See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/376488789

ESTIMATING THE ELECTRICAL ANISOTROPY COEFFICIENT, STRIKE

DIRECTION, AND DIP ANGLE USING SQUARE ELECTRODE ARRAY DATA WITH
GENERALIZED REGRESSION NEURAL NETWORKS

Conference Paper · November 2023

CITATIONS READS

0 16

3 authors:

Ertan Peksen Doğukan Durdağ

Kocaeli University Kocaeli University
43 PUBLICATIONS 353 CITATIONS 11 PUBLICATIONS 7 CITATIONS

SEE PROFILE SEE PROFILE

Elnur Gasimov
Kocaeli University
6 PUBLICATIONS 2 CITATIONS

SEE PROFILE

All content following this page was uploaded by Elnur Gasimov on 14 December 2023.

The user has requested enhancement of the downloaded file.

 Eurasia Geoscience Congress and Exhibition 2023
November, 10-13, 2023, Antalya, TÜRKİYE
Organized by

ABSTRACT BOOK
 Eurasia Geoscience Congress and Exhibition 2023
November, 10-13, 2023, Antalya, TÜRKİYE
Organized by

10.1 ESTIMATING THE ELECTRICAL ANISOTROPY COEFFICIENT, STRIKE

DIRECTION, AND DIP ANGLE USING SQUARE ELECTRODE ARRAY
DATA WITH GENERALIZED REGRESSION NEURAL NETWORKS
E. Pekşen1*, D. Durdağ2 and E. Gasimov2

1
Assoc. Prof., Department of Geophysical Engineering, Kocaeli University, Kocaeli, Türkiye
2
Ph. D.(c), Department of Geophysical Engineering, Kocaeli University, Kocaeli, Türkiye
*Corresponding author e-mail: [email protected]

ABSTRACT
The Direct Current Resistivity (DCR) method is established as one of the oldest and most
extensively used techniques in applied geophysics. DCR measurements are influenced by
electrical anisotropy (EA). Electrical anisotropy characterizes the directional dependence of
electrical resistivity measurements. This study focuses on the integration of a square electrode
array with electrical anisotropy, offering an effective tool for estimating the strike, dip, and
electrical resistivities of subsurface layers along both the x and y directions of the medium using
a generalized regression neural network.

KEYWORDS: square array, electrical anisotropy, generalized regression neural network

INTRODUCTION
The Direct Current Resistivity (DCR) method is one of the oldest and most extensively employed
techniques in applied geophysics (Telford et al., 1990). DCR measurements are influenced by
several factors, including formation porosity, cementation factor, water saturation, fracturing,
resistivity of formation water, temperature, and electrical anisotropy (EA). Electrical anisotropy,
denoting the directional dependence of electrical resistivity measurements, has been a subject of
investigation since Maillet’s pioneering work in 1947. Numerous studies on electrical anisotropy
have been conducted to date. In this study, we focus on the application of a square electrode
array with electrical anisotropy. Square arrays have proven valuable for investigating the strike
and dip of subsurface layers (Habberjam, 1972; 1975), a particularly relevant technique given the
prevalence of karstic geological features in Türkiye, as demonstrated by Şener et al. (2021) in their
work on the Menekşe karst plateau.
Herein, we present an innovative approach that uses this effective square electrode array in
conjunction with the Generalized Regression Neural Network (GRNN) to estimate the strike and
dip of subsurface layers, as well as the electrical resistivities along both the x and y directions of
the medium.

138 | P a g e
 Eurasia Geoscience Congress and Exhibition 2023
November, 10-13, 2023, Antalya, TÜRKİYE
Organized by

METHOD and APPLICATION

Figure 1 illustrates a geological model incorporating a square electrode array. Within this model,
ρx represents the resistivity measured along the x direction, while ρy corresponds to the resistivity
measured perpendicular to the x-direction. The apparent resistivity value (ρa) can be calculated
using Equation (1):

2 1
( 1) − ( 1)
2 2
(1 + (𝑁 − 1) 𝑐𝑜𝑠 (𝜃))2
𝜌𝑛 (2 + (𝑁 2 − 1)(1 + 𝑠𝑖𝑛(2𝜃)))2
𝜌𝑎 =
2 − √2 1
−( 1)
{ (2 + (𝑁 2 − 1)(1 − 𝑠𝑖𝑛(2𝜃)))2 }
1
𝜌
where 𝑁 = ((1 + (𝑓 2 − 1) 𝑠𝑖𝑛( 𝛼)2 )) is the effective anisotropy coefficient. 𝑓 = √𝜌ℎ𝑦 is the electrical
2
ℎ𝑥

anisotropy coefficient of electrical. 𝜌𝑛 = √𝜌ℎ𝑦 𝜌ℎ𝑥 is the mean resistivity.  is the strike and  the
dip angle (Habberjam, 1972). Figure 1 also provides examples of the horizontal electrical
anisotropy model, depicting variations in electrical anisotropy, strike directions, and dip angles.

Figure1 (a) A square array and electrical anisotropic model. C1 ve C2 are current electrodes. P1 ve P2 are potential
electrodes. The distance among electrodes is a. (b) Various electrical anisotropies of the earth model response.
(c) Various strike directions of the earth model response. (d) Various dip angles of the earth model response.

139 | P a g e
 Eurasia Geoscience Congress and Exhibition 2023
November, 10-13, 2023, Antalya, TÜRKİYE
Organized by

Artificial neural networks (ANNs) are a class of nonlinear models designed to mimic biological
nervous systems. ANNs have been widely applied to solve many difficult problems, including in
different fields such as pattern recognition, signal processing, language learning, function
approximation, prediction, and modeling. Typically, a biological nervous system (consisting of
several layers, each composed of multiple neural units (neurons)) can process information in a
parallel manner. Models with these features are known as ANN models (Haykin, 1999).
GRNN was proposed by Specht in 1991. Its general structure consists of four layers divided into
input pattern collection, and output sections. GRNN has many advantages over other nonlinear
neural network techniques. A benefit of GRNN is that it does not require an iterative training
phase. In other words, the network learns directly (single pass learning) from the input data.
Another advantage is that predicted model individuals can encounter minimums and maximums
of the input data. Moreover, although local regime methods converge to undesirable solutions
corresponding to local minima, the GRNN method does not. The calculation time of GRNN
includes fast calculation time, database-based prediction ability, and low error rate.

RESULTS

Figure2 (a) Polar plot of a model with two parameters estimated by GRNN using synthetic data. (b) Polar plot of
a model with four parameters estimated by GRNN using synthetic data. (c) Training set of parameters space for
model (a) with two unknowns. (d) Training set of parameters space for model (b) with four unknowns.

140 | P a g e
 Eurasia Geoscience Congress and Exhibition 2023
November, 10-13, 2023, Antalya, TÜRKİYE
Organized by

Figure 2 shows the estimated model parameters using the GRNN method with synthetic data. The
estimated model parameters are given in Table 1.
Table1 Estimated model parameters using the GRNN.

Parameters ρx (ohm.m) ρy (ohm.m)  (degrees)  (degrees)

Model 100 15 120 45
Noise-Free 2 Parameters 119.21 44.94
Noise-Free 4 Parameters 100.94 14.58 120.17 45.70
Training Set Range (0-200] (0-200] [0-180] [0-90]

CONCLUSION
This study demonstrates the successful application of the GRNN method to estimate the strike
direction, dip angle, and anisotropy coefficient of an anisotropic earth model. We rigorously
tested this approach using synthetic data, both with and without noise, highlighting its
adaptability to karst environments and various geophysical problems. Effective parameter
estimation relies on defining the solution space for the model parameters, with the training set
range playing a crucial role in computational efficiency. Selecting an appropriate parameter range
is essential, because an excessively large range can lead to increased computational demands.

ACKNOWLEDGEMENT
This work is supported by the Scientific and Technical Research Council of Turkey (TÜBİTAK) under
grant 123Y277.

REFERENCES
Habberjam, G.M. 1972. “The effects of anisotropy on square array resistivity measurements”, Geophysical
Prospecting, 20 (2), 249–266.
Habberjam, G.M. 1975. “Apparent resistivity, anisotropy and strike measurements”, Geophysical Prospecting,
23 (2), 211–247.
Haykin, S. 1999. Neural Networks: A Comprehensive Foundation (second ed). Prentice Hall.
Maillet, R. 1947. “The fundamental equations of electrical prospecting”, Geophysics, 12 (4), 529–556.
Specht, D.F. 1991. “A general regression neural network”, IEEE Trans Neural Netw, 2(6):568 576.
Şener, A., Pekşen, E., Yolcubal, İ. 2021. “Application of square array configuration and electrical resistivity
tomography for characterization of the recharge area of a karst aquifer: A case study from Menekse
karst plateau (Kocaeli, Turkey)”, Journal of Applied Geophysics, 195/ 104474.
Telford, W. M., Geldart, L. P., Sheriff, R.E. 1990. Applied geophysics. Cambridge: Cambridge University Press.

141 | P a g e

View publication stats