Sunteți pe pagina 1din 4


of the IEEE International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA)
25-27 November 2014, Kuala Lumpur, Malaysia

A Complete Quantitative Analysis of Self-Potential

Anomaly Using Singular Value Decomposition
Arya Dwi Candra #1, Wahyu Srigutomo*2, Sungkono#3, Bagus Jaya Santosa#4
Department of Physics, Institut Teknologi Sepuluh Nopember
Surabaya 60111, Indonesia

Physics of Earth & Complex System Research Division, Institut Teknologi Bandung
Bandung 40116, Indonesia

AbstractA new quantitative interpretation method of self

potential anomaly related to geometric-shaped models such as
horizontal cylinder, vertical cylinder, and sphere object has been
proposed in this paper. This method is based on the concept of
solving least-squares algorithm with singular value
decomposition approach which is designed and implemented to
calculate the depth, the electric dipole moment, the polarization
angle, and the geometric shape factor of self potential anomaly.
This approach uses singular value decomposition algorithm to
solve non-linear inversion of self potential anomaly. The singular
value decomposition algorithm was randomly tested on
theoretical synthetic data which was generated by a chosen
statistical distribution from a known model with different
random noise level. The result shows there is a close agreement Fig. 1. Cross-sectional view of the vertical cylinder, the horizontal cylinder,
and the sphere models.
between the assumed and calculated parameters. Finally the
method validity is tested on the real self potential data anomaly decomposition (SVD) algorithm is proposed to overcome
which is obtained from a cylindrical object that was buried at almost all the drawbacks of all the above methods, particularly
certain depth. the disadvantage of handling random errors and noisy data.

KeywordsSelf-potential anomaly, non-linear inversion, In the other hand, calculation problem sometimes find
singular value decomposition algorithm, non-intrusive during the inversion process when the matrix is singular. If the
measurement matrix has a singular solution, then we will get a very small
eigenvalues which can lead to errors in the calculation. To
overcome these problems, then we have to develop a new non-
linear inversion technique using the SVD algorithm. This
Self-potential (SP) method is one of the oldest geophysics algorithm is effective to calculate the depth (z), the electric
methods. This method is firstly used by Robert Fox in 1830 to dipole moment (K), the polarization angle (), and the
find sulphide deposit in Cornwall, USA. The principle of this geometric shape factor (q) of SP data. The fitting between
method is to measure the static natural voltage which is located observed and calculated SP anomaly represent the accuracy of
at specific points on the ground surface. SP method is one of the result.
the simplest methods of geophysics that is inexpensive, non-
intrusive and fast deployment among other geophysical
methods. Further, the quantitative interpretation of SP
anomalies has been widely applied in field measurement such A. Self Potential
as mining exploration [1], cave investigation [2], volcanic
activity and geothermal [3], archeological surveying [4], and The general SP anomaly mathematical expression triggered
embankment seepage control [5]. by most polarized structures along a principal profile in a
Cartesian coordinate system is given by the following equation
Several methods have been developed to calculate the SP (Fig. 1) [7]:
parameters, such as curves matching [6], characteristic points
[7], least-squares [8], Levenberq-Marquardt [9], derivatives xi cos + z sin
V ( xi , z , , q ) = K
and gradient analysis [10]. However, most of these methods are
strongly affected by the noise in the measured data. In this
(x 2
i + z2 ) (1)
paper, a new least-squares method using singular value (i = N ,..., 1, 0,1,..., N )

978-1-4799-0841-3/14/$31.00 2014 IEEE

Where K is the electric dipole moment, z is the depth, xi is a 1
J T J J T = V 1U T (10)
discrete point where there are anomalies, is the polarization
angle between the axis of polarization and the horizontal, and
q is a shape factor which are 0.5 for vertical cylinder object, 1.0 Where J is Jacobi matrix, U and V are eigenvector of the
for horizontal cylinder object and 1.5 for sphere objects. matrix, and is eigenvalue of the matrix. Finally, SVD
algorithm is expressed by the equation:
Standard error () is used as a statistical criterion and
applied on field examples to demonstrate the efficiency and mn +1 = mn + V 1U T ( dobs dcal ) (11)
accuracy of the proposed interpretation technique.
Calculation using non-linear inversion has a disadvantage,
(V ( obs ) V ( cal ) )
m 2

= i =1 i i
(2) when invers of Jacobi matrix has a singular solution or
mn approach to singular so the eigenvalues of the matrix become
very small then the solution has gone beyond the linear region
Where is the standart error, Vi(cal) is the calculated SP (overshoot). Although the Jacobi matrix inversion is not
anomaly, and Vi(obs) is the observed SP anomaly at discrete singular matrix, the iteration result is diverging or very slowly
point xi. converging. SVD method is proposed to overcome the
disadvantage of the inversion process, particularly to perform
B. Singular Value Decomposition Algorithm more stability on matrix inversion.
SVD algorithm is derived from least-squares method that To test the stability of our method in dealing with noise,
expressed by the equation: each SP anomaly is calculated by adding 5% and 10% noise
level using the following equation:
mn +1 = mn + J T J J T ( dobs dcal ) (3)
Vrand ( xi ) = V ( xi ) + k (randn(i ))V ( xi ) (12)

Where m is the calculated solution at nth iteration, dobs is Where Vrand(xi) is the contaminated anomaly value at xi, k
observed data, dcal is calculated data, and J is Jacobi matrix that is the noise level, and randn(i) is a random normal number
can be shown in the equation below: which range is (0,1). The random normal number interval does
not include the extremes value, 0 and 1.
Vi Vi Vi Vi
z K q
J = # # # # (4)
A. Theoretical Synthetic Example
z Theoretical synthetic examples of SP anomaly has been
K q
used to study the SP anomalies across geometric models
shaped like a sphere, cylinder horizontal, and vertical cylinders
Where i is the number of iteration, V is the SP function and using following parameters assumptions: z= 5 m, = 10 , and
J is the Jacobi matrix. K= -600 mV. Parameter assumptions have been used in
The first derivative expression of eq. (1) is given by: equation (1) to produce the appropriate theoretical curves. A
new random data is produced by applying a continuous
V ( K sin( ) ) uniform distribution with maximum random error 5% and 10%
= (5)
z (2 zKq( z sin( ) + xi cos( ))) respectively, see Table 1. The real purpose of adding random
(z + x )
2 2 q
noise is to create data that is very close to the field
( z 2 + xi2 ) q +1
measurements. This allows the addition of random noise to
V ( z sin( ) + xi cos( )) validate the method that is developed according to the results
= (6)
K ( z 2 + xi2 ) q of synthetic data interpretation in Table 1. It shows that the
( K ( z cos( ) xi sin( ))) SVD method has the ability to solve non-linear equations very
= (7) well, which is characterized by the value of the inversion
( z 2 + xi2 ) q results in accordance with the assumed parameters.
V ( K log( z 2 + xi2 )( z sin( ) + xi cos( ))) In the 5% noise addition indicates that the data inversion
= (8)
q ( z 2 + xi2 ) q results still give a value corresponding to the model parameters
assumptions. The same inversion result of 10% noise addition
If the Jacobi matrix has a singular solution, so SVD still show a corresponding value to the parameter assumptions.
algorithm is expressed as follows: Based on Table 1, we can also see that the error percentage of
each parameter is less than random noise. Similarly, curve
J = U VT (9) fitting result of the synthetic data shows very good results for
each noise addition. Furthermore, the result of each subject
Therefore we obtain shape also shows very good result, which are shown in Fig. 2.
This is suggests that the SVD method is able to process
inversion data more accurately.
Synthetic data Calculated parameters
Geophysical Assumed No noise % of error 5% random % of error 10% random % of error
parameter parameter noise noise
Case 1
z 5.0000 5.0000 0.0000 4.9586 -0.8278 4.9896 -0.2075
k -600.0000 -599.9999 0.0000 -597.9330 -0.3445 -624.1518 4.0253
10.0000 9.9660 -0.3400 10.3626 3.6257 10.5793 5.7927
q 0.5000 0.5000 0.0000 0.5004 0.0891 0.5082 1.6335
Case 2
z 5.0000 5.0000 0.0000 4.9584 -0.8317 4.9027 -1.9456
k -600.0000 -599.9999 0.0000 -585.3769 -2.4372 -619.7067 3.2844
10.0000 9.9663 -0.3370 10.1900 1.9001 9.9300 -0.7004
q 1.0000 1.0000 0.0000 0.9934 -0.6597 1.0083 0.8272
Case 3
z 5.0000 5.0000 0.0000 5.0144 0.2886 4.9905 -0.1896
k -600.0000 -599.9999 0.0000 -615.0901 2.5150 -571.2692 -4.7885
10.0000 9.9655 -0.3450 9.9200 -0.7997 9.4804 -5.1958
q 1.5000 1.5000 0.0000 1.5050 0.3329 1.4944 -0.3762

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

Fig. 2. A SP anomaly curve fitting profile on synthetics data buried at depth z= 5 m with polarization angle = 40 and electric dipole moment K= -600 mV.
(a)(b)(c) shows SP anomaly profile at the surface over a vertical cylinder model (q= 0.5) with no noise, 5% and 10% respectively. (d)(e)(f) shows SP anomaly
profile at the surface over a horizontal cylinder model (q= 1) with no noise, 5% and 10% respectively. (g)(h)(i) shows SP anomaly profile at the surface over a
sphere model (q= 1.5) with no noise, 5% and 10% respectively.
Methods of Srigutomo et al Proposed method
Parameter Derivative analysis Least-squares Levenberg- Singular Value Decomposition
method method Marquardt method method
z 1.75 0.57 1.234 1.2325
k -18.20 -6.15 -10.733 -10.7308
38.34 67.91 45.780 44.3241
q 1.05 0.56 0.916 0.9225
% of error 1.115 0.748 0.537 0.587

that the proposed method has high accuracy and efficiency. In

sum, the present algorithm works very well in both theoretical
synthetic and field example data. The advantage of this method
is efficient in computational, and rapidly convergence to the
optimal estimation of the SP parameters. Therefore, the
proposed method can be recommended for routine analysis of
SP field data to determine the geophysical parameters
associated with the studied structures. Furthermore, this
method can be used to obtain subsurface geological
information, as described by the examples of the SP field data.

We would like to thank reviewers and editors for improving
the quality of the article. The main author greatly
acknowledges the Directorate General of the Higher Education
of the Republic of Indonesia (DIKTI) for supporting him
through DIKTI Graduate Scholarship grant. We also thank
Fig. 3. A curve fitting SP anomaly profile on field data. Alwi Husein and Yekti Widyaningrum for refining the English
used in this paper.
B. Field data example
To validate the applicability, the proposed method was
applied in SP field data from Srigutomo et al [9]. This SP [1] M. Sato and H. M. Mooney, The electrochemical mechanism of sulfide
self-potentials, Geophysics, vol. 25, no. 1, pp. 226249, Feb. 1960.
anomaly obtained from a buried cylindrical conductive object [2] Y. Vichabian and F. Morgan, Self potentials in cave detection, Lead.
at certain depth. The observed and calculated parameters for Edge, vol. 21, no. 9, pp. 866871, Sep. 2002.
implementation of the SVD method in real SP field data are [3] K. Yasukawa, T. Ishido, and I. Suzuki, Geothermal reservoir
shown in Fig. 3. From the Fig. 3, it can be seen that the curve monitoring by continuous self-potential measurements, Mori geothermal
fitting results of SVD method is better than the method field, Japan, Geothermics, vol. 34, no. 5, pp. 551567, Oct. 2005.
[4] M. G. Drahor, Application of the self-potential method to
developed by Srigutomo et al [9]. According to the result of archaeological prospection: some case histories, Archaeol. Prospect.,
Table 2, SVD method is an efficient and accurate method to vol. 11, no. 2, pp. 77105, 2004.
analyze SP anomaly. Therefore, as long as we consider the [5] J. R. Moore, A. Boleve, J. W. Sanders, and S. D. Glaser, Self-potential
fitting between the observed parameters and the calculated investigation of moraine dam seepage, J. Appl. Geophys., vol. 74, no.
parameters, the SVD methods provides best overall accuracy in 4, pp. 277286, Aug. 2011.
[6] B. V. S. Murty and P. Haricharan, Nomogram for the complete
the estimation of SP parameters. interpretation of spontaneous potential profiles over sheet-like and
cylindrical two-dimensional sources, Geophysics, vol. 50, no. 7, pp.
IV. CONCLUSION 11271135, Jul. 1985.
[7] S. H. Yungul, Interpretation of spontaneous polarization anomalies
A new interpretation method, based on the concept of non- caused by spheroidal ore bodies, Geophysics, vol. 15, no. 2, pp. 237
linear optimization using the SVD algorithm has developed in 246, Apr. 1950.
this study to determine the SP anomaly parameters. The [8] E. M. Abdelrahman, K. S. Essa, E. R. Abo-Ezz, M. Sultan, W. A.
proposed method is very well validate using theoretical Sauck, and A. G. Gharieb, New least-squares algorithm for model
synthetic data with various random noise, where the assumed parameters estimation using self-potential anomalies, Comput. Geosci.,
vol. 34, no. 11, pp. 15691576, Nov. 2008.
and calculated parameters is very close to an agreement. The [9] W. Srigutomo, E. Agustine, and M. H. Zen, Quantitative Analysis of
validity of this method is tested and proven in the SP anomalies Self-Potential Anomaly: Derivative Analysis, Indones. J. Phys., vol.
field obtained from cylindrical objects buried at a certain depth, 17, no. 2, pp. 4955, 2006.
where good agreement is obtained between the observed and [10] E. M. Abdelrahman, A. A. B. Ammar, H. I. Hassanein, and M. A.
calculated SP anomalies. The proposed method has a standard Hafez, Derivative analysis of SP anomalies, Geophysics, vol. 63, no.
3, pp. 890897, Jun. 1998.
error of 5% when applied on field example. This is indicates