based on borehole litho-logical and geo-electrical information.
Figure 3.1b shows that in the overburden, the aquifer is dominated by soil/fractured
laterite. The middle layer is chiefly dominated by clay/clayey sand. According to
the borehole litho-logical information provided by Groundwater Survey and
Development Agency (GSDA), Sindhudurg, Maharashtra, the basement consists of
fractured granulites/Phyllite. The fractured basement is exposed to the surface in
many places in the survey area (Deshpande 1998).
Recently electrical resistivity modeling supplemented with hydro-geochemical
analysis has become indispensable for exploration of groundwater and water quality
studies (Zohdy 1989; Asfahani 2007; Song et al. 2007; Chandra et al. 2008; CGWB
2009; Maiti et al. 2011; Mondal et al. 2013; Gupta et al. 2014; Kumar et al. 2014;
Machiwal and Jha 2015; Das et al. 2017; Maiti et al. 2017). This is because the
underlying approach is fast and sensitive to conductivity of rock/fluid. Additionally,
the method gives continuous coverage of apparent resistivity distribution pattern of
sub-surface earth (Telford 1990). However, modeling and interpretation of true
3 Integrated Geoelectrical and Hydrochemical Investigation … 41
Fig. 3.1 a General cross-sectional hydro-geological map of the study area, b representative
hydro-geological section along an arbitrary line
especially Bayesian neural networks (BNN) with multiple algorithms have been
used to solve geo-electrical inverse inverse problems (Lampinen and Vehtari 2001;
Maiti et al. 2011, 2012, 2013a, b). The Bayesian neural network modeling has been
shown to approximate any continuous function with arbitrary precession (Lampinen
and Vehtari 2001; Karmakar et al. 2018). Particularly, the HMC-BNN has been
used successfully in VES data inversion and the robustness of methods is examined
when the data is assorted with different levels of correlated noise (Maiti et al. 2011).
However, none of the prior research explores the robustness of HMC-BNN for VES
data inversion in noisy (particularly chaotic noise) environment. The presence of
chaotic noise could result in erratic variation of measured VES data. Here it is
examined the HMC-BNN to invert new set of VES data acquired over coastal
Maharashtra by optimizing optimal hyper parameters to deal with erratic variation
in VES data component and then interpret hydro-geological models combining
litho-logy and hydro-chemical information collected from the study area. The
present technique is different from previously employed technique of HMC-BNN in
the sense that it focuses to obtain a set of optimal hyper parameters which can able
to recover the to recover layered parameters from noise (chaotic) intervened VES
data. Consequently, based on the literature reviewed, it is found that a joint tech-
nique for interpretation of hydro-geological models combining VES, lithology, and
hydro-geochemical information was not examined before in the southern part of
Konkan coast, India.
The main aims of the present study are, therefore to: (i) examine the sensitivity
of hyper-parameter of HMC-BNN approach for modeling/ simulation of signal
corrupted with chaotic noise. (ii) compare the HMC-BNN-based results with the
inversion results by singular value decomposition (SVD) for estimating earth layer
parameters when VES data is corrupted with different level of chaotic noise
(iii) interpret hydro-geological model constructed from a new VES data with
lithology and hydro-chemical information and (iv) establish empirical relation
between the hydro-geochemical parameters with earth resistivity in the current
study.
Fig. 3.2 a Geological map of the study area, b location map of the Schlumberger resistivity
sounding, well for hydro-chemical data analysis and well from where lithological information is
obtained from Groundwater Survey and Development Agency (GSDA), Sindhudurg, Maharashtra
Z1
qa ðsÞ ¼ s 2
TðwÞJ1 ðwsÞwdw ð3:1Þ
0
Ti þ 1 ðwÞ þ qi tanhðwhi Þ
Ti ðwÞ ¼ ; i ¼ n 1; . . .1 ð3:2Þ
½1 þ Ti þ 1 ðwÞ tanhðwhi Þ=qi Þ
where, n represents layer number, qi and hi denote true resistivity and true thickness
of the ith layer of earth model respectively.
The inverse modeling technique uses iterative scheme to minimize the objective
function set between the field data and the data computed by the forward modeling
program. Because the geo-electrical inverse problem is inherently non-linear, the
inversion solution is usually associated with non-uniqueness and instability. It may
be treated by introducing a damping factor in the governing equations (Lines and
Treitel 1984; Roy 1999). The widely used damped least-squares solution gives rise
to (Roy 1999)
the sources of instability. It has been popularly used for solving a small-scale, rank
deficient geophysical inverse problem (Inman 1975; Menke 1984; Maiti et al.
2012).
3.3.3 HMC-BNN
1XN
ES ¼ fxk ok ðdk ; wk Þg2 ð3:4Þ
2 k
PðsjwÞPðwÞ
PðwjsÞ ¼ ð3:6Þ
PðsÞ
framework, it is practical to use the Monte Carlo based sampling scheme to gather
the samples directly from the target distribution. The maximization of likelihood
PðsjwÞ results in the maximization of posterior PðwjsÞ and thus it is not required to
estimate scaling parameter explicitly to find the posterior distribution in eq. 3.6.
The output can be given in integral form using the conditional probability (Bishop
1995),
Z
Pðxjd; sÞ ¼ Pðxjd; wÞPðwjsÞdw ð3:7Þ
Using the Markov Chain Monte Carlo (MCMC) method, one can approximate
eq. (3.7) as
1X N
Pðxjd; sÞ ¼ Pðxjd; wm Þ ð3:8Þ
N n¼1
where fwm g denotes a weight vector which contains MCMC samples collected the
target probability PðwjsÞ and N is the total number of points where sampling is
done. The HMC algorithm (Duane et al. 1987) is a sampling based inversion
scheme where each trajectory of energy equation defined by position and
momentum of particle was solved by using Hamiltonian differential equations with
leapfrog scheme. The algorithm samples the model space using HMC and gathers a
number of equally probable models through simulation process and provides a
mean solution along with hundreds of equally probable solutions. Thus in addition
to a mean solution, it allows estimating uncertainty from probability of the posterior
pdf of the network output. The uncertainty analysis is very essential in order to have
confidence in interpretation of VES data inversion. Complete depiction of the HMC
method may be available elsewhere (Bishop 1995; Tiwari and Maiti 2011; Maiti
et al. 2011).
Assuming that the real earth data is tamed by some complex erratic/non-Gaussian/
chaotic noise, it generated VES data according to Ghosh (1971) of a typical
(i) 3-layer (ii) 4-layer and (iii) 5-layer earth model and mixed with different level of
chaotic noise. In order to create the complex chaotic time/space series, it
48 S. Maiti and G. Gupta
Fig. 3.3 a The graph shows the simulation of a synthetic underlying function using 50 numbers
of data points in the presence of chaotic noise via Hybrid Monte Carlo (HMC) method when the
data are sampled at point 0.25. The chaotic noise level 10%, b examples when no regularization is
used and the chaotic noise level 10%, c examples when the data are sampled at point 0.25 and 0.75
and the chaotic noise level 30%
essentially followed the chaotic model invented by May (1976) where the constant
may take any value in the range 2.0–4.0 Then the algorithm is tested to estimate the
layer parameters from synthetic VES data mixed with chaotic (erratic) noise. The
noise level in the data was increased from 5 to 50%. In total 1500 representative
3 Integrated Geoelectrical and Hydrochemical Investigation … 49
input (i.e. apparent resistivity)/output (i.e. true resistivity and thickness) training
pairs were considered with all geologically plausible models (via permutation and
combination) for VES data inversion. Utmost care is taken while preparing the
training data sets for HMC-BNN-based inversion. In this paper, pseudo 2D resis-
tivity section is constructed from collated HMC-BNN derived 1D VES inversion
results. A triangle based cubic interpolation was used. Grid data was fitted a surface
to non-uniformly distributed data and allowed interpolation at desired locations to
produce a smooth surface. The underlying Matlab-based approach used here is
based on Delaunay triangulation of the data. It may also be noted that a key part of
the cubic interpolation is defined over which the underlying functions are to be
valid (Menke and Menke 2012).
The true resistivity cross-sections at different depths are constructed based on the
inverted result (Fig. 3.5). The cross-section suggests that low resistivity zone
continues up to a depth of 10 m from the ground in the western and SW part of the
50 S. Maiti and G. Gupta
Table 3.1 Estimation of model parameter from synthetic data set corrupted with 30% chaotic
noise over 3-layer earth model
Model True Estimated value Estimated value Average uncertainty
parameter value SVD-based conventional HMC-BNN HMC-BNN method
method method
q1 400 Xm 395.0 Xm 401.8 Xm ±5.0 Xm
q2 40 Xm 38 Xm 39.5 Xm ±3.0 Xm
q3 100 Xm 102 Xm 99.5 Xm ±4.0 Xm
h1 30 m 28.50 m 29.51 m ±2 m
h2 10 m 12.0 m 11.04 m ±1.0 m
Fig. 3.4 The comparison of Hybrid Monte Carlo (HMC)-based Bayesian neural network
(BNN) inversion results with singular value decomposition (SVD)-based conventional resistivity
inversion results for sounding and interpreted lithology. a apparent resistivity versus AB/2 for
sounding s6, b inversion results for sounding s6 and c interpreted lithology for sounding s6.
d apparent resistivity versus AB/2 for sounding s8, e inversion results for sounding s8 and f
interpreted lithology for sounding s8. The HMC-BNN results lies within the standard deviation
(STD) limit. ‘Obs’ denotes observed, and ‘Cal’ denotes the calculated resistivity values, ‘SVD-Inv’
represents SVD based conventional resistivity inversion results, and ‘HMC-BNN’ represents Hybrid
Monte Carlo-based Bayesian Neural Network results
3 Integrated Geoelectrical and Hydrochemical Investigation … 51
Fig. 3.5 a Geology map of the study area. True resistivity surface map at different depth b surface
map at depth = 1 m c surface map at depth = 10 m d surface map at depth = 50 m
current survey area. The trend of low resistivity is further continued up to the coast
of Arabian Sea. This could be interpreted as continuation of sea water intrusion
from western to north western side starting from Sagareshwar to Redi, in coastal
Maharashtra. The NE and SE parts of survey depicts the occurrence of high
resistive features that is interpreted by the presence of lateritic rock and/or meta-
morphic hard rock continuing up to 10 m depth from ground (Fig. 3.5). At the
depth level of 50 m, relatively low resistivity zones are dominated at southern and
SE part which could be due to saltwater intrusions at deeper. The major lineaments
are identified on true resistivity section marked by sharp changes of electrical
properties (Fig. 3.5). The intersection points among the lineaments could be pos-
sible locale of groundwater recharge in the study area.
Figure 3.6a depicts the apparent resistivity distribution against electrode spacing
(AB/2). A low anomalous resistivity zone is observed below the station s25 at
Sagareshwar (Fig. 3.6a). From pseudo 2D resistivity section, one can see the
conductive feature below station s25 which continues up to 12 m depth from
ground surface (Fig. 3.6b). This conductive feature arises due to saltwater
52 S. Maiti and G. Gupta
Fig. 3.6 Resistivity structure of Hybrid Monte Carlo (HMC)-based Bayesian neural network
(BNN) approach (a, b) for profile AA′ and (c, d) for profile BB′ and below (e, f) their interpreted
2-D geological cross-section
intrusions and observed along the profile AA′. The borehole lithology which
constrained the local geology is represented by well number 1 (w1). The saturated
weathered rock is expected to continue along Sagareshwar-Tulas-Hodawade up to
the depths of 5–10 m, which might act as a conduit (Fig. 3.6b). Hydro-geochemical
data analysis of Sagareshwar village also confirms the saline water intrusion with
high TDS (total dissolved solids) (Table 3.2). The anomalous low resistive zone
below Shiroda indicates that there is an extensive intrusion of saline water
(Fig. 3.6c–d), which is also corroborated by the hydro-geochemical data with high
TDS (Table 3.2). Thus in this detecting station, resistivity of the rock is mainly
influenced by the TDS of fluid saturating the rock (Table 3.2). The borehole
lithology is supplemented by well number 2 (w2). Figures 3.7a-b indicate the
presence of a conductive zone below Redi at depth range 40–50 m. These con-
ductive zones are due to saline water intrusion. The weathered rock saturated with
water which might serve as a conduit is found at depth range of 5–7 m along DD′
profile at Malewad-Matond (Fig. 3.7d). Borehole lithology at Matond is plotted by
well number 3 (w3). The 2-D geological cross-sections have been constructed for
the profile AA′, BB′, CC′ and DD′ (Figs. 3.6 and 3.7) where it is suggested to
indicate the presence of soil/fractured laterite, and clay/clayey sand (resistivity
range: 0.001–100 Xm), fractured granulites (resistivity range: 100–1000 Xm) and
granulites/phyllite (resistivity range: 1000–10,000 Xm) to match with resistivity
pattern over the transect.
3 Integrated Geoelectrical and Hydrochemical Investigation … 53
Sagareshwar 6.8 690 442 64.1 26.7 22.7 0.39 0.8 146.4 174.2 21.1
Mochemar 5.9 330 212 96.1 7.29 31.6 4.6 0.2 122 218 14.2
Redi 6.9 415 266 46.2 28.5 37.4 2.3 0.7 83.2 154.3 38.5
Shiroda 7.3 1240 792 104.2 34 136.8 0.4 2.3 219.7 160.9 255.2
Aronda 7.0 511 301 80.4 27.3 62.2 5.2 ND 165 163.2 46.3
Ajgao 6.9 311 200 37.2 16.4 28.3 2.8 0.1 69.3 131 19.4
Matond 7.2 200 130 116.2 38.8 14.9 1.5 10.1 195.3 293.4 20.1
pH in log10H+, EC electrical conductivity in lS/cm, TDS total dissolved solids in mg/l, all detected ions
are in mg/l. ND denotes not detected
Fig. 3.7 Resistivity structure of Hybrid Monte Carlo (HMC)-based Bayesian neural network
(BNN) approach (a, b) for profile CC′ and (c, d) for profile DD′ and below (e, f) their interpreted
2-D geological cross-section
The statistical relation between the resistivity of the earth ðqe Þ and the TDS is
established in the study area by the curvilinear regression technique (Fig. 3.8a). The
error function is defined in the framework of L2 norm. The curvilinear regression
analysis between the earth resistivity with the TDS gives the statistical relation
between qe and the TDS as qe ¼ 0:0002 TDS2 0:395 TDS þ 144:34. The corre-
lation coefficient (r) given by Pearson and coefficient of determination (COD) ðr 2 Þ
54 S. Maiti and G. Gupta
are estimated as 0:84 and 0:71 respectively. The earth resistivity is found to be
inversely related to the TDS. It is therefore interpreted this inverse relation to
indicate the resistivity of the rock/soil is controlled notably by the TDS (e.g. Na, K,
Cl, Mg etc).
The statistical relation between the water resistivity ðqw Þ and the TDS is derived by
the curvilinear regression technique (Fig. 3.8b). The error function is used in the
framework of L2 norm. The curvilinear regression analysis between the water
resistivity with the TDS demonstrate the statistical relation between qw and the TDS
as qw ¼ 0:0001 TDS2 0:214 TDS þ 70:255. The r and r 2 are estimated as 0:83
and 0:63 respectively. The water resistivity is found to be inversely related to the
3 Integrated Geoelectrical and Hydrochemical Investigation … 55
TDS. The decrease in resistivity of rock sample is controlled by the porous space
saturated with the TDS because the fluid resistivity is controlled by the TDS.
Likewise the statistical relation between the earth resistivity ðqe Þ and the water
resistivity ðqw Þ is derived by the curvilinear regression technique (Fig. 3.8c). The
curvilinear regression analysis between the earth resistivity with the water resis-
tivity describes the statistical relation between qe and qw as
qe ¼ :0488 q2w þ 5:071 qw 35:72. The r and r 2 are estimated as 0:80 and 0:65
respectively.
It may be noted that the earth resistivity value was obtained by the inversion of
27-sounding stations spread over the survey area. Consequently, the empirical
relationship is derived using the VES inverted results and the hydro-geochemical
information obtained from chemical test of water samples collected from the nearby
well of the VES location. The empirical relationship may be useful to know the
TDS value where a well for water sample collection is unavailable. In that sense,
the empirical relationship yields an additional scope to interpret the results. The
comparison of HMC-BNN inversion results with available litho-logical information
exhibits more or less matching pattern. However, the number of layer from VES
inversion may not exactly match with the numbers interpreted from litholog. This is
due to due to the possible great differences in resistivity value between the bedrock
and the regolith (Ghosh 1971). Figure 3.9 shows an average litholog which reveals
as many as six different geological layer. An average resistivity range given by
Telford et al. (1976), corresponding to various typical rocks is useful to interpret
inversion results of the VES data collected in basement complex. In the present
case, it is interpreted that the aquifer system mainly consists of clayey sand and
fractured/weathered granites/granulites. It is observed that the clay layer is inter-
preted with the lowest resistivity range among all geological formations encoun-
tered in the study area (Fig. 3.9). The low resistivity of clay layer is explained by
high cation exchange capacity and the presence of electrical double layer that
contributes the enhanced electrochemical properties of the clay mineral in the
mineral-water interface (Revil and Leroy 2001; Leroy and Revil 2004). The
basement rock is interpreted as granite/granulites which contains minerals with high
resistivity. Although, the coefficient of determination ðr 2 Þ obtained between TDS
56 S. Maiti and G. Gupta
Fig. 3.9 (a, b) Generalized representative borehole logs and its corresponding resistivity values in
the study area
and resistivity of the earth r 2 ¼ 0:71 is not so high as indicated in Fig. 3.8, but this
procedure allows by using resistivity information to gather authentic information on
how TDS is varied at places where no wells are available for sample analysis.
It may be noted that in the typical hard rock setting, one to one interpretation of
resistivity with lithology/hydro-geological section is somewhat difficult where
3 Integrated Geoelectrical and Hydrochemical Investigation … 57
3.5 Conclusions
Acknowledgements We thank directors IIT(ISM), Dhanbad and IIG, New Panvel for giving
permission to publish the work. SM is grateful to Ministry of Earth Sciences (MoES), Govt. of
India, New Delhi, India, for partial financial support through the Grant No. MoES/P.O. (Geosci)/
44/2015.
References
American Public Health Association (APHA) (1985) Standard methds for the examination of
water and waste, 16th edn. American Public Health Association, Washington, p 100
Asfahani J (2007) Electrical earth resistivity surveying for delineating the characteristics of ground
water in a semi-arid region in the Khanasser Valley, northern Syria. Hydrol Process 21:1085–
1097
Batte AG, Muwanga A, Sigrist PW, Owor M (2008) Vertical electrical sounding as an exploration
technique to improve on the certainty of groundwater yield in the fractured crystalline
basement aquifers of eastern Uganda. Hydrogeol Jour 16:1683–1693
Bhattacharya BB, Shalivahan, Sen MK (2003) Use of VFSA for resolution, sensitivity and
uncertainty analysis in 1D DC resistivity and IP inversion. Geophys Prosp 51:393–408
Biswas A, Sharma SP (2015) Interpretation of self-potential anomaly over idealized bodies and
analysis of ambiguity using very fast simulated annealing global optimization technique. Near
Surf Geophys 13:179–195
Bishop CM (1995) Neural networks for pattern recognition. Oxford University Press
Brown E, Shougstad MW, Fishmen MJ (1974) Methods for collection and analysis of water
sample for dissolved minerals and gases. Government Printing Office, Washington
Central Ground Water Board (CGWB) (2009) Ground water information Sindhudurg district
Maharashtra by S.C. Paranjape, pp 1–16
3 Integrated Geoelectrical and Hydrochemical Investigation … 59
Chandra S, Ahmed S, Ram A, Dewandel B (2008) Estimation of hard rock aquifers hydraulic
conductivity from geoelectrical measurements: a theoretical development with field applica-
tion. J Hydrol 357:218–227
Chandra A, Agarwal A, Shalivahan and Singh RK, (2017) Grey wolf optimisation for inversion of
layered earth geophysical datasets. Near Surf Geophys 15:499–513
Constable S, Parker RL, Constable CG (1987) Occam’s inversion: a practical algorithm for
generating smooth models from electromagnetic sounding data. Geophysics 52(3):289–300
Das A, Maiti S, Naidu S, Gupta G (2017) Estimation of spatial variability of aquifer parameters
from geophysical methods: a case study of Sindhudurg district, Maharashtra, India. Stoch
Environ Res Risk Assess. 31:1709–1726. https://doi.org/10.1007/s00477-016-1317-4
Deshpande GG (1998) Geology of Maharashtra. Geological Society of India, Bangalore
Duane S, Kennedy AD, Pendleton B, Roweth D (1987) Hybrid Monte Carlo. Phys Lett B
195:216–222
Ekinci YL, Demirci A (2008) A damped least-squares inversion program for the interpretation of
Schlumberger sounding curves. J Appl Sci 8(22):4070–4078
El-Qady G, Ushijima K (2001) Inversion of DC resistivity data using artificial neural networks.
Geophys Prospect 49:417–430
Ghosh D (1971) The application of linear filter theory on the direct interpretation of geoelectrical
resistivity sounding measurements. Geophys Prosp 19:192–217
Gupta G, Maiti S, Erram VC (2014) Analysis of electrical resistivity data in resolving the saline
and fresh water aquifers in west coast Maharashtra. J Geol Soc India 84:555–568
Inman JR (1975) Resistivity inversion with ridge regression. Geophysics 40:798–817
Karmakar M, Maiti S, Singh A, Ojha M, Maity B (2018) Mapping of rock types using a joint
approach by combining the multivariate statistics, self-organizing map and Bayesian neural
networks: an example from IODP 323 site. Marine Geophysical Research 39(3):407–419.
http://dx.doi.org/10.1007/s11001-017-9327-2
Koefoed O (1970) A fast method for determining the layer distribution from the raised kernel
function in geoelectrical soundings. Geophys Prosp 18:564–570
Kumar D, Rai SN, Thiagaranjan S, Ratna Kumari Y (2014) Evaluation of heterogeneous aquifers
in hard rocks from resistivity sounding data in parts of Kalmeshwar taluk of Nagpur district,
India. Curr Sci 107(7):1137–1145
Lampinen J, Vehtari A (2001) Bayesian approach for neural networks-review and case studies.
Neural Netw 14:7–24
Leroy P, Revil A (2004) A triple-layer model of the surface electrochemical properties of clay
minerals. J Colloid Interface Sci 270:371–380
Lines LR, Treitel S (1984) Tutorial: A review of least-squares inversion and its application to
geophysical problems. Geophys Prospect 32:159–186
Machiwal D, Jha MK (2015) Identifying sources of groundwater contamination in a hard-rock
aquifer system using multivariate statistical analyses and GIS-based geostatistical modeling
techniques. J Hydrol 4:80–110
Maiti S, Gupta G, Erram VC, Tiwari RK (2011) Inversion of Schlumberger resistivity sounding
data from the critically dynamic Koyna region using hybrid Monte Carlo-based neural network
approach. Nonli Proceses Geophys 18:179–192
Maiti S, Erram VC, Gupta G, Tiwari RK (2012) ANN based inversion of DC resistivity data for
groundwater exploration in hard rock terrain of western Maharashtra (India). J Hydrol 464–
465:281–293
Maiti S, Gupta G, Erram VC, Tiwari RK (2013a) Delineation of shallow resistivity structure
around Malvan, Konkan region, Maharashtra by neural network inversion using vertical
electrical sounding measurements. Environ Earth Sci 68:779–794
Maiti S, Erram VC, Gupta G, Tiwari RK, Kulkarni UD, Sangpal RR (2013b) Assessment of
groundwater quality: a fusion of geochemical and geophysical information via Bayesian neural
networks. Environ Monit Assess 185:3445–3465
60 S. Maiti and G. Gupta
Maiti S, Das A, Shah R, Gupta G (2017) Application of automatic relevance determination model
for groundwater quality index prediction by combining hydro-geochemical and geo-electrical
data. Model Earth Syst Environ 3:1371–1382. https://doi.org/10.1007/s40808-017-0369-x
Mondal NC, Singh VP, Ahmed S (2013) Delineating shallow saline groundwater zones from
Southern India using geophysical indicators. Environ Monit Assess 185:4869–4886
May R (1976) Simple mathematical models with very complicated dynamics. Nature 261:459–467
Menke W (1984) Geophysical data analysis: discrete inverse theory. Academic Press, Inc., New
York
Menke W, Menke J (2012) Environmental data analysis with Matlab. Elsevier, ISBN:
9780123918871, 288 page
O‘Neill DJ, Merrick NP (1984) A digital linear filter for resistivity sounding with a generalized
electrode array. Geophys Prospect 32:105–123
Revil A, Leroy P (2001) Hydroelectric coupling in a clayey material. Geophys Res Lett 28:1643–
1646
Roy IG (1999) An efficient nonlinear least squares 1D inversion scheme for resistivity and IP
sounding data. Geophys Prospect 47:527–550
Song SH, Lee JY, Park N (2007) Use of vertical electrical soundings to delineate seawater
intrusion in a coastal area of Byunsan, Korea. Environ Geol 52:1207–1219
Telford NW, Geldart LP, Sheriff RS, Keys DA (1976) Applied geophysics. Cambridge University
Press, London
Telford WM, Geldart LP, Sheriff RE (1990) Applied Geophysics. 2nd Edition, Cambridge
University Press, London
Tiwari RK, Maiti S (2011) Bayesian neural network modeling of tree-ring temperature variability
record from the Western Himalayas. Nonlinear Processes in Geophysics 18(2):515–528. http://
dx.doi.org/10.5194/npg-18-515-2011
Todd DK (1959) Groundwater hydrology. Wiley, New York, pp 277–294
Uchida T (1991) Two-dimensional resistivity inversion for Schlumberger sounding. Geophys
Explor Jpn 44:1–17
Verma RK, Rao MK, Rao CV (1980) Resistivity investigations for ground water in metamorphic
areas near Dhanbad, India. Ground Water 18:46–55
Zohdy AR (1989) A new method for automatic interpretation of Schlumberger and Wenner
sounding curves. Geophysics 54:245–253
Chapter 4
Modeling Streaming Potential in Porous
and Fractured Media, Description
and Benefits of the Effective Excess
Charge Density Approach
4.1 Introduction
Fig. 4.1 a Sketch of the electrical double layer. Distribution of b the excess charge and c the pore
water velocity as a function of the distance from the shear plan (modified from Jougnot et al. 2012)
Pozzi 1995). However, the HS equation cannot be applied for partially saturated
conditions and the evolution of the streaming potential coupling coefficient when
the water saturation decreases is still the subject of important debates in the com-
munity (e.g., Allègre et al. 2014; Fiorentino et al. 2016; Zhang et al. 2017).
An alternative approach to model the electrokinetic coupling phenomena is
based on the excess charge located in the EDL which is dragged by the water flow
in the pore space. It was first formulated by Kormiltsev et al. (1998) as the elec-
trokinetic coefficient, and later physically developed by Revil and co-workers using
different up-scaling methods (e.g., Revil and Leroy 2004; Linde et al. 2007; Revil
et al. 2007; Jougnot et al. 2012). This chapter aims at describing the theory and the
usefulness of the effective excess charge density approach to better understand and
model the generation of the streaming potential. First, the theory of this approach
will be described, linking it to the more traditional approach that uses the coupling
coefficient. Then, the evolution of the effective excess charge with different rock
properties and environmental variables will be studied. Finally, this approach will
be used to simulate the generation of the streaming potential in two complex media:
a partially saturated soil and a fractured domain.
4.2 Theory
Figure 4.1a shows a schematic description of the EDL that develops at the interface
between a charged mineral and the pore water solution. The amount and the sign of
the surface charge can vary from one mineral to another or with varying pH (e.g.,
Leroy and Revil 2004). We here call Q0 the surface charge of the mineral (in C
m−2) that are counterbalanced by the charge (i.e., counterions) located in the EDL.
These counterions are distributed between: (1) the Stern layer, sometimes called
fixed layer as the ions are sorbed onto the mineral surface, and (2) the diffuse layer
(also called Gouy–Chapman layer), where ions are less affected by the surface
charges and can diffuse more freely. At thermodynamic equilibrium and in satu-
rated conditions, these charges respect the following charge balance equation:
Ssw
v ¼ 0;
Q0 þ Qb þ Q ð4:1Þ
Vw
where Ssw is the surface of the mineral (in m2), Vw is the water volume in the pore
space (in m3), Qb is the charge of the Stern layer (in C m−2), and Q v is the
volumetric charge density in the diffuse layer (in C m−3). In partially saturated
conditions, that is, when the pore space contains air and water, an additional
interface and electrical double layer are present in the porous media (e.g., Leroy
et al. 2012). The specific surface area of the air-solution interface is considered to be
4 Modeling Streaming Potential in Porous … 65
negligible by many authors compared to the mineral-solution one (e.g., Revil et al.
2007; Linde et al. 2007). However, some works have recently challenged this
hypothesis (e.g., Allègre et al. 2015; Fiorentino et al. 2017).
While the Stern layer contains only counterions and has negligible thickness, the
diffuse layer contains both counterions and co-ions and its thickness strongly
depends on the pore solution chemistry. The distribution of ions in the diffuse layer
is determined by the local electrical potential w (in V) distribution as a function of
the distance from the shear plane, x:
x
wðxÞ ¼ f exp ; ð4:2Þ
lD
where f is the so-called zeta potential (in V), the local electrical potential at the
shear plane, and lD is the Debye length (in m) defined as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
e w kB T
lD ¼ ; ð4:3Þ
2NA Ie20
1X N
I¼ z2 C0 ; ð4:4Þ
2 i¼1 i i
where N is the number of ionic species i, zi and Ci0 are the valence and the
concentration (in mol L−1) of the ith ionic species. More precisely, Ci0 is the
concentration of the ionic species outside the EDL (i.e., in the free electrolyte). In
the diffuse layer, and under the assumption that the pores are larger than the diffuse
layer (i.e., thin layer assumption), the concentration of each ionic species follows:
zi e0 wðxÞ
Ci ðxÞ ¼ Ci0 exp : ð4:5Þ
kB T
The excess charge distribution in the diffuse layer can be expressed by the sum
of charges from each species (see Fig. 4.1b):
X
N
v ðxÞ ¼ NA
Q zi e0 Ci ðxÞ: ð4:6Þ
i¼1
66 D. Jougnot et al.
From the above equations, it becomes easy to see that the thickness of the diffuse
layer is related to the Debye length. The diffuse layer extension corresponds to the
fraction of the pore space for which a significant amount of excess charge is not
negligible: i.e., roughly 4 lD (Fig. 4.1b).
The constitutive equations describing the coupling between the electrical field and
the water flow can be written as follow (e.g., Nourbehecht 1963):
j ru
¼ L ð4:7Þ
u rðpw qw gzÞ
where j is the electrical current (in A m−2), u is the water flux (in m s−1), u is the
electrical potential (in V), pw is the water pressure (in Pa), g is the gravitational
constant (in m s−2), z is the elevation (in m) and qw the water density (kg m−3). The
coupling matrix L is defined as:
r LEK
L¼ ð4:8Þ
LEK k
gw
where r is the electrical conductivity of the medium (in S m−1), k is the medium
permeability (in m2), and gw is the dynamic viscosity of the water (Pa s). From this
coupling matrix, one can easily identify the Ohm’s law and the Darcy’s law through
L11 (i.e., r) and L22 (i.e., k=gw ), respectively. Following Onsager (1931), the two
non-diagonal terms should be equal and correspond to the electrokinetic coupling
coefficient LEK. It can be used to describe both the electrokinetic coupling (i.e., a
water flow induces an electrical current) and the electro-osmotic coupling (i.e., an
electrical current induces a water flow) in porous media. However, for most
environmental applications (except for compacted clay rocks), the effect of
electro-osmosis on the water flow can be safely neglected (e.g., Revil et al. 1999).
In this case, the system can be simplified by neglecting L21:
k
u¼ rðpw qw gzÞ: ð4:10Þ
gw
Using this simplification and considering that there is no external current in the
system (i.e., no current injection, and thus r j ¼ 0), Sill (1983) proposes the
following Poisson’s equation for describing the streaming potential generation:
4 Modeling Streaming Potential in Porous … 67
r ðrruÞ ¼ r jS ; ð4:11Þ
where jS is the streaming current density (in A m−2) resulting from the electroki-
netic coupling phenomenon that can be written as:
Note that Eq. (4.12) is often expressed as a function of the hydraulic head
gradient H (in m), which yields to:
LEK ew f
CHS ¼ ¼ ; ð4:14Þ
r gw rw
where rw is the pore water electrical conductivity (in S m−1). See also the complete
derivation in Rice and Whitehead (1965). The HS equation has proven to be very
useful as, in absence of external current, it relates the electrical potential difference
Du that can be measured at the boundaries of a sample to the pressure difference
Dpw to which it is submitted:
Du
C HS
¼ : ð4:15Þ
Dpw j¼0
However, Eq. (4.14) is only valid when the surface conductivity of the minerals
can be neglected. When it is not the case, modified versions of Eq. (4.14) have been
proposed in the literature (e.g., Revil et al. 1999; Glover and Déry 2010). Another
limitation with the HS coupling coefficient is when considering a porous medium
under partially saturated conditions. Many models have been proposed to describe
the evolution of the coupling coefficient with variable water saturation (e.g., Perrier
and Morat 2000; Guichet et al. 2003; Revil and Cerepi 2004; Allegre et al. 2010,
2015). Nevertheless, as illustrated in Zhang et al. (2017) (their Fig. 4.1), no con-
sensus has been found on the behavior of the coupling coefficient as a function of
water saturation as it seems to differ from one medium to another.
In order to deal with these two issues (i.e., surface conductivity and partially
saturated media), an alternative approach can be used to describe the coupling
coefficient. In this case, the electrokinetic coupling variable becomes the excess
charge which is effectively dragged by the water flow in the pore space.
68 D. Jougnot et al.
where the middle term LEK gkw is expressed in C m−3 and corresponds to a volumetric
excess charge as defined in Sect. 4.2.1. It is therefore possible to re-write Eq. (4.12)
as:
^ v u;
jS ¼ Q ð4:17Þ
where Q^ v (in C m−3) is the volumetric excess charge which is effectively dragged
by the water flow in the pore space (called a in Kormiltsev et al. 1998).
Independently from Kormiltsev et al. (1998), Revil and Leroy (2004) developed a
theoretical framework for various coupling properties based on this effective excess
charge approach for saturated porous media. In this work, a formulation for the
electrokinetic coupling coefficient is given as an alternative to the HS coupling
coefficient (Eq. 4.14):
^ vk
Q
C EK ¼ : ð4:18Þ
rgw
with krel ðSw Þ the relative permeability function comprised between 0 and 1. In the
following, the upper/lower script rel refers to the value of a parameter relatively to
its value under fully water saturated conditions.
4 Modeling Streaming Potential in Porous … 69
Following the definition of Guichet et al. (2003), the relative coupling coefficient
EK
Crel (unitless) can then be expressed as relative to the value in saturated conditions
ðCsat Þ which yield to (Linde et al. 2007; Jackson 2010):
EK
From Eq. (4.19), it is interesting to note that the coupling coefficient results from
the product of three different petrophysical properties of the porous medium: k, r,
and Q ^ v . Therefore, the coupling coefficient strongly depends on these parameters
and their evolution. The permeability, k, and the electrical conductivity, r, are two
extensively studied properties that have been shown to vary by orders of magnitude
between the different lithologies, but also for varying water saturation and, for r,
different pore water conductivities.
Various petrophysical relationships exist to describe k and r. The permeability
can be expressed as a function of the porosity and the medium tortuosity (e.g.,
Kozeny 1927; Carman 1937; Soldi et al. 2017) or the water saturation (e.g., Brooks
and Corey 1964; van Genuchten 1980; Soldi et al. 2017). On the other side, the
electrical conductivity depends on the porosity, the water saturation and the pore
water conductivity (e.g., Archie 1942; Waxman and Smits 1968; Linde et al. 2006).
However, the evolution of the effective excess charge density still remains
unknown. The present contribution aims at better describing this property, its
evolution, and its usefulness to understand and model the streaming current gen-
eration in porous and fractured media.
The determination of the effective excess charge density has been the subject of
only a couple of studies during the last two decades. One can identify two main
ways to determine this crucial parameter: (1) empirically from experimental mea-
surements and (2) numerically or analytically through an up-scaling procedure.
Based on previous studies from the literature and the theoretical framework
described by Kormiltsev et al. (1998), Titov et al. (2002) first showed that Q ^v
strongly depends on the medium permeability. Then, Jardani et al. (2007) proposed
a very useful and effective empirical relationship:
^ v Þ ¼ A1 þ A2 logðkÞ;
logðQ ð4:21Þ
Fig. 4.2 Effective excess charge density of various porous media as a function of the permeability
(modified from Guarracino and Jougnot 2018)
provide a fairly good first approximation for all kinds of water saturated porous
media that range from gravels to clay (Fig. 4.2). Note that other empirical rela-
^ v to the
tionships can be found in the literature (e.g., Bolève et al. 2012). Linking Q
permeability seems fairly logical since both properties depend on the interface
between mineral and solution: the permeability through viscous energy dissipation
and the effective excess charge density through the EDL. However, the use of this
relationship is limited by the fact that it does not take into account other physical
properties like porosity and the chemical composition of the pore water. This
particular point has been discussed by Jougnot et al. (2015) while modeling the SP
response of a saline tracer infiltration in the near surface.
The second approach to obtain the effective excess charge density is through an
up-scaling procedure. In this approach the transport of the excess charge density by
the water flux in the medium is explicitly considered. In order to perform this
up-scaling, one must simplify the problem using geometrical approximations to
describe the porous medium. Following the original work of von Smoluchovski
(1903), it is possible to consider electrokinetic coupling phenomena occurring in a
capillary (e.g., Rice and Whitehead 1965; Packard 1953) or in a bundle of capil-
laries (e.g., Bernabé 1998; Jackson 2008, 2010; Jackson and Leinov 2012).
More recently, Guarracino and Jougnot (2018) proposed an analytical mecha-
nistic model to determine the effective excess charge under saturated conditions for
a bundle of capillaries. This model is based on a two-steps up-scaling procedure
that was proposed numerically by Jougnot et al. (2012): (1) from the EDL scale to
the effective excess charge in a single capillary and then (2) from one capillary to a
bundle of capillaries (i.e., the REV).
4 Modeling Streaming Potential in Porous … 71
Based on the EDL description and the assumptions presented in Sect. 4.2.1,
Guarracino and Jougnot (2018) derived a closed-form equation for the effective
^ R (in C m−3):
excess charge density in a single capillary with the radius R (in m), Q v
" 3 #
^ R ðRÞ ¼ 8NA e 0 C 0
e 0 f e 0 f
Q w
2 : ð4:22Þ
v
ðR=lD Þ kB T 3kB T
Then, by considering a fractal law for the pore size distribution, that is a power
law distribution relating the pore size R to the number of pores in the medium N
(R) (e.g., Guarracino et al. 2014; Tyler and Wheatcraft 1990; Yu et al. 2003):
REV D
R
NðRÞ ¼ ; ð4:23Þ
R
where the parameters controlling Q ^ REV can be decomposed in two main parts
v
(1) the geometrical properties (i.e., petrophysical properties): porosity /, perme-
ability k, and hydraulic tortuosity s and (2) the electro-chemical properties: ionic
concentration, Debye length, and Zeta potential. Note that all these properties can
be estimated independently. By arranging Eq. (4.24), it is possible to derive the
empirical relationship proposed by Jardani et al. (2007) (Eq. 4.21) and to obtain
expressions for the fitting constants A1 and A2 in terms of fractal dimension and
chemical parameters. The performance of the model is tested with the extensive
data set presented in Fig. 4.2.
Under partially saturated conditions, that is, when the water volume in the pore
space diminishes, the behavior of the effective excess charge is still under discus-
sion. One could see two different up-scaling approaches to determine it: (1) the
volume averaging approach and (2) the flux-averaging approach.
The volume averaging approach to determine the evolution of Q ^ REV ðSw Þ was
v
first proposed by Linde et al. (2007) to explain the data from a sand column
drainage experiment and described in detail by Revil et al. (2007) in a very com-
plete electrokinetic framework in partially saturated porous media. This up-scaling
approach is built on the fact that no matter the medium water saturation, the surface
72 D. Jougnot et al.
charge to counterbalance is constant. That is, when the water volume decreases, the
total excess charge diminishes but its density increases linearly. It yields:
^ REV;sat
^ REV ðSw Þ ¼ Qv
Q : ð4:25Þ
v
Sw
where Q ^ R ðRÞ is the effective excess charge density (in C m−3) in a given capillary
v
R as expressed by Eq. (4.21), vR ðRÞ is the pore water velocity in the capillary (in
m s−1), and fD ðRÞ is the capillary size distribution of the considered medium.
Although this flux-averaging model can consider any kind of capillary size distri-
bution, Jougnot et al. (2012) propose to infer fD ðRÞ from the hydrodynamic
properties of the considered porous medium. It yields two approaches: (1) the water
retention (WR) and (2) the relative permeability (RP) based on the corresponding
hydrodynamic functions. From various studies, it has been shown that the WR
approach tends to better predict the relative evolution of the effective excess charge
density as a function of saturation, while the RP approach performs better for
amplitude prediction (e.g., Jougnot et al. 2012, 2015). Therefore, following the
proposition of Jougnot et al. (2015), we suggest that the effective excess charge
density under partially saturated conditions can be obtained by:
^ REV ðSw Þ ¼ Q
Q ^ REV;rel ðSw ÞQ
^ REV;sat ; ð4:27Þ
v v v
4 Modeling Streaming Potential in Porous … 73
where the saturated effective excess charge density Q ^ REV;sat can be obtained from
v
Eq. (4.24) and the relative excess charge density Qv^ REV;rel
ðSw Þ can be determined
using Eq. (4.26).
It is worth noting that Jougnot and Linde (2013) shown that the predictions of
Eqs. (4.25) and (4.26) can overlap over a large range of saturations for certain
sandy materials (e.g., the one used in Linde et al. 2007), which explains why the
volume averaging model performed well in Linde et al. (2007) and possibly in
Mboh et al. (2012) as they used a similar material.
From the theory section, it clearly appears that the pore water salinity strongly
influences the electrokinetic coupling. Indeed, the pore water electrical conductivity
explicitly appears in the coupling coefficient definition (Eqs. 4.14 and 4.18).
Nevertheless, the pore water salinity also strongly affects the properties of the EDL.
Equation (4.3) shows its effect on the extension of the diffuse layer, while many
studies show that it also changes the value of the f-potential (e.g., Pride and Morgan
1991; Jaafar et al. 2009; Li et al. 2016). In the present approach, we use the Pride
and Morgan (1991) model:
where a = −6.43 mV and b = 20.85 mV for silicate-based materials and for NaCl
brine according to Jaafar et al. (2009) if f is expressed in mV and Cw0 in mol L−1.
Note that the behavior of the f-potential as a function of the salinity is challenged in
the literature (e.g., see the discussion in Fiorentino et al. 2016).
Figure 4.3 illustrates the evolution of the effective excess charge density as a
function of the pore water salinity (i.e., ionic concentration of NaCl). The experi-
mental data come from the study of Pengra et al. (1999) for different porous media,
while the model is the one proposed by Guarracino and Jougnot (2018) where the
hydraulic tortuosity (i.e., the only parameter not measured by Pengra et al. 1999) is
optimized to fit the data. The overall fit is pretty good, indicating that the
Guarracino and Jougnot (2018) model correctly takes into account the effect of the
salinity on the EDL and the resulting effective excess charge density.
74 D. Jougnot et al.
Fig. 4.3 Effective excess charge density of various porous media as a function of the ionic
concentration of the NaCl in the pore water. The experimental data have been extracted from
Pengra et al. (1999)
From the previous section, it is clear that the effective excess charge density
depends on petrophysical properties like permeability, porosity, and hydraulic
tortuosity. In contrast to other models, Guarracino and Jougnot (2018) explicitly
express Q ^ REV as a function of these three parameters.
v
Glover and Déry (2010) conducted a series of electrokinetic coupling mea-
surements on well-sorted glass bead samples of different radii at two pore water
salinities. They also performed an extensive petrophysical characterization of each
sample, providing all the necessary parameters to test the model proposed by
Guarracino and Jougnot (2018), except for the hydraulic tortuosity. Figure 4.4a
shows the Q ^ REV predicted by this model (using s ¼ 1:2) and by the empirical
v
relationship from Jardani et al. (2007) (Eq. 4.21). Figure 4.4b compares the cou-
pling coefficient measured by Glover and Déry (2010) with the coupling coeffi-
cients calculated using the Q ^ REV predicted by the models of Guarracino and
v
Jougnot (2018) and Jardani et al. (2007), respectively. One can see that the model
informed by the measured petrophysical parameters performs better and is able to
reproduce the entire dataset with a single value of hydraulic tortuosity. A better fit
can be obtained by optimizing the hydraulic tortuosity for each sample.
The link between effective excess charge density and hydraulic tortuosity can be
explicitly seen in Eq. (4.24). Unfortunately, the hydraulic tortuosity is not an easy
4 Modeling Streaming Potential in Porous … 75
Fig. 4.4 a Effective excess charge density and b coupling coefficient of various porous media as a
function of the permeability for s ¼ 1:2
parameter to measure for all type of porous media; Clennell (1997) provides an
extensive review of the different definitions and models to estimate tortuosities in
porous media. Among others, Windsauer et al. (1952) proposes a simple way to
relate the hydraulic tortuosity to the formation factor F, which is easier to measure:
pffiffiffiffiffiffiffi
se ¼ F/; ð4:29Þ
Fig. 4.5 Predicted versus best-fit tortuosities for the data from Glover and Déry (2010) and
Pengra et al. (1999). The plain black line corresponds to 1:1 values (i.e., s ¼ se )
The effect of the saturation on the effective excess charge density remains a vivid
area of investigation as explained in the theory section. In the present chapter we
compare the volume averaging approach of Linde et al. (2007) with the flux
averaging approach of Jougnot et al. (2012). Figure 4.6a and b show the evolution
Fig. 4.6 Effective excess charge density of various soil types as a function of the saturation
(modified from Jougnot et al. 2012)
4 Modeling Streaming Potential in Porous … 77
Sw Srw
Se ¼ ; ð4:30Þ
1 Srw
to remove the effect of the residual water saturation Srw differences between the soil
types. It explains why all the volume averaging curves are not superposed.
It can be noted that the effective excess charge always increases as the water
saturation decreases. For the volume averaging model, it is due to decreasing volume
of water in the pores while the amount of charges to compensate remains constant. For
the flux averaging model, this behaviour is due to the fact that larger pores (i.e., smaller
relative volume of EDL in the capillary) are desaturating first, letting the water flow
through the smaller pores (i.e., smaller relative volume of EDL in the capillary).
Hence, the model proposed by Jougnot et al. (2012) yields a soil-specific function
^ REV ðSw Þ which strongly depends on the soil texture and shows very important
Q v
changes with saturation, i.e., up to 9 orders of magnitudes (see also Soldi et al. 2019).
Following the formalism exposed in Bernabé (1998), the hydraulic flux Q and the
electrical flux J in a single capillary of radius r and length l are given by the two
coupled equations:
8
> Rr
>
> pr4 ðPu Pd Þ
Q ¼ 8g l þ g pee0 r2 f
1 r2 f rwðr Þdr ðVu V
2 dÞ
>
> l
>
> 0
< Rr
J ¼ peeg0 r 1 r22 f rwðr Þdr ðPu P dÞ
2
ð4:31Þ
>
>
l
>
0
> r
2
>
> 2pe2 e20 R ð Þ Rr
>
: g
dw
r dr r
dr þ 2prf r cosh kT dr ðVu V
zew ð r Þ
l
dÞ
0 0
78 D. Jougnot et al.
where Pu (resp. Vu) is the upstream hydraulic pressure (resp. the electrical potential)
and Pd (resp. Vd) the downstream pressure (resp. potential). The computation of the
local electrical potential distribution w inside the capillary is obtained by solving the
Poisson-Boltzmann equation inside infinite cylinders, as done by Leroy and
Maineult (2018).
The set of Eq. (4.31) can be written as:
Ql ¼ ch ðPu Pd Þ þ cc ðVu Vd Þ
ð4:32Þ
Jl ¼ cc ðPu Pd Þ ce ðVu Vd Þ
where ch is the modified hydraulic conductance (in m4 Pa−1 s−1), ce the electrical
conductance (in S m), and cc the coupling conductance (in m4 V−1 s−1).
We consider a square tube network as depicted in Fig. 4.7, for which all tubes are of
length l (in m).
Writing the conservation laws (Kirchhoff’s laws 1845) for the hydraulic flux and
the electrical flux at each node of the network, combined with the appropriate
boundary conditions, provides a linear system to be solved, whose unknown are the
hydraulic pressures and electrical potential Pi,j and Vi,j at all nodes and the electrical
potential V0 (for more details, see Appendix 1).
DV 0 V0
CEK ¼ ¼ ¼ V0 : ð4:33Þ
DP 01
^ v ¼ grC
EK
Q ; ð4:34Þ
k
neglecting the surface conductivity and introducing the formation factor gives:
^ v ¼ grw C
EK
Q : ð4:35Þ
kF
For the computation of the quantities k/1 and F/, see Appendix 2.
4.4.4 Applications
where SD is the standard deviation. We explored different values of rpeak (i.e., 0.1,
0.2, 0.3, 0.5, 1, 2, 3, 5 and 10 µm), and took SD = 0.5.
Note that to compute the fluid conductivity rw associated with the concentration
Cw0 , we used the empirical relation given by Sen and Goode (1992) for NaCl brine:
80 D. Jougnot et al.
2:36 þ 0:099T 3
rw ¼ 5:6 þ 0:27T 1:5 104 T 2 M M2 ð4:37Þ
1 þ 0:214M
where T is the temperature (in °C) and M is the molality (in mol kg−1). To convert
the concentration Cw0 into molality, we use the CRC Handbook Table at 20 °C (Lide
2008). The f-potential is then obtained from the relation given by Jaafar et al.
(2009) (Eq. 4.28).
Figure 4.9 shows the electrokinetic coupling coefficients calculated for different
2D pore networks having different permeabilities. For ionic concentrations larger
than 0.01 mol/L, the coupling coefficient appears not to be dependent on the per-
meability despite the influence of the permeability in its definition (Eq. 4.18). This
is a result of the linearly dependence on the permeability of the effective excess
charge density, canceling the permeability in Eq. (4.18). That can be clearly seen in
Fig. 4.10, where the analytical model of Guarracino and Jougnot (2018) predicts
4 Modeling Streaming Potential in Porous … 81
Fig. 4.9 Coupling coefficient of the 2D pore networks as a function of permeability for different
NaCl concentrations
Fig. 4.10 Evolution of the excess charge density as a function of permeability for different NaCl
concentrations: comparison between the 2D pore network results and model predictions of
Guarracino and Jougnot (2018) for the corresponding ionic concentrations and s ¼ 2:3
accurately the evolution of the effective excess charge density for the synthetic 2D
pore network. Note that this very good fit is obtained from all the calculated
parameters, with only one unknown, which has been fitted: s ¼ 2:3.
82 D. Jougnot et al.
Then, for 0.001 mol/L, the coupling coefficient tends to decrease for the lowest
permeabilities (below 10−12 m2), i.e., the smallest pore sizes, which also correspond
to the poorer fit of Eq. (4.24) on the synthetic data. This can be expected from the
assumptions of Guarracino and Jougnot (2018) model which is only valid when the
EDL thickness is small enough in comparison to the pore size (see discussion in
Jougnot et al. 2019). Low permeabilities and low salinities therefore show a limi-
tation of their model, as the local potential distribution in the EDL must be com-
puted by solving the Poisson-Boltzmann equation (see Leroy and Maineult 2018).
The present section illustrates the usefulness of the effective excess charge approach
to model the streaming potential distribution in two kinds of complex media: a
partially saturated soil and a fractured aquifer.
Figure 4.11 describes the numerical framework that we use to simulate the
streaming potential distribution resulting from a rainfall infiltration in a sandy loam
soil. As explained in the theory section, the results of the hydrological simulation
are used as input parameters for the electrical problem. In this scheme, it is clear
that the electrokinetic coupling parameter is the effective excess charge density
Fig. 4.11 Numerical framework for the simulation of the streaming potential distribution in a
partially saturated porous medium
4 Modeling Streaming Potential in Porous … 83
Fig. 4.12 Simulation results of the rainwater infiltration: a precipitation, b water saturation, and
c streaming potential as a function of time
even if the water saturation distribution also plays a role through the electrical
conductivity, affecting the amplitude of the SP signals.
We consider a homogeneous sandy loam soil subjected to a rainfall event
(Fig. 4.12a). The initial hydraulic conditions of the soil are set to hydrostatic
equilibrium with a water table localized at 2.5 m depth. Following the work of
Jougnot et al. (2015), the hydrological problem is solved using Hydrus 1D. This
code solves the Richards equation to determine the evolution of the water saturation
(Fig. 4.12b) and Darcy velocity as a function of depth and time. We choose the van
Genuchten model to describe the water retention and the relative permeability
84 D. Jougnot et al.
Fig. 4.13 a Comparison of the effective excess charge density as a function of the water
saturation using Jougnot et al. (2012) RP and WR approaches and Linde et al. (2007). b Vertical
^ ðSw Þ
distribution of the SP signal resulting for the rainwater infiltration using the corresponding Qv
function at two different times, t = 2 and 10 d, for the plain and the dashed lines, respectively
functions, using the average hydrodynamic properties for a sandy loam soil pro-
posed by Carsel and Parrish (1988).
The electrical problem is solved using a home-made code (for details please refer
to Jougnot et al. 2015). As illustrated in Fig. 4.11, the hydrological simulation
ouputs (i.e., the water saturation and the Darcy velocity distribution in both space
and time) are used as input parameters for the electrical problem. The electrical
conductivity is determined using Archie (1942) with the following petrophysical
parameters: m = 1.40 the cementation exponent and n = 1.57 the saturation
exponent. The effective excess charge is determined using Eq. (4.27) in which
^ REV;rel ðSw Þ can be obtained from the WR or the RP flux averaging approach of
Q v
Jougnot et al. 2012, or using the volume averaging approach of Linde et al. (2007)
as explained in Sect. 4.2.4.2 (Fig. 4.13a).
Figure 4.12c shows the results of the numerical simulation of the streaming
potential for virtual electrodes localized at different depths in the soil. Note that the
reference electrode is localized at a depth of 3 m. As the rainwater infiltration front
progresses in the soil, the SP signal starts to increase. An electrode localized at the
soil surface should be able to capture the highest signal amplitude during the
rainfall, while the deeper electrodes show a time shift related to the time needed for
the water flow to reach the electrode. The signal amplitude also decreases with
depth as the Darcy velocity diminishes during the infiltration. The multimodal
nature of the rainfall also vanishes, showing only a single SP peak at a depth of
5 cm. The Q ^ ðSw Þ function used to plot Fig. 4.12c is the RP approach from Jougnot
v
et al. (2012). Figure 4.13b shows the strong influence of the chosen approach on the
vertical distribution of the signal amplitude at two different times (t = 2 and 10 d).
These results are consistent with the findings of Linde et al. (2011), that is, the
4 Modeling Streaming Potential in Porous … 85
volume averaging model of Linde et al. (2007) does not allow to reproduce the
large vertical SP signals that can be found in the literature (e.g., Doussan et al.
2002; Jougnot et al. 2015).
The effective excess charge can also be used for modeling the streaming potential
arising from groundwater flow in fractured media (e.g., Fagerlund and Heinson
2003; Wishart et al. 2006, 2008; Maineult et al. 2013). Existing studies focusing on
this phenomenon in fractured rocks suggest that monitoring the corresponding
streaming potential under pumping conditions can help to identify the presence of
fractures that interact with the surrounding matrix (Roubinet et al. 2016; DesRoches
et al. 2017). This has been demonstrated with numerical approaches relying on a
discrete representation of the considered fractures that are coupled to the matrix by
using either the finite element method with adapted meshing (DesRoches et al. 2017)
or the finite volume method within a dual-porosity framework (Roubinet et al. 2016).
The latter method is used here to illustrate the sensitivity of SP signals to
hydraulically active fractures, and in particular to fractures subject to important
fracture-matrix exchanges. For this purpose, we consider the coupled fluid flow and
streaming potential problem described in Fig. 4.11 that we apply to fractured porous
domains under saturated conditions. In this case, the fluid flow problem is solved by
considering Darcy’s law and Darcy-scale mass conservation under steady-state
conditions, and the effective excess charge is evaluated from the fracture and matrix
permeability by adapting the strategy proposed in Jougnot et al. (2012) to two
infinite plates having known separation and using the empirical relationship defined
by Jardani et al. (2007), respectively. As shown in Roubinet et al. (2016), both fluid
flow and streaming current must be simulated in the fractures and matrix to ade-
quately solve this problem, even if the matrix is characterized by a very low per-
meability. Furthermore, relatively small fracture densities should be considered in
order to individually detect the fractures that are hydraulically active.
Figure 4.14a, b, and c show three examples of fractal fracture network models
defined by Watanabe and Takahashi (1995) for characterizing geothermal reservoirs
and used in Gisladottir et al. (2016) for simulating heat transfer in these reservoirs.
In these models, the number of fractures and the relative fracture lengths (i.e., the
ratio of fracture to domain length) are defined from the fracture density, the smallest
fracture length, and the fractal dimension that are set to 2.5, 0.1, and 1 m,
respectively, considering a square domain of length L ¼ 100 m. The positions of
these fractures are randomly distributed, their angle can be equal to h1 ¼ 25 or
h2 ¼ 145 with equal probability, and their aperture is set to 10−3 m. Note that we
also add a deterministic fracture whose center is located at the domain center and
whose angle is set to h1 (represented in red in Figs. 4.14a–c). Finally, the fracture
and matrix conductivity are set to 5 102 and 5 104 S m1 , respectively, and
86
50 50 50
0 0 0
0 50 100 0 50 100 0 50 100
60 60 60
50 50 50
40 40 40
20 20 20
0 0 0 0 0 0
0 50 100 0 50 100 0 50 100
Fig. 4.14 a–c Studied fractured domains where the red cross represents the position of the considered pumping well. d–f Spatial distribution of the SP signal
Dux;y (in mV) with respect to a reference electrode located at position (x, y) = (0,0). g–i Polar plots of the SP signal Dur (in mV) along the dashed white circle
plotted in (d–f) with respect to the minimum value measured along this circle and represented with a white cross
D. Jougnot et al.
4 Modeling Streaming Potential in Porous … 87
the matrix permeability to 1015 m2 .The fluid flow and streaming potential problem
is solved by considering (i) a pumping rate of 103 m3 s1 applied at the domain
center, (ii) gradient head boundary conditions with hydraulic head set to 1 and 0 m
on the left and right sides of the domain, respectively, and (iii) a current insulation
condition on all borders. Figure 4.14 shows the resulting difference in potential
Dux;y and Dur where the white (Figs. 4.14d–f) and black (in Figs. 4.14g–i) dots
represent the two largest SP signals measured along the dashed white circles that are
plotted in Figs. 4.14d–f. These results show that a strong SP signal is observed for
the primary fracture in which the pumping rate is applied when this fracture is not
intersected by secondary fractures that are close to the pumping well
(Figs. 4.14d and g). On the contrary, when the primary fracture is intersected by
secondary fractures that are close to the pumping well and not connected to the
domain borders, strong SP signals are observed at the extremities of the single
secondary fracture (Figs. 4.14e and h) or the pair of secondary fractures
(Figs. 4.14f and i). As demonstrated in existing studies (DesRoches et al. 2017;
Roubinet et al. 2016), these results suggest that strong SP signals are associated
with hydraulically active fractures, and that the largest values of SP measurements
are related to important fracture-matrix exchanges.
Modeling of the streaming current generation and the corresponding electrical field
requires a good understanding of electrokinetic coupling phenomena that occur
when the water flows in porous and fractured media. This complex process can be
modeled with two electrokinetic coupling parameters: coupling coefficient and
effective excess charge. In this chapter we focused on the latter.
Considering the effective excess charge approach is quite recent (Kormiltsev
et al. 1998) in comparison with the use of the coupling coefficient. Unlike the
coupling coefficient, the effective excess charge density shows a strong dependence
on petrophysical parameters (permeability, porosity, ionic concentration in the pore
water). This has been highlighted by both empirical (Titov et al. 2002; Jardani et al.
2007) and mechanistic (Jougnot et al. 2012; Guarracino and Jougnot 2018)
approaches. The mechanistic approaches that we discuss in this chapter are based
on the up-scaling process called flux-averaging as they propose an effective value
for the excess charge density which is related to pore scale properties of the EDL
and water flow patterns.
Under saturated conditions, Guarracino and Jougnot (2018) model shows a
linear dependence with geometrical properties (permeability, porosity, hydraulic
tortuosity) and non-linears ones to chemical properties (ionic concentration, zeta
potential). In Sects. 4.4.3 and 4.4.4, we show that it provides good match with
published laboratory data for various types of media as long as the model
assumptions are respected (i.e., the pore radius should be 5 times larger than the
88 D. Jougnot et al.
This appendix details the calculation of the pressure and the electrical potential in
the pore network. The Kirchhoff (1845) laws for the water flow and the electrical
current at node of coordinates (i, j), which express the conservation of mass and the
conservation of charge respectively, write:
8 h
>
> ci1;j!i;j Pi;j Pi1;j þ cci1;j!i;j Vi;j Vi1;j
> h
>
>
> ci þ 1;j!i;j Pi;j Pi þ 1;j þ cciþ 1;j!i;j Vi;j Vi þ 1;j
>
>
>
> chi;j1!i;j Pi;j Pi;j1 þ cci;j1!i;j Vi;j Vi;j1
>
>
< ch Pi;j Pi;j þ 1 þ cci;j þ 1!i;j Vi;j Vi;j þ 1 ¼ 0
i;j þ 1!i;j ; ð4:38Þ
>
> cci1;j!i;j Pi;j Pi1;j cei1;j!i;j Vi;j Vi1;j
>
> þ cc
>
> i þ 1;j!i;j Pi;j Pi þ 1;j ceiþ 1;j!i;j Vi;j Vi þ 1;j
>
>
>
> þ ci;j1!i;j Pi;j Pi;j1 cei;j1!i;j Vi;j Vi;j1
c
>
>
:
þ cci;j þ 1!i;j Pi;j Pi;j þ 1 cei;j þ 1!i;j Vi;j Vi;j þ 1 ¼ 0
0
chi1;j!i;j Pi1;j þ chiþ 1;j!i;j Pi þ 1;j jhi;j Pi;j þ chi;j1!i;j Pi;j1 þ chi;j þ 1!i;j Pi;j þ 1
B
B cci1;j!i;j Vi1;j cciþ 1;j!i;j Vi þ 1;j þ jci;j Vi;j cci;j1!i;j Vi;j1 cci;j þ 1!i;j Vi;j þ 1 ¼ 0
B ;
B cc
@ i1;j!i;j Pi1;j ci þ 1;j!i;j Pi þ 1;j þ ji;j Pi;j ci;j1!i;j Pi;j1 ci;j þ 1!i;j Pi;j þ 1
c c c c
þ cei1;j!i;j Vi1;j þ ceiþ 1;j!i;j Vi þ 1;j jei;j Vi;j þ cei;j1!i;j Vi;j1 þ cei;j þ 1!i;j Vi;j þ 1 ¼ 0
ð4:39Þ
with:
8
>
> j h
¼ c h
þ c h
þ c h
þ c h
>
< i;j
i1;j!i;j i þ 1;j!i;j i;j1!i;j i;j þ 1!i;j
jci;j ¼ cci1;j!i;j þ cciþ 1;j!i;j þ cci;j1!i;j þ cci;j þ 1!i;j ; ð4:40Þ
>
>
>
: je ¼ c e
i;j i1;j!i;j þ c e
i þ 1;j!i;j þ c e
i;j1!i;j þ c e
i;j þ 1!i;j
In i = 1 (no outward flux), j 2 2; Nj 1 , we have (see Fig. 4.1):
8 h
>
> c P2;j jh1;j P1;j þ ch1;j1!1;j P1;j1 þ ch1;j þ 1!1;j P1;j þ 1
< 2;j!1;jc
c2;j!1;j V2;j þ jc1;j V1;j cc1;j1!1;j V1;j1 cc1;j þ 1!1;j V1;j þ 1 ¼ 0
; ð4:41Þ
>
> cc P2;j þ jc1;j P1;j cc1;j1!1;j P1;j1 cc1;j þ 1!1;j P1;j þ 1
: 2;j!1;j
þ c2;j!1;j V2;j j1;j V1;j þ c1;j1!1;j V1;j1 þ c1;j þ 1!1;j V1;j þ 1 ¼ 0
e e e e
with:
8
>
> j h
¼ c h
þ c h
þ c h
>
< 1;j
2;j!1;j 1;j1!1;j 1;j þ 1!1;j
jc1;j ¼ cc2;j!1;j þ cc1;j1!1;j þ cc1;j þ 1!1;j : ð4:42Þ
>
>
>
: je ¼ ce
1;j 2;j!1;j þ c e
1;j1!1;j þ c e
1;j þ 1!1;j
In i = Ni (no outward flux), j 2 2; Nj 1 (see Fig. 4.7), we have:
8 h
>
> c P
i ;j Ni 1;j
jhNi ;j PNi ;j þ chNi ;j1!Ni ;j PNi ;j1 þ chNi ;j þ 1!Ni ;j PNi ;j þ 1
< Ni 1;j!N
cNi 1;j!Ni ;j VNi 1;j þ jcNi ;j VNi ;j ccNi ;j1!Ni ;j VNi ;j1 ccNi ;j þ 1!Ni ;j VNi ;j þ 1 ¼ 0
c
;
>
> cc P
i ;j Ni 1;j
þ jcNi ;j PNi ;j ccNi ;j1!Ni ;j PNi ;j1 ccNi ;j þ 1!Ni ;j PNi ;j þ 1
: Ni 1;j!N
þ cNi 1;j!Ni ;j VNi 1;j jNi ;j VNi ;j þ cNi ;j1!Ni ;j VNi ;j1 þ cNi ;j þ 1!Ni ;j VNi ;j þ 1 ¼ 0
e e e e
ð4:43Þ
with:
8
>
> jh
¼ c h
þ c h
þ c h
>
< Ni ;j
Ni 1;j!Ni ;j Ni ;j1!Ni ;j Ni ;j þ 1!Ni ;j
jNi ;j ¼ cNi 1;j!Ni ;j þ cNi ;j1!Ni ;j þ ccNi ;j þ 1!Ni ;j :
c c c
ð4:44Þ
>
>
>
: je ¼ ce
Ni 1;j!Ni ;j þ cNi ;j1!Ni ;j þ cNi ;j þ 1!Ni ;j
e e
Ni ;j
90 D. Jougnot et al.
Pi;1 ¼ 1; ð4:45Þ
Vi;1 ¼ V0 ; ð4:46Þ
in such a way that the total entering electrical flux is equal to 0, that is to say:
Ni
X
Ni X
Ji;1!i;2 l ¼ cci;1!i;2 Pi;2 Pi;1 cei;1!i;2 Vi;2 Vi;1 ¼ 0; ð4:47Þ
i¼1 i¼1
Pi;Nj ¼ 0; ð4:49Þ
Vi;Nj ¼ 0: ð4:50Þ
This set of equations forms a linear system, whose unknowns are the hydraulic
pressures and electrical potential Pi,j and Vi,j at all nodes and the electrical potential
V0.
In the classical case, the hydraulic flux Fx!y through a tube linking two nodes x and
y writes (Poiseuille law):
prx!y
4
Px Py
Fx!y ¼ ¼ ghx!y Px Py : ð4:51Þ
8g l
prx!y
4
Uhx!y ¼ Fx!y l ¼ Px Py ¼ chx!y Px Py : ð4:52Þ
8g
Under the assumption that the surface conductivity can be neglected, the elec-
trical flux Jx!y writes:
Vx Vy
Jx!y ¼ rf prx!y
2
¼ gex!y Vx Vy : ð4:53Þ
l
Then, we use the modified electrical flux Uex!y to simplify from l and the fluid
conductivity rf
l
Uex!y ¼ Jx!y ¼ prx!y
2
Vx Vy ¼ cex!y Vx Vy : ð4:54Þ
rf
Kirchhoff’s law (1845) applies at any node inside the square network (Fig. 4.7):
with Z equal to F or J respectively. Using Eqs. (4.51) or (4.53), this leads to:
ai;j1!i;j Xi;j1 þ ai1;j!i;j Xi1;j ai;j1!i;j þ ai1;j!i;j þ ai þ 1;j!i;j þ ai;j þ 1!i;j Xi;j
þ ai þ 1;j!i;j Xi þ 1;j þ ai;j þ 1!i;j Xi;j þ 1 ¼ 0
ð4:56Þ
with a = r4 and X = P for the hydraulic case, and a = r2 et X = V for the electrical
case.
For the nodes on the borders of the network, Eq. (4.56) is easily modified to take
into account the boundary conditions (i.e., no outward flow for i = 1 and i = Ni,
P = 1 or V = 1 for j = 1, and P = 0 and V = 0 for j = Nj).
A linear system is obtained; the hydraulic pressure or electrical potential at the
nodes of the network are the NiNj unknowns. The modified fluxes can be computed
after the system is solved through the use of Eqs. (4.52) and (4.54).
One can compute the permeability of the network using Darcy’s law:
g Nj 1 URout=in
h
gQL
k¼ ¼ 2 ð4:57Þ
SjDPj l Ni 1 jDPj
where Q is the hydraulic flux, L is the length of the network along the flow direction
(i.e., j-direction), S the transversal section, and the total out-flowing and in-flowing
fluxes are given by:
92 D. Jougnot et al.
8 NP
i 1
>
>
< URout ¼ Uhi;Nj 1!i;Nj
h
NP
i¼1
: ð4:58Þ
>
> i 1
: UhRin ¼ Uhi;1!i;2
i¼1
To estimate S and the porosity of the network, we extend the 2D network into a
virtual 3D one, by adding at each node two vertical tubes of length l/2, which do not
contribute to the transport. The section S is thus equal to (Ni–1)l2, and the porosity
/ can be expressed as:
ðNi 1ÞNj þ ðNi 1ÞNj þ Ni Nj p r 2 l
/¼ : ð4:59Þ
ðNi 1Þ Nj 1 l3
1 Nj 1 URout=in
e
1 rr 1 JL
¼ ¼ ¼ 2 : ð4:61Þ
F rw rw SjDV j l Ni 1 jDV j
References
Winsauer WO, Shearin H Jr, Masson P, Williams M (1952) Resistivity of brine-saturated sands in
relation to pore geometry. AAPG Bull 36(2):253–277
Yu B, Li J, Li Z, Zou M (2003) Permeabilities of unsaturated fractal porous media. Int J Multiph
Flow 29(10):1625–1642
Zhang J, Vinogradov J, Leinov E, Jackson MD (2017) Streaming potential during drainage and
imbibition. J Geophys Res: Solid Earth 122(6):4413–4435. https://doi.org/10.1002/
2017JB014242
Chapter 5
Forward Modeling and Inversion
of Very Low Frequency Electromagnetic
Data Over Rugged Topography Using
2D Triangular Elements
Abstract Very Low Frequency (VLF) method is the simplest among all electro-
magnetic induction techniques with a variety of geophysical applications. Various
qualitative and quantitative interpretation techniques are developed to interpret VLF
data. However, most of the quantitative interpretation techniques using numerical
methods are based on rectangular grid based modeling. Such discretization may
produce a large error while modeling the surface topography and also, number of
the model parameters will increase drastically to model the surface topography. An
effort is made to develop an inversion approach based on triangular grid to reduce
the problem size by utilizing minimal number of model parameter and model the
rugged topography accurately. First, an analytical expression is derived to compute
the vertical component of the magnetic field due to a triangular block of an assigned
current density in the block. Subsequently, subsurface structure is optimized in
terms of apparent current density using preconditioned conjugate gradient method
by considering a model including the surface topography. Since, noise reduction in
VLF data is also an important step of interpretation, nonlinear empirical mode
decomposition technique is used to minimize the noise from the data. Efficacy of
the developed interpretation approach is discussed on one synthetic and one field
data set. Inversions for the synthetic and field data sets were carried out with and
without topographical variation. The study reveals that depth to the top of the
subsurface conductor is correlated with the surface topography. Inversion of VLF
data measured over topographical variation with the flat surface model resulted in
A. Singh (&)
Department of Earth Sciences, Indian Institute of Technology-Bombay, Powai, Mumbai,
India
e-mail: anandsinghiitb@gmail.com
S. K. Maurya S. P. Sharma
Department of Geology and Geophysics, Indian Institute of Technology-Kharagpur,
Kharagpur, West Bengal, India
e-mail: satishmaurya@iitkgp.ac.in
S. P. Sharma
e-mail: spsharma@gg.iitkgp.ac.in
shifting of the depth to the top of the conductor to an incorrect position (shallower
depth). Therefore, consideration of topography in the inversion scheme is vital for
delineation of subsurface structures accurately. The efficacy of triangular element in
depicting the topography accurately and inversion is demonstrated in the present
study.
Keywords VLF-EM Forward modeling Inversion Rugged topography
Subsurface conductor
5.1 Introduction
The very low frequency (VLF) electromagnetic method is the most straightforward
electromagnetic induction method. The VLF method uses very high frequency (15–
30 kHz) in comparison to the other electromagnetic methods. The transmitter which
we utilized in the VLF method uses located in coastal areas worldwide. These
transmitted signals travel a long distance and act like plane wave electromagnetic
signals. Further, these signals diffuse into the Earth’s subsurface and produce
electromagnetic induction because of its low frequency. First, Paal (1965, 1968)
discovered these signals and proposed VLF electromagnetic methods for detection
of conductive structures in the Earth’s subsurface. Subsequently, VLF method has
been routinely used for different applications and widely used for shallow sub-
surface investigations with mineral, groundwater, and environmental investigations
(Table 5.1).
Interpretation of measured field data is the most significant part of any geo-
physical investigation. Various approaches have been developed for the qualitative
and quantitative interpretations of VLF electromagnetic data (see Table 5.2 for
more detail). The first qualitative approach using filtering of VLF data proposed by
Fraser (1969) yields the approximate location of the conductor while Karous and
Hjelt (1983) approach yields the apparent current density at pseudo depths. Saydam
(1981) proposed an understanding outline to identify the depth of vertical sheet like
conductors and the conductivity-thickness product from tilt angle and ellipticity
measurements of VLF data. Boukerbout et al. (2003) used continuous wavelet
transform method to depict the source in the potential field and demonstrated its use
for VLF data. Padersen and Becken (2005) developed subsurface imaging approach
using equivalent current density. On the other hand, different quantitative inter-
pretations employing modeling and inversion of VLF data have been developed to
delineate 2D and 3D subsurface structures (see the detail in Table 5.2). First,
quantitative approach was developed by Kaikkonen (1979). Development of
quantitative interpretation techniques needs very specialized knowledge of
numerical computations, large memory for computing resources as well as signif-
icant amount of computation time. Instead of qualitative in nature, Karous and Hjelt
(1983) filtering approach has been widely used till today due to its computational
simplicity and depicting the approximate subsurface structure. However, the
5 Forward Modeling and Inversion of Very Low … 99
apparent current density at pseudo depth is the main limitation of above filtering
approach. It is rather difficult to estimate the actual depth as well as shape of the
causative body from filtering approach. Even vertical structures appear as dipping
structures in apparent current density cross-sections. Singh and Sharma (2015)
developed and compared the results of different qualitative and quantitative tech-
niques. It was observed that the images were not reliable in the filtering technique
developed by Karous and Hjelt (1983) and Pedersen and Becken (2005). They also
discuss that the resistivity inversion will yield more reasonable results.
Most of the above mentioned interpretation techniques have been established for
the flat earth models. Since VLF technique operates with very high frequency range
(15–30 kHz), very little topographic variations (amplitudes in the order of 10 m)
can also affect the VLF response (Singh and Sharma 2016). To deal with the
100 A. Singh et al.
topography, Baranwal et al. (2011) developed first finite element based 2D inver-
sion approach for VLF data. However, their forward formulation is based on the
unstructured triangle grids and inversion was performed on rectangle and octree
grid based meshes. Further, Singh and Sharma (2016) developed VLF imaging
technique and also compared their results with the resistivity distribution of
Baranwal et al. (2011). Their tests have been carried out on the octree-mesh grids.
In the present study, an attempt is made to optimize the actual current density at
actual depth by developing a new computational procedure on triangular grids.
There are two important aspects of using the triangular grids (1) reduction of the
problem size, and (2) ability to model topography exactly. In Sect. 5.2, triangular
grid based two dimensional forward modeling of VLF data was computed in terms
of current density. Section 5.3 deals with inverse modeling of VLF data using
preconditioned conjugate gradient method. In Sect. 5.4, nonlinear empirical mode
decomposition (EMD) technique is presented to reduce noise from VLF data.
Finally in Sect. 5.5, testing of the forward formulation and the efficiency of the
inversion algorithm has been investigated through one synthetic and one field
example having variable topography.
5 Forward Modeling and Inversion of Very Low … 101
In VLF method, the horizontal and vertical components of the magnetic field are the
quantity measured in the field and the VLF real component/anomaly (ratio of the
vertical and horizontal components of the magnetic field) is represented in per-
centage (Smith and Ward 1974)
Hz
% Real anomaly ¼ 100 Real : ð5:1Þ
Hx
where, H x ¼ H Px þ H Sx (the terms P and S denote primary and secondary field part,
subscripts x and z are the horizontal and vertical part of field, respectively). In
Eq. 5.1, H x can be considered uniform in the study area as H Sx is very small
compared to H Px . In the present work, H x is considered as one in order to balance
the relationship between the vertical component of the magnetic field and the real
anomaly. In the present work, the subsurface structure below the measured topo-
graphical profile can be discretized into small 2D triangular meshes with each cell
having different current density. A uniform current density is considered in each
triangular element. Further, the vertical component of the magnetic field due to a
particular triangular cell is computed along a topographical profile. At any topo-
graphical measurement point, the total vertical component of the magnetic field is
considered as the algebraic sum of the vertical component of the magnetic field
computed for the individual triangular cells.
The final expression for the vertical component of the magnetic field for a triangular
cell ABC at the measurement point (0, 0) can be written as (see Eq. 5.23 of
Appendix)
" #
j X3
x k k
z z k k
x k
k þ þ
DHzk ð0; 0Þ ¼ i i 1 i i 1
ðzi þ 1 zki Þðhkiþ 1 hki Þ;
2p i¼1 ðxkiþ 1 xki Þ2 þ ðzkiþ 1 zki Þ2
#
rik
þ ðxi þ 1 xi Þln k
k k
ð5:2Þ
ri þ 1
where, DHzk is the vertical component of the magnetic field due to the kth triangular
cell, jk is the current density of the kth cell, and ðxki ; zki Þ is the coordinate of the kth
triangular cell. In Eq. 5.2, the coordinates ðxk4 ; zk4 Þ are representing ðxk1 ; zk1 Þ at i ¼ 3.
As mentioned earlier, combined/cumulative vertical magnetic field at the
102 A. Singh et al.
measurement point (0, 0) can be expressed as summation of the vertical field due to
each single triangular cell as
X
M
Hz ð0; 0Þ ¼ DHzk ; ð5:3Þ
k¼1
where, M is the number of the triangular cells. Let ðx0j ; z0j Þ (j =1, 2, 3, …, N) is the
coordinate locations of N data point. Here, N is the number of the datasets along the
measured profile. The vertical component of the magnetic field at the jth data can be
written as
X
M
dj ¼ aj;k jk ; ð5:4Þ
k¼1
here jk is the current density of the kth triangular cell and aj;k is the matrix element
signifying the influence of the kth triangular cell at jth measurement point. Using
Eq. (5.2), the precise expression for aj;k is
2 3
ðxki x0j Þðzkiþ 1 z0j Þðzki z0j Þðxkiþ 1 x0j Þ
6 ððxkiþ 1 x0j Þðxki x0j ÞÞ2 þ ððzkiþ 1 z0j Þðzki x0j ÞÞ2 7
3 66 ððzk z j Þ ðzk z j ÞÞðhk hk Þ
7
1 X 6
7
7;
aðj; kÞ ¼ iþ1 0 i 0 iþ1 i ð5:5Þ
2p i¼1 6
6 k
# 7
7
4 j j ri 5
þ ððxi þ 1 x0 Þ ðxi x0 ÞÞ ln k
k k
ri þ 1
where,
zk z j
hki ¼ tan1 xik x0j ;
i 0
zk z j
hkiþ 1 ¼ tan1 xikþ 1 x0j ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
iþ1 0
and qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rikþ 1 ¼ ðxkiþ 1 x0j Þ2 þ ðzkiþ 1 z0j Þ2 :
d ¼ G:m ð5:6Þ
where, d is the measured vertical magnetic field, m represent model parameter i.e.
the current density, G is the N M dimensional kernel matrix of which the ele-
ments are the anomalies caused by the triangular cells.
5 Forward Modeling and Inversion of Very Low … 103
P ¼ ‘b I: ð5:7Þ
where b is a constant related to the rapid attenuation of the VLF anomaly with the
distance ‘ between the triangular block and the observation points. Pilkington
(1997) inverted the magnetic data considering b equal to 3. Liu et al. (2013)
performed the inversion results with different values of b and suggested that b 4
for 2D inversion of magnetic data. The algorithm for very low frequency imaging
using the conjugate gradient method is as follows.
Algorithm
(1) Choose a start model m0
(2) Set i ¼ 0. Compute dpre 0 ¼ Gm0 and Dd0 ¼ d
obs
dpre
0 .
(3) while Gmi d tol and i\imax do
obs 2
and rjþ 1 ¼ rj tj H j
(7) if pj ePCG , proceed to step 8; otherwise, j ¼ j þ 1:
Proceed to step 5.
104 A. Singh et al.
(8) mi þ 1 ¼ mi þ Dmi .
(9) if mi þ 1 mmin , then mi þ 1 ¼ mmin . if mi þ 1 mmax , then
mi þ 1 ¼ mmax :
(10) i ¼ i þ 1.
(11) end
Since the VLF transmitters work in the frequency range 15–30 kHz. Thus, it can
result in relatively high background geologic noise. Everett and Weiss (2002)
showed that background geologic noise is signal generated broadband noise and
exhibit a 1/frequency spectrum. There are other noises associated with VLF data
which are nonlinear and harmonic. For example, Park and Helliwell (1978) iden-
tified the characteristics of power line radiation waves (i.e. harmonic waves) which
are associated with signals of VLF transmitters. VLF noises associated with the
earthquake, volcanic eruptions, tornadoes and dust storms are also nonlinear and
non-stationary (Barr et al. 2000). It is highlighted that most data sets recorded over
natural phenomena are either nonlinear or non-stationary which can be identified by
the distorted waveforms in the data. Since VLF response is the convolution of the
causative geologic features and background geologic noise; therefore, the associ-
ated noise is difficult to remove using conventional digital linear filtering techniques
of Fraser (1969) and Karous and Hjelt (1983). Since interpretation of VLF anomaly
is strongly dependent on geologic noise (Jeng et al. 2007), therefore, recovery of
signals with significant geologic information is necessity. For this purpose, we used
empirical mode decomposition technique developed by (Huang et al. 1998) which
was applied in the field of geophysics in particular VLF EM method by Jeng et al.
(2007).
A complicated data can be divided into a small and finite intrinsic mode function
(IMF) components with time-variable frequencies and amplitudes through
Empirical Mode Decomposition (EMD). The IMF which can be both frequency and
amplitude module characterizes the oscillation mode embedded in the data. It is not
restricted to a narrow-band but it can be non-stationary signal. A single profile VLF
data is presented in the spatial domain, we can divide a VLF signal into different
intrinsic oscillatory modes and instantaneous wave numbers of the data. Following
is the algorithm steps of the EMD (Huang et al. 1998):
5 Forward Modeling and Inversion of Very Low … 105
1. Identify all the local maxima and minima of the VLF electromagnetic signal
sðxÞ.
2. Create the upper and lower envelope of the extrema point (maxima and minima)
trough interpolation.
3. Generate the mean function using the upper and lower envelope, mðxÞ.
4. Compute the difference deðxÞ ¼ sðxÞ mðxÞ.
5. Stop the iteration, if deðxÞ becomes a zero-mean and make deðxÞ is an IMF1
named c1 ðxÞ; otherwise, go to step (1) and replace sðxÞ with deðxÞ.
6. Compute the residue signal rðxÞ ¼ sðxÞ c1 ðxÞ.
7. Repeat the procedure from steps (1) to (6) to obtain IMF2 , named c2 ðxÞ. To
obtain cn ðxÞ, continue steps (1)–(6) after n iterations.
8. Stop the process when residual signal rðxÞ becomes a monotonic function.
As a output of the algorithm, we can have a collection of n IMFs named from
c1 ðxÞ to cn ð xÞ and a residue rðxÞ. The VLF-EM signal can be written as:
X
n
sðxÞ ¼ ci ðxÞ þ rðxÞ: ð5:8Þ
i¼1
Fig. 5.1 a Synthetic real part of VLF-EM data (green solid line), white Gaussian noise added (S/
N = 5) data (blue dot) and reconstructed data using EMD algorithm (red solid line).
b Decomposition of noisy VLF-EM data into IMFs ðc1 c4 Þ and a residue r using Eq. 5.8
106 A. Singh et al.
5.5 Results
The results have been presented into two sections. In the first section, validity of the
developed forward formulation has been examined. Since, vertical component of
the magnetic field due to entire subsurface has been estimated using the principle of
superposition, therefore, principle of superposition has also been demonstrated. In
the second section, inversion results have been presented on synthetic and field
data.
In the present study, for the first time an attempt is made to compute the vertical
component of the magnetic field for a given current density distribution in con-
ductive subsurface structure using triangular cross-section elements. The validity of
Eq. 5.2 was demonstrated by computing the forward response for a given current
density using different approaches. The results are compared with (1) computation
of the vertical component of magnetic field (Hz) along a profile ranging from 0 to
1000 m using Biot-Savart law for a current carrying cable of infinite extent of same
current magnitude placed at a distance of 510 m in the profile direction and at a
depth of 30 m (Fig. 5.2a); (2) by computing the Hz response using equation
developed by Singh and Sharma (2016) along the above mentioned profile for a 2D
block of rectangular cross-section of dimension 20 20 m such that its center lies
location of 510 m in the profile direction with same current value of 1 A or
equivalent uniform current density of 0.0025 Amp/m2 (j = 1A/20 20) extended
for 500 to 520 m on profile direction and extending from 20 to 40 m depth value
Singh and Sharma 2016 (Fig. 5.2b). Finally, the 2D rectangular block is divided
into two triangular blocks with same uniform current density value (Fig. 5.2c).
Responses are computed in term of the vertical component of the magnetic field
from the formulation shown in Eq. 5.2 for individual triangular block and alge-
braically added for both blocks. This response exactly matches with the response
computed in the first approach for a current carrying cable and with the response of
the second approach for a rectangular block of uniform current density. This
expresses that developed Eq. 5.2 follows the physical principle.
Physically all the three above mentioned conditions are the same as they are
centered at the same location and same depth value and have same uniform current
density. Therefore all the three cases must yield the same value of Hz along the
given profile. Figure 5.3 establishes this fact. This approves the validity of the
newly derived Eq. 5.2. The principle of superposition was also confirmed with the
presently developed equation and it can be used to compute Hz due to 2D current
density distribution with a triangular cross-section due to entire subsurface.
5 Forward Modeling and Inversion of Very Low … 107
Fig. 5.2 Current density distribution model for different approaches. a One ampere current
carrying cable placed at 510 m in the profile direction and at a depth of 30 m. b A 2D block of
rectangular cross-section extended for 500–520 m on profile direction and extending from 20 to
40 m depth with an equivalent uniform current density of 0.0025 Amp/m2. c The rectangular block
shown in (b) assumed to consist of two triangular cross-sections with same uniform current density
value
Fig. 5.3 The vertical component of the magnetic field for the different approaches as shown in
Fig. 5.1
The synthetic model demonstrated in Fig. 5.4a comprises of two anomalous regions
having current density of 0.08 A/m2 respectively, in a zero current density
half-space with a smooth, but prominent sinusoidal elevated topography. These two
conductors are kept at different depth with different dipping direction. To show the
importance of topography, first synthetic data is inverted by considering exact
topography and subsequently the same VLF data is inverted by considering the flat
earth subsurface. Synthetic VLF data from −480 to 480 m with interval of 20 m is
inverted using the present inversion approach. Figure 5.4b is the result of model
including topography whereas Fig. 5.4c is the current density distribution by
considering the flat earth only. It is important to note that the position of conducting
body is seen to have shifted upwards from its true position. It is due to the coupling
of the response between the undulating surface and conductive bodies. True
positions of the conductive bodies are represented by the solid white line in
108 A. Singh et al.
Fig. 5.4 Inversion results on the synthetic VLF data. a Two conducting body (one vertical and
one dipping dyke) of same current density 0.08 A/m2 respectively in zero current density half
space including topography. 2D current density distribution image using inversion of real anomaly
of VLF data considering, b topography and c flat earth model
Fig. 5.4. Figure 5.4 clearly demonstrates that the topographical observation sur-
faces may become important when the scale length of topographic surface varia-
tions (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 accurate
results as all the inverted current density sections (Fig. 5.4) will fit the data pre-
cisely (Fig. 5.5).
The study area lies in the central part of the Eastern Indian Shield (Fig. 5.6a) and
associated with the South Purulia Shear Zone (SPSZ). The SPSZ is a contact zone
between the North Singhbhum Mobile Belt and the South Chotanagpur Granite
Gneissic Complex. The SPSZ is an extended area with strike trending in
ENE-WSW to E-W to ESE-WNW direction (Acharyya et al. 2006). It is to be noted
5 Forward Modeling and Inversion of Very Low … 109
Fig. 5.5 Fittings comparison between the observed real anomaly, modeled real anomaly (with
topography), and modeled real anomaly (without topography) for the inverted model shown in
Fig. 5.4
that there is a good similarity between Singhbhum Shear Zone and SPSZ in terms
of nature of mineral deposits. In the present work, VLF data acquisition is carried
out in December 2016 in Bhanratongri, West Bengal (Fig. 5.4b), India. Most of the
study region covered by ferruginous kaolinite (clay) and cultivated land soil.
ABEM-Wadi instrument is used to measure the data. Since strike of the geologic
feature is along E-W direction (Fig. 5.6b), VLF data is collected on two N-NE
trending traverses using 19.2 kHz frequency. The field data set was measured
across hill whose topography is varying with 10–20 m (Fig. 5.6b). The existence of
uranium mineralization (Kaolinite zone, host rock of uranium mineralization) is
expected beneath the hillocks. Therefore, a preliminary VLF survey was carried out
at the mentioned site.
Measurements were taken by a VLF system with 5 m interval along the two
profiles (P1 and P2) as shown in Fig. 5.6b. First, VLF data is processed using
empirical mode decomposition (EMD) technique as described in Sect. 5.4.1 to
remove the noise from the VLF data (Fig. 5.7). Figure 5.8 is the result of six IMFs
components and residue for the profile P1. It is noted that first two IMFs i.e. c1 and
c2 are clearly a random noise associated with VLF data. In the present study, these
two IMFs are excluded from the data and rest IMFs and residual are the result of
recovered signal of VLF data (Fig. 5.7a). Similar kind of strategy was applied for
profile 2 (P2) data and results are shown in Fig. 5.7b.
After processing the VLF data to remove the high frequency noise, inversion
was performed by considering the model including topography and flat earth model.
The, minimum and maximum value of current density was constrained within
−0.006 A/m2 and 0.015 A/m2, respectively. In Eq. 5.7, b was taken as 3 as con-
sidered by several authors (Liu et al. 2015; Singh and Sharma 2016). Inverted
model current density sections of profile P1 are illustrated in Fig. 5.9a, b.
Figure 5.9a depicts the result of the model including topography whereas Fig. 5.9b
is the result of flat earth consideration. A common color scale is used to generate
both the Figures. Similarly, Fig. 5.10a, b are the result of current density distri-
bution for the Profile 2. Three main anomalous bodies can be distinguished from the
110 A. Singh et al.
Fig. 5.6 a Central part of the Eastern Indian Shield, b location map and aerial view of the study
area (Bhanratongri, West Bengal) with VLF electromagnetic profile. White zone is kaolinite rich
rock which is the possible host rock of uranium mineralization
5 Forward Modeling and Inversion of Very Low … 111
Fig. 5.8 Collection of all IMFs ðc1 c6 Þ and a residue (r) of the field data along the profile P1
5 Forward Modeling and Inversion of Very Low … 113
Fig. 5.9 Inverted current density image section of profile P1 a with topography and b without
topography of the region
Fig. 5.10 Inverted current density image section of profile P2 a with topography and b without
topography of the region
A triangular grid based forward modeling and inversion algorithm was developed
with special emphasis on the surface topography to interpret VLF electromagnetic
data. First a new forward formulation is derived to compute vertical component of
the magnetic field using triangular cell (Appendix). The derivation which was
shown in Appendix is obtained from the Ampere’s law for a current carrying cable.
114 A. Singh et al.
The problem size (no of model parameter) is also reduced while performing the
modeling and inversion on triangular grids. One important feature of triangular
grids is the ability to represent the topography exactly.
Appendix
Same expression (Eq. 5.10) for the vertical component of magnetic field due to a
two dimensional body of trapezoidal cross section (as shown in Figure 5.13) can be
written in terms of z and h:
Zh2 Zz2
jy
DHz ð0; 0Þ ¼ cothdhdz ð5:10Þ
2p
h1 z1
Let us consider that the subsurface consist of 2D triangular cell whose strike is
perpendicular to the plain of Fig. 5.14. Electromagnetic induction in the subsurface
gives induced current build up in the subsurface conducting body. Thus there will
be distribution of current density in the subsurface according to the conductivity
distribution in the subsurface. Due to an elementary strip of triangular cross section
(see Fig. 5.14), the vertical component of the magnetic field at point P (0, 0) on the
earth surface will be given:
I
jy
DHz ð0; 0Þ ¼ z cot hdh ð5:12Þ
2p
The total vertical component of magnetic field on the surface at a point P due to
an elementary strip of triangular cross section can be written as summation of
vertical component of magnetic field due to its individual segment AB, BC and BA
as
z ¼ x tan h ¼ ðx a1 Þ tan w1
ð5:14Þ
¼ x tan w1 a1 tan w1
5 Forward Modeling and Inversion of Very Low … 117
and
a1 tan h tan w1
z ¼ x tan h ¼ ; ð5:16Þ
ðtan w1 tan hÞ
and
x2 x1
a1 ¼ PC 0 QC 0 ¼ x2 z2 cot w1 ¼ x2 z2 ð5:21Þ
z2 z1
Putting all the values from Eqs. 5.19–5.21 into 5.18, the expression for the
vertical component of magnetic field due to segment AB written as
" #
jy x1 z2 z1 x2 r1
DHz;AB ð0; 0Þ ¼ ðz2 z1 Þðh2 h1 Þ þ ðx2 x1 Þln
2p ðx2 x1 Þ2 þ ðz2 z1 Þ2 r2
ð5:22Þ
ð5:23Þ
118 A. Singh et al.
Where, DHz ð0; 0Þ is the vertical magnetic field component due to the triangular
cell, jy is the current density of the cell, and ðxi ; zi Þ where i = 1, 2, 3 are the
coordinates of the triangular cell and for third coordinate i = 3, the value of i þ 1
corresponds to the first coordinate i.e. i = 1.
References
Acharyya A, Ray S, Chaudhuri BK, Basu SK, Bhaduri SK, Sanyal AK (2006) Proterozoic rock
suites along South Purulia Shear Zone, Eastern India: evidence for rift—related setting. J Geol
Soc India 68:1069–1086
Al-Tarazi E, Rajab JA, Al-Naqa A, El-Waheidi M (2008) Detecting leachate plumes and
groundwater pollution at Ruseifa municipal landfill utilizing VLF-EM method. J Appl
Geophys 65:121–131
Baker HA, Myers JO (1980) A topographic correction for VLF-EM profiles based on model
studies. Geoexploration 18:135–144
Baranwal VC, Sharma SP (2006) Integrated geophysical studies in the east-Indian geothermal
province. Pure appl Geophys 163:209–227
Baranwal VC, Franke A, Borner RU, Spitzer K (2011) Unstructured grid based 2-D inversion of
VLF data for models including topography. J Appl Geophys 75:363–372
Barr R, Jones DL, Rodger CJ (2000) ELF and VLF radio waves. J Atmos Solar Terr Phys
62:1689–1718
Beamish D (1994) Two-dimensional regularized inversion of VLF data. J Appl Geophys 32:357–
374
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 AK, Payne KL, Stubben MA (1997) Mapping groundwater contamination using DC
resistivity and VLF geophysical method—a case study. Geophysics 62:80–86
Bernard J, Valla P (1991) Groundwater exploration in fissured media with electrical and VLF
methods. Geoexploration 27:81–91
Bosch FP, Müller I (2005) Improved karst exploration by VLF-EM-gradient survey: comparison
with other geophysical methods. Near Surf Geophys 3:299–310
Boukerbout H, Gibert D, Sailhac P (2003) Identification of sources of potential fields with the
continuous wavelet transform: application to VLF data. Geophys Res Lett 30(8) article 1427
Chouteau M, Zhang P, Chapellier D (1996) Computation of apparent resistivity profiles from
VLF-EM data using linear filtering. Geophys Prospect 44:215–232
Drahor MG (2006) Integrated geophysical studies in the upper part of Sardis archaeologicalsite.
Turkey J Appl Geophys 59:205–223
Drahor MG, Berge MA (2006) Geophysical investigations of the Seferihisar geothermal area,
Western Anatolia, Turkey. Geothermics 35(3):302–320
Ellis RG, Oldenburg DW (1994) Applied geophysical inversion. Geophys J Int 116:5–11
Everett ME, Weiss CJ (2002) Geological noise in near‐surface electromagnetic induction data.
Geophys Res Lett 29(1)
Fraser DC (1969) Contouring of VLF-EM data. Geophysics 34:958–967
Guerin R, Benderitter Y (1995) Shallow karst exploration using MT-VLF and DC resistivity
methods. Geophys Prospect 43:635–653
Günther T, Rücker C, Spitzer K (2006) Three-dimensional modelling and inversion of DC
resistivity data incorporating topography—II. Inversion. Geophys J Int 166:506–517
Gürer ÖF, Sarica-Filoreau N, Özburan M, Sangu E, Doğan B (2009) Progressive development of
the Büyük Menderes Graben based on new data, western Turkey. Geol Mag 146(5):652–673
5 Forward Modeling and Inversion of Very Low … 119
Huang NE, Sheen Z, Long SR, Wu MC, Shih HH, Zheng Q, Yen NC, Tung CC, Liu HH (1998)
The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary
time series analysis. In: Proceedings of the Royal Society of London A: mathematical, physical
and engineering sciences (Vol. 454, No. 1971, 903–995). The Royal Society
Jeng Y, Lin MJ, Chen CS, Wang YH (2007) Noise reduction and data recovery for a VLF-EM
survey using a nonlinear decomposition method. Geophysics 72(5)
Kaikkonen P (1979) Numerical VLF modeling. Geophys Prospect 27(4):815–834
Kaikkonen P, Sharma SP (1998) 2D nonlinear joint inversion of VLF and VLF-R data using
simulated annealing. J Appl Geophys 39:155–176
Kaikkonen P, Sharma SP, Mittal S (2012) 3D modeling and inversion of VLF and VLF-R
electromagnetic data. Geophysics 77(4):WB219–WB231
Kamm J, Pedersen LB (2014) Inversion of airborne tensor VLF data using integral equations.
Geophys J Int 198:775–794
Karous M, Hjelt SE (1983) Linear filtering of VLF dip-angle measurements. Geophys Prospect
31:782–794
Liu S, Hu X, Liu T, Feng J, Gao W, Qiu L (2013) Magnetization vector imaging for borehole
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 observation
surface using constrained Delaunay triangulation. Comput Geosci 76:18–30
Mackie RL, Madden TR (1993) Three-dimensional magnetotelluric inversion using conjugate
gradients. Geophys J Int 115:215–229
Mohanty WK, Mandal A, Sharma SP, Gupta S, Misra S (2011) Integrated geological and
geophysical studies for delineation of chromite deposits: a case study from Tangarparha,
Orissa, India Chromite exploration at Tangarparha. Geophysics 76:B173–B185
Monteiro Santos FA, Mateus A, Figueiras J, Gonçalves MA (2006) Mapping groundwater
contamination around a landfill facility using the VLF-EM method—a case study. J Appl
Geophys 60:115–125
Newman GA, Alumbaugh DL (2000) Three-dimensional magnetotelluric inversion using
non-linear conjugate gradients. Geophys J Int 140:410–424
Paal G (1965) Ore prospecting based on VLF-radio signals. Geoexploration 3:139–147
Paal G (1968) Very low frequency measurements in northern Sweden. Geoexploration 6:141–149
Park CG, Helliwell RA (1978) Magnetospheric effects of power line radiation. Science 200
(4343):727–730
Paterson NR, Ronka V (1971) Five years of surveying with very low frequency—electromagnetic
method. Geoexploration 9:7–26
Pazzi V, Tapete D, Cappuccini L, Fanti R (2016) An electric and electromagnetic geophysical
approach for subsurface investigation of anthropogenic mounds in an urban environment.
Geomorphology 273:335–347
Pedersen LB, Becken M (2005) Equivalent images derived from very low frequency (VLF) profile
data. Geophysics 70(3):G43–G50
Philips WJ, Richards WE (1975) A study of the effectiveness of the VLF method for the location
of narrow-mineralized fault zones. Geoexploration 13:215–226
Pilkington M (1997) 3D magnetic imaging using conjugate gradients. Geophysics 62:1132–1142
Poddar M, Rathor BS (1983) VLF survey of the weathered layer in southern India. Geophys
Prospect 31(3):524–537
Rodi W, Mackie RL (2001) Nonlinear conjugate gradients algorithm for 2D magnetotelluric
inversion. Geophysics 66:174–187
Saydam AS (1981) Very low-frequency electromagnetic interpretation using tilt angle and
ellipticity measurements. Geophysics 46:1594–1605
Sharma SP, Baranwal VC (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 SP, Kaikkonen P (1998) Two-dimensional non-linear inversion of VLF-R data using
simulated annealing. Geophys J Int 133:649–668
120 A. Singh et al.
Singh A, Sharma SP (2015) Fast imaging of subsurface conductors using very low frequency
electromagnetic data. Geophys Prospect 63:1355–1370
Singh A, Sharma SP (2016) Interpretation of very low frequency electromagnetic measurements in
terms of normalized current density over variable topography. J Appl Geophys 133:82–91
Singh A, Sharma SP (2018) Identification of different geologic units using fuzzy constrained
resistivity tomography. J Appl Geophys 148:127–138
Smith BD, Ward SH (1974) On the computation of polarization ellipse parameters. Geophysics
39:867–869
Spitzer K (1995) A 3-D finite-difference algorithm for DC resistivity modelling using conjugate
gradient methods. Geophys J Int 123:903–914
Sundararajan N, Ramesh Babu V, Chaturvedi AK (2011) Detection of basement fractures
favorable to uranium mineralization from VLF-EM signals. J Geophys Eng, 8:330–340
Tezkan B (1999) A review of environmental applications of quasi-stationary electromagnetic
techniques. Surv Geophys 20:279–308
Timur E (2009) Joint İnversion of Magnetic and Electromagnetic Data. DokuzEylul University,
The Graduate School of Natural and Applied Sciences, Ph.D. Thesis, İzmir
Timur E (2014) Magnetic susceptibility and VLF-R investigations for determining geothermal
blowout contaminated area: a case study from Alaşehir (Manisa/Turkey). Environ Earth Sci
72:2497–2510
Yamaguchi TI, Yamano M, Nagao T, Goto S (2001) Distribution of radioactive heat production
around an active fault and in accretionary prisms of southwest Japan. Phys Earth Planet Inter
126:269–277
Zhang J, Mackie RL, Madden TR (1995) 3D resistivity forward modeling and inversion using
conjugate gradients. Geophysics 60:1313–1325
Zlotnicki J, Vargemezis G, Mille A, Bruère F, Hammouya G (2006) State of the hydrothermal
activity of Soufriere of Guadeloupe volcano inferred by VLF surveys. J Appl Geophys 58:265–
279
Chapter 6
Forward and Inverse Modeling of Large
Loop TEM Data Over Multi-layer Earth
Models
6.1 Introduction
The TEM method is an inductive method in which the earth is energized by a strong
current in a rectangular/square/circular loop usually placed over the surface of the
earth. The current in the surface loop generates a time varying magnetic field that
passes through the ground and induces eddy currents in the subsurface. When
current is turned off, the primary time varying magnetic field produces an elec-
tromagnetic induction during this time and results in generation of secondary field
in the subsurface.
According to Kaufman and Keller (1983) the intensity of these currents at a
certain time and depth depends on the conductivity, size, and shape of subsurface
conductor and the voltage response, which is proportional to the time rate of change
of the secondary magnetic field, is measured. The system of induced currents at any
point of time after the current is turned-off, can be represented by a simple current
filament of the shape of transmitter loop which spreads outward and downward with
decreasing velocity and diminishing amplitude with time (Fig. 6.1) in form of
smoke ring (Nabighian 1979).
The velocity Vz at any time t with which the ring spreads away is given by
(Nabighian 1979)
2
Vz ¼ pffiffiffiffiffiffiffiffiffiffi ð6:1Þ
prlt
The subsurface conductivity and time interval after the transmitter current is turned
off controls the depth of investigation (Eq. 6.2). With increase in time, the current
intensity enters to greater depths. The current in transmitter loop is of equal periods of
on-time and off-time (Fig. 6.2). The TDEM signal is measured in the transmitter off
time period, when primary field is absent. The plot of TEM response plotted against
time on a logarithmic scale in a homogeneous medium shows three stages; the early
Fig. 6.2 The basic measuring principles of the TEM Central loop method with theoretical current
shown in blue and real current shown in red, over the time (after Christiansen 2003)
stage, an intermediate stage, and the late stage. Grant and West (1965) has shown that
in early time stage, the induced currents is independent of the conductivity and at the
time when transmitter current is switched off, it is confined to the surface of the
conductor (Weaver et al. 1970). In this stage, the distribution of current is only con-
trolled by the size and shape of the conductor (Grant and West 1965).
In intermediate stage the inward diffusion of the current pattern occurs due to
ohmic losses in region immediately inside the conductor showing a decrease in
magnetic field and induced emf and flow of new current. Thereafter, the late-time
stage occurs at which current distribution becomes roughly invariant with time. The
inductance and resistance of current ring becomes stable and both the currents and
related magnetic fields start decaying with a time constant, s
rla2
s¼ ð6:3Þ
p2
A large loop laid on the surface of the earth is one of the most widely used EM
source with option of measurements taken either inside or outside the loop source.
The source loop is usually a square/rectangular or circular to facilitate the inte-
gration over the loop analytically. A square loop can be approximated by a circular
loop of the same area if the receiver is located at the center of the loop or at some
distances away from the loop. A large loop TEM system represents a class of TEM
sounding methods, which consist of a large transmitter loop for generating the
primary field and a small receiver loop or magnetometer for recording the transient
voltage or magnetic field in the air or on the surface of the earth with receiver at the
center of the loop or at any arbitrary point either outside or inside the source
loop. In general, with a large loop source, one can procure the data in any one of the
configurations, namely, central loop configuration (with receiver at center of the
loop), in-loop configuration (receiver at an arbitrary in-loop point), coincident-loop
configuration (receiver loop coincident with the transmitter loop) and offset-loop
configuration (receiver at an arbitrary offset loop point), respectively. Of the three
configurations, the central loop system (for ground surveys) and coincident loop
system (for ground as well as airborne surveys) have been developed and used
extensively for exploration of mineral and geothermal resources, for mapping
contaminated ground water caused by hazardous waste and thickness of permafrost
layer, because of mathematical simplicity associated with the expressions of EM
fields, as compared to the in-loop and offset-loop systems. However, the capability
a large loop ground system for taking measurement inside as well as outside the
loop, can be realized only if there would be similar development for forward and
inverse modeling techniques for the in-loop and offset loop systems as for the
central loop and coincident loop systems.
Constable and Srnka 2007; Moreno and Abner 2008; Morgan 2013; Commer et al.
2015; Expsito et al. 2016).
3-D numerical EM modeling involves countless computer storage and time
because the increase in accuracy requires denser discretization which results in a
bigger matrix equation. With advancement in inversion or optimization methods,
the computational requirements for 3-D modeling increased drastically. The prime
objective of numerical inversion is to find the parameters of best fit model which
produces the minimum misfit between the computed and measured data.
Further based on problem faced in optimization of initial model, the inversion
methods are classified as local and global optimization methods, viz. the steepest
descent, conjugate gradient, and linearized least squares inversions are local opti-
mization methods, and the grid search, random search, simulated annealing, and
neural network algorithms are global optimization methods. The global methods
search the parameter space more expansively for the optimal model, and therefore
possibly resolve the true minimum among possible local minimums in a better
sense than the local optimization methods, especially in cases when the number of
optimization parameters are large. The success of local optimization methods
depend greatly on selection of initial model parameters, and local methods are much
faster than global methods. The detailed information on geophysical inversion can
be found in Lines (1988), Oristaglio and Spies (1999), Zhdanov (2010), and Sen
and Stoffa (2013). With all the advancement in modeling and inversion of TEM
sounding data, the 1-D layer earth model still remain important and useful because
they give an idea about the conductivity distribution in the subsurface and are used
as initial model for optimization processes for getting more refined and accurate
subsurface model. Further, there is lack of sophisticated modeling and inversion
techniques for 1-D layered earth models because of the computational intricacies
involved in computation of TEM responses from frequency domain responses
through application of transforms which are very sensitive to frequency domain
calculation to such an extent that an inaccuracy even after 4rth place of decimal
point in frequency computation may lead to entirely different and inaccurate time
domain computation.
The researches related to large loop of finite size in frequency domain can be
found in Patra and Mallick (1980) and Kaufman and Eaton (2001) for the central
loop soundings, in Morrison et al. (1969), Ryu et al. (1970), Poddar (1982, 1983),
and Ward and Hohmann (1988) for induction depth sounding for points outside the
loop, and in Singh and Mogi (2002) for central, in-loop and offset loop induction
sounding, whereas those in time domain can be found in Newman et al. (1987),
Buselli et al. (1990), Hoekstra and Blohm (1990), Nabighian and Macnae (1991),
Hadi and Piket-May (1997), Mitsuhata (2000), Haber et al. (2004, 2007), Streich
(2009), Um et al. (2010), Commer et al. (2015), and Li et al. (2016). There are
number of researches related to use of large loop TEM sounding in groundwater,
environmental, waste management, mineral exploration and geothermal problems
(Weiland 1996; Pellerin et al. 1996; Cohen 2003; Taflove and Hagness 2005;
Avdeev 2005; Constable and Srnka 2007; Moreno and Abner 2008; Morgan 2013;
Commer et al. 2015; Expsito et al. 2016). The TEM sounding data are often
128 S. P. Maurya et al.
Z1
@hz ðtÞ 2
¼ Re½Hz ðx; q; hÞ cosðxtÞdx ð6:4Þ
@t p
0
Z1
@hz ðtÞ 2
¼ Im½Hz ðx; q; hÞ sinðxtÞdx ð6:5Þ
@t p
0
where Re½Hz ðx; q; hÞ and Im½Hz ðx; q; hÞ are the real and imaginary parts of the
vertical magnetic field in frequency domain over a layered earth model. The
component of vectors q and h are the resistivities and thickness of different layers of
layered earth model, and x is the angular frequency.
The expressions of EM field components at a point on or above the surface of an
n-layered earth due to a finite horizontal circular loop of radius a, carrying a current
I eixt and placed at the height z ¼ h above the surface of layered earth in fre-
quency domain is given in Ward and Hohmann (1988). Accordingly, the expression
of Hz field at a measurement point on the surface of n-layered earth (i.e. at z = 0)
can be written as,
Z1
Ia eu0 h ð1 þ rTE Þ k2
Hz ðx; q; hÞ ¼ J1 ðkaÞJ0 ðkrÞdk ð6:6Þ
2 u0
0
6 Forward and Inverse Modeling of Large Loop TEM Data … 129
Y0 b
where rTE ¼ Y1
with Y0 ¼ ixl
u0
(intrinsic admittance of free space)
Y0 þ b
Y1 0
TE
and c
Hy TE
Hx
Y1 ¼ ETE ¼ ETE (surface admittance at z = 0).
x y
For layered earth, these surface admittances are related by the recurrence
relation,
b b
b1 ¼ Y1 Y 2 þ Y1 tanhðu1 h1 Þ ;
Y bn ¼ Yn Y n þ 1 þ Yn tanhðun hn Þ
Y
Y1 þ Yb2 tanhðu1 h1 Þ Yn þ Y bn þ 1 tanhðun hn Þ
bn ¼ Yn with Yn ¼ un ,
and Y ixl n
1=2 1=2
un ¼ kx2 þ ky2 kn2 ¼ k2 kn2 and kn2 ¼ x2 ln en ixln rn
@h
VðtÞ ¼ l0 M ð6:7Þ
@t
where M is the area-turns product of the receiver coil. These voltage data can be
further transformed into the apparent resistivity because sometime it is preferable to
use apparent resistivity transformation to have a direct relation with the
geo-electrical section and having an initial estimate of layer resistivities, which are
often mandatory in non-linear inversion for interpretation of data.
In general, there are two methodologies for solving TEM non-linear inverse
problems. In first, we find the classes of conductivity models that fit the data
without making any assumption about the conductivity distribution in the earth,
while in second approach, we assume an initial model representative of the earth
under consideration. Thereafter, the model parameters are appraised using an
optimization technique. The main aspect of the second approach is the assumption
of correct class of model, that allows the incorporation of geological and geo-
physical information into the inverse problems. The major disadvantage of this
130 S. P. Maurya et al.
1
VðFÞ ¼ ½VðRNORMÞ2 ð6:8Þ
2
1 X n
VðRNORMÞ ¼ pffiffiffi RðIÞ ð6:9Þ
n i¼1
where YðIÞ and WTðIÞ are the I-th data point and corresponding weight factor, and
F is corresponding calculated value.
Using the mathematical steps described in above forward and inverse modeling
procedure, an inversion algorithm is generated and computations for forward and
inversion results for voltage response data for the homogeneous, 2-layer, 3-layer
and 4-layer resistive and conductive models are performed for central loop, in-loop
and offset loop systems respectively and are presented in the next section.
6 Forward and Inverse Modeling of Large Loop TEM Data … 131
The plan view of large loop TEM method with central loop, in-loop and offset loop
configurations over a homogeneous, 2 layer, 3 layer and 4 layer earth model is as in
Fig. 6.3. The large loop presents a source loop and small loop at center of the
source loop (X1), represents receiver position corresponding to the central loop
configuration (Abubakar and vanden Berg 2004). Similarly, the large loop presents
a source loop and small loop at arbitrary position inside the source loop (X2),
represents receiver position corresponding to the in-loop configuration and small
loop at arbitrary point outside the source loop (X3) represents receiver position
corresponding to the arbitrary offset loop configuration. Figure 6.3a depicts
homogeneous earth model with conductivity of 0.01 S/m whereas Fig. 6.3b depicts
two layer earth model with conductivity 0.01 and 0.1 S/m for first and second layer,
respectively. Figure 6.3c demonstrate three layer earth model with conductivity
0.01, 0.03 and 0.05 S/m for first, second and third layer, respectively whereas
Fig. 6.3d demonstrate four layer earth model of conductivity 0.01, 0.03, 0.05 and
0.09 S/m for first, second, third and fourth layers, respectively. These models are
used for testing the forward and inverse algorithm in this study.
For checking the reliability and accuracy of the forward computation method, we
applied it for computation of TEM (impulse) response of a large circular loop
source over the surface of a homogeneous earth model for central loop configu-
rations, and the results are compared with the TEM responses of a large loop source
generated using the central loop analytical expression for impulse response (Ward
and Hohman 1988; Singh et al. 2009). The analytic expression for impulse response
of the hz field at the center of a large loop source of radius (a) over the homoge-
neous earth model of conductivity, r can be written as (Ward and Hohman 1988;
Singh et al. 2009)
@hzjStep
hzjImpulse ¼
@t ð6:12Þ
I 2
2 2 h2 a2
¼ 3erf ðhaÞ 1 3 þ 2h a e
l0 ra3 p2
1=2
where h ¼ l4t0 r , erf is error function and t is delay time. This impulse response
of vertical magnetic field is related with the voltage response in accordance with
Eq. 6.7. Figure 6.4 shows comparison of voltage response obtained by the present
study forward computational algorithm, Ward and Hohmann (1988) and Singh
et al. (2009) for central loop configuration over homogeneous earth model. The
figure shows that all the curves follow each other very well. After getting satis-
factory results for homogeneous earth model, the program is applied to the
multi-layer earth models for checking efficiency and accuracy of the program.
132 S. P. Maurya et al.
Fig. 6.3 Schematic diagram depicting all the possible configurations due to large loop TEM
source and models for which analyses are performed
6 Forward and Inverse Modeling of Large Loop TEM Data … 133
Fig. 6.4 Comparison of results computed using present method, Ward and Hohmann (1988), and
Singh et al. (2009) for vertical magnetic field at the center of large circular loop of radius (50 m),
over the homogeneous earth model of conductivity of 0.01 S/m
Figure 6.5 presents TEM responses of a large loop source over multi-layer earth
models (as shown in Fig. 6.3b) for central loop configurations for source loop
radius 50 m. Figure 6.5a depicts TEM response for homogeneous earth model with
conductivity 0.01 S/m whereas Fig. 6.5b–d depicts TEM response for the Two
layer, three layer and four layer earth model respectively for central loop config-
uration. The computed results depict characteristic features of TEM response over
the multi-layer earth models.
Figure 6.6 present TEM responses of a large loop sources over multi-layer earth
model (as shown in Fig. 6.3c) for In-loop configurations for source loop radius
50 m. Figure 6.6a depicts TEM response for homogeneous earth model with
conductivity 0.01 S/m whereas Fig. 6.6b–d depicts TEM response for the Two
layer, three layer and four layer earth models, respectively for in-loop configuration.
The computed results depict characteristic features of TEM response over the
multi-layer models. For all layer, the response varies in similar way with time and
depict the capability of developed techniques for generating TEM response data
over layered earth model.
Figure 6.7 present TEM responses of a large loop sources over the multi-layer
earth model (as shown in Fig. 6.3d) for offset loop configurations for source loop
radius 50 m. Figure 6.7a depicts TEM response for homogeneous earth model with
conductivity 0.01 S/m whereas Fig. 6.7b–d depicts TEM response for the Two
layer, three layer and four layer earth models, respectively for offset loop config-
uration. The computed results illustrate the distinctive features of TEM response
over the multi-layer earth models. For all the layer earth models, the offset loop
134 S. P. Maurya et al.
Fig. 6.5 Transient voltage response of a large circular loop source of radii 50 m for central loop
configuration over the a homogeneous, b 2-layer, c 3-layer, and d 4-layer earth model
response varies in similar way with time and depicts the capability of developed
techniques to generate the offset loop TEM response over layered earth model.
From these figures (Figs. 6.5, 6.6 and 6.7), it is clear that the TEM response
show smooth and well defined characteristic. Moreover, it is also noticed that the
central loop TEM responses are more regular as compared to the offset loop and
arbitrary in-loop responses.
This section contains the application of the developed algorithm for interpretation
of large loop central loop, in-loop and offset loop TEM data over the layered earth
model.
To demonstrate the efficacy of the program, the program is applied for inverting
the central loop, in-loop and offset loop TEM data over the homogeneous as well as
multi-layer earth models, namely two, three and four layer models comprising of
conductivity and thickness of the various layers, for loop sizes of radius 50 m with
impulse source excitation. The inversion results are shown for noise free as well as
6 Forward and Inverse Modeling of Large Loop TEM Data … 135
Fig. 6.6 In-loop transient voltage response due to large circular loop source of radii 50 m over the
a homogeneous, b 2-layer, c 3-layer, and d 4-layer earth model
noisy large loop TEM data with 5% random noises, over 2-layer, 3-layer and
4-layer earth models for all the possible configurations. For all the inverted results,
the inversion starts with an initial homogeneous earth model of conductivity
0.001 S/m.
The analysis is performed for the inversion of large loop sounding data and the
results are presented in sequence of inversion results for homogeneous, two layer,
three layer and four layer earth models for the central loop TEM sounding noise
free as well as noisy data, for the in-loop TEM sounding noise free as well as noisy
data, and for the offset loop TEM sounding noise free as well as noisy data. The
analysis is carried out for noise free and noisy data with addition of 1, 3, 5 and 7%
random noise in the data but the results are illustrated only for noise free and 5%
noisy data for simplicity.
Figure 6.8 presents inversion results for the finite loop noise free TEM data over
the homogeneous earth model for central loop, in-loop and offset loop system for
with source loop of radius 50 m. Figure 6.8a, c, e shows voltage response whereas
136 S. P. Maurya et al.
Fig. 6.7 Transient voltage response due to large circular loop source of radii 50 m over the
a homogeneous, b 2-layer, c 3-layer, and d 4-layer earth model for the offset loop configuration
Fig. 6.8b, d, f show conductivity model for central loop, In-loop and offset loop
configurations, respectively. The inversion was stared with an initial homogeneous
model of conductivity 0.001 S/m for noise free, 1% random noise, 3% random
noise and 5% random noise data, however, the results are presented only for noise
free data. From Fig. 6.8a, c, e, it is observed that there is good match between the
original and inverted voltage response data. Figure 6.8b, d, f depict that inverted
model are in good agreement with the original model with which data were gen-
erated. The conductivity of homogeneous layer is reproduced with difference of as
little as 0.1%.
Figure 6.9 shows inversion results for the voltage response data over the 2-layer
earth model for central loop system and source loop of radius 50 m. Figure 6.9a, c
show original and inverted voltage response curve, while Fig. 6.9b, d shows
original and inverted 2-layer conductivity models for noise free and 5% random
noises in the data respectively. The inversion was started with a homogeneous
model of conductivity 0.001 S/m. Figure 6.9a, c depict that there is good agreement
6 Forward and Inverse Modeling of Large Loop TEM Data … 137
Fig. 6.8 Inversion results for TEM data over homogeneous earth model for the central loop
configuration, a the synthetic and inverted best fit curve and b the original and inverted
conductivity models; inversion results for in-loop configuration, c the synthetic and inverted best
fit curve and d the original and inverted conductivity models; inversion results for the offset loop
configuration, e the synthetic and inverted best fit curves and f the original and inverted
conductivity models
between the observed and inverted voltage response data whereas from Fig. 6.9b, d,
it is observed that the inverted models are in agreement with the original synthetic
model with which data were generated even after addition of 5% random noises in
the data. The program is also tested for more than 5% random noises, however, the
inverted results start deviating from real model significantly.
Figure 6.10 depicts inversion results for large loop TEM voltage response data
over the 2-layer earth model for the in-loop configuration for source loop of radius
138 S. P. Maurya et al.
Fig. 6.9 Inversion results for central loop TEM data over two layer earth models, a the synthetic
and inverted best fit curve and b the original and inverted conductivity models for noise free data;
c the synthetic and inverted best fit curve and d the original and inverted conductivity models for
5% noise data
50 m. Figure 6.10a, c show original and inverted voltage response, and Fig. 6.10b,
d show original and inverted 2-layer earth conductive earth models for noise free
and 5% random noise data respectively. The inversion was started with a homo-
geneous model of conductivity 0.001 S/m. From Fig. 6.10a, c, it is clear that there
is good agreement between the observed and inverted voltage response data. From
Fig. 6.10b, d, it is clear that the inverted model is in agreement with the original
synthetic model even after adding 5% random noise in the data. The inverted
parameters are recovered with significantly high accuracy, i.e. conductivities and
thickness of different layers interpreted from inversion indicates that there is an
average variation of less than 0.5%.
Figure 6.11 shows inversion results for large loop TEM voltage response data
over the 2-layer earth model for the offset loop configuration. The source loop
radius is 50 m. Figure 6.11a, c show the original and the inverted voltage response
curves for noise free and 5% random noise data, respectively, and depicts that there
is good matching between the original and inverted curves. Figure 6.11b, d show
original and inverted 2-layer conductivity models for noise free and 5% random
noise in the data respectively. The inversion was started with a homogeneous model
of conductivity 0.001 S/m. Figure 6.11b, d depicts that there is good agreement
6 Forward and Inverse Modeling of Large Loop TEM Data … 139
Fig. 6.10 Inversion results for TEM data over two layer earth model for in-loop configuration,
a the synthetic and inverted best fit curves and b the original and inverted conductivity models for
noise free data; c the synthetic and inverted best fit curves and d the original and inverted
conductivity models for 5% noise data
between the inverted and original conductivity models with which data were
generated. The inverted parameters, i.e. conductivities and thickness of different
layers are close to the original parameters. The average variation of inverted
parameters is less than 1%. These results depict accuracy and capability of the
method for inversion of large loop TEM data acquired using offset loop configu-
ration. Further, the inversion was carried out for more than 5% Gaussian noise in
the data and it is noticed that the inversion algorithm does not work properly and
gives ambiguous results.
Thereafter, the algorithm is further applied for inverting the voltage time data over a
3 layer earth model with addition of 1, 3, 5% and more than 5% random noises.
However, the results up to 5% random noise are shown. Figure 6.12 shows
inversion results for the large loop central loop TEM voltage response data over the
3-layer model for loop radius of 50 m. Figure 6.12a, c show original and inverted
voltage response curves for noise free and 5% random noise in the data respectively
140 S. P. Maurya et al.
Fig. 6.11 Inversion results for TEM data over two layer earth model for offset loop configuration,
a the synthetic and inverted best fit curves and b the original and inverted conductivity models for
noise free data; c the synthetic and inverted best fit curves and d the original and inverted
conductivity models for 5% noise data
and depict that there is good matching between the data points and inverted curves.
Figure 6.12b, d show original and inverted 3-layer conductivity models. The
inversion was started with a homogeneous model of conductivity 0.001 S/m.
Figure 6.12b, d depict that the inverted models match significantly with the original
conductivity model with which data was generated. The inverted parameters, i.e.
conductivities and thicknesses of different layers are close to the original parame-
ters. The average variation of inverted parameters is less than 3%. These results
depict accuracy and capability of the method for inversion of large loop central loop
TEM data over 3 layer earth models.
Figure 6.13 shows inversion results for the large loop TEM voltage response
data over 3-layer earth model for in-loop configuration for source loop radius 50 m.
Figure 6.13a, c show original and inverted voltage response curves for noise free
and 5% random noise data respectively, which depict that there is good match
between the original and inverted curves. Figure 6.13b, d show original and
inverted 3-layer conductivity models for noise free and 5% random noise data
respectively. The inversion was started with a homogeneous model of conductivity
0.001 S/m. Figure 6.13b, d depict good agreement between the inverted and
original model with which data was generated. The inverted parameters, i.e.
6 Forward and Inverse Modeling of Large Loop TEM Data … 141
Fig. 6.12 Inversion results for TEM data over three layer earth model for central loop
configuration, a the synthetic and inverted best fit curves and b the original and inverted
conductivity models for noise free data; c the synthetic and inverted best fit curves and d the
original and inverted conductivity models for 5% noise data
conductivities and thicknesses of different layers are close to the original parame-
ters. The average variation of inverted parameters is less than 4%. These results
depict accuracy and capability of the method for inversion of large loop TEM data
acquired using in-loop configuration over 3 layer earth models. Further, the
inversion is carried out for the more than 5% noise in the data and it is noticed that
the inversion algorithm is not working properly and giving ambiguous results.
These results depict the capability of the program for interpreting large loop in-loop
TEM data acquired over 3 layer earth models.
Figure 6.14 depicts inversion results for the large loop TEM voltage response
data over 3-layer earth model for offset loop configuration for source loop radius
50 m. Figure 6.14a, c show the original and the inverted voltage response curves
for noise free and 5% random noise data respectively, and show that there is good
matching between the original and inverted curves. Figure 6.14b, d show original
and inverted 3-layer conductivity models for noise free and 5% random noise data
respectively. The inversion was started with a homogeneous model of conductivity
0.001 S/m. Figure 6.14b, d depict that the inverted model is in well agreement with
the original conductivity model with which data was generated. The inverted
parameters, i.e. conductivities and thickness of different layers are close to the
142 S. P. Maurya et al.
Fig. 6.13 Inversion results for TEM data over three layer earth model for in-loop configuration,
a the synthetic and inverted best fit curves and b the original and inverted conductivity models for
noise free data; c the synthetic and inverted best fit curves and d the original and inverted
conductivity models for 5% noise data
original parameters. The average variation of inverted parameters is less than 4%.
These results depict accuracy and capability of the method for inversion of large
loop TEM data acquired using offset loop configuration. Further, the inversion was
carried out for more than 5% Gaussian noise in the data and it is noticed that the
inversion algorithm does not work properly and gives ambiguous results. These
results indicate the capability of the program to invert large offset loop TEM data
over 3 layer earth models.
Figure 6.15 depict inversion results for the large loop TEM voltage response data
over the 4-layer earth model for central loop configuration and the source loop
radius 50 m. Figure 6.15a, c show original and inverted voltage response curves for
noise free and 5% random noise data respectively. It is observed that there is good
match between the original and inverted curves. Figure 6.15b, d show original and
inverted 4-layer conductivity models for noise free and 5% random noise data
respectively. The inversion was started with a homogeneous model of conductivity
6 Forward and Inverse Modeling of Large Loop TEM Data … 143
Fig. 6.14 Inversion results for TEM data over three layer earth model for offset loop
configuration, a the synthetic and inverted best fit curves and b the original and inverted
conductivity models for noise free data; c the synthetic and inverted best fit curves and d the
original and inverted conductivity models for 5% noise data
0.001 S/m. Figure 6.15b, d depict that there is good matching between the inverted
and original model with which data was generated. The inverted parameters, i.e.
conductivities and thicknesses of different layers are close to the original parame-
ters. The average variation of inverted parameters is less than 4%. These results
depict accuracy and capability of the method for inversion of large loop TEM data
acquired using central configuration over the 4 layer earth models. Further, the
inversion is carried out for more than 5% Gaussian noise in the data and it is noticed
that the inversion algorithm is not working properly and giving ambiguous results.
These results depict the capability of the program for interpreting large loop in-loop
TEM data acquired over 4 layer earth models.
Figure 6.16 depict inversion results for the large loop TEM voltage response
data over 4-layer earth model for in-loop configuration for source loop of radius
50 m. Figure 6.16a, c show original and inverted voltage response curves for noise
free and 5% random noise data respectively. It is noticed that there is good match
between the original and inverted curves. Figure 6.16b, d show original and
inverted 4-layer conductivity models for noise free and 5% random noise data
respectively. The inversion was started with a homogeneous model of conductivity
0.001 S/m. Figure 6.16b, d depict that there is good agreement between the
144 S. P. Maurya et al.
Fig. 6.15 Inversion results for TEM data over four layer earth model for central loop
configuration, a the synthetic and inverted best fit curves and b the original and inverted
conductivity models for noise free data; c the synthetic and inverted best fit curves and d the
original and inverted conductivity models for 5% noise data
inverted and original model with which data was generated. The inverted param-
eters, i.e. conductivities and thicknesses of different layers are close to the original
parameters. The average variation of inverted parameters is less than 5%. These
results depict accuracy and capability of the method for inversion of large loop
TEM data acquired using In-loop configuration over the 4 layer earth models.
Further, the inversion is carried out for more than 5% Gaussian noise in the data and
it is noticed that the inversion algorithm is not working properly and giving
ambiguous results. These results depict the capability of the program for inter-
preting large loop in-loop TEM data acquired over 4 layer earth models.
Figure 6.17 depict inversion results for the large loop TEM voltage response
data over the 4-layer earth model for offset loop configuration for source loop radius
50 m. Figure 6.17a, c show original and inverted voltage response curves for noise
free and 5% random noise data respectively. It is noticed that there is good match
between the original and inverted curves. Figure 6.17b, d shows original and
inverted 4-layer conductivity models for noise free and 5% random noise data
respectively. The inversion was started with a homogeneous model of conductivity
0.001 S/m. Figure 6.17b, d depict that there is good agreement between the
inverted and original model with which data was generated. The inverted
6 Forward and Inverse Modeling of Large Loop TEM Data … 145
Fig. 6.16 Inversion results for TEM data over four layer earth model for in-loop configuration,
a the synthetic and inverted best fit curves and b the original and inverted conductivity models for
noise free data; c the synthetic and inverted best fit curves and d the original and inverted
conductivity models for 5% noise data
parameters, i.e. conductivities and thicknesses of different layers are close to the
original parameters. The average variation of inverted parameters is less than 5%.
These results depict accuracy and capability of the method for inversion of large
loop TEM data acquired using In-loop configuration over 4 layer earth models.
Further, the inversion is carried out for more than 5% Gaussian noise in the data and
it is noticed that the inversion algorithm is not working properly and giving
ambiguous results. These results depict the capability of the program for inter-
preting large loop offset loop TEM data acquired over 4 layer earth models.
Figure 6.18 presents variation of RMS (residual error) against iteration number for
the inverted results of central loop system over homogeneous earth model for
source loop of radius 50 m. Figure 6.18a shows RMS error for noise free data and
Fig. 6.18b shows variation of RMS error with iteration number for 5% random
noise in the data. From these Figure, it is observed that the noise free data converges
146 S. P. Maurya et al.
Fig. 6.17 Inversion results for TEM data over four layer earth model for offset loop configuration,
a the synthetic and inverted best fit curves and b the original and inverted conductivity models for
noise free data; c the synthetic and inverted best fit curves and d the original and inverted
conductivity models for 5% noise data
Fig. 6.18 Plot of residual error/RMS error versus iteration number for central loop TEM data over
homogeneous earth model. a For noise free data and b for 5% random noise data
rapidly as compared to the noisy data. Similar results are also found for the in-loop
and offset loop systems over 2-layer, 3-layer and 4-layer earth models but they are
not discussed here for simplicity.
6 Forward and Inverse Modeling of Large Loop TEM Data … 147
From the results presented in Figs. 6.8, 6.9, 6.10, 6.11, 6.12, 6.13, 6.14, 6.15,
6.16 and 6.17, it is observed that the program is capable of interpreting large loop
TEM data acquired using all the possible configurations over the homogeneous,
2-layer, 3-layer and 4-layer earth models and therefore have possibility of its
successful application for inversion of large loop TEM data over multi-layer earth
models.
6.4 Conclusions
This research embodies the results on forward modeling and inversion of large loop
TEM sounding data over the layered earth models acquired using central loop,
arbitrary in-loop and offset loop configurations for different source loop sizes with
the objective of presenting a forward and an inversion schemes suitable for inter-
pretation of real field large loop TEM sounding data. In this work, a forward
computation scheme based on Fourier cosine and sine transform of frequency
domain solutions computed using Singh and Mogi (2003) and an inversion scheme
based on minimization of residual misfit function are developed and applied for the
modeling and inversion of large loop transient electromagnetic (TEM) data
acquired using central loop, in-loop and offset-loop configurations over the layer
earth models. The developed inversion scheme is based on the modification of
NLSTCI program (Anderson 1982) in accordance with the need of the problem.
Accordingly the forward computation scheme in NLSTCI has been changed by the
EMLCLLER (Singh and Mogi 2003, 2005, 2010), for computation of frequency
domain EM response of large loop sources over the layer earth models. Thereafter
these frequency domain responses are converted into the TEM responses over
layered earth with the use of Fourier cosine and sine transforms.
Both the programs are tested for synthetic models and data generated using
central loop, in-loop and offset loop configurations over homogeneous (conduc-
tivity 0.01 S/m), two layer (conductivity 0.01 and 0.1 S/m), three layer (conduc-
tivity 0.001, 0.1 and 0.01 S/m) and four layer (conductivity 0.01, 0.03, 0.05,
0.09 S/m) earth models with and without the addition of random noises. The the-
oretical examples illustrating the accuracy and efficacy of forward and inversion
programs for inverting the large loop TEM data with or without random noise
demonstrate the potential of the programs for its further application for interpre-
tation of real field data. The scheme work satisfactorily and produces reliable
forward and inversion results for up to four layer earth model and addition of up to
5% random noise in original data. The inversion program works satisfactorily for
central loop, in-loop and offset loop TEM sounding data due to a large loop source
even for data with 5% random noises and hence have possibility of its successful
application for real field data. This single program is capable of forward compu-
tation and inverting large loop TEM data due to all the configurations of a large
148 S. P. Maurya et al.
loop; central, in-loop and offset loop configurations. The programs in their present
forms are developed for voltage response (voltage-time decay curve), but have the
option for forward computation and inversion of apparent resistivity TEM data with
some modification in constraints and the input parameters in the program.
References
Abubakar A, vanden Berg PM (2004) Iterative forward and inverse algorithms based on domain
integral equations for three-dimensional electric and magnetic objects. J Comput Phys 195
(1):236–262
Alumbaugh DL, Newman GA (1997) Three-dimensional massively parallel electromagnetic
inversion—II. Analysis of a crosswell electromagnetic experiment. Geophys J Int 128:355363.
https://doi.org/10.1111/j.1365-246X.1997.tb01560.x
Aminzadeh F (2013) Geophysical applications of artificial neural networks and fuzzy logic, vol 21.
Springer Science & Business Media
Anderson WL (1979) Numerical integration of related Hankel transform of order 0 and 1 by
adaptive digital filtering. Geophysics 44:1287–1305
Anderson WL (1982) Nonlinear least square inversion of transient soundings for a central
induction loop system (Program NLSTCI), U.S. Geological Survey Open File. Report, pp 82–
1129
Annan AP (1974) The equivalent source method for electromagnetic scattering analysis and its
geophysical application. Doctoral dissertation, Memorial University of New found land)
Auken E, Christiansen AV, Kirkegaard C, Fiandaca G, Schamper C, Behroozmand AA, Binley A,
Nielsen E, Effers F, Christensen NB, Srensen K (2014) An overview of a highly versatile
forward and stable inverse algorithm for airborne, ground-based and borehole electromagnetic
and electric data. Explor Geophys 46(3):223–235
Avdeev DB (2005) Three-dimensional electromagnetic modelling and inversion from theory to
application. Surv Geophys 26(6):767–799
Baumgartner F (1996) A new method for geoelectrical investigations underwater. Geophys
Prospect 44(1):71–98
Buselli G, O’Neill B (1977) SIROTEM: a new portable instrument for multichannel transient
electromagnetic measurements. Explor Geophys 8(3):82–87
Buselli G, Barber C, Davis GB, Salama RB (1990) Detection of groundwater contamination near
waste disposal sites with transient electromagnetic and electrical methods. Geotech Environ
Geophys 5:27–39
Butler DK (2005) Electromagnetic induction methods for environmental problems. In:
Near-surface geophysics, pp 1–6
Chave AD, Jones AG (2012) The magnetotelluric method: theory and practice. Cambridge
University Press
Christensen JH (1997) The Danish Eulerian hemispheric model—a three-dimensional air pollution
model used for the Arctic. Atmos Environ 31(24):4169–4191
Christensen NB (2014) Sensitivity functions of transient electromagnetic methods. Geophysics 79
(4):E167–E182
Christensen NB, Srensen KI (1998) Surface and borehole electric and electromagnetic methods for
hydrogeological investigations. Eur J Environ Eng Geophys 3(1):75–90
Christiansen AV, Christensen NB (2003) A quantitative appraisal of airborne and ground-based
transient electromagnetic (TEM) measurements in Denmark. Geophysics 68(2):523–534
Christiansen AV, Auken E, Kirkegaard C, Schamper C, Vignoli G (2016) An efficient hybrid
scheme for fast and accurate inversion of airborne transient electromagnetic data. Explor
Geophys 47(4):323–330
6 Forward and Inverse Modeling of Large Loop TEM Data … 149
Strack KM (1992) Exploration with deep transient electromagnetics, vol 373. Elsevier, Amsterdam
Strack KM, Vozoff K (1996) Integrating long offset transient electromagnetics (LOTEM) with
seismics in an exploration environment. Geophys Prospect 44(6):997–1017
Strack KM, Hanstein T, LeBrocq K, Moss DC, Vozoff K, Wolfgram PA (1989) Case histories of
LOTEM surveys in hydrocarbon prospective area. First Break 7(12):467–477
Strack KM, Lschen E, Ktz AW (1990) Long-offset transient electromagnetic (LOTEM) depth
soundings applied to crustal studies in the Black Forest and Swabian Alb, Federal Republic of
Germany. Geophysics 55(7):834–842
Streich R (2009) 3D finite-difference frequency-domain modeling of controlled-source electro-
magnetic data: Direct solution and optimization for high accuracy. Geophysics 74(5):F95–
F105
Streich R (2016) Controlled-source electromagnetic approaches for hydrocarbon exploration and
monitoring on land. Surv Geophys 37(1):47–80
Taflove A, Hagness SC (2005) Computational electrodynamics: the finite-difference time domain
method. Artech house
Teixeira FL, Chew WC, Straka M, Oristaglio ML, Wang T (1998) Finite-difference time-domain
simulation of ground penetrating radar on dispersive, inhomogeneous, and conductive soils.
IEEE Trans Geosci Remote Sens 36(6):1928–1937
Torres-Verdin C, Habashy TM (1994) Rapid 2.5-dimensional forward modeling and inversion via
a new nonlinear scattering approximation. Radio Sci 29(04):1051–1079
Um ES, Harris JM, Alumbaugh DL (2010) 3D time-domain simulation of electromagnetic
diffusion phenomena: A finite-element electric-field approach. Geophysics 75(4):F115–F126
Vozoff K (1986) Magnetotelluric methods. Soc Expl Geophys Reprint Ser (5)
Wang T, Hohmann GW (1993) A finite-difference, time-domain solution for three dimensional
electromagnetic modeling. Geophysics 58(6):797–809
Wannamaker PE, Hohmann GW, SanFilipo WA (1984) Electromagnetic modeling of
three-dimensional bodies in layered earths using integral equations. Geophysics 49(1):60–74
Ward SH (1980) Electrical, electromagnetic, and magnetotelluric methods. Geophysics 45
(11):1659–1666
Ward SH, Hohmann GW (1988) Electromagnetic theory for geophysical applications.
Electromagn Methods Appl Geophys (1)
Weaver PF, Yuen PC, Prolss GW, Furumoto AS (1970) Acoustic coupling into the ionosphere
from seismic waves of the earthquake at Kurile Islands on August 11, 1969. Nature 226
(5252):1239
Weidelt P (1975) Electromagnetic induction in three-dimensional structures: J Geophys 41(1):85–
109
Weidelt P (1975b) Inversion of two-dimensional conductivity structures. Phys Earth Planet Inter
10(3):282–291
Weiland T (1996) Time domain electromagnetic field computation with finite difference methods.
Int J Numer Model Electron Netw Devices Fields 9(4):295–319
Wright DA, Ziolkowski AM, Hobbs BA (2005) Detection of subsurface resistivity contrasts with
application to location of fluids. U.S. Patent 6,914,433
Xiong Z (1992) Electromagnetic modeling of 3-D structures by the method of system iteration
using integral equations. Geophysics 57(12):1556–1561
Xue Y, Chang L, Kjaer SB, Bordonau J, Shimizu T (2004) Topologies of single-phase inverters
for small distributed power generators: an overview. IEEE Trans Power Electron 19(5):1305–
1314
Zhang Z, Xiao J (2001) Inversion of surface and borehole data from large-loop transient
electromagnetic system over a 1-D earth. Geophysics 66:10901096
Zhdanov MS (2009a) Geophysical electromagnetic theory and methods, vol 43. Elsevier
Zhdanov MS (2009b) New advances in regularized inversion of gravity and electromagnetic data.
Geophys Prospect 57(4):463–478
Zhdanov MS (2010) Electromagnetic geophysics: notes from the past and the road ahead.
Geophysics 75(5):75A49–75A66
6 Forward and Inverse Modeling of Large Loop TEM Data … 153
Abstract In this study, global optimizations of the data sets obtained from
potential field method applications including self-potential (SP), magnetic and
gravity have been presented through particle swarm optimization (PSO), differential
evolution algorithm (DEA) and differential search algorithm (DSA), respectively.
Both synthetically produced and real field anomalies due to various kinds of geo-
logical sources (e.g. sheet, horizontal cylinder, sphere and fault) have been used to
show the capability of those population-based metaheuristic algorithms. Firstly,
error energy maps have been produced for each model parameter pairs in every
synthetic example to reveal the mathematical nature of the inverse problems under
consideration. These maps have clearly helped us assess the resolvability of the
model parameters for the given problem. Metropolis–Hastings (M–H) sampling
algorithm has been used to perform uncertainty analyses. Produced histograms of
synthetic data cases have provide insight to the reliability of the estimated
parameters. Additionally, the reliabilities of the solutions have been also tested via
probability density function (PDF) applications. After performing successful syn-
thetic studies on model parameter estimations, three real data examples including an
SP anomaly from a sulphide mineralization zone (India), a total field magnetic
anomaly from a fold belt (Australia) and a gravity anomaly from a chromite deposit
(Cuba) have been inverted through PSO, DEA and DSA, respectively. Satisfactory
solutions in well agreement with the results of previous studies have been obtained.
Y. L. Ekinci (&)
Department of Archaeology, Bitlis Eren University, 13100 Bitlis, Turkey
e-mail: ylekinci@beu.edu.tr
Career Application and Research Center, Bitlis Eren University, 13100 Bitlis, Turkey
Ç. Balkaya
Department of Geophysical Engineering, Süleyman Demirel University, 32260 Isparta,
Turkey
e-mail: caglayanbalkaya@sdu.edu.tr
G. Göktürkler
Department of Geophysical Engineering, Dokuz Eylül University, 35390 İzmir, Turkey
e-mail: gokhan.gokturkler@deu.edu.tr
Keywords Potential field data Particle swarm optimization Differential
evolution algorithm Differential search algorithm Metaheuristics
Uncertainty analyses
7.1 Introduction
As it is well known one of the most important aims of the geophysical applications
is to explore and determine the causative geological sources by means of measured
geophysical data sets (Grasmueck et al. 2004; Sari and Şalk 2006; Rucker et al.
2009; Bekler et al. 2011; Balkaya et al. 2012; Ekinci and Yiğitbaş 2012, 2015;
Erginal et al. 2013; Shan et al. 2014; Kosaroglu et al. 2016; Oruç et al. 2017; Lang
et al. 2017). A function in a domain of definition can be used to make sense out of
these geophysical responses. Optimization procedure clearly plays a key role in
geophysics like in most of the other natural and applied sciences by trying to find
the best value of this function. However, nonuniform and complex nature of the
earth’s inside makes the problem more difficult. Besides, due to the existence of
noise content and an insufficient number of observation points in most cases,
optimization of the geophysical data set is considerably complicated, and therefore,
the solutions are typically ambiguous and error-prone (Meju 1994; Menke 1989;
Zhdanov 2002; Tarantola 2005). In addition to those mentioned difficulties,
ill-posed nature of the inverse modelling problems give rise to non-unique solu-
tions, and therefore, possible minor changes in the measured data may give rise to
remarkable perturbations in the solutions that strongly affect the stability of the
solution unfavourably (Roy 2002; Zhdanov 2002; Ekinci and Demirci 2008; Haber
et al. 2008; Gholami and Siahkoohi 2010). This challenging non-uniqueness
problem may be resolved using geophysical or geological a priori information and
foreknown constraints on the variable values (Li and Oldenburg 1996, 2003;
Portniaguine and Zhdanov 2002; Čuma et al. 2012; Bektaş et al. 2013; Cai and
Zhdanov 2015; Balkaya et al. 2017).
Converging to an optimal mathematical solution including the model parameters
is the most significant task in the inverse modelling studies. The error/objective
function has a complex nature and it is the measure of mismatch between the
measured and the estimated data. This function is mostly characterized by multiple
hills and valleys (maxima and minima) like shown in Fig. 7.1. Global minimum is
the minimum of all the valleys and is unique. Local minima involve all other
valleys (Sen and Stoffa 1995). Geophysical inverse modelling problems are
addressed using either local or global optimization methods to reach to the global
minimum of all possible points in the model space (Zhdanov 2002; Gallardo and
Meju 2004; Tarantola 2005; Fernández-Martínez et al. 2010; Mehanee et al. 2011;
7 Global Optimization of Near-Surface Potential … 157
Global maximum
Local maximum
Local
minimum
F
X1
X2
Global minimum
Ulugergerli 2011; Biswas and Sharma 2014; Mehanee 2014; Mehanee and Essa
2015; Ekinci et al. 2017; Alkan and Balkaya 2018). Although traditional
gradient-based local-search optimization methods have a fast convergence char-
acteristic, their achievement strongly requires a well-established initial guess
(starting model), namely one initial candidate solution (Menke 1989; Maiti et al.
2011; Ogunbo 2018). They characteristically try to reach a local minimum in the
close neighbourhood of the starting model without guaranteeing the global mini-
mum (Menke 1989; Chunduru et al. 1997; Başokur et al. 2007; Ekinci and Demirci
2008; Göktürkler 2011; Balkaya et al. 2017). Thus, using a good starting model is
crucial to abstain from a local minimum (Ekinci et al. 2016). It must be noted that
the findings obtained from the previous geological and geophysical investigations,
drilling data and some constraints about the subsurface structure may help to choose
a good initial guess (Zhdanov 2002). However, in most of the inverse problems the
error functions have very complex nature as explained by Vasant (2012). These
types of problems are quite difficult to solve, and therefore, global optimization
algorithms which have been gaining more attention in the recent decades can be
performed to obtain some approximate solutions (Vasant 2012). Additionally, the
success of the global-search algorithms is not dependent on the goodness of the
initial guess (Göktürkler and Balkaya 2012; Balkaya et al. 2017; Ekinci et al. 2017).
Since these algorithms try to find the optimum solution among all possible solutions
that exist in model space, satisfactory solutions may be obtained by using an
efficient global optimization algorithm even starting with a bad initial guess (Sen
and Stoffa 1995). Moreover, using a stochastic search type in the model space, these
algorithms can avoid local minima without requiring a well-constructed initial
model. These special measures which provide a robust search process (Ekinci et al.
2016) mostly increase the chance of reaching a global minimum (Weise 2009).
More importantly, these methods try to reach a global minimum of the error
158 Y. L. Ekinci et al.
Table 7.1 Some of the global optimization applications for the potential field data inversions
Researcher(s) Metaheuristics Geophysical
method
ACO DEA GA GPA PSO SA
Abdelazeem and SP
Gobashy (2006),
Sweilam et al. (2007)
Fernández-Martínez
et al. (2010)
Santos (2010)
Pekşen et al. (2011)
Göktürkler and Balkaya
(2012)
Sharma and Biswas
(2013)
Biswas and Sharma
(2014)
Di Maio et al. (2016)
Biswas (2017)
Toushmalani (2013) Gravity
Liu et al. (2014)
Biswas (2015)
Pallero et al. (2015)
Pallero et al. (2017)
Ekinci et al. (2016)
Biswas (2016)
Singh and Biswas
(2016)
Biswas et al. (2017)
Srivastava and Agarwal Magnetic
(2010)
Liu et al. (2014)
Liu et al. (2015)
Ekinci (2016)
Biswas (2016)
Biswas and Acharya
(2016)
Ekinci et al. (2017)
Balkaya et al. (2017)
Biswas et al. (2017)
Kaftan (2017)
Essa and Elhussein
(2018)
7 Global Optimization of Near-Surface Potential … 159
function. The stochastic global optimization algorithms update their current posi-
tion by using more global details about the nature of the misfit plane (Sen and Stoffa
1995). Another priority of these algorithms is that they use wider search bounds
involving the model parameters (Fernández-Martínez et al. 2010). On the other
hand, the weakness of those global optimization algorithms is their intense com-
putational nature because of the huge amount of error function evaluations and also
the challenging forward modelling equations (Ekinci et al. 2016; Balkaya et al.
2017). However, they may be a right choice when the noise content is high and in
the lack of a geophysical or geological priori information when the success is
greatly initial guess-dependent (Fernández-Martínez et al. 2010). Nowadays, the
mentioned-above weakness of those algorithms started to lose its significance due
to the advancements of fast and capable computer systems. Hence, using
derivative-free nature-inspired global optimization algorithms such as ant colony
optimization (ACO), differential evolution algorithm (DEA), genetic algorithm
(GA), genetic-price algorithm (GPA), particle swarm optimization (PSO) and
simulated annealing (SA) has become widespread for the model parameter esti-
mation studies in geophysics. Some of the published works performed using these
algorithms are listed in Table 7.1. Due to the mentioned-above advantages and
conveniences, some of these algorithms are presented in this study. Here, funda-
mentals of PSO, DEA and recently suggested differential search algorithm
(DSA) with application to geophysical inversion using self-potential (SP), magnetic
and gravity data sets are described in sufficient detail. Additionally, some statistical
studies to determine the convergence characteristics of the presented algorithms and
also to estimate the possible uncertainties in the obtained results are presented to
provide some insights.
7.2 Methodology
7.2.1 PSO
these parameters are modified in the multi-dimensional search space through the
equations given below (Kennedy and Eberhart 1995; Shi and Eberhart 1998)
vki þ 1 ¼ w vki þ c1 r1 pki xki þ c2 r2 gki xki
ð7:1Þ
xki þ 1 ¼ xki þ vki ;
where, vki represents particle velocity i at the kth iteration, xki is model i at kth
iteration, w denotes inertia weight value, and c1 and c2 are the controllers of the
particle’s individual and social behaviours, respectively, r1 and r2 denote the
arbitrary numbers between 0 and 1 (Srivastava and Agarwal 2010). The inertia
weight balances the exploration and exploitation (Bansal et al. 2011). In PSO
algorithm, the iterations are terminated after reaching the user-defined maximum
iteration number or reaching the predefined error value (Shi and Eberhart 1998; Poli
et al. 2007; Luke 2009; Salmon 2011; Pekşen et al. 2011, 2014; Göktürkler and
Balkaya 2012; Ekinci 2016; Essa and Elhussein 2018). The stages of the PSO
algorithm are shown in Fig. 7.2 (left panel) via a flowchart. In the applications
presented here, 30 independent runs were performed by using 150 particles in each
run. The constant values suggested by Carlisle and Dozier (2001) were assigned to
w, c1 and c2 .
Start
Start
Initialize Particles Start
Set Control Parameters
Calculate fitness values of each particles Set Control Parameters
G=G+1
Crossover
Calculate each particle’s velocity
Fig. 7.2 Generalized flowcharts of PSO (left panel), DEA (right panel) and DSA (mid panel)
7 Global Optimization of Near-Surface Potential … 161
7.2.2 DEA
DEA, which is used to optimize real parameters and real-valued functions (Storn
and Price 1995, 1997; Storn 1996), is mainly considered as an efficient population-
and vector-based metaheuristic. It is a type of evolutionary algorithms such as GAs
(Holland 1975), evolution strategies (Rechenberg 1973; Schwefel 1977) and evo-
lutionary programming (Fogel et al. 1966). During the optimization, it, therefore,
displays a general evolutionary algorithm procedure including initialization,
mutation, recombination and selection stages (Fig. 7.2, right panel).
In the optimization of a function having D real parameters, population number
(Np), crossover probability (Cr) and the mutation constant (orweighting factor,
F) are
selected first. The parameter vector shows the form xi;G ¼ x1i;G ; x2i;G ; . . .; xD
i;G ; i ¼
ð1; . . .; NpÞ where i represents the index for each individual and G denotes the
current generation. An initial population is then generated via the following
equation
j
xi;G ¼ xlj þ rand(0; 1Þ xuj xlj ; j ¼ 1; 2; . . .; D; ð7:2Þ
where, x is target vectors, j indicates parameters, rand denotes the arbitrary number
between 0 and 1, l and u are the lower and upper parameter limits, respectively.
Mutation, crossover and selection operations are then applied to each of the Np
parameter vectors by using the following mathematical notations
vi;G ¼ xr i3; G þ F xr i1; G xr i2; G ; ð7:3Þ
( j
j
vi;G if ðrandð0; 1Þ Cr or j ¼ jrand Þ
ui;G ¼ j
xi;G otherwise; ð7:4Þ
j ¼ 1; 2; . . .; D;
ui;G if f ui;G f xi;G
xi;G þ 1 ¼ ð7:5Þ
xi;G otherwise:
In DEA terminology, DE/x/y/z is used for its evolution strategies. Here, x represents
the mutation scheme chosen, y shows the number of difference vectors, and
z symbolises binomial or exponential crossover scheme (Balkaya et al. 2017). Of
the strategies in DEA, DE/best/1/bin was applied in this study due to its advantages
(Balkaya 2013; Ekinci et al. 2016). Considering at each generation G, best men-
tioned above displays the best individual vector in the population. The trial vector
ui;G is achieved from both donor vector elements, vi;G , and target vector, xi;G . Donor
vector elements get involved in the trial vector with a user-defined crossover
probability Cr which is a value between 0 and 1, and jrand denotes randomly
selected integer changing between 1 and D. Thus, recombination combines suc-
cessful solutions considering the previous G. Lastly, after comparing the target
vector with the trial vector, the one having the lowest misfit is transferred to the next
generation. These routines in the evolution cycle continue until reaching the pre-
defined maximum iteration number or obtaining the desired objective function
value. Here, in both synthetic and real data applications, 30 independent runs were
performed using the values 0.5, 0.9 and 180 for F, Cr and Np, respectively.
7.2.3 DSA
DSA (Fig. 7.2, mid panel) is one of the recent swarm-based metaheuristics
(Civicioglu 2012) used for the optimization of real-valued problems. It is essentially
inspired by the migration behaviour of superorganisms composed of a large number
of individuals such as fire ants, honeybees, monarch butterflies, and many bird spe-
cies. The superorganism may exhibit seasonal migratory behaviour in some parts of
the year because vital resources such as water resources and pastures may be reduced
for some reasons such as periodic climate changes (Civicioglu 2012). The superor-
ganism settles temporarily in the new stopover site after checking its suitability. Then,
their migration starts again for discovering better sites. In the algorithm, a population
consists from an artificial-superorganism ðSuperorganismG ¼ ½XiÞ which involves
artificial-organisms
ðXi ; i ¼ f1; 2; 3; . . .; NpgÞ. Considering dimension of the prob-
lem Xi;j ; j ¼ f1; 2; 3; . . .; Dg , each individual of artificial-organism contains ran-
dom solution vectors aiming to reach the global minimum. Here, while G represents
the number of generations, i and j are counters of the individual’s number in the
superorganism (Np) and unknown parameter number (D), respectively. An
artificial-organism xi;j in the predefined upper ðup ¼ fu1 ; u2 ; . . .uD gÞ and lower
ðlow ¼ fl1 ; l2 ; . . .lD gÞ parameter bounds is generated as given in DEA section.
Therefore, artificial-organisms are shown by Xi ¼ xi;j .
7 Global Optimization of Near-Surface Potential … 163
7.2.4.1 SP Anomaly
The general expression of SP anomaly caused by an inclined sheet body (Fig. 7.3a)
at an observation point on the surface is defined as follows (Murthy and Haricharan
1985; Santos 2010; Sharma and Biswas 2013)
" #
½ðx xoÞ a cos a 2 þ ½zo a sin a 2
VðxÞ ¼ K ln ; ð7:9Þ
½ðx xoÞ a cos a 2 þ ½zo þ a sin a 2
where, x and xo denote the horizontal distance and the location of the source body,
respectively, K represents the electric dipole moment, a is the angle of inclination in
degrees, zo and a denote the depth and the half-width of the causative source,
respectively.
(a) 40 x
20 P (x,z)
SP Anomaly [mV]
xo
0 zo
-20 α
α
-40
-60 2a
-80 z
(b) x
50
P (x,z)
Gravity Anomaly [mGal]
40 xo
r zo
r
30
r
20
10
0 z
(c) 500
xo x
Magnetic Anomaly [nT]
400 P (x,z)
r1 θ1
z1 θ2
300
200
z2
100 r2
0
θ
-100 φ
0 10 20 30 40 50 60 70 80 90 100
Distance [m] J
Fig. 7.3 Geological models and corresponding synthetic anomalies for a SP, b magnetic and
c gravity applications, respectively
7 Global Optimization of Near-Surface Potential … 165
r2
TðxÞ ¼ 2J 0 sin h cosðh þ /0 Þðh1 þ h2Þ þ sinðh þ /0 Þ ln ; ð7:10Þ
r1
h i 1=2
r2 ¼ ððx xoÞ þ ðz2 z1Þ cot hÞ2 þ z22 ; ð7:14Þ
where, x and xo are the horizontal distance along the observation line and the origin,
respectively, / denotes the dip of effective magnetization vector, z1 and z2 rep-
resent top and bottom depths, respectively, a is the strike of the source, Dm denotes
direction of measurement, being equal to p=2 for the vertical component, 0 for the
horizontal component and earth’s magnetic field inclination for the total field.
The general gravity anomaly due to some idealized sources at an observation point
(Fig. 7.3c) is given as below (Abdelrahman et al. 1989; Essa 2014; Mehanee 2014;
Ekinci et al. 2016)
166 Y. L. Ekinci et al.
Table 7.2 The details of the parameters used in gravity anomaly computation
Model A η q
Semi-infinite vertical cylinder pcrr2 0 0.5
Infinitely long horizontal cylinder 2pcrr2 1 1
Sphere 4pcrr3/3 1 1.5
c: The universal gravitational constant (6.67384 10−11 m3 kg−1 s2), r: The density contrast
(g cm−3), r: The radius of the source (m)
zog
gðxÞ ¼ A h iq ; ð7:17Þ
ðx xoÞ2 þ zo2
where, A is the amplitude coefficient, zo denotes the depth of the causative body,
q and ƞ indicate the geometric shape factors, x and xo denote the horizontal distance
along the observation profile and the location of the source, respectively. Details of
the variables are given in Table 7.2.
7.3 Applications
where, T represents the transposition of the matrix, N denotes the data number, and
dobs and dcal symbolize observed and calculated data, respectively. The square root
of Eq. 7.18 is rms error which was considered in the applications.
Using the anomaly equation given in Eq. 7.9 a synthetic SP anomaly was produced
(Fig. 7.3a) along a 100-m long profile assuming 1 m sampling interval. Table 7.3
7 Global Optimization of Near-Surface Potential … 167
Table 7.3 Optimizations parameters and estimated values through the PSO for SP sheet model
Model True Search Estimated parameters Estimated parameters
parameters values spaces (noise-free case) (noisy case)
min max
K (mV) 50 1 500 50.05 43.65
xo (m) 50 1 100 50 49.79
zo (m) 10 1 50 10 9.68
a (m) 5 1 50 4.99 5.66
a (°) 30 1 180 30 29.9
rms (mV) 2.00e−03 1.68
shows sheet body parameters. Before the inversion studies, prediction error maps
(Figs. 7.4 and 7.5) were produced to understand the relationships between the
parameter pairs. In these procedures values of unused parameters were set to the
actual ones. Error maps were produced with small search ranges and true param-
eters were highlighted by white circles. Both horizontal and vertical axes limits are
equal to search space bounds for every parameter. The a–K map (Fig. 7.4) consists
of narrow valley topographies nearly parallel to each other. The global minimum is
positioned in one of these unclosed valleys. This unclosed contour topography
implies the existence of many solutions having almost the same error energy values.
The slopping closed elliptical contours in the maps of zo–K and a–zo (Fig. 7.4)
denote that these parameter pairs are correlated with each other. In the error energy
maps of a–K, a–zo and a–a (Figs. 7.4 and 7.5) the uncorrelated natures of the
parameter pairs are clearly identified due to nearly circular contour topographies
enclosing the global minima and therefore these parameters can be resolved inde-
pendently. Elliptical closed contours nearly parallel to the horizontal axes are
evident in the maps of parameter pairs of xo–a, xo–K, xo–zo and xo–a (Figs. 7.4
and 7.5). This behaviour indicates the uncorrelated nature of these parameters and
they can be resolved independently too. Additionally, it must be noted that the short
axes of the ellipses indicate the parameter xo, and thus this parameter can be
determined more sensitively than the other ones. Considering the produced pre-
diction error maps, it can be mentioned here that the sheet parameters of the SP
problem can be estimated considerably via an efficient inversion procedure.
Figure 7.6 (upper panels) shows the fit between theoretical noise-free and noisy
sheet anomalies and calculated corresponding anomalies using the best solution
achieved through PSO applications. Table 7.3 displays search space bounds and the
obtained parameters of both cases. In noise-free anomaly case it is obvious that
although relatively wide search space bounds were used, accurate parameter values
were obtained. Noisy anomaly case also produced satisfactory results with an rms
error of 1.68 mV. This value is well-matched to the standard deviation (std) of
artificial noise content (1.74 mV). The rapid decreases in the error energy values
against generation number for both cases in the best run are shown in Fig. 7.7
(upper panels). In addition to inversion studies, uncertainty appraisal analyses in the
168 Y. L. Ekinci et al.
14 7
40
12 6
35
zo [m]
a [m]
α [ o]
10 30 5
25
8 4
20
6 3
30 40 50 60 70 30 40 50 60 70 30 40 50 60 70
K [mV] K [mV] K [mV]
70 7
40
60 6
35
xo [m]
a [m]
α [o ]
50 5 30
25
40 4
20
30 3
30 40 50 60 70 6 8 10 12 14 6 8 10 12 14
K [mV] zo [m] zo [m]
Fig. 7.4 Error energy maps obtained from noise-free SP anomaly case. White circles show the
true solutions
40
35
α [o]
30
70 70
25
60 60
20
xo [m]
xo [m]
50 3 4 5 6 7 50
a [m]
70
40 40
60
30 30
6 8 10 12 14 3 4 5 6 7
xo [m]
zo [m] a [m]
50
40
30
20 25 30 35 40
α [o]
Fig. 7.5 Error energy maps obtained from noise-free SP anomaly case. White circles show the
true solutions
7 Global Optimization of Near-Surface Potential … 169
40 40
20 20
SP anomaly [mV]
0 0
-20 -20
-40 -40
-60 -60
-80 -80
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Distance [m] Distance [m]
100
0
SP anomaly [mV]
-100
-200
-300
-400
-150 -100 -50 0 50 100
Distance [m]
Fig. 7.6 Theoretical and calculated noise-free SP anomalies (upper left panel), noisy case (upper
right panel), observed and calculated SP anomalies of the field case (lower panel)
104 103
Error energy [mV2]
102
102
100
10-2
101
10-4
10-6 100
0 50 100 150 200 250 300 0 50 100 150 200 250 300
Generation number Generation number
104
Error energy [mV2]
103
102
101
0 50 100 150 200 250 300
Generation number
Fig. 7.7 Error energy values versus generation number in SP data applications. Upper left, upper
right and lower panels illustrate noise-free, noisy and field data cases
170 Y. L. Ekinci et al.
0.12 1 1
50.05 50.00 10.00
0.1
0.8 0.8
Relative frequency
0.08
0.6 0.6
0.06
0.4 0.4
0.04
0.2 0.2
0.02
0 0 0
40 50 60 49 50 51 9 10 11 12
K [mV] xo [m] zo [m]
1.2 0.3
4.99 30.00
1 0.25
Relative frequency
0.8 0.2
0.6 0.15
0.4 0.1
0.2 0.05
0 0
4 5 6 26 28 30 32 34
a [m] α [ o]
Fig. 7.8 Histograms produced using the M–H algorithm for noise-free synthetic SP data case.
Estimated values are shown at the upper left corners of the plots
The total field magnetic anomaly of an arbitrary magnetized fault model shown in
Fig. 7.3b was produced along a 100-m long profile assuming 1 m sampling interval
via Eq. 7.10. The model parameters of the magnetized fault are shown in Table 7.4.
The relationships between the parameter pairs were analysed through error energy
maps (Figs. 7.10 and 7.11). The actual values of parameters pairs in the error
energy maps are indicated by white circles. Both horizontal and vertical axes limits
7 Global Optimization of Near-Surface Potential … 171
0.01 3 0.12
50.00 50.00 10.00
true value 2.5 0.1
0.008
2 0.08
0.006
PDF
1.5 0.06
0.004
1 0.04
0.002 0.5 0.02
0 0 0
-50 0 50 100 49.8 50 50.2 5 10 15
K [mV] xo [m] zo [m]
0.1 0.025
5.00 30.00
0.08 0.02
0.06 0.015
PDF
0.04 0.01
0.02 0.005
0 0
-5 0 5 10 0 20 40 60
a [m] α [o]
PDF between 95% Mean The best value
parameter value PSO
Fig. 7.9 PDF plots obtained from 30 independent runs for noise-free synthetic SP data case. True
values are shown at the upper left corners of the plots
are equal to search space bounds for every parameter. Unclosed contours nearly
parallel to the horizontal axis are dominant in the map of parameter pair of xo–h
(Fig. 7.11). However, a narrow elliptical closed contour surrounding the global
minimum increase the independence and resolvability of the parameters. Nearly
circular closed contours surrounding the global minima are clearly seen in the error
energy maps of h–z1, /–z1, h–z2, /–z2, J–h and J–/ pairs (Figs. 7.10 and 7.11).
Thus, these parameter pairs are uncorrelated with each other, and they can be
estimated accurately independently. The slopping closed elliptical contours are
evident in the maps of z2–z1, J–z1, J–z2, /–h and xo–/. These parameter pairs are
correlated with each other. If one parameter is resolved the other one can be
resolved as well. xo–z1, xo–z2 and xo–J error maps exhibit elliptical closed con-
tours parallel to the horizontal axes. These parameters are uncorrelated with each
other. Alike to the SP case, the short axis of the ellipses indicates the parameter xo.
Thus, more precise solutions can be obtained for the parameter xo. These maps
(Figs. 7.10 and 7.11) indicate that all parameters can be resolved using an efficient
inversion process. The noise-free and noisy anomalies and calculated corresponding
anomalies obtained from the best run are demonstrated in Fig. 7.12 (upper panels).
Table 7.4 shows search space bounds and the obtained parameters of both cases.
Noise-free case produced a solution providing precise parameter values. In the
172 Y. L. Ekinci et al.
Table 7.4 Optimizations parameters and estimated values through the DEA for magnetic fault
model
Model True Search spaces Estimated parameters Estimated
parameters values min max (noise-free case) parameters (noisy
case)
z1 (m) 3 0.1 10 3 2.81
z2 (m) 12 10 100 12 12.45
h (°) 60 1 180 60 65.09
/ (°) 70 −150 150 70 69.8
J (nT) 180 1 1000 180 167.79
xo (m) 50 1 100 50 49.8
rms (nT) 1.00e−06 6.32
16 80
90
14 70 80
z2 [m]
φ [o]
θ [o]
12 60 70
10 50 60
50
8 40
60
70
200
xo [m]
J [nT]
θ [o]
50 60
160
50
40
120 40
30
2 2.5 3 3.5 4 2 2.5 3 3.5 4 8 10 12 14 16
z1 [m] z1 [m] z2 [m]
70
240
90
60
80
200
xo [m]
J [nT]
φ [o]
70 50
160
60
40
50
120
30
8 10 12 14 16 8 10 12 14 16 8 10 12 14 16
z2 [m] z2 [m] z2 [m]
Fig. 7.10 Error energy maps obtained from noise-free magnetic anomaly case. White circles
show the true solutions
7 Global Optimization of Near-Surface Potential … 173
70
240
90
60
80
200
xo [m]
J [nT]
φ [o]
70 50
160
60
40
50
120
30
40 50 60 70 80 40 50 60 70 80 40 50 60 70 80
θ [o] θ [o] θ [o]
70 70
240
60 60
200
xo [m]
xo [m]
J [nT]
50 50
160
40 40
120
30 30
50 60 70 80 90 50 60 70 80 90 120 160 200 240
φ [o] φ [o] J [nT]
Fig. 7.11 Error energy maps obtained from noise-free magnetic anomaly case. White circles
show the true solutions
500 500
400 400
Magnetic anomaly [nT]
300 300
200 200
100 100
0 0
-100 -100
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Distance [m] Distance [m]
120
100
Magnetic anomaly [nT]
80
60
40
20
-20
0.5 1 1.5 2 2.5 3 3.5 4
Distance [km]
Fig. 7.12 Theoretical and calculated noise-free magnetic anomalies (upper left panel), noisy case
(upper right panel), observed and calculated magnetic anomalies of the field case (lower panel)
174 Y. L. Ekinci et al.
104 104
102
Error energy [nT2]
100
103
10-2
10-4
102
10-6
-8
10
10-10 101
0 50 100 150 200 250 300 0 50 100 150 200 250 300
Generation number Generation number
4
10
Error energy [nT2]
103
102
101
100
0 50 100 150 200 250 300
Generation number
Fig. 7.13 Error energy values versus generation number in magnetic data applications. Upper left,
upper right and lower panels illustrate noise-free, noisy and field data cases
3 1.2 0.15
3.00 12.00 60.00
2.5 1
Relative frequency
2 0.8 0.1
1.5 0.6
1 0.4 0.05
0.5 0.2
0 0 0
2.6 2.8 3 3.2 11 11.5 12 12.5 13 55 60 65
z1 [m] z2 [m] θ [o]
0.7 0.05 3
70.00 180.00 50.00
0.6 2.5
0.04
Relative frequency
0.5
2
0.4 0.03
1.5
0.3 0.02
1
0.2
0.01
0.1 0.5
0 0 0
68 69 70 71 160 170 180 190 200 49.6 49.8 50 50.2 50.4
φ [o] J [nT] xo [m]
Fig. 7.14 Histograms produced using M‒H algorithm for noise-free synthetic magnetic data case.
Estimated values are shown at the upper left corners of the plots
7 Global Optimization of Near-Surface Potential … 175
1000 800
1500
PDF
600
1000
500 400
500 200
0 0 0
2.9998 3 3.0002 11.9996 12 12.0004 59.9995 60 60.0005
z1 [m] z2 [m] θ [o]
× 10 4
600 800 2.5
70.00 180.00 50.00
500 2
600
400
1.5
PDF
300 400
1
200
200
100 0.5
0 0 0
69.999 70 70.001 179.999 180 49.99998 50 50.00002
φ [o] J [nT] xo [m]
PDF between 95% Mean The best value
parameter value
Fig. 7.15 PDF plots obtained from 30 independent runs for noise-free synthetic magnetic data
case. True values are shown at the upper left corners of the plots
noisy data case, an rms error of 6.32 nT was obtained which matches well to the
added noise content (std = 6.57 nT). Additionally, the fast convergence charac-
teristic of DEA in the best run is clearly observed from Fig. 7.13 (upper panels).
Both cases verify the efficiency of DEA even using very wide search limits.
Histogram reconstructions (Fig. 7.14) produced through the M–H algorithm
showed that the estimated parameters are within the confidence intervals indicating
the reliability of the solutions. Additionally, PDF plots (Fig. 7.15) using the best
solutions of 30 runs clearly indicated that DEA yielded model parameter solutions
within confidence interval without uncertainties, confirming the information
obtained from the error energy maps (Figs. 7.10 and 7.11).
Table 7.5 Optimizations parameters and estimated values through the DSA for gravity horizontal
cylinder model
Model True Search spaces Estimated parameters Estimated
parameters values min max (noise-free case) parameters (noisy
case)
A (mGal m) 500 1 1000 556.67 836.39
zo (m) 10 1 50 10 9.77
q 1 0.5 1.5 1 0.98
ƞ 1 0 1 0.95 0.72
xo (m) 50 1 100 50 49.96
rms (mGal) 2.80e−05 1.07
energy maps of each parameter pairs (Fig. 7.16). Like the previous examples, white
circles in the error energy maps indicate the real values. Both horizontal and vertical
axes limits are equal to search space bounds for every parameter. Narrow and long
valleys characterized by unclosed contours seen in the maps of ƞ–A, q–A and ƞ–
q pairs draw the attention which indicate the existence of many solutions having
almost the same error energy values. Positive and negative correlations are clearly
seen in the error maps of q–A and ƞ–A, respectively (Fig. 7.16). Thus, accurate
estimations for these parameters are not expected concurrently. On the other hand,
contours lying nearly parallel to the ƞ axis are evident in the map of q–ƞ. This
response indicates that q can be resolved while the possibility of estimating ƞ
successfully is very low. Nearly circular and elliptical contours in the other error
energy maps imply that xo and zo are resolvable with a high degree of probability.
The inversion results showing the fit between theoretical and calculated anomalies
for both cases are demonstrated in Fig. 7.17 (upper panels). Search limits and the
estimated model parameters for both cases are also given in Table 7.5. The beha-
viours of error energy curves against the generation number for both cases in the
best run are shown in Fig. 7.18 (upper panels). Although exact fits between theo-
retical and calculated anomalies were obtained in both cases, satisfactory parameter
estimates could not be obtained for the model parameters ƞ and A. It must be noted
that this finding is not surprising based on the information about the nature of
inverse problem gained from the error energy maps (Fig. 7.16). However, satis-
factory solutions were obtained for the parameters xo, zo and q albeit the use of
wide search spaces. Hence depth, location and the source geometry can be deter-
mined by the inversion process. Histograms produced via M–H algorithm
(Fig. 7.19) show the validity of the estimated parameter values. Moreover, PDF
plots (Fig. 7.20) clearly show that xo, zo and q are within the confidence intervals
while the other couple has considerable uncertainties, confirming the information
obtained from prediction error energy maps (Fig. 7.16).
7 Global Optimization of Near-Surface Potential … 177
14 1.4 1.4
12 1.2 1.2
zo [m]
10 1 1
q
8 0.8 0.8
0.6 0.6
300 400 500 600 700 300 400 500 600 300 400 500 600 700
A [mGal.m] A [mGal.m] A [mGal.m]
70 1.4 1.4
60 1.2 1.2
xo [m]
50 1 1
q
η
40 0.8 0.8
30 0.6 0.6
300 400 500 600 700 6 8 10 12 14 6 8 10 12 14
A [mGal.m] zo [m] zo [m]
70 1.4 70
60 1.2 60
xo [m]
xo [m]
50 1 50
η
40 0.8 40
30 0.6 30
6 8 10 12 14 0.6 0.8 1 1.2 1.4 0.6 0.8 1 1.2 1.4
zo [m] q q
Fig. 7.16 Error energy maps obtained from noise-free gravity anomaly case. White circles show
true solutions
50 50
40 40
30 30
20 20
10 10
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Distance [m] Distance [m]
0.25
Gravity anomaly [mGal]
0.2
0.15
0.1
0.05
0
0 20 40 60 80 100 120
Distance [m]
Fig. 7.17 Theoretical and calculated noise-free gravity anomalies (upper left panel), noisy case
(upper right panel), observed and calculated gravity anomalies of the field case (lower panel)
4
10 102
Error energy [mGal2]
2
10
0
10
10-2
101
10-4
10-6
10-8
10-10 100
0 50 100 150 200 250 300 0 50 100 150 200 250 300
Generation number Generation number
10-2
Error energy [mGal2]
10-3
10-4
10-5
0 50 100 150 200 250 300
Generation number
Fig. 7.18 Error energy values versus generation number in gravity data applications. Upper left,
upper right and lower panels illustrate noise-free, noisy and field data cases
7 Global Optimization of Near-Surface Potential … 179
0.06 0.7 10
646.08 10.00 1.00
0.05 0.6
8
Relative frequency
0.5
0.04
0.4 6
0.03
0.3 4
0.02
0.2
2
0.01 0.1
0 0 0
600 700 800 9 10 11 12 0.9 1 1.1
A [mGal.m] zo [m] q
6 0.8
0.89 50.00
5
0.6
Relative frequency
3 0.4
2
0.2
1
0 0
0.7 0.8 0.9 1 1.1 49 50 51 52
η xo [m]
Fig. 7.19 Histograms produced using M‒H algorithm for noise-free synthetic gravity data case.
Estimated values are shown at the upper left corners of the plots
of the error energy values in the first 30 generations obtained in best run shows the
efficiency of the PSO algorithm (Fig. 7.7, lower panel). Optimization produced a
result of 31.15 m depth which is close to the results of some published studies such
as 21 m (Paul 1965), 30.48 m (Rao et al. 1970), 29.88 m (Jagannadha et al. 1993),
27.65 m (Sundararajan et al. 1998), 26.52 m (Radhakrishna Murthy et al. 2005),
27.78 m (El-Kaliouby and Al-Garni 2009) and 31.40 m (Santos 2010). It must be
noted that Santos (2010) estimated the model parameters by using PSO. However,
in his paper, there is no information about the number of the independent runs and
sampling interval of the inverted SP data. Additionally, we used more number of
particles as compared to his study. The inertia weight and the learning factors
controlling cognitive and social behaviours are also different in our study, which are
given previously in the synthetic example section. Using the obtained model
parameters given in Santos (2010), we produced the SP model response using a data
sampling interval of 5 m. Then, rms error value of the inversion study of Santos
(2010) was computed, and it was compared with the one obtained in the present
study. Although we used a wider search space bound, our solution produced a
lower rms error (6.04 mV) in comparison with the study of Santos (2010) that is
6.23 mV which emphasises the significance of the selection of user-defined control
parameters in PSO.
180 Y. L. Ekinci et al.
× 10 -3
2.5 600 2500
500.00 10.00 1.00
300
1 1000
200
0.5 100 500
0 0 0
600 800 1000 9.999 10 10.001 0.9998 1 1.0002
A [mGal.m] zo [m] q
4 2500
1.00 50.00
2000
3
1500
PDF
2
1000
1
500
0 0
0.6 0.7 0.8 0.9 1 49.9998 50 50.0002
η xo [m]
PDF between 95% Mean The best value
parameter value
Fig. 7.20 PDF plots obtained from 30 independent runs for noise-free synthetic gravity data case.
True values are shown at the upper left corners of the plots
likely due to the abrupt thinning of Douro volcanic against a fault (Qureshi and
Nalaye 1978). Douro volcanic consists mainly of coarse tuffs and are about 1000 m
thick (Qureshi and Nalaye 1978). Here, the magnetic anomaly was inverted using a
fault model via DEA. We used the same user-defined control parameters as men-
tioned in the synthetic study section for the optimization. The 4-km long magnetic
anomaly was sampled using 80 m data intervals. First 10 data points at the
beginning of the profile which are not thought to be associated with the magnetic
anomaly source were not taken into consideration. Search space and parameter
estimations are listed in Table 7.7. After performing 30 independent runs a good
agreement between observed magnetic anomaly and the magnetic response of the
best-fitting model parameters was obtained (Fig. 7.12, lower panel). The error
energy curve against the generation number, obtained in the best run, is shown in
Fig. 7.13 (lower panel). Considering the flight height, the best solution having the
lowest rms error yielded 135.81 and 1149.84 m depths to the top and bottom of the
fault, respectively. Thus, the thickness of about 1014 m was determined for the
fault structure. This solution is in well agreement with the ones obtained by Qureshi
and Nalaye (1978). They reported 115 m depth to the top of the fault from the
ground surface and a fault thickness of 990 m, which yield a bottom depth of
1105 m.
obtained. The decrease of the error energy values against the generation number,
obtained in the best run, is shown in Fig. 7.18 (lower panel). Estimated shape factor
q (1.5) strongly indicated that this gravity anomaly is due to a spherical body.
A geologically reasonable solution, that is 23.23 m depth to the centre was esti-
mated for the spherical body. There are some results in the literature obtained via
different algorithms about this residual gravity anomaly. Robinson and Coruh
(1988) reported a depth of 21 m to centre using a spherical model. Roy (2001)
reported of 19.4 m depth to the top a finite vertical line. 23.8 m depth to the centre
of a sphere was reported by Salem et al. (2003). Asfahani and Tlas (2008) reported
a horizontal cylinder buried at 17.7 m depth. Mehanee (2014) suggested a hori-
zontal cylinder body with a depth of 16 m to its centre. Ekinci et al. (2016) sug-
gested a sphere model having a depth of 23.22 m to the centre. It is clear that
previously published results are close to the one determined via DSA.
7.4 Conclusions
and DSA, model parameter estimations from both theoretical and real field
anomalies due to near-surface geological sources were presented. Each optimization
algorithm was used to invert different type of potential field data sets. Prediction
error energy images for each parameter couples were produced to characterize the
nature of the presented inverse problems. By this way, parameter resolutions and
dependencies were identified before the model parameter estimation studies.
Although pleasing solutions were obtained in theoretical experiments even in the
existence of a reasonable amount of noise content, some efficient statistical analyses
which give insight into possible uncertainties in the obtained solutions were also
performed. For this purpose, M–H sampling algorithm constructed on the basis of
SA without cooling schedule, and PDF plots using the best solutions of independent
runs were utilized. These analyses clearly indicated that the most of the obtained
model parameters are within the confidence intervals despite the wide search space
bounds used in the optimization. Additionally, rapid and stable convergence
characteristics of the suggested algorithms were demonstrated by producing error
energy versus generation number plots.
In the real data cases, an SP anomaly from a sulphide mineralization zone
(India), a total field magnetic anomaly from a fold belt (Australia) and a gravity
anomaly from a chromite deposit (Cuba) were inverted to estimate some model
parameters of causative geological structures through PSO, DEA and DSA,
respectively. Each field anomalies were assumed to be caused from a different type
of geological structure such as sheet, fault and sphere as stated in the literature
previously. Model parameter estimation studies yielded geologically reasonable
results that are in well agreements with the ones of published studies.
Since geophysical inverse problems have an ill-posed nature, there may be many
solutions in the model space having nearly the same error rate. Therefore, con-
sidering only the match between observed and calculated anomaly may lead to
unreasonable interpretations about the sources. In addition, each independent run
may produce different model solutions when working with global optimization
algorithms. In particular, in the lack of a priori information and some constraints, it
seems difficult to choose the proper parameter solution among the best-fitting
solutions obtained from independent runs. However, information about the model
confidence intervals and uncertainties in the solutions may facilitate interpreter’s
work. Thus, some efficient statistical analyses should be implemented in the
inversion studies performed through stochastic global optimization algorithms. It is
also worth noting that the wide search space limits used in our cases did not affect
the solutions unfavourably. However, in other studies, some test studies by using
various search space limits should be carried out by the researchers to understand
the effects of the search space limits on the solutions. Finally, it is mentioned here
that stochastic global optimization algorithms are powerful tools for geophysical
parameter estimation studies.
Acknowledgements We thank Prof. Dr. Arkoprovo Biswas for inviting us to participate in this
edited book as authors, and also thank anonymous reviewers for their many constructive com-
ments, which have greatly improved the earlier version of this book chapter.
184 Y. L. Ekinci et al.
References
Abdelazeem M, Gobashy M (2006) Self-potential inversion using genetic algorithm. JKAU Earth
Sci 17:83–101
Abdelrahman EM, Bayoumi AI, Abdelhady YE, Gobashy MM, El-Araby HM (1989) Gravity
interpretation using correlation factors between successive least-squares residual anomalies.
Geophysics 54:1614–1621
Alkan H, Balkaya Ç (2018) Parameter estimation by differential search algorithm from horizontal
loop electromagnetic (HLEM) data. J Appl Geophys 149:77–94
Asfahani J, Tlas M (2008) An automatic method of direct interpretation of residual gravity
anomaly profiles due to spheres and cylinders. Pure Appl Geophys 165:981–994
Balkaya Ç, Göktürkler G, Erhan Z, Ekinci YL (2012) Exploration for a cave by magnetic and
electrical resistivity surveys: Ayvacık sinkhole example, Bozdağ İzmir (Western Turkey).
Geophysics 77(3):B135–B146
Balkaya Ç (2013) An implementation of differential evolution algorithm for inversion of
geoelectrical data. J Appl Geophys 98:160–175
Balkaya Ç, Ekinci YL, Göktürkler G, Turan S (2017) 3D non-linear inversion of magnetic
anomalies caused by prismatic bodies using differential evolution algorithm. J Appl Geophys
136:372–386
Bansal JC, Singh PK, Saraswat M, Verma A, Jadon SS, Abraham A (2011) Inertia weight
strategies in particle swarm optimization. In: IEEE third world congress nature and biologically
inspired computing, pp 633–640
Başokur AT, Akça I, Siyam NWA (2007) Hybrid genetic algorithms in view of the evolution
theories with application for the electrical sounding method. Geophys Prospect 55:393–406
Bekler T, Ekinci YL, Demirci A, Erginal AE, Ertekin C (2011) Characterization of a landslide
using seismic refraction, electrical resistivity and hydrometer methods, Adatepe-Çanakkale,
Turkey. J Environ Eng Geophys 16(3):115–126
Bektaş Ö, Büyüksaraç A, Rozimant K (2013) 3D modelling and structural investigation of the
central volcanics in Slovakia using magnetic data. Carpath J Earth Env 124:27–33
Biswas A (2015) Interpretation of residual gravity anomaly caused by a simple shaped body using
very fast simulated annealing global optimization. Geosci Front 6:875–893
Biswas A (2016) Interpretation of gravity and magnetic anomaly over thin sheet-type structure
using very fast simulated annealing global optimization technique. Model Earth Syst Environ
2:30
Biswas A (2017) A review on modeling, inversion and interpretation of self-potential in mineral
exploration and tracing paleo-shear zones. Ore Geol Rev 91:21–56
Biswas A, Acharya T (2016) A very fast simulated annealing method for inversion of magnetic
anomaly over semi-infinite vertical rod-type structure. Model Earth Syst Environ 2:198
Biswas A, Sharma SP (2014) Optimization of self-potential interpretation of 2-D inclined
sheet-type structures based on very fast simulated annealing and analysis of ambiguity. J Appl
Geophys 105:235–247
Biswas A, Parija MP, Kumar S (2017) Global nonlinear optimization for the interpretation of
source parameters from total gradient of gravity and magnetic anomalies caused by thin dyke.
Ann Geophys 60:G0218
Cai H, Zhdanov MS (2015) Modeling and inversion of magnetic anomalies caused by
sediment-basement interface using 3D Cauchy-type integrals. IEEE Geosci Remote Sens
Lett 12:477–481
Carlisle A, Dozier G (2001) An off-the-shelf PSO. In: Proceedings of workshop on particle swarm
optimisation (Indianapolis, IN), pp 1–6
Chunduru RK, Sen MK, Stoffa PL (1997) Hybrid optimization for geophysical inversion.
Geophysics 62:1196–1207
Civicioglu P (2012) Transforming geocentric Cartesian coordinates to geodetic coordinates by
using differential search algorithm. Comput Geosci 46:229–247
7 Global Optimization of Near-Surface Potential … 185
Civicioglu P (2013) Circular antenna array design by using evolutionary search algorithms. Prog
Electromagn Res B 54:265–284
Čuma M, Wilson GA, Zhdanov MS (2012) Large-scale 3D inversion of potential field data.
Geophys Prospect 60:1186–1199
Das S, Abraham A, Konar A (2008) Particle swarm optimization and differential evolution
algorithms: technical analysis, applications and hybridization perspectives. Stud Comput Intell
116:1–38
Davis WE, Jackson WH, Richter DH (1957) Gravity prospecting for chromite deposits in
Camaguey province, Cuba. Geophysics 22:848–869
Di Maio R, Rani P, Piegari E, Milano L (2016) Self-potential data inversion through a
genetic-price algorithm. Comput Geosci 94:86–95
Ekinci YL (2016) MATLAB-based algorithm to estimate depths of isolated thin dike-line sources
using higher-order horizontal derivatives of magnetic anomalies. SpringerPlus 5:1384
Ekinci YL, Demirci A (2008) A damped least-squares inversion program for the interpretation of
Schlumberger sounding curves. J Appl Sci 8:4070–4078
Ekinci YL, Yiğitbaş E (2012) A geophysical approach to the igneous rocks in the Biga Peninsula
(NW Turkey) based on airborne magnetic anomalies: geological implications. Geodin Acta 25
(3):267–285
Ekinci YL, Yiğitbaş E (2015) Interpretation of gravity anomalies to delineate some structural
features of Biga and Gelibolu peninsulas, and their surroundings (north-west Turkey). Geodin
Acta 27(4):300–319
Ekinci YL, Balkaya Ç, Göktürkler G, Turan S (2016) Model parameter estimations from residual
gravity anomalies due to simple-shaped sources using differential evolution algorithm. J Appl
Geophys 129:133–147
Ekinci YL, Özyalın Ş, Sındırgı P, Balkaya G, Göktürkler G (2017) Amplitude inversion of 2D
analytic signal of magnetic anomalies through differential evolution algorithm. J Geophys Eng
14:1492–1508
El-Kaliouby HM, Al-Garni MA (2009) Inversion of self-potential anomalies caused by 2D
inclined sheets using neural networks. J Geophys Eng 6:29–34
Erginal AE, Ekinci YL, Demirci A, Bozcu M, Öztürk MZ, Avcıoğlu M, Öztura E (2013) First
record of beachrock on Black Sea coast of Turkey: implications for Late Holocene sea-level
fluctuations. Sediment Geol 294:294–302
Essa KS (2014) New fast least-squares algorithm for estimating the best-fitting parameters due to
simple geometric-structures from gravity anomalies. J Adv Res 5:57–65
Essa KS, Elhussein M (2018) PSO (particle swarm optimization) for interpretation of magnetic
anomalies caused by simple geometric structures. Pure Appl Geophys. https://doi.org/10.1007/
s00024-018-1867-0
Fernández-Martínez JL, García-Gonzalo E, Fernández Álvarez JP, Kuzma HA, Menéndez Pérez
CO (2010) PSO: a powerful algorithm to solve geophysical inverse problems. Application to a
1D-DC resistivity case. J Appl Geophys 71:13–25
Fogel LJ, Owens AJ, Walsh AJ (1966) Artificial intelligence through simulated evolution. Wiley,
New York
Gallardo LA, Meju MA (2004) Joint two-dimensional dc resistivity and seismic travel-time
inversion with cross-gradients constraints. J Geophys Res 109:B03311
Gholami A, Siahkoohi HR (2010) Regularization of linear and non-linear geophysical ill-posed
problems with joint sparsity constraints. Geophys J Int 180:871–882
Göktürkler G (2011) A hybrid approach for tomographic inversion of crosshole seismic
first-arrival times. J Geophys Eng 8:99–108
Göktürkler G, Balkaya Ç (2012) Inversion of self-potential anomalies caused by simple geometry
bodies using global optimization algorithms. J Geophys Eng 9:498–507
Grasmueck M, Weger R, Horstmeyer H (2004) Three-dimensional ground-penetrating radar
imaging of sedimentary structures, fractures, and archaeological features at submeter
resolution. Geology 32:933–936
186 Y. L. Ekinci et al.
Haber E, Horesh L, Tenorio L (2008) Numerical methods for experimental design of large-scale
linear ill-posed inverse problems. Inverse Prob 24:055012
Holland JH (1975) Adaptation in natural and artificial systems. MIT Press, Cambridge
Jagannadha Rao S, Rama Rao P, Radhakrishna Murthy IV (1993) Automatic inversion of
self-potential anomalies of sheet-like bodies. Comput Geosci 19:61–73
Kaftan İ (2017) Interpretation of magnetic anomalies using a genetic algorithm. Acta Geophys
65:627–634
Kennedy J, Eberhart R (1995) Particle swarm optimization. IEEE Int Conf Neural Netw 4:1942–
1948
Kosaroglu S, Buyuksarac A, Aydemir A (2016) Modelling of shallow structures in the Cappadocia
region using gravity and aeromagnetic anomalies. J Afr Earth Sci 124:214–226
Lang J, Sievers J, Loewer M, Igel J, Winsemann (2017) 3D architecture of cyclic-step and
antidune deposits in glacigenic subaqueous fan and delta settings: integrating outcrop and
ground-penetrating radar data. Sediment Geol 362:83–100
Li Y, Oldenburg DW (1996) 3-D inversion of magnetic data. Geophysics 61:394–408
Li Y, Oldenburg DW (2003) Fast inversion of large-scale magnetic data using wavelet transforms
and a logarithmic barrier method. Geophys J Int 152:251–265
Liu B (2014) Composite differential search algorithm. J Appl Math 2014:1–15
Liu S, Hu X, Liu T (2014) A stochastic inversion method for potential field data: ant colony
optimization. Pure Appl Geophys 171:1531–1555
Liu S, Hu X, Liu T, Xi Y, Cai Zhang H (2015) Ant colony optimisation inversion of surface and
borehole magnetic data under lithological constraints. J Appl Geophys 112:115–128
Luke S (2009) Essentials of metaheuristics (Lulu), p 233. http://cs.gmu.edu/*sean/book/
metaheuristics/
Maiti S, Gupta G, Erram VC, Tiwari RK (2011) Inversion of Schlumberger resistivity sounding
data from the critically dynamic Koyna region using the hybrid Monte Carlo-based neural
network approach. Nonlinear Proc Geophys 18:179–192
Mehanee SA (2014) Accurate and efficient regularized inversion approach for the interpretation of
isolated gravity anomalies. Pure Appl Geophys 171:1897–1937
Mehanee S, Essa KS (2015) 2.5D regularized inversion for the interpretation of residual gravity
data by a dipping thin sheet: numerical examples and case studies with an insight on sensitivity
and non-uniqueness. Earth Planets Space 67:130
Mehanee S, Essa KS, Smith PD (2011) A rapid technique for estimating the depth and width of a
two-dimensional plate from self-potential data. J Geophys Eng 8:447–456
Meju MA (1994) Geophysical data analysis: understanding inverse problem theory and practice:
SEG Course notes series, vol 6. Society of Exploration Geophysicists, Tulsa, 296 p
Menke W (1989) Geophysical data analysis—discrete inverse theory. Academic Press, San Diego
Murthy BVS, Haricharan P (1985) Nomograms for the complete interpretation of spontaneous
potential profiles over sheet like and cylindrical 2D structures. Geophysics 50:1127–1135
Ogunbo JN (2018) MATLAB code for data-driven initial model of 1D Schlumberger sounding
curve. Geophysics 83:F21–F28
Oruç B, Gomez-Ortiz D, Petit C (2017) Lithospheric flexural strength and effective elastic
thickness of the Eastern Anatolian (Turkey) and surrounding region. J Asian Earth Sci 150:1–
13
Pallero JLG, Fernandez-Martinez JL, Bonvalot S, Fudym O (2015) Gravity inversion and
uncertainty assessment of basement relief via particle swarm optimization. J Appl Geophys
116:180–191
Pallero JLG, Fernandez-Martinez JL, Bonvalot S, Fudym O (2017) 3D gravity inversion and
uncertainty assessment of basement relief via particle swarm optimization. J Appl Geophys
139:338–350
Paul MK (1965) Direct interpretation of self-potential anomalies caused by inclined sheets of
infinite horizontal extensions. Geophysics 30:418–423
Pekşen E, Yas T, Kayman AY, Özkan C (2011) Application of particle swarm optimization on
self-potential data. J Appl Geophys 75:305–318
7 Global Optimization of Near-Surface Potential … 187
Pekşen E, Yas T, Kıyak A (2014) 1-D DC resistivity modeling and interpretation in anisotropic
media using particle swarm optimization. Pure Appl Geophys 171:2371–2389
Poli R, Kennedy J, Blackwell T (2007) Particle swarm optimization: an overview. Swarm Intell
1:33–57
Portniaguine O, Zhdanov MS (2002) 3-Dmagnetic inversion with data compression and image
focusing. Geophysics 67:1532–1541
Qureshi IP, Nalaye AM (1978) A method for direct interpretation of magnetic anomalies caused by
2-D vertical faults. Geophysics 43:179–188
Radhakrishna Murthy IV (1998) Gravity and magnetic interpretation in exploration geophysics.
Geological Society of India, Bangalore, India
Radhakrishna Murthy IV, Swamy KV, Jagannadha Rao S (2001) Automatic inversion of magnetic
anomalies of faults. Comput Geosci 27:315–325
Radhakrishna Murthy IV, Sudhakar KS, Rama Rao P (2005) A new method of interpreting
selfpotential anomalies of two-dimensional inclined sheets. Comput Geosci 31:661–665
Rao BSR, Radhakrishna Murthy IV, Jeevananda Reddy S (1970) Interpretation of self-potential
anomalies of some simple geometric bodies. Pure Appl Geophys 78:66–77
Rechenberg I (1973) Evolutionsstrategie: Optimierung technischer Systeme nach Prinzipien der
biologischen Evolution. Frommann-Holzboog, Stuttgart
Robinson ES, Coruh C (1988) Basic Exploration Geophysics. Wiley, New York, NY
Roy L (2001) Short note: source geometry identification by simultaneous use of structural index
and shape factor. Geophys Prospect 49:159–164
Roy IG (2002) A robust descent type algorithm for geophysical inversion through adaptive
regularization. Appl Math Mod 26:619–634
Rucker DF, Glaser DR, Osborne T, Maehl C (2009) Electrical resistivity characterization of a
reclaimed gold mine to delineate acid rock drainage pathways. Mine Wat Environ 29:149–157
Salem A, Elawadib E, Ushijima K (2003) Depth determination from residual gravity anomaly data
using a simple formula. Comput Geosci 29:801–804
Salmon S (2011) Particle swarm optimization in Scilab. http://forge.scilab.org/index.php/p/pso-
toolbox/downloads/
Santos FAM (2010) Inversion of self-potential of idealized bodies’ anomalies using particle swarm
optimization. Comput Geosci 36:1185–1190
Sari C, Şalk M (2006) Sediment thicknesses of the western Anatolia graben structures determined
by 2D and 3D analysis using gravity data. J Afr Earth Sci 26:39–48
Schwefel HP (1977) Numerische Optimierung von Computer-Model len mittels der
Evolutionsstrategie. In: Interdisciplinary systems research, vol 26. Birkhauser, Basel
Sen M, Stoffa PL (1995) Global optimization methods in geophysical inversion. In: Advances in
exploration geophysics, vol 4. Elsevier, Amsterdam
Shan C, Bastani M, Malehmir A, Persson L, Engdahl M (2014) Integrated 2D modelling and
interpretation of geophysical and geotechnical data to delineate quick clays at a landslide site in
southwest Sweden. Geophysics 79:EN61–EN71
Sharma SP, Biswas A (2013) Interpretation of self-potential anomaly over a 2D inclined structure
using very fast simulated-annealing global optimization—an insight about ambiguity.
Geophysics 78:WB3–WB15
Shi Y, Eberhart RC (1998) Parameter selection in particle swarm optimization. In: Proceedings of
the 7th international conference on evolutionary programming VII, New York, pp 591–600
Singh A, Biswas A (2016) Application of global particle swarm optimization for inversion of
residual gravity anomalies over geological bodies with idealized geometries. Nat Resour Res
25:297–314
Srivastava S, Agarwal BNP (2010) Inversion of the amplitude of the two-dimensional analytic
signal of magnetic anomaly by the particle swarm optimization technique. Geophys J Int
182:652–662
Storn R (1996) On the usage of differential evolution for function optimization. In: 1996 Biennial
Conference North American fuzzy information processing society, Berkeley, CA, pp 519–523
188 Y. L. Ekinci et al.
Storn R, Price K (1995) Differential evolution—a simple and efficient adaptive scheme for global
optimization over continuous spaces. Technical Report TR-95-012. International Computer
Science Institute, Berkeley, USA
Storn R, Price K (1997) Differential evolution—a simple and efficient heuristic for global
optimization over continuous spaces. J Glob Optim 11:341–359
Sundararajan N, Srinivasa Rao P, Sunitha V (1998) An analytical method to interpret selfpotential
anomalies caused by 2D inclined sheets. Geophysics 63:1551–1555
Sweilam NH, El-Metwally K, Abdelazeem M (2007) Self potential signal inversion to simple
polarized bodies using the particle swarm optimization method: A visibility study. J Appl
Geophys 6:195–208
Tarantola A (2005) Inverse problem theory and methods for model parameter estimation. Society
of Industrial and Applied Mathematics, Philadelphia
Toushmalani R (2013) Comparision result of inversion of gravity data of a fault by particle swarm
optimization and Levenberg-Marquardt methods. SpringerPlus 2:462
Ulugergerli EU (2011) Two-dimensional combined inversion of short- and long-normal dc
resistivity well log data. J Appl Geophys 73:130–138
Vasant P (2012) Meta-heuristics optimization algorithms in engineering, business, economics and
finance. IGI Publishing, Hersley, 734 p
Weise T (2009) Global optimization algorithms theory and applications. University of Kassel,
http://www.it-weise.de/
Zhdanov MS (2002) Geophysical inverse theory and regularization problems. Elsevier,
Amsterdam, 609 p
Chapter 8
Global Optimization of Model
Parameters from the 2-D Analytic Signal
of Gravity and Magnetic Anomalies
Over Geo-Bodies with Idealized
Structure
West) with the presence of sulfide bodies buried at different depths and two from
Turkey (Kesikköprü and Marmara anomaly) related to iron ore deposit and deep
seated magnetic deposit.
8.1 Introduction
Gravity and magnetic surveys are consistently applied in various purposes such as
for regional geological studies, subsurface crustal studies as well as for exploration
purposes. Moreover, both the method has also been used in various other appli-
cations such as hydrological, geodetic analysis, seismology and other subsurface
modeling (Telford et al. 1990; Reynolds 1997). The method has been effectively
used from mineral exploration and mining studies (Mandal et al. 2013, 2015;
Biswas et al. 2014).
The quantitative interpretation for gravity and magnetic anomalies was discussed
by Nabighian et al. 2005a, b). The quantitative interpretation of gravity and mag-
netic anomalies commonly includes estimations of source locations (depth and
exact origin), shape factor (source geometry) and amplitude (Parasnis 1986;
Srivastava and Agarwal 2010; Ekinci et al. 2017). The general methods to interpret
the data is to execute a competent inversion approach. Moreover, it is well-known
that all geophysical data processing and interpretation is quite difficult because of
the problem of non-uniqueness, non-linearity and ill-posed nature. In case of linear
problems, the presence of smaller number of identified quantities than the number
of unidentified geological model parameters produces problems in the interpretation
which is regarded as the ambiguity in the interpretation and is intrinsic in all
geophysical problems (Roy 1962; Srivastava and Agarwal 2010). Thus, all geo-
physical inversion problem mostly requires some constraints and a priori infor-
mation so as to recover the interpretable and accurate model parameter solutions
relating the geological source (Li and Oldenburg 1996, 2003; Srivastava and
Agarwal 2010; Ekinci et al. 2017).
Some of the recently developed methods for the interpretation of gravity and/or
magnetic data are euler deconvolution (Thompson 1982; Salem and Ravat 2003;
Agarwal and Srivastava 2008, 2009), hilbert transforms (Mohan et al. 1982), walsh
transformation (Shaw and Agarwal 1990), edge approximation (Ekinci et al. 2013),
regularized inversion (Mehanee 2014; Mehanee and Essa 2015); simplex algorithm
(Tlas and Asfahani 2015), finite element and derivative based algorithm (Ekinci and
Yiğitbaş 2015), very fast simulated annealing (Biswas et al. 2017; Biswas 2016a, b;
Biswas 2015), particle swarm optimization (Singh and Biswas 2016) and differ-
ential evolution (Ekinci et al. 2016; Balkaya et al. 2017). Various other interpre-
tation methods for both gravity and/or magnetic anomalies has been developed
considering the idealized structure (Roy et al. 2000; Essa 2007; Biswas 2015,
8 Global Optimization of Model Parameters from the 2-D Analytic … 191
2016a; Ekinci et al. 2016; Singh and Biswas 2016; Abdelrahman and Essa 2015;
Abdelrahman et al. 2012, 2009; Asfahani and Tlas 2004, 2007, 2012; Tlas and
Asfahani 2011a, b; Tlas et al. 2005; Abo-Ezz and Essa 2016; Ekinci 2016).
However, in most of the interpretation either horizontal or the vertical gradient/
derivative has been taken to interpret the data.
It is noteworthy to mention that there are a number of techniques which can
invert the 3-D analytic signal amplitude (ASA) over presumed source geometry
(Keating and Sailhac 2004; Chen et al. 2009; Srivastava and Agarwal 2010; Ekinci
et al. 2017). However, there are problems associated with the 3-D ASA and some
hypothesis typically causes to acquire unstable as well as unrealistic model
parameters (see Ekinci et al. 2017). To overcome those problems, 2-D ASA of the
observed gravity and magnetic anomalies can be applied (Srivastava and Agarwal
2010). The intricacies do not occur in the 2-D ASA of gravity and magnetic
anomalies (Nabighian 1972). Moreover, a limited inversion was carried out over
idealized models because of the complications associated in mathematical
formulation.
In the present study, 2-D ASA or Total Gradient (TG) derived from the square
root of the sum of the squares of both horizontal and vertical derivatives of the
observed gravity and magnetic anomaly has been taken. This TG is equal to the
amplitude of the analytical signals (Nabighian et al. 2005a, b; Srivastava et al. 2014,
Ekinci et al. 2017). We study the expression of analytic signal for 2-D causative
sources as established by Nabighian (1972). The AAS over 2-D sources is a regular
function with crests nearly concurring with the upper bends of the causative source
(s). It is worth mentioning that a few techniques such as PSO, ACO has been used
to invert the 2-D ASA. Hence, we try to visualize the viability of a global inversion
process from 2-D ASA over 2-D sources to compute model parameters and detect
the source geometry. The main objective of the current work is to expand a com-
bined method for quantitative interpretation of gravity and magnetic anomalies over
various idealized geo-bodies which will fulfill the Laplace’s equation and further
extend the work taking help of different subsurface geo-bodies with single and
multiple source bodies (lacking in almost all interpretation methods) using Very
Fast Simulated Annealing (VFSA) global optimization method. The VFSA method
is capable of escaping neighborhood minima by playing out a stochastic search
inside the model space. It does not require all around developed of initial model
giving a hard and flexible search space without discussing the resolution (Sen and
Stoffa 2013; Sharma 2012; Sharma and Biswas 2011, 2013a, b; Biswas and Sharma
2015, 2016, Biswas 2016b). The method has been applied in interpreting the
gravity and magnetic anomalies taking noise free and noise corrupted synthetic data
and two well-known field examples from Canada and two recent data from Turkey.
The present method has also been compared with other optimization methods such
as Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO) and
Differential Evolution (DE) for four field examples in terms of its ability, consis-
tency and the error in the final interpretation.
192 S. Trivedi et al.
8.2 Methodology
The expression for the ASA or TG of potential field anomaly can be expressed in
terms of different idealized geo-bodies (Sphere, Cylinder, Fault/Contact) (Fig. 8.1)
after Nettleton (1971), Nabighian (1972), Srivastava and Agarwal (2010),
Srivastava et al. (2014) and Ekinci et al. (2017) is given as
2 3
6 1
iq 7
ASA=TGð xÞ ¼ k 4h 5 ð8:1Þ
2 2
ð x x0 Þ þ ð z Þ
where, k is the amplitude coefficient, z is the depth from the surface to the centre
of the sphere and horizontal cylinder and top of the body for Fault/contact, x0
(i = 1, …, N) is the coordinate of the horizontal position on the surface, q is the
shape factor. The shape factors (q) for, Sphere, Horizontal/Vertical Cylinder and
thin dyke/Fault or Contact are 1.5, 1.0, and 0.5 for Gravity anomaly (Srivastava
et al. 2014) and shape factors (q) for, Sphere, Horizontal/Vertical Cylinder, thin
dyke and Fault or Contact are 2.0, 1.5, 1.0 and 0.5 for magnetic anomaly respec-
tively (Srivastava et al. 2014). Nabighian (1972) first introduced the concept of
Analytical Signals (AS) derived from the horizontal and vertical gradient of mag-
netic anomaly which is also called as Total Gradient (TG) by Srivastava et al.
(2014). A comprehensive derivation for the forward problem can be found in
various literatures (Nettleton 1971; Nabighian 1972; Nabighian et al. 2005a, b;
Srivastava et al. 2014; Biswas et al. 2017).
Equation 8.1 can be used to interpret single structure. Moreover, observed field
data are always linked with multiple structures and to interpret such data, the above
equation can also be rewritten as (after Biswas and Sharma 2014a)
X
M
V ðxi Þ ¼ Vj ðxi Þ ð8:2Þ
j¼1
where Vj(xi) is either gravity or magnetic anomaly at xi location for jth body and
M is the number of bodies.
Fig. 8.1 A diagram showing cross-sectional views, geometries and parameters for sphere, cylinder and fault/contact-type structure
193
194 S. Trivedi et al.
2 3
6 z
oq 7
gðxÞ ¼ k4n 5 ð8:3Þ
2 2
ð x x0 Þ þ ð z Þ
where, q = 1.5 (Sphere), 1(Horizontal Cylinder) and 0.5 (Vertical Cylinder) and
2
k ¼ 43 pGrR3 for q = 1.5; k = 2 pGrR2 for q = 1 and k ¼ pGzrR for q = 0.5. k is
the amplitude coefficient, z is the depth from the surface to the center of the body
(sphere or horizontal cylinder) or the depth from the surface to the top (vertical
cylinder), q is the geometric shape factor, x0 is the horizontal position coordinate, r
is the density contrast between the source and the host rock, G is the universal
gravitational constant, and R is the radius of the buried structure.
The general expressions for magnetic anomaly m(x) for sphere, horizontal
cylinder, thin dyke and thin sheet can be written as (see Biswas 2018 and reference
therein):
2 3
6 2z n sin h 3zxi cos h7
2
x2i
mð xÞ ¼ kz3 4 oq 5 ð8:4Þ
ðxi Þ2 þ ðzÞ2
2 3
2
6 z nxi cos h þ 2zx i sin h7
2
m ð xÞ ¼ k 4 oq 5 ð8:5Þ
2 2
ð xi Þ þ ð z Þ
2 3
6xi sin h þ z cosoh7
mð xÞ ¼ kz4n q5 ð8:6Þ
ð xi Þ 2 þ ð z Þ 2
2 3
6z cos h xi sinoh7
mð xÞ ¼ k4 n q5 ð8:7Þ
ð xi Þ 2 þ ð z Þ 2
In all the above equations k is the amplitude coefficient, z is the depth from the
surface to the center of the body (sphere, cylinder) and depth from the surface to the
top of the body (dyke and thin sheet), xi (i = 1, …, N) is the horizontal position
coordinate, h is the effective magnetization angle or the index parameter and q is the
shape factor. The shape factor for sphere is 2.5, horizontal cylinder 2, and thin dyke
and thin sheet 1.
8 Global Optimization of Model Parameters from the 2-D Analytic … 195
Simulated Annealing (SA) comes from the classic Boltzmann annealing method
(Kirkpatrick et al. 1983). The method was widely applied in many geophysical
optimization problems (Sen and Stoffa 2013; Sharma and Biswas 2011, 2013a, b;
Biswas 2015; Biswas and Sharma 2014a, b, 2015). In the present study, a modi-
fication of SA termed as Very Fast Simulated Annealing (VFSA) (Ingber and Rosen
1992), is applied to interpret the potential field data. The main feature of SA is that
it offers a technique to escape local optima by permitting a hill-climbing moves in
search of finding a global optimum solution. VFSA has a quicker cooling schedule
due to its sharper Cauchy probability distribution and have the capacity to take any
number/value in the model space. It has also better resolution and needs very small
memory (Sen and Stoffa 2013; Biswas and Sharma 2015; Biswas et al. 2017). To
minimize the objective function, the following misfit or the objective function (u) is
used to interpret the potential field data (Sharma and Biswas 2013a).
!2
1X N
V0 Vci
u¼ 0 i0 ð8:8Þ
N i¼1 Vi þ Vmax V0min =2
where, N is number of data point, Vi0 and Vic are the ith observed and model
0 0
responses and Vmax and Vmin are the maximum and minimum values of the observed
response respectively. The above misfit function is chosen so as to negate the data
near zero crossing which affects (increases) the objective function (Biswas 2018).
In the present study, different optimization parameters, for example Initial
temperature was taken as 1.0, cooling schedule as 0.4, number of iterations as 2000
and number of moves per temperature as 50. These parameters were taken from the
previous studies based on VFSA (Sharma and Biswas 2013a, b). For finding global
optimum solution, Probability Density Function (PDF) within one standard devi-
ation limit and uncertainty analysis, was carried out based on the procedures
established by Mosegaard and Tarantola (1995), Sen and Stoffa (1996). The
explanation of PDF and uncertainty analysis is well documented in various liter-
atures (Sen and Stoffa 2013; Sharma and Biswas 2013a, b) and are not discussed
here for brevity. For the present interpretation of gravity and magnetic field data, the
basic computation was carried out in a Window 7 environment by means of
MS FORTRAN Developer studio on a desktop PC with Intel Core i7 processor.
The computation time for the study is 40 s (not CPU) for a single structure and it
increases to another 40% for successive multiple structures. A comprehensive
review of VFSA method can be seen in various literatures published till date (Sen
and Stoffa 2013; Sharma 2012; Biswas 2015).
196 S. Trivedi et al.
It is well known that the parameters such as horizontal location of the source body
can be guessed from the peak anomaly (negative or positive), depth can be pre-
dicted from the half width of the anomaly and the amplitude coefficient from the
peak amplitude value of the anomaly as well as the depth (Nabighian 1972) for the
qualitative interpretation of the potential field from TG. It was observed in various
literatures that the horizontal location (x0) and shape factor (q) are the maximum
stable parameters in finding out the utmost dependable results (Biswas et al. 2017;
Srivastava and Agarwal 2010). Hence, to find out the best solution for every
parameter, firstly the search range for all the parameters were put in large range.
Next, after looking at the value of every model parameter are within the search
range, it is reduced to find the most probable solution with minimum uncertainty.
8.3.1.2 Gravity
For gravity data, initially three types of structures (Sphere, Cylinder and Fault/
Contact) were selected for theoretical modeling and inversion. Using Eq. 8.1,
synthetic data were first generated for different structures. Next, inversion was
carried out for each and every models and final interpreted model parameters (k, x0,
z, and q) were studied. As discussed above, the search spaces were selected and
inversion was repeated again and again to reduce the ambiguity. The convergences
of each model parameter were simultaneously studied for every inversion. Next, to
get the mean model, 10 inversion runs were performed and histogram as well as
convergences were studied where the misfit remains lower than 10−4. Moreover, a
statistical mean model was also calculated using the lower misfits which exist
within one standard deviation.
Inversion of sphere like structure was first executed and the whole inversion
procedure was applied as discussed above. Figure 8.2a shows the histogram of
model parameters (k, x0, and z) and the convergence pattern for noise free data. The
fits amongst the observed and model response are presented in Fig. 8.5a.
Interpreted mean model parameters are shown in Table 8.1. It is important to
mention that using Eq. 8.3, we have compared the TG anomaly with the normal
gravity anomaly (GA). This exercise was carried out to check how TG is different
from normal GA. A comparison of TG and GA is shown in the Fig. 8.2a and also
the comparison of model parameters in Table 8.1.
Next, the data for cylinder type structure was taken and inversion procedure was
repeated again. Histogram and convergence of model parameters were shown in
Fig. 8.3a. Observed and model responses are shown in Fig. 8.5b and the mean
8 Global Optimization of Model Parameters from the 2-D Analytic …
Fig. 8.2 Histogram and convergence for noise free and noisy data for sphere-type structure for gravity data
197
198
Table 8.1 Actual value of model parameters, search space and interpreted mean model for noise free, random noise with misfit for gravity data
Model parameters Actual value Search space Noise free (TG) Noise free (GA) Noisy (TG) Noisy (GA)
Sphere
k (mGal/m) 8000 0–10,000 8000.6 ± 37.7 8001.8 ± 29.6 7516.4 ± 112.6 8137.4 ± 92.5
x0 (m) 250 0–500 250.0 ± 0.1 250.0 ± 0.1 250.0 ± 0.1 249.6 ± 0.4
z (m) 50 0–100 50.0 ± 0.1 50.0 ± 0.1 48.6 ± 0.3 50.8 ± 0.5
q 1.5 0–2 1.5 ± 0.0 1.5 ± 0.0 1.5 ± 0.0 1.5 ± 0.0
Misfit 6.5 10−7 9.5 10−7 2.1 10−4 2.9 10−4
Model parameters Actual value Search space Noise free Noisy
Cylinder
k (mGal/m) 500 0–1000 499.8 ± 1.8 500.7 ± 5.6
x0 (m) 200 0–500 200.0 ± 0.0 200.0 ± 0.1
z (m) 60 0–100 60.0 ± 0.1 60.2 ± 0.5
q 1.0 0–2 1.0 ± 0.0 1.0 ± 0.0
Misfit 1.1 10−7 5.3 10−4
Fault/contact
k (mGal/m) 200 0–500 199.9 ± 0.4 202.5 ± 1.9
x0 (m) 300 0–500 300.0 ± 0.0 300.0 ± 0.1
z (m) 20 0–50 20.0 ± 0.0 22.0 ± 0.4
q 0.5 0–2 0.5 ± 0.0 0.5 ± 0.0
Misfit 5.9 10−10 1.7 10−4
S. Trivedi et al.
8 Global Optimization of Model Parameters from the 2-D Analytic …
Fig. 8.3 Histogram and convergence for noise free and noisy data for cylinder-type structure for gravity data
199
200 S. Trivedi et al.
model parameters are shown in Table 8.1. Again, the synthetic data for fault or the
contact type structure were taken and inversion was repeated. Histogram, conver-
gence and the fittings between observed and model responses are shown in
Figs. 8.4a and 8.5c. Interpreted mean model are shown in Table 8.1. For brevity,
the comparison of GA is not shown as it can be understood from the sphere model.
Analysis of histogram demonstrates that all the model parameters of the ideal-
ized geo-bodies can be interpreted after inversion. Moreover, it has been established
that there is a very less error in the interpretation of the model parameters from the
convergence pattern as well as from the interpreted mean model parameters from
Table 8.1.
8.3.1.3 Magnetic
For magnetic data interpretation, as discussed for gravity data, three different
structures were taken for theoretical modeling and inversion. Inversion was repe-
ated for magnetic data as discussed in gravity data. Histogram of model parameters
(k, x0, and z) and the convergence pattern for noise free data are shown in Fig. 8.6a.
The fits amongst the observed and model response are illustrated in Fig. 8.9a.
Interpreted mean model parameters for magnetic data are presented in Table 8.2.
For cylinder type structure, inversion procedure was repeated. Histogram and
convergence of model parameters were displayed in Fig. 8.7a. Observed and model
responses are shown in Fig. 8.9b and the mean model parameters are presented in
Table 8.2. Synthetic data for fault type structure were taken and inversion was
carried out. Histogram, convergence and the fittings amongst observed and model
responses are presented in Figs. 8.8a and 8.9c. Interpreted mean model are pre-
sented in Table 8.2. As discussed for gravity data, analysis of histogram, conver-
gence also shows that the present inversion method was competent enough to find
out a very accurate solution for magnetic data.
It is well recognized that the actual field data are always corrupted with noise and it
is very difficult to understand what degree of noise is added in such data. Hence, to
delineate the efficiency of the method, all the synthetic noise free data were cor-
rupted with 10% random noise (multiplied by a random draw between 1 and 1.10)
and the inversion was repeated for all kind of structures for both gravity and
magnetic data.
Figures 8.2b, 8.3b and 8.4b show the histogram and convergence for noise
corrupted sphere, cylinder and fault like structure for gravity anomaly. Table 8.1
displays the interpreted mean model for noise corrupted data and the fits amongst
the observed and model response for noise corrupted model is shown in Fig. 8.5d–
f. For magnetic anomaly, the histogram and convergence were plotted and shown in
Figs. 8.6b, 8.7b and 8.8b. Table 8.2 show the interpreted mean model for noise
8 Global Optimization of Model Parameters from the 2-D Analytic …
Fig. 8.4 Histogram and convergence for noise free and noisy data for fault/contact-type structure for gravity data
201
202 S. Trivedi et al.
Fig. 8.5 Fittings between the observed and Model responses for sphere (a, d), cylinder (b, e) and
fault/contact (c, f)-type structure for noise free and noisy synthetic data for gravity data
corrupted data and the fits amongst the observed and model response are shown in
Fig. 8.9d–f.
A 3D volume space cross-plot analysis was also carried out to see how different
models get close to the true value. Figure 8.10a show the cross plots between k,
z and x0. It can be seen that constraining the x0 to its actual value gives other
parameters close to their true value. Figure 8.10b shows that the cross plots
between k, z and q. Moreover, it can also be seen that constraining q to its real value
also gives the other parameters points close to their actual value. Analysis of
cross-plots demonstrates the model parameters are very close to their actual value
8 Global Optimization of Model Parameters from the 2-D Analytic …
Fig. 8.6 Histogram and convergence for noise free and noisy data for sphere-type structure for magnetic data
203
204
Fig. 8.7 Histogram and convergence for noise free and noisy data for cylinder-type structure for magnetic data
S. Trivedi et al.
8 Global Optimization of Model Parameters from the 2-D Analytic …
Fig. 8.8 Histogram and convergence for noise free and noisy data for fault/contact-type structure for magnetic data
205
206 S. Trivedi et al.
Fig. 8.9 Fittings between the observed and model responses for sphere (a, d), cylinder (b, e) and
fault/contact (c, f)-type structure for noise free and noisy synthetic data for magnetic data
(Yellow) and the final model parameters were inside the uncertainty limits (one
standard deviation) and with high PDF (Red). Figure 8.10c and d demonstrate the
same taking the noisy synthetic data.
It is highlighted from the analysis of histogram, convergence and cross-plots, the
main model parameters (k and z) are interpreted precisely when the other two
parameters are constrained to its actual value. It can be clearly visualized that the
estimated model parameters for the entire structures are within the uncertainty limits
8 Global Optimization of Model Parameters from the 2-D Analytic … 207
Table 8.2 Actual value of model parameters, search space and interpreted mean model for noise
free, random noise with misfit for magnetic data
Model parameters Actual value Search space Noise free Noisy
Sphere
k (nT/m) 5000 0–10,000 4999.9 ± 54.9 6099.5 ± 446.3
x0 (m) 250 0–500 250.0 ± 0.0 250.3 ± 0.2
z (m) 5 0–10 5.0 ± 0.0 5.4 ± 0.2
q 2.0 0–3 2.0 ± 0.0 2.0 ± 0.0
Misfit 8.3 10−11 2.0 10−5
Cylinder
k (nT/m) 600 0–1000 599.9 ± 3.9 626.8 ± 13.5
x0 (m) 200 0–500 200.0 ± 0.0 200.2 ± 0.2
z (m) 10 0–20 10.0 ± 0.0 10.2 ± 0.2
q 1.5 0–2 1.5 ± 0.0 1.5 ± 0.0
Misfit 9.2 10−9 1.5 10−5
Fault/contact
k (nT/m) 300 0–500 299.9 ± 0.6 298.6 ± 3.0
x0 (m) 300 0–500 300.0 ± 0.0 300.8 ± 0.2
z (m) 15 0–20 15.0 ± 0.0 16.6 ± 0.3
q 0.5 0–2 0.5 ± 0.0 0.5 ± 0.0
Misfit 1.5 10−9 1.2 10−4
(one standard deviation from the mean) and inside high PDF region. This suggest
that the inversion approach established for this study can precisely determine every
single model parameter even though the data are extremely corrupted with noise.
To demonstrate the effectiveness of the present approach four field examples from
total gradient of gravity and magnetic data were presented from published
literatures.
A residual gravity anomaly map of Noranda Mining District, Quebec, Canada was
taken (Grant and West 1965; Roy et al. 2000) above a massive base metal sulfide
deposit, mainly of pyrite ore body (Fig. 8.11a). The main residual gravity anomaly
shows a prominent peak anomaly (G#1). Interpretation for this data only
208 S. Trivedi et al.
Fig. 8.10 a 3D cross-plots between amplitude coefficient (k), depth (z), location (x0) for all
models having misfit < threshold (10−4 for noise-free data) (yellow), and models with PDF (one
standard deviation) (red) for noise free data; b 3D cross-plots between amplitude coefficient (k),
depth (z), shape factor (q) for all models having misfit < threshold (10−4 for noise-free data)
(yellow), and models with PDF (one standard deviation) (red) for noise free data; c 3D cross-plots
between amplitude coefficient (k), depth (z), location (x0) for all models having misfit < threshold
(10−2 for noisy) (yellow), and models with PDF (one standard deviation) (red) for noisy data; d 3D
cross-plots between amplitude coefficient (k), depth (z), shape factor (q) for all models having
misfit < threshold (10−2 for noisy) (yellow), and models with PDF (one standard deviation)
(red) for noisy data
considering single structures was carried out in various studies (Mehanee 2014) and
also in the present study. This anomaly was interpreted following Biswas (2015) to
retrieve the probable structure. Interpretation of this anomaly suggest a horizontal
cylinder type structure. Interpreted model parameters are presented in Table 8.3a.
TG was recalculated for the gravity data following Agarwal and Srivastava (2008).
It can be seen that there is only one peak (G#1) from the residual anomaly data,
8 Global Optimization of Model Parameters from the 2-D Analytic … 209
Fig. 8.11 Gravity field data: a fittings between the observed and model data and b subsurface
structure for Mobrun Massive Sulphide Body, Noranda, Quebec, Canada
210
Table 8.3 Mobrun Massive Sulphide Body, Noranda, Canada—(a) Search space and interpreted mean model for single anomaly for normal and TG gravity
anomaly and (b) for multiple bodies
Model parameters Search space Present method G#1 (TG/ASA) G#1 (GA)
(a) Single body
k (mGal/m) 0–1000 423.3 ± 2.9 795.9 ± 5.2
x0 (m) 290–320 301.9 ± 0.1 291.9 ± 0.3
z (m) 0–50 15.7 ± 0.1 48.7 ± 0.4
q 0–2 1.0 1.0
Misfit 7.7 10−3 7.2 10−3
Model Parameters Search Space Present method TG#1 Search Space Present method TG#2 Search Space Present method TG#3 Search Space Present method TG#4
(b) Multiple bodies
k (mGal/m) 0–100 37.2 ± 0.3 0–2000 1158.2 ± 21.6 0–500 310.0 ± 4.0 0–100 90.6 ± 1.4
x0 (m) 250–270 258.7 ± 0.1 280–300 286.7 ± 0.1 290–320 302.9 ± 0.1 320–330 321.5 ± 0.2
z (m) 0–50 9.9 ± 0.0 0–50 9.9 ± 0.1 0–20 13.2 ± 0.1 0–50 11.2 ± 0.1
q 0–2 1.0 0–2 1.5 0–2 1.0 0–2 1.0
Misfit 6.2 10−3 3.3 10−3 7.2 10−4 1.3 10−3
Search Space Peak 1 ACO/PSO Search Space Peak 2 ACO/PSO Search Space Peak 3 ACO/PSO Search Space Peak 4 ACO/PSO
Srivastava et al. (2014) (ACO/PSO)
k (mGal/m) 1–250 38/144 1–750 67/2 1–1125 100/103 1–500 44/421
x0 (m) 250–280 257/249 270–300 284/286 280–320 302/299 310–340 326/315
z (m) 10–100 31.8/27.9 10–100 29/36 10–100 39/23 10–100 33/11
b (=2q) 2–4 2.3/3.3 2–4 3.1/3.6 2–4 2.3/2.1 2–4 2.3/3.7
RMS ACO = 0.04, PSO = 0.21
S. Trivedi et al.
8 Global Optimization of Model Parameters from the 2-D Analytic … 211
however, for TG, there are four peaks (TG#1, TG#2, TG#3 and TG#4). TG
anomaly was delineated by Srivastava et al. (2014) using Ant Colony and Particle
Swarm Optimization. The anomaly was again interpreted using VFSA for both
single and multiple structures from the TG of gravity anomaly. The TG anomaly
displays a big magnitude peak anomaly with three other lesser anomalies. Here, the
central peak anomaly was interpreted as single structure. The depth interpreted for
single body was found to be 15.7 m and horizontal location of 301.9 m. The
structure was deduced as a cylindrical body. For multiple structures, the inversion
procedure was repeated for four dissimilar structures considering four peak values.
The depth obtained by the main peak anomaly was found to be 13.2 m and hori-
zontal location of 302.9 m. The structure was inferred to be a cylindrical body. It is
highlighted that the drilling results show a depth of around 30 m from the top of the
massive sulfide deposits (Grant and West 1965). However, in the present study, the
average depth from the top of the bodies are from 10 to 15 m. This inconsistency is
due to the fact that the drilling results were taken from the core mineralized
deposits; however, in general, due to inhomogeneous nature of the subsurface, the
presence of mineralized bodies might also be at a shallower depth which was
interpreted in the present study. Additionally, the location of the four bodies was
well interpreted in the present inversion. The error estimated for all the four bodies
were quite excellent and well comparable with other methods. The other interpreted
depth and horizontal locations and the interpreted structures along with the misfit
are shown in Table 8.3b along with a comparison of other works. It is also high-
lighted that the objective functions used here is different from Srivastava et al.
(2014) and hence, the errors were different. Figure 8.11a shows the fits amongst the
field data and the model responses for single as well as for the multiple structures
and Fig. 8.11b depicts the corresponding subsurface structures.
This magnetic anomaly was taken over the Kesikköprü iron ore deposit. The
deposit is located at the central Anatolian crystalline complex and it is the principal
iron ore reserve in turkey. The generalized geological map of this area can be seen
from Ekinci et al. (2017). The iron ore deposit is a skarn-type developed adjacent to
the granitoid contact (Oruç 2013). This anomaly was interpreted by Ekinci et al.
(2017) using DE algorithm. The TG anomaly was derived from the magnetic
anomaly following the method of Agarwal and Srivastava (2008). The derived TG
anomaly was interpreted using VFSA optimization. It was earlier reported that the
probable structure of this mineralized body is a spherical type structure and at a
larger depth of 42 m (Oruç 2013). However, in the present investigation, the
mineralized body interpreted to be a dike-like structure (q = 1.0) with a depth of
20.2 m. The depth interpreted in this study are in respectable agreement with that of
212 S. Trivedi et al.
Fig. 8.12 Magnetic field data: fitting between the observed and model data for a Kesikköprü,
Turkey and b Marmara, Turkey
the interpretation method developed by Ekinci et al. (2017). The fits between the
observed and mean model are presented in Fig. 8.12a. Table 8.4a summarises the
interpreted mean model and a comparison with other interpretation methods such as
NGF, EUL and DE following Ekinci et al. (2017).
8 Global Optimization of Model Parameters from the 2-D Analytic … 213
Table 8.4 (a) Search space and interpreted mean model for Kesikköprü anomaly and (b) Search
space and interpreted mean model for Marmara anomaly
VFSA Ekinci et al. (2017)
Model Search Present method M#1 DE EUL NGF
parameters space (TG/ASA)
(a) Kesikköprü anomaly
k (nT/m) 100–106 388,195.8 ± 4441.2 1,825,814.9 – –
x0 (m) 30–90 63.4 ± 0.2 62.9 63.16 60
z (m) 0–50 20.2 ± 0.2 24.4 15.87 22.5
q 0–2 1.0 1.19 – –
Misfit 4.2 10−3 – – –
VFSA Ekinci et al. (2017)
Model Search Present method M#1 (TG/ DE EUL NGF
parameters space ASA)
(b) Marmara anomaly
k (nT/km) 0–1000 4478.9 ± 37.9 1499 – –
x0 (km) 0–30 19.6 ± 0.0 19.97 20.56 19.5
z (km) 0–20 5.7 ± 0.0 3.87 3.52 3.58
q 0–2 1.0 0.79 – –
Misfit 3.0 10−3 – – –
different structure interpreted and complex geological setting where fractures were
developed in the fault plane. However, the fits amongst the observed and mean
model matches well and are shown in Fig. 8.12b. Table 8.4b summarises the
interpreted mean model and a comparison with other interpretation methods such as
NGF, EUL and DE, as given by Ekinci et al. (2017).
Fig. 8.13 Magnetic field data: a fittings between the observed and model data and b subsurface
structure for Barraute, North-West, Quebec, Canada
8 Global Optimization of Model Parameters from the 2-D Analytic … 215
Both gives a consistent result however, the fits for thin sheet type structure matches
better than thin dyke type structure and the misfit for thin sheet is less than thin
dyke. Hence, from the interpretation of magnetic anomaly, it can be said that the
anomalies were formed due to the presence of thin dyke to sheet type structure.
Interpreted model parameters are shown in Table 8.5a. Next, TG anomalies of
magnetic data was derived from the magnetic anomaly. TG anomaly shows four
(TM#1, TM#2, TM#3, TM#4) prominent anomalies. These anomalies were inter-
preted by Srivastava et al. (2014) using Ant Colony and Particle Swarm
Optimization. Initially, based on the magnetic anomaly (M#1 and M#2), only two
peak TG anomaly (TM#2 and TH#3) were interpreted. The depth acquired for
TM#2 was found to be 6.9 m and horizontal location of 36.9 m. The structure was
inferred as a thin dyke. The second anomaly TM#3 was found to be at a depth of
17.6 m with a horizontal location of 94.2 m. Next, considering four structures, the
inversion process was repeated again. The depths obtained by four peaks namely
TM#1, TM#2, TM#3, and TM#4 is 6.8 m, 5.8 m, 12.6 m and 7.9 m on a horizontal
location of 25.6 m, 37.4 m, 90.8 m and 105.4 m respectively. The first three
structures were interpreted as thin dyke or a combination of thin and thick dyke like
structure and the fourth one as a vertical fault or contact like structure. This
interpretation is in good agreement with the published drill-hole measurements
(Telford et al. 1990). The estimated error for the present study is comparable to
Srivastava et al. (2014). Srivastava et al. (2014) did not consider the first anomaly
for their interpretation, however, in the present study, TM#1 was also considered as
an anomaly generated from mineralization. However, the third anomaly considered
by Srivastava et al. (2014) was not considered here because of a very weak anomaly
which could associated with the main peak anomaly at TM#3. However, the
structure interpreted from the four anomalies was equivalent. The error estimated
for all the four bodies are quite excellent. The interpreted depth and horizontal
locations and the interpreted structures along with the misfit are shown in
Table 8.5b along with a comparison with other methods. It is worth to mention that
the objective functions used here is different from Srivastava et al. (2014).
Figure 8.13a shows the fittings amongst the TG anomaly of field magnetic data and
the model responses for two and four structures and Fig. 8.13b shows the corre-
sponding subsurface structures.
Table 8.5 Barraute, North-West, Canada—(a) Search space and interpreted mean model for normal magnetic anomaly and (b) Search space and interpreted
216
8.4 Conclusions
We have tried to assess the application and viability of VFSA on the appraisals of
model parameter from the analytic signal or the total gradient derived from the
horizontal and vertical gradient of potential field anomalies. The present study was
validated using noise free and noise corrupted synthetic data with four field
examples. The total gradient potential field data were interpreted in terms of
amplitude coefficient, location, depth and shape of a buried structure. The model
parameters were inferred using VFSA global optimization technique. The presents
study shows that optimizing all parameters gives reliable results where the location
and shape of the idealized structure gives very accurate estimates of the true value
of the model parameters. Hence, constraining the location and shape in the final
interpretation reduces the ambiguity in the final interpretation. Analysis of his-
togram, convergence as well as cross-plots reveal that the parameters were accu-
rately determined and very small errors were estimated for the present study.
Comparisons of results from other interpretation methods suggest that the present
method is as efficient as other methods such as PSO, ACO and DE. The present
method can be effectively applied to interpret numerous structures. The estimated
model parameters interpreted for field data were in respectable agreement with other
interpretation approaches in terms of near probable values and uncertainty as well
as from the drilling results.
Acknowledgements The authors would like to thank the reviewers for their suggestions and
constructive criticism which helped us to improve the quality of the manuscript. The authors are
also thankful to one reviewer for suggesting to interpret two new field data. This work is a Masters
Dissertation of ST who acknowledges the necessary facilities and support from the Head,
Department of Geology, and Director, Institute of Science, Banaras Hindu University, Varanasi.
References
Abdelrahman EM, Essa KS (2015) A new method for depth and shape determinations from
magnetic data. Pure Appl Geophys 172(2):439–460
Abdelrahman EM, Soliman KS, El-Araby TM, Abo-Ezz ER, Essa KS (2009) A least-squares
standard deviation method to interpret magnetic anomalies due to thin dikes. Near Surf
Geophys 7:41–46
Abdelrahman EM, Abo-Ezz ER, Essa KS (2012) Parametric inversion of residual magnetic
anomalies due to simple geometric bodies. Explor Geophys 43:178–189
Abo-Ezz ER, Essa KS (2016) A least-squares minimization approach for model parameters
estimate by using a new magnetic anomaly formula. Pure Appl Geophys 173:1265–1278
Agarwal BNP, Srivastava S (2008) FORTRAN codes to implement enhanced local wave number
technique to determine location, depth and shape of the causative source using magnetic
anomaly. Comput Geosci 34:1843–1849
Agarwal BNP, Srivastava S (2009) Analyses of self-potential anomalies by conventional and
extended Euler deconvolution techniques. Comput Geosci 35:2231–2238
Asfahani J, Tlas M (2004) Nonlinearly constrained optimization theory to interpret magnetic
anomalies due to vertical faults and thin dikes. Pure Appl Geophys 161:203–219
8 Global Optimization of Model Parameters from the 2-D Analytic … 219
Asfahani J, Tlas M (2007) A robust nonlinear inversion for the interpretation of magnetic
anomalies caused by faults, thin dikes and spheres like structure using stochastic algorithms.
Pure Appl Geophys 164:2023–2042
Asfahani J, Tlas M (2012) Fair function minimization for direct interpretation of residual gravity
anomaly profiles due to spheres and cylinders. Pure Appl Geophys 169:157–165
Ateş A, Büyüksaraç A, Bilim F, Bektaş Ö, Şendur Ç, Komanovali G (2009) Spatial correlation of
the aeromagnetic anomalies and seismogenic faults in the Marmara region, NW Turkey.
Tectonophysics 478:135–142
Balkaya Ç, Ekinci YL, Göktürkler G, Turan S (2017) 3D non-linear inversion of magnetic
anomalies caused by prismatic bodies using differential evolution algorithm. J Appl Geophys
136:372–386
Biswas A (2015) Interpretation of residual gravity anomaly caused by a simple shaped body using
very fast simulated annealing global optimization. Geosci Front 6(6):875–893
Biswas A (2016a) Interpretation of gravity and magnetic anomaly over thin sheet-type structure
using very fast simulated annealing global optimization technique. Model Earth Syst Environ 2
(1):30
Biswas A (2016b) A comparative performance of least square method and very fast simulated
annealing global optimization method for interpretation of self-potential anomaly over 2-D
inclined sheet type structure. J Geol Soc India 88(4):493–502
Biswas A (2018) Inversion of source parameters from magnetic anomalies for mineral/ore deposits
exploration using global optimization technique and analysis of uncertainty. Nat Resour Res 27
(1):77–107
Biswas A, Sharma SP (2014a) Resolution of multiple sheet-type structures in self-potential
measurement. J Earth Syst Sci 123(4):809–825
Biswas A, Sharma SP (2014b) Optimization of self-potential interpretation of 2-D inclined
sheet-type structures based on very fast simulated annealing and analysis of ambiguity. J Appl
Geophys 105:235–247
Biswas A, Sharma SP (2015) Interpretation of self-potential anomaly over idealized body and
analysis of ambiguity using very fast simulated annealing global optimization. Near Surf
Geophys 13(2):179–195
Biswas A, Sharma SP (2016) Integrated geophysical studies to elicit the structure associated with
Uranium mineralization around South Purulia Shear Zone, India: a review. Ore Geol Rev
72:1307–1326
Biswas A, Mandal A, Sharma SP, Mohanty WK (2014) Delineation of subsurface structure using
self-potential, gravity and resistivity surveys from South Purulia Shear Zone, India: implication
to uranium mineralization. Interpretation 2(2):T103–T110
Biswas A, Parija MP, Kumar S (2017) Global nonlinear optimization for the interpretation of
source parameters from total gradient of gravity and magnetic anomalies caused by thin dyke.
Ann Geophys 60(2):G0218, 1–17
Chen J, Chouteau M, Keating P (2009). Joint inversion of magnetic anomaly due to kimberlite
pipe and its analytic signal. In: 11th SAGA biennial technical meeting & exhibition,
Swaziland, pp 222–226
Ekinci YL (2016) MATLAB-based algorithm to estimate depths of isolated thin dike-like sources
using higher-order horizontal derivatives of magnetic anomalies. Springer Plus 5(1):1384
Ekinci YL, Yiğitbaş E (2015) Interpretation of gravity anomalies to delineate some structural
features of Biga and Gelibolu peninsulas, and their surroundings (north-west Turkey). Geodin
Acta 27(4):300–319
Ekinci YL, Ertekin C, Yiğitbaş E (2013) On the effectiveness of directional derivative based filters
on gravity anomalies for source edge approximation: synthetic simulations and a case study
from the Aegean graben system (western Anatolia, Turkey). J Geophys Eng 10:035005
Ekinci YL, Balkaya Ç, Gokturkler G, Turan S (2016) Model parameter estimations from residual
gravity anomalies due to simple-shaped sources using differential evolution algorithm. J Appl
Geophys 129:133–147
220 S. Trivedi et al.
Sen MK, Stoffa PL (2013) Global optimization methods in geophysical inversion, 2nd edn.
Cambridge Publisher, London
Sharma SP (2012) VFSARES—a very fast simulated annealing FORTRAN program for
interpretation of 1-D DC resistivity sounding data from various electrode array. Comput Geosci
42:177–188
Sharma SP, Biswas A (2011) Global nonlinear optimization for the estimation of static shift and
interpretation of 1-D magnetotelluric sounding data. Ann Geophys 54(3):249–264
Sharma SP, Biswas A (2013a) Interpretation of self-potential anomaly over a 2D inclined structure
using very fast simulated-annealing global optimization—an insight about ambiguity.
Geophysics 78:WB3–WB15
Sharma SP, Biswas A (2013b) A practical solution in delineating thin conducting structures and
suppression problem in direct current resistivity sounding. J Earth Syst Sci 122(4):1065–1080
Shaw RK, Agarwal BNP (1990) The application of Walsh transforms to interpret gravity
anomalies due to some simple geometrically shaped causative sources: a feasibility study.
Geophysics 55:843–850
Singh A, Biswas A (2016) Application of global particle swarm optimization for inversion of
residual gravity anomalies over geological bodies with idealized geometries. Nat Resour Res
25(3):297–314
Srivastava S, Agarwal BNP (2010) Inversion of the amplitude of the two-dimensional analytic
signal of the magnetic anomaly by the particle swarm optimization technique. Geophys J Int
182:652–662
Srivastava S, Datta D, Agarwal BNP, Mehta S (2014) Applications of Ant Colony Optimization in
determination of source parameters from total gradient of potential fields. Near Surf Geophys
12:373–389
Telford WM, Geldart LP, Sheriff RE (1990) Applied geophysics. Cambridge University Press
Thompson DT (1982) EULDPH-a new technique for making computer-assisted depth estimates
from magnetic data. Geophysics 47:31–37
Tlas M, Asfahani J (2011a) Fair function minimization for interpretation of magnetic anomalies
due to thin dikes, spheres and faults. J Appl Geophys 75:237–243
Tlas M, Asfahani J (2011b) A new-best-estimate methodology for determining magnetic
parameters related to field anomalies produced by buried thin dikes and horizontal cylinder-like
structures. Pure Appl Geophys 168:861–870
Tlas M, Asfahani J (2015) The simplex algorithm for best-estimate of magnetic parameters related
to simple geometric-shaped structures. Math Geosci 47(3):301–316
Tlas M, Asfahani J, Karmeh H (2005) A versatile nonlinear inversion to interpret gravity anomaly
caused by a simple geometrical structure. Pure Appl Geophys 162:2557–2571
Tuncer MK, Oshiman N, Baris S, Kamaci Z, Kaya MA, Isikara AM, Honkura Y (1991) Further
evidence for anomalous magnetic structure along the active fault in western Turkey.
J Geomagn Geoelectr 43:937–950
Chapter 9
Role of Euler Deconvolution in Near
Surface Gravity and Magnetic
Applications
9.1 Introduction
Fig. 9.1 Leonhard Euler (born: 15 April 1707 in Basel, Switzerland, died: 18 Sept 1783 in
St. Petersburg, Russia)
Slack et al. 1967), but the method itself was a newly built and developed in 1980–
1990s (starting with the fundamental papers: Thompson 1982; Reid et al. 1990).
The overall popularity of the technique has been gained mainly thanks to an
understandable and straightforward algorithm, which is also easily programmable.
Shortly after publication of the essential paper for the 3D version (Reid et al. 1990),
it became a part of some professional interpretation packages (e.g. extension
GRIDEPTH from Geosoft in 1994). Then, in next two decades, the Euler decon-
volution method “went through” a big boom and many modifications and
improvements of the basic concept have been published (to make here a complete
list of contributions is quite difficult—the most important ones are given in the
references list). As a side effect, for many users the Euler deconvolution represent
today a broad spectrum of methods or some modification of the original method,
rather than the original method. This contribution has to explain the principles of
the Euler deconvolution method and also describing the interesting contributions
that improved or enhanced the method over time. Great advantage of the method is
that it is a pure depth-estimator, it can give estimations about the source positions
without assuming any information about their physical properties (density or sus-
ceptibility contrasts). On the other hand, it is suitable only for the interpretation of
well-developed and separated anomalies from isolated sources.
9 Role of Euler Deconvolution in Near Surface Gravity … 225
9.2 Theory
As it was shortly mentioned in the introduction, main idea of the Euler deconvo-
lution method is based on the so called Euler theorem for homogeneous functions:
a function f(x, y, z) is called to be homogeneous in Euler sense, when the following
equation is valid (t is a real number) (Euler 1936, p. 93; in Stavrev and Reid 2007):
@f @f @f
x þy þ z ¼ nf ðx; y; zÞ: ð9:2Þ
@x @y @z
When we will take f(x, y, z) as a typical rational function g(x, y, z), which is used for
the description of effects of simple sources in potential theory:
(G, N are constants, r is the distance source—calculation point), then we will obtain
a very interesting result. Describing the situation in Cartesian coordinate system
with calculation point coordinates (x, y, z) and source position in point (0, 0, 0) we
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
can write for the distance r following simple expression: r ¼ x2 þ y2 þ z2 .
Entering Eq. 9.3 into Eq. 9.2 we obtain:
@g @g @g ðNxÞ ðNyÞ
x þy þz ¼x þy
@x @y @z ðx þ y þ z Þ
2 2 2 N=2 þ 1
ðx þ y2 þ z2 ÞN=2 þ 1
2
ðNzÞ Nðx2 þ y2 þ z2 Þ
þz ¼ ð9:4Þ
ðx2 þ y2 þ z2 ÞN=2 þ 1 ðx2 þ y2 þ z2 ÞN=2 þ 1
N N
¼ N=2
¼ N ¼ Ngðx; y; zÞ:
ðx þ y þ z Þ
2 2 2 r
Comparing this result with the right-hand side of Eq. 9.2 we can easily see that it
can be interpreted as: n = −N.
This can be also verified directly from Eq. 9.1:
G G
gðtx; ty; tzÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiN ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiN ¼ tN gðx; y; zÞ
t 2 x2 þ t 2 y2 þ t 2 z 2 tN x2 þ y2 þ z 2
ð9:5Þ
So, with both approaches it can be shown that this type of rational function is a
homogeneous function with the degree of homogeneity n = −N.
Thompson (1982) has named this parameter N as so called structural index
(SI) and described it as a “fall of rate—change with distance of the potential
function”. Structural index has different values for different types of sources in
gravimetry and magnetometry (summarized in Table 9.1; derivations of selected
bodies are given in details in Appendix 2). The visualization of the SI value as a fall
of the rate of the analysed function can be helpful—but only in the case of positive
and nonzero values of SI. For zero value and even negative values this idea is
misleading and would induce a concept where the field has no change or even
grows with the distance from the source. This is of course a nonsense. But zero and
negative SI values are real for some kind of models (Table 9.1) and therefore it is
better to understand the role of SI simply as source parameter. Also Fedi (2016)
9 Role of Euler Deconvolution in Near Surface Gravity … 227
Table 9.1 Values of structural index (SI) for various simple bodies (compiled from Yaghoobian
et al. 1992, Geosoft User guide for GRIDEPTH 1994, Stavrev 1997, Stavrev and Reid 2007)
Model Number of infinite Magnetometry Gravimetry
dimensions N N
Sphere, point pole/dipole 0 3 2
Vertical cylinder, rod 1 (z) 2 1
Horizontal cylinder 1 (x-y) 2 1
Sheet, dike 2 (z and x-y) 1 0
Slab, semi-infinite slab 2 (x and y) 1 0
(sill)
Contact, step 3 (x, y and z) 0 −1 (!)
analyses the actual situation with SI definition and he writes that “SI is a source
parameter and homogeneity degree is a field parameter”.
It is evident that there is a constant shift between SI values for identical bodies in
magnetometry and gravimetry (Ngrav ¼ Nmagn 1). This can be easily explained by
the fact that gravitational and magnetic fields are bounded via Poisson theorem
(magnetic field uses higher derivative of the gravitational field—and each differ-
entiation step increases the SI value by one).
An interesting concept of the relation between the SI values and number of
infinite dimensions of anomalous structure was recognized by various authors (e.g.:
Yaghoobian et al. 1992; Geosoft User guide for GRIDEPTH® 1994), but it was
generalized by Stavrev and Reid (2007) in a form of a simple algebraic relation:
SI ¼ s þ kd; ð9:6Þ
where s is the index of the gravity or magnetic source (1-pole, 2-dipole), k is the
derivative order (0-potential, 1-intensity, 2-gradient) and d is the number of
dimensions of the source (0-sphere, point-pole, point-dipole, 1-horizont or vertical
cylinder, line of poles, line of dipoles, 2-slab, sheet, sill, dike [vertical, horizontal,
inclined], 3-step, contact).
As it was shown by various authors (e.g.: Thompson 1982; Yaghoobian et al.
1992), an incorrect estimation of SI value yields wrong determination of the source
depth (this will be shown later in the part with synthetic examples). It is interesting
that horizontal coordinates of the source are not influenced by SI value—this was
theoretically derived by Barbosa et al. (1999).
In the classical (standard) Euler deconvolution (see section with algorithm
description), SI values are predicted (prescribed) and coordinates of the source are
determined. In various newer modifications of the method SI values are determined
together with the source position and a type of source can be interpreted. Here
originates a question: Should the SI value be an integer number (like in Table 9.1)?
Strictly speaking, an “integer character” of the SI values does not follow from the
Euler theorem, homogeneity degree can be any real number. On the other hand, we
can derive in an analytical way (in close solutions) only SI values for models with
228 R. Pašteka and D. Kušnirák
one typical singularity in their source characterization—and these are all integer
numbers (Table 9.1). Non-integer values should by typical for model bodies, which
are in some “intermediate stadium” between two typical clear one-singularity
bodies, like vertically limited rod or sheet, thick step, etc. In practical applications
of classical Euler deconvolution, many authors used such values (e.g. Yaghoobian
et al. (1992) used 0.5 for a thick step in magnetometry; Marson and Klinglele
(1993) used −0.5 for the same structure in gravimetry). Some authors (e.g., Reid
and Thurston 2014; Melo and Barbosa 2017) started to discuss and criticise this use
of noninteger SI values. The problem is that anomalous fields for such models are
not homogeneous functions in Euler sense and their SI are no longer a constant.
This was shown in the early stage of the method development by Ravat (1996),
which can finally yield misleading results. This is true and classical Euler decon-
volution method really produce in such cases more scattered solutions in both
horizontal and mainly vertical directions (when compared with integer SI values).
In order to deal with this drawback a proper window selection or clustering algo-
rithm (this will be discussed later) can partly “heal” the problem. In various newer
modification of Euler deconvolution, where the SI value is estimated together with
source coordinates (see further in this text), noninteger values of SI values are often
interpreted as transition stages between the basically interpreted models. Some
authors write (Fedi et al. 2015; Fedi 2016) that fractional values are normally
estimated when dealing with real sources, since they are more complex than ide-
alized sources, unless we are in asymptotic regions, that is too far or too close to the
sources. Finally, we can state that one should always stay cautious when processing
and interpreting noninteger SI values, but authors of this contribution do not think
that it should be formulated as some kind of “a prohibition” against their use in
Euler deconvolution method.
As it was mentioned, basic idea of the method is built upon the solution of the Euler
equation (Eq. 9.2). When we will write this equation for n = −N and expect that the
field source is in point (x0, y0, z0), it will change its form (3D-case, grid data
interpretation):
@f @f @f
ð x x0 Þ þ ðy y0 Þ þ ðz z0 Þ ¼ Nf ðx; y; zÞ: ð9:7Þ
@x @y @z
@f @f
ð x x0 Þ þ ðz z0 Þ ¼ Nf ðx; zÞ: ð9:8Þ
@x @z
In both equations, N and source position are unknown. As we can see, we have
always a situation with one equation, containing several unknowns (for both—
profile and grid type of interpretation). In applied geophysics, such situation is
usually solved in a way that the solved equation is written for several different
positions of measurement (processing) points and we can then solve a system of
equations (more details will be given later).
Thompson (1982) has performed tests with profile data interpretation. He real-
ized that direct solution of Eq. (9.8) yield unreliable and erratic results. This is due
the existence of regional background field and influence from other closed
anomalies. He suggested that during practical data interpretation it is necessary to
exclude such kind of influence by means of adding a constant background term into
the interpretation equation—so instead of function f() we have to work with
expression (f − B), where B is a constant. For a 2D-case we get:
@f @f
ðx x 0 Þ þ ðz z0 Þ ¼ Nðf BÞ: ð9:9Þ
@x @z
@f @f @f
ð x x0 Þ þ ðy y0 Þ þ ðz z0 Þ ¼ Nðf BÞ: ð9:10Þ
@x @y @z
@f @f @f @f @f @f
x0 þ y0 þ z0 þ NB ¼ Nf þ x þy þz ; ð9:11Þ
@x @y @z @x @y @z
@f @f @f @f
x0 þ z0 þ NB ¼ Nf þ x þz : ð9:12Þ
@x @z @x @z
In both Eqs. (9.11) and (9.12) the unknown parameters are the source coordi-
nates and the B term. N is no more a “free” parameter, it must pre predicted or
known. This is due to reason that the last term of the right-hand side of (9.11) and/
or (9.12) is a multiplication of two unknown variables (N and B), which cannot be
solved without some introduced constrains.
It is easy to demonstrate that the core of the classical (standard) Euler decon-
volution evaluation in the case of the 2D-case (Eq. 9.12). Basically, we have two
possibilities:
(a) We will write Eq. 9.12 for three different positions along the profile (x1, x2, x3
and f1, f2, f3) (see Fig. 9.2) and then we solve three equations for three
230 R. Pašteka and D. Kušnirák
@f @f @f
ð x x0 Þ þ ð y y0 Þ þ ðz z0 Þ ¼ A; ð9:13Þ
@x @y @z
9 Role of Euler Deconvolution in Near Surface Gravity … 231
where the constant A incorporates amplitude, strike and dip factors (in the case of
inclined contact in magnetometry).
In the interpretation of anomalous gravity field from a step/contact structure in
gravimetry (SI = −1) a next modification of the basic interpretation equation can be
helpful (Pašteka 2006), where instead of constant background term B a polynomial
function is incorporated:
2D-case:
@f @f
ðx x 0 Þ þ ðz z0 Þ ¼ Nf þ A0 þ A1 x þ A2 x2 þ þ An xn ; ð9:14aÞ
@x @z
3D-case:
@f @f @f
ðx x0 Þ þ ðy y0 Þ þ ðz z0 Þ ¼ Nf þ A0 þ A1 x þ A2 y þ A3 xy þ A4 x2 þ A5 y2 þ :
@x @y @z
ð9:14bÞ
This idea was adopted from the concept of interference polynomial in Werner
deconvolution (Werner 1953), but unfortunately it dramatically decreases the sta-
bility of the solved equation systems and can be hardly applied for noisy real
datasets (Pašteka 2006).
Typical outputs from Euler deconvolution method are sets of depth estimations
(almost for each position of the moving window one solution is obtained). Such
solutions build concentrated groups (clusters) at the positions of field sources—
centres in case of poles/dipoles (spheres) and lines of poles/dipoles (horizontal
cylinders); in the case of sheet-like bodies (sheets, dikes, sills) these are the edges.
Clustering principle is one of the strongest decision tools in Euler deconvolution—
strong and well separated sources manifestations attract intensively built clusters of
solutions (Fig. 9.3).
Fig. 9.3 Results from 2D-Euler algorithm applied on anomalous gravitational field from two
horizontal cylinders (black circles are obtained depth solutions from different positions of the
interpretation window, moving along whole profile). Sampling step = 1 km, window size = 15
points, used SI = 1
Fig. 9.4 Results from 2D-Euler algorithm applied on anomalous gravitational field from a
magnetic dike (x0 = 40 m, z0 = 7 m), sampling step = 1 km, window size = 10 points, used
SI = 1. a All received solutions, b selected solutions by means of the so called “window” criterion
Also the size of window plays an important role during the interpretation of
Euler deconvolution method results. It is good to try several different sizes and
analyse or cluster the received data. In general, it is valid that the window should be
9 Role of Euler Deconvolution in Near Surface Gravity … 233
Fig. 9.5 Typical deformations of solutions clusters, caused by large window size (in profile data
interpretation it is sometimes good to connect the solutions with a solid or dashed line—then we
can follow their origin and “movement” around source positions)
no too small (an important part of the anomaly should “fit in”) and not too large,
because then an interference of neighbouring anomalies can become active. For
small window sizes we often get defocused solutions and for large sizes we get
often solutions formed in strange shapes (tails, trails and arcs) (Fig. 9.5). These
cannot be interpreted as shapes of interpreted structures—e.g. like shapes of
anticline/syncline structures or dipping parts of contacts (this was well shown also
by Kuttikul et al. 1995).
(b) Standard deviation of the depth solution (SDz criterion)
As it was mentioned above, solution of an overdetermined equation system by
means of the LSQ method gives us not only solutions itself, but also their standard
deviations (SD) (e.g. Press et al. 2007). Standard deviation gives us a kind of
precision assessment—when SD has large values (even greater than the estimated
value itself) then it is with high probability a wrong solution. Important is to focus
on the SDz values from depth estimates—these carry the most important infor-
mation (not these for lateral coordinates). Reid et al. (1990) write about a level of
15%, which is acceptable, Yaghoobian et al. (1992) give different levels from 5 to
15%, decreasing then with the rise of SI values (higher SI values estimations need
smaller acceptance levels, because they are much more stable in comparison with
solutions for lower SI values) (Table 9.2). This behaviour was later on theoretically
explained by Barbosa et al. (1999). Thompson (1982) gives a simple formula for
the evaluation of the acceptance TOL level, based on SD value:
this means that only these solutions which have values of parameter z0/(NSDz)
larger than 20, will be accepted. Usually we accept solutions, which have values of
SDz below the level of 10–15% of its maximum values for whole interpreted dataset
(this acceptance level is usually adapted by user). Later, with synthetic and real data
examples, we will show the role of this SDz criterion.
(c) Condition number of the depth solution (CNz criterion)
This criterion is very similar to the SD criterion. Condition number (CN) is a special
parameter, which follows as an output from the so called Singular Value
Decomposition method them (e.g. Press et al. 2007). It is the ratio between the first
and last term of the central diagonal matrix, which originates from the SVD
method. It gives information about the stability of the received solution. When the
main matrix of the system is close to be singular (tendency for unstable solutions)
than the CNz will reach high values (>105–107). On the other hand, when CNz is
small (<102–103), then solutions tend to be stable. Sometimes we can get very high
values of CNz (e.g. over 1050, when adopting the interference polynomial into Euler
equation), but these solutions still build clusters around correct source positions
(Pašteka 2006). Usually we accept solutions, which have values of CNz below the
level of 10–15% of its maximum values for whole interpreted dataset (also this
acceptance level is usually adapted by user). Later, with synthetic and real data
examples, we will show the role of this CNz criterion.
(d) “Euler error”
This idea is very simple and straightforward and comes from Cooper (2004). It is
based on the substitution of received solutions back to the interpretation Eq. 9.11 or
Eq. 9.12 (all terms from the right-hand side are moved to the left-hand side). Ideally
in a case of a correct solution, we should get as result of this back-substitution zero
value in the right-hand side of the equation. This is not the case in real world with
all its limitations (noise, numerical errors, etc.) and we usually get a small number.
We can now order all obtained solutions, based on this value and usually the first
10% of all solutions build the most correct ones (Cooper 2004 has made experi-
ments even with smaller limits, like 2%).
(e) Background term properties
Barbosa et al. (1999) have realized (and theoretically proved) that best obtained
Euler solutions show an interesting property—their estimated constant background
term B has the smallest correlation with the interpreted field in a moving window.
Such a property can be used also for the selection of correct SI value, which was
further developed by Melo et al. (2013) and Melo and Barbosa (2017).
9 Role of Euler Deconvolution in Near Surface Gravity … 235
An important part of the input into the Euler interpretation Eqs. (9.9 and 9.10,
reformulated into 9.11 and 9.12) is built by the three orthogonal derivatives of the
interpreted field: ∂f/∂x, ∂f/∂y and ∂f/∂z. These derivatives can be numerically
evaluated by means of standard methods: either in space or spectral domain (using
Fourier transformation). These evaluations are instable calculations in the sense of
mathematical physics definition and their improper computation can contribute to
the growth of the instability of the whole method. This means that they amplify
noise and errors, which Sare in the input data (interpreted field f) and aggravate
therefore results from the whole method. Later in the section with synthetic models,
we will give some example of this behaviour. A solution of this problem is
available—introduction of controlled smoothing into the process of numerical
derivatives evaluation. There exist simple methods of smoothing in space domain
(e.g. the Hamming window) or low-pass filtration in the spectral (Fourier) domain
(e.g. the Butterworth filter). Some more pretentious methods use the Wiener opti-
mum filtering approach (e.g. Pawlowski and Hansen 1990) or enhanced derivatives
(Fedi and Florio 2001). One fashionable and ingenuous method to solve this
problem is the Tikhonov regularization approach (Tikhonov and Arsenin
1977)—this was used in a development of a low-pass filter in spectral domain
(Pašteka et al. 2009):
1
iuFff g; ð9:16Þ
1 þ au2
Fig. 9.6 a Synthetic Vz field for a 2D-horizontal cylinder with added artificial 5% noise;
b Constructed C-norm function for the evaluation of regularized x-derivative (component Vxz)
this minimum defines the optimum value of regularization parameter a. In Fig. 9.7
reader can see the horizontal derivative in x-direction (Vxz component) without and
with incorporation of regularization. In the nonregularized solution, there is a strong
dominating effect of the amplified noise and in the case of regularized solution, the
derivative is very close to the theoretical one (dashed line in both graphs). In papers
Pašteka et al. (2009) and Florio et al. (2014), there are given several examples of
this approach of stable numerical derivatives evaluation and its influence of Euler
method solutions. More information about this approach will be given in the part
Fig. 9.7 Non-regularized and regularized x-derivative of the Vz field from Fig. 9.6a. Dashed line
in both graphs shows values of synthetically evaluated component Vxz
9 Role of Euler Deconvolution in Near Surface Gravity … 237
After the classical Euler deconvolution was introduced (Thompson 1982; Reid et al.
1990), many developments and new modifications were published in the following
238 R. Pašteka and D. Kušnirák
Fig. 9.8 Screen-shot of the REGDER menu, selected is the 3D-case (upper part: menu for
derivatives evaluation, bottom part: menu for Euler deconvolution)
two decades. Here we conclude a short summary of a few most interesting modi-
fications of the Euler deconvolution method, some of them have also been used in
the part with real-world data interpretations.
(a) Derivative Euler Deconvolution
Due to the problems of SI value direct estimation from Eqs. 9.11 and 9.12 (where N
is multiplied with another unknown parameter B), some authors (e.g. Hsu 2002;
Cooper 2004) have got an idea to apply the Euler equation not directly to the field
f() itself, but to its vertical derivative fz = ∂f/∂z. Great advantage in such a case is
that the constant term B disappear, because ∂B/∂z 0. Basic Euler Eq. (9.10) will
now became (3D-case):
@f z @f z @f z
ðx x 0 Þ þ ð y y0 Þ þ ð z z0 Þ ¼ Nfz f z ; ð9:17Þ
@x @y @z
9 Role of Euler Deconvolution in Near Surface Gravity … 239
where Nfz is the structural index, valid for this new situation—its value is increased
by 1, when compared with the standard N: N = Nfz − 1. Here we can easily see that
after reformulation, we can directly estimate from Eq. 9.17 beside the source
coordinates also the structural index Nfz (this is the great advantage of this modi-
fication). On the other hand, great attention must be paid to the evaluation of the
initial vertical derivative of the interpreted field fz, because of the unstable beha-
viour of numerical derivatives evaluation. Also here, we suggest to use the concept
of regularized derivatives. Example of this modification application will be given
later on.
(b) Analytical signal Euler deconvolution
Reasons and main idea is very similar to the Derivative Euler deconvolution
modification. When we apply the basic Euler equation to the analytical signal
(AS) function (Keating and Pilkington 2004), then we can write a similar equation
than in the case (9.17):
where bx and by are constants (which are normally eliminated, except for N = 0)
and the horizontal Hilbert transformations Hx and Hy can be evaluated in Fourier
domain (Nabighian 1984). Again, great advantage of this modification is the pos-
sibility to obtain the estimate of the structural index together with the source
coordinates. FitzGerald et al. (2004, p. 462) give a quite large list of different
combinations of different equations solutions, when applying these new Eqs. 9.19a
and 9.19b, even in combination with the classical Euler equation (authors of this
contribution named it as a hybrid Euler deconvolution).
(d) Ridge Euler deconvolution
The main idea of a group of methods in this category is the application of Euler
equation to selected data in vertical direction—these were obtaining by means of
upward continuation, vertical differentiation and separation along so called ridges
(obtained by connecting local maxima of transformed function in vertical direction).
Such a multiscaling approach has a great advantage—only the important infor-
mation (along the ridges) is extracted from the interpreted field (simplified Euler
equations are then solved). Also here the direct estimation of SI values is per-
formed. Method went a quite large development—from single-ridge version (Florio
and Fedi 2006) to multiridge variant (Fedi et al. 2009; Florio and Fedi 2014), where
ridges of different types (RI, RII, RIII) are defined and analyzed. Another variant of
this approach is the DEXP method (Depth from Extreme Points) (Fedi 2007; Florio
et al. 2009).
In this part we would like to show typical properties of classical Euler deconvo-
lution, when applied on simple models, used in approximation of sources of
potential fields. In Fig. 9.9 we can see that for synthetic data (from various typical
2D sources) we get very reliable results—regarding the horizontal position of the
source (in the case of a cylinder it is the centre, in the case of sheet-like bodies it is
the edge). Depth is strongly dependent on the used value of SI—higher values give
incorrect deeper solutions, smaller values give shallower depths. Also the received
clusters are better developed in the case of correct SI value (it is important to
mention that on all these examples, only the “window condition” was applied). It is
interesting that in the case of vertical sheet (Fig. 9.9c, e), solutions for different SI
values are aligned along the vertical extension of the sheet. But this arrangement is
only apparent—its vertical shift is caused only due to the change of SI and this
process runs always and is independent on the slope of the sheet (it happens also in
the case of inclined or horizontal sheets—see Fig. 9.9d, f, g, h, where the
9 Role of Euler Deconvolution in Near Surface Gravity … 241
Fig. 9.9 Simple bodies and results from classical 2D-Euler deconvolution; a 2D-horizontal
cylinder, gravimetry; b 2D-horizontal cylinder, magnetometry; c 2D-vertical sheet, grav.;
d 2D-semiinfinite horizontal slab, grav.; e 2D-vertical sheet, magn.; f 2D-semiinfinite horizontal
slab, magn.; g 2D-inclined sheet, grav.; h 2D-inclined sheet, magn. Symbols for Euler solutions for
different SI values are in attached legend
“trajectory” of solutions did not followed the shape of the structure, it has simply a
vertical direction).
Contact structure is a little bit more complicated structure for classical Euler
deconvolution. In magnetometry, SI value 0 was derived by Reid et al. (1990), but
the slightly modified equation (Eq. 9.13) should be used. Figure 9.10a, b shows
that for this value of SI, the edge of the contact can be quite well defined. With the
contact structure in gravimetry, the situation is much more complicated. The
structural index N = 1, derived by Reid et al. (1990) was wrong, as it was pointed
by Pašteka (2001) and later on corrected in (Stavrev and Reid 2007, 2010) to the
value N = −1. In Fig. 9.10c, d we can see that solution for N = 0 and N = 1 are too
deep. It is interesting that the algorithm of classical Euler deconvolution does not
work well with N = −1 (this is purely empirical issue; we are not able to explain it
now from the theoretical point of view). Adopting the idea of interference poly-
nomial into the description of the background field (Eqs. 9.14a and 9.14b) helps
much in this situation—the edge of the contact can be very well defined for N = −1
(Fig. 9.11). Unfortunately, this modification makes the solution of the main matrix
242 R. Pašteka and D. Kušnirák
Fig. 9.10 Contact structure (in magnetometry and gravimetry) with 2D-Euler deconvolution
solutions for N = 0, 0.5 and 1.0
Fig. 9.11 Contact structure in gravimetry) with 2D-Euler deconvolution solutions for correct
structural index N = −1.0 (method modification with interference polynomial of 2. degree)
of the equation system too instable (as it was mentioned before) and it can be hardly
used for the interpretation of real noisy datasets.
One important aspect of various algorithms of Euler deconvolution is the
influence of the window size (number of points in the window). In previous figures
with simple model studies only the best solutions have been displayed, we always
run the calculations for various sizes of the window (in profile and grid interpre-
tation). In Fig. 9.12 can be nicely seen that the window size has an important
influence on the creation of solution clusters and their deformations. As it was
mentioned in the part about the algorithm, small windows give weak results (there
are only few and they are not well clustered) (Fig. 9.12, upper part). On the other
hand, used large window have tendency to produce deformations in the solutions
alignments (existence of “tails”, “trails” and “arcs”), as it can be seen in Fig. 9.12
(bottom part). In this case of two adjacent dikes (magnetometry) the window
size = 6 seems to be the best solution. Some authors suggested calculation with
several different sizes of the window and then applied afterwards some kind of
clustering algorithm to get the central value from all obtained solutions (e.g.
Gerovska and Araúzo-Bravo 2003; Mikhailov et al. 2003).
9 Role of Euler Deconvolution in Near Surface Gravity … 243
Fig. 9.12 Results for 2D-Euler deconvolution, applied on the anomalous magnetic field of two
dikes (SI = 1) for different sizes of the used window
Next important point is the influence of the noise on the Euler deconvolution
solutions. In general, a noise content in the interpreted data causes an origin of
many false solutions and defocusing of clusters (moreover depths of solutions are
often shallower). In this situation, the concept of regularized derivatives (Pašteka
et al. 2009) is very helpful—it can at least minimize the amplifying of the noise
244 R. Pašteka and D. Kušnirák
Fig. 9.13 Results for 2D-Euler deconvolution, applied on the anomalous magnetic field from a
thin belt fault structure. SI = 2, window size = 10 points. In the upper part the DT field curve is
displayed. Noise free case
content in the evaluated derivatives. In Figs. 9.13, 9.14 and 9.15 we bring one
synthetic example, where the model of a thin belt fault structure (magnetometry) is
studied. It was analytically shown by Reid (2003) that this model has a values of SI
equal to 2, which is quite unexpected (sheet-like forms in magnetometry have in
general N = 1). For a noise-free case, results from the classical 2D-Euler decon-
volution algorithm give a well developed cluster—but in the depth of the bottom
part of the bodies system (Fig. 9.13). In the case with a 5% added synthetic noise
(white noise), and without any smoothing or regularization of the derivatives
(Fig. 9.14), solutions are defocused and there occurred a quite large amount of false
solutions. Regularization of derivatives helps in this situation—received solutions
are better focused and the amount of false solutions was decreased (Fig. 9.15).
9 Role of Euler Deconvolution in Near Surface Gravity … 245
Fig. 9.14 Results for 2D-Euler deconvolution, applied on the anomalous magnetic field from a
thin belt fault structure. SI = 2, window size = 10 points. In the upper part the x- and z-derivatives
(nonregularized) of the DT field are displayed. Noise level in original dataset = 5% (white noise)
Here we present results from various models, which are composed of several
anomalous bodies in different positions. In general, it is valid that classical (stan-
dard) Euler deconvolution detect well shallower parts of the structures (Figs. 9.14,
9.16, 9.18 and 9.19), even in a case of a sub horizontal boundary with a
micro-relief. In the case of rectangular prisms only the upper edges of them can be
detected (Fig. 9.16). Very important situation occurs, when several structures are
superposed: when these are isolated bodies (Fig. 9.17), then the whole system of
them can be well defined (but transition between them not). But when we have a
246 R. Pašteka and D. Kušnirák
Fig. 9.15 Results for 2D-Euler deconvolution, applied on the anomalous magnetic field from a
thin belt fault structure. SI = 2, window size = 10 points. In the upper part the x- and z-derivatives
(regularized) of the DT field are displayed. Noise level in original dataset = 5% (white noise)
Fig. 9.16 Results for 2D-Euler deconvolution, applied on the anomalous gravitational field from
two adjacent prisms with inclined contact (and opposite differential densities). SI = 0, window
size = 6 points
Fig. 9.17 Results for 2D-Euler deconvolution, applied on the anomalous magnetic field from two
superposed prisms with polygonal cross-section. Only induced magnetisation is taken (I = 65°,
D = 0°). SI = 0, window size = 6 points
248 R. Pašteka and D. Kušnirák
Fig. 9.18 Results for 2D-Euler deconvolution, applied on the anomalous magnetic field from a
subhorizontal susceptibility contact. Only induced magnetisation is taken (I = 65°, D = 0°).
SI = 0, window size = 4 points
Fig. 9.19 Results for 2D-Euler deconvolution, applied on the anomalous magnetic field from a
superposition of two subhorizontal susceptibility contacts (with identical suscept. contrasts). Only
induced magnetisation is taken (I = 65°, D = 0°). SI = 0, window size = 4 points
estimates—which were then used for structure detection or for further interpretation
(manual modelling, solution of inverse methods, etc.).
(a) Cavities detection in environmental microgravity research (Wolfsberg, Austria)
This microgravity research was realized in 2009 year by company G-trend Ltd.
together with Austrian company AGS Ltd. in southern Austria—name of the site is
St. Marein (near Wolfsberg). It is an area above an abandoned coal mine, where
extracted volume of coal bed (Neogene age) causes collapses on the surface, which
can be dangerous. Excavated coal bed is in an approximate depth of 20 m below the
surface, plunging East-wards. Anomalous gravity field was acquired in a net of
2 2 m and incomplete Bouguer anomalies (without terrain correction evaluation)
have been calculated for the correction density 2.2 g cm−3 (Fig. 9.20, left). In this
anomalous field a strong dominating trend is visible (due to the effect of plunging
Neogene layers) and it was removed by means of a simple planar LSQ approxi-
mation. Resulting residual field (Fig. 9.20, right) shows several interesting
anomalous zones, but one of them is significant (in the NE part of the grid)—the
negative anomaly amplitude reaches more than −30 microGal, which could be
connected with a near surface cavity.
Independently from our microgravity research drilling works have been realized
by the company GKB Bergbau GmbH in the site and one of them (3N16) has
opened a cavity exactly on the edge of our detected negative anomaly (Fig. 9.21).
The edge of the cavity was approx. 3.5 m below the surface, its height was approx.
3–4 m (even greater) and the estimated diameter about 5 m. We have tried to
estimate the depth to the centre of the cavity by means of classical 3D Euler
deconvolution (structural index N = 2, centre of a sphere in gravimetry)—with and
without stabilization (by means of regularization) of the numerical derivatives
evaluation. Areal position of the best selected solutions is approximately identical
for both approaches (see figure below for solutions without regularized derivatives),
but the depths have been completely different, it can be clearly seen in the figure
with a vertical geological cross-section (Fig. 9.21, right).
Fig. 9.20 Incomplete Bouguer anomalies from Wolsfberg site (lef-hand—original field,
right-hand—residual anomalies after subtracting a linear trend). Correction density 2.2 g cm−3
250 R. Pašteka and D. Kušnirák
The solutions without regularized derivatives involvement (red circles) are too
shallow (over the real cavity position), these with regularized derivatives (blue
circles) are situated in the space of the cavity. The situation is not perfect—solu-
tions are focused in the upper part of the cavity and not in the centre of it, but such a
result we got. Interesting is also a fact that in planar visualization of the received
Euler solutions (Fig. 9.21, left), the cluster is shifter in the direction to the southern
edge of the negative anomaly. We meet this situation quite often in processing of
real microgravity data and are not 100% in explanation of this phenomenon—we
think that this is caused by the deformation of the acquired anomaly due to small
measurement errors.
(b) Cavities detection in archaeological microgravity research (Trnava, Slovakia)
Next example comes from typical microgravity survey for cavities in archaeology.
Microgravity data were acquired in the St. Nicholas baroque church in Trnava,
Western Slovakia in year 2006 (Pašteka et al. 2011). Acquisition was realized in a
net of 1 1 m points using a Scintrex CG-3M and CG-5 gravity meters. Bouguer
anomaly field was evaluated for the correction density of 1.80 g cm−3 and large
attention was paid to the realization of the correction of gravitational effects of walls
and other architectonic features. Results of the microgravity survey have been
verified by parallel GPR (Ground Penetrating Radar) survey. In this survey, seven
unknown crypts have been found by the two given methods and consequently
verified by video-inspection with a small borehole camera. We present here results
of the Derivative Euler deconvolution (DED) method, applied on a chosen portion
of residual Bouguer anomaly field (Fig. 9.22a). Chosen part of the residual Bouguer
anomaly field contains two small negative anomalies (min.: approx. −45 microGal),
9 Role of Euler Deconvolution in Near Surface Gravity … 251
caused by two crypts with equivalent shapes and depth positions (Pašteka et al.
2011). Also here, the solutions without regularized derivatives were diffused and
too shallow (not shown here). Incorporation of regularized derivatives into the DED
algorithm yield again interesting results (Fig. 9.22b, c). We can see well in the
section in Fig. 9.22c that the best developed cluster was for the solutions with SI
around 1, which is typical for the cylinder structure in gravimetry. This can be
accepted, because both crypts build a system of interconnected anomalies and both
are elongated in y-direction. The depth of the received cluster is in a very good
acceptance with the result from the GPR (Fig. 9.22c, the cross-section of the found
crypt based on GPR interpretation is outlined with blue line), where these solutions
are concentrated in the depth of approx. 1.4 m, which is close to the centre of the
crypt. These estimates (together with GPR vertical sections) have been used as
constrains for subsequent 3D density modelling of the crypts. All of the 7 detected
crypts by means of microgravimetry and GPR method (Pašteka et al. 2007) have
been confirmed and verified by video inspection (Fig. 9.23).
(c) UXO objects detection in environmental research (Studienka-Rohoznik,
Slovakia)
Next real data case-study will be a high-sensitivity magnetometric survey for
UneXploded Ordnance (UXO) detection survey in SW Slovakia in year 2006. Use
of Euler deconvolution in UXO-survey is well known and efficient (e.g. Davis et al.
2010, Paoletti et al. 2019). On the site, there was a need to clean a 24 m wide and
2 km long strip along a planned reflection seismic survey, as it was projected in a
close vicinity of an active military shooting site (Studienka-Rohoznik). Anomalous
Fig. 9.22 Selected part of incomplete Bouguer anomalies from the St. Nicholas baroque church in
Trnava, a together with the results from the Derivative Euler deconvolution method (b, c);
b xy-view with the received solutions, c xz-section with the received solutions (the cross-section of
the found crypt from GPR interpretation is outlined with blue line)
252 R. Pašteka and D. Kušnirák
Fig. 9.23 Some selected photos, captured from the video inspection of detected crypts in the St.
Nicholas church in Trnava (archive of M. Terray, personal communication)
magnetic induction field DT (Fig. 9.24a) was acquired using a Cs-vapor sensor
magnetometer with sampling interval of 0.1 m along lines separated by 1.0 m. For
this study we have selected a part of the explored site with dimensions 24 50 m
(Fig. 9.24a). Standard diurnal corrections and removal of the heading error have
been applied on each acquired line by means of a median filter technique.
Anomalous character of the DT field indicates a presence of few iron objects, some
of them can be significant for the UXO detection survey. As a first step, regularized
derivatives (DT/∂x, ∂DT/∂y and ∂DT/∂z) have been evaluated, using the concept of
C-norm function for the choice of the optimum regularization parameter (Pašteka
et al. 2009). Later the classical Euler deconvolution (CED) with presumed value of
structural index N = 3 (dipole in magnetometry) was applied. Received solutions
were filtered based on a standard deviation of estimated depths, where only the
lowest 15% were selected. Condition number criterion did not work well in this
case—some from the important clusters had high CNz values. Received results
(8 important clusters) are displayed in Fig. 9.24b and summarized in Table 9.3.
Beside this, also Derivative Euler deconvolution (DED) algorithm was used with
the aim of direct estimation of SI values (summarized also in Table 9.3). As we can
see, SI values are lower than 3 (2.5–2.7), which can be interpreted as a manifes-
tation of a body with not spherical, but elongated shape (projectiles are of approx.
ellipsoidal shape, see Fig. 9.25). Separation between projectiles and fragments,
which would be based on different values of estimated SI is not possible, values
from both objects are very similar (Table 9.3). Finally, these 8 positions were
signed directly in the field and excavated by EOD technicians (EOD = Explosive
Ordnance Disposal). Five objects were projectiles with 100 mm, the rest were
fragments from exploded projectiles (EOD technicians did not measure the depth in
the case of fragments). Estimated depths from the CED method are in the majority
deeper than the real depth of projectiles, only in one case (object nr. 1) the estimated
depth is too shallow. Largest error in the depth (object nr. 2) is probably caused by a
strong interference with anomaly from object nr. 3 (Fig. 9.24a). Typical examples
of excavated projectiles (100 mm tank gun projectile) are in Fig. 9.25.
9 Role of Euler Deconvolution in Near Surface Gravity … 253
Fig. 9.24 Results from UXO magnetometric survey site Studienka-Rohoznik. a Anomalous
magnetic induction field DT from a selected part of the survey. b Results from the CED method (8
clusters of depth solutions)
Table 9.3 Results from CED and DED methods together with real excavated depths of objects in
the site Studienka-Rohoznik (UXO magnetometric survey)
Object Depth from CED Estimated SI value Real depth Object
nr. (m) (DED) (m) description
1 1.4 2.8 2.0 100 mm
projectile
2 2.4 2.6 2.0 100 mm
projectile
3 1.3 ok 2.7 1.2 100 mm
projectile
4 1.7 2.4 1.4 100 mm
projectile
5 1.8 ok 2.5 1.7 100 mm
projectile
6 0.9 2.4 N/A Fragmentation
7 0.9 2.7 N/A Fragmentation
8 0.8 2.5 N/A Fragmentation
254 R. Pašteka and D. Kušnirák
9.5 Conclusions
important and reliable depth information, which can be used further in next inter-
pretation steps, like forward modelling or inversion. When using the Euler
deconvolution method, we should be aware of its quite high sensitivity to noise in
the input datasets. The concept of regularized (smoothed) derivatives can be helpful
in solving such situations.
Acknowledgements We would like to thank to our colleagues and friends, who helped us in
better understanding of the Euler deconvolution method (P. Richter for help with the derivation in
Appendix 1, R. Karcol for the help with several derivations of SI values); we are also thankful to
G. Florio, M. Fedi, A. Reid, R. O. Hansen, D. FitzGerald and B. Meurers for many interesting
discussions, concerning the method. Authors would also like to express thanks to companies GKB
Bergbau GmbH, AGS—Angewandte Geo-Systemtechnik GmbH and G-trend Ltd. for the per-
mission to publish the microgravity dataset from the site Wolfsberg—St. Marein and
UXO-geomagnetic data from site Studienka. Great thanks belong also to Miroslav Terray, who
offered to us his results from video-inspection and GPR-method from the site Trnava. Authors
have realized this study in the frame of the national project VEGA 1/0462/16 and it was also
based upon work from COST Action SAGA: The Soil Science & Archaeo-Geophysics Alliance—
CA17131 (https://www.saga-cost.eu), supported by COST (European Cooperation in Science and
Technology).
This is a very simple derivation of the Euler equation from the homogeneity the-
orem. As it cannot be found in common text-books of mathematical analysis, it is
given here. We work with homogeneous function f(x), where x has components (x1,
x2, x3) (in Cartesian coordinates can be written x1 = x, x2 = y, x3 = z).
Euler homogeneity theorem can be written:
Now both sides of Eq. (9.20) will be differentiated with respect to the real
number t:
Now in the right-hand side of Eq. (9.22) (in the sum) the expression from the
left-hand side of Eq. (9.21) will be entered:
X
M
@ ½ f ð xÞ
ntn1 f ðxÞ ¼ tn1 xi ;
i¼1
@xi
ð9:23Þ
X
M
@ ½f ðxÞ
nf ðxÞ ¼ xi :
i¼1
@xi
And this is the Euler’s equation, which is solved in the Euler deconvolution
method. For M = 3 and x1 = x, x2 = y, x3 = z, Eq. (9.23) can be written in the more
common form:
@f @f @f
nf ðx; y; zÞ ¼ x þy þz : ð9:24Þ
@x @y @z
For bodies with elementary shape (Table 9.1 in the main text), values of structural
index (SI) can be derived analytically. This was done by several authors (Reid et al.
1990; Pašteka 2001; Reid 2003; Stavrev and Reid 2010, etc.). There is no space to
give all of these derivations here for each of the source type. We give here few
typical examples, so reader can have an idea how it works and can derive other
types of bodies by themselves. Simplest derivations are of course for simple bodies,
like poles/dipoles and line of poles/dipoles. For tabular bodies (dike, sill), deriva-
tions are little bit more complicated, but can be done with tools of simple mathe-
matical analysis. The most complicated source is the step (contact) in gravimetry,
which has a value of −1. Such a negative value doesn’t work well with the
9 Role of Euler Deconvolution in Near Surface Gravity … 257
algorithm of classical Euler deconvolution and method has to use either modified
equations (Stavrev and Reid 2010) or a non-constant background term in Euler
equation (Pašteka 2006). This negative value was also hardly accepted by the
community, because negative indices contradict the idea of SI as a fall of rate of
interpreted function (as it is mentioned in the main text). It seems that the
anomalous field of this structure is not homogeneous in Euler’s sense (at least that
from vertically limited step) and several authors (e.g. Marson and Klinglele 1993)
recommend to apply firstly a vertical derivative evaluation and then the application
of the Euler deconvolution itself (with SI value around 0, which is valid for this
type of structure in magnetometry—see Table 9.1 in main text).
In the analytical derivation of SI values we can follow two concepts—based on
Eqs. 9.1 and 9.2 from the main text. From Eq. 9.1 (homogeneity equation) it fol-
lows after few simple adjustments (Stavrev and Reid 2007, p. L2) that for SI value
we get a simple formula:
We take a 2D-horizontal cylinder (rod) with the length density k, (e.g. Telford et al.,
1990, p. 37), which central axis is identical with the y-axis (crossing the xz-plane in
point [0, 0]). For the vertical component of gravitational acceleration Vz and
derivatives with respect to x- and z-direction (derivative with respect to y-direction
is equal to zero) we can write:
¼ 1:
ð9:29Þ
In the case of magnetized inclined sheet (dike), positioned with its edge along the
y-axis) we can write for the DT field following formula (Werner 1953):
Ax þ Bz
DTðx; zÞ ¼ ; ð9:30Þ
x2 þ z 2
where: A = −2b (Mx sin I + Mz cos I sin a) and B = 2b (−Mx cos I sin a + Mz sin I).
Mx and Mz are components of the magnetization vector, 2b is the thickness of the
sheet, I is inclination of the magnetization vector and a is the azimuth of the profile.
For the derivatives with respect to x and z (derivative with respect to y-direction is
equal to zero) we can write:
þ Bz þ Bz
ln½DTðtx; tyÞ=DTðx; yÞ x þz
2 2
x2 þ z 2
N¼ ¼ ¼
lnðtÞ lnðtÞ lnðtÞ
ð1Þ lnðtÞ
¼ ¼1 ð9:32Þ
lnðtÞ
N¼ ¼ ¼ 1:
Ax þ Bz Ax þ Bz
x2 þ z 2 x2 þ z 2
ð9:33Þ
Both ways came to an identical result (SI = N = 1) for the 2D-inclined sheet
(dike) in magnetometry. In gravimetry it would be N = 0.
We take a 2D-semiinfinite horizontal sheet (sill) with the surface m (e.g. Telford
et al. 1990, p. 40), which starts in the begin of the coordinate system [0, 0] and runs
along the positive part of x-axis-axis for the vertical component of gravitational
acceleration Vz and derivatives with respect to x- and z-direction (derivative with
respect to y-direction is equal to zero) we can write:
p x
Vz ðx; zÞ ¼ 2Gl þ arctg ;
2 z ð9:34Þ
@Vz ðx; zÞ z @Vz ðx; zÞ x
¼ 2Gl 2 ; ¼ 2Gl 2 :
@x x þz 2 @z x þ z2
2Gl x2xz
þ z2
2Gl x2zx 0
þ z ¼
2
N¼ p x p ¼ 0: ð9:35Þ
2Gl 2 þ arctg z 2Gl 2 þ arctg xz
References
Barbosa VCF, Silva JBC, Medeiros WE (1999) Stability analysis and improvement of structural
index estimation in Euler deconvolution. Geophysics 64:48–60
Cooper GRJ (2004) Euler deconvolution applied to potential field gradients. Explor Geophys
35:165–170
Davis K, Yaoguo L, Nabighian M (2010) Automatic detection of UXO magnetic anomalies using
extended Euler deconvolution. Geophys 75(3):G13–G20
Euler L (1936) Introduction to the analysis of infinitesimal. ONTI (Russian translation)
Fedi M (2007) DEXP: a fast method to determine the depth and the structural index of potential
fields sources. Geophysics 72(1):I1–I11
Fedi M (2016) An unambiguous definition of the structural index. Expanded abstracts from 86th
annual SEG international exposition and meeting, pp 1537–1541
Fedi M, Florio G (2001) Detection of potential fields source boundaries by enhanced horizontal
derivative method. Geophys Prospect 49:40–58
Fedi M, Florio G, Quarta T (2009) Multiridge analysis of potential fields: geometrical method and
reduced Euler deconvolution. Geophys 74(4):L53–L65
Fedi M, Florio G, Paoletti V (2015) MHODE: a local-homogeneity theory for improved source
parameter estimation of potential fields: Geophys J Int 202:887–900
FitzGerald D, Reid AB, McInerney P (2004) New discrimination techniques for Euler
deconvolution. Comput Geosci 30:461–469
Florio G, Fedi M (2006) Euler deconvolution of vertical profiles of potential field data. SEG 76th
Annual International Meeting. Expand Abstr 958–962
Florio G, Fedi M (2014) Multiridge Euler deconvolution. Geophys Prospect 62:333–351
Florio G, Fedi M, Pašteka R (2006) On the application of Euler deconvolution to the analytic
signal. Geophysics 71(6):L87–L93
Florio G, Fedi M, Rapolla A (2009) Interpretation of regional aeromagnetic data by multiscale
methods: the case of Southern Apennines (Italy). Geophys Prospect 57:479–489
Florio G, Fedi M, Pašteka R (2014) On the estimation of the structural index from low-pass filtered
magnetic data. Geophysics 79(6):J67–J80
Gerovska D, Araúzo-Bravo MJ (2003) Automatic interpretation of magnetic data based on Euler
deconvolution with unprescribed structural index. Comput Geosci 29(8):949–960
Hood P (1965) Gradient measurement in aeromagnetic surveying. Geophysics 30:891–902
Hsu SK (2002) Imaging magnetic sources using Euler’s equation. Geophys Prospect 50:15–25
http://www-groups.dcs.st-and.ac.uk/*history/Biographies/Euler.html (cit 2018-04-20)
Keating P, Pilkington M (2004) Euler deconvolution of the analytical signal and its application to
magnetic interpretation. Geophys Prospect 52:165–182
Kuttikul P, Barritt S, Hanssen R (1995) Maximization of geological information in 3D Euler
deconvolution. In: 57th annual international EAEG meeting, Glasgow, D031
Marson I, Klinglele EE (1993) Advantages of using the vertical gradient of gravity for 3-D
interpretation: Geophysics 58:1588–1595
Melo FF, Barbosa VCF (2017) Base level estimates to define the correct structural index in Euler
deconvolution. In: Expanded abstracts from 87th annual SEG international exposition and
meeting, pp 1770–1775
Melo FF, Barbosa VCF, Uieda L, Oliveira VC Jr, Silva JBC (2013) Estimating the nature and the
horizontal and vertical positions of 3D magnetic sources using Euler deconvolution.
Geophysics 78(6):J87–J98
Mikhailov V, Galdeano A, Diament M, Gvishiani A, Agayan S, Bogoutdinov S, Graeva E,
Sailhac P (2003) Application of artificial intelligence for Euler solutions clustering. Geophys
68(1):168–180
Mushayandebvu MF, van Driel P, Reid AB, Fairhead JD (2001) Magnetic source parameters of
two-dimensional structures using extended Euler deconvolution. Geophysics 66:814–823
9 Role of Euler Deconvolution in Near Surface Gravity … 261
Tikhonov AN, Glasko VB (1965) Application of the regularization method to nonlinear problems.
Zh. vychislit. matem. i matem. fiz. 5(3):463–473 (in Russian)
Werner S (1953) Interpretation of magnetic anomalies at sheet-like bodies. Sveriges Geologiska
undersekning, ser. C. Arsbok 43, N. 06
Yaghoobian A, Boustead GA, Dobush TM (1992) Object delineation using Euler’s Homogeneity
Equation. Location and depth determination of buried Ferro-Metallic bodies: Manuscript,
Proceedings of SAGEEP 92, San Diego, California
Chapter 10
Magnetic Data Interpretation
Using Advanced Techniques:
A Comparative Study
Abstract It can be demonstrated that magnetic data have diverse applications and
can be used to study different geological provinces. We used six distinct numerical
strategies for the body parameters (depth, origin, half-width, index angle and
amplitude Coefficient) assessment of a 2D inclined dike like a geologic structure.
These different strategies are the second horizontal gradient (SHG) method, the
Particle Swarm Optimization (PSO) technique, Werner deconvolution method,
Euler deconvolution method, the moving average method and the Modular Neural
Network (MNN). The utilization of these techniques was validated by using syn-
thetic data free of noise and including various range of random noise (5, 10 and
15%) and connected on the field data from Egypt and Canada. The estimated
inverse model parameters values from each single approach were compared toge-
ther and from those published in literature.
10.1 Introduction
measuring the variability in the magnetic field of the earth in view of the impacts of
magnetic susceptibility contrasts amongst the subsurface rocks studied (Essa et al.
2018; Ekinci 2016). Different automatic methods have emerged for computing the
magnetic model parameters, for example, depth, width, dip, susceptibility contrast,
and location of structures. The furthermost popular techniques are graphical
methods (Nettleton 1976), curves matching standardized techniques (Gay 1963;
McGrath and Hood 1970), characteristic points and distance methods (Grant and
West 1965), Fourier transform method (Bhattacharyya 1965), Werner deconvolu-
tion method (Werner 1953; Ku and Sharp 1983; Essa and Elhussein 2018a), Euler
deconvolution method (Thompson 1982; Mushayandebvu et al. 2001; Nabighian
and Hansen 2001; Silva and Barbosa 2003; Li 2003; Keating and Pilkington 2004).
Others are Hilbert transforms method (Mohan et al. 1982), statistical methods
(Spector and Grant 1970; Syberg 1972), symmetric and anti-symmetric analysis
method (Naudy 1971), and analytic signal method (Nabighian 1974; Agarwal 1984;
Roest et al. 1992; Bastani and Pedersen 2001; Beiki 2010), source parameter
imaging method (Thurston and Smith 1997), local wave number method (Smith
et al. 1998; Thurston et al. 2002; Salem et al. 2005; Smith and Salem 2005) and
continuous and/or discrete wavelets transform method (Moreau et al. 1997; Hornby
et al. 1999; Ridsdill-Smith and Dentith 1999; Fedi et al. 2004), Continuous least
squares minimization approaches (Abdelrahman and Essa 2005; Abdelrahman et al.
2003; McGrath and Hood 1973; Lines and Treitel 1984; Abo-Ezz and Essa 2016),
Fair function minimization procedure (Tlas and Asfahani 2011), DEXP method
(Fedi 2007), Simplex algorithm (Tlas and Asfahani 2015), Very fast simulated
annealing (Biswas and Acharya 2016; Biswas and Sharma 2016), the gradient
methods (Abdelrahman et al. 2007a), moving average methods (Abdelrahman et al.
2009). Lately, some inversion techniques found in the literature have become very
popular for interpreting of magnetic data (Abdelrahman and Essa 2005; Dondurur
and Pamukçu 2003; Al-Garni 2017).
The main goal of this chapter is focused on studying the magnetic effect of
dipping or so-called “inclined” dike that is considered in most geologic situation as
the source of magnetic field. The model may not be totally geologically sensible,
yet rather generally utilized in geophysical clarification to discover the depth, width,
and the thickness of inclined dike structures (Mehanee and Essa 2015). Different
graphical and numerical techniques have been established to estimate the source
parameters of the inclined dike (Gay 1963; McGrath and Hood 1970; Abdelrahman
et al. 2007a, b; Dondurur and Pamukçu 2003; Peters 1949; Hutchison 1958;
Bruckshaw and Kunaratnam 1963; Koulomzine et al. 1970; Bhattacharya 1971; Am
1972; Rao et al. 1973; Sengupta and Das 1975; Bhimasankaram et al. 1978; Rao
et al. 1981; Cooper 2012, 2015; Al-Garni 2015; Essa and Elhussein 2017a; Kara
et al. 2017).
The elucidation of the magnetic data over an inclined dike with an infinite depth
extent is endeavoured here utilizing six approaches: the gradient method
(Abdelrahman et al. 2007a), the particle swarm optimization (Essa and Elhussein
2017b), Euler deconvolution (Thompson 1982), Werner deconvolution (Hinze
et al. 2013; Werner 1953; Kilty 1983), the moving average method (Abdelrahman
10 Magnetic Data Interpretation Using Advanced Techniques … 265
et al. 2009) and the Modular Neural Network method (MNN) (Cooper 2015).
Investigation of the magnetic anomalies can permit acquiring extra detailed
information.
Different six algorithms utilized to infer the magnetic anomaly (nT) for a 2D
inclined dike are represented by the following equation (McGrath and Hood 1970;
Hood 1964):
1 xj þ w 1 xj w
M xj ; h; w; a ¼ A sin a tan tan
h h
2 !
cos a xj þ w þ h 2
ln 2 ; j ¼ 1; 2; 3; 4; . . .; N
2 xj w þ h2
where xj is the horizontal distance (m), h is the depth to the top (m), w is the
half-width (m), a is the index angle (°), and A is the amplitude factor (nT).
where s is an integer value called a window length. The amplitude factor can be
evaluated from the next formula:
h=8m
α = 50o
2w = 5 m
A = 2000 nT
Fig. 10.1 Magnetic field anomaly due to a 2D inclined dike with (A = 2000 nT, w = 2.5 m,
h = 8 m, and a = 50°) (top panel), and the model geometry with its parameters (bottom panel)
10 Magnetic Data Interpretation Using Advanced Techniques … 267
where
2Mxx ð þ sÞ tan1 2s hþ w tan1 2s h w 2 tan1 w
h
V¼
Mxx ð0Þ
3s þ w 3s w
tan1 tan1
h h
s þ w s w
tan1 þ tan1 :
h h
3s þ w
hf ¼ ; ð10:5Þ
tanðYÞ
where
2s þ w 2s w w
Y¼F tan1 tan1 2tan1
hj hj hj
3s w sþw sw
þ tan1 þ tan1 tan1 ;
hj hj hj
where
The particle swarm optimization (PSO) was developed by Kennedy and Eberhart
(1995) and it has many different applications in Model building, Biomedical,
268 K. S. Essa et al.
where Vkj is the velocity of the molecule j, Pkj is the recent i model at the k-th
iteration, rand() is a random number amongst (0, 1), c1 and c2 are constant numbers
that govern the individual and the social conduct, they are usually equal 2 (Sweilam
et al. 2007) but some up-to-date studies show that choosing c1 greater than c2 but
c1 + c2 4 may offer well results (Parsopoulos and Vrahatis 2002), c3 is the
inertial factor that regulate the velocity of the particle, it is frequently taken value
˂1, xkj is the location of the particle j at the k-th iteration.
The five source parameters (h, w, A, a, and xo) can be assessed by using the
PSO-algorithm on the subsequent objective function (Obj):
PN m
c
2 j¼1 M j M j
Obj ¼ P P ; ð10:8Þ
N c N m c
j¼1 M m
j M j þ j¼1 M j þ M j
Werner deconvolution technique (Hinze et al. 2013; Werner 1953; Kilty 1983) was
initially well-known for magnetic interpretation. The method is principally valuable
and suitable once the anomaly profile of importance can be communicated as an
objective formula. Werner deconvolution is not only effective for determining the
position, depth, dips, and magnetic susceptibility contrast times thickness product
of thin vertical or dipping sheets, but can also be employed to determine the
characteristics of 2D contacts or interfaces of prismatic bodies. More detail about
the Werner deconvolution method found in Hinze et al. (2013).
This method was developed by Thompson (1982) to define location and depth of
different magnetic sources. This method makes utilization of the first horizontal and
vertical gradients of the magnetic anomaly. A group of regular formulas are then
resolved to decide the position and depth of the causative supposing source con-
figuration by utilizing the next formula:
@M @M @M
ð x xo Þ þ ð y yo Þ þ ðz zo Þ ¼ SI ðB M Þ; ð10:9Þ
@x @y @z
where x, y and z are the observation point while xo, yo and zo are the source
coordinates, SI is the structural index, which relies upon the shape of the target in
our case (dike) SI = 1.
A
M xj ; h; w; a; s ¼ V xj ; h; w; a; s ; ð10:10Þ
2
270 K. S. Essa et al.
where,
x w
xj þ w
tan1 tan1
j
V xj ; h; w; a; s ¼ 2 sin a
h h
2 !#
cos a xj þ w þ h 2
ln 2
2 xj w þ h 2
1 xj þ s þ w 1 xj þ s w
sin a tan tan
h h
2 !#
cos a xj þ s þ w þ h 2
ln 2
2 xj þ s w þ h2
1 xj s þ w 1 xj s w
sin a tan tan
h h
2 !#
cos a xj s þ w þ h 2
ln 2
2 xj s w þ h2
The depth and the width can be gauged from the next equation:
fðhÞ
hf ¼ hj ; ð10:11Þ
f ðhÞ
The root mean square error (RMS) amongst the data and model anomalies is:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uP h i
u N Mm x Mc x 2
t j¼1 j j j j
RMS ¼ ; ð10:12Þ
N
where N is the data points numbers, Mm j is the observed magnetic anomaly and Mj
c
is the predicted magnetic anomaly at the point xj. This is deliberated as a standard in
estimating the best fit source parameters (h, w, A, a, and xo) of the buried
structures.
A noise free magnetic anomaly for a 2D inclined dike with A = 2000 nT, h = 8 m,
w = 2.5, a = 50° and profile length = 120 m (Fig. 10.1). We begin our analyses by
using the 2nd-horizontal derivative separation method (Eq. 10.2) to the magnetic
anomaly utilizing numerous s-values (s = 4, 5, 6, 7, 8 and 9 m) (Fig. 10.2). By
using this inversion approach, we assessed h, a and A values at various w for each
s-value and after that discovered the average depth and RMS (Table 10.1).
Table 10.1 shows the appraisal results of the elucidation of free data. The param-
eters obtained from the anticipated method are in a good covenant with the pro-
posed model. Thus, we can observe that the lowest RMS (RMS = 0 nT) happens as
the true ones.
The PSO-method was employed to a similar synthetic magnetic anomaly.
Moreover, there is no noise in the data, so we commence to investigate our method
employing 100 models. The best model was obtained after 800 cycles and the
scopes of the parameters are presented in Table 10.2. The gauged parameters have a
respectable covenant with the synthetic values (Table 10.2) at which the RMS
breaks even with zero.
Besides, we use Werner deconvolution technique to investigate a similar mag-
netic anomaly. We utilized 11 clustered solutions to ascertain in the average
appraised depth (8.77 m) (Fig. 10.3). Also, we used Euler deconvolution technique
to the same magnetic anomaly. We utilized 14 clustered solutions for computing the
average estimated depth which is 8.88 m (Fig. 10.4).
Moreover, the moving average method was used for the above-mentioned
synthetic example which is dependent on window curves. The estimated depth and
half width are 8 and 2.5 m respectively (Fig. 10.5). Finally, we have used the MNN
method to the noise free data. Table 10.3 shows the estimated parameters.
In geophysics, noise is regarded as component of data. With a specific ultimate
target to explore the effect of noise tainted data, we imposed random noise of
different level 5, 10 and 15% on the magnetic anomaly data. The 2nd-horizontal
272 K. S. Essa et al.
Fig. 10.2 Data analysis of Fig. 10.1 using the SHG method for a noise free data, b 5% random
noise, c 10% random noise and d 15% random noise imposed on the magnetic anomaly
derivatives were figured for identical s-values (Fig. 10.2). Table 10.1 in addition
shows the computational results for the elucidation of noisy magnetic data. The
average depth of 7.93 m with minimum RMS (6.35 nT) solution corresponds to 5%
random noise while in the case of 10% random noise the average depth is 8.10 m
with minimum RMS (16.94 nT). Finally, using 15% random noise, the average
evaluated depth is 8.13 m with minimum RMS (13.38 nT).
By utilizing the PSO-technique for the same diverse level noisy data of the
inclined dike model, the gauged parameters (h, w, A, a, and xo) are presented in
Table 10.2. Table 10.2 exhibits that the RMS error is 5.95 nT at the best solutions
using 5% random noise. Thus, an occurrence of 10% random noise on the synthetic
example, the RMS error is 10.72 nT at the correct solutions. Toward the end,
subsequent to applying 15% random noise on the same magnetic data, the RMS
error is 13.78 nT at the best solution.
10
Table 10.1 Numerical results of the magnetic field anomaly caused by a 2D inclined dike with (A = 2000 nT, h = 8 m, a = 50°, and w = 2.5 m)
contaminated with and without different levels of random noise (5, 10 and 15%) using the SHG method
w (m) Parameters s (m) Average RMS (nT)
4 5 6 7 8 9
Without random noise
0.5 h (m) 8.54 8.49 8.44 8.40 8.37 8.34 8.43 8.04
a (°) 50.08 49.93 49.85 49.76 49.71 49.67 49.83
A (nT) 10,544.45 10,468.90 10,383.85 10,331.16 10,277.37 10,226.50 10,372.04
1 h (m) 8.53 8.48 8.43 8.39 8.36 8.33 8.42 9.84
a (°) 49.90 49.82 49.78 49.73 49.70 49.69 49.77
A (nT) 5328.40 5272.62 5220.08 5185.87 5154.38 5126.49 5214.64
1.5 h (m) 8.44 8.38 8.34 8.30 8.27 8.25 8.33 8.27
a (°) 49.82 49.84 49.81 49.79 49.77 49.76 49.80
A (nT) 3535.51 3481.97 3452.20 3428.56 3410.60 3398.12 3451.16
2 h (m) 8.25 8.21 8.19 8.17 8.15 8.14 8.19 4.62
a (°) 49.89 49.91 49.90 49.88 49.87 49.86 49.89
A (nT) 2587.12 2562.76 2549.63 2539.41 2532.59 2526.50 2549.67
2.5 h (m) 8.00 8.00 8.00 8.00 8.00 8.00 8.00 0
Magnetic Data Interpretation Using Advanced Techniques …
Table 10.2 Numerical results of the magnetic field anomaly caused by a 2D inclined dike with
(A = 2000 nT, h = 8 m, a = 50°, and w = 2.5 m) contaminated with and without different levels
of random noise (5, 10 and 15%) using the PSO-technique
Parameters Used ranges Result RMS (nT)
Without random noise
A (nT) 500–3000 2000 0
h (m) 3–12 8
a (°) 10–80 50
w (m) 0.5–8 2.5
xo (m) −80 to 30 0
With 5% random Gaussian noise
A (nT) 500–3000 2201.08 5.95
h (m) 3–12 8
a (°) 10–80 50.14
w (m) 0.5–8 2.3
xo (m) −80 to 30 0
With 10% random Gaussian noise
A (nT) 500–3000 2250.5 10.72
h (m) 3–12 8.1
a (°) 10–80 49.85
w (m) 0.5–8 2.4
xo (m) −80 to 30 −0.04
With 15% random Gaussian noise
A (nT) 500–3000 2350 13.78
h (m) 3–12 8.1
a (°) 10–80 49.89
w (m) 0.5–8 1.9
xo (m) −80 to 30 0.14
Fig. 10.3 Werner solutions for magnetic field anomaly due to a 2D inclined dike with
(A = 2000 nT, w = 2.5 m, h = 8 m, and a = 50°) noise free data and contaminated data with
different levels of random noise (5, 10 and 15%)
10 Magnetic Data Interpretation Using Advanced Techniques … 279
Fig. 10.4 Euler solutions for magnetic field anomaly due to a 2D inclined dike with
(A = 2000 nT, w = 2.5 m, h = 8 m, and a = 50°) noise free data and contaminated data with
different levels of random noise (5, 10 and 15%)
Fig. 10.5 Different window curves of depth (h) versus half width (w) using s = 4, 5, 6, 7 and 8 m
for noise free synthetic example
280 K. S. Essa et al.
Fig. 10.6 Different window curves of depth (h) versus half width (w) using s = 4, 5, 6, 7 and 8 m
for 5% noisy synthetic example
A total magnetic field anomaly (A-A′) taken by Salem et al. (2000) from the total
airborne magnetic anomaly map that was conducted in 1996, over the Quseir region
including offshore extension to the Red Sea (Fig. 10.9). The profile length is
13,274.4 m and was digitized at 165.93 m sampling interim. The anomaly is
interpreted using the same approaches as discussed earlier. Firstly, the interpreted
results by using the SHG method are shown in Table 10.4 and The SHG anomalies
are exposed in Fig. 10.9 for s-values (s = 331.86, 497.79, 663.72 and 829.65 m).
Besides, by using the PSO-approach, the interpretive results obtained are summa-
rized in Table 10.5. Secondly, the depth and the location appraised by utilizing
282 K. S. Essa et al.
Fig. 10.7 Different window curves of depth (h) versus half width (w) using s = 4, 5, 6, 7 and 8 m
for 10% noisy synthetic example
Werner and Euler deconvolution methods are presented in Table 10.6. Thirdly, the
moving average estimated parameters (depth and half width) are shown in
Fig. 10.10 (h = 596.93 m and w = 207.41 m). Finally, the interpreted results from
the MNN technique are shown in Table 10.6. The results obtained particularly the
depth and half-width are compared with the results published by Salem et al. (2000)
(h = 763 m and w = 231 m) and are found to be in good agreement.
Fig. 10.8 Different window curves of depth (h) versus half width (w) using s = 4, 5, 6, 7 and 8 m
for 15% noisy synthetic example
utilizing the SHG technique are exhibited in Table 10.7 and The SHG anomalies re
appeared in Fig. 10.11 for s-values (s = 154, 176, 198, and 220 m). Additionally,
by utilizing the PSO-technique, the acquired results are condensed in Table 10.8.
Besides, the depth and the location evaluated by utilizing Werner and Euler
deconvolution methods are introduced in Table 10.9. Thirdly, the moving average
estimated parameters (depth and half width) are clearly shown in Fig. 10.12
(h = 304.5 m and w = 38 m). At long last, the results get from the MNN technique
are shown in Table 10.9. The evaluated parameters (h and w) are fully correlated
with McGrath and Hood (1970), Abdelrahman et al. (2012) and are found to be in
good correspondence.
284 K. S. Essa et al.
Fig. 10.9 A total magnetic field anomaly profile of length 13,274.4 m over the Quseir area, Egypt
(top panel) and the SHG anomalies using different s-values (bottom panel)
10 Magnetic Data Interpretation Using Advanced Techniques … 285
Table 10.4 Numerical results for the magnetic anomaly over the Quseir area, Egypt, Egypt using
the SHG method
w (m) Parameters s (m) Average RMS (nT)
331.86 497.79 663.72 829.65
165.93 h (m) 563.96 645.44 591.27 602.26 600.73 11.946
a (°) 49.47 38.81 34.11 30.46 38.21
A (nT) 119.42 178.93 179.64 201.27 169.82
186.67 h (m) 556.74 636.91 585.08 597.06 593.95 6.021
a (°) 49.52 38.87 34.29 30.93 38.40
A (nT) 115.35 157.49 158.57 162.00 148.35
207.41 h (m) 541.93 629.44 577.93 591.30 585.15 16.168
a (°) 49.59 38.93 34.28 30.62 38.36
A (nT) 92.05 140.17 141.52 159.31 133.26
228.15 h (m) 528.65 620.03 570.29 584.69 575.91 18.41
a (°) 49.66 39.00 34.39 30.72 38.44
A (nT) 81.80 125.86 127.56 143.88 119.78
248.90 h (m) 513.44 609.53 561.57 577.50 565.51 16.349
a (°) 49.74 39.08 34.50 30.83 38.54
A (nT) 73.01 113.78 115.75 130.96 108.37
269.64 h (m) 496.30 597.83 551.68 569.62 553.86 16.461
a (°) 49.85 39.17 34.64 30.95 38.65
A (nT) 65.37 103.39 105.58 119.94 98.57
290.38 h (m) 476.91 585.05 541.16 561.08 541.05 16.583
a (°) 49.96 39.27 34.78 31.09 38.78
A (nT) 58.60 94.37 96.83 110.43 90.06
10.4 Discussion
Table 10.6 The inverted model parameters estimated for the magnetic anomaly over the Quseir
area, Egypt, Egypt by using the six distinguished techniques
Methods Parameters
A (nT) h (m) a (°) w (m) xo (m)
The SHG method 148.35 ± 19.13 593.95 ± 28.80 38.40 ± 7.01 186.67 –
The PSO-technique 168 ± 8.89 603 ± 12.65 40 ± 6.47 192.5 ± 3.19 0.48 ± 0.27
Werner – 589 – – 0.15
deconvolution
method
Euler deconvolution – 587 – – 0.1
method
The moving average – 596.93 – 207.41 –
method
The MNN method 154 598.82 39.3 197.41 0.61
Fig. 10.10 Different window curves of depth (h) versus half width (w) using s = 331.86, 497.79
and 663.72 m for the field example, Egypt
10 Magnetic Data Interpretation Using Advanced Techniques … 287
Fig. 10.11 A total magnetic field measured over an outcropping of gabbroic olivine diabase dike,
Pishabo lake, Canada (top panel) and the SHG anomalies using different s-values (bottom panel)
288 K. S. Essa et al.
Table 10.7 Numerical results for the Pishabo Lake field example, Canada
w Parameters s (m) Average RMS
(m) (nT)
154 176 198 220
22 h (m) 296.74 301.40 319.25 334.87 313.07 32.488
a (°) 53.72 53.75 53.51 53.40 53.60
A (nT) 2618.89 2710.88 2989.20 3232.42 2887.85
33 h (m) 296.56 301.20 319.08 334.76 312.90 32.745
a (°) 53.61 53.67 53.45 53.35 53.52
A (nT) 1755.79 1815.47 2000.52 2162.49 1933.57
44 h (m) 295.35 299.94 318.24 332.29 311.46 14.340
a (°) 53.54 53.63 53.50 53.30 53.49
A (nT) 1497.76 1560.90 1502.67 1626.95 1547.07
55 h (m) 292.16 296.92 315.38 331.90 309.09 32.099
a (°) 53.58 53.68 53.45 53.33 53.51
A (nT) 1042.10 1077.41 1190.87 1293.02 1150.85
66 h (m) 288.28 293.38 312.30 328.93 305.72 31.328
a (°) 53.63 53.72 53.48 53.36 53.55
A (nT) 856.45 887.49 983.74 1068.67 949.09
77 h (m) 283.68 289.33 308.51 325.43 301.74 30.454
a (°) 53.68 53.76 53.53 53.41 53.59
A (nT) 722.30 751.19 833.81 907.29 803.64
88 h (m) 278.32 284.46 304.10 321.41 297.07 29.509
a (°) 53.74 53.82 53.58 53.46 53.65
A (nT) 620.24 647.02 720.19 785.29 693.19
99 h (m) 272.22 278.85 299.07 316.91 291.76 28.484
a (°) 53.79 53.88 53.64 53.51 53.70
A (nT) 539.78 564.79 630.79 689.81 606.29
110 h (m) 265.07 272.50 293.41 311.69 285.67 27.420
a (°) 53.86 53.95 53.70 53.57 53.77
A (nT) 473.54 498.00 558.40 612.07 535.50
Table 10.9 The inverted model parameters estimated for the total magnetic anomaly over the
Pishabo Lake field example, Canada by using the six distinguished techniques
Methods Parameters
A (nT) h (m) a (deg) w (m) xo (m)
The SHG method 1547.07 ± 52.38 311.46 ± 14.77 53.49 ± 0.12 44 –
The PSO-technique 1555.27 ± 34.62 314.32 ± 8.35 57.36 ± 2.47 46.24 ± 1.87 1.2 ± 0.18
Werner deconvolution – 305 – – 0.24
method
Euler deconvolution – 308 – – 0.17
method
The moving average – 304.5 – 38 –
method
The MNN method 1551.38 312.84 52.27 41.39 0.87
Fig. 10.12 Different window curves of depth (h) versus half width (w) using s = 154, 176 and
198 m for the field example, Pishabo lake, Canada
290 K. S. Essa et al.
the results published in literature. The first technique (the SHG method) has been
used to infer all model parameters except the origin of the body (xo) in the case of
utilizing synthetic example and the real data. The second technique (the PSO
method) was been used to estimate all model parameters using synthetic example
and the real data, while the Euler and Werner deconvolution techniques have been
utilized to synthetic and real data to decide the location of buried structure and the
depth (xo, h). The fifth technique is the moving average method which employed to
appraise the width and depth (w, h) of the buried structure in case of synthetic and
real cases. The last technique is the MNN technique, which determines all the body
parameters for synthetic and real cases. For example, Tables 10.6 and 10.9 show a
comparison between the estimated results for the real data cases using the six
different methods.
10.5 Conclusions
Six different methods were utilized for analysing magnetic anomaly for a 2D
inclined dike like a geologic structure. The feasibility of the approaches (the gra-
dient technique, particle swarm optimization, Werner deconvolution technique,
Euler deconvolution technique, the moving average method and the modular neural
network technique) is utilized on a theoretical case including noise free data, tainted
with different levels of noise (5, 10 and 15%), and real data from Egypt and Canada.
The six methods showed that their applications can improve data quality and give
better attributes and computational productivity. The contrast of the outcomes with
published information described in the state-of-arts revealed the pervasiveness of
the six methods and its potential for elucidating geologic and geophysical problems.
It is recommended that additional and better analytical models should be proposed
to make up for the shortcomings of the existing methods for inspecting magnetic
anomaly due to 2D inclined dike.
References
Abdelrahman EM, Soliman KS, Abo-Ezz ER, EL-Araby TM, Essa KS (2009) A least-squares
standard deviation method to interpret magnetic anomalies sue to thin dikes. Near Surf
Geophys 7:41–46
Abdelrahman SM, Abo-Ezz ER, Essa KS (2012) Parametric inversion of residual magnetic
anomalies due to simple geometric bodies. Explor Geophys 43:178–189
Abo-Ezz ER, Essa KS (2016) A least-squares minimization approach for model parameters
estimate by using a new magnetic anomaly formula. Pure Appl Geophys 173:1265–1278
Agarwal BNP (1984) Quantitative interpretation of self-potential anomalies. In: The 54th annual
international meeting, SEG, expanded abstracts, pp 154–157
Al-Garni MA (2015) Interpretation of magnetic anomalies due to dipping dikes using neural
network inversion. Arab J Geosci 8:8721–8729
Al-Garni MA (2017) Inversion of magnetic anomalies due to isolated thin dike-like sources using
artificial neural networks. Arab J Geosci 10:337
Alvarez JPF, Martìnez F, Gonzalo EG, Pèrez COM (2006) Application of the particle swarm
optimization algorithm to the solution and appraisal of the vertical electrical sounding inverse
problem. In: Proceedings of the 11th annual conference of the international association of
mathematical geology (IAMG06), Liège, Belgium, CDROM
Am K (1972) The arbitrary magnetized dike; interpretation by characteristics. Geoexploration
10:63–90
Bastani M, Pedersen LB (2001) Automatic interpretation of magnetic dyke parameters using the
analytic signal technique. Geophysics 66:551–561
Beiki M (2010) Analytic signals of gravity gradient tensor and their application to estimate source
location. Geophysics 75:I59–I74
Bhattacharya B (1971) Analysis of a vertical dyke infinitely deep, striking north by Fourier
transform. Pure Appl Geophys 89:134–138
Bhattacharyya BK (1965) Two-dimensional harmonic analysis as a tool for magnetic interpre-
tation. Geophysics 30:829–857
Bhimasankaram VLS, Mohan NL, Seshagiri Rao SV (1978) Interpretation of magnetic anomalies
of dikes using Fourier transforms. Geoexploration 16:259–266
Biswas A (2018) Inversion of source parameters from magnetic anomalies for mineral/ore deposits
exploration using global optimization technique and analysis of uncertainty. Nat Resour Res 27
(1):77–107. https://doi.org/10.1007/s11053-017-9339-2
Biswas A, Acharya T (2016) A very fast simulated annealing method for inversion of magnetic
anomaly over semi-infinite vertical rod-type structure. Model Earth Syst Environ 2:198
Biswas A, Sharma SP (2016) Integrated geophysical studies to elicit the structure associated with
Uranium mineralization around South Purulia Shear Zone, India: a review. Ore Geol Rev
72:1307–1326
Bruckshaw JM, Kunaratnam K (1963) The interpretation of magnetic anomalies due to dikes.
Geophys Prospect 11:519–522
Cedeno W, Agrafiotis DK (2003) Using particle swarms for the development of QSAR models
based on K-nearest neighbor and kernel regression. J Comput Aided Mol Design 17:255–263
Chau WK (2008) Application of a particle swarm optimization algorithm to hydrological
problems. In: Robinson LN (ed) Water resources research progress. Nova Science Publishers
Inc. New York, pp 3–12
Ciuprina G, Ioan D, Munteanu I (2002) Use of intelligent-particle swarm optimization in
electromagnetics. IEEE Trans Magn 38:1037–1040
Cooper GRJ (2012) The semi-automatic interpretation of magnetic dyke anomalies. Comput
Geosci 44:95–99
Cooper GRJ (2015) Using the analytic signal amplitude to determine the location and depth of thin
dikes from magnetic data. Geophysics 80:1–6
Dondurur D, Pamukçu OA (2003) Interpretation of magnetic anomalies from dipping dike model
using inverse solution, power spectrum and Hilbert transform methods. J Balk Geophys Soc
6:127–136
292 K. S. Essa et al.
Ekinci YL (2016) MATLAB-based algorithm to estimate depths of isolated thin dyke-like sources
using higher-order horizontal derivatives of magnetic anomalies. Springer Plus 5:1384
Essa KS, Elhussein M (2017a) A new approach for the interpretation of magnetic data by a 2-D
dipping dike. J Appl Geophys 136:431–433
Essa KS, Elhussein M (2017b) 2D dipping dike magnetic data interpretation using a robust particle
swarm optimization. Geosci Instrum Methods Data Syst Discuss (in review). https://doi.org/10.
5194/gi-2017-39
Essa KS, Elhussein M (2018a) Gravity data interpretation using new algorithms: a comparative
study. In: Zouaghi Z (ed) Gravity-geoscience applications, industrial technology and quantum
aspect. InTech. https://doi.org/10.5772/intechopen.68576
Essa KS, Elhussein M (2018b) PSO (particle swarm optimization) for interpretation of magnetic
anomalies caused by simple geometrical structures. Pure Appl Geophys 175:3539–3553
Essa KS, Nady AG, Mostafa MS, Elhussein M (2018) Implementation of potential field data to
depict the structural lineaments of the Sinai Peninsula, Egypt. J Afr Earth Sci 147:43–53
Fedi M (2007) DEXP: a fast method to determine the depth and the structural index of potential
fields sources. Geophysics 72:I1–I11
Fedi M, Primiceri R, Quarta T, Villani AV (2004) Joint application of continuous and discrete
wavelet transform on gravity data to identify shallow and deep sources. Geophys J Int 56:7–21
Gay SP (1963) Standard curves for the interpretation of magnetic anomalies over long tabular
bodies. Geophysics 28:161–200
Grant RS, West GF (1965) Interpretation theory in applied geophysics. McGraw-Hill Book Co.,
New York
Hinze WJ, Von Frese RRB, Saad AH (2013) Gravity and magnetic exploration: principles,
practices and applications. Cambridge University Press, New York
Hood P (1964) The Königsberger ratio and the dipping-dyke equation. Geophys Prospect 12:440–
456
Hornby P, Boschetti F, Horowitz FG (1999) Analysis of potential field data in the wavelet domain.
Geophys J Int 137:175–196
Hutchison RD (1958) Magnetic analysis by logarithmic curves. Geophysics 23:749–769
Kara I, Bal OT, Tekkeli AB, Karcioğlu G (2017) A different method for interpretation of magnetic
anomalies due to 2-D dipping dikes. Acta Geophys 65:237–242
Kearey P, Brooks M, Hill I (2002) An introduction to geophysical exploration. Blackwell
Publishing, Oxford
Keating P, Pilkington M (2004) Euler deconvolution of the analytic signal and its application to
magnetic interpretation. Geophys Prospect 52:165–182
Kennedy J, Eberhart R (1995) Particle swarm optimization. In: IEEE international conference on
neural networks, Perth, Australia, vol 4. IEEE Service Center, Piscataway, pp 1942–1948
Kilty TK (1983) Werner deconvolution of profile potential field data. Geophysics 48:234–237
Koulomzine T, Lamontagne Y, Nadeau A (1970) New methods for the direct interpretation of
magnetic anomalies caused by inclined dikes of infinite length. Geophysics 35:812–830
Ku CC, Sharp JA (1983) Werner method for automated magnetic interpretation and its refinement
using Marquardt inverse modelling. Geophysics 48:754–774
Li X (2003) On the use of different methods for estimating magnetic depth. Leading Edge
22:1090–1099
Lines LR, Treitel S (1984) A review of least-squares inversion and its application to geophysical
problems. Geophys Prospect 32:159–186
Lu WZ, Fan HY, Leung AYT, Wong JCK (2002) Analysis of pollutant levels in central Hong
Kong applying neural network method with particle swarm optimization. Environ Monit
Assess 79:217–230
Mandal A, Mohanty WK, Sharma SP, Biswas A, Sen J, Bhatt AK (2015) Geophysical signatures
of uranium mineralization and its subsurface validation at Beldih, Purulia District, West
Bengal, India: a case study. Geophys Prospect 63:713–724
McGrath PH, Hood PJ (1970) The dipping dike case: a computer curve-matching method of
magnetic interpretation. Geophysics 35:831–848
10 Magnetic Data Interpretation Using Advanced Techniques … 293
McGrath PH, Hood PJ (1973) An automatic least-squares multi-model method for magnetic
interpretation. Geophysics 38:349–358
Mehanee S, Essa KS (2015) 2.5D regularized inversion for the interpretation of residual gravity
data by a dipping thin sheet: numerical examples and case studies with an insight on sensitivity
and non-uniqueness. Earth Planets Space 67:130
Mohan NL, Sundararajan N, Seshagiri Rao SV (1982) Interpretation of some two dimensional
magnetic bodies using Hilbert transforms. Geophysics 46:376–387
Moreau F, Gibert D, Holschneider M, Saracco G (1997) Wavelet analysis of potential fields.
Inverse Problem 13:165–178
Mushayandebvu MF, van Driel P, Reid AB, Fairhead JD (2001) Magnetic source parameters of
two dimensional structures using extended Euler deconvolution. Geophysics 66:814–823
Nabighian MN (1974) Additional comments on the analytic signal of two-dimensional magnetic
bodies with polygonal cross-section. Geophysics 39:85–92
Nabighian MN, Hansen RO (2001) Unification of Euler and Werner deconvolution in three
dimensions via the generalized Hilbert transform. Geophysics 66:1805–1810
Naudy H (1971) Automatic determination of depth on aeromagnetic profiles. Geophysics 36:717–
722
Nettleton LL (1976) Gravity and magnetics in oil prospecting. McGraw-Hill Book Co.
Ourique CUO, Biscaia ECJ, Pinto JC (2002) The use of particle swarm optimization for dynamical
analysis in chemical processes. Comput Chem Eng 26:1783–1793
Parsopoulos KE, Vrahatis MN (2002) Recent approaches to global optimization problems through
particle swarm optimization. Nat Comput 1:235–306
Peters LJ (1949) Direct approach to magnetic interpretation and its practical application.
Geophysics 14:290–320
Rao BSR, Murthy IVR, Rao CV (1973) Two methods for computer interpretation of magnetic
anomalies of dikes. Geophysics 38:710–718
Rao DA, Babu HVR, Narayan PVS (1981) Interpretation of magnetic anomalies due to dikes: the
complex gradient method. Geophysics 46:1572–1578
Reynolds JM (1997) An introduction to applied and environmental geophysics. Wiley, Chichester
Ridsdill-Smith TA, Dentith MC (1999) The wavelet transform in aeromagnetic processing.
Geophysics 64:1003–1013
Roest WR, Verhoef J, Pilkington M (1992) Magnetic interpretation using 3D analytic signal.
Geophysics 57:116–125
Salem A, Ushijima K, ElSirafi A, Mizunaga H (2000) Spectral analysis of aeromagnetic data for
geothermal reconnaissance of Quseir area, Northern Red Sea, Egypt. In: Proceedings world
geothermal congress, Kyushu—Tohoku, Japan, pp 1669–1674
Salem A, Ravat D, Smith RS, Ushijima K (2005) Interpretation of magnetic data using an
enhanced local wave number (ELW) method. Geophysics 70:L7–L12
Santos FAM (2010) Inversion of self-potential of idealized bodies anomalies using particle swarm
optimization. Comput Geosci 36:1185–1190
Sengupta S, Das SK (1975) Interpretation of magnetic anomalies of dikes by Fourier transforms.
Pure Appl Geophys 113:625–633
Shaw R, Srivastava S (2007) Particle swarm optimization: a new tool to invert geophysical data.
Geophysics 72:75–83
Silva JBC, Barbosa VCF (2003) Euler deconvolution: theoretical basis for automatically selecting
good solutions. Geophysics 68:1962–1968
Smith RS, Salem A (2005) Imaging the depth, structure and susceptibility from magnetic data: the
advanced source parameter imaging method. Geophysics 70:L31–L38
Smith RS, Thurston JB, Dai TF, Macleod IN (1998) iSPITM: the improved source parameter
imaging method. Geophys Prospect 46:141–151
Sousa T, Silva A, Neves A (2004) Particle swarm based data mining algorithms for classification
tasks. Parallel Comput 30:767–783
Spector A, Grant F (1970) Statistical models for interpreting aeromagnetic data. Geophysics
35:293–302
294 K. S. Essa et al.
Sweilam NH, El-Metwally K, Abdelazeem M (2007) Self potential signal inversion to simple
polarized bodies using the particle swarm optimization method: a visibility study. J Appl
Geophys 6:195–208
Syberg FJR (1972) A Fourier method for the regional-residual problem of potential fields.
Geophys Prospect 20:47–75
Thompson DT (1982) EULDPH-a new technique for making computer-assisted depth estimates
from magnetic data. Geophysics 47:31–37
Thurston JB, Smith RS (1997) Automatic conversion of magnetic data to depth, dip, and
susceptibility contrast using the SPITM method. Geophysics 62:807–813
Thurston JB, Smith RS, Guillon JC (2002) A multimodel method for depth estimation from
magnetic data. Geophysics 67:555–561
Tlas M, Asfahani J (2011) Fair function minimization for interpretation of magnetic anomalies due
to thin dikes, spheres and faults. J Appl Geophys 75:237–243
Tlas M, Asfahani J (2015) The simplex algorithm for best-estimate of magnetic parameters related
to simple geometric-shaped structures. Math Geosci 47:301–316
Wachowiak MP, Smolíková R, Zheng Y, Zurada JM, Elmaghraby AS (2004) An approach to
multimodal biomedical image registration utilizing particle swarm optimization. IEEE Trans
Evol Comput 8:289–301
Werner S (1953) Interpretation of magnetic anomalies of sheet-like bodies. Sveriges Geologiska
Undersokning, Series C, Arsbok, p 43
Zavala AEM, Aguirre AH, Villa Diharce ER, Rionda SB (2008) Constrained optimisation with an
improved particle swarm optimisation algorithm. Int J Intell Comput Cybern 1:425–453
Zong-Ran D, Yan L (2017) A particle swarm optimization based approach for ship pipe route
design. Int Shipbuild Prog 63:59–84
Chapter 11
Coal Fire Study Over East Basuria
Colliery
Abstract Coal seam fire and coal mine fire as well as fire in coal waste or storage
piles are worldwide general phenomenon and pose a great threat to the national
economy, environment, ecology and health of peoples living in its proximity. In
17th century the coal mining in India first started in Raniganj coalfield in an ill-
planned way and some planned mining started in an about 1820. In early days, the
collieries were owned by several small companies and owners. In 1971, the Indian
coal industry was nationalized under Coal India (CIL) and Bharat Coking Coal
Limited (BCCL) was formed as a subsidiary of CIL. The haphazard coal mining
without suitable mine planning pose several paramount threats with environmental
hazards. At present it is spread over more than 70 areas. The mapping and
understanding of hidden subsurface coal fire is a challenging task. In view of this,
the present study was attempted for Coal fire mapping and detection of previously
burnt coal fire regions in East Basuria colliery, a part of the Jharia coal field, India.
Magnetic and Self Potential (SP) methods of geophysical investigation have been
used for coal fire mapping over parts of Jharia Coalfield, India mainly due to
cost-effectiveness with good accuracy and faster data acquisition. SP study works
based on redox potential generated by oxidation of coal and also on Thomson
potential due to temperature gradient. Whereas, magnetic study allows to map the
previously burned, currently burning and unburned locations based on the magnetic
properties of the materials and their changes with temperatures above or below
Curie temperature. SP data have been analysed using Particle Swarm Optimization
(PSO) technique. Magnetic data have been corrected for diurnal variation followed
by correction using reduced to magnetic pole (RTP). The corrected magnetic data
further, enhanced using north-south and east-west horizontal derivatives, first ver-
tical derivative, second vertical derivative, Total horizontal derivative, Analytical
signal, Tilt Derivative and downward continuation techniques. Radially averaged
power spectrum (RAPS) of magnetic data has been calculated to estimate the
average depth to the top of different layers. Different sets of 3D Euler’s depth
solutions have been estimated separately for each various source geometry (SI = 0,
0.5, 1.0, 1.5, 2, 2.5 and 3) to understand the possible source geometry of complex
fire activities. PSO inversion of SP data reveals that the depth of the coal fire
extends between 10 and 12 m below the surface. The study based on SP method
over East Basuria reveals that the geometry of subsurface coal combustion is
possibly similar to the inclined sheet with relatively large horizontal extension. The
large variations in inclination angles of the causative sources may possibly indicate
complex nature of fire propagation along different inclined fractured planes which
are generated in multiphase coal seam combustion. It is observed that the total
magnetic field intensity anomaly of the area varies approximately from 44,850 to
47,460 nT and the residual magnetic anomaly varies approximately from −1323 to
1253 nT. The range of the magnetic anomaly after RTP is approximately 1050–
1450 nT. Twelve low (L1–L12) and nine high (H1–H9) RTP magnetic anomalies
have been delineated from RTP of East Basuria Colliery for 2012. The average
depth to the top of different layers estimated using radially averaged power spec-
trum (RAPS) over East Basuria colliery are (i) 10 m, (ii) 18 m and (iii) 38 m for
2012. It is observed that the average depth of the estimated Euler’s solution for
different SI is varies from 10 to 26 m for 2012. Possible fire affected, non-fire and
possible baked and cool areas have been delineated as (i) 21%, (ii) 36%, (iii) 42%
for the years 2012 based on RTP anomaly distribution over East Basuria colliery.
The results prove the efficacy of the SP and magnetic methods for characterization
of causative sources associated with coal fires over East Basuria colliery, Jharia
coalfield, India.
11.1 Introduction
Coal is the leading energy resource of most of the countries around the world.
Different types of coal fires (Coal seam fire, mine fire, coal waste fires or storage
piles fires) are universally common occurrence and pose a danger to the national
economy, environment, ecology and health of peoples in nearby affected areas.
Coal fires pose several environmental threats which release toxic and greenhouse
gases and creates vegetation corrosion. Besides these, coal fires have a direct impact
on the surrounding areas degradation through significant aerosol input to water
resources and agricultural lands (Saraf et al. 1995; Rathore and Wright 1993;
Livingood et al. 1999; Ghose and Majee 2000; Bell et al. 2001).
The coal fires cause air, water and land pollution. The coal fire emits harmful
poisonous gases which causes lungs, skin and respiratory diseases via chronic
bronchitis and asthma while the emission of gases also leads to the global warming
to the exposed population. Coal mining methane emanation depends on the mining
methods, depth of coal seam, coal quality and entrapped gas of coal seams. The
11 Coal Fire Study Over East Basuria Colliery 297
water pollution from coal fires pollute water and increases its acidity, which is due
to the presence of certain amount of sulphur in the coal. Fire leads to land erosion
and does not permit any vegetation in that area.
Moreover, land subsidence occurs due to the underground volume loss, when a
coal seam burns and goes into ash, coal seam with its overlying strata degrades and
subsidized. A gradual or very sudden subsidence creates a threat to the infras-
tructure, local inhabitants and miners (Chen 1997; Guan 1989; Rathore and Wright
1993; Munshi 1995; Zhang et al. 2010; Zhang 2004, 1998; Prakash et al. 2010a, b;
Page et al. 2002). A number of human casualties occur due to coal fires in mining
accidents (Zhang et al. 2004; Boekemeier et al. 2002; Jia 2002). Valuable coal
resources burned out due to the coal fires particularly that can be easily exploited.
Coal fire leads to stoppage and devaluation of subsurface coal resources in the
seams below, above and around them. Forest coal fires lead to the destruction of
large forest land, including valuable re-establishment habitat for endangered spe-
cies. Coal fires contribute to the global warming problem due to releasing of huge
amount of CO2 in the atmosphere.
Coal fire is defined as burning of coal below underground surface (pile of storage
or waste coal piles), which has ability of burning for a greater time both extending
in dip and strike of coal seams (within the pile). They can be visualized by naked
eye. Coal fires occur below subsurface provides oxygen supply through longer
cracks to the burning subsurface coal through channelling. Various researchers
grouped coal fires as different types (Guan and van Genderen 1997; van Genderen
and Guan 1997; Yang 1995; Zhang 1998). Burning of coal fires occurs in different
ways, either by spontaneous combustion of coal, forest fires (Bustin and Mathews
1982, 1985), coal mining accidents, careless handling fires on the coal seam out-
crop, lightening strike (Guan et al. 1996) or by careless human interaction (Jia
2002). During excavation of coal mining provides supply of oxygen to the previ-
ously covered coal which leads to the coal waste and storage piles accumulation in
surrounding areas (Chaiken 1980; Chaiken et al. 1983, 1998; Kim and Chaiken
1993; Walker 1999; Van Genderen and Guan 1997).
Spontaneous coal combustion can be classified as: (i) surface and subsurface
coal fires as depth of combustion; (ii) paleo and recent burning coal fires as the
beginning time of ignited coal fire; (iii) extinct, dormant and active coal fires as
state of coal fire combustion; and (iv) coal field fire, coal mine fire and stock piles
fire as the starting point of coal fire ignition (Yang 1995). According to Guan and
Van Genderen (1997), Coal fires can be categorized into three types: (i) coal field
fire, (ii) coal mine fire, (iii) waste or stock fires and these categories could be
classified into several secondary categories. In 1998, Zhang classified coal fires into
four groups: subsurface mine fire, coal seam fire (coal field fire), coal refuse fire,
and coal stack fire.
The oxygen adsorption at inner and outer coal surface and its resulting oxidation
is an exothermic process which increases temperature within the coal accumulation.
If the subsurface temperature rises above 80 °C, the coal ignites and started to burn.
This process is defined as “spontaneous combustion”, which is the root cause of
coal fires for greater extent. Control of human activities in coal mining areas can
298 S. K. Pal and J. Vaish
reduce the spontaneous combustion of coal burning which occurs due to the heat
accumulation generated during the interaction of oxygen with coal, and poor
thermal conductivity favouring heat accumulation. The interaction of oxygen with
coal is mainly due to the oxidation of carbonaceous matter present in coal. The
other factors of heat generation are the oxidation of pyrite present in coal and
absorption of water vapour in coal. The most significant factors for spontaneous
combustion of coal are coal rank, coal type, geomorphologic setting, geological,
geographical and hydrological conditions and human interactions (Banerjee et al.
1972; Banerjee 1982, 1985; Gijbels and Bruining 1982; Schmal 1987).
Subsurface coal fires are severe problems in majority of countries, viz., China,
USA, South Africa (Stracher and Taylor 2004; Lohrer et al. 2005; Zhang 2004;
Zhang et al. 2004; Revil et al. 2013; Revil and Jardani 2013); India (Prakash et al.
1995; Michalski 2004; Gangopadhyay 2006; Kuenzer et al. 2008; Mishra et al.
2011; Bharti et al. 2014, 2016a, b; Singh et al. 2015; Srivardhan et al. 2016; Vaish
and Pal 2015, 2016) Australia, Russia, Poland and Indonesia (Ide et al. 2011; Ide
and Orr 2011), Germany, Czech Republic (Glover 1998; Deng et al. 2001; Pennig
2003; Boekemeier et al. 2002). Global distribution of coal fire is shown in
Fig. 11.1a. Globally, the ancient known coal fire, have been started burning in New
South Wales, Australia since 2000 years ago (Zhang et al. 2004). The subsurface
coal fire of Centralia in Pennsylvania, U.S., comes into great attention in 1962
which leads to the evacuation of whole city and rehabilitated during 1980 and 1998.
The gradually progressing coal fire underneath the city has affected infrastructure
and endangered houses. During recent times the fire still leads to the formerly
inhabited ground collapse. History of coal fires in China are mentioned in the travel
report of Li Dao Yuan as early as 1000 AC while the travel documents of Marco
Polo referred the “burning mountains along the silk road” as well as paleo-coal-fires
in Xinjiang, which are dated back to be of Pleistocene age (Schneider 1996; de Boer
et al. 1997; Gielisch 2002; Zhang and Kroonenberg 1996). During recent times
about 2030 Million tonnes of coal losses due to the burning of coal fires every year.
Actually, the tenfold amount of the coal resource is lost as it becomes inaccessible
to the nearby affected areas of a burning coal fire. In China, the financial loss of
valuable coal resource due to burning coal fire is estimated as about total 4.2 Billion
tons till now. Thus, China faces the major losses in terms of economical as well as
environmental due to burning coalfire every year.
The coal mining in India was started roughly in the year 1774 near Sitarampur in
Raniganj coal field, West Bengal (Pant and Murthy 2004; Chandra 1992; Coal
Atlas of India 1993). The coal mining was started roughly in the year 1777 over
Jharia (Chandra 1992) while planned mining has started in around 1820. The
geological survey for coal was started in 1845 and Geological Survey of India
(GSI) was officially established in 1851. The first geological map of the coal fields
was prepared on 1:63,360 scale in the year 1866 (Coal Atlas of India 1993; Coal
Directory of India 1998–99; Coal Research 1998). National Coal Development
Corporation was formed in 1956. Initially, the collieries were owned by several
small companies and private owners. Later, the coal industry of India was
nationalized in 1971 under Coal India (CIL) and being its subsidiary, the formation
11 Coal Fire Study Over East Basuria Colliery 299
(a)
(b)
Damodar
Fig. 11.1 a Global distribution of coal fire showing in star marks. b Coal fire affected area in
Jharia Coal field (modified after Michalski 2004). c Location map, East Basuria colliery along with
generalized geological map of Jharia Coal field and d subsurface mine workings plan (after BCCL
2009; Pal et al. 2017; Vaish 2018)
300 S. K. Pal and J. Vaish
of Bharat Coking Coal Limited, BCCL, occurred. Head quarter of BCCL is located
in Dhanbad, coal capital of India and ISM (presently IIT-ISM) was envisioned from
its initiation to operate within the 450 km2 of entire Jharia coal field.
Burning of coal mine fires in Jharia coalfield have been started since 1916 when
the first coal fire was reported near Bhowrah, Jharia coal field which later spread to
the nearby mining areas (Jogta—1941, Nadkhurki—1952, Rajapur—1956, West
Modidih—1957 and Jogidih and Koeridih—1965) (http://www.tehelka.com/2016/
05/jharias-infinite-inferno). Mining operations in Jharia coal fields have been the
major factor of spontaneous combustion of coal which is the main cause of sub-
surface coal fire that are continuously burning since decade and still on extension
during recent times (Mukherjee et al. 1991; Sinha 1986). During recent times, it
extends over more than 70 colliery areas.
Coal is the worldwide single largest energy generation resource which is applied
as resource for electricity generation and deals about 40% of the total energy power
generation of the world (Holtberg 2011; Mohalik et al. 2016). In India, the coal
offers 54.5% of the commercial energy supply which leads an important role in
supporting the Indian economy backbone and industrial growth irrespective of the
environmental causes. Thus, coal consumption in India has increased from 70
million tons (MT) in 1970s to 554 MT in 2011–2012 (Ministry of Coal Annual
Report, Government of India 2014). India is the third largest country in coal pro-
duction with approximately 80% from surface coal mining (Mohalik et al. 2016).
At present, the Indian coal industry having many problems owing to social and
environmental issues. The potentially mineable coal reserves are burning due to the
spontaneous combustion is the major economic loss experienced by the industry.
Since last 140 years, the coal fields of India have experienced a widespread open
and concealed coal fires (Dhar 1996). 40% of the total mining disasters and about
50% of the human causalities of Indian coal mining were because of coal mine fires
and explosion in both underground and opencast mine workings, since indepen-
dence up to 2010. More than 30 coal mine disasters have been reported in the last
ten decades regarding fire and explosion (Pandey et al. 2016; DGMS, Government
of India 2011). The major coal fires of India occurred in the Jharia, Raniganj,
Singrauli and Singareni coalfields (Mohalik et al. 2004, 2016). Spontaneous
combustion is the main cause of about 70% of these coal fires (Zutshi et al. 2001).
During the Jharia mine fire control technical assistance project, Michalski (2004)
identified that the majority of collieries are affected by subsurface mine fires as
presented in Fig. 11.1b.
The present study occupies an approximate of about 0.15 km2 area lying within
the geographical coordinates 23° 47′ 34″N to 23° 47′ 53″N latitudes and
86° 21′ 59″E to 86° 22′ 16″E longitudes. Normally, coal seams occur in the
Barakar formation of Lower Gondwana group of rocks of the Early Permian period
which mostly comprises of sandstone of different grain size, intercalation of
sandstone and shale, grey and carbonaceous shale and coal seams. The natural
topography is severely destroyed due to old mining activities, severe subsurface
coal fire activities which leads to formation of different potholes, cracks and fis-
sures, land sagging and small overburden dumping. The overburden materials have
302 S. K. Pal and J. Vaish
been dumped over coal fire affected regions to avoid and combat against further
coal fire exaggeration in the surroundings. In this regard, self-potential (SP) method
was not used for detail mapping of East Basuria Colliery except over a limited
part. Location map of the East Basuria Colliery, Jharia coal field, India along with
generalized geological map of Jharia Coal field (Fig. 11.1c) and underground mine
workings (Fig. 11.1d) are presented in Fig. 11.1.
The first coal fire at East Basuria colliery was reported in V/VI seam quarry in
1976. After that, in 1995, the intense fire was noticeable at three different galleries
of V/VI coal seams. To limit the extension of coal fires, water ponds were con-
structed in 1996. During that time, the coal fire has been extended up to the VIII and
VIII A coal seams, results in subsidence, fractures parallel cracks and potholes. By
using V/VI seam as a base, a trench has been made for cutting the eastward
advancement of coal fire during 1997–98. But the coal fire extended through trench
during cutting down of trenching (Bharat Coking Coal 2003; CMPDI 2003; World
Bank Group 2003). Lithology of boreholes SBJ-56, SBJ-6 and SBJ-4, represent
coal and shaly coal seams at various depth intervals are presented in Fig. 11.2
(BCCL 2009). The major coal bearing seams are VIII, V/VI, IV, III and II with
average thickness of 7.2 m, 9.45 m, 9.57 m, 3.66 m, 10.57 m respectively.
Initially, coal fire study of East Basuria Colliery has been carried out using
self-potential (SP) method and its PSO inversion. Further, detailed study of coal fire
of East Basuria colliery has been carried out using magnetic method.
The data collection of Self Potential method has been conducted with two
non-polarizable Pb–Pb Cl2 NaCl electrodes with high impedance digital
multi-meter (Fluke 287/289). Initially, base station has been fixed over non burning
region while the Self data have been collected with 1 m station increment at
working electrode. For drift correction, a every 30 min reading have been taken
repeatedly at the base station electrode. Holes were dug at every station to remove
the dry soil surface for efficient measurements. The data variations were less than
5 mV during conducted measurements. Figure 11.1d shows the self potential data
acquisition during the traverse of various profiles. Drift corrections measurements
have been superimposed on the conducted SP data.
According to Bhattacharya and Roy (1981), the synthetic Self Potential data that
has been acquired for cylindrical and spherical bodies using the Eq. 11.1:
11 Coal Fire Study Over East Basuria Colliery 303
Cave in zone
11.5-12.4(L3 seam) Indurated sediments
18.5-28.5(V/VI seam) due to surface fire and 14.57-15.25m(VIII B seam)
burnt outcrop 19.45 -22.8m (VIII A seam)
X /IX seam
26.20- 33.9m(VIII seam)
19.5 -22.8m (L3 seam)
41.4-41.98(L2 seam)
46.-46.14
40.7-41.3m (L2 seam)
48.1-52.5(IV-Top seam) 44.28 -46.55m(VII seam)
49.05-49.30m(L-3 seam)
54.4-58.5(IV-BoƩom seam)
54.15 65.10m(V/VI seam)
71.1-71.9m (VIII B
seam)
82.95 -83.2m(L2 seam)
91-95.8(III- seam), Shaly coal 90.05-93.8m(IV-Top seam)
95.08-99.6m(IV-BoƩom seam)
120.48-130.55(II-Top seam)
132.7-133.15(II-BoƩom seam)
132.3-137.9m(III seam)Shaly
coal
150.5-151.07(I- seam)
Legend
Coal seam
Shaly coal seam
162.8 -165.3m(II-Top)
Local Coal seam
166.7- 173.3m (II-Middle
Matrix rock (sandstone
seam)
/siltstone/shale)
Fig. 11.2 Borehole lithologs (BCCL 2009) representing coal and shaly coal seams at various
depth intervals. Locations of a SBJ-4, b SBJ-56 and c SBJ-6 are presented in Fig. 11.1 (after Pal
et al. 2017; Vaish 2018)
304 S. K. Pal and J. Vaish
ðx xo Þ cos h þ Z sin h
VðxÞ ¼ K q ð11:1Þ
ðx xo Þ2 þ ðZ Þ2
Here Z is the causative source depth, h signifies the angle of polarization, q is the
shape factor, xo is the origin of the anomaly and K. the electric current dipole
moment. The causative spherical and cylindrical sources and their geometrical
arrangements are represented in Fig. 11.3a, b respectively.
The SP response for an infinitely horizontally extended inclined sheet at right
angles to the strike is governed by Eq. 11.2 (Murthy and Haricharan 1985)
2h i3
ððx xo Þ a cos aÞ2 þ Z ða sin aÞÞ2
VðxÞ ¼ K ln4 h i 5 ð11:2Þ
ððx xo Þ þ a cos aÞ2 þ Z þ ða sin aÞÞ2
The PSO inversion that are applicable on inversion parameters are listed in
Table 11.3a. The inversion results have been represented in Fig. 11.4 and are in
good agreement as being recommendable misfit errors are precisely described in
Table 11.3b. The inversion results propose that the depth of the coal fire extends
between 10 and 12 m below the surface that are well correlated with reported
burning coal seam L3. At the site, blazing fire through crevices/fractures on the
ground shows the burning activity at the surface. The inversion results pointed out
the dissimilarities in the original and estimated values with higher misfit
11 Coal Fire Study Over East Basuria Colliery 305
(a)
Voltage (mv)
Offset (m)
(b)
Voltage (mv)
Offset (m)
(c)
Voltage (m)
Offset (m)
Fig. 11.3 Simulated SP anomaly response for a sphere, b cylinder, c inclined sheet (after
Srivardhan et al. 2016)
percentages errors at profiles 4 (1.0%) and 5 (1.034) that may be possibly little
noise in the data recording and recommends a robust PSO inversion. The low
values of shape factor denote the acquired anomaly is extended downward. This
possibly shows extension of the coal fire to the seams at deeper levels. The results
of the PSO optimization described that for all profiles, half widths are equal to the
depths of the causative sources and endorses large spreading of the coalfire within
306 S. K. Pal and J. Vaish
Table 11.1 (a) Input parameters for the synthetic dataset for the different causative sources
(Srivardhan et al. 2016). b Input range for the PSO inversion for different causative sources
(Srivardhan et al. 2016)
(a)
Body Z (m) xo (m) h (°) q a (°) a (m) Noise (%) K (mV)
Sphere 8 1.5 10 1.5 – – 21 −1450
Cylinder 25 2 30 1 – – 23 −1500
Inclined sheet 15 0 – – 10 10 20 −1300
(b)
Body Z (m) xo h (°) q a (°) a (m) Vo (mV) K (mV)
(m)
Sphere 1–40 0–10 1–60 1– – – −100 to 100 −1000 to −1600
2
Cylinder 1–40 0–10 1–60 1– – – −100 to 100 −1000 to −1600
2
Inclined 1–40 0–10 – – 1–60 1–30 −5000 to 5000 −1000 to −1600
sheet
Table 11.2 Inversion results for the synthetic models of different causative sources (Srivardhan
et al., 2016)
Body Z (m) xo h (°) q a a (m) Vo (mV) Misfit % K (mV)
(m) (°)
Sphere 8.3 1.34 11.1 1.5 – – −11.37 0.134 −1445.24
Cylinder 23.24 1.87 29.34 1 – – −334.72 0.036 −1495.27
Inclined 13.1 0.18 – – 9.41 10.1 −1346.4 4.18 10−5 −1300.08
sheet
Table 11.3 (a) The parameters used for the PSO inversion of the collected SP data (after
Srivardhan et al. 2016). (b) Summary of the results after PSO Inversion of the collected SP data
(after Srivardhan et al. 2016)
(a)
z (m) K (mV) a (m) h (°) Vo (mV) xo (m)
1–80 100–350 1–100 0.1–350 −100 to 100 0.1–10
(b)
Line Profile length Z h (°) a xo K q Misfit
number (m) (m) (m) (m) (mV) (%)
1 36 11.2 76.18 11.2 1.34 201.24 0.0 0.0773
2 35 10.1 9.41 10.1 0.18 200.08 0.0 0.0412
3 36 12.3 135.5 12.3 2.38 202.28 0.0 1.0001
4 12 10.0 5.00 10.0 0.10 200.00 0.0 1.0342
5 24 10.1 9.41 10.1 0.18 200.08 0.0 0.0870
6 12 11.1 70.19 11.1 1.24 201.14 0.0 0.1297
7 37 10.0 5.00 10.0 0.10 200.00 0.0 0.1295
11 Coal Fire Study Over East Basuria Colliery 307
(a) Line-1
Voltage (mv)
Offset (m)
(b) Line-2
Voltage (mv)
Offset (m)
(c) Line-3
Voltage (mv)
Offset (m)
Fig. 11.4 PSO inversions results for the profiles 1–7 (a–g) (after Srivardhan et al. 2016; Vaish
2018)
308 S. K. Pal and J. Vaish
(d) Line-4
Voltage (mv)
Offset (m)
(e) Line-5
Voltage (mv)
Offset (m)
(f) Line-6
Voltage (mv)
Offset (m)
(g) Line-7
Voltage (mv)
Offset (m)
seam. This proposes the geometrical arrangement of the burning coalfire zone is
equivalent to the inclined sheet with extensively spreading in plane. The causative
sources inclination angles are variable from 5° to 135.50°. The inversion results
exhibit a complex coal fire spreading along various fracture planes that produced in
multiphase burning of coal. Coalfire management based on plugging of burning
coal fire fissures is an exciting job and the analysis of the acquired SP data based on
PSO technique can be effectively used for coalfire hazard mitigation and prior
finding to save the economy, environment and people of neighbouring areas.
Before magnetic data acquisition, a base station has been established where coal fire
is not burning i.e. non coal fire region. At base station, data has been collected every
30 min for removing diurnal fluctuation readings. A sum of 2865 magnetic data
measurements has been collected along 42 profiles with 10 m increment in different
directions and data station spacing of 5 m increment along the profiles.
The diurnal corrections superimposed on collected magnetic readings is based on
the drift measured at the base station. The diurnal fluctuation readings are about
50 nT. The spikes and dropped signals were removed and only the good signal
readings were selected for preparation of total magnetic field intensity map. Total
field magnetic intensity map after International Geomagnetic Reference Field
(IGRF) correction is presented in Fig. 11.5a. The center coordinate points of the
East Basuria colliery is 86° 21′ 51.8″E longitude and 23° 47′ 45.56″N latitude.
According to the IGRF, the geomagnetic declination, inclination and total magnetic
field of the East Basuria colliery were −0.34°, 35.91° and 46,173.9 nT, respectively
during the field survey period in October 2012 while total magnetic field measured
over natural topography was about 46,160 nT ± 10 nT. IGRF corrected magnetic
310 S. K. Pal and J. Vaish
(a)
0N
0 0 0
E 0
E 0
E E E 0
E
0N
0N
0N
0N
0N
0N
0N
0N
0N
0N
0N
0 0 0
E 0
E 0
E E E 0
E
(b) 0 0 0 0 0 0
E E E E E E
L8
0N
H7
L5 0N
0N
L6
H6
H1
0N
L7
0N
H5 L1
L4
0N
L2 H2
L12 H4
0N
H8 L11 L3
L10
0N
L9
0N
H3
H9
0 0 0 0 0 0
E E E E E E
Fig. 11.5 a IGRF corrected magnetic anomaly map for the 2012 over East Basuria colliery,
b RTP anomaly map for the 2012 showing twelve magnetic low (L1–L12) and nine magnetic high
(H1–H9) anomaly regions over East Basuria colliery (after Pal et al. 2016; Vaish 2018)
anomaly was about −1360 to 1310 nT. Further, this magnetic data has been cor-
rected by reduced to north magnetic pole (RTP) magnetic anomaly map have been
generated for removal the lateral shifts and asymmetries produced by the local
magnetic anomaly. RTP magnetic anomaly map was about −1090 to 1490 nT, is
presented in Fig. 11.5b. Comparing, Fig. 11.5a, b it clearly denotes that RTP
11 Coal Fire Study Over East Basuria Colliery 311
(a) (b)
AcƟve
smoke/fire/blaz
Wide cracks
(c) (d)
Subsidence
Wide cracks
(e) (f)
Smoke through cracks
Wide cracks
(g) (h)
Cracks
Smoke through cracks
Fig. 11.6 a–h are field photographs over dynamic fire regions through fissures, cracks, potholes
etc. near low magnetic anomaly regions (L1–L8). i, j Field photograph showing temperature
measurements near L2 (340 °C) and L5 (195.4 °C) (after Pal et al. 2016; Vaish 2018)
11 Coal Fire Study Over East Basuria Colliery 313
blaze free zones/cooled baked rock regions (Fig. 11.7). Nineteen distinct first
vertical derivative low anomaly regions (FVL1–FVL19) have been identified which
are severely affected by surface coal fire activities (Fig. 11.6). Furthermore, in
Fig. 11.10b, eleven distinct second vertical derivative high anomaly regions
(SVH1–SVH11) have been identified that are feasibly related to non coalfire/blaze
free zones cooled baked rock regions (Fig. 11.7). Whereas, twenty distinct second
vertical derivative low anomaly regions (SVL1–SVL20) have been identified by
second vertical which are severely affected by active coal fire activity (Fig. 11.6).
Figure 11.11a shows the total horizontal derivative (THD) anomaly map gen-
erated from square root of the summation of squares of first order horizontal
derivatives (DX and DY) of RTP magnetic anomaly. About seventeen
(17) anomalies (THDL1–THDL17) have been identified with approximate centers
of the causative sources. Highest value contour of THD anomaly indicates the
horizontal extension of the causative source, whereas the lowest of THD anomaly
indicates the centre of the causative source. This technique is suitable for shallow
surface anomaly enhancement (Wijns et al. 2005).
The analytical signal (AS) anomaly map (Fig. 11.10b) has been estimated using
square root of the summation of squares of first order (DX, DY and FV) derivatives
of the RTP anomaly. About thirteen (13) high anomalies (ASH1–ASH13) have
been identified with their approximate centres of the bodies. Usually, analytic signal
produce highest value over the causative source (Nabighian 1972, 1974, 1984;
Roest et al. 1992; Roest and Pilkington 1993; Cooper and Cowan 2004; Narayan
et al. 2016). The low anomaly centres on THD could be related well with the high
anomalies on AS in their equivalent locations (Fig. 11.10a, b).
Tilt derivative (TDR) map (Fig. 11.11) is generated using inverse tangent of the
ratio of first vertical (FV) to total horizontal derivative (THD) of the study area. Tilt
derivative map of East Basuria colliery have been generated from RTP magnetic
anomaly map (Fig. 11.5b). The TDR image expresses 0 and ±p/4 (0 to
±0.785 rad) contours. The zero contours evaluate the horizontal location of abrupt
lateral changes in the subsurface features. The TDR map highpoints short wave-
lengths. The half distance between ±p/4 contours have been used for the depth of
magnetized sources edges. Eleven distinct causative source zones are identified
314 S. K. Pal and J. Vaish
(a) (b)
(c) (d)
(e) (f)
(g) (h)
Fig. 11.7 a–h are field photographs of burnt/baked rocks near high magnetic anomaly regions
(H1–H3, H5–H9) (after Pal et al. 2016, 2017; Vaish 2018)
11 Coal Fire Study Over East Basuria Colliery 315
(a) nT/m
DXH7
DXL6
DXL5
DXH6 DXH1
DXH5
DXL7
DXL1 DXH4
DXL4
DXL2
DXH10 DXH2
DXL10
DXL3
DXH8DXL9
DXL8
DXH3
DXH9
(b) nT/m
DYL7
DYH7
DYL5
DYL6 DYL11
DYH6
DYH1
DYH5 DYH4
DYL1
DYL4
DYL2
DYH10 DYH2
DYL10
DYL9
DYH8 DYL3
DYL8
DYH3
DYH9
Fig. 11.8 a East-West Horizontal derivative (DX) map showing ten high anomalies (DXH1–
DXH10) and ten low anomalies (DXL1–DXL10) zones for the year 2012 over East Basuria
colliery (after Vaish 2018). b North-South horizontal derivative (DY) map showing ten high
(DYH1–DYH10) and eleven low (DYL1–DYL11) anomaly zones for the year 2012 over East
Basuria colliery (after Vaish 2018)
316 S. K. Pal and J. Vaish
(a) 0
E 0
E 0
E 0
E 0
E 0
E
FVL8
0N
FVH7
FVL19
0N
FVL5
0N
FVL6
FVL9
FVH6 FVL18
FVH1
0N
FVL7 FVL10
0N
0N
FVH10 FVL2 FVH2
FVH11 FVL16
FVL12
0N
FVL3
FVL15
FVL14
FVH8
0N
FVH3
FVL13
0N
FVH9
FVL17
0 0 0 0 0
E 0
E E E E E
(b) 0
E 0
E 0
E 0
E 0
E 0
E
SVL8
0N
SVH7
SVL19 SVL20
0N
SVL5
0N
SVL6
SVL9
SVH6 SVL18
SVH1
0N
SVL7 SVL10
0N
SVL2
SVH10 SVL16 SVH2
SVH11
0N
SVL12
SVL3
SVL15
SVL14
0N
SVH8 SVH3
SVL13
0N
SVH9 SVL17
0 0 0 0 0
E E E E 0
E E
Fig. 11.9 a First vertical derivative anomaly (FV) map showing nineteen low (FVL1–FVL19)
and eleven high (FVH1–FVH11) anomaly zones for the year 2012 over East Basuria colliery (after
Pal et al. 2016; Vaish 2018). b Second vertical derivative anomaly (SV) map showing twenty low
(SVL1–SVL20) and eleven high (SVH1–SVH11) anomaly zones for the year 2012 over East
Basuria colliery (after Pal et al. 2016; Vaish 2018)
11 Coal Fire Study Over East Basuria Colliery 317
Table 11.4 (a) Magnetic anomaly (low) and its vertical derivatives with their surface
characteristics (after Pal et al. 2016; Vaish 2018). (b) Magnetic anomaly (high) and its vertical
derivatives with their surface characteristics (after Pal et al. 2016; Vaish 2018)
(a)
S. No. Magnetic FV SV Central Central latitude RTP Surface
anomaly longitude magnetic characteristics
anomaly
(nT)
1 L1 FVL1 SVL1 86° 21′ 56.25″ 23° 47′ 45.98″ −1043 Active fire/blaze/
E N smoke through wide
fractures/fissures/
cracks
2 L2 FVL2 SVL2 86° 21′ 58.53″ 23° 47′ 44.41″ −563 Active fire/blaze/
E N smoke through
fissures/cracks
3 L3 FVL3 SVL3 86° 22′ 0.11″ 23° 47′ 43.38″ −998 Active fire through
E N pot hole and cracks
and subsidence
4 L4 FVL4 SVL4 86° 21′ 50.58″ 23° 47′ 45.61″ −547 Wide spread cracks
E N and subsidence due
to ire/blaze
5 L5 FVL5 SVL5 86° 21′ 50.28″ 23° 47′ 50.952″ −258 Cracks and
E N subsidence due to
fire/blaze
6 L6 FVL6 SVL6 86° 21′ 47.47″ 23° 47′ 50.34″ −618 Smoke through pot
E N hole/ fissure
7 L7 FVL7 SVL7 86° 21′ 44.69″ 23° 47′ 47.637″ −181 Cracks and
E N subsidence
8 L8 FVL8 SVL8 86° 21′ 42.98″ 23° 47′ 55.047″ −150 Pot hole and cracks
E N and subsidence
9 – FVL9 SVL9 86° 21′ 49.10″ 23° 47′ 49.35″ −56.15 Cracks and
E N subsidence
10 – FVL10 SVL10 86° 21′ 48.55″ 23° 47′ 47.77″ −46.8 Cracks and
E N subsidence
11 – FVL11 SVL11 86° 21′ 48.46″ 23° 47′ 46.7″N −40.72 Smoke through
E fractures/fissure
12 – FVL12 SVL12 86° 21′ 47.73″ 23° 47′ 43.05″ −51.74 Coal fire burning but
E N intensity decreased
13 L9 FVL13 SVL13 86° 21′ 58.31″ 23° 47′ 40.53″ −328 Shallow fire/smoke
E N through fractures/
14 L10 FVL14 SVL14 86° 21′ 58.40″ 23° 47′ 41.9″N −286 fissure
E
15 L11 FVL15 SVL15 86° 21′ 56.94″ 23° 47′ 42.8″N −336
E
16 L12 FVL16 SVL16 86° 21′ 55.61″ 23° 47′ 43.69″ −378
E N
17 – FVL17 SVL17 86° 22′ 0.066″ 23° 47′ 38.09″ −31.69 Shallow fire/smoke
E N through fractures/
fissures
18 – FVL18 SVL18 86° 21′ 59.43″ 23° 47′ 49.14″ −28.08 Shallow fire/smoke
E N through fractures/
fissures
19 – FVL19 SVL19 86° 21′ 44.61″ 23° 47′ 53.02″ −33.49 Shallow fire/pot hole
E N and cracks
20 – – SVL20 86° 21′ 47.77″ 23° 47′ 51.92″ −3.03 Shallow fire/pot hole
E N and cracks
(continued)
318 S. K. Pal and J. Vaish
nT/m
(a)
THDL8
THDL1 THDL5
THDL17
THDL9
THDL7 THDL6
THDL1
THDL10 THDL4
THDL2
THDL15
THDL14
THDL11 THDL3
THDL13
THDL12
THDL16
nT/m
(b)
ASH13
ASH6 ASH5
ASH7
ASH12
ASH4 ASH1
ASH2 ASH3
ASH11
ASH8
ASH10
ASH9
Fig. 11.10 a Total horizontal derivative map of East Basuria colliery showing seventeen centre of
the causative anomaly (THDL1–THDL17) (after Vaish 2018). b Analytical signal map of East
Basuria colliery showing thirteen centre of the causative anomaly (ASH1–ASH13) (after Vaish
2018)
Fig. 11.11 Tilt Derivative map of East Basuria Colliery showing eleven positive (TDH1–TDH11)
and twenty negative Tilt derivative (TDL1–TDL20) (after Pal et al. 2017; Vaish 2018)
Table 11.5 Detail enhanced Magnetic anomaly anomalies delineated using different derivatives
techniques (after Vaish 2018)
S. Magnetic DX DY FV SV THD Analytical TD
No. anomaly signal
1 L1 DXL1 DYL1 FVL1 SVL1 THDL1 ASH1 TDL1
2 L2 DXL2 DYL2 FVL2 SVL2 THDL2 ASH2 TDL2
3 L3 DXL3 DYL3 FVL3 SVL3 THDL3 ASH3 TDL3
4 L4 DXL4 DYL4 FVL4 SVL4 THDL4 ASH4 TDL4
5 L5 DXL5 DYL5 FVL5 SVL5 THDL5 ASH5 TDL5
6 L6 DXL6 DYL6 FVL6 SVL6 THDL6 ASH6 TDL6
7 L7 – – FVL7 SVL7 THDL7 – TDL7
8 L8 – DYL7 FVL8 SVL8 THDL8 – TDL8
9 – – – FVL9 SVL9 THDL9 ASH7 TDL9
10 – – – FVL10 SVL10 THDL10 – TDL10
11 – DXL7 – FVL11 SVL11 – – TDL11
12 – – – FVL12 SVL12 THDL11 ASH8 TDL12
13 L9 DXL8 DYL8 FVL13 SVL13 THDL12 ASH9 TDL13
14 L10 – – FVL14 SVL14 THDL13 – TDL14
15 L11 DXL9 DYL9 FVL15 SVL15 THDL14 ASH10 TDL15
16 L12 DXL10 DYL10 FVL16 SVL16 THDL15 ASH11 TDL16
17 – – – FVL17 SVL17 THDL16 – –
18 – – DYL11 FVL18 SVL18 THDL17 – TDL18
19 – – – FVL19 SVL19 – ASH13 TDL19
(continued)
11 Coal Fire Study Over East Basuria Colliery 321
with their probable locations, e.g. L1 could be correlated with DXL1, DYL1,
FVL1, THDL1, ASH1 and TDL1 for DX, DY, FV, SV, THD, AS and TD,
respectively. Nevertheless, a few minor mismatch in orientations, extensions and
spatial occurrences with some offsets have been observed for the delineated
anomalies. These are due to differences in different edge enhancement techniques
and their limitations. These offsets are possibly associated with different complex
causative anomalies occurring nearby/superimposed with varying inclination
(Grauch and Cordell 1987).
Magnetic anomaly distribution at the surface and subsurface at various depths viz.,
20, 40 and 80 m (downward continued anomaly) have been produced and is rep-
resented by 3D schematic model (Fig. 11.12).
The red regions showing the increase in magnetic enhancement with respect to
ambient field where as the blue regions representing the reduction in magnetic
anomalies with respect to natural conditions. Figure 11.6a–h shows the low magnetic
anomaly regions (L1–L8), are that depicted over coal fire as fissures, cracks and pot
holes while the high magnetic anomaly regions viz H1–H3, H5–H9, of baked/burnt
rocks are presented in Fig. 11.7a–h; respectively. This signifies that red regions are
those which were heated and subsequently cooled while the blue regions are those
where some portion of the formation is under fire and heated above the Curie tem-
perature. The light green regions are denoted as non burning coal fire regions.
322 S. K. Pal and J. Vaish
SBJ-56
SBJ-6
SBI-4
L8 H7 L5 SBJ-6 H1
SBJ-4 L6 H6
L7 L1 H2
H5 L4 L2 L3
L12 L11 L10
VII
H8 H4
L9 H3
H9
Surface
VIII B
SBJ-56
L3
VIII BIII A
20 20
L5 H6 20
20 L3 20
L8 20 20
L1 H2
V / VI
H10
L3
20 20 20
L6 H5 20 L2
20 H4
20
L9 H8 20
20
H3
H9
20m
40
H740
L4 40
H11
40
L6 40H640 40
H1
L2
L3
IV Top L2
40
L7 40
VII
40
H10 40
L1 H2
40 40 40
L5 H5 40 L2 40
40
40
H8 H4 H3
V / VI L-3
L8 40
40
H9
40
L9 L10
40m
IV Bot
80
80
H6 L8
80
80
H10 80
H5 80
H180
80
L1 L2 80H2
80 H9
VIII B
L3 80
H4 80
80
L4 80
L5 80 H3 80
H7 80
L2
L6 80
H8
VIII A
80
L7
L2
80m
IV Bot IV Top
Fig. 11.12 3D schematic model of magnetic anomaly distribution indicating low anomaly (L1–
L12) at surface, 20L1–20L9 at 20 m depth, 40L1–40L10 at 40 m depth and 80L1–80L8 at 80 m
depth) and high anomaly (H1–H9 at surface, 20H1–20H10 at depth of 20 m, 40H1–40H10 at 40 m
depth and 80H1–80H9 at 80 depth) regions are generally caused by coal active fire and coal fire-free
regions, respectively (after Pal et al. 2017; Vaish 2018)
about 140 10−5 SI to 160 10−5 SI while some samples are diamagnetic
(−50 10−5 SI to −90 10−5 SI) in nature. Magnetic susceptibility of baked rock
samples of old coal fire regions is about 420 10−5 SI to 590 10−5 SI whereas
fresh rock samples of non coalfire regions are about 12 10−5 SI to 18 10−5 SI.
A positive magnetic susceptibility is produced from the presence of iron, nickel or
cobalt bearing minerals within the rocks or magnetic alignment moments within
these minerals (Hooper 1987). Sandstones, shales being overburden rocks have
magnetic mineral particles or traces. When these overburden rocks are subjected to
heating to the temperatures approaching to the Curie temperature, any natural
magnetic alignment of the magnetic moments and thus magnetization is random-
ized that are present in the study area.
So the heating above Curie temperature leads to the loss of natural magnetism
loss; a reduction in the total magnetic field locally in the coal fire region. When the
burning fire migrates from its region and the overburden is cooled down, the
magnetic moments become aligned to the Earth’s ambient magnetic field, produces
the enhancement in the magnetic field which is ascribed to the newly formed
magnetite in overburden and the new preferential alignment of magnetic moments
(Gielisch 2007; Ide et al. 2011). In the thermally altered rock, the magnetic
enhancement is due to the formation of magnetite in the coal fire affected shales (Ide
and Orr 2011). The magnetite and hematite formation process in overburden is a
strong function of O2 that are obtained at temperatures (200–1000 °C). Magnetite is
formed at low O2 concentrations in deeper multi seam fire or area covered with
overburden formation of while hematite is formed at high O2 concentrations in
shallower coal fires with high O2 circulation through wide spread fissures, cracks,
pot holes (Hooper 1987). Sedimentary rocks (siltstones, shales) over coal seam fires
contains magnetic minerals so it is suitably being the perfect environment for the
magnetite formation. Hematite (Fe2O3) is produced from the shales and siltstones
that are affected due to subsurface coal fire and thus does not increase the mag-
netization of a region upon the formation (Ide et al. 2011).
Being usefulness of magnetometer surveys, along with other coal fire charac-
terisation methods, magnetometer surveys could create interruptions at some
coalfire sites. Metallic materials (water tanks, train and power lines, or other
metallic objects) on nearby surface will considerably distort the results which
generate spikes and dropped signals. In present study, only good quality signal
readings are used for analysis while spikes and dropped signals are removed. If
metallic objects are left out which were used during active mining actives in
abandoned mines, coal fire characterisation may not be useful for that suitability
conditions for magnetometer surveys. Hidden metallic objects create hindrances as
it produces greater anomalies that will outshine the smaller anomalies associated
with coal fires (Ide et al. 2011).
324 S. K. Pal and J. Vaish
The radially averaged power spectrum (RAPS) of magnetic data has been calculated
in MAGMAP of Oasis Montaj for estimation of depth to the different layers over
study area. RAPS and depth estimate for East Basuria Colliery, Jharia Coal field are
shown Fig. 11.13. Best fit slopes are drawn on the RAPS for depth evaluation. The
average depth to the top of different layers are (i) 10 m (VII Seam, L3 Seam SBJ4;
Fig. 11.2), (ii) 18 m (V/VI Seam, SBJ4, Fig. 11.2) and (iii) 38 m (L2 Seam, VIII
Seam, Fig. 11.2).
Further, different sets of Euler’s depth solutions have been estimated separately
for each SI (0, 0.5, 1.0, 1.5, 2, 2.5 and 3) to find possible tightest cluster corre-
sponding to possible complex source geometry using various window sizes
(10 10, 15 15 and 20 20) and depth tolerances (10, 15 and 20%). Different
grid cell size has also been attempted during calculations. Relatively better clus-
tering observed for window size of 10 10 and depth tolerance 15%. The solu-
tions for SI = 0, 0.5, 1.0, 1.5, 2.0, 2.5 and 3.0 values are plotted on SV map which
are shown in Fig. 11.14a–g (Vaish 2018), respectively, using colour symbols for
different depth range.
Only a few number of solutions are found for SI = 0 (Fig. 11.14a) with depth
range of about 8–40 m with prominent linear clustering pattern which indicate
infinite contact/fault. This could be supported very well by the in situ field pho-
tograph of a wide and long fracture of the same area (Fig. 11.6d). However, the
Euler’s depth solutions with SI = 0.5 and more (SI = 1, 1.5, 2, 2.5 and 3) generate
progressively larger numbers of clustered solutions with a particular prominent
pattern. It is interesting to mention that the clustering pattern, position, extension
and depth range of the solutions estimated using SI = 0.5 remain almost same for
all higher order of fractional and integer structural index (SI = 1, 1.5, 2, 2.5 and 3)
as well, except some additional clustered solutions of deeper depth. A structural
index that is too low gives too shallow depth, while structural index that one that is
too high gives estimate too deeper depth (Reid et al. 1990). Depth estimation
provide better results for higher index sources than lower sources. Even through the
SI is correct, it is obvious that depth estimation is much better for SI sources than
for low index sources (Reid and Thurston 2014; Reid et al. 1990). Previous studies
indicate that SI variation within 0.25–0.5 range for the causative body generally
produce well defined and focussed solutions (Geosoft Inc. M-TR40). Moreover, the
clustering patterns are well correlated with all the delineated low anomaly patches
as observed on SV anomaly map. As such, Euler deconvolution technique is both a
depth as well as boundary locator of magnetic anomaly. The depth range of the
estimated solutions could be well interrelated with the various coal seam occur-
rences of Borehole lithology (Fig. 11.2).
The wide range of variations in the SI values of both fraction and integer with
similar clustering pattern in only particular locations indicate complex nature of
possible fire source/fire transmission along various inclined/horizontally extend
fractured planes generated in multiphase coal seam combustion associated with
11 Coal Fire Study Over East Basuria Colliery 325
Fig. 11.13 RAPS and depth estimate of East Basuria Colliery, for the year 2012 showing three
major layers with depth to the top of (i) 1st layer is 10 m, (ii) 2nd layer is 18 m and (iii) 3rd layer is
38 m (after Vaish 2018)
complex source geometries (fire through contact/infinite contact, thick step/sill i.e.,
underground fire in horizontal coal seam/dike i.e., fracture with long and wide
horizontally extended fracture/vertical cylindrical opening through potholes)
(Fig. 11.14).
11.4 Conclusions
Primarily, PSO inversion technique has been carried out effectively utilized using
Self potential data and characterise significantly the synthetic models of various
geometrical shapes via sphere, cylinder and inclined sheet with wide horizontal
extension. In study area, the PSO inversion of Self potential data provide results
that are in very good resemblance showing the low misfit error percentages
(0.0412–1.0342%). It illustrates that the extension of coal fire is between 10 and
12 m depth below subsurface. The PSO inversion of Self potential data indicates
that the geometrical shape of subsurface coal combustion zone is feasibly in good
agreement to the inclined sheet with wide horizontal extension. The larger variation
in inclined angles may possibly indicate complex nature of coal fire propagation
along different inclined fractured planes generated in multiphase coal seam com-
bustion zone.
Most of the coal fire affected areas are delineated in RTP magnetic anomaly
map, which are correlated well as observed in surface. Pole reduction causes
removal of lateral shifts and asymmetries due by the local orientation of the
magnetic field. The First vertical derivative (FV) anomaly (Fig. 11.9a) of pole
326 S. K. Pal and J. Vaish
(a)
nT/ m2
11.60
6.10
3.60
2.50
1.60
1.00
0.68
0.40
0.22
0.05
-0.19
-0.45
-0.70
-1.18
-1.67
-2.70
-4.60
-7.90
-18.90
(b)
nT/ m2
11.60
6.10
3.60
2.50
1.60
1.00
0.68
0.40
0.22
0.05
-0.19
-0.45
-0.70
-1.18
-1.67
-2.70
-4.60
-7.90
-18.90
Fig. 11.14 a Second vertical derivative map with superimposed Euler’s depth solution with
SI = 0.0, b second vertical derivative map with superimposed Euler’s depth solution with SI 0.5,
c second vertical derivative map with superimposed Euler’s depth solution with SI = 1.0, d second
vertical derivative map with superimposed Euler’s depth solution with SI 1.5, e second vertical
derivative map with superimposed Euler’s depth solution with SI, f second vertical derivative map
with superimposed Euler’s depth solution with SI 2.5, g second vertical derivative map with
superimposed Euler’s depth solution with SI 3.0
11 Coal Fire Study Over East Basuria Colliery 327
(c)
nT/ m2
11.60
6.10
3.60
2.50
1.60
1.00
0.68
0.40
0.22
0.05
-0.19
-0.45
-0.70
-1.18
-1.67
-2.70
-4.60
-7.90
-18.90
(d)
nT/ m2
11.60
6.10
3.60
2.50
1.60
1.00
0.68
0.40
0.22
0.05
-0.19
-0.45
-0.70
-1.18
-1.67
-2.70
-4.60
-7.90
-18.90
(e)
nT/ m2
11.60
6.10
3.60
2.50
1.60
1.00
0.68
0.40
0.22
0.05
-0.19
-0.45
-0.70
-1.18
-1.67
-2.70
-4.60
-7.90
-18.90
(f)
nT/ m2
11.60
6.10
3.60
2.50
1.60
1.00
0.68
0.40
0.22
0.05
-0.19
-0.45
-0.70
-1.18
-1.67
-2.70
-4.60
-7.90
-18.90
(g)
nT/ m2
11.60
6.10
3.60
2.50
1.60
1.00
0.68
0.40
0.22
0.05
-0.19
-0.45
-0.70
-1.18
-1.67
-2.70
-4.60
-7.90
-18.90
reduction magnetic data provides sharper resolution for delineation of coal fire
affected and non-coal fire regions than the simple pole reduction magnetic data
(Fig. 11.5b). Whereas, the Second vertical derivative (SV) anomaly (Fig. 11.9b) of
pole reduction magnetic data demarcates approximate shape and frameworks of
coal fire and fire free regions better than that of the First vertical derivative
(FV) anomaly (Fig. 11.9a). Tilt derivative (Fig. 11.11) provides better response
from both weak and strong anomalies. Most of the magnetic anomaly, vertical
derivative anomaly and tilt derivatives anomaly are matched. From 3D schematic
model of downward continue anomaly distribution, principal anomalies with
approximate source depth are well correlated with multi seam occurrences, mine
working levels and the possible depth inferred by Tilt derivative method. The
magnetometer data along with an integrated vertical derivative, Downward con-
tinuation of RTP data and Tilt derivative approach offers a noble means for
understanding of mine fire development and its detail study. This could be done
very efficiently by repeated study of the affected area in time lapsed magnetic
survey. It is observed that most of the corresponding anomalies (DX, DY, FV, SV,
THD, AS, TD and downward continuations at different depths are well correlated
with their probable locations.
The estimated depth range based on Euler solutions and RAPS are interrelated
well with the different coal seam occurrences as of Borehole lithology (Fig. 11.2).
Interestingly, depth estimated based on SP data also correlate well with Euler
solution, RAPS, Borehole lithology (Fig. 11.2).
330 S. K. Pal and J. Vaish
References
Banerjee SC (1982) A theoretical design to the determination of risk index of spontaneous fires in
coal mines. J Mines Met Fuels 30:399–406
Banerjee SC (1985) Spontaneous combustion of coal and mine fires. A. A. Balkema, Rotterdam.
ISBN: 9061915740
Banerjee SC, Nandy DK, Banerjee DD, Chakravorty RN (1972) Classification of coal with respect
to their susceptibility to spontaneous combustion. Trans Min Metall Insti India 59(2):15–31
BCCL (2009) East Basuria Colliery plan of underground working and Borehole lithologs
Bell FG, Bullock SET, Hälbich TFJ, Lindsay P (2001) Environmental impacts associated with an
abandoned mine in the Witbank Coalfield, South Africa. Int J Coal Geol 45(2–3):195–216
Bharat Coking Coal (2003) Brief history of BCCL Dhanbad, Bharat Coking Coal Ltd. http://bccl.
nic.in/About-us.htm
Bharti AK, Pal SK, Vaish J (2014) Application of self-potential method for coal fire detection over
Jharia Coal field. In: 51st Annual convention of Indian Geophysical Union, Kurukshetra
University, Kurukshetra, 19–21 November 2014, pp 59–62
Bharti AK, Pal SK, Priyam P, Kumar S, Srivastava S, Yadav PK (2016) Subsurface cavity
detection over Patherdih colliery, Jharia Coalfield, India using electrical resistivity tomography.
Environ Earth Sci 75(5)
Bharti AK, Pal SK, Priyam P, Pathak VK, Kumar R, Ranjan SK (2016) Detection of illegal mine
voids using electrical resistivity tomography: The case-study of Raniganj coalfield (India). Eng
Geol 213:120–132
Bhattacharya BB, Roy N (1981) A note on the use of nomograms for self-potential anomalies.
Geophys Prospect 29:102–107
Boekemeier R, Wang H, Zhu L, Elleringmann, S (2002) Höllenfahrt durch China. Geo 9, p 2003
Bustin RM, Mathews WH (1982) In situ gasification of coal, a natural example: history, petrology,
and mechanics of combustion. Canadian J Earth Sci 19(3):514–523
Bustin RM, Mathews WH (1985) In situ gasification of coal, a natural example: additional data on
the Aldridge Creek coal fire, southeastern British Columbia. Canadian J Earth Sci 22(12):
1858–1864
Coal Atlas of India (1993) Central Mine Planning and Design Institute Publications, Ranchi,
pp 1–139
Coal Directory of India (1998–99) Government of India, Ministry of Coal, Coal Controller’s
Organization, Calcutta
Coal Research (1998) Success stories: highlights and applications. Publication of Ministry of Coal,
Govt. of India, New Delhi
Chaiken RF (1980) Controlled burnout of wasted coal on abandoned coal mine lands.
BuMines RI 8478, 23pp
Chaiken RF, Brennan RJ, Heisey BS, Kim AG, Malenka WT, Schimmel JT (1983) Problems in
the control of anthracite mine fires: a case study of the Centralia Mine Fire (August 1980)
Chaiken RF, Brennan RJ, Heisey BS, Kim AG, Malenka WT, Schimmel JT (1998) Problems in
the Control of Anthracite Mine Fires: A Case Study of the Centralia Mine Fire. Report of
Investigations 8799. U.S. Department of the Interior, Pittsburgh, PA, 68p
Chandra D (1992) Jharia Coalfields, Geol Soci India, Bangalore, pp 149
Chen L (1997) Subsidence assessment in the Ruqigou coalfield, Ningxia, China, using a
geomorphological approach. M.S. Thesis, ITC, Enschede, The Netherlands
CMPDI (2003) Scheme for dealing with fire for protection of KT link line at East Basuria colliery,
pp 1–16
Cooper GRJ, Cowan DR (2004) Filtering using variable order vertical derivatives. Comput Geosci
30(5):455–459
de Boer JF, Milner TE, van Gemert MJC, Stuart Nelson J (1997) Two-dimensional birefringence
imaging in biological tissue by polarization-sensitive optical coherence tomography. Optics
Letters 22(12):934
11 Coal Fire Study Over East Basuria Colliery 331
Deng W, Wan YQ, Zhao RC (2010) Detecting coal fires with a neural network to reduce the effect
of solar radiation on Landsat Thematic Mapper thermal infrared images. Int J Remote Sens
22(6):933–944
DGMS Government of India (2011) Statistics of mines in India, Volume 1 (Coal) Directorate
General of Mines Safety, Ministry of Labour & Employment, Government of India
Dhar BB (1996) Keynote address on status of mine fires – trends and challenges. In: Proceedings
of the conference on prevention and control of mine and industrial fires – trends and
challenges, Calcutta, India, pp 1–8
Gangopadhyay PK (2006) Application of remote sensing to identify coalfires in the Raniganj
coalbelt, India. Int J App Earth Obser Geoinfor 8:188–195
Ghose MK, Majee SR (2000) Sources of air pollution due to coal mining and their impacts in
Jharia coalfield. Environ Int 26(1–2):81–85
Gielisch H (2002) Loeschung von Kohlebraenden – Statusbericht zum GTZ Projekt. Internal
unpublished report of the DMT (GermanMontan Technology) on a project on coal fire
extinction in China within the framework of a German GTZ project, 52 pp
Gielisch H (2007) Detecting concealed coal fires. Rev Eng Geol 18:199–210
Gijbels MR, Bruining J (1982) Spontaneous ignition time of underground coal-estimation from
oxidation rates determined in the laboratory. Erdol & Kohle erdgas Petrochemie 35(8):376–379
Glover L (1998) Burning beneath the surface. Tribune Review, May 3, USA
Grauch VJS, Cordell L (1987) Limitations of determining density or magnetic boundaries from the
horizontal gradient of gravity or pseudogravity data. Geophysics 52:118–124
Guan H (1989) Applications of remote sensing techniques in coal geology. Acta Geologica Sinica
2:253–269
Guan H, Van Genderen JL (1997) Report on environment monitoring of spontaneous combustion
in the coal fields of North China. Aerophotogrammetry and Remote Sensing Bureau of China
Coal, Xi’an, China, and International Institute for Aerospace Survey and Earth Sciences,
Enschede, The Netherlands (in Chinese)
Guan H, Van Genderen JL, Schalke H (1996) Study and survey on the geological hazards of coal
fire in North China. In: Proceedings of the 30th International Geological Congress Abstracts,
Issue 1, p 458
Holtberg P (2011) International Energy Outlook 11
Hooper RL (1987) Factors affecting the magnetic susceptibility of baked rocks above a burned
coal seam. Int J Coal Geol 9:157–169
Ide TS, Orr FM Jr (2011) Comparison of methods to estimate the rate of CO2 emissions and coal
consumption from a coal fire near Durango, CO. Int J Coal Geol 86:95–107
Ide TS, Crook N, Orr FM Jr (2011) Magnetometer measurements to characterize a subsurface coal
fire. Int J Coal Geol 87:190–196
Jia Y (2002) Personal interviews during the field campaigns (Y. Jia is the chief engineer of the coal
fire fighting team Wuda)
Kim AG, Chaiken RF (1993) Fires in abandoned coal mines and waste banks, vol. 9352. US
Department of the Interior, Bureau of Mines
Kuenzer C, Zhang J, Hirner A, Bo Y, Jia Y, Sun Y (2008) Multi temporal insitu mapping of the
Wuda coal fires from 2000 to 2005 assessing coal fire dynamics. UNESCO Beijing, 2008,
Spontaneous Coal Seam Fires: Mitigating a Global Disaster: ERSEC ecological book series 4,
pp 132–148
Livingood M, Winicaties J, Stein J (1990) Centralia mine fire analysis: presence of sulfur-bearing
mineral deposits at thermal vents. Unpublished report at West Chester University,
Pennsylvania 9
Lohrer C, Schmidt M, Krause, U (2005) Influence of Environmental Parameters on the Self
Ignition of Coal. In: Proc. of the Int. Conf. on Coal Fire Research, Beijing, China, pp110–112
Ma G (2013a) Edge detection of potential field data using improved local phase filter. Explor
Geophys 44:6–41
Ma G (2013b) Combination of horizontal gradient ratio and Euler (HGR-EUL) methods for the
interpretation of potential field data. Geophysics 78:J53–J60
332 S. K. Pal and J. Vaish
Ma G, Li L (2012) Edge detection in potential fields with the normalized total horizontal
derivative. Comput Geosci 41:83–87
Michalski SR (2004) The Jharia mine fire control technical assistance project: an analysis. Int J
Coal Geol 59(1–2):83–90
Miller HG, Singh V (1994) Potential field tilt—a new concept for location of potential field
sources. J Appl Geophys 32:213–217
Mishra RK, Bahuguna PP, Singh VK (2011) Detection of coal mine fire in Jharia Coal Field using
Landsat-7 ETM+ data. Int J Coal Geol 86(1):73–78
Mohalik NK, Singh RVK, Sural G, Barnwal RPJ, Pandey J, Singh VK (2004) Environmental
impacts of coal mine fire during excavation of developed galleries by opencast method. Indian
Min Eng J. 33:30–35
Mohalik NK, Lester E, Lowndes IS, Singh VK (2016) Estimation of greenhouse gas emissions
from spontaneous combustion/fire of coal in opencast mines – Indian context. Carbon
Management 7(5–6):317–332
Mukherjee TK, Bandopadhyay TK, Pande SK (1991) Detection and delineation of depth of
subsurface coalmine fires based on an airborne multispectral scanner survey in a part of the
Jharia Coalfield, India. Photogram Eng Remote Sens 57:1203–1207
Munshi PL (1995) Experience of dealing with fires in Jharia coalfield – Introduction to diagnostic
technique,national seminar on minefires. Dept. of Mining Engineering, Institute of Technology,
Banaras Hindu University, pp 82–90
Murthy BVS, Haricharan P (1985) Nomograms for the complete interpretation of spontaneous
potential profiles over sheet like and cylindrical 2D structures. Geophysics 50:1127–1135
Nabighian MN (1972) The analytic signal of two-dimensional magnetic bodies with polygonal
cross-section: its properties and use for automated anomaly interpretation. Geophysics 37:
507–517
Nabighian MN (1974) Additional comments on the analytical signal of two dimensional magnetic
bodies with polygonal cross section. Geophysics 39(1):85–92
Nabighian MN (1984) Toward a three dimensional automatic interpretation of potential field data
via generalized Hilbertz transforms: fundamental relations. Geophysics 49(6):780–786
Narayan S, Sahoo SD, Pal SK, Kumar U, Pathak VK, Majumdar TJ, Chouhan A (2016)
Delineation of structural features over a part of the Bay of Bengal using total and balanced
horizontal derivative techniques. Geocarto Int. 32(1):1–16
Page SE, Siegert F, Rieley JO, Boehm H-DV, Jaya A, Limin S (2002) The amount of carbon
released from peat and forest fires in Indonesia during 1997. Nature 420(6911):61–65
Pal SK, Vaish J, Kumar S, Bharti AK (2016) Coalfire mapping of East Basuria Colliery, Jharia
coal field using vertical derivative technique of magnetic data. J Earth Syst Sci 125(1):165–178
Pal SK, Vaish J, Kumar S, Priyam P, Bharti AK, Kumar R (2017) Downward continuation and Tilt
Derivative of magnetic data for delineation of concealed coal fire in East Basuria Colliery,
Jharia coal field, India. J Earth Syst Sci 126(53):1–17. https://doi.org/10.1007/s12040-017-
0826-y
Pandey J, Kumar D, Singh VK, Mohalik NK (2016) Environmental and socio-economic impacts
of fire in Jharia coalfield, Jharkhand, India: an appraisal. Curr Sci 110(9):1639
Pant PR, Murty BVS (2004) Geophysical exploration for coal: the Indian scenario (No. 59). Geol
Soci India
Pennig L (2003) Vergleich zweier Kohlebrandregionen in Nordchina mit Hilfe von
Fernerkundungs-daten und GIS. M.S. Thesis, University of Jena, Germany. Ph.D thesis at
the Ludwig Maximilians University Munich, Germany, 155 pp
Prakash A, Saraf AK, Gupta RP, Sundaram RM (1995) Surface thermal anomalies associated with
underground fires in Jharia coal mines, India. Int J Rem Sens 16:2105–2109
Prakash A, Fielding EJ, Gens R, Van Genderen JL, Evans DL (2010a) Data fusion for
investigating land subsidence and coal fire hazards in a coal mining area. Int J Rem Sens 22(6):
921–932
Prakash A, Gupta RP, Saraf AK (2010b) A Landsat TM based comparative study of surface and
subsurface fires in the Jharia coalfield, India. Int J Rem Sens 18(11):2463–2469
11 Coal Fire Study Over East Basuria Colliery 333
Rathore CS, Wright R (1993) Monitoring environmental impacts of surface coal mining. Int J Rem
Sens 14(6):1021–1042
Reid AB, Thurston JB (2014) The structural index in gravity and magnetic interpretation: errors,
uses, and abuses. Geophysics 79(4):J61–J66
Reid AB, Allsop JM, Granser H, Millett AJ, Somerton IW (1990) Magnetic interpretation in three
dimensions using Euler deconvolution. Geophysics 55:80–91
Revil A, Jardani A (2013) The self-potential method: theory and applications in environmental
geosciences. Cambridge University Press, pp 367
Revil A, Karaoulis M, Srivastava S, Byrdina S (2013) Thermo electric self-potential and resistivity
data localize the burning front of underground coal fires. Geophysics 78(5):B259–B273
Roest WR, Pilkington M (1993) Identifying remanent magnetization effects in magnetic data.
Geophysics 58(5):653–659
Roest W, Verhoef J, Pilkington M (1992) Magnetic interpretation using the 3-D analytic signal.
Geophysics 57:116–125
Santos FAM (2010) Inversion of self-potential of idealized bodies anomalies using particle swarm
optimization. Comput Geosci 36(9):1185–1190
Saraf AK, Prakash A, Sengupta S, Gupta RP (2007) Landsat-TM data for estimating ground
temperature and depth of subsurface coal fire in the Jharia coalfield, India. Int J Rem Sens
16(12):2111–2124
Schmal D (1987) A model for the spontaneous heating of stored coal. 2003 Scheme for dealing
with fire for protection of KT link line at East Basuria colliery, pp 16
Schneider W (1996) The coal-bearing Jurassic at the southern margin of the Junggar basin,
Xinjiang. Geowissenschaften 14:285–287
Singh BB, Srivardhan V, Pal SK, Kanagaraju SK, Kumar S, Vaish J (2015) Particle swarm
optimization inversion of self potential anomaly for detecting coal fires, a case study - Jharia
coal field. In: Third sustainable earth and sciences conference in Celle, Germany, EAGE, 13
October. https://doi.org/10.3997/2214-4609.201414282
Sinha PR (1986) Mine fires in Indian coalfields. Energy 11(11–12):1147–1154
Srivardhan V, Pal SK, Vaish J, Kumar S, Bharti AK, Priyam P (2016) Particle swarm optimization
inversion of self-potential data for depth estimation of coal fires over East Basuria colliery,
Jharia coalfield, India. Environ Earth Sci 75(8):688, 1–12. https://doi.org/10.1007/s12665-015-
5222-9
Stracher GB, Taylor TP (2004) Coal fires burning out of control around the world: thermodynamic
recipe for environmental catastrophe. Int J Coal Geol 59(1–2):7–17
Vaish J (2018) Time-lapse study of coal fire characterization over parts of Jharia coalfield, India.
Ph.D. thesis, Department of Applied Geophysics, IIT (ISM) Dhanbad, p 204
Vaish J, Pal SK (2015) Subsurface coal fire mapping of East Basuria Colliery, Jharkhand. J Geol
Soci India 86(4):438–444
Vaish J, Pal SK (2016) Subsurface coal fire mapping of Patherdih Colliery, a part of Jharia coal
field, India. J Geol Soc India Spec Publ 4:80–85. https://doi.org/10.17491/cgsi/2016/95899
Van Genderen, J L, Guan, H (1997) Environmental monitoring of spontaneous combustion in the
North China coalfields. Final report to European Commission under contract No. Cl1–CT93–
0008
Walker S (1999) Uncontrolled fires in coal and coal wastes. London, Great Britain, International
Energy Agency, IEA on Coal Research, pp 73
Wijns C, Perez C, Kowalczyk P (2005) Theta map: edge detection in magnetic data. Geophysics
70(4):39–43
World Bank Group (2003) Jharia mine fire control technical assistance project. Project ID
INP010411, Borrower: Bharat Coking Coal Limited. Retrieved 4 June, Web. http://www4.
worldbank.org/sprojects/Project.asp?pid=P010411
Yang H (1995) Detection of areas of spontaneous combustion of coal using airborne and TM data
in Xinjian, China. MSc thesis, International Institute for Aerospace Survey and Earth Sciences
Zhang XM (1998) Coal fires in Northwest China — Detection, monitoring, and prediction using
remote sensing data. Ph.D. dissertation, Int. Inst. Geo-Inform. Sci. and Earth Observ. (ITC)
334 S. K. Pal and J. Vaish
Zhang J (2004) Spatial and statistical analysis of thermal satellite imagery for extraction of coal fire
related anomalies. PhD Thesis, TU Vienna and ITC-Enschede, Vienna
Zhang X, Kroonenberg SB (1996) Pleistocene coal fires in Xinjiang, northwest China. In:
Abstracts 30th International Geological Congress, 457
Zhang X, Kroonenberg SB, De Boer CB (2004) Dating of coal fires in Xinjiang, north-west China.
Terra Nova 16(2):68–74
Zhang X, Van Genderen JL, Kroonenberg SB (2010) A method to evaluate the capability of
Landsat-5 TM band 6 data for sub-pixel coal fire detection. Int J Rem Sens 18(15):3279–3288
Zhang T, Xu J, Zeng J, Lou K (2013) Diversity of prokaryotes associated with soils around
coal-fire gas vents in MaNasi Country of Xinjiang, China. Antonie Van Leewenhoek
103:23–36
Zutshi A, Ray SK, Bhowmick BC (2001) Indian coal vis-à-vis spontaneous heating problems.
J Mines Met Fuels 44:123–128
Chapter 12
Geothermal Potential and Circulation
Depth of Hüdai Thermal Springs
(Sandıklı-Afyonkarahisar, Türkiye)
Using Magnetic, Geothermometry
and Heat Flow Data
N. MADEN (&)
Department of Geophysics, Gümüşhane University, Gümüşhane, Türkiye
e-mail: nmaden@gumushane.edu.tr
M. Afşin
Department of Geology, Aksaray University, Aksaray, Türkiye
e-mail: mafsinbaglum@gmail.com
F. Aksever A. Davraz
Department of Geology, Süleyman Demirel University, Isparta, Türkiye
e-mail: fatmaaksever@sdu.edu.tr
A. Davraz
e-mail: adavraz@gmail.com
Keywords Western Anatolia Curie point depth Heat flow Geothermal fields
Reservoirs temperature Circulation depth
12.1 Introduction
Fig. 12.1 Generalized Turkish tectonic map indicating major neotectonic provinces and structures
(Barka 1992; Bozkurt 2001). EAF, East Anatolian Fault; AEZ, Aegean Extension Zone; NAF,
North Anatolian Fault; SLF, Salt Lake Fault; EF, Ecemis Fault. The plate motion direction is
denoted with bold black arrows. Strike-slip faults are shown with half arrows pointing relative
movement of plate. The fold and thrust belt are indicated with filled triangles. Active subduction
zone is shown with open triangles. The thick lines with hachures mark normal faults where
hachures show the down-thrown side. The black box shows the study region
12 Geothermal Potential and Circulation Depth of Hüdai Thermal … 337
Fig. 12.2 Tectonic and Geological sketch map of the Hüdai-Sandıklı basin (Öngür 1973; Afşin
1991). 1. Qal: Alluvium; 2. Travertine; 3. Tpl: Gravel, sand, silt, clay; 4. Tms: Conglomerate,
marl, limestone; 5. a: Andesite; 6. T: Tuff; 7. Basalt; 8. Jka: Karstic limestone-Akdağ formation; 9.
Jd: Shale_Derealanı formation; 10. Conglomerate, siltstone-Karatepe formattion; 11. Es: Siltstone,
marl, limestone-Seydişehir formation; 12. Ec: Dolomitic limestone-Çaltepe formation; 13. Eh:
Quartzite-Hüdai formation; 14. Pek: Phyllite-Kocayayla/Kestel formation; 15. Peks:
Marble-Kocayayla/Kestel formation; 16a. Fault; 16b. Possible fault; 17a. Normal fault; 17b.
Possible normal fault; 18a. Horizontal layer; 18b. Foliation strike and dip; 19a. Anticlinal axis;
19b. Synclinal axis;20a. Cold spring; 20b. Thermal-mineral spring; 21a. Karst features; 21b.
Stratum strike and dip; 22a. Stream (perennial); 22b. Stream (temporary); 23a. Highway; 23b.
Railway
magnetite) at which rocks vanish their magnetic characteristics. This signifies that
the deepest sources at greater temperatures are intrinsically non-magnetic. Former
studies state that the depth of Curie point differs with respect to geological factors.
Tanaka et al. (1999) stated that the depth levels of the Curie point are higher than
30 km in trenches, greater than 20 km in plateaus, equal to 15–25 km at
mid-ocean ridges and volcanic arcs, and less than approximately 10 km in
geothermal and volcanic areas. Generally, the oceanic Curie point depth values are
shallower than continental regions. High heat flow values appropriate to shallow
Curie point depths, which are observed in young volcanic zones, trenches and
back-arc basins (Yamano 1995).
The aim of this study is to identify the origin and geothermal waters circulation
depth in the Hüdai-Sandıklı geothermal field using magnetic, geothermometry and
heat flow data. The Fourier’s heat conduction law is employed to specify the heat
flow data using the value of geothermal gradient. Finally, a relationship is proffered
to establish between heat flow and Curie temperature depth.
Table 12.1 Lithostratigraphic relations of the geologic units and hydrogeological properties in
the Hüdai-Sandıklı region
Period Formation Symbol Lithology Hydrogeological
Properties
Quaternary Alluvium Qal Gravel, sand, and Aquifer
mudstone
Quaternary Travertine Qtr Travertine Aquifer
Pliocene Hamamçay Tp Limestone, sandstone, Aquifer
formation siltstone, conglomerate
Miocene Sandıklı Tsl Trachyte, trachyandesite, Basalts are
lavas andesite and basalt permeable
Miocene Soğucak Tsp Tuff, tuffite and agglomerate Agglomerate is
pyroclastics aquifer
Cretaceous Akdağ Cra Dolomitic limestone, Karstic aquifer
formation micritic limestone
Jurassic Derealanı Jd Carbonated sandstone, Generally
formation siltstone, claystone and aquiclude
clayey limestone
Triassic Karatepe P-TRk Conglomerate, sandstone, Aquitard
formation shale and sandy, clayey
limestone
Permian Seydişehir Pzs Sandstone, locally clayey Aquiclude
formation limestone lenses
Ordovician Çaltepe Pzç Dolomite, limestone Reservoir rock
formation
Cambrian Hüdai Pzh Quartzite and schist Reservoir rock
formation
Precambrian Kestel Pk Quartzitic sandstone, Aquiclude
formation phyllite,
quartz-chlorite-sericite
schist,
quartz-feldspar-chlorite
schist, feldspar-quartz
sericite schist and quartz
sericite schist
A number of authors (Ateş et al. 2005; Aydın et al. 2005; Dolmaz et al. 2005;
Bektaş et al. 2007) performed researches about the Curie temperature depth values
utilizing magnetic anomaly data in Western, Central and Eastern Turkey. In the
Eastern Anatolia plateau, Bektaş et al. (2007) stated that the depth to the Curie
temperature ranges from 22.6 to 12.9 km. In the Western Anatolia basin, the values
340 N. MADEN et al.
of Curie point level obtained by Dolmaz et al. (2005) were exhibited to change
between 8.2 and 19.9 km. The Curie temperature depth was calculated in Central
Anatolia varying between 7.9 and 22.6 km (Ateş et al. 2005). Around the Erciyes
and Cappadocia Volcanic regions, minimum Curie temperature depths were found.
Aydın et al. (2005) prepared the Turkish Curie isotherm map using the method of
power spectrum analysis utilizing magnetic anomaly data. In the map, the minimum
Curie temperature depths are observed in central and western Anatolia regions
changing between 6 and 10 km. However, the maximum Curie point depths are
evaluated in the Western Taurus and Eastern Pontide orogenic belts (20–29 km).
The mean Curie temperature depth value was estimated by Maden et al. (2009)
using spectral analysis technique implemented to the magnetic data of the Eastern
Pontides (22.2 km). Also, the Curie temperature depth levels were found to range
between 14.3 and 27.9 km within the southern portion of eastern Pontides mag-
matic chain. Maden (2009) computed the depth of Curie point values vary between
14.8 and 21.8 km from south to north of the Central Pontides. On the other hand,
Maden (2010) evaluated the magnetic data of the Erciyes and surrounding regions
and applied spectral analysis technique to compute the Curie temperature depth
value (13.7 km). The value of geothermal gradient was computed to be 42.3 °
C km−1 while the surface heat flow value was found to be 88:8 mWm2 . Maden
(2010) determined that heat flow is inversely proportional to Curie temperature
depth with an equation of q ¼ 1216:12 Zb1 .
The preliminary map of Turkish heat flow data was presented by Tezcan and
Turgay (1987) estimating geothermal gradients using constant geothermal con-
ductivity data collected from exploration boreholes. This map was then improved
by Tezcan (1995) using temperature data gathered from boreholes in the Thrace
basin and southeast Turkey. İlkışık (1992) prepared another Turkish heat flow map
using silica contents of thermal waters. It is apparent from this map that the mean
heat flow values for the central, western and eastern Anatolia region are
102:8 mWm2 , 110:7 mWm2 , and 112:8 mWm2 , respectively. The maximum
silica heat flow values were observed on the north of the Lake Van (80 mWm2 ),
Eastern Pontides orogenic belt (80 mWm2 ) and East Anatolian Fault zone
(100 mWm2 ). In another major study, the maximum heat flow values was iden-
tified in the inner Eastern Anatolia region increasing from 89 to 99 mWm2 by
employing the steady-state geothermal approach (Bektaş 2013).
The mean heat flow value (158:5 mWm2 ) conducted by İlkışık et al. (1997) was
obtained in the Ihlara valley using conventional and silica techniques. However, the
maximum surface heat flow rate (247 mWm2 ) was observed close to the Gediz
basin. İlkışık (1995) evaluated the average surface heat flow rates depending on the
conventional and silica data are 97 mWm2 and 107 mWm2 , respectively, in the
western Anatolia. Heat flow values vary between 110 and 130 mWm2 on the
Menderes Massif demonstrating a close relationship between younger volcanism
and high heat flow value (İlkışık 1995). Negative magnetic and Positive gravity
anomalies and high surface heat flow values indicated that the asthenosphere is
uplifted in the Menderes Massif (Düzgit et al. 2006). Maden (2009, 2010, 2012)
12 Geothermal Potential and Circulation Depth of Hüdai Thermal … 341
investigated the variations of heat flow rates in the Central Pontides of northern
Turkey and Central Anatolia Plateau. The rates of heat flow in Central Pontides
achieved from the depth of Curie temperature values which vary between
94:1 mWm2 in the south and 63:8 mWm2 in the north proving that the northern
portion has lower geothermal potential than southern portion of the region (Maden
2009). Eurasian plate motion beneath the Anatolian block might be possible reason
for this phenomenon. In the Central Pontides, Maden (2010) appreciated the heat
flow values of 106:5 and 88:8 mWm2 derived from the conventional and silica
heat flow techniques, respectively. The tectonic processes following Late Triassic
age as recommended by Bozkurt (2001) may be triggered to maximum surface heat
flow rates ranging from 66:5 to 104:7 mWm2 (Maden 2012) in the Central
Pontides.
The gravity and magnetic data of the study region with a grid interval of 5 km were
taken from the General Directorate of Mineral Research and Exploration (MTA) of
Turkey. The gravity and magnetic anomaly maps of the Afyon-Sandıklı basin are
shown in Fig. 12.3a, b. The gravity anomaly is dominated by negative anomalies
which are generally accompanied with the tectonic elements, grabens, low-density
sedimentary units and thin crust. The gravity anomaly values vary between −30 and
−130 mGal. While the positive gravity anomalies are dominant in the southwestern
part of the region, negative gravity anomalies are situated in the center of the
region. The negative gravity anomalies sighted around the Sandıklı basin express
the high geothermal structures and thick sedimentary succession. Mostly, sedi-
mentary succession is related to negative gravity anomalies because of low-density
sediment accumulation.
The magnetic flight lines, which are perpendicular to the geological and tectonic
structures, are 1–2 km. The interval of the magnetic data measurements throughout
the flight tracks is approximately 70 m. The International Geomagnetic Reference
Field (IGRF) was extracted to generate magnetic anomaly data (Ateş et al. 1999).
The reduced to North Pole magnetic anomaly data of the Hüdai-Sandıklı basin is
shown in Fig. 12.3. The magnetic anomaly values vary from −80 nT in the center
to 50 nT in the southeastern part of the study area. The high-frequency positive
high-amplitude magnetic events were observed over the Hüdai-Sandıklı region. The
negative magnetic anomalies are oriented NE-SW and NW-SE directions.
Topography, regional geological elements and magnetic core fields form total field
magnetic anomalies characterized by low frequencies and may influence the
determination of the depth to the centroid of a source and likely cause unsteady
342 N. MADEN et al.
Fig. 12.3 The gravity (a) and magnetic (b) anomaly map of the Afyon-Sandıklı region. The
gravity and magnetic counter interval is 5 mGal and 5 nT. Color bar at the right side indicates
counter interval values
Curie depths estimation (Okubo et al. 1985). A low-pass filter should be applied to
the magnetic anomaly data to intercept the topographical effects and then reduced to
North Pole (Maden 2013).
Spectral analysis method to find out the depth to underground geological structures
was recommended originally by Spector and Grant (1970). This method has pre-
viously been used to derive the average depths to the magnetic and gravity anomaly
sources (Bhattacharyya and Leu 1975; Byerly and Stolt 1977; Shuey et al. 1977;
Connard et al. 1983; Okubo et al. 1985; Blakely 1995; Okubo et al. 1989, 1991;
Poudjom Djomani et al. 1992; Rivero et al. 2002; Maden et al. 2009; Maden 2010,
2013).
Spector and Grant (1970) provided a technique to state the mean depths for the
upper boundary of magnetic sources, thorough the log power spectrum slope. This
technique supplies a relationship between the 2D FFT power spectrum of the data
and the depth of anomaly sources. Dimitriadis et al. (1987) have generalized this
technique to gravity anomaly data to determine the bottom depth of gravity
ensembles. The depth to the crustal density interfaces and the Curie temperature
values are determined (Nnange et al. 2000; Ateş et al. 2005; Dolmaz et al. 2005)
when the power spectral analysis technique is applied to the potential field data. The
anomaly data is transformed into the wave-number domain to analyze the
wave-number content of the data. A plot of the logarithm power spectrum versus
wave-number mostly indicates several straight-line portions which increase in
12 Geothermal Potential and Circulation Depth of Hüdai Thermal … 343
where k, m, and N are the wave number, the average spectral slope and number of
gravity data, respectively. The basement
upper boundary depth (d) could be simply
computed, since the value of 2p N d m equal to zero. The top boundary of the
anomaly sources is estimated by the traditional method of least-squares for fitting
line segments with low and high wave-number in the power spectrum curves.
The method of FFT has been used to transform the data from space domain to
frequency domain. Figure 12.4 demonstrates the plot of the logarithmic amplitude
spectra of the radial wave-number to estimate the depth of the crustal interfaces. In
this figure, the power spectrum analysis reveals three straight-line segments asso-
ciated with the depths of anomaly sources. The lowest wave-number segment of 0–
0.09 km−1 revealing the deepest anomaly source is appreciated by the density
contrast in the Moho surface (31.4 km). A middle wave-number portion between
0.09 and 0.25 km−1 is considered to be Conrad density interface (15.2 km). The
higher wave-number segment of 0.3–0.55 km−1 supplies the low-density sedi-
mentary units (5.3 km). On the other hand, Sari and Şalk (2006) estimated the
maximum thickness of sedimentary layer as 2.5 and 3.5 km in the Büyük Menderes
Graben and between 0.5 and 2.0 km in the Gediz Graben.
Some authors determined the crustal structure of the western Anatolia region.
Saunders et al. (1998) and Horasan et al. (2002) computed the depth of Moho in
western Anatolia as 30–33 km, respectively. In western Turkey, Akyol et al. (2006)
constructed 1D four layers crustal structure model. In that model, the Moho depth
and upper mantle velocity were defined as 29 km and 7.8 km s−1, respectively.
Bilim (2007) stated the crustal thickness of western Anatolia region changes
between 28 and 36 km by using empiric relationship proposed by Riad et al.
(1981).
To find out the bottom depth ðZb Þ of the underground geological structures,
Okubo et al. (1985) proposed a two steps technique employing the natural power
spectrum approach of Spector and Grant (1970). The discrete magnetic anomaly
data must be reduced to the pole before low-pass filter process. At first, the centroid
depth ðZo Þ, secondly, the top bound ðZt Þ of the magnetic source are derived from
the slope of the longest and second longest wavelength portion of the spectrum.
Finally, the Curie point depth ðZb Þ is computed from centroid and top bound depth
values employing the equation Zb ¼ 2Z0 Zt (Okubo et al. 1985; Tanaka et al.
1999).
The plot of power spectrum data of the magnetic anomaly given in Fig. 12.5a, b
is obtained with Fourier transformation method. The depth to the upper boundary
and centroid of the magnetic anomaly sources computed from the power spectrum
curves as Zt ¼ 3:3 km and Zo ¼ 7:4 km, respectively (Fig. 12.5a, b). The Curie
temperature depth (Zb ¼ 11:5 km) is estimated from the upper and centroid depths
to magnetic anomaly of geologic structures in the Hüdai-Sandıklı region. The
estimated depth level of Curie temperature is in harmony with the values of Dolmaz
et al. (2005) changing between 8.2 and 19.9 km in western Anatolia.
Fig. 12.5 Power spectrum plot of the magnetic anomaly data for the study region. a Top
boundary of the magnetic sources ðZt Þ; b centroid of the magnetic sources ðZo Þ acquired from the
slope of the power spectrum graph
12 Geothermal Potential and Circulation Depth of Hüdai Thermal … 345
dT
q¼k ð12:2Þ
dZ
where k is the geothermal conductivity and q is the heat flow value. In this formula,
the temperature was assumed to be vertically changing and the geothermal gradient
dZ is fixed. The rate of change of temperature (dT) with depth (dZ) is described as
dT
geothermal gradient within the earth crust. The Curie temperature value ðhÞ could
be determined from geothermal gradient dTdZ and Curie temperature depth ðZb Þ data
with following formula;
dT
h¼ Zb ð12:3Þ
dZ
data as presented by Tezcan (1979). Tezcan and Turgay (1987) estimated the
regional heat-flow values greater than 120 mWm2 in Western Anatolia. The mean
heat flow value is computed to be 107 mWm2 and 97 27 mWm2 using silica
geothermometry and conventional heat flow technique in Western Anatolia. On the
other hand, İlkışık (1995) determined that the heat flow values for Afyon are
108 mWm2 in the Sandıklı, 144 mWm2 in Gecek and to 155 mWm2 in mineral
water. In the Western Anatolia basin, the surface heat flow rates and depth to the
Curie point level range from 62 to 150 mWm2 and 8.2 to 19.9 km, respectively
(Dolmaz et al. 2005). The surface heat flow versus Curie temperature depth plot for
the research area reveals an inversional relationship with the equation Q ¼
1223:03 Zb1 (Fig. 12.6). The computed values of heat flow and Curie point
temperature are coherent with the values obtained by Ateş et al. (2005) in the
Central Anatolia and Bilim (2007) in western Anatolia.
substances were in equilibrium at depth (İlkışık 1995; Maden 2013; Davraz et al.
2016).
The geothermometers depend primarily on one or more dissolved constituents in
the geothermal fluid whose concentrations vary depending on the temperature of the
fluids. These constituents may be solutes, gases or isotopes. Therefore, geother-
mometers have been classified into three groups such as solute (water), gas (steam)
and isotope geothermometers. The solute and gas geothermometers are called
chemical geothermometers (Afşin et al. 2013; Davraz et al. 2016). Chemical
geothermometry has become an important tool for estimating reservoir temperatures
of hydrothermal systems and has proved very useful in determining the geothermal
resource potential of a specific region. The two worthy sorts of geochemical
geothermometers that are usually employed in geothermal exploration are based on
mineral solubility (silica-SiO2) and exchange reactions (Na–K, Na–K–Ca, etc.).
These geochemical geothermometers generally refer to silica and cation
geothermometers.
Silica geothermometers has been prepared on the chemical reaction affecting the
aquifer temperature or cooling of the hot water while rising. Various geother-
mometers equations used to estimate of reservoir temperature are formulated based
on dissolution of quartz increases with temperature (Table 12.2). The silica
geothermometry is based on experimentally determined variations in the solubility
TSiO2 ¼ mq þ b ð12:5Þ
where b is the national average air temperature value specified locally. The value of
the slope (m) is related to the average groundwater circulation depth. By using
Eq. (12.5), the silica heat flow could be computed from,
TSiO2 b
q¼ ð12:6Þ
m
It is accepted that the maximum circulation depth is between 1.5 km for m = 0.7
and 2 km for m = 1 (İlkışık 1995; Maden 2013).
@2T A0
¼ ð12:7Þ
@z2 k
The numerical solution of differential Eq. (12.7) establishes the origins for all
computations of steady-state geotherm vertically. In the event of constant RHP, the
heat equation from analytic solution can be derived from the following formula,
Qs A0
T ðzÞ ¼ T0 þ z z2 ð12:8Þ
k 2k
where k is the geothermal conductivity value in Wm1 K1 , Qs is heat flow rate in
the surface in mWm2 and A0 is the surface RHP value in lWm3 , T0 is the surface
temperature of the Earth in °C and T(z) is the temperature at depth z within the Earth
in °C (Cermák et al. 1991a, b).
It is essential to know the RHP and the geothermal conductivity values to
compute temperatures of the crustal layers. Table 12.3 demonstrates the crustal
velocity model used in this study to compute geothermal model of the study region.
12 Geothermal Potential and Circulation Depth of Hüdai Thermal … 349
Table 12.3 The crustal model, thermal conductivity and radiogenic heat production parameters
used for the geotherm calculation. The values of crustal thickness and P-wave velocity are taken
from Kalafat et al. (1987) and Akyol et al. (2006)
Layer Depth VP Geothermal conductivity Heat production
(km) (km s n−1) (Wm−1 K−1) (lWm3 )
Sediments 0–5 5.21 3.14a 2.20c
Upper 5–15 6.00 3.50b 1.80b
crust
Middle 15–21 6.25 3.00b 0.83d
crust
Lower 21–31.6 6.43 2.50b 0.37d
crust
Mantle 7.80 4.00b 0.002b
a
Rao et al. (1970)
b
Jokinen and Kukkonen (1999)
c
Rimi (1999)
d
He et al. (2009)
k0 ð1 þ bzÞ
k¼ ð12:9Þ
1 þ cT
k0
k¼ ð12:10Þ
1 þ cT
Ramalho (1999) suggested to utilize the geothermal conductivity values for the
upper, the middle and the lower crust as 2.7, 2.5 and 2:1 Wm1 K1 , respectively.
The value of geothermal conductivity for the sediments is 3:14 Wm1 K1 (Rao
et al. 1970). The geothermal conductivity values of granites and basalts are 2.4–3.8
and 1.3–2.9 Wm1 K1 , respectively, as recommended by Turcotte and Schubert
(1982). The geothermal conductivity values of the crustal model are given in
Table 12.3 in the study region.
Crustal radioactivity is influenced by magmatic intrusion, tectonics, and meta-
morphism in the uppermost crust (10 km). The values of RHP are associated with
the geochemical and lithological characteristics of the rocks. In the continental
lithosphere, the radiogenic heat producing elements distribution supplies a sub-
stantial thermal evolution restraint, the mantle depletion and the crustal evolution of
the Earth (Perry et al. 2006; Kumar et al. 2007). While high RHP values are
observed in upper crustal rocks, low RHP values indicates depleted mantle products
(Moisio and Kaikkonen 2006). Jokinen and Kukkonen (1999) utilized the RHP
rates for the upper, the middle and the lower crust and the lithospheric mantle as
1.8, 0.6, 0.2 and 0:002 lWm3 , respectively. The RHP value for sediments was
proposed by Rimi (1999) as 2:20 lWm3 . The RHP values of metamorphic rocks
vary from 1:5 to 3:5 lWm3 in the upper crust. According to the He et al. (2009),
the RHP values for the upper, the middle and the lower crusts are 1.10, 0.83, and
0:37 lWm3 , respectively. The RHP rates for the Hüdai-Sandıklı region used in
this computation are listed in Table 12.3.
The surface RHP values are estimated from concentrations of radioactive ele-
ments. Therefore, the P-wave velocity data is used to obtain the lower crustal RHP
values with the following equation suggested by Rybach and Buntebarth (1984),
where Vp (20,100) is the velocity (km s1 ) at room temperature (20 °C) and
pressure (100 MPa) and A is in lWm3 . The RHP values are specified as 3.64,
0.66, 0.38, 0.26 and 0:01 lWm3 for the sediments, upper, middle, lower crusts and
mantle by using P-wave seismic velocity data obtained from seismologic research
(Kalafat et al. 1987; Akyol et al. 2006).
Numerous models have been proposed in the literature that the RHP values
depending on the processes of magmatic differentiation are decreased as a function
of depth in the crustal layers (Cermák et al. 1991a, b). The widely referenced and
affirmed exponential model of Lachenbruch (1970) is given with the following
equation,
z
AðzÞ ¼ A0 e D ð12:12Þ
obtained from the reduced and the surface heat flow data ranges mostly between 5
and 15 km (Morgan and Sass 1984; Pasquale 1987).
The RHP as a function of depth plot is seen in Fig. 12.7a for the research area. In
this graph, a sudden decrease is observed in the crustal radioactivity differentiation
rates. The radioactively enriched crustal layer thickness was computed 7.8 km for
the study region. In the uppermost crustal layer, motion of groundwater along
micro-crack systems might be induce an enrichment of radioactive elements in the
shallow subsurface rocks (Čermák and Rybach 1989). The heat flow and temper-
ature values for the crustal layer of radioactively enriched are revealed as
97:6 mWm2 and 368 °C, respectively. The rates of RHP owing to the concen-
trations of K, Th and U isotopes supply major heat flow source within the crust (Rai
and Thiagarajan 2006).
The one-dimensional geothermal model is introduced to understand thermal
regime of the Hüdai-Sandıklı geothermal field (Fig. 12.7b). The numerical geo-
therm results confirm that the temperature at the base of Moho is 1140 °C based on
the crustal model for the study region (Fig. 12.7b). The depth of Curie temperature
is found at 13.8 km, which more or less consistent with the Curie point depth value
given by Dolmaz et al. (2005) and Bilim (2007). The heat flow value determined
from the geotherm calculations in the mantle and Curie point depth is determined as
80:9 and 91 mWm2 , respectively, in the region (Fig. 12.7c).
Fig. 12.7 a Heat production, b temperature and c heat flow versus depth profile for the Hüdai
geothermal field. The crustal thickness values are taken from Kalafat et al. (1987) and Akyol et al.
(2006). RE stands for radioactively enriched crustal layer depth denoted by D on the graphs
352 N. MADEN et al.
12.5 Results
CaCO3 by CO2-rich meteoric water leakage through groundwater flow paths. The
dissolution of carbonate mineral might be reduced along the groundwater circula-
tion from the recharge areas to the thicker portion of the aquifer due to the absence
of carbon dioxide (Afşin 1997). The SO42− and Cl− contents of geothermal waters
also change from 237.87 to 575.52 mg lt−1 and from 49.40 and 94.21 mg lt−1,
respectively.
The dominant constituent chemical composition yields are summarized in a
Piper diagram from which the hydrogeochemical facies/major water types were
concluded (Fig. 12.8). According to Piper diagram, geothermal waters in the
Hüdai-Sandıklı geothermal field are Na–Ca–SO4–HCO3 and Na–Ca–HCO3–SO4
hydrogeochemical facies. These facies represented very high ion content because of
extensive contact with rocks and old waters which have deep circulation within the
low hydraulic conductive aquifers. The TDS values of the geothermal waters are
highest within the groundwater samples. While the cations are controlled by Na+
with lower relative contents of Ca2+ and Mg2+, the anions are controlled by HCO3–
and SO42− (Afşin et al. 2013).
stability has been achieved (Fournier 1985). The most common cation geother-
mometers listed in Table 12.2 are silica geothermometers associated with temper-
ature changes in the solubility of various silica minerals and cation
geothermometers based on changes of Na, K, Ca, Mg contents.
This study associates hydrogeochemical data with conventional heat flow data,
after the silica heat flow interpretation method of Swanberg and Morgan (1980). In
this method, quartz geothermometer is applied to non-thermal ground waters. The
quartz geothermometer provides a useful tool to determine the subsurface tem-
perature. In the application stage, Truesdell (1976) and Verma (2000) discussed the
quartz geothermometer and its assumptions in detail.
The geothermal springs of Hüdai-Sandıklı geothermal field were appreciated in
triangular Na–K–Mg ternary diagram suggested by Giggenbach (1988). According
to Na–K–Mg ternary plot, whole data for geothermal waters close to the immature
water region seen in Fig. 12.9. The fluid–mineral equilibrium cannot be accom-
plished with this triangular ternary diagram. On the other hand, this diagram shows
the duration of water–rock interaction in a geothermal region. Hence, the results
acquired with the cation geothermometers should be considered as debatable
(Giggenbach 1988; Gupta and Roy 2007).
Silica geothermometers were employed to predict the geothermal waters tem-
peratures in the reservoir. According to chalcedony, quartz and quartz steam loss
geothermometers, geothermal waters reservoir temperature values range from 73.96
to 122.06 °C (Table 12.5). These results indicate that the Hüdai geothermal field
Table 12.5 Silica geothermometry results of Hüdai-Sandıklı geothermal waters for different
periods
Location Period Measured Chalcedony Quartz Quartz
temperature (°C) steam loss
AFS-11 October 54 86.64 115.47 114.30
2010
AFS-12 October 73 87.10 115.88 114.65
2010
AFS-15 October 65.36 90.20 118.74 117.08
2010
AFS-16 October 82.40 93.84 122.06 119.92
2010
AFS-11 May 2011 66.18 73.96 103.75 104.23
AFS-12 May 2011 73.40 74.91 104.63 104.98
AFS-13 May 2011 75.65 77.40 106.94 106.97
AFS-15 November 70.50 75.48 105.16 105.44
2011
AFS-16 November 76.80 84.10 113.13 112.29
2011
AFS-7 May 2012 66.40 80.47 109.78 109.42
Mineral May 2012 63 80.30 109.61 109.27
Spring
has a higher temperature than the highest measured temperature of 82.40 °C and
that the potential of geothermal energy is higher in the Hüdai geothermal region.
On the other hand, it is tried to define the reservoir temperature and depth of the
geothermal sources located in the Hüdai-Sandıklı region. Generally, geother-
mometers related to ratios are more robust to dilution effects than those based on
absolute concentrations (Banks et al. 1998). The conductive cooling equation for
quartz is used in this study to determine the reservoirs temperature, developed by
Fournier (1977), constitutes the principle of the quartz geothermometer. This
equation is expressed as follows:
1309
TSiO2 ¼ 273:15 ð12:13Þ
5:19 log SiO2
where SiO2 is the dissolved silica concentration in mg lt1 and TSiO2 is the tem-
perature of silica in °C. The estimated temperatures computed from geochemical
data collected by Akkuş et al. (2005) and Memiş (2010) based on the geother-
mometer equation of (12.13) are plotted in Fig. 12.10a–d and shown in Table 12.6.
The SiO2 content of the geothermal sources located in the Hüdai-Sandıklı region,
ranging between 63:32 and 76:8 mg lt1 , is shown in Fig. 12.10a. Temperatures of
reservoir rock of geothermal resources of the study region, whose surface tem-
peratures range from 66 to 75 °C, were determined between 108.02 °C (AFS-4)
356 N. MADEN et al.
Fig. 12.10 a SiO2, b surface and reservoir temperatures, c heat flow values and d depths to the
reservoir of geothermal sources located in the Hüdai-Sandıklı region
12.6 Conclusions
Table 12.6 The Reservoir temperature, heat flow, Curie point depth, geothermal gradient and
reservoir depth values of Hüdai-Sandıklı geothermal waters determined from silica geothermom-
etry results
Spring Surface SiO2 Reservoir Heat flow Curie Geothermal Reservoir
temperature (mg lt−1) temperature (mWm−2) point gradient (° depth
(°C) (°C) depth C km−1) (km)
(km)
AFS-1 70a 64a 113.69 102.39 11.94 48.56 0.90
AFS-2 69a 64a 113.69 102.39 11.94 48.56 0.92
AFS-4 66a 57a 108.02 96.72 12.64 45.87 0.92
AFS-6 69b 63.32b 113.16 101.86 12.01 48.31 0.91
AFS-9 69b 64.77b 114.29 102.99 11.88 48.84 0.93
AFS-11 73.6b 68.62b 117.18 105.88 11.55 50.21 0.87
AFS-12 73.5b 68.32b 116.96 105.66 11.57 50.11 0.87
AFS-15 75b 76.8b 122.96 111.66 10.95 52.95 0.91
Average 114.99 103.69 11.81 49.18 0.90
a
Akkuş et al. (2005)
b
Memiş (2010)
Acknowledgements This study was partially supported by the Scientific and Research Council of
Turkey (Grant No: 110Y034). We thank Mümin Hakkıoğlu (GU) and Necati ÇELİK (GU) for
their vigorous English-language editing. The authors are grateful to Arkoprovo Biswas (Ph.D.) for
his valuable comments and editorial advice to improve the quality of this manuscript. We would
also like to thank the General Directorate of the Mineral Research and Exploration (MTA) of
Turkey for supplying the data used in this study.
References
Afşin M (1991) Hydrogeological investigation of Afyon Sandıklı Kurucay plain and Hudai spa.
Ph.D. thesis, Ankara University, Ankara
Afşin M (1997) Hydrochemical evolution and water quality along the groundwater flow path in the
Sandıklı plain, Afyon, Turkey. Environ Geol 31:221–230
Afşin M, Canik B (1996) Hydrogeological and hydrochemical study of the Hudai (Sandikli,
Afyon, Turkey) thermal mineral springs and interpretation of their genesis. J Geosound 28:69–
86 (in Turkish)
Afşin M, Dağ T, Davraz A, Aksever F, Karakaş Z (2012) The origin and sustainability of Hudai
geothermal waters, Sandıklı, Afyonkarahisar, Turkey. 39th IAH Congress, 16–21 Sept 2012.
Niagara Falls, Canada, pp 288–289
Afşin M, Davraz A, Karakas Z, Hınıs MA, Aksever F (2013) Hydrogeological, hydrogeochemical
and isotopic of Hudai geothermal field and its vicinity, Sandikli-Afyonkarahisar, Turkey.
TUBITAK project no. CAYDAG-110Y034, unpublished report
Akkuş İ, Akıllı H, Ceyhan S, Dilemre A, Tekin Z (2005) Turkish geothermal sources inventory.
MTA, Ankara
Aksever F (2011) Hydrogeological investigations of the Sandıklı (Afyonkarahisar) Basin. Ph.D.
thesis, Suleyman Demirel University, Isparta, p 224
Akyol N, Zhu L, Mitchell BJ, Sözbilir H, Kekovalı K (2006) Crustal structure and local seismicity
in western Anatolia. Geophys J Int 166:1259–1269
Arnorsson S, Gunnlaugsson E, Svavarsson H (1983) The chemistry of geothermal waters in
Iceland. III. Chemical geothermometry in geothermal investigations. Geochim Cosmoschim
Acta 47:567–577
Ateş A, Kearey P, Tufan S (1999) New gravity and magnetic maps of Turkey. Geophys J Int
136:499–502
12 Geothermal Potential and Circulation Depth of Hüdai Thermal … 359
Ateş A, Bilim F, Büyüksaraç A (2005) Curie point depth investigation of Central Anatolian
Turkey. Pure Appl Geophys 162:357–371
Aydın I, Karat HI, Koçak A (2005) Curie-point depth map of Turkey. Geophys J Int 162:633–640
Banks D, Sletten RS, Haldorsen S, Dale B, Heim M, Swensen B (1998) The thermal springs of
Bockfjord, Svalbard: Occurrence and major ion hydrochemistry. Geothermics 27(4):445–467
Barka AA (1992) The north anatolian fault zone. Ann Tecton 6:164–195
Bektaş O (2013) Thermal structure of the crust in inner east Anatolia fromaeromagnetic and
gravity data. Phys Earth Planet Inter 221:27–37
Bektaş O, Ravat D, Büyüksaraç A, Bilim F, Ateş A (2007) Regional geothermal characterization
of East Anatolia from aeromagnetic, heat flow and gravity data. Pure Appl Geophys 164:
975–998
Bhattacharyya BK, Leu LK (1975) Spectral analysis of gravity and magnetic anomalies due to
two-dimensional structures. Geophysics 40:993–1013
Bilim F (2007) Investigation into the tectonic lineaments and thermal structure ofKutahya-Denizli
region western Anatolia, from using aeromagnetic, gravityand seismological data. Phys Earth
Planet Inter 165:135–146
Blakely RJ (1988) Curie temperature isotherm analysis and tectonic implications of aeromagnetic
data from Nevada. J Geophys Res 93:11817–11832
Blakely RJ (1995) Potential theory and its applications. Cambridge University Press, New York
Bozkurt E (2001) Neotectonics of Turkey—a synthesis. Geodin Acta 14:3–30
Byerly PE, Stolt RH (1977) An attempt to define the Curie point isotherm in northern and central
Arizona. Geophysics 42(7):1394–1400
Čermák V, Bodri L (1986) Temperature structure of the lithosphere based on 2D temperature
modeling applied to Central and Eastern Europe. Thermal modeling in sedimentary basins.
Editions Technip, Paris, pp 7–32
Čermák V, Rybach L (1989) Vertical distribution of heat production in the continental crust.
Tectonophysics 159(3–4):217–230
Čermák V, Král M, Krešl M, Kubík J, Šafanda J (1991a) Heat flow, regional geophysics and
lithosphere structure in Czechoslovakia and adjacent part of Central Europe. In: Terrestrial heat
flow and the lithosphere structure. Springer, Berlin, pp 133–165
Čermák V, Bodri L, Rybach L (1991b) Radioactive heat production in the continental crust and its
depth dependence. In: Cermak V, Rybach L (eds) Terrestrial heat flow and the lithosphere
structure. Springer, New York, pp 23–69
Chapman DS, Furlong KP (1992) Thermal state of the continental crust, Continental Lower
Crust DM Fountain. R Arculus, RW Kay, pp 179–200
Connard G, Couch R, Gemperle M (1983) Analysis of aeromagnetic measurements from the
Cascade Range in central Oregon. Geophysics 48(3):376–390
Correia A, Jones FW (1995) A magnetotelluric survey in a reported geothermal area in southern
Portugal. Proc World Geothermal Congress 2:927–931
Correia A, Ramalho EC (1999) One-dimensional thermal models constrained by seismic velocities
and surface radiogenic heat production for two main geotectonic units in southern Portugal.
Tectonophysics 306(3):261–268
Davraz A, Afşin M, Aksever F, Karakas Z, Hınıs MA (2016) The interference of a deep thermal
system with a shallow aquifer and health risk assessment: the case of Sandıklı (Afyonkarahisar)
Basin, Turkey. Environ Earth Sci 75(4):332
Demer S, Memiş Ü, Özgür N (2013) Investigation of hydrogeochemical properties of the Hüdai
(Afyon-Sandıklı) geothermal systems, SW Turkey. J Earth Syst Sci 122(4):1081–1089
Dimitriadis K, Tselentis GA, Thanassoulas K (1987) A basic program for 2-D spectral analysis of
gravity data and source depth determination. Comput Geosci 13:549–560
Dolmaz MN, Hisarli ZM, Ustaomer T, Orbay N (2005) Curie point depths based on spectrum
analysis of aeromagnetic data, west Anatolian extensional province, Turkey. Pure Appl
Geophys 162:571–590
360 N. MADEN et al.
Düzgit Z, Hisarli ZM, Sayin N, Orbay N (2006) Correlation between gravity and magnetic
anomalies of Western Anatolia and its relation to tectonic structures. Earth Planets Space
58:943–949
Ercan T, Ölmez E, Matsudo I, Wagao K, Kıta I (1994) Kuzey ve Batı Anadolu’da sıcak ve
mineralize sular ile içerdikleri gazların kimyasal ve izotopik değerleri. Turkiye Enerji Bulteni.
Cilt: 1. Sayı: 2. TMMOB Jeoloji Muh. Odası yayını, pp 20–21
Erkan K (2015) Geothermal investigations in western Anatolia using equilibrium temperatures
from shallow boreholes. Solid Earth 6:103–113
Fournier RO (1977) Chemical geothermometers and mixing models for geothermal systems.
Geothermics 5:41–50
Fournier RO (1985) The behaviour of silica in hydrothermal solutions. Rev Econ Geol 2:45–61.
Berger BR, Bethke PM (eds) Geology and geochemistry of epithermal systems. Society of
economic geologists
Giggenbach WF (1988) Geothermal solute equilibria. Derivation of Na–K–Mg–Ca Geoindicators.
Geochim Cosmochim Acta 52:2749–2765
Gupta H, Roy S (2007) Geothermal energy, an alternative resource for the 21st Century. Elsevier,
Amsterdam, 279 pp
Gürsu S, Göncüoğlu MC (2005) The geology and petrography of Late Neoproterozoic and Early
Paleozoic units of Western Taurus (SW-Sandıklı, Afyon). Bull Min Res Exp 130:29–55
He L, Hu S, Yang W, Wang J (2009) Radiogenic heat production in the lithosphere of Sulu
ultrahigh-pressure metamorphic belt. Earth Planet Sci Lett 277:525–538
Horasan G, Gülen L, Pınar A, Kalafat D, Özel N, Kuleli S, Işıkara AM (2002) Lithospheric
structure of the Marmara and Aegean Regions, Western Turkey. Bull Seism Soc Am 92(1):
322–329
İlkışık OM (1992) Silica heat flow estimates and lithospheric temperature in Anatolia. In:
Proceedings, XI. congress of world hydrothermal organization, İstanbul, 13–18 Mayıs 1992,
pp 92–104
İlkışık OM (1995) Regional heat flow in western Anatolia using silica temperature estimates from
thermal springs. Tectonophysics 244:175–184
İlkışık OM, Gürer A, Tokgöz T, Kaya C (1997) Geoelectromagnetic and geothermic investigations
in Ihlara valley geothermal field. J Volcanol Geoth Res 78:297–308
Jokinen J, Kukkonen IT (1999) Random modelling of the lithospheric thermal regime: forward
simulations applied in uncertainty analysis. Tectonophysics 306(3):277–292
Kalafat D, Gürbüz C, Üçer SB (1987) Batı Türkiye’deki Kabuk ve Üst Manto Yapısının
Araştırılması, Turkish Bull. Earthq Res 59:43–64 (in Turkish)
Karamanderesi İH (2008) The geological and mineralogical evaluation of the Afyonkarahisar
geothermal fields. In: Türker AE, Yıldız A (eds) Proceedings conference of thermal and mining
waters, 24–25 Apr 2008, Afyonkarahisar, pp 55–69
Kukkonen IT, Jokinen J, Seipold U (1999) Temperature and pressure dependencies of thermal
transport properties of rocks: implications for uncertainties in thermal lithosphere models and
new laboratory measurements of high-grade rocks in the central fennoscandian shield. Surv
Geophys 20(1):33–59
Kumar PS, Menon R, Reddy GK (2007) The role of radiogenic heat production in the thermal
evolution of a Proterozoic granulite-facies orogenic belt: Eastern Ghats, Indian Shield. Earth
Planet Sci Lett 254(1):39–54
Lachenbruch AH (1970) Crustal temperature and heat production: implications of the linear
heat-flow relation. J Geophys Res 75(17):3291–3300
Maden N (2009) Crustal thermal properties deduced from spectral analysis of magnetic data in
Central Pontides (Turkey). Turk J Earth Sci 18:383–392
Maden N (2010) Curie-point depth from spectral analysis of magnetic data in Erciyes
stratovolcano (Central TURKEY). Pure Appl Geophys 167:349–358
Maden N (2012) Two-dimensional geothermal modeling along the Central Pontides magmatic arc
(Northern Turkey). Surv Geophys 33:275–292
12 Geothermal Potential and Circulation Depth of Hüdai Thermal … 361
Maden N (2013) Geophysical approach for the detection and evaluation of geothermal energy
potential stimulated from geology and tectonics in Cappadocia region (Central Turkey). In:
Yang J (ed) Geothermal energy, technology and geology. Nova Science Publishers, NY
Maden N, Gelişli K, Eyüboğlu Y, Bektaş O (2009) Determination of tectonic and crustal structure
of the Eastern Pontide Orogenic Belt (NE Turkey). Pure Appl Geophys 166:1987–2006
Mall DM, Sharma SR (2009) Tectonics and thermal structure of western Satpura, India. J Asian
Earth Sci 34(3):450–457
Memiş Ü (2010) Determination of hydrogeochemical features and investigation of trace element
pollution of the Hüdai geothermal field in Sandıklı, Afyon (in Turkish). M.Sc. thesis, Süleyman
Demirel University, 64 p
Moisio K, Kaikkonen P (2006) Three-dimensional numerical thermal and rheological modelling in
the central Fennoscandian Shield. J Geodyn 42(4):95–114
Morgan P, Sass JH (1984) Thermal regime of the continental lithosphere. J Geodyn 1(2):143–166
Mutlu H (1996) Geochemical assessment of thermal waters from the Afyon geothermal area:
geothermometry applications and fluid-mineral equilibria. Ph.D. thesis, Middle East Technical
University, Graduate School of Natural and Applied Sciences, 169 p
Nnange JM, Ngako V, Fairhead JD, Ebinger CJ (2000) Depths to density discontinuities beneath
the Adamawa plateau region, Central Africa, from spectral analyses of new and existing gravity
data. J Afr Earth Sci 30:887–901
Okubo Y, Graf RJ, Hansen RO, Ogawa K, Tsu H (1985) Curie point depths of the island of
Kyushu and surrounding areas, Japan. Geophysics 50:481–494
Okubo Y, Tsu H, Ogawa K (1989) Estimation of Curie point temperature and geothermal structure
of island arcs of Japan. Tectonophysics 159(3–4):279–290
Okubo Y, Makino M, Kasuga S (1991) Magnetic model of the subduction zone in the Northeast
Japan arc. Tectonophysics 192:103–115
Öngür T (1973) Geologic situation of Sandıklı (Afyon) geothermal investigation field and
geothermal energy facilities. General Directorate of Mineral Research and exploration report
no. 5520
Özpınar Y (2008) Sandıklı (Afyon) yöresinde, sabazit ve fillipsit mineralleri içeren tüflerin
minerolojik petrografik ve iyon değiştirme özellikleri ve tarımda kullanılması, GB Anadolu,
Türkiye. Bull Min Res Exp 137:27–47
Öztaş Y (1989) Homa-Akdağ yöresi, Sandıklı-Şuhut kuzeylerinin jeolojisi ve petrol olanakları;
TPAO Ar, siv Rap. No. 2584, Ankara
Pasquale V (1987) Possible thermal structure of the eastern part of the Central Alps. Il Nuovo
Cimento C 10(2):129–141
Perry HKC, Jaupart C, Mareschal JC, Bienfait G (2006) Crustal heat production in the Superior
Province, Canadian Shield, and in North America inferred from heat flow data. J Geophys Res:
Solid Earth 111(B4)
Poudjom Djomani YH, Diament M, Albouy Y (1992) Mechanical behaviour of the lithosphere
beneath the Adamawa Uplift (Cameroon, West Africa) based on gravity data. J Afr Earth Sci
15:81–90
Rai SN, Thiagarajan S (2006) A tentative 2D thermal model of central India across the
Narmada-Son Lineament (NSL). J Asian Earth Sci 28(4):363–371
Rao RUM, Verma RK, Rao GV, Gupta ML (1970) Heat flow at Damua and Mohapani, Satpura
gondwana basin, India. Earth Planet Sci Lett 7(5):406–412
Riad S, Refai E, Ghalib M (1981) Bouguer anomalies and crustal structure in the Eastern
Mediterranean. Tectonophysics 71:253–266
Rimi A (1999) Mantle heat flow and geotherms for the main geologic domains in Morocco. Int J
Earth Sci 88(3):458–466
Rivero L, Pinto V, Casas A (2002) Moho depth structure of the eastern part of the Pyrenean belt
derived from gravity data. J Geodyn 33:315–332
Rybach L, Buntebarth G (1984) The variation of heat generation, density and seismic velocity with
rock type in the continental crust. Tectonophysics 103:309–344
362 N. MADEN et al.
Sari C, Şalk M (2006) Sediment thicknesses of the western Anatolia graben structures determined
by 2D and 3D analysis using gravity data. J Asian Earth Sci 26:39–48
Saunders P, Priestly K, Taymaz T (1998) Variations in the crustal structures beneath western
Turkey. Geophys J Int 134:373–389
Serpen U, Aksoy N, Öngür T (2010) Present status of geothermal energy in Turkey. In:
Proceedings, thirty-fifth workshop on geothermal reservoir engineering, 1–3 Feb 2010,
Stanford University, Stanford, CA, 2010
Shuey RT, Schellinger DK, Tripp AC, Alley LB (1977) Curie depth determination from
aeromagnetic spectra. Geophys J Int 50(1):75–101
Spector A, Grant FS (1970) Statistical models for interpreting aeromagnetic data. Geophysics
35:293–302
Stüwe K (2007) Geodynamics of the lithosphere: an introduction. Springer Science & Business
Media
Swanberg CA, Morgan P (1979) The linear relation between temperatures based on the silica
content of groundwater and regional heat flow: a new heat flow map of the United States.
Pageoph I 17:227–241
Swanberg CA, Morgan P (1980) The silica heat flow interpretation technique: assumptions and
applications. J Geophys Res 85:7206–7214
Tanaka A, Okubo Y, Matsubayashi O (1999) Curie-point depth based on spectrum analysis of the
magnetic anomaly data in East and Southeast Asia. Tectonophysics 306:461–470
Tezcan AK (1979) Geothermal studies, their present status and contribution to heat flow
contouring in Turkey. Terrestrial heat flow in Europe, pp 283–291
Tezcan AK (1995) Geothermal explorations and heat flow in Turkey Terrestrial Heat Flow and
Geothermal Energy in Asia Rotterdam (Rotterdam: August Aime Balkema), pp 23–42
Tezcan AK, Turgay MI (1987) Heat flow density distribution in Turkey. Publication of the Mineral
Research and Exploration (MTA), Ankara
Truesdell AH (1976) Summary on section III geochemical techniques in exploration. In:
Proceedings, second United Nations symposium on the Development and Use of Geothermal
Resources, San Francisco
Turcotte DL, Schubert G (1982) Geodynamics applications of continuum physics to geologic
problems. Wiley, New York, p 450
Verma MP (2000) Chemical thermodynamics of silica: a critique on its goethermometer.
Geothermics 29:323–346
Verma SP (2001) Geochemical and Sr-Nd-Pb isotopic evidence for a combined assimilation and
fractional crystallisation process for volcanic rocks from the Huichapan caldera, Hidalgo,
Mexico. Lithos 56:141–164
Yamano M (1995) Recent heat flow studies in and around Japan. In: Gupta ML, Yamano M
(eds) Terrestrial heat flow and geothermal energy in Asia. A. A. Balkema, Rotterdam,
pp 173–201
Chapter 13
Geophysical Characterization
of Chumathang (Ladakh) Hot Spring
Keywords Audio magnetotelluric Geothermal hot spring Indus Tsangpo suture
zone (ITSZ) Resistivity
13.1 Introduction
Himalaya, result of subduction and collision of Indian plate with Eurasian plate,
host many geothermal field all along its arc. Majority of the geothermal region or
springs are located around major structural features e.g. Main Central Thrust
(MCT), Indo-Tsangpo Suture zone (ITSZ). Proximity of geothermal fields to these
tectonic and structural features suggest the control of tectonics and structure in
manifestation of these geothermal springs on the surface. Although, there is con-
siderable geological and geophysical investigations in Ladakh region (Thakur and
Mishra 1984; Sigoyer et al. 2004; Yin 2006; Arora et al. 2007; Rao and Rai 2009;
Jain 2017, etc.) mainly focused for understanding of regional characteristic of
Himalayan tectonics but due to very limited integrated studies dedicated to
geothermal fields, barring few e.g. Puga in Ladakh region (Thussu 2002;
Harinarayana et al. 2004; Azeez et al. 2007) and unpublished reports, there is still
poor understanding about the source and subsurface extent of these geothermal
regions. To improve upon the present understanding, We have attempted a localized
study of Chumathang hot spring area in Ladakh region through Audio
magnetotelluric (AMT) investigations around it. Here we discuss some results of
our AMT studies in brief.
The Chumathang geothermal area located north of Indus Tsangpo suture zone
(ITSZ) and is considered as a part of the Puga—Chumathang geothermal field.
Chumathang, a small town of Indus valley in eastern Ladakh region of the North
West Himalaya, can be reached from Upshi by road (Fig. 13.1a). In the three
distinct tectonic belts around Chumathang (Fig. 13.1b), the top northern belt
comprises of sedimentary rocks of shallow marine to fluvial facies. Ladakh granites,
including Chumathang granite, intruded the sedimentary sequence. Chumathang
Granite is exposed close to the site of the Chumathang geothermal field. In the
central belt, thick sequence of basic, ultrabasic rocks and associated sedimentary
rocks (fossiliferous limestone) belonging to the Sumdo Group is present. The
central belt corresponds to the Indus Suture Zone itself. NE–SW trending Mahe
fault limited the Chumathang region from the south. The Puga valley and the Puga
geothermal system is further southern side of Chumathang. Thick succession of
sediments, meta-sediments and metamorphic rocks intruded by granites
(Harinarayana et al. 2006) are observed in Puga valley. The intensity of meta-
morphism varies in increasing manner from northwest to southeast. Highly
deformed and of medium to high metamorphic grade rocks are exposed in the
southern belt.
13 Geophysical Characterization of Chumathang (Ladakh) Hot Spring 365
Fig. 13.1 a Geological cross-section through the N.W. Himalaya showing the relationship
between the Puga Valley and Chamuthang geothermal areas and the deep structure of the Indus
Suture Zone (Modified from Epard and Steck 2008). b Geological map of Puga Valley and
Chamuthang area, Ladakh showing the major tectonic structures (Modified from Shankar et al.
1991 and Azeez and Harinarayana 2007)
366 G. Rawat et al.
The Chumathang geothermal area is located in the northern belt. The area may
be related to a prominent NNE–SSW trending fault (sub-parallel to the Kiagar Tso
Fault at Puga) which crosses the Indus valley at this point (Fig. 13.2). Recent talus,
aeolian sand, fluvial-glacial sediments and glacial moraines encrusted with borax,
sulphur and other hot spring deposits comprise the alluvial cover in the
Chumathang area. It is believed that this loose deposits extends to a depth of
between 15 and 65 m below which lies hard breccia and conglomerate, compacted
from the sediments by the action of geothermal fluids (Azeez et al. 2007).
Geomorphological observations indicate the area tectonically active. The area is
intensely affected by Neogene and Quaternary vertical or steeply-dipping faults,
broad regional doming and basin and pull-apart structures developed in a zone of
dextral transpression parallel to the Indus Suture Zone and the Zildat-Ribil fault
system (Craig et al. 2013). It is interpreted that these and the other major faults in
the area, probably act as conduits for the hot water.
At Chumathang, signature of thermal activity is in the form of hot springs,
travertine deposits and stained ground covers mostly on the north bank of the Indus
river. Geochemical analysis of hot spring water shows high pH values with elevated
amount of TDS and presence of Trace elements like As and Sr. Enriched Cl− in the
hot water spring of Chumathang geothermal spring may be from chloride phase of
metamorphic crustal fluid (Thussu 2002; Tiwari et al. 2016). Other magnetotelluric
surveys identified a low resistivity zone of 13–30 Xm up to a depth of 300 m
indicative of vigorous geothermal activity at depth (Thussu 2002). Although, low
resistivity values in AMT results may be the due the presence of clay-rich sedi-
ments or alteration. Geochemical thermometry suggests a deep reservoir tempera-
ture of at least 145–145 °C. However, silica–quartz geothermometry of current data
indicate reservoir temperature of 107 °C (Tiwari et al. 2016).
13 Geophysical Characterization of Chumathang (Ladakh) Hot Spring 367
13.3 Methodology
E ¼ Z:H
Ex Zxx Zxy Hx
¼
EY Zyx Zyy Hy
medium. For deeper information (crustal scale), longer periods (1–1000 s) are
required whereas for shallow depth (*1 to 5 km) high frequency signals (10 kHz—
1 s (AMT band)) are required. Therefore, recording period of electric and magnetic
field variations, i.e., occupancy period of each site, control the depth penetration of
signals.
High frequency magnetotelluric data (10 kHz—1 s) is collected in Chumathang
at 16 sites along 1.5 km profile cutting across the regional structural setup
(Fig. 13.1). The valley is divided by river Indus and most of the surface geothermal
expression is on the north side of river. Cultural habitats are also on the north side
where as south side of the river is barren and accessible through a temporary bridge
made by villagers. The profile was selected in a way that it crossed the hot spring
and traversed approximately orthogonal to regional trend of structural setup in the
region. Solid circle in Fig. 13.3 shows the location of AMT site with reference to
the place of space heating (star). Total 16 sites were occupied during the period.
The ADU07e magnetotelluric systems, MFS07E Induction coils and EFP07e
electrodes were used for recording three component magnetic field variations in
different frequency bands. Due to its remoteness and locality in high altitude region,
there was no apparent source of contamination of MT source signals. The only
source of local noise was electricity generators. The electric supply in the region
was due to individual generators or community generators which contributed noise
in the AMT data collection, during their operation. Although operating time of
these generators is very limited in day time and 3–4 h during night time. The sites at
the north side of river had larger contribution of noise due to proximity with
habitats and their activity. We, therefore, planned simultaneous recording of MT
sites one at southern and one at northern side of river so that remote reference
technique can be applied to reduce the effect of local noise.
Fig. 13.4 Representative apparent resistivity curves from both side of river
13 Geophysical Characterization of Chumathang (Ladakh) Hot Spring 371
Fig. 13.5 Observed resistivity and phase pseudo section of E-W (left) and N-S (right) polarization
372 G. Rawat et al.
Fig. 13.6 Inverted resistivity model (TM mode) with r.m.s. misfit of 1.7
place below site S03 whereas toward north eastern side of Indus river these deposits
are in pockets ‘B’, ‘C’, ‘D’ and ‘E’ delimited with intrusion of granitic rocks.
Perhaps hot conduits are moving upward through the boundaries of these pockets.
Therefore, most of the geothermal exposure is visible toward north of river. The ‘F’
and ‘G’ feature are resistive feature where ‘F’ is more resistive than ‘G’ and
represents granitic rock of different stages. South of the Chumathang region is
suture zone known as Indus Tsangpo suture zone. The ITSZ in a regional geo-
physical modelling of the Ladakh region manifest itself in a sub vertical conductive
feature merging with the mid crustal conductor of low resistivity. The low resis-
tivity is indicative of presence of fluid, partial melt (Arora et al. 2007).
Figure 13.7 depicts the model and calculated response. Model response and
calculated response matches to larger extent. The difference is due to primarily two
reasons. First is due to the fact that entire response does not exhibit regional 2D
conductivity distribution and second is modelling of rotated response which may
have some projection of noise in rotated co-ordinate frame. Therefore, there is
further scope of improvement in model and suggest three dimensional modelling
for getting improved picture of subsurface. However, the model can be considered
as representative signature of subsurface in Chumathang region, within scope of 2D
modelling.
374 G. Rawat et al.
Fig. 13.7 Model and observed apparent resistivity and phase pseudosection along the line P1
13 Geophysical Characterization of Chumathang (Ladakh) Hot Spring 375
13.6 Conclusion
The model obtained after inverting AMT data sets of Chumathang region reveals
subsurface features and is consistent with regional geological and geophysical
model of Ladakh region. The Chumathang region at depth of 200–300 m is
occupied with a thick crystalline basement which is seen as high resistive feature in
geoelectrical model. AMT studies are able to peep into the subsurface of
Chumathang hot spring region but due to modeling of rotated tensor, there is
significant scope of improvement in geoelectrical modelling which may reveal
some more features pertaining to geothermal field in the region.
References
Arora BR, Unsworth MJ, Rawat G (2007) Deep resistivity structure of the northwest Indian
Himalaya and its tectonic implications. Geophys Res Lett 34:L04307.1–L04307.4
Azeez KKA, Harinarayana T (2007) Magnetotelluric evidence of potential geothermal resource in
Puga, Ladakh, NW Himalaya. Curr Sci 93(3):323–329
Cagniard L (1953) Basic theory of the magnetotelluric method of geophysical prosepecting.
Geophys 18:605–645
Craig J, Absar A, Bhat G, Cadel G, Hafiz M, Hakhoo N, Kashkari R, Moore J, Ricchiuto TE,
Thurow J, Thusu B (2013) Hot springs and the geothermal energy potential of Jammu &
Kashmir State, NW Himalaya, India. Earth-Sci Rev 126:156–177. https://doi.org/10.1016/j.
earscirev.2013.05.004
Epard JL, Steck A (2008) Sturcutral development of Tso Morari ultra high pressure nappe of the
Ladakh Himlaya. Tectonophys 451:242–264
Harinarayana T, Azeez KKA, Naganjaneyulu K, Manoj C, Veeraswamy K, Murthy DN, Rao SPE
(2004) Magnetotelluric studies in Puga valley geothermal field, NW Himalaya, Jammu and
Kashmir, India. J Volcanol Geotherm Res 138:405–424
Harinarayana T, Abdul Azeez KK, Murthy DN, Veeraswamy K, Eknath Rao SP, Manoj C,
Naganjaneyulu K (2006) Exploration of geothermal structure in Puga geothermal field, Ladakh
Himalayas, India by magnetotelluric studies. J Appl Geophys 58(4):280–295
Jain AK (2017) Continental subduction in the NW-Himalaya and Trans-Himalaya. Ital J Geosci
136(1):89–102. https://doi.org/10.3301/IJG.2015.43
Jones AG (1992) Electrical properties of the lower continental crust. In: Fountain DM, Arculus R,
Kay RW (eds) Developments in geotectonics, vol 23, pp 81–143
Rao DR, Rai H (2009) Geochemical studies of granites from Shyok tectonic zone of
Khardung-Panamik section, Ladakh, India. J Geol Soc India 73(4):553–566
Sigoyer JD, Dick P (2004) Exhumation of ultra-high pressure Tso Morari Unit in eastern Ladakh
(NW Himlaya): a case study. Tectonics 23:TC3003
Simpson F (2001) Fluid trapping at the brittle-ductile transition re-examined. Geofluids 1:123–136
Simpson F, Bahr K (2005) Practical magnetotellurics. Cambridge University Press, p 254
Thakur VC, Mishra DK (1984) Tectonic framework of the Indus and Shyok suture zones in
Eastern Ladakh. Northwest Himalaya Tectonophys 101:207–220
Thussu JL (2002) Geothermal energy resources of India, Special Publication No 69, Geological
Society of India, p 210
Tikhonov AN (1986) On determining electrical characteristics of the deep layers of the Earth’s
crus. In: Vozoff K (ed) Magnetotelluric methods. Society of Exploration Geophysicsts, Tulsa,
pp 2–3
376 G. Rawat et al.
Tiwari KS, Rai SK, Bartarya SK et al (2016) Stable isotopes (d13CDIC, dD, d18O) and
geochemical characteristics if geothermal springs of Ladakh and Himachal (India): evidence
for CO2 discharge in northwest Himalaya. Geothermics 64:314–330
Vanyan LL, Tezkan B, Palshin N (2001) Low electrical resistivity and seismic velocity at the base
of the upper crust as indicator of rheologically weak layer. Surv Geophys 22(2):131–154
Yin A (2006) Cenozoic tectonic evolution of the Himalayan orogen as constrained by long-strike
variation of structural geometry, exhumation history, and foreland sedimentation. Earth Syst
Rev 36:1–131
Chapter 14
Airborne Geophysical Surveys
and Their Integrated Interpretation
14.1 Introduction
lines with a few hundred meters flight line spacing (100–200 m). The survey is
performed either using a helicopter or a fixed-wing airplane. Generally, three types
of geophysical data namely magnetic, radiometric and electromagnetic (EM) are
collected simultaneously in a typical airborne survey. Sometime gravity data is also
collected by airborne surveys though it is not covered in this chapter. Readers are
recommended to consult Reeves (2005) and Camara and Guimarães (2016) for
details of the airborne magnetic survey, IAEA (2003) and Connor et al. (2016) for
details of airborne radiometry survey and Siemon et al. (2009), Legault (2015) and
Auken et al. (2017) for airborne EM surveys. In this chapter, a brief description of
various aspects of airborne geophysical data acquisition, processing and interpre-
tation is provided using examples and practices from Geological Survey of Norway
(NGU).
The airborne geophysical data assist geologists and geophysicists in various
types of the geological and environmental studies e.g. bedrock mapping
(Christensen et al. 2015; Oldenborger et al. 2016), mineral prospecting (Spector and
Lawler 1995; Smith 2010; Elkhadragy et al. 2017), radiogenic heat production
(Youssef 2016), tunnel construction (Okazaki et al. 2011), clay characterization
(Baranwal et al. 2015, 2017b) and groundwater explorations (Dickson et al. 2014;
Høyer et al. 2015). The airborne geophysical data plays a very crucial role to
enhance the general understanding of the regional geology. The data are used to
map geological contacts and important structural features. It can help to define the
potential of mineralization zones, the geological settings, and to identify new areas
of interest. Airborne geophysical data are also used for foreseeing possible tunnel
construction problems due to deep weathering (Olesen et al. 2007), and evaluation
of possible radon problems in houses (Smethurst et al. 2017) and characterization of
soft sediments (Supper et al. 2013).
The airborne survey utilizes a highly sensitivity cesium (Cs) vapor magne-
tometer, gamma ray spectrometer, EM transmitters and receivers. Radar and laser
altimeters are used to measure altitude of the sensors. A GPS navigation enabled
computer system showing flight path ensures an accurate positioning of the geo-
physical data. Generally, gamma-ray spectrometer is mounted inside an airplane or
under the belly of a helicopter. Magnetometer, EM transmitters and EM receivers
are either in-housed in a bird which is towed below a helicopter or they are mounted
on a fixed-wing airplane.
both onshore and offshore. More flight lines will obviously require more economic
resources therefore a pre-survey planning to find a trade-off between required data
resolution and budget of the survey is beneficial.
A high-sensitivity cesium vapor magnetometer is generally used to measure total
magnetic field. A gamma ray spectrometer consisting of 4 or 5 NaI detector packs
of 4 L each is used to register incidence of gamma rays emitted either by natural
radioactive sources e.g. potassium (K), uranium (U) and thorium (Th) or man-made
sources (e.g. Cs) due to nuclear accidents. Modern spectrometer e.g. RSX-5 (from
Radiation Solutions Inc.) measures a 1024 channel gamma ray spectrum between 0
and 3 MeV energy.
A typical RSX-5 (from Radiation Solutions Inc.) gamma ray spectrometer with
four downward detectors and one upward detector is shown in Fig. 14.1. Details of
magnetometer and gamma ray spectrometer instrumentation are provided in
Table 14.1.
Different types of EM transmitters and EM receivers and their combinations are
used to collect airborne electromagnetic (EM) data. The measurements are performed
either in frequency domain or in time domain. Well known fixed wing time domain
airborne EM systems are TEMPEST, GEOTEM and MEGATEM. Well known time
domain helicopter EM systems are HeliTEM, VTEM, AeroTEM and SkyTEM.
Fig. 14.1 An RSX-5 gamma-ray spectrometer used in airborne survey by NGU. Four crystals of
4.1 L each work as downward looking detectors and one crystal of 4.1 L (on the top) works as
upward looking detector
380 V. C. Baranwal and J. S. Rønning
Table 14.1 Instrument specification of airborne geophysical surveys performed by NGU (from
NGU website and various NGU reports)
Instrument Producer/model Accuracy/resolution Sampling
frequency/
interval
Magnetometer Scintrex Cs-2 0.002 nT 5 Hz
Base GEM GSM-19 0.1 nT 3s
magnetometer
Electromagnetic Geotech 1–2 ppm 10 Hz
hummingbird
Gamma Radiation 1024 channel, 16.4 L 1 Hz
spectrometer solutions RSX-5 downward, 4.1 L upward
Radar altimeter Bendix/King ±3% 0–500 ft 1 Hz
KRA 405B ±5% 500–2500 ft
Pressure/ Honeywell PPT ±0.03% FS 1 Hz
temperature
Navigation Topcon ±5 m 1 Hz
GPS-receiver
Acquisition NGU in-house
software software
All three types of the airborne geophysical data i.e. magnetic, radiometric and EM
are collected for a general geological mapping of the area with flight line spacing
normally of 200 m. A helicopter survey is generally flown at ca. 60 m altitude at a
speed of ca. 100 km/h depending on the terrain condition with the bird towed ca.
30 m below. A fixed-wing survey is generally performed at 60–100 m altitude at a
speed of ca. 200 km/h. However, flight line spacing and flying height can be
adapted to the particular project depending on the need to the required resolution in
the data. If possible then survey is generally performed perpendicular to the geo-
logical strike.
14 Airborne Geophysical Surveys and Their Integrated Interpretation 381
Fig. 14.2 Hummingbird system for airborne magnetic and electromagnetic survey. a Shows
hummingbird tied below helicopter during flying, b shows hardware unit mounted inside
helicopter for airborne data acquisition and c inside of hummingbird showing five sets of
transmitter and receiver coils for five different frequencies together with a Cs-vapor magnetometer
in the middle
Occurrence of gamma rays emitted from natural decay of the radioactive elements
is registered by the spectrometer and normally accumulated for 1 s. Earlier gamma
ray spectrometers were not very stable and a known source was required to check
position of the peak energy for that particular source e.g. 137Cs everyday during the
survey. This would help to calculate a correct energy calibration of the spectrom-
eter. Modern gamma ray spectrometers e.g. RSX-5 (from Radiation Solutions) are
stabilized against natural isotopes of K, U and Th sources at the surface. A typical
gamma ray spectrum and corresponding energies for various natural radio-elements
(K, U, Th) are shown in Fig. 14.3.
A set of calibrations are required before and during the airborne survey for
processing of the radiometric data. Various corrections are applied on airborne
radiometry data to calculate ground concentration of the elements e.g. K, U, Th and
sometime Cs. Standard calibration procedures are well documented by IAEA
(2003).
Very high altitude flights (several thousand meters) are required to calculate
cosmic and aircraft background coefficients which is done annually. Radon cor-
rection parameters are calculated from calibration of upward detectors or by flying
14 Airborne Geophysical Surveys and Their Integrated Interpretation 383
Fig. 14.3 Example of a typical gamma-ray spectrum showing the position of the K, Th, U and
Total count windows (from Stampolidis et al. 2013)
over test lines and adjacent water bodies that are performed during the survey.
Mobile calibration pads or fixed calibration pads are needed to calculate stripping
ratios to remove the spectral overlap among the radio-elements (Compton scatter-
ing) which is also done annually. Calibration flights within a range of nominal
survey height are needed to calculate height attenuation coefficients which should
be ideally done during each survey or at least annually. Sensitivity coefficients are
needed to convert gamma ray count rates from different radioelement windows to
ground concentrations. These sensitivity coefficients are ideally calculated from
ground measurements of a calibration range test strip by a calibrated handheld
spectrometer and flying the spectrometer at nominal flying height over this test
strip. Alternatively, sensitivity coefficient is also calculated from measurements by
placing spectrometer directly over calibration pads and using attenuation coeffi-
cients to calculate the count rates at the ground from the nominal height though this
procedure is not accurate.
These calibration coefficients and parameters are used in general processing of
the airborne gamma ray spectrometry data for geological mapping. Total count
(TC), equivalent uranium (eU), equivalent thorium (eTh) and potassium
(K) window data are corrected for the aircraft and cosmic background, radon in the
surrounding air, stripping correction due to spectral overlap, attenuation correction
for variations in the flying height to a nominal flying height and to convert count
rates in effective ground element concentrations. Equivalent term is used for ura-
nium and thorium because concentration is calculated from daughter isotopes
assuming decay series is in equilibrium.
384 V. C. Baranwal and J. S. Rønning
Final processed data from airborne magnetic, radiometry and electromagnetic are
generally presented as interpolated grids. Different types of gridding approach can
be used for it e.g. kriging, minimum curvature, natural neighbor, inverse distance
weighting (IDW), spline etc. Kriging and minimum curvature are widely used in
geology and geophysics. The data is gridded generally with cell size of one fourth
of the line spacing to avoid aliasing which results in distortion and artifacts due to
under-representation of the data. Therefore, data collected at 200 m line spacing is
gridded generally with 50 m cell size and results in cells of 50 m 50 m. Before
gridding, flight data are split in flight lines. Micro-leveling is performed to the data
before final gridding if there are visible noise as along-line artifacts. A convolution
14 Airborne Geophysical Surveys and Their Integrated Interpretation 385
filter is also used sometime to smooth the final grid image. Shaded relief
enhancement method is used to illuminate the image from a specific direction which
adds spatial resolution to a color image. It also gives more insight in quality of the
gridded data and processing errors.
Processed magnetic data is generally presented as total magnetic field anomaly.
Magnetic anomaly reflects magnetic sources from near the surface and down to
10 km below the subsurface. Vertical, horizontal and tilt derivatives and analytic
signal (Salem et al. 2008) can also be calculated to enhance the presence of the
lineaments, faults or small variation in the rock types. Many other types of filters
e.g. reduce to pole, reduce to equator, pseudo density etc. are also used for specific
purposes (Reeves 2005). For further detailed interpretation, forward modeling and
inversion are performed to the processed magnetic data with the help of various
types of advanced modeling software and routines (Mishra 2011; Fairhead 2015).
Footprint of the airborne radiometry measurements is roughly an oval having
width twice the flying height, and length twice the flying height and the distance
travelled during the measurement, i.e. roughly 120 m 150 m (for a 60 m sensor
height with a speed of 100 km/h). An average concentration of the radio-elements
K, eU and eTh from rough area of this oval (footprint) is measured in an airborne
radiometry survey. Final processed data from radiometry is represented in ground
concentration of radio-elements K, eU and eTh. A ternary image can be drawn from
combining all three elements and assigning colors blue, green and red to eU, eTh
and K, respectively (IAEA 2003). The ternary image is very useful to highlight
different rock types and structures. Sometime ratio of various radio-elements e.g.
K and eTh or K and eU are also plotted to differentiate among different rock types.
In frequency domain EM (FEM) methods, EM signal attenuates through the
conducting medium and higher the frequency, faster it attenuates, giving less
penetration depth. Therefore lower frequencies contain information from deeper
parts of subsurface and higher frequencies contain information from shallower part
of the earth. For time domain systems, the induced current diffuses in the ground
down- and outwards with time after current is switched off. Early measured time
gates represent shallow depth while late time gates give response from deeper
targets. EM methods provide average volumetric conductivity information of the
subsurface. Multi-frequency data in frequency domain and decayed EM signal at
various time gates are used to obtain the conductivity or resistivity information from
different depths of the subsurface. SKYTEM, VTEM and other time domain sys-
tems have a depth of investigation range down to ca. 500 m however RESOLVE,
Hummingbird and other frequency domain systems have a depth of investigation
range down to ca. 150 m. Footprint (horizontal resolution) of AEM method is
dependent on flying height, frequencies and type of transmitter and receiver coils.
Hummingbird system has a footprint around 70 m for sensor height of 30 m (Smith
et al. 2011). The near surface vertical resolution is higher for FEM systems, but
recent developments with faster electronics are about to give the same resolution for
TEM as well (Auken et al. 2017).
386 V. C. Baranwal and J. S. Rønning
Processed time domain or frequency domain AEM data is inverted with the help
of 1D, 2D or 3D inversion routines and true resistivity/conductivity of the sub-
surface is calculated.
A crude homogenous half-space inversion of frequency domain AEM data at a
single frequency provides a qualitative interpretation of the lateral location of the
conductors and its resistivity/conductivity. However, it cannot tell about depth of
the conductor. In this approach, a half-space resistivity is calculated at each data
locations using iterative least-squares method to fit the observed data (Geosoft
Manual 2005).
The first example shows location of Alum shale near Drammen, Norway by air-
borne radiometry and electromagnetic survey. Alum shale has high content of
uranium, graphite and sulfide which responds to airborne radiometry and EM
surveys, respectively. Airborne magnetic data was also collected in the same survey
but it did not distinguish Alum share areas and therefore not discussed here. The
decay of uranium leads to emission of radon gas which may seep into buildings and
also dissolve in the groundwater and cause serious health issues such as lung cancer
(Smethurst et al. 2017). The whole region belongs to Kongsberg-Modum complex
of southern Norway which is part of the Mesoproterozoic Sveco-Norwegain orogen
of southwest Scandinavia and internationally known as silver mining district
(Baranwal et al. 2010). Base metals and cobalt has been mined from several places
in the region (Lutro and Nordgulen 2008).
Figure 14.4 shows helicopter EM data at four frequencies from one flight.
Location of this flight path is shown in Fig. 14.5 by semi transparent gray lines.
Data at 880 Hz was very noisy so it was not processed. Figure 14.4a and c depict
unleveled, spikes removed, non-linear and low-pass filtered in-phase and quadra-
ture component of the HEM data, respectively. Figure 14.4b and d show leveled
in-phase and quadraure data that is final data to be used for inversion. Leveling of
the HEM data was done for each flight using sections of high altitude flight data (ca.
300 m or more) as shown in Fig. 14.4e. Details of airborne EM processing are
described by Baranwal et al. (2013). Higher HEM anomalies are observed over
areas with saline water and alum shale as shown in Figs. 14.5 and 14.6.
Figure 14.5 depicts apparent resistivity calculated from 6.6 kHz in-phase and
quadrature data using a half-space inversion method (Geosoft Manual 2005).
6.6 kHz data can reveal conductivity information from few tens of meters below
subsurface. The image shows location of low resistivity areas caused by either
saline water, marine clay, conductive minerals individually or due to all of them
together. Location of water bodies are shown by thin black lines. Electrical and
Electromagnetic methods alone cannot differentiate whether low resistivity is
caused by saline water or conductive mineral or due to other source. However, it is
very likely that conducting areas adjacent to water bodies are caused by sulfides
present in the alum shale because it matches well with high concentration of ura-
nium (Fig. 14.6).
Figure 14.6 shows uranium concentration from processed airborne radiometry
data. Thin white lines show water bodies in the area. Details of processing steps for
radiometry data are detailed by Baranwal et al. (2013). Alum shale areas are clearly
visible with higher concentration of uranium. Gamma-rays will be attenuated by
few centimeters of water and soil layers therefore it mostly shows nearly-exposed
alum shale areas. There could be hidden alum shale areas below subsurface which
are reveled by HEM data (Fig. 14.5) due to highly conductive sulfide present in the
388 V. C. Baranwal and J. S. Rønning
Fig. 14.4 Spike removed, non-linear filtered, low-pass filtered and unlevelled a in-phase data and
c quadrature data at four frequencies. Levelled b in-phase and d quadrature data using high altitude
flight data as shown in (e), e altitude of the flight. Location of the flight lines are shown in
Fig. 14.5 by semi-transparent thin gray lines
alum shale. Comparing Figs. 14.5 and 14.6 reveals that alum shale could be present
below water and soil layers and could be continuous. This effect is well demon-
strated from the lake in the northern part of Figs. 14.5 and 14.6.
Fig. 14.5 Apparent resistivity calculated from 6.6 kHz in-phase and quadrature data using a
homogeneous half-space inversion routine. Thin black lines show location of water bodies. Thin
gray lines show flight lines along which EM data are shown in Fig. 14.4
Fig. 14.6 Concentration of uranium obtained from airborne gamma-ray spectrometry data. High
uranium areas mark location of alum shale
390 V. C. Baranwal and J. S. Rønning
Fig. 14.7 A quaternary geology map of Byneset showing location of geophysical surveys
(modified after Baranwal et al. 2017b). Black lines represent helicopter survey lines. Bedrock
exposure locations are shown by black dots. Orange and red lines are ERT lines performed in 2012
and 2014, respectively. Blue lines are seismic refraction lines. RCPTU was performed at S1 and S2
locations, shown by white triangles
deglaciation period around 10,000 years ago. The sediments are now exposed on
land at ca. 160 m a.s.l. due to glacio-isostatic rebound (Reite et al. 1999). Old seabed
is also preserved in some parts. Ravines and ca. hundred landslide scars have
dominated central parts of Byneset (Solberg et al. 2016). Common bedrock in the
area are chlorite slate in west and phyllites in the east. The phyllite are known to
contain graphite (an electronically conductive mineral) (Solberg et al. 2016).
14 Airborne Geophysical Surveys and Their Integrated Interpretation 391
Fig. 14.8 Subsurface resistivity image along line L2 a from ERT data and b from HEM data
(slightly modified after Baranwal et al. 2017b). Dashed black lines show seismic velocity
boundaries with lower dashed-black line as bedrock boundary
392 V. C. Baranwal and J. S. Rønning
line) and suggests bedrock layer is continuous below the landslide location and not
ruptured as suggested by ERT. HEM could not detect bedrock because of low depth
of penetration through a very conductive marine clay layer. Resistivity cone pen-
etration test unit (RCPTU) was also performed at two locations near the landslide
location and it gave similar vertical resistivity to HEM and ERT (Solberg et al.
2016; Baranwal et al. 2017b). Quick clay at landslide location was confirmed by
laboratory testing of the soil samples as well (Solberg et al. 2016).
Figure 14.9 shows 3D resistivity image of subsurface from the area in form of
resistivity slices at different depths from surface to ca. 30 m depth that was obtained
by spatially constrained inversion of five frequency HEM data. The image shows
that highly resistive bedrock at very shallow depth surrounded the region.
Outcropping bedrock (black dots in Fig. 14.7) was observed in the northern part
area and same is indicated by HEM data interpretation. The central part of the area
Fig. 14.9 Resistivity slices from 0 to 31 m depth extracted from SCI resistivity models obtained
from HEM data (slightly modified after Baranwal et al. 2017b)
14 Airborne Geophysical Surveys and Their Integrated Interpretation 393
This is an example for graphite exploration in north Norway. NGU acquired high
resolution airborne electromagnetic (EM), magnetic and radiometric data during
2011–2015 as part of a government funded program for mineral exploration in
Norway (MINN and MISN). A correlation of low magnetic anomalies and low
apparent resistivity was noticed from locations of graphite deposits observed at the
surface and for old graphite showings (Rønning et al. 2018). Figure 14.10 shows
apparent resistivity calculated from 7 kHz coaxial HEM data using a homogeneous
half-space model (Geosoft Manual 2005). Location of exposed graphite are shown
by black squares. Water bodies are plotted with semi-transparent blue color to
distinguish the saltwater areas where high conductivity/low resistivity could be
caused by saltwater instead of graphite. However, we see few graphite deposits very
close to the coast, almost in the seawater in Fig. 14.10. Seawater shows apparent
resistivity <1 Ωm and graphite deposits are located in structures with <300 Ωm
apparent resistivity. The observed apparent resistivity is an average of resistivities
covered by footprint area of each EM observation.
Fig. 14.10 Apparent resistivity image zoomed in the area around Vikeid, Norway and obtained
from half-space inversion of 7 kHz coaxial HEM data. Known and exposed graphite deposit
locations are shown by black squares. Semi-transparent water bodies are shown in light blue colour
394 V. C. Baranwal and J. S. Rønning
Fig. 14.11 Total magnetic field anomaly image zoomed in the area around Vikeid, Norway.
Known graphite deposit locations are shown by black squares. Semi transparent water bodies are
shown in light blue colour
Figure 14.11 shows processed total magnetic field anomaly data from the heli-
copter survey. We can see most of the graphite deposits occur in low magnetic
anomaly areas with locally noticeable amplitude which can be explained either due
the occurrence of negative induced magnetization i.e. diamagnetism from graphite
(Reynolds 2011) or due to a reducing magnetic environment. For an accurate
estimation of the quantity of the graphite occurrence in the subsurface, a good
knowledge of the petrophysical characteristics of not only the graphite but also for
the surrounding rocks is crucial, since the low magnetic anomaly can also be due to
a low magnetized host rock or even remanent magnetization.
The correlation of low apparent resistivity and low magnetic anomaly were
considered together to refine the possible graphite occurring areas for a rather large
survey area. Anomalous zones with resistivity <300 Ωm and magnetic anomaly
<300 nT from whole survey area (42 km 48 km) covering Langøya in
Vesterålen are shown in Fig. 14.12. Exposed graphite locations are plotted with red
squares. This image indicates likelihood of higher grade graphite occurrence from
the area which can be further investigated in more detail by ground geophysical
observations and core drillings. Some of these identified locations with and without
observed graphite at the surface were followed by rock samples collection, ERT,
shallow ground EM survey (i.e. EM31), self potential (SP) and induced polarization
(IP) measurements in summer 2018. Drilling and thereafter borehole logging was
also performed at few locations. Rønning et al. (2018) summarize results of the
geophysical and geological follow up. Occurrence of graphite deposits from various
locations marked in Fig. 14.12 is confirmed in the follow up and some of them are
reported to be with very high carbon content (up to 25%). Some of the graphite
deposits not exposed at the surface are confirmed by the drilling.
14 Airborne Geophysical Surveys and Their Integrated Interpretation 395
Fig. 14.12 Possible graphite deposit zones obtained by combining magnetic anomaly <300 nT
and apparent resistivity <300 Ωm from Langøya, Vesterålen area, northern Norway (From
Rønning et al. 2018). Known graphite deposits are shown by red squares with name of the areas
14.6 Discussions
Airborne geophysical surveys collect data from air either using a helicopter or a
fixed wing aircraft. Generally magnetic, radiometry and electromagnetic data are
collected from the air at the same time. Sometime gravity data is also collected
together with other geophysical data. This chapter discussed details of airborne
magnetic, radiometry and electromagnetic data. In airborne electromagnetic data,
only frequency domain helicopter data is discussed in detail though airborne time
domain data is also very common or even more popular nowadays.
396 V. C. Baranwal and J. S. Rønning
airborne magnetic and electromagnetic data refines a large survey area to increase
chances of finding graphite deposits. Many of the locations identified by the inte-
grated interpretation are confirmed by drilling and ground surveys (Gautneb et al.
2017; Rønning et al. 2017, 2018). Some of the deposits are not even exposed at the
surface and are covered by soil to make it impossible to find it without using such
geophysical techniques.
14.7 Conclusions
Airborne geophysical data can be collected very fast normally without harming the
local environment. Airborne survey can be performed as a reconnaissance survey
for scanning a vast survey area and narrowing it down to smaller areas for ground
follow up and the drilling. With advancement in hardware and computing tech-
nology, airborne geophysical data are as precise as ground geophysical data if
collected with proper sampling intervals and processed in a proper way. Nowadays,
it is very common to collect high resolution airborne geophysical data for general
geological mapping and to solve several other geophysical, geological, geotechnical
and environmental problems through them. Almost every country either has already
initiated or going to initiate soon to collect airborne geophysical data to refine and
update the pre-existing geological maps of the country. Airborne geophysical data
has direct applications in ground water exploration, mineral prospecting, geological
mapping, and radioactive and other environmental hazard management. These data
are also used in oil exploration and urban planning.
Three examples showed application of airborne EM surveys and their integrated
interpretation. First example showed use of airborne radiometry and airborne EM
data in exploration of a uranium rich alum shale area. Second example detailed use
of airborne EM data for quick-clay mapping and its comparison with ERT surveys
results. Third example highlighted integrated use of airborne magnetic and airborne
EM data for graphite exploration.
Acknowledgements We will like to acknowledge all our colleagues at NGU who were involved
in one or other way with different types of geological and geophysical work presented in this
chapter.
References
AarhusInv (2015) Manual for inversion program ver. 6.3 HydroGeophysics Group (HGG),
University of Aarhus, Denmark. http://www.hgg.geo.au.dk/HGGsoftware/em1dinv/em1dinv_
manual.pdf
Abraham JD, Cannia JC, Bedrosian PA, Johnson MR, Ball LB, Sibray SS (2011) Airborne
electromagnetic mapping of the base of aquifer in areas of western nebraska. USGS scientific
investigations report 2011-5219
398 V. C. Baranwal and J. S. Rønning
nonlinear inverse problems through the use of model reduction and forward model
approximations. This approach is also valid for classification problems where the
forward model is unknown and it has to be constructed. The aim of this paper is not
to provide new original results but showing the researchers why uncertainty
assessment is a compulsory step in inversion and how to deal with it in a simpler
way.
The inverse problems where the relation F is constructed are usually called
regression problems. In geology, this second kind of problems usually involve
machine learning techniques (Caté et al. 2017). Classification problems could be
used for instance to learn from historical data if a given situation or pattern will or
not happen.
The simplest case of an inverse problem consists in solving the linear system
Fm ¼ dobs , where F is a matrix with as many rows as observed data, dobs , and as many
columns as the dimension of the model m that we would like to identify. The linear
problem admits at least one solution if dobs can be written as a linear combination of
Pn
the column vectors of F since Fm ¼ mk ck with F ¼ ½c1 ; c2 ; . . .; cn . This implies
k¼1
that dobs 2 colðFÞ, whose dimension is the rank of matrix F. Therefore, colðFÞ is the
space of predictions and it is a subspace of the data space Rs . Besides the solution is
unique if the system fc1 ; c2 ; . . .; cn g is a basis set of colðFÞ. In this case the observed
data dobs can be uniquely expressed in the basis set fc1 ; c2 ; . . .; cn g and its coordinates
are ðm1 ; m2 ; . . .; mn Þ. Thus, solving a linear system is changing the system of reference
from the canonic base of Rn to fc1 ; c2 ; . . .; cn g. This situation happens when s ¼ n and
the rank of F is n. This is the case of full-rank endomorphisms represented by squared
matrices.
The linear system has an infinity of solutions when dobs 2 colðFÞ and
rankðFÞ\n. In this case the null space of F, kerðFÞ ¼ fw : Fw ¼ 0g, is nontrivial
and its dimension is n rankðFÞ. The null space of F is the uncertainty space of
F since it is formed by the vectors of the model space Rn that have no influence in
the data predictions. The solutions of the linear system form a linear variety ori-
ented by kerðFÞ that can be written as m ¼ mp þ kerðFÞ, where mp is a particular
solution of Fm ¼ dobs , that is, Fmp ¼ dobs .
In this case, it is possible to look for the minimum norm solution that has
no components in kerðFÞ. The solution belongs
to the orthogonal subspace of
kerðFÞ which coincides with col FT , that is, mMN ¼ FT z therefore
404 J. L. Fernández-Martínez et al.
1
mMN ¼ FT FFT dobs . The minimum norm solution is obviously unique. This
happens for purely underdetermined linear systems, which are those where s\n and
rankðFÞ ¼ s.
If the observed data does not belong to colðFÞ, due to the effect of noise in data or
due to modelling errors in F, then the linear system has not solution. In the case of
incompatible linear systems two different cases can be distinguished: purely over
determined and rank deficient linear systems.
A linear system is purely over determined when the number of observed data is
greater than the number of unknowns and the rank of F coincides with the number
of model parameters. In this case the least squares solution is unique and it is found
by performing the orthogonal projection of the observed data onto colðFÞ,
p ¼ projColðFÞ dobs , and solving the linear system Fm ¼ p. This is equivalent to
find the model mLS such as its prediction error E ¼ Fm dobs is orthogonal to
colðFÞ, what is equivalent to impose that the prediction error E to belong to
ker FT , which is the orthogonal subspace to colðFÞ. Therefore, the least squares
solution fulfils the well-known normal equations: FT FmLS dobs ¼ 0. The
least-squares solution is unique if the column vectors of F are independent. In any
other case, there will exist an infinity set of least squares solutions, that as in the
purely underdetermined case, lie in a linear variety that is oriented by kerðFÞ.
The second case is the rank deficient systems where the rank of F is smaller than
the minimum of the number of rows and columns in F. This implies that both, the
observed data and the model parameters are redundant. This situation usually
happens because the data and model spaces are oversampled. This kind of systems
usually have not a solution, that is, dobs does not belong to the space of predictions
colðFÞ. As in the previous case, the least-squares solution is found by orthogonally
projecting the observed data into colðFÞ, but in this case the column vectors of F do
not form a basis set, and the coordinates of the least squares solution cannot be
uniquely determined. Therefore, in this case the least squares solution is not unique
and belongs to a linear variety of the form m ¼ mLS þ kerðFÞ, where mLS is a
particular least-squares solution of the least-squares system FT Fm ¼ FT dobs . It is
important to remark that the null space of FT F coincides with the null space of
F. Therefore, the least-squares system is compatible but rank deficient, having an
infinity of solutions. The generalized least squares solution provided by the
Moore-Penrose pseudo inverse coincides with the minimum norm solution of the
least-squares system, that, as in the purely underdetermined case, has no compo-
nents in the null space of FT F.
Figure 15.1 shows a schematic representation of the purely underdetermined,
purely overdetermined and rank-deficient linear systems. The model space Rn can
15 How to Deal with Uncertainty in Inverse and Classification … 405
be written as direct sum of two orthogonal subspaces: kerðFÞ and col FT . In the
s
same way, the data space R can be written as direct sum of two orthogonal
subspaces: ker F T
and colðFÞ. Figure 15.1a shows the solution of purely
underdetermined linear problems. It can be observed that the linear operator has a
non-trivial kernel that orientates the linear variety of solutions. The minimum norm
solution is the one that lies in the
linear
variety and it is orthogonal to kerðFÞ. This
implies a projection onto col FT . Setting the non-informed coordinates of the
model is the right mathematical way to minimize the energy solution. Figure 15.1b
shows the solution of the purely over determined linear problem. It can be observed
that after orthogonally projecting dobs onto colðFÞ, the least squares system has a
unique solution. Finally, Fig. 15.1c shows the case of rank deficient systems where
T projection of d onto colðFÞ and those of the least squares
obs
a double orthogonal
solution onto col F are performed.
In the case of linear problems, the region of uncertainty (called Ltol ) it is defined
as follows:
kF m dk2
m 2 Ltol : Etol : ð15:5Þ
kdk2
Ltol is a hyper-ellipsoid whose axes are oriented following the right singular
V vectors of matrix F, given by its singular value decomposition F ¼ URVT , and
their axes are inverse proportional to their singular values (Fernández-Martínez
et al. 2012a). This result implies that the uncertainty has an anisotropic character
since it is not the same in all the directions. In fact, the ill-condition number of the
linear system is related to the eccentricity of this hyper-ellipsoid (Fernández-
Martínez et al. 2013). In the case of purely underdetermined and rank-deficient
systems the hyper-quadric of equivalence becomes an elliptical cylinder, that is, a
406 J. L. Fernández-Martínez et al.
rectilinear elongated valley with infinite length in the directions of the kernel of
F. In the case of nonlinear inverse problems, the region of uncertainty is a curvi-
linear valley and might be composed by several disconnected basins. The region if
uncertainty becomes curvilinear due to the dependency of the Jacobian of F on the
model m (Fernández-Martínez et al. 2012a).
15 How to Deal with Uncertainty in Inverse and Classification … 407
A partial sampling of the data introduces uncertainty in the identification. This fact
can be visualized via a simple linear regression problem that consists in fitting a
straight line y ð xÞ ¼ a0 þ a1 x to set of noisy data points fðxk ; yk Þ; k ¼ 1; . . .; sg.
The goal is finding the parameters m ¼ ða0 ; a1 Þ such the distance between the
observed data yobs and the corresponding predictions ypre is minimum according to
the Euclidean distance in Rs . This is a purely overdetermined system ypre ¼
a0 1Rs þ a1 x that hasno solution but it is possible to find mLS so that the prediction
obs
error FmLS yobs 2 is the minimum. For that FmLS ¼ Proj? h1Rs ;xi y , and the
T
prediction error E ¼ FmLS y 2 ker F , where F ¼ ½1R ; x.
obs s
coincides with the region of the model space inside the ellipse
2
mT FT Fm 2mT FT yobs þ yobs T yobs ¼ yobs 2 tol2 ; ð15:7Þ
that is centered in the least squared solution and oriented in the directions of the
eigenvectors of FT F. Performing the singular value decomposition of F,F ¼ URVT ,
the ellipse (15.7) referred to its principal axes is
2 2
a0V aLS a1V aLS
obs 2 þ obs 2 ¼ 1; ð15:8Þ
0V 1V
tolky k2 tolky k2
pffiffiffiffi pffiffiffiffi
k1 k2
where a0V ; a1V are referred to the principal directions of FT F (V basis set) and
a0V ; a1V are the model parameters and the least squares solution in V.
LS LS
Figure 15.2 shows the ellipses of uncertainty for relative error predictions of 10–
15% of a simple linear regression problem (that are determined analytically), the
least-squares solution located in the center of the ellipse, and the different sets of
parameters aLS 0k ; a1k k¼1;...;NS obtained in the least squares fitting of different NS ¼
LS
100 experiments constructed with NS partial random sets generated with the whole
set of observed data. It can be observed that these parameter models belong to the
408 J. L. Fernández-Martínez et al.
Fig. 15.2 Ellipses of uncertainty for relative misfit of 10, 12 and 15%, corresponding to the linear
regression problem, and the model parameters that have been sampled by the data kit inversion
procedure in the region of relative misfit
region of uncertainty and the sampling is denser along the axis of maximum
uncertainty of this ellipse. Therefore, if we associate the center of the ellipse to the
solution achieved with full information, a partial sampling in the data space
introduces uncertainty in the parameter identification. Also, the data bag procedure
used above is able to sample the region of uncertainty of this simple linear
regression problem, mainly along the axis of higher uncertainty. Figure 15.3 shows
in the previous example the posterior histograms of the model parameters in the
region of relative error of 10%. It can be observed that these histograms approxi-
mate quite well the Gaussian posterior distribution that is expected for this linear
regression problem.
The modelling errors come from the fact that forward model is approximate and
does not coincide with the true forward model. In the case of a linear problem
F ¼ Ftrue þ dF and the solution of the linear system ðFtrue þ dFÞm ¼ dobs provides
the same solution as Ftrue m ¼ dobs if m belongs to the null space of the modelling
errors dF. This is not usually the case. Therefore, the modelling errors are equiv-
alent to a noise effect dFm that depends on the model parameters, and are partic-
ularly hard to unravel when the forward model provide low data error misfits.
forward problem and computing the data misfit. If the likelihood increases the new
model is accepted, and if the likelihood decreases the new model can still be
accepted with a given probability that depends on the likelihood ratio of both
models, the initial one and the one that has been proposed. This algorithm generates
a Markov chain that converges with iterations to the posterior distribution of the
model parameters, which is the probabilistic counterpart of the nonlinear equiva-
lence region. The Metropolis algorithm has the drawback of being very slow and it
has as a major feature its exploratory character. Although it has not been theoreti-
cally proved, a similar approximate sampling can be performed faster via explora-
tory global algorithms in a procedure that it is known as sampling while optimizing.
This is the case of binary genetic algorithms when the mutation rate is high
(Fernández-Álvarez et al. 2008). This approximate sampling can be performed via
exploratory versions of Particle Swarm Optimization (PSO) (Fernández-Martínez
et al. 2008, 2010). Some examples of their implementation in geophysics can be
consulted in Pallero et al. (2015, 2017), Arung Laby et al. (2016), etc. This analysis
can be also performed using simulated annealing (see for instance Sharma and
Biswas 2013; Sen and Stoffa 2013; Biswas and Sharma 2014).
Therefore, exploration of the nonlinear equivalence is a crucial feature to per-
form a correct posterior sampling. The probabilistic approach of inverse problems
has been formerly proposed by Tarantola and Valette (1982a, b). The Bayesian
approach of inverse problems can be consulted in Scales and Tenorio (2001).
Type 2 decisions concern what is called sensitivity analysis, that is, the models
that have been sampled in Mtol are used to run the forward model of a second
experiment to study how the ambiguity in the “correct” determination of these
parameters affects the results obtained by using the optimum parameters that pro-
vide the minimum misfit in the corresponding inverse problem. Most of the times it
would suffice to be able to locate the presence of very different plausible scenarios
from the optimum solution that has been found and how this impacts the decision
that is going to be taken.
In any case some crucial questions remain: why to sample randomly if we can do
it smartly? Are practitioners lost in the Bayesian forest? Modeling methods should
theoretically correct but also feasible and efficient. We try in the next section how to
do that. Also, these approaches are general and not only concern geophysical
inversion but other fields such as biometry (Fernández-Martínez and Cernea 2015)
and biomedicine (DeAndrés-Galiana et al. 2016).
many solutions are compatible with the prior information and the observed data
within the same error bounds. Random sampling of these equivalent models is
impossible because of the curse of dimensionality (Curtis and Lomax 2001) and the
high computational cost needed for providing the corresponding forward
predictions.
The curse of dimensionality is clearly explained by Albert Tarantola (2005): “…
large-dimensional spaces tend to be terribly empty. Hitting by chance the circle
inscribed in a square is easy. Hitting by chance the sphere inscribed in a cube is a
little bit more di cult. When the dimension (n) of the space grows, the probability of
hitting the hyper-sphere inscribed in a hyper-cube rapidly tends to zero (for
n > 10).” This probability is even lower in the case of a linear inverse problem due
to the anisotropic character of the uncertainty region (Fernández-Martínez 2015). In
Tarantola’s words, “… sampling uncertainty in a nonlinear problem is like finding
a needle in a haystack” (Tarantola 2006).
These two complexities should be addressed separately. The high computational
cost can be solved by parallelization and also by using surrogate models with
simplified physics assumptions (if possible), or using data-driven black-box models
that try to mimic the response of the forward simulator based in a limited number of
intelligent chosen examples (inputs and outputs) (see for instance Gong et al. 2016;
Pereyra 2016). These are generalized interpolation models (Schölkopf and Smola
2002).
Reducing the dimensionality in inverse problems is needed since the model
parameters are correlated. This correlation is introduced by the forward problem to
fit the observed data. Some examples of how to use the model reduction combined
with approximate sampling can be consulted in Fernández-Martínez et al. (2011,
2012b) and using prior and posterior reduced basis sets (Fernández-Martínez 2014,
2015; Fernández-Martínez et al. 2017).
The P general expansion that it is used in both cases is
m ¼ l þ qk¼1 ak vk ¼ l þ Va, where l is a given trend model and
V ¼ v1 ; v2 ; . . .; vq , being v1 ; v2 ; . . .; vq the reduced basis set. In the case of
linear inverse problem reduces to FVa ¼ dobs Fl.
Therefore, the new matrix FV of the linear has only q unknowns, which is much
lower than original number of model parameters (n).
The prior bases vectors vk are constructed via Principal Component Analysis
(PCA) of a set of geological scenarios that are generated via geostatistical simu-
lation. Then, PCA reduces the dimension by finding an orthogonal basis set that
serves to summarize the prior geological variability that is present in these scenarios
(Fernández-Martínez et al 2012b). The inverse problem serves to constraint the
coefficients and fit the observed data. The posterior basis sets can be found per-
forming the spectral decomposition of a posterior model located in the nonlinear
equivalence region (Fernández-Martínez 2014; Fernández-Martínez et al. 2017).
These basis can be combines with different sampling methods as in the geometric
sampling approach (Tompkins et al. 2011a, b). This algorithm is an alternative way
of sampling the posterior as it has been shown in a simple 1D electromagnetic
412 J. L. Fernández-Martínez et al.
inverse problem (Tompkins et al. 2013). Other strategies concern the use of
orthogonal transformation to concentrate the energy of the Jacobian of the forward
operator into the first harmonics and computing an approximate pseudo-inverse
(Fernández-Martínez et al. 2017b; Fernández-Muñiz et al. 2017).
15.7 Conclusions
References
Arung Laby D, Santosa S, Santosa BJ, Bahri AS (2016) RR-PSO: fast and robust algorithm to
invert Rayleigh waves dispersion. Contemp Eng Sci 9(15):735–741
Biswas A, Sharma SP (2014) Optimization of Self-Potential interpretation of 2-D inclined
sheet-type structures based on very fast simulated annealing and analysis of ambiguity. J Appl
Geophys 105:235–247
Caté A, Perozzi L, Gloaguen E, Blouin M (2017) Machine learning as a tool for geologists. Lead
Edge 36(3):215–219
Curtis A, Lomax A (2001) Prior information, sampling distributions and the curse of
dimensionality. Geophysics 66(2):372–378
DeAndrés-Galiana EJ, Fernández-Martínez JL, Sonis ST (2016) Design of biomedical robots for
phenotype prediction problems. J Comput Biol 23(8):678–692
Fernández-Álvarez JP, Fernández-Martínez JL, Menéndez-Pérez CO (2008) Feasibility analysis of
the use of binary genetic algorithms as importance samplers. Application to a 1-D DC
resistivity inverse problem. Math Geosci 40(4):375–408
15 How to Deal with Uncertainty in Inverse and Classification … 413
Sen MK, Stoffa PL (2013) Global optimization methods in geophysical inversion, 2nd edn.
Cambridge University Press
Sharma SP, Biswas A (2013) Interpretation of self-potential anomaly over a 2D inclined structure
using very fast simulated-annealing global optimization-An insight about ambiguity.
Geophysics 78(3):WB3–WB15
Tarantola A (2005) Inverse problem theory and methods for model parameter estimation. Applied
mathematics series. SIAM
Tarantola A (2006) Popper, Bayes and the inverse problem. Nat Phys 2:492–494
Tarantola A, Valette B (1982a) Inverse problems = quest for information. J Geophys 50(3):159–
170
Tarantola A, Valette B (1982b) Generalized nonlinear inverse problems solved using the least
squares criterion. Rev Geophys Space Phys 20:219–232
Tibshirani R (1996) Regression Shrinkage and Selection via the lasso. J R Stat Soc Ser B
(Methodol) 58(1):267–288
Tompkins MJ, Fernández-Martínez JL, Alumbaugh, DL, Mukerji T (2011a) Scalable uncertainty
estimation for nonlinear inverse problems using parameter reduction, constraint mapping, and
geometric sampling: marine CSEM examples. Geophysics F263. https://doi.org/10.1190/1.
3581355
Tompkins MJ, Fernández-Martínez JL, Fernández-Muñiz Z (2011b) Marine electromagnetic inverse
solution appraisal and uncertainty using model-derived basis functions and sparse geometric
sampling. Geophys Prospect 1–19. https://doi.org/10.1111/j.1365-2478.2011.00955.x
Tompkins MJ, Fernández-Martínez JL, Fernández Muñiz Z (2013) Comparison of sparse grid
geometric and random sampling methods in nonlinear inverse solution uncertainty estimation.
Geophys Prospect 61–1:28–41