Research Collection: New Developments in Electrical Resistivity Imaging

Research Collection
Doctoral Thesis
New developments in electrical resistivity imaging
Author(s):
Stummer, Peter
Publication Date:
2003
Permanent Link:
https://doi.org/10.3929/ethz-a-004526076
Rights / License:
In Copyright - Non-Commercial Use Permitted
This page was generated automatically upon download from the ETH Zurich Research Collection. For more
information please consult the Terms of use.
ETH Library
DISS. ETH NO. 15034
NEW DEVELOPMENTS IN ELECTRICAL

RESISTIVITY IMAGING
A dissertation submitted to the

SWISS FEDERAL INSTITUTE OF TECHNOLOGY ZURICH
for the degree of
Doctor of Natural Sciences
presented by
PETER STUMMER
Diplom-Ingenieur (M.Sc.)
University of Leoben (Austria)
born July 16, 1970
citizen of Austria
accepted on the recommendation of
Prof. Dr. Alan G. Green, examiner

PD Dr. Hansruedi Maurer, co-examiner
Dr. David E. Boerner, co-examiner
2003
Dedicated to Phil
Contents
Zusammenfassung ix
Abstract xiii
1 Introduction 1
1.1 Electrical Properties of Rocks . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Background Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Survey Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Depth of investigation . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Signal-to-noise ratio and geometric factor . . . . . . . . . . . . . . . 14
1.3.3 EM coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.4 Lateral resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Instrument Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.1 Vertical electric sounding (VES) . . . . . . . . . . . . . . . . . . . . 20
1.5.2 Electric profiling or areal mapping . . . . . . . . . . . . . . . . . . . 21
1.5.3 2D and 3D imaging . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.6 Motivation of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.7 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2 New Multi-Electrode DC Resistivity System 33

2.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 Adaptive Data Acquisition - Real-Time
Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4 Multi-Electrode Recording Systems . . . . . . . . . . . . . . . . . . . . . . 38
vi Contents
2.4.1 Centralized systems . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.4.2 Distributed systems . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4.3 Generation of the source currents . . . . . . . . . . . . . . . . . . . 41
2.4.4 Commercially available systems . . . . . . . . . . . . . . . . . . . . 42
2.5 A Novel Digital Multi-Electrode Recording System . . . . . . . . . . . . . . 42
2.5.1 Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5.2 Acquisition speed . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5.3 Data example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6 Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.7 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3 Laboratory Testing 53
3.1 Test Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2 Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4 Experimental Design 65
4.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3 Images Determined from Synthetic Comprehensive, Wenner, Dipole-Dipole,
and Combined Wenner - Dipole-Dipole Data Sets . . . . . . . . . . . . . . . 70
4.4 Identifying Optimal Data Sets . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.1 A fast approximate experimental design algorithm . . . . . . . . . . 73
4.4.2 Application of the simple experimental design procedure to synthetic
data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4.3 First attempts to account for the influence of heterogeneous resistivity
distributions in the experimental design procedure . . . . . . . . . . 79
4.5 Application to a Field Data Set . . . . . . . . . . . . . . . . . . . . . . . . 81
4.5.1 The Stetten landfill test site . . . . . . . . . . . . . . . . . . . . . . 81
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.7 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5 Outlook 101
5.1 Improvements to the ETH-DCMES . . . . . . . . . . . . . . . . . . . . . . 103
5.1.1 Weight reduction and mobility . . . . . . . . . . . . . . . . . . . . . 103
Contents vii
5.1.2 Reducing data acquisition time by distributing the data

analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.1.3 Surveying with long electrode spreads . . . . . . . . . . . . . . . . . 104
5.1.4 System expandability . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.1.5 Acquisition of induced polarization data . . . . . . . . . . . . . . . . 107
5.2 Influence of Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.3 Future Developments of the Experimental Design Strategy . . . . . . . . . . 110
A Resolution Capabilities 113
B ETH-DCMES Set-Up 117

B.1 Field Set-Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
B.2 Software Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
B.2.1 Array configuration . . . . . . . . . . . . . . . . . . . . . . . . . . 120
B.2.2 System preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
B.2.3 Acquisition parameters . . . . . . . . . . . . . . . . . . . . . . . . . 122
B.2.4 Optional parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 123
B.3 Measurement Sequence and Data Preparation . . . . . . . . . . . . . . . . 124
B.3.1 Measurement sequence . . . . . . . . . . . . . . . . . . . . . . . . 124
B.3.2 Data preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
B.4 Resetting ID’s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
B.5 DCINVIN Matlab Script . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
C ETH-DCMES Technical Details 141

C.1 Data Acquisition Unit (DAU) . . . . . . . . . . . . . . . . . . . . . . . . . 141
C.2 Interface Box (IFB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
D Ext. Abstract SEG 1999 149

D.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
D.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
D.3 Designing DC Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
D.4 Real-time Experimental Design for
Three-Dimensional DC/IP Problems . . . . . . . . . . . . . . . . . . . . . . 154
D.5 A New Digital Multi-Electrode Array . . . . . . . . . . . . . . . . . . . . . 154
viii Contents
D.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
E Ext. Abstract SPIE 2001 159

E.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
E.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
E.3 Real-Time Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . 160
E.4 A Novel Digital Multi-Electrode Recording System . . . . . . . . . . . . . . 162
E.5 System Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
E.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
E.7 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Acknowledgments 183
Curriculum Vitae 185

Zusammenfassung
Die Gleichstromgeoelektrik spielt seit dem Beginn der Anwendung geophysikalischer Messme-
thoden eine wichtige Rolle in einer Vielzahl umwelt-, hydro-, und ingenieurgeologischer Frage-
stellungen. Mit der Einführung von Multielektrodeninstrumenten und Inversionsalgorithmen
für zwei- bzw. dreidimensionale (2D, 3D) Anwendungen wurden hochauflösende Widerstands-
messungen zu einem essentiellen Werkzeug zur Untersuchung des untiefen Untergrundes.
Jedes geophysikalische Feldexperiment umfasst prinzipiell die Punkte (i) Planung des Fel-
dexperimentes, (ii) Datenaufnahme und (iii) Datenanalyse, wobei diese drei Punkte üblicherweise
unabhängig voneinander bearbeitet werden. Die Motivation dieser Doktorarbeit war die Kom-
bination der genannten Aspekte zu einem integrierten Prozess zur Steigerung der Gesamtef-
fizienz bzw. Erhöhung der Daten- und Modellqualität. Zwei Problemkreisen wurde dabei
besonderes Augenmerk gewidmet:
• Entwicklung und Aufbau eines Multielektrodenmesssystems, welches die Auswahl von

beliebigen Elektrodenkonfigurationen erlaubt.
• Erarbeitung eines Algorithmus zur Bestimmung von Elektrodenkonfigurationen, welche

die Akquisition von optimierten Daten zur Darstellung der Widerstandsverteilung des
untiefen Untergrundes erlauben.
Zur Ermittlung der räumlichen Lage und der elektrischen Eigenschaften von relativ kleinen
Untersuchungsobjekten die über große Gebiete verteilt sind, ist es in der Regel notwendig,
mit einer ausgedehnten, engmaschigen Anordnung von Elektroden über längere Zeitperioden
zu messen. Für die meisten akademischen wie auch kommerziellen Projekte ist ein derartiger
Aufwand jedoch zumeist nicht realisierbar bzw. ökonomisch nicht sinnvoll. Das neue Kon-
zept der Echtzeitoptimierung von Feldexperimenten stellt eine mögliche Lösung dar. Hierbei
strebt man die kostenoptimierte Akquisition begrenzter Datensätze an, welche die wichtigsten
Untergrundsinformationen enthalten. Dieses Verfahren beruht auf einer Anzahl von “Daten-
x Zusammenfassung
akquisition - Datenanalyse - Aktualisierung der Experimentgeometrie” - Zyklen und erfordert

daher den Einsatz einer hochflexiblen Messapparatur. Um dieser Anforderung gerecht zu
werden, habe ich ein verteiltes Multielektrodensystem entwickelt, welches Spannungssignale
direkt an den Elektroden digitalisiert bevor diese an den zentralen Steuerungsrechner trans-
feriert und für weitere Bearbeitungsschritte gespeichert werden. Die Akquisition qualitativ
hochwertiger, digitaler Daten wird durch den Einsatz von Datenaufnahmeeinheiten (oft auch
als intelligente Elektroden bezeichnet) mit sehr hohem Eingangswiderstand und großer dy-
namischer Bandbreite gewährleistet. Software mit grafischer Bedienungsoberfläche dient zur
Kontrolle aller Elektrodenfunktionen (z.B. ob diese als Strom- oder Spannungselektroden
arbeiten) bzw. der Wellenform der eingespiesenen Stromsignals. Das System konnte seine
Überlegenheit gegenüber konventionellen Widerstandsinstrumenten in Labormessungen bzw.
die Feldtauglichkeit während einer ausgedehnten Messkampagne unter schlechtesten Wet-
terbedingungen unter Beweis stellen.
Obwohl einfache Multielektrodenmesssysteme seit mehr als einem Jahrzehnt kommerzi-

ell erhältlich sind, basieren Anwendungen der Widerstandstomographie zumeist auf einer
oder mehrerer Standardelektrodenanordnungen (z.B.: Wenner oder konventionelle Dipol-
Dipol Anordnungen). Um die Möglichkeiten moderner Messsysteme besser zu nutzen, habe
ich ein Verfahren zur Optimierung von Elektrodengeometrien eingeführt. Mit dessen Hilfe
können Gruppen von Elektrodenkonfigurationen ermittelt werden welche einem vordefinier-
ten Optimierungskriterium genügen. Der Algorithmus besteht aus einer Gütefunktion zur
Klassifizierung der Sensitivitäten aller möglichen Elektrodenkonfigurationen in Abhängigkeit
von Veränderungen der elektrischen Untergrundsparameter. Zur Beurteilung des neuen Algo-
rithmus wurde ein umfangreicher Datensatz, der jegliche Informationen von Standard- und
Nicht-Standardelektrodenkonfigurationen enthält, für (a) ein komplexes 2D Widerstands-
modell errechnet bzw. (b) über einem ausführlich untersuchten Testgelände aufgezeichnet.
Untergrundsmodelle resultierend aus den umfangreichen Datensätzen wurden folglich als Re-
ferenz für Modelle aus optimierten Datensätzen herangezogen. Modelle aus relativ kleinen,
optimierten Datensätzen mit 265 - 282 Messwerten lieferten mehr Untergrundsinformation
als Modelle von Standarddatensätzen gleicher Größe. Mit Abstand die besten Modelle resul-
tierten aus optimierten Datensätzen mit 1000 - 6000 Datenpunkten. Diese Modelle lieferten
verlässliche Informationen aus etwa dreimal größeren Tiefenbereichen als Modelle basierend
auf Standarddatensätzen. Die ersten rund 600 durch den Optimierungsalgorithmus gewählten
Zusammenfassung xi
Elektrodenanordnungen waren hierbei asymmetrische Dipol-Dipol Konfigurationen, wohin-

gegen die folgenden ∼4800 Konfigurationen zu etwa gleichen Teilen aus asymmetrischen
Doppeldipolen bzw. verschachtelten (d.h., zumeist Gradienten oder andere asymmetrische
verschachtelte Anordnungen) Dipolen bestand.
Abstract
Direct-current (DC) resistivity techniques have played an important role in addressing a wide
variety of environmental, hydrogeological, and engineering issues. With the introduction of
multi-electrode data acquisition systems and two- and three-dimensional (2D, 3D) imaging
software, high-resolution resistivity surveying has become an essential element in many near-
surface investigations.
Conducting any geophysical field experiment includes (i) survey design, (ii) data acqui-
sition, and (iii) data analysis. These three aspects of an experiment are commonly treated
independently. The purpose of this doctoral project has been to combine these tasks for the
resistivity method into an integrated process that increases overall efficiency and improves
data and model quality. Two specific issues have been addressed:
• Design and construction of a multi-electrode data acquisition system that allows com-
pletely arbitrary electrode configurations to be selected.
• Development of an algorithm to identify electrode configurations that provide optimum

data for resolving the resistivity distribution of the shallow subsurface.
To determine the location and electrical characteristics of relatively small targets distributed
over wide areas, deployments of laterally extensive arrays of densely spaced electrodes for long
recording periods may be required. For many academic and commercial projects, such large
expenditures of effort are not plausible. The concept of real-time experimental design, in
which limited data sets that contain the most important subsurface information are recorded
in a cost-optimized sense, could be a possible solution. Real-time experimental design, which
involves a number of “data acquisition - data analysis - update survey design” cycles, requires
highly flexible recording equipment. To meet this need, I have developed a distributed data
acquisition system in which waveforms are digitized and partially processed at each electrode
before being rapidly transmitted to a central computer for storage, further processing, and
xiv Abstract
display. The data acquisition unit at each electrode has a high input impedance and a
large dynamic range, thus ensuring the recording of high quality digital data. Versatile
software controls the functions of all electrodes (e.g., whether they operate as current or
potential electrodes) and the waveform of the transmitted current. Besides demonstrating
its superiority over conventional resistivity equipment in laboratory testing, the new system
has shown its field-worthiness during an extensive campaign under harsh weather conditions.
Although primitive multi-electrode systems have been commercially available for more
than a decade, resistivity imaging of the subsurface continues to be based on data sets
recorded using one or more of the standard electrode arrays (e.g., the Wenner or conventional
dipole-dipole array). In an attempt to exploit better the full capabilities of multi-electrode
acquisition systems, I have introduced an experimental design procedure to identify suites
of electrode configurations that provide subsurface information according to predefined opti-
mization criteria. The experimental design algorithm includes a goodness function that ranks
the sensitivity of every possible electrode configuration to changes in the subsurface param-
eters. To examine the potential and limitations of the new algorithm, comprehensive data
sets that included data from all standard and non-standard electrode configurations were (a)
generated for a complex 2D resistivity model and (b) recorded across a well-studied test site
in Switzerland. Images determined from the resultant comprehensive data sets were used as
benchmarks against which the images derived from the optimized data sets were assessed.
Images from relatively small optimized data sets containing 265 - 282 data points provided
more information than those from standard data sets of equal size. By far the best images,
comparable to those determined from the much larger comprehensive data sets, were ob-
tained from optimized data sets with 1000 - 6000 data points. These images supplied reliable
information over depth ranges that were three times greater than the depth ranges resolved
by the standard images. The first ∼600 electrode configurations selected by the experimental
design procedure were non-standard dipole-dipole-type arrays, whereas the following ∼4800
electrode configurations were an approximately equal mix of non-standard dipole-dipole-type
arrays and nested configurations (i.e., mostly gradient and other non-standard arrays).
Chapter 1
Introduction
Electrical properties of earth materials vary over the broadest range of all petrophysical
parameters (Figure 1.1). As a consequence, electric and electromagnetic techniques can
be more sensitive to changes in the subsurface than other geophysical methods used in
environmental and hydrogeological applications. Electrical methods are defined by their
frequency of operation, the origin of the source signal and the manner by which the sources
and receivers are coupled to the ground. Signal frequencies range from a few Hz in direct-
current (DC) resistivity measurements up to several GHz in ground-penetrating radar (GPR)
surveys. Passive methods use electromagnetic (EM) fields created by natural phenomena
(e.g., telluric currents), whereas active techniques employ signal generators to produce the
necessary input signal. Sources and receivers can be coupled to the ground through galvanic
contacts (e.g., planted electrodes) or through EM induction (e.g., coils of wire). Taken
together, those possibilities result in a greater variety of field techniques than any other
prospecting method. Here, I focus on the well-established DC resistivity technique.
I will first summarize the basics of the DC resistivity method before discussing past re-
search in (i) survey design, (ii) instrument design and (iii) data analysis techniques. Progress
in these three research fields has not been uniform since the introduction of exploration resis-
tivity techniques in 1912, when Conrad Schlumberger conducted his first field experiments.
Furthermore, they are usually treated quite independently.
Interestingly, only a few survey design studies have been conducted in the past, such that
the Wenner, Schlumberger and dipole-dipole configurations continue to be regarded as state-
of-the-art. During the past 20 years, instrument design has improved significantly and data
analysis techniques have advanced markedly, allowing the construction of sophisticated three-
2 1. Introduction
dimensional (3D) models. These disparate levels of progress were an important motivation
for the initiation of this thesis.
Attempts will be made to improve survey and instrument design. With the concept of
real-time experimental design, I will outline how survey design, data acquisition and data
analysis can be efficiently interlinked.
1.1 Electrical Properties of Rocks
The electrical resistance R of a material sample is related to its physical dimensions, cross-
sectional area A and length l, through the resistivity ρ or conductivity σ:
1 RA
ρ= = , (1.1)
σ l
where ρ and σ are given in Ωm and S/m, respectively. Bulk resistivities of earth materials
in the depth range relevant to ore prospection or environmental investigations are mainly
controlled by electrolytic conduction. The current-carrying medium is usually an aqueous so-
lution of salts distributed through a complex structure of interconnected pores and fractures.
Resistivities of water-bearing rocks depend on the amount of water present, the minerals
contained in the water and the degree of interconnection between the pores.
Ions that conduct current in an electrolyte result from the dissociation of salts dissolved
in water. Since each ion can only carry a finite quantity of charge, the more ions available in
an electrolyte, the greater the charge that can be carried. Increases in temperature enhances
the mobility of ions, which results in decreases of resistivity (Keller and Frischknecht, 1966).
Since temperature variations in the subsurface are generally small, this influence is usually
negligible. However, in certain cases, such as the measurement of resistivity in permafrost
regions, the temperature influence can be significant. The presence of clay minerals decreases
the resistivity of water-bearing rock significantly through ion-exchange processes (Schön,
1996).
In general, hard rocks are poor conductors of electricity, but geological processes can
alter rock to reduce resistivity. Weathering, dissolution, hydrothermal alteration, faulting and
shearing can increase the porosity and permeability of rock, and hence decrease resistivity. By
comparison, compaction of sedimentary rock and metamorphism of all rock types may result
in lower porosities and permeabilities. Resistivity is, therefore, a widely varying parameter,
1. Introduction 3
Figure 1.1: Typical ranges of electrical resistivities / conductivities of earth materials (after Ward,
1990a).
not only from lithology to lithology, but also within a particular formation (Figure 1.1).
A series of petrophysical measurements has been made on rock samples to derive empirical
relations between lithology and resistivity. To address the needs of the oil industry, Archie
(1942) suggested an equation for calculating formation resistivity ρf from the pore-fluid
resistivity ρw and the porosity Φ of sandstones in the absence of clay minerals:
ρf = F ρw = aΦm ρw , (1.2)
where F is referred to as the formation factor, a is the saturation coefficient and m is the
cementation factor. F depends only on the pore geometry, because Φ = Vw /Vr only includes
measured volumes of rock Vr and pore water Vw . Values for a and m are chosen to make
equation 1.2 fit a particular group of measurements; a varies between 0.6 and about 1.0 and
m has values between 1.4 and 2.2. Values for the various factors and coefficients in Archie’s
equation are given by Keller and Frischknecht (1966).
Texture may also have a strong influence on resistivity (Figure 1.2). Pore space that con-
4 1. Introduction
Figure 1.2: Various textures of rocks (after Ward, 1990a).
tains electrolytes provides the means for current conduction through a rock. However, if the
pores are not connected to each other (i.e., so-called unconnected or dead-end pore space),
no closed path for current conduction can be formed, thus resulting in low permeability and
high resistivity of the rock. A typical example of a high porosity rock with low conductivity
due to its low permeability is basalt (Figure 1.2f). Recently, more complex petrophysical
models for current conduction through rocks have been developed. Schön (1996) reviews
these new attempts and gives an excellent overview of the newly gained insights.
1. Introduction 5
1.2 Background Theory
The DC resistivity method employs very low-frequency alternating currents (AC) as source
signals for determining subsurface resistivity distributions. Since magnetic properties can usu-
ally be ignored (e.g., Keller and Frischknecht, 1966; Telford et al., 1990) Maxwell’s equations
reduce to:
~ = 1 q,
∇·E (1.3)
ε0
~ = 0,
∇×E (1.4)
~ is the electric field vector in V/m, ε0 is the permittivity of free space (ε0 ≈
where E
8.854 · 10−12 F/m) and q is the charge density in C/m3 . Geophysical techniques generally
have to deal with 3D distributions of physical properties. Therefore, the following derivation
is given for all three spatial dimensions x, y and z. Although the equations are strictly only
applicable to continuous direct-current flow, they may also be used to represent the effects
of alternating currents (AC) at very low frequencies, such that displacement currents and
induction effects can be neglected. The electrostatic field can be described by the gradient
of the electric potential U :
~ = −∇U.
E (1.5)
Combining equations 1.3 and 1.5 results in the fundamental Poisson equation for electrostatic
fields:
1
∇2 U (x, y, z) = − q(x, y, z). (1.6)
ε0
The equation of continuity for a point in 3D space defined by the Dirac delta function is:
∂q(x, y, z, t)
∇ · ~j(x, y, z, t) = − δ(x)δ(y)δ(z), (1.7)
∂t
where ~j is the current density vector and t is time. Together with equation 1.5 and Ohm’s
law:
6 1. Introduction
~
~ = E,
~j = σ E (1.8)
ρ
equation 1.7 can be rearranged to:
∂q(x, y, z, t)
−∇ · [σ(x, y, z)∇U (x, y, z)] = δ(x − xs )δ(y − ys )δ(z − zs ), (1.9)
∂t
where xs , ys and zs define a point source of injected charge. The source term in equation
1.9 can be rewritten in a more practical form by considering an elemental volume ∆V about
the charge injection point:
∂q(x, y, z, t) I
δ(x − xs )δ(y − ys )δ(z − zs ) = δ(x − xs )δ(y − ys )δ(z − zs ). (1.10)
∂t ∆V
Here, I is the current driven by a point source, which is a good approximation of real field
situations, where metal stakes are usually used to inject current. Substituting equation 1.10
into 1.9 results in a partial differential equation for the electric potential in an isotropic
non-uniform 3D medium generated by a point source:
I
−∇ · [σ(x, y, z)∇U (x, y, z)] = δ(x − xs )δ(y − ys )δ(z − zs ). (1.11)
∆V
Numerical solutions of this equation can be used to model potential distributions within
arbitrarily conductive half-spaces. This will be discussed later.
For a homogenous isotropic medium, the electric field due to a point source can be
derived analytically. Integrating the equation of continuity 1.7 over a volume and applying
Gauss’ divergence theorem results in a surface integral of the current density ~j. Substituting
~j from Ohm’s law 1.8 and integrating over the surface of a sphere with radius r yields:
~ Iρ
E(r) = , (1.12)
4πr2
from which it is easy to show that:
Iρ
U (r) = . (1.13)
4πr
1. Introduction 7
Equation 1.13 describes the potential due to a single point source within a homogenous
space at a distance r from the injection point. This situation would apply to the so-called
mise-à-la-masse method, in which one of the two current electrodes is located at depth.
Resistivity techniques usually use electrodes deployed on the surface. For these solutions,
equation 1.12 only has to be integrated over a hemispherical surface to yield:
Iρ
U (r) = (1.14)
2πr
for the homogenous half-space.
To enable current flow through the conducting medium, a single point current source
can be achieved in theory by placing a corresponding current sink at infinity . Determination
of subsurface resistivities requires knowledge of the potential distribution in addition to the
input current (see equation 1.13). Given two current electrodes A and B in Figure 1.3 and
applying equation 1.14, the potential at arbitrary point M is:

Iρ 1 1
UM = − , (1.15)
2π r1 r2
where r1 is the distance between M and A and r2 the distance between M and B.
B A M N
Figure 1.3: Distribution of current and potential lines for two current electrodes at the surface of
a homogenous half-space (modified after Van Nostrand and Cook, 1966).
To measure potential differences, two electrodes are needed. Theoretically, the inject-
ing electrodes A and B could be used to measure the response signal. However, transition
8 1. Introduction
resistances between the electrodes and the subsurface would influence the measurements in
an unknown fashion (e.g., Cheng et al., 1990). A dedicated pair of electrodes for measur-
ing voltage differences completes the four-electrode array commonly used in DC resistivity
surveying. Subtracting the potential at point N from that at point M gives the potential
difference ∆U between M and N :

Iρ 1 1 1 1 Iρ
∆U = − − + = , (1.16)
2π r1 r2 r3 r4 K
where r3 is the distance between N and A and r4 the distance between N and B. Since K
only contains distances between electrodes, it is called the geometric factor. It depends only
on the relative distribution of electrodes. Finally, on rearranging equation 1.16 we obtain:
∆U
ρ=K . (1.17)
I
For an inhomogenous earth, this equation will produce values that vary according to the
geometrical arrangement of electrodes on the surface. Values obtained from equation 1.17
for an inhomogenous underground are referred to as apparent resistivities ρa .
1.3 Survey Design
In general, the four electrodes A, B, M and N (Figure 1.3) can be placed at arbitrary lo-
cations on the surface. However, a variety of specific electrode arrangements are commonly
employed. Each layout offers advantages in equipment handling or in measurement instru-
mentation. Several popular arrays are illustrated in Figure 1.4, together with their geometric
factors.
Different logistical and interpretational procedures may lead to preferences of one or more
arrays (e.g., Wenner, Schlumberger or dipole-dipole) over other arrangements. However,
selection of an electrode array for a field experiment should be based on the problem definition
(i.e., the aim of the survey). A variety of factors that influence the applicability of the arrays
to a particular situation have previously been investigated. In the following sections, four
aspects that need to be considered when comparing and selecting electrode arrays will be
discussed.
1. Introduction 9
e Electrode Arrangement Geometric Factor
ner
Schlumberger
Dipole-Dipole
Pole-Dipole
Pole-Pole
Gradient
Figure 1.4: Common electrode arrangements used in DC resistivity surveying and their geometric
factors (modified after Knödel et al., 1997).
1.3.1 Depth of investigation
Definitions of investigation depth
A central question in the characterization of resistivity measurements is the ability of an

electrode array to resolve certain resistivity contrasts at depth. The “depth of investiga-
tion”concept was introduced by Schlumberger and Schlumberger (1932). Early attempts
(e.g., Muskat and Evinger, 1941) assumed that only the depth of current penetration in the
ground was decisive in estimating the investigation depth. A higher percentage of current
10 1. Introduction
flowing through deeper portions of the subsurface would mean a greater depth of investi-
gation. Muskat and Evinger (1941) presented computations for two- and three-layer earth
models considering only the current electrodes A and B. The current penetration is only a
function of the distance between the A and B electrodes, whereas the depth of investigation
is determined by the geometry of all four electrodes.
In a paper ahead of its time, Evjen (1938) presented an alternative approach by inves-
tigating the influence that a buried thin layer had on surface measurements. This method
was further investigated by Roy and Apparao (1971) and Roy (1972, 1978). They calculated
the depth to a thin horizontal layer that resulted in a maximum response at the surface
electrodes. The results were presented as depth of investigation characteristic (DIC) curves
for different arrays overlying homogenous ground and two-layer models. A table of approxi-
mate depths of investigation relative to the characteristic lengths of the arrays was presented
together with a ranking of the vertical resolution of the individual arrangements. Edwards
(1977) modified the DIC concept by relating the median value of the response signal to
the depth of a thin layer. Although Edwards (1977) demonstrated the advantages of his
definition, later publications that extended the DIC concept to homogenous anisotropic half-
spaces (Bhattacharya and Sen, 1981) and gradient arrays (Bhattacharya and Dutta, 1982)
continued using Roy’s definition. Although Barker (1989) concluded that Edwards (1977)
definition of depth of investigation was the most reliable, Guptasarma (1981) showed that
the DIC concept produces questionable results for non-homogenous half-space models (e.g.,
layered structures).
Sensitivity analyses
More recently, the advent of affordable computing power has led to the calculation of sen-
sitivity patterns that consider subsurface conductivity distributions. Sensitivity calculations
determine the influence that small perturbations to the resistivities ∂ρ of a small volume
have on the measured values ∂U (for details see Section 1.5). The sensitivity S for a de-
fined source and receiver configuration can be expressed as the inner product of the current
densities j s and j r produced by a current source of strength I at the source and receiver
positions, respectively, integrated over the perturbed volume dV (after Geselowitz, 1971):
Z Z Z
∂U 1
S= = j s j r dV. (1.18)
∂ρ I V
1. Introduction 11
Spitzer (1998) described four different methods for determining sensitivities in three dimen-
sions and analyzed sensitivity patterns of several arrays for homogenous and layered models.
Friedel (2000) also discussed sensitivity patterns of various electrode configurations overlying
a homogenous half-space (Figure 1.5).
Figure 1.5: Examples of 2D sensitivity distributions of a homogenous half-space for different arrays:
(a,b) pole-pole; (c,d) pole-dipole; (e,f) dipole-dipole ABMN; (g) asymmetric Schlumberger; (h)
symmetric Schlumberger; (i,j) dipole-dipole AMBN; (k,l) Wenner (modified after Friedel, 2000).
Positive / negative sensitivities (red / blue areas) in Figure 1.5 indicate regions where
an increase in resistivity will lead to an increase / decrease in apparent resistivity. The
results can be interpreted as follows. Figures 1.5k and 1.5l for the Wenner array consist
12 1. Introduction
of a sequence of positive and negative sensitivity areas in close proximity to the electrodes.
Sensitivities beneath the midpoint of the array are nearly horizontal, suggesting that the
Wenner array is sensitive to vertical changes (i.e., horizontal structures) in the subsurface
resistivity and relatively insensitive to horizontal changes. The symmetric Schlumberger
array sensitivities (Figure 1.5h) have a focused pattern of positive sensitivities beneath the
array center. This directional characteristic makes the Schlumberger array well suited for 1D
sounding applications, because resistivity changes under the array center will result in large
variations of the measured values at M and N . In contrast to the previously mentioned
arrays, sensitivities for the dipole-dipole arrangement in Figure 1.5e and 1.5f have steeply
dipping positive values between the injecting and the measuring dipoles. These features
indicate high sensitivity to horizontal resistivity variations, which makes this array a good
choice for mapping vertical structures, such as dykes and cavities. Finally, sensitivities for
the pole-pole arrangement (Figure 1.5a, b) indicate that this array has the widest horizontal
coverage and greatest investigation depth, but the relatively wide spacing of the contours
reflect the poorest resolution of all arrays.
The above analyses do not provide details on the information content of the sum of all
measurements acquired during a field survey (i.e., the field data set). Moreover, there is little
information about the depth at which features in the resulting models can be interpreted.
Because investigation depths strongly depend on the unknown subsurface structure (i.e., the
distribution of resistivities), studies based on a homogenous isotropic half-space are a logical
starting point in designing a field experiment.
Forward modeling and inversion
Beside sensitivity analyses, two other approaches need to be mentioned in this context, (i)
forward modeling and (ii) inversion (discussed in more detail in Section 1.5). Both techniques
can be useful for evaluating the performance of arrays.
One of the first modeling attempts was made by Van Nostrand (1953). He developed
an analytic solution for the apparent resistivity measured by a Wenner array over a buried
conducting sphere. His solution shows that spherical bodies buried at depths greater than
their radius cannot be detected using the Wenner array.
With the availability of mainframe computer systems in the 1970’s, finite-difference (FD),
finite-element (FE), integral equation (IE), and transmission-network analogy (TNA) meth-
1. Introduction 13
ods became practical. Coggon (1973) implemented a FE solution (Coggon, 1971) to compute
the responses of dipole-dipole, pole-dipole and gradient arrays to ten different subsurface
structures. He summarized his results in a table of array performances.
Dey et al. (1975) use an implementation of TNA to analyze the perturbing effects due
to a vertical dike at various depths on Schlumberger, dipole-dipole, pole-dipole and unipole
measurements. To help compare apparent resistivities determined by these different arrays,
Dey et al. (1975) classify their results by introducing the anomaly index (AI):
Exmax − Exmin
AI = 100, (1.19)
E0
where Exmax − Exmin is the variation of electric field intensity measured across the buried
dike and E0 = Is ρ1 /6.25πa2 is the intensity at the center of a Schlumberger array with
potential dipole length a located over a homogenous half-space of resistivity ρ1 and source
current Is . This treatment of the results has the advantage of making the responses to
the inhomogeneity independent of the resistivity of the host rock ρ1 and dimensions of
the injecting and measuring dipoles. Dey et al. (1975) conclude their investigations with
characteristic diagrams for each array and a statement that the pole-dipole and unipole
(often referred to as focused array) configurations are significantly more effective in terms
of their responses to buried dykes than other arrays.
Another approach for tackling the problem of array classification was presented by Ap-
parao et al. (1992). They employed a tank filled with water as host medium and target
structures (sheet, cylinder and sphere) made of aluminium to model physically the resistivity
surveying problem. Based on the work of Van Nostrand (1953), they defined a depth of
detection as the depth below which a target cannot be detected with a given array assuming
that the value of the minimum detectable anomaly was 10 % above the resistivity of the host
rock. Based on their results, they concluded that the pole-pole array had the greatest depth
of detection followed by the pole-dipole array, whereas the Wenner array had the shallowest
depth of detection.
The introduction of low-cost computing (i.e., inexpensive fast personal computers) led
to further improvements in analyses of common array types. It was not only possible to
develop and run numerical modeling codes for arbitrary 2D and 3D resistivity problems,
but inversion schemes became feasible. Inverting field data sets became a powerful tool
in the quest to understand the influence of electrode geometries on measurements. Beard
14 1. Introduction
and Tripp (1995) implemented a volume IE solution to invert dipole-dipole, pole-dipole and
pole-pole measurements over four different structures. They compared the resolutions of the
arrays based on the results of 2D inversions and estimated the efficiency of the arrays for
resolving the structures under investigation. Dahlin and Loke (1998) used different inversion
algorithms to compare models derived from synthetic Wenner data sets and to estimate the
influence of data density on the results. Oldenburg and Li (1999) applied inversion theory
to develop a new parameter, the so-called “depth of investigation index”(DOI), which is
the depth below which surface data are insensitive to physical property variations. Based
on a code developed for induced polarization (IP) data (Oldenburg and Li, 1994), they
varied damping parameters and starting models to obtain DOI sections that they superposed
on a reference model. The resulting representations provided clear guidelines for further
interpretation, because areas where data do not constrain the model (i.e, regions at depth
not resolved) were clearly outlined. This technique is by far the most advanced approach for
estimating investigation depth.
1.3.2 Signal-to-noise ratio and geometric factor
Investigation depth is not the only issue that needs to be considered when selecting an
electrode array. For example, signal-to-noise ratios (SNR) may be key elements in the decision
process (e.g., Ward, 1990a). Assuming a colinear electrode arrangement, the SNR depends
on whether the potential electrodes M and N are placed within or outside of the current
dipole A and B. Voltages between M and N and the SNR will be higher for the former
than for the latter case. Of the classical arrangements, the Wenner array rates highest in this
regard, because the separation of the potential electrodes is a large fraction of the total array
length. Voltages and SNR are expected to be lower for Schlumberger measurements, because
the potential electrode separation is usually significantly smaller than the current electrode
separation (see Figure 1.4). A typical example of an array with an exterior potential dipole
is the dipole-dipole array. SNR’s for this array typically suffer from low voltages, especially
when the dipole separation factor n (see Figure 1.4) is large.
Directly related to the SNR is the geometric factor K (see Section 1.2). Values of K
reflect the range of potential differences to be expected with a particular electrode arrange-
ment. This fact can be used to assess the expected quality of a measurement. High values
of K indicate that low voltages will be recorded and vice versa. For example, assume a
1. Introduction 15
dipole-dipole arrangement with large separation factor n, such that the geometric factor can
take on values in the range 104 m to 105 m, then the observed potentials and the SNR will
be low.
Before starting data acquisition, it is useful to estimate the maximum geometric factor
for the planned configurations by rearranging equation 1.17 to:
Imax ρmean
Kmax = , (1.20)
∆Umin
where Imax is the maximum current the instrument can inject into the ground, ρmean is the
estimated mean resistivity of the subsurface under investigation and ∆Umin represents the
accuracy of the measurable voltage. ∆Umin not only depends on the maximum resolution
of the analogue-to-digital converter of the instrument, in most cases it will be governed by
the ambient noise level. This means that low measured voltages in combination with large
K-values may be overwhelmed by noise. Any electrode configuration with a geometric factor
greater than Kmax should not be considered for the survey.
1.3.3 EM coupling
Frequencies of source signals used in DC resistivity surveying are usually very low (from DC
up to about 50 Hz) to avoid electromagnetic effects. Yet most commercial instruments
use square waves or pulsed direct currents as source signals. Spectral analysis demonstrates
that such signals yield high frequency harmonics, which may result in coupling between
the two dipoles and the wires connecting them to the recording instrument. The basic
mathematical formulation of this problem dates back to Wait (1951), who developed an
analytic expression for transient fields due to a step function of current in a grounded wire
of finite length embedded in an infinite medium.
Based on this work the following conclusions can be drawn: EM coupling increases with
(i) frequency, (ii) the electrode configuration and (iii) the conductivity of the half-space. Gen-
erally, the coupling expressions are rather complex (Bhattacharyya, 1957). Analytic solutions
only exist for a few simple cases. For complicated earth models, numerical approximations
for the coupling have been developed in the context of IP surveying, because electromagnetic
coupling has a similar effect on measurements as induced polarization does (Dey and Mor-
rison, 1973; Pelton et al., 1978). Consequently, it is important to understand and eliminate
16 1. Introduction
electromagnetic coupling effects before interpreting IP data. Stoyer and Greenfield (1976)
have developed a FD formulation for modeling the electromagnetic coupling of a 2D sub-
surface, and Unsworth et al. (1993) have used a FE approximation to expand this idea to
model a 3D source over 2D structures (often referred to as the 2.5D problem).
None of the above mentioned publications provides guidelines with which to judge the
degree of coupling associated with each electrode configuration. For the dipole-dipole array,
large distances between the current and potential dipoles and separation of their connecting
cables solves the problem by simply avoiding its cause. The pole-dipole, gradient, Schlum-
berger and, Wenner arrays are progressively more susceptible to coupling effects.
Beside the array geometry, the frequency is an important parameter when considering
electromagnetic coupling. For a square-wave input signal, cycle times of the injecting signal
should be chosen long relative to the decay time, such that the measured response has
reached a stable value. If a slowly varying sinusoidal signal is employed, coupling effects
would be suppressed due to the lack of rapidly changing currents and the low frequency.
Unfortunately, whether the recorded peak value of the response signal represents the DC
part of the resistivity or wether complex resistivities have to be considered would not be
clear. In the latter case, a series of measurements with different source-signal frequencies
would be necessary to evaluate the phase shift together with the resistivity amplitude (this
technique is known as complex resistivity or spectral IP).
1.3.4 Lateral resolution
Only a limited amount of research has been directed toward determining the lateral resolving
power of electrode arrays. Coggon (1973) compared three arrays with respect to their lateral
resolving capabilities. His analysis focused on the dipole-dipole, pole-dipole and gradient
arrays, of which the latter one is rarely used in practical field work. He demonstrated the
superiority of the gradient array for accurately locating steeply dipping inhomogeneities. In
contrast, Dey et al. (1975) established that the Schlumberger array ranks with the Wenner
array in its lateral resolving power, and further suggested that a focused array is superior
to all conventional arrays. A focused electrode arrangement for surface use is basically a
Schlumberger array for which the positive current is split between electrodes A1 and A2
(situated at positions A and B of a conventional Schlumberger array) and electrode B is
situated at infinity (i.e., at great distance from the array). For this configuration, the current
1. Introduction 17
is focused at a depth that is dependent on the dimensions of the array.

Most research over the past 70 years has focused on a small number of well-established
electrode arrangements (e.g., Figure 1.4). Even the advent of modern multi-channel instru-
ments (see following Section 1.4) has made little impact on the way resistivity data have
been collected. Only recently did Noel and Xu (1991) start investigations that considered
the number of measurements needed to extract maximum subsurface information. They
refined and extended their ideas in a paper on the completeness of data sets (Xu and Noel,
1993), in which they describe a complete data set that contains all linearly independent
measurements of a specific array. Lehmann (1995) adopted Xu and Noel’s (1993) non-
classical arrangement, established its completeness, and tested its performance in the field.
All three publications have shown that classical electrode arrangements result in incomplete
data sets, whereas other arrangements may deliver more details on the subsurface resistivity
distribution.
1.4 Instrument Design

An important advantage of the DC resistivity method has always been the low-cost equipment
necessary to start surveying (see Figure 1.6). The classical set of tools consists of the
following parts:
• a constant current source, which in the simplest case is a battery pack connected to a
commutator circuit (to change polarity of the source current);
• an ammeter to measure the injecting current;
• a sensitive voltmeter to measure the response signal;
• four metal stakes (electrodes);
• and four cable reels to connect the electrodes to the source and voltmeter.
Instead of commutated DC sources, low frequency AC signals (Christensen, 1989) can

be employed as a means to suppress the effect of electrode polarization. To minimize the
influence of the skin effect and electromagnetic coupling (Section 1.3.3), the frequencies are
usually low. To estimate a resistance value, it is necessary to measure the current of the
injecting signal and the voltage response of the subsurface. The ammeter, which is placed
18 1. Introduction
Figure 1.6: Basic set-up for DC-resistivity measurements using four electrodes (after Robinson
and Coruh, 1988).
in series, should have a low internal impedance to minimize its effect on the measuring
circuit. For the same reason, the voltmeter, which is placed in parallel, should have a
high input impedance. Generally, the source and both meters are contained within a single
box. For conventional DC resistivity surveying under normal field conditions (i.e., good
coupling conditions between the electrodes and the ground, etc.), no special requirements
apply to the electrodes. In some regions, low-resistance coupling to the soil is needed to
avoid unwanted voltage-drops. This would require the use of high-quality metal stakes.
For surveys demanding the highest accuracy, non-polarizing Cu − CuSO4 or P b − P bCl2
potential electrodes should be employed. These have very-low impedance characteristics.
Using the four-electrode set-up shown in Figure 1.6, acquisition of a simple vertical-
sounding data set is time consuming, and the recording of large 2D or 3D data sets is
impractical. In particular, moving the electrodes and the resistivity instrument is inefficient
and expensive for large scale investigations.
1. Introduction 19
Cable Drums + Switching Units

Digital Multicore
Portable Interface Interface Cable
Computer Box
Resistivity
Meter
Electrodes 1 2 3 4
Figure 1.7: Layout of a microprocessor-controlled 25-electrode data acquisition system using ad-
dressable switches that enable any four electrodes to be connected to the resistivity meter (modified
after Griffiths and Barker, 1993).
Multi-electrode data acquisition systems (Figure 1.7) have partly resolved this prob-
lem. It was the rapidly evolving electronics industry that led to the initial development
of multi-electrode systems. Inspired by research in medical resistivity imaging, the first
microprocessor-controlled systems became commercially available in the late 1980’s. Finally,
the advent of inexpensive analogue-to-digital converters (ADC), computing power and large
mass storage devices led to fully automated resistivity systems comprising up to 1000 elec-
trodes. Different hardware design strategies that offer advantages either in field handling
(e.g., systems designed for roll-along acquisition of long profiles) or acquisition speed (multi-
channel design) have evolved. Unfortunately, there continues to be a lack of flexibility in
terms of switching capabilities of commercial multi-electrode systems. Most systems con-
tinue to be based on the classical electrode arrangements (e.g., Wenner, Schlumberger, etc.).
A discussion of instrument design philosophies is provided in Chapter 2.
20 1. Introduction
1.5 Data Analysis
Depending on the field procedure employed, several investigation techniques for delineating
resistivity contrasts in the subsurface can be distinguished:
• vertical electric sounding (VES)
• electric profiling or areal mapping
• 2D and 3D imaging.
The first two methods will only be mentioned for the sake of completeness. I will focus
on the analysis of 2D and 3D data sets, which are common in modern resistivity surveying.
1.5.1 Vertical electric sounding (VES)
VES is designed to provide vertical profiles of resistivity versus depth. This technique is
based on the general observation that current penetrates deeper into the subsurface with
increasing separation of electrodes. The actual depth of penetration depends on various
factors (Section 1.3.1). Although most electrode configurations can be used for VES, the
Schlumberger arrangement (Figure 1.4) offers important logistical advantages (only two
electrodes have to be moved at a time).
Interpretations of VES data are based on the assumption that the area under investigation
is underlain by a finite number of horizontal layers. Primative analysis involves comparing
the sounding curve shape (logarithmic apparent resistivity versus logarithmic half-distance
of current electrodes) with model curves of typical resistivity structures. Figure 1.8 shows
a suite of three-layer model curves. The presence of turning points indicates boundaries
between the layers.
During the 1960’s and 1970’s, a variety of numerical algorithms for the computation of
apparent resistivity versus depth were developed (e.g., Mooney et al., 1966; Koefoed, 1979).
This was soon followed by the introduction of inversion schemes that allowed 1D models
to be derived on computers (Inman et al., 1973; Zhody, 1989), most recently on personal
computers in the field.
As for most other geophysical techniques, interpretations of resistivity data (1D, 2D or
3D) are inherently non-unique. This shortcoming, expressed by the principles of equivalence
1. Introduction 21
Figure 1.8: Characteristic apparent resistivity curve shapes derived from three-layer models (after
Lowrie, 1997).
and suppression, is nicely described by Kunetz (1966). For example, only the product of layer
thickness and resistivity of a resistive layer between two conductive layers can be uniquely
determined. As long as the product remains constant, layer thickness and resistivity may be
varied without affecting the resulting model. A similar situation exists for a conductive layer
between two resistive layers. Here, only the quotient of thickness and resistivity affects the
result. Suppression refers to the fact that a relatively thin layer of intermediate resistivity
between two layers of increasing or decreasing resistivity has very limited influence on the
VES curve.
1.5.2 Electric profiling or areal mapping
Profiling and areal mapping are recording techniques for mapping lateral variations in re-
sistivity at approximately constant depths. This is achieved by moving a fixed-length array
along traverses or across an areal grid. Figure 1.9 shows a typical profile acquired along a
traverse across a clay-filled feature within limestone host-rock. Measurements were made
every 5 m using a Wenner array with constant 10 m electrode separation (see Figure 1.4).
Variations in apparent resistivity highlight anomalous areas.
The Wenner array is especially well suited for profiling and areal mapping due to the
22 1. Introduction
Figure 1.9: A constant-separation profiling data set using a Wenner array with 10 m electrode
separation (after Reynolds, 1997).
equidistant spacing between electrodes. Often, a few VES data sets are recorded prior to
a profiling or mapping campaign to estimate the necessary electrode spacing (Zhody et al.,
1973). Besides the well-established Wenner arrangement, the gradient array (see Figure 1.4)
might, in some situations, be a good choice for profiling or mapping, because only two
electrodes have to be moved (Furness, 1993).
1.5.3 2D and 3D imaging
Most applications in engineering or environmental geophysics require high-resolution resistiv-

ity information in two or three dimensions. By combining the previously discussed techniques
of sounding and profiling / areal mapping, 2D and 3D images can be determined. In the
field, this combination can be realized by either collecting VES data at each surface location
or by repeating the profiling / areal mapping with successively increasing electrode spacings.
Alternatively and preferably, multi-electrode arrays are increasingly employed. Recording of
3D data sets is often realized by using the pole-pole configuration on a square grid of regu-
larly spaced electrodes (e.g., Loke and Barker, 1996a). If the available system does not offer
enough electrodes to cover a full grid, a series of closely spaced parallel lines may suffice.
Sørensen (1994, 1996) has developed a pulled-array recording system (PACEP) that adapts
1. Introduction 23
the well-known streamer concept from off-shore seismic surveying. A set of eight heavy
steel electrodes connected to a heavy-duty cable is pulled by a small catapiller over areas of
interest. While two electrodes inject current, data are recorded continuously from the other
six electrodes in three different configurations. Shallow images extending over distances of
several kilometers can be recorded during a single day with this system.
A common way to present apparent resistivity data is to plot the recorded values beneath
the array midpoints at depths equal to a specified fraction (usually 1/3 or 1/2) of the array
lengths (Figure 1.10). Contouring of the resistivity values reflects variations in apparent
resistivity along the surveying line. Since the depths are not true depths, such a plot is called
pseudosection (Hallof, 1957). Even though true depth information cannot be directly inferred
from pseudosections, they are valuable tools for qualitative analyses and quality control (e.g.,
Dahlin, 1993, 1996).
Once a 2D or 3D data set is acquired, it is necessary to model or invert the data to obtain
the resistivity information. As for most geophysical problems, the data are related to the
model parameters in a non-linear way, such that standard inversion schemes developed for
linear problems cannot be employed. Non-linear inversion problems are handled by iterative
approximation schemes that require the forward problem to be solved at every iteration step.
Figure 1.10: Construction of a pseudosection for dipole-dipole resistivity data (after Lowrie, 1997).
24 1. Introduction
The forward problem
Solving the forward problem involves deriving a theoretical response (synthetic data) for a
suite of input parameters, given the appropriate equations that relate the model to the data.
In DC resistivity, this means solving Poisson’s equation 1.6 or equation 1.11. The elliptic
partial-differential equation 1.11 can be solved either analytically or approximately using
numerical approaches. Since analytic solutions exist for only a few rare and rather simple
cases, almost all practical cases (e.g., arbitrary resistivity distributions) require numerical
solutions. Generally, FD and FE techniques are used to solve the differential equations of
the DC resistivity forward problem in two or three dimensions.
Mufti (1976) first developed an FD-scheme for the modeling of arbitrary resistivity struc-
tures in two dimensions. His scheme was based on line sources. Dey and Morrison (1979b)
enhanced Mufti’s implementation by introducing point-sources for current injection. This al-
gorithm was further expanded to cover 3D problems (Dey and Morrison, 1979a) and formed
the basis of several later publications. A different discretization scheme and an enhanced
algorithm for solving the resulting difference equations has been suggested by Spitzer (1995).
Although the algorithms of Dey and Morrison (1979b) and Spitzer (1995) provide accurate
results for the voltages at most FD-grid points, errors of 5 - 15 % are encountered in the
close vicinity of the source points. In fact, whenever the right side of equation 1.11 is not
smooth, a singularity arises in the solution of the partial-differential equation. This situa-
tion is problematic, because forward solutions for common four-point electrode arrays are
generated by the superposition of four pole-pole solutions according to:
U4−point = UAM − UAN − UBM + UBN , (1.21)
where U4−point is the desired potential difference of a standard four-point array (e.g., Wenner)
and UAM is the computed voltage at grid-point M for current injected at grid-point A, etc.
Errors caused by singularities at grid points surrounding the injection point may accumulate,
leading to unacceptable results. The problem can be solved by simply refining the FD
grid near the singularities, thus reducing the truncation error. Unfortunately, to achieve an
acceptable effect, the grid spacing has to be chosen very small, which in turn results in
significant increases in computation time.
1. Introduction 25
An alternative and more effective solution has been proposed by Lowry et al. (1989)
and refined by Zhao and Yedlin (1996). Their so-called singularity removal scheme splits the
potential U (x, y, z) in equation 1.11 into a regular part Ur (x, y, z) caused by the conductivity
distribution of the FD-grid cells and a singular portion Us (x, y, z) due to the singularity:
U (x, y, z) = Ur (x, y, z) + Us (x, y, z). (1.22)
The singular part of the field Us (x, y, z) can easily be calculated for simple resistivity distri-
butions (e.g., homogenous or layered half-space). It is given by the equation:
I
−∇ · [σs (x, y, z)∇Us (x, y, z)] = δ(x − xs )δ(y − ys )δ(z − zs ), (1.23)
∆V
where σs (x, y, z) is the conductivity at the source point in the case of a homogenous half-
space. Substituting equation 1.22 into the left side of equation 1.11, subtracting equa-
tion 1.23 and rearranging yields:
∇ · [σ(x, y, z)∇Ur (x, y, z)] = −∇ · {[σ(x, y, z) − σs (x, y, z)]∇Us (x, y, z)}, (1.24)
where all known parts are on the right side of the equation and the unknown potential dis-
tribution Ur (x, y, z) is on the left. Equation 1.24 is discretized using the same scheme as
described by Dey and Morrison (1979a). The advantage gained by this method is twofold.
The error at the current injection point is reduced to machine precision and the method
pre-conditions the source vector of the resulting matrix equations, thus accelerating conver-
gence. Weller et al. (1996) expanded this methodology to IP measurements using complex
conductivities.
Although FE solutions are less frequently used for resistivity modeling, surface topog-
raphy is easier to model using FE approaches than using FD techniques. Pridmore et al.
(1981) discussed the application of FE methods to resistivity and EM modeling, and Sasaki
(1994) has implemented this technique in his inversion algorithm. A simplification of the
FE technique is the boundary element method, in which only the boundaries between media
of constant resistivity need to be discretized and integrated. Lesur et al. (1999a) used this
approach to formulate the forward problem for crosshole tomography.
In contrast to the FD and FE implementations, which are somewhat comparable in
26 1. Introduction
their underlying principles, the TNA method approximates the subsurface with a network of
lumped impedances, such that the impedances are proportional to the model resistivities.
The TNA method has not been widely used in DC resistivity forward modeling. However, a
3D implementation of this method was presented by Zhang et al. (1995).
The inverse problem
DC resistivity data are acquired by measuring response voltages (see Section 1.4). Written
~ obs are related to the true model parameters ρ~
in vector notation, the observed voltages U
(subsurface resistivities):
~ obs = F(~
U ρ ), (1.25)
where F is the non-linear forward operator. Simple linearization procedures such as those
employed for the refraction delay-time problem, are rarely found for highly non-linear prob-
lems. A general strategy for handling such problems was first suggested by Gauss. His
scheme involves the conversion of the non-linear problem into an approximately linear form
ρ ) in a Taylor series about an initial model guess ρ~ 0 , followed by application
by expanding F(~
of methods to solve the linear problem. The starting model ρ~ 0 might be based on some a
priori information (e.g., results from previous investigations) or is simply an intelligent guess.
~ calc corresponding to the initial model ρ~ 0 is:
The theoretical response U
~ calc = F(~
U ρ 0 ). (1.26)
ρ ) is assumed to be linear around ρ~ 0 , such that a small perturbation

According to Gauss, F(~
of the model responses ρ~ − ρ~ 0 about ρ~ 0 can be approximately expressed by only maintaining
the first term of a Taylor expansion:
p 
X
0 ∂F(ρ) 
F(~
ρ ) ≈ F(~
ρ )+  ρj − ρ~j0 ),
(~ (1.27)
∂ρj 
j=1 ρ ρ0
~=~
~ obs will differ

where j = 1, ..., p model parameters have been assumed. The observed data U
~ calc by the prediction error ~e:
from the theoretical response U
~ obs − U
~e = U ~ calc = F(~ ρ 0 ).
ρ ) − F(~ (1.28)
1. Introduction 27
Combining equations 1.28 and 1.27 leads to:
p 
X ∂F(ρ) 
~e =  ρj − ρ~j0 ).
(~ (1.29)
j=1
∂ρj  ρ ρ0
~=~
Introduction of δ~ ρ − ρ~ 0 ) the unknown model adjustments and denoting ∂F(ρ)/∂ρj as

ρ = (~
S̄, results in the desired linear equation:
~e = S̄δ~
ρ. (1.30)
Matrix S̄ containing the partial derivatives of F with respect to each model parameter ρj
is the Jacobian matrix and the derivatives are the sensitivities discussed in Section 1.3.1.
Equation 1.30 allows the inversion problem to be formulated as an iterative optimization
process. The objective is to find model adjustments δ~
ρ that minimize the difference between
observed and predicted data ~e. The function to be minimized is:
2
ρ ⇒ minimum.
~e − S̄δ~ (1.31)
The least-squares approach is the method of choice for many such minimization problems,
primarily due to its mathematical robustness when dealing with noisy data. The resulting
system of normal equations after application of least-squares minimization (e.g., Menke,
1984) is:
ρ = (S̄ T S̄)−1 S̄ T ~e = Ḡ−g~e,

δ~ (1.32)
where S̄ T is the transpose of the Jacobian matrix and Ḡ−g = (S̄ T S̄)−1 S̄ T is denoted as
the generalized inverse (Menke, 1984). This form of inversion is known as unconstrained
least-square fitting or the Gauss-Newton Method. The inversion process is stopped once a
predefined error-level is reached. Unfortunately, this scheme is impractical for DC resistivity
data: poor starting models may result in spurious convergence or even divergence. Since
the number of model parameters may exceed the number of observed data values, the
problem is said to be under-determined, such that (S̄ T S̄)−1 does not exist. One solution
to this problem involves introducing a damping factor that controls the absolute values of
the parameter changes δ~
ρ. This method, called the ridge-regression or Marquardt-Levenberg
method, results in the following formula:
28 1. Introduction
δ~ ¯ −1 S̄ T ~e,
ρ = (S̄ T S̄ + β I) (1.33)
where β is the damping factor. Inman (1975) first applied this method to his previously
published 1D inversion scheme, which was based on generalized inverse theory (Inman et al.,
1973). Lehmann (1995) implemented an inversion principle based on ridge-regression for 2D
structures.
Another important approach for stabilizing least-squares schemes is Occam’s inversion,
in which only smooth solutions are accepted. Model parameters are only allowed to vary
slowly with position. This is achieved by adding a weighting matrix (or smoothness matrix)
W̄ to the objective function:
ρ = (S̄ T S̄ + W̄ )−1 S̄ T ~e.

δ~ (1.34)
Constable et al. (1987) applied this principle to EM sounding data and deGroot Hedlin
and Constable (1990) extended it to 2D magnetotelluric data. LaBrecque et al. (1996)
investigated the technique’s limitations in applications to resistivity tomography. Combining
damping and smoothing constraints leads to a more general approach to tackling the under-
determined problem.
Tarantola (1987) suggested the inversion procedure could be constrained by the inverse
−1
model covariance matrix C̄m :
−1 −1 T
ρ = (S̄ T S̄ + C̄m
δ~ ) S̄ ~e. (1.35)
Park and Van (1991) presented an inversion scheme based on equation 1.35 for 3D resis-
tivity structures, whereas Ellis and Oldenburg (1994) solved the problem by implementing a
conjugate-gradient minimization scheme to avoid the computation of sensitivities. An im-
proved form of the smoothness- and damping-constrained inversion that enhances the spatial
resolution of resistivity images has been suggested by Lesur et al. (1999b). It assumes that
resistivity contrasts are known a priori.
As an alternative to solving the normal equations, Maurer et al. (1998) generalized the
least-squares solution to include stochastic constraints. Hence, row action matrix solvers can
be applied instead of full matrix inversions.
1. Introduction 29
The calculation of the forward problem is usually the most time consuming step in the
inversion procedure. Consequently, most effort has been expended in deriving more efficient
forward algorithms. In addition, to reduce computing time research has focused on finding
approximate inversion procedures. Li and Oldenburg (1994), introduced an algorithm that
utilizes an approximate inverse mapping formalism. Loke and Barker (1996b,a) described
a fast inversion algorithm based on a low contrast approximation. Their scheme requires
sensitivities to be calculated only for the first iteration. For subsequent iterations, a quasi-
Newton method is used to estimate the partial derivatives. A slightly different low-contrast
approximation was introduced by Beard et al. (1996). They achieved a fast forward solution
by computing only the diagonal elements of the volume IE solution of the 3D resistivity prob-
lem. More recently, Møller et al. (2001) developed a multichannel deconvolution inversion
scheme that can be run on field computers.
Solution appraisal of the inverse problem
When dealing with inexact field data, it is essential to determine how well the inverse solution
represents the true subsurface. The goodness of fit is one means to assess the quality of a
solution. Assuming n independent observations with known uncertainties (i.e., experimental
errors) s, the goodness of fit is usually expressed by the root-mean-square error (RMS):
n~ obs − U
~ calc ]2
1 X [U
RM S = . (1.36)
n s2
The smaller the RMS-value, the better the fit between the observed and calculated data.
Generally, the RMS-error should be in the same range as the observational errors.
A more meaningful quality parameter is the model resolution. The inversion process can
be described using a similar mathematical notation to that used for the forward problem
~ obs ], where noise-free data are assumed. Supposing that there
(equation 1.25) ρ~ est = F −1 [U
~ obs = F[~
is a true but unknown set of model parameters ρ~ true that solves U ρ true ], a model
estimate can be derived as:
~ obs ] = F −1 [F(~
ρ~est = F −1 [U ρ true )] = R̄[~
ρ true ], (1.37)
where R̄ is the model resolution matrix. If R̄ is equal to the identity matrix, then the model
parameters are uniquely determined (Menke, 1984). Information contained in rows of the
30 1. Introduction
model resolution matrix indicate how well the true model parameters can be resolved. For
example, narrow peaks near the main diagonal of the matrix indicate that the model is well
resolved.
A similar measure can be computed for the data by asking how well the model estimate
ρ~ est fits the data:
~ pred = F[~
U ~ obs )] = R̄d [U
ρ est ] = F[F −1 (U ~ obs ]. (1.38)
~ pred are the predicted data and R̄d is the data resolution matrix. If R̄d is equal to
Here, U
the identity matrix, then the predicted equals the observed data. However, if R̄d differs from
the identity matrix, the prediction error is nonzero. Information contained in the rows of R̄d
reveals how well neighboring data can be independently resolved. Often, only the diagonal
elements of R̄d are considered, because they indicate how much a datum contributes to the
model estimate (i.e, how important a datum is). This sub-set of R̄d is called an importance
matrix.
1. Introduction 31
1.6 Motivation of the Thesis
In recent years, the availability of affordable digital electronics has resulted in the introduction
of numerous multi-electrode data acquisition systems. Furthermore, a lot of effort has been
put into the development of accurate forward modeling and inversion schemes for both 2D
and 3D data sets. The advent of fast computers together with the development of efficient
algorithms for solving matrices has resulted in inversion techniques capable of handling very
large data sets. Ongoing research into approximative inversion methods has led to accurate
and fast algorithms for computing provisional models in the field.
By comparison, relatively little attention has been given to the way resistivity data are
collected (e.g., the way electrodes are arranged in the field). Conventional four-electrode con-
figurations (e.g., Wenner, Schlumberger and dipole-dipole arrays) continue to be employed
in surveying with modern equipment. The question arises as to whether these configurations
are efficient when using modern multi-electrode equipment and whether the resulting sub-
surface models contain the maximum possible information about the resistivity distribution.
Gathering the maximum resistivity information with a minimum number of measurements is
the ultimate goal of research associated with this thesis.
To tackle these questions, statistical experimental design has been proposed by Maurer
and Boerner (1998). This approach integrates the design phase of the field experiment with
the data acquisition and analysis phases. The idea is to acquire an initial data set using a
standard electrode array and, subsequently, to invert this data set to give an initial resistivity
model. Based on this initial model, suitable (arbitrary) four-point electrode configurations
that may contribute to an increase in model resolution would be identified. Measurements
using these configurations and subsequent inversion of the updated data set would provide
an improved resistivity model. To achieve an optimum result, the iterative cycle of data
acquisition - data inversion - experimental design should eventually be performed at the
field-site in near real-time.
Even though new data acquisition equipment has appeared on the market in recent years,
no system capable of selecting arbitrary electrode arrangements existed before the initiation
of this thesis. Thus, as a first step, a multi-electrode system that offered the flexibility to
implement the new acquisition strategy had to be developed.
32 1. Introduction
1.7 Outline of the Thesis

In addition to this introductory chapter, my thesis comprises four chapters. Chapters 2 and
4 are self-contained manuscripts that describe research efforts undertaken to reach the goals
described above (Section 1.6). The development and construction of a new versatile multi-
electrode resistivity acquisition system (ETH-DCMES) that fulfills the needs of adaptive
experimental design and 3D resistivity imaging were the key objectives of the manuscript
presented in Chapter 2. It has been published in the IEEE Transactions on Geoscience and
Remote Sensing. In Chapter 3, the results of various laboratory tests designed to evaluate
the operational capabilities of the ETH-DCMES system are presented. The introduction
of a fast experimental design algorithm for identifying electrode configurations that provide
optimum subsurface information was the focus of Chapter 4. The new algorithm was applied
successfully to extensive 2D synthetic and field data sets. These results have been submitted
for publication in Geophysics. Finally, some ideas for improving the ETH-DCMES and for
advancing the experimental design scheme are outlined in Chapter 5.
Chapter 2
Optimization of DC Resistivity Data

Acquisition: Real-Time Experimental
Design and a New Multi-Electrode
System
Peter Stummer, Hansruedi Maurer, Heinrich Horstmeyer, and Alan G. Green
IEEE Transactions on Geoscience and Remote Sensing, published
2.1 Abstract
With the recent introduction of multi-electrode acquisition systems and two- and three-
dimensional imaging software, there has been a marked increase in the use of direct-current
(DC) resistivity techniques for addressing a wide variety of environmental, hydrogeological,
and engineering issues. To determine the location and physical characteristics of relatively
small targets distributed over wide areas, deployments of laterally extensive arrays of densely
spaced electrodes for long recording periods may be required. For many academic and
commercial projects, such large expenditures of effort are not plausible. We introduce the
34 2. New Multi-Electrode DC Resistivity System
concept of real-time experimental design, in which relatively limited data sets that contain
the most important subsurface information are recorded in a cost-optimized sense. Real-time
experimental design, which involves a number of “data acquisition - data analysis - update
survey design”cycles, requires highly flexible recording equipment. To meet this need, we
have developed a distributed data acquisition system in which waveforms are digitized and
partially processed at each electrode before being rapidly transmitted to a central computer
for storage, further processing, display, and preliminary interpretation. The individual data
acquisition unit at each electrode has a high input impedance and a large dynamic range that
ensures the recording of high quality digital data. Versatile software controls the functions
of all electrodes (e.g., whether they operate as current or potential electrodes) and the
waveform of the transmitted current.
2.2 Introduction
Direct-current (DC) resistivity methods are extensively used for investigating the shallow
subsurface. They are particularly appropriate for resolving diverse environmental, hydrogeo-
logical, and civil engineering problems. Resistivity contrasts between subsurface targets and
their host sediments or rocks can be large. Compared to other physical properties such as
seismic velocities, densities, and relative permittivities, which may vary from a few percent to
a factor of 10, electrical resistivities of earth materials may extend over several decades (e.g.,
Keller and Frischknecht, 1966). Although this characteristic makes electrical prospecting a
powerful investigative technique, the resultant high degree of non-linearity between the sub-
surface resistivities and observed data needs to be accounted for in modeling and inversion
algorithms.
The most important component of non-linearity in DC resistivity investigations is due to
the interdependence of current flow paths and the distribution of resistivities. This results in
data analysis problems and complications in survey design. For example, an electrode layout
suitable for delineating subhorizontal boundaries may be inadequate for resolving steeply
dipping features (e.g., Broadbent and Habberjam, 1971). One solution would be to deploy a
very large number of closely spaced electrodes and to record data for all possible combinations
of voltage and current electrodes. Since there exist a very large number of combinations for
even a limited number of electrodes (e.g., for only 30 electrodes there exist 657 720 possible
2. New Multi-Electrode DC Resistivity System 35
voltage - current electrode configurations), such an approach for many situations may not
be commercially practical.
Most DC resistivity studies published in the geophysical literature are based on standard
electrode patterns, such as the Wenner, Schlumberger, dipole-dipole, or pole-pole configu-
rations (e.g., Telford et al., 1990). The ongoing popularity of these configurations suggests
that they are reasonably good choices. Nevertheless, it is well known that standard electrode
configurations have limitations in terms of model resolution (e.g., Aspinall and Gaffney, 2001;
Dahlin and Loke, 1998). To complement the information content of standard DC resistivity
data sets in a cost-optimized sense, we introduce in the first part of the paper a novel adap-
tive data acquisition method: real-time experimental design. Implementation of this new
acquisition strategy requires a recording system that has a high degree of flexibility not pro-
vided by commercially available systems. This motivated us to develop a new multi-electrode
system. As described in the second part of the paper, the new system not only offers the
versatility necessary for real-time experimental design, but it also includes a number of other
appealing features, such as parallel data scans and arbitrary waveforms of the transmitted
currents. We describe technical aspects of the new system and provide key information on
its performance.
2.3 Adaptive Data Acquisition - Real-Time

Experimental Design
With the introduction of multi-electrode systems, the acquisition speed of DC resistivity

surveys has increased markedly (e.g., Dahlin, 1996; Dahlin and Loke, 1998; Griffiths and
Turnbull, 1985; Griffiths et al., 1990). These systems offer efficient means to acquire data in
arbitrary four-point configurations. Commonly used configurations, such as Wenner, Schlum-
berger or dipole-dipole, are known to be incomplete in the sense of Xu and Noel (1993).
These authors define a complete data set as a suite of noise-free observations from which any
four-point-configuration data set can be reconstructed via superposition. Pole-pole data, for
example, represent a complete data set. However, Xu and Noel’s (1993) data completeness
does not necessarily guarantee that maximum subsurface information can be extracted from
the data. This may be due to the presence of ambient noise and/or the limited quality
of recording instruments, either of which could preclude the reliable measurement of small
voltages. In particular, pole-pole measurements are known to be susceptible to ambient
noise (e.g., Beard and Tripp, 1995). Instead of considering complete data sets, we intro-
duce the concept of optimized data sets, which provide maximum subsurface information for
minimum acquisition costs. The amount of subsurface information is constrained by (i) the
number of available electrodes, (ii) the ambient noise level in the investigation area and (iii)
the fidelity of the recording system. Acquisition costs are primarily controlled by the number
of measurements to be made.
During the past decade, several suggestions have been made to improve the information
content of DC resistivity data. Based on methods developed for biomedical imaging (Isaac-
son, 1986), Cherkaeva and Tripp (1996) propose maximizing the electrical response due to
a perturbing geological body by finding the optimum intensity distribution of the injecting
current. This problem, which is solved by a singular value decomposition of the impedance
matrix, results in a pattern of current injection points with varying intensity. A more gen-
erally applicable concept for optimizing data acquisition is statistical experimental design, a
relatively new method described by Maurer and Boerner (1998). In this approach, optimum
survey layouts can be determined by using global optimization techniques. An application of
statistical experimental design to DC resistivity surveys is provided by Maurer et al. (2000).
A major shortcoming of these two approaches is the underlying assumption that sufficient
a priori information about the earth exists before a survey is conducted. If there is considerable
uncertainty about subsurface structures and non-linearity is severe, both methods may fail.
Moreover, several authors (Beard and Tripp, 1995; Dahlin and Loke, 1998) have shown that
noise needs to be considered in any practical optimization scheme. Finally, because many
investigations are concerned with delineating isolated structures (e.g., cavities, ore bodies or
fracture zones), only parts of an investigation area may need to be resolved accurately, but
the locations of the critical parts may be imperfectly known or completely unknown.
A possible solution that accounts for these problems involves integrating data acquisition,
analysis, and interpretation into a real-time iterative process (Figure 2.1). The maximum area
planned for surveying and the number of available electrodes are the only input parameters
required to start this process. Initially, the investigation area would be crossed by a restricted
number of profiles or covered with a sparse regular grid of electrodes, and a limited number
of measurements would be made. Pole-pole recording would be the obvious choice for the
first measurements, because this type of data provides good subsurface coverage in low noise
environments (Loke and Barker, 1996a).
A fast approximate inversion algorithm, such as the quasi-Newton method of Loke and
Barker (1996b) or the multichannel deconvolution approach of Møller et al. (2001), operating
on a fast field computer would provide an initial estimate of the resistivity distribution at
the survey site. At this stage, the field geophysicist would identify areas characterized
by unusually high noise conditions and regions with interesting anomalies. To enhance
information in the poorly resolved areas and to characterize better important anomalies,
additional data would be acquired using various four-point configurations. In many regions it
may be sufficient to record denser data sets using one of the standard configurations, whereas
in others it may be necessary to design a suite of recording arrays that address certain key
issues (e.g., improved spatial resolution or greater depth penetration at specified locations).
Maurer et al. (2000) describe approaches for determining optimized add-on data sets. The
fast approximate inversion algorithm would provide an updated resistivity distribution model
based on the combined data.
Indeed, a series of “data acquisition - data analysis - update survey design”cycles may
be required to reach optimized solutions. Root-mean square (RMS) differences between
the current and the updated model could be used as a basis for deciding if more data
should be recorded. If the RMS differences exceed a predetermined threshold, acquisition
of additional data may be necessary to constrain the model further. Besides model RMS
differences, it is equally important to analyze measures of the model goodness, for example
the model resolution matrix or the model covariance matrix (e.g., Menke, 1984). If these
quantities do not differ significantly for consecutive models, further data may not provide
additional subsurface information. Once convergence to an acceptable preliminary model
is achieved, the end stage, which could be accomplished in the laboratory, would involve a
more sophisticated full 2D or 3D inversion of the final data set.
The above described scenarios are based on the assumption that significant interven-
tion from the field geophysicist is required. We are presently in the process of developing
approaches that minimize such intervention by taking advantage of the results of previous
surveys.
Our final combined data set would be optimized in the sense that its information content is
limited only by the physics of the DC resistivity method and the total number of electrodes
employed. Furthermore, the data acquisition process would be cost-optimized, since the
iterative cycle shown in Figure 2.1 should avoid unnecessary measurements.
Of course, this new approach does not address the well-known problem of equivalence
(ambiguity) in interpretations of DC resistivity data. We emphasize that wherever a-priori
information is available, it should be used to constrain the inversion process. In particular,
well-log data may be used to establish geological boundaries or electrical properties at a
limited number of locations in the model space.
2.4 Multi-Electrode Recording Systems
Implementation of the real-time experimental design concept requires a highly flexible multi-
electrode data acquisition system that includes (i) an extendable bank of electrodes, (ii)
simultaneous fast scans of many different electrode configurations, (iii) software-controlled
selection of electrode configurations, and (iv) sufficient computing power to perform rudi-
mentary data processing, data interpretation, and statistical experimental design. Besides
the specific requirements of the real-time experimental design scheme, a modern DC re-
sistivity recording system should provide high acquisition speed, high overall accuracy, a
programmable transmitter waveform, data acquisition unit(s) with large dynamic range and
high input impedance (to minimize interference of the measuring circuit with the earth).
Conceptually, there exist two end-member design philosophies for such a multi-electrode
system: centralized and distributed systems (Figure 2.2 and 2.3).
2.4.1 Centralized systems
The simplest centralized multi-electrode systems are realized by adding arrays of electrodes
with addressable switches to existing standard single-channel resistivity instruments and mak-
ing conventional 4-point measurements in a serial fashion (e.g., Griffiths et al., 1990). This
option is not efficient for extensive 2D or 3D surveys that require thousands of measurements.
A more sophisticated centralized multi-electrode system is shown in Figure 2.2. A single
casing encloses most or all of the system electronics, which may be controlled by an external
computer. Analogue signals from groups of electrodes are transmitted through multi-core
cables to the central unit. A built-in switch matrix allows arbitrary pairs of electrodes to
inject currents into the ground and arbitrary potential electrode configurations. To provide
parallel measurement capabilities, the potential signals are passed through an analogue mul-
tiplexer (MUX) before being fed into a single analog-to-digital converter (ADC). Finally, the
signals are processed and displayed on a built-in screen or on the optional external computer
(Figure 2.2). As an alternative to an analogue multiplexer, a number of parallel ADC’s could
be used to increase acquisition speed.
In general, simultaneous measurements can be made in two different modes: (i) one of
the potential electrodes is selected as a reference and the remaining electrodes (except for
the injecting electrodes) are used to determine the potential differences with respect to it;
(ii) a number of pre-selected electrode pairs are used to determine voltage signals.
Advantages of the centralized system shown in Figure 2.2 are:
• A single box includes most or all of the necessary electronics.
• Only a single instrument has to be handled and operated in the field.
• Purchase costs for an entire system are, in most cases, lower than for distributed
systems.
• By having relatively few electrical connections between system components, the pos-
sibility of leaking currents is reduced and overall reliability is increased.
Disadvantages of such a system are:
• Source signals and measured voltages are usually transferred via heavy multi-core ca-
bles. The long lengths of some cables make the system susceptible to ambient elec-
tromagnetic noise and cross-talk.
• For a large number of electrodes, the switch matrix is a relatively bulky unit that may
introduce cross-talk between the injected and measured signals (due to their proximity
to each other).
• With presently affordable ADC technology, it is possible to multiplex only a few tens
of channels with the required sample rate and resolution, thus limiting the flexibility
of the system when many electrodes are required.
• Time skews introduced by the MUX switching between channels precludes the deter-
mination of spectral-induced polarization parameters, such as phase shifts between the
current and potential signals.
• The dependability of the system depends on the reliability of the central unit compo-
nents. If one of the critical components fails, the survey has to stop.
2.4.2 Distributed systems
For distributed systems, each electrode is equipped with a data acquisition unit (DAU; Fig-
ure 2.3). The DAU’s enable data to be digitized directly at the individual electrodes. Data
transfer from the daisy-chained DAU’s is routed via a bi-directional data bus (e.g., RS-485)
or LAN (e.g., TCP/IP) and an interface board to a controlling field computer. By employing
a central source-signal generator, the DAU’s can be made compact and lightweight. Positive
aspects of this design are:
• Parallel scan capabilities provide high acquisition speeds.
• Digital data transfer minimizes noise pick-up on long cables.
• The DAU-concept introduces redundancy, such that if one DAU fails, the data from
only a single electrode is missing.
• The flexibility is high in terms of possible array configurations.

• The electrode array can be arbitrarily expanded.
Separate data acquisition systems also have disadvantages:
• The distance between electrodes is limited by the maximum length of the digital data
bus or LAN.
• Acquisition performance is restricted by the data transfer rate between the DAU’s and
the field computer.
• For the same number of electrodes, the purchase cost of the distributed system of
Figure 2.3 is higher than for most commercial centralized systems.
• Operational control of a distributed system requires greater programming effort than

a centralized one.
• To avoid leakage of currents, multi-pole connectors at each DAU have to be of high

quality.
2.4.3 Generation of the source currents
Most commercially available systems include current sources for which predefined waveforms
(square or low-frequency sine waves) can be generated. Theoretically, other waveforms could
also be employed for DC resistivity measurements. Choice of the most appropriate waveform
depends on the electrical subsurface properties and the characteristics of the ambient elec-
tromagnetic noise. Earth materials characterized by low chargeabilities may be investigated
using sinusoidal source signals, because the asymptotically stable value of the response is
reached within a very short time. In contrast, if induced polarization (IP) effects are to be
expected, a low-frequency square wave would better account for the decay of voltage accumu-
lation. Injection of a swept signal could account for IP effects as well as providing additional
information on the phase relationships between current and measured voltage responses. It
is therefore desirable to be able to generate waveforms to meet specific needs. This can
be achieved by digital-to-analogue conversion and appropriate amplification of waveforms
designed and generated on the field computer.
2.4.4 Commercially available systems
Table 2.1 provides an overview of the principal technical features of commercial multi-
electrode systems as of January, 2002. We distinguish between several types of centralized
system: those comprising only a single-channel (C), multi-channel units realized with either
a MUX or several ADC’s (CM), those with passive switches at the electrodes (CS), and
multi-channel instruments with switches at the electrodes (CSM). Distributed systems are
marked (D) in Table 2.1.
In principle, real-time experimental design can be implemented with any of the com-
mercial multi-channel systems. Because of potential cross-talk problems with the multi-core
cables of the C and CM systems and the limited choice of electrode configurations of the
CS and CSM systems, we judge the type D systems to be most appropriate. Only one com-
mercial type D system is listed in Table 2.1. This is the SIP-256 system from the “Deutsche
Arbeitsgemeinschaft für SIP-Anwendungen”. Unfortunately, it is not possible to select sep-
arately the operational mode of individual electrodes; only a choice from a set of predefined
arrays (e.g., Wenner, dipole-dipole, pole-dipole) can be made, such that general layouts can-
not be defined. Consequently, we decided to design and build a new acquisition system that
satisfies our scientific and technical requirements.
2.5 A Novel Digital Multi-Electrode Recording System
We have adopted the concept of a distributed system (Figure 2.3). An internal switching
matrix allows each DAU (Figure 2.4) to be configured either as a current source, current sink,
voltage reference, or measuring device (Figure 2.5). Signals from the electrodes are fed into
24-bit ADC’s that have very high input impedances (≈ 1 GΩ). A microcomputer controls
data flow from and to the DAU’s internal random access memory (RAM), while a RS-485
controller enables communication with the PC. The serial RS-485 data bus is well-established
in industrial applications due to its robustness, it’s low implementation costs, and the fact
that only two lightweight wires are needed for data transfer.
The maximum number of DAU’s is only limited by the number of RS-485 channels. Each
channel can handle 31 DAU’s, and our current field computer can accommodate up to 5
four-channel RS-485 interface cards. With such a setup it is possible to handle 620 electrodes
(Table 2.1). The number of channels and electrodes can be increased significantly beyond
this limit by employing modern Universal Serial Bus (USB) technology.
A drawback of all digital data buses is their limited overall length, which is required
to maintain the voltage drops within predefined limits (e.g., for the RS-485 the maximum
recommended length is 1200 m). Even with this restriction, our current system, which is
designed primarily for shallow environmental, hydrogeological and civil engineering applica-
tions, is capable of simulating a wide variety of electrode configurations. Profile lengths can
be up to 2400 m (1200 m on either side of the control unit), and areal arrays can be up to
800 x 800 m2 . The maximum depths of investigation will depend on the employed electrode
configuration (e.g., Roy and Apparao, 1971; Edwards, 1977; Barker, 1989), the maximum
distances between electrodes, and the subsurface resistivities (e.g., Oldenburg and Li, 1999).
As for most modern multi-electrode DC resistivity systems, polarization in the ground

around the injecting electrodes can significantly distort nearby voltage measurements. This
charge-up effect can be orders-of-magnitude larger than variations in signal due to resistivity
contrasts (Dahlin, 2000), perhaps even resulting in saturation of the ADC. Problems due
to induced polarization can be avoided by carefully designing and implementing appropriate
measurement sequences, such that electrodes very close to and including the current elec-
trodes are not used for voltage measurements within a certain time after current injection.
The recovery time can vary from a few milliseconds for highly resistive ground to several
minutes for conductive clayey soils.
A unique feature of the DAU design is the internal rechargeable battery, which is con-
nected to an automatic charging circuit. After the controlling PC has initiated a measure-
ment, all DAU’s disconnect from the digital data bus and operate independently until the
requested amount of data has been acquired. No data transfer takes place during a measur-
ing sequence, so that cross-talk between wires carrying the source signal and the data bus is
avoided.
Together, the data and source-signal buses (Figure 2.4) consist of four wires within a
single cable. A separate parallel coaxial cable feeds the reference potential to each DAU.
The lightweight nature of both cables is important for efficient handling of the system in the
field. One DAU always operates as a reference unit; the associated ADC is inactive and the
electrode is connected to the reference line (coaxial cable) on the field data bus (Figure 2.5).
All active DAU’s in the array measure with respect to this reference electrode (i.e., reference
potential). Note, that an arbitrary number of DAU’s can be active (Figure 2.5), allowing
parallel scans to be performed.

After data acquisition is completed, the DAU’s reconnect to the RS-485 bus to enable
data transfer, and the power amplifier disconnects from the source-signal bus (Figure 2.4).
Concurrently, a 24 V DC source connects to the source bus to recharge the DAU’s inter-
nal batteries.Current waveforms designed on the PC are fed through a digital-to-analogue
converter board to a class A power amplifier. Before the resulting waveform is sent to the
injecting electrodes, it is routed through a dedicated DAU that precisely determines the input
current.
Figure 2.6 shows a photograph of the entire system. To protect the electronics from poor
weather conditions, the central parts of the system are stored in a box. An uninterruptable
power supply (UPS) enhances operational security of the system.
2.5.1 Resolution
As previously mentioned, the ADC used in the DAU’s offers a theoretical “resolution”of 24
bits. A programmable gain amplifier (PGA) built into the ADC allows the selection of eight
different amplification settings (1, 2, 4, 8, 16, 32, 64, 128), such that for a gain of “1”the
maximum input voltage range of a DAU is ± 2.5 V and for a gain of “128”the maximum
input voltage range is ± 19.5 mV. Because the analogue front-ends of the DAU’s do not
include any additional pre-amplification before analogue-to-digital conversion, the noise level
of the DAU’s is controlled by the noise level of the ADC’s. For typical sampling frequencies
between 50 and 100 Hz, these levels range from 4 to 0.8 µV, corresponding to an effective
“resolution”of 20 to 15.5 bits.
2.5.2 Acquisition speed
Principal goals in designing our new multi-electrode system included high acquisition speeds
and maximum control of signal quality. In achieving these aims, some trade-offs need to be
considered. The ability to perform parallel scans facilitates fast data acquisition, whereas
the transfer of digital full-waveform signals to the computer provides the user with complete
control over the data processing. For example, full-waveform signals provide critical infor-
mation on induced-polarization effects. However, the transfer of full waveforms significantly
limits acquisition speed, because a large amount of data needs to be transferred via the serial
RS-485 connection. Since each DAU is equipped with a microcomputer and random access
memory, data processing could be performed directly at the electrode, and only selected re-
sults transmitted to the computer. This option, which will further improve data acquisition
speed, is currently under development.
2.5.3 Data example
Initial tests have been performed across a buried waste disposal site near Stetten, which lies
about 25 km west of Zürich, Switzerland (Lanz et al., 1996, 1998, 1999; Green et al., 1999).
Figure 2.7 shows a typical recorded time series (small upper panel) and it’s corresponding
frequency spectrum (injecting dipole spacing: 40 m; recording dipole spacing: 60 m). As
expected, there is a narrow peak of about 0.2 V at the source signal frequency of 1 Hz. A
moderately strong signal of 1 mV at 13 - 14 Hz is likely caused by a sorting machine situated
within a nearby gravel pit and a peak of about 2 mV at 16.67 Hz is generated by the Swiss
railway system.
The data shown in Figure 2.7 were recorded with a gain of 4 at a sampling frequency of
62 Hz. For these settings, the resolution limit is 1.3 µV (dashed line in Figure 2.7), which
corresponds to an effective “resolution”of 20 bits. By contrast, the ambient noise level is of
the order of 65 µV (dotted line in Figure 2.7), which is 50 times higher than the instrumental
noise. This implies that the system does not impose any resolution limitations, even when
the ambient noise is very low.
Digital signal processing techniques can be used to determine apparent resistivities from
the stored full-waveform current and voltage responses. Our extensive tests have shown that
the best results are obtained by dividing the sum of voltage values in a narrow band centered
about the dominant source frequency by the sum of current values in the same frequency
band.
2.6 Discussion and Conclusions
We have introduced real-time experimental design, a novel concept for acquiring DC resistiv-
ity data. Technical realization of this concept required a versatile data acquisition scheme.
Comparisons of the advantages and disadvantages of different acquisition schemes revealed
that a distributed system is the most appropriate solution. To meet the requirements of
high flexibility and fast acquisition of high quality data, we have built a new multi-electrode
system.
Our current research efforts involve the evaluation of suitable electrode configurations.
The simultaneous measurement of a large number of potential differences with a single
current dipole offers new possibilities that need to be explored. With an n electrode array,
n − 3 waveforms can be recorded concurrently. However, the time required for the data
transfer and processing of these waveforms can be large. As a consequence, it might be
appropriate to record a subset of the n−3 waveforms that maximizes the returned subsurface
information while minimizing the data acquisition time. This is a global optimization problem
that can be solved using statistical experimental design techniques.
2.7 Acknowledgments
We gratefully acknowledge financial support from ETH (project no. 0-20394-97).
Figure 2.1: Basic steps in real-time experimental design showing the iterative nature of the process.
Figure 2.2: Schematic layout of a centralized data acquisition system for DC resistivity surveying.
Figure 2.3: Schematic layout of a distributed data acquisition system for DC resistivity surveying.
Figure 2.4: Block diagram of one Data Acquisition Unit (DAU) and the field data buses.
Figure 2.5: Switching scheme for an arbitrary four-pole arrangement showing the three different
operational modes of a data acquisition unit (DAU).
Figure 2.6: A typical field deployment of the entire data acquisition system. UPS - uninterruptable
power supply.
Figure 2.7: Data recorded by a single data acquisition unit (DAU): Acquired time series (small
upper display) and corresponding amplitude spectrum.
Table 2.1: Comparison of commercially available multi-electrode DC resistivity systems with the new ETH-DCMES system. Channels - number of
2. New Multi-Electrode DC Resistivity System
independent data recording units realized using either a multiplexer or several analogue-to-digital converters. Electrodes - maximum number of electrodes
operable at a time. Type - refers to the definition of the different system designs as follows: C - single-channel centralized system, CM - multi-channel
centralized system, CS - system with switches at the electrodes, CSM - multi-channel system with switches at the electrodes, and D - distributed system.
Source - signal source built-in or external unit.
No. Name Supplier Channels Electrodes Type Source
1. E-Scan 3D Premier Geophysics Inc., Canada 1 >100 CS built-in
2. Tomoplex Campus Ltd., United Kingdom 64 64 CM built-in
Geopulse-Tigre 1 256 C built-in
3. McOhm-El Oyo Geoelectric, Japan 1 750 C built-in
4. Saris Scintrex Ltd., Canada 1 65 000 CS built-in
5. Super Sting R8 / Swift AGI Inc., USA 8 254 CSM built-in
Sting R1 / Swift 1 254 CS built-in
6. SAS 4000 Lund ABEM Instrument AB, Sweden 4 75 CM built-in
SAS 1000 Lund 1 75 C built-in
7. RESECS / RESITOMO DMT GmbH, Germany 6 960 CSM built-in
8. GDP-32 II Zonge Inc., USA 16 32 CM external
9. SYSCAL Switch-48 IRIS Instruments, France 1 48 C built-in
SYSCAL Switch-24 1 24 C built-in
E-NODE / Syscal Rx 1 255 CS built-in
10. GeoTom Res/IP Geolog2000, Germany 1 250 C built-in
11. GMS 125A GeoSys, Germany 1 125 C built-in
GMS 150 1 150 C built-in
12. GRM-30.2 KBFI-TRIAS Ltd., Hungary 1 30 C built-in
RESP-12 1 60 C built-in
13. PA-CEP Aarhus University, Denmark 6 6 CM built-in
14. SIP-256 ARGE SIP-Anwendungen, Germany 256 256 D built-in
ETH-DCMES ETH Zurich, Switzerland 620a 620a D external
a
The current number of channels and electrodes for the new ETH-DCMES system is only limited by the capacity of our present field computer: 4 isolated channels
52
per RS-485 interface card and 5 PCI slots. With the introduction of the Universal Serial Bus (USB), the number of channels and electrodes can be greatly increased.
Chapter 3
Laboratory Testing the New

ETH-DCMES System
Ambient EM noise can seriously distort DC resistivity data, especially in urban areas. Here,
I provide a brief overview of a simple technique for estimating the influence of noise on
measured potentials (resistances) and determining operational characteristics of any multi-
electrode system.
For most commercial DC resistivity systems, it is not clear how raw data are processed to
yield final resistivity values. Consequently, it is not possible to estimate the influence of noise
on the recorded values. There are a number of approaches for addressing this issue. For
example, measurements using standard electrode configurations (e.g., dipole-dipole) along a
profile across a calibration site is one possibility. However, changing subsurface and contact
resistivities with varying weather conditions may preclude meaningful comparisons. As an
alternative, I have constructed a test tool that allows the following tests to be performed on
any multi-electrode system under controlled laboratory conditions (Brunner, 2001):
• operational check
• noise suppression test
• comparison of systems.
54 3. Laboratory Testing
3.1 Test Tool
1 2 3 19
RC1 RC2 RC3 RC19
R1 R2 R19
Noise Signal
R25 R24 R20
RC25 RC24 RC20
25 24 20
Figure 3.1: Layout of the resistor network for testing multi-electrode DC resistivity systems with
up to 25 electrodes (after Brunner, 2001).
The tool suggested by Brunner (2001) is basically a lumped resistor network. Figure 3.1
shows the layout of the network, in which resistors marked Rci (with i = 1 - 25) simulate
varying contact resistances ranging from 0 Ω to 12 kΩ. The actual measuring circuit consists
of resistors Ri (with i = 1 - 25) ranging from 12 Ω to 1200 Ω. Details on the values and
the nominal powers of all resistors can be found in Table 3.1. A major advantage of this
approach versus a series of field-tests is that the resistance distribution is exactly known,
such that the expected values can be calculated and compared to those produced by the
instrument under investigation.
The network is designed for a maximum input current of 200 mA. To test the noise
suppression capabilities of multi-electrode instruments, an additional signal can be fed into
the network via the input terminals marked “Noise Signal”(see Figure 3.1). Any signal
generator connected to the noise signal terminal must have an internal impedance of at least
100 kΩ, otherwise the network will be short-circuited by the noise source.
3. Laboratory Testing 55
no. Ri [Ω] P (Ri )[W ] RCi [Ω] P (RCi )[W ]

1 1k 1 12k 5
2 680 1 22 1
3 470 1 6k8 5
4 330 1 470 4
5 220 1 2k2 11
6 150 4 330 4
7 100 1 1k 11
8 68 4 680 11
9 47 4 2·340 22
10 33 4 2·340 22
11 22 1 0 -
12 15 1 0 -
13 12 1 3.3 1
14 18 1 5.6 1
15 27 4 22 1
16 39 4 56 4
17 56 4 150 11
18 82 4 220 11
19 120 4 2·340 22
20 180 1 1k 11
21 270 1 1k8 5
22 390 1 1k 4
23 560 1 5k6 5
24 820 1 1k2 1
25 1k2 1 2k7 5
Table 3.1: Dimensions of the resistors used in the network for test measurements with multi-
electrode DC resistivity systems (after Brunner, 2001).
This tool is designed for testing systems with a dipole-dipole configuration. For such a
configuration, resistances may vary over four decades, with values being very small for the
longest simulated offsets, thus subjecting the tested systems to extreme conditions. Assume
that current I is applied to the circuit through terminals 1 and 2 (Figure 3.1). Current I will
then be divided into partial current I1 through R1 and partial current I2 through the sum of
resistors R2 to R25 , such that:
I = I1 − I2 . (3.1)
The minus sign accounts for the different flow directions from the node. Kichhoff’s second
law (the sum of all voltages in a closed loop is equal to zero) gives the voltage drop across
R1 as:
25
X
UR1 = I1 R1 = −I2 Ri − R1 . (3.2)
i=1
If, for instance, the voltage across terminals 24 and 25 is to be evaluated, partial current I2
has to be derived from the known values I and Ri . Combining equations 3.1 and 3.2 leads
to:
R1
I2 = −I 25
. (3.3)
P
Ri
i=1
The desired voltage drop across resistor R24 is then:
R1 R24
UR24 = −I2 R24 = I 25
. (3.4)
P
Ri
i=1
Equation 3.4 can be re-written in a general form for voltage drops across terminals m and
n for a current applied through terminals a and b as:
Rmn Rab
Umn = I 25
, (3.5)
P
Ri
i=1
where Rmn and Rab are the resistors across terminals m and n and a and b, respectively.
Most commercial resistivity instruments do not provide separate voltage and current values.
Instead, they only show resistance values. Therefore, equation 3.5 simply has to be divided
by the applied current I to yield the resistance at the measuring terminals:
calc Rmn Rab

Rmn = 25
. (3.6)
P
Ri
i=1
In case the array geometry is taken into account by the system the resulting resistivity values
have to be corrected by division with the appropriate geometric factor.
3.2 Test Results
To evaluate the operational capabilities of the ETH-DCMES system and to determine its
accuracy under varying measurement conditions, a series of tests employing the resistor
network have been carried out. A sinusoidal source signal of 1 Hz frequency has been
employed for all tests. For comparison, all tests were repeated with an ABEM commercial
resistivity system, which is using a square-wave as source signal. Furthermore, different data
processing strategies employed in the two systems were compared.
Controlling PC
Source Signal Generator
Data Bus
DAU DAU DAU DAU DAU

1 2 3 24 25
1 2 3 24 25
Resistor Network
Noise Signal
Noise Signal Generator
Figure 3.2: Typical laboratory test setup: ETH-DCMES system hooked-up to the resistor network.
All tests were conducted with 25 electrodes deployed in a dipole-dipole configuration.

Figure 3.2 shows a sketch of the ETH-DCMES system hooked-up to the test tool. The dipole
separation factor n was successively increased from 1 to 9, resulting in a total of 162 data
points for each pseudosection. Figure 3.3a shows the reference pseudo-section calculated
with equation 3.6. All test measurements were compared to this pseudosection. ETH-
DCMES measurements without superimposed noise signals agree well with the calculated
data, as shown by the percent difference pseudosection in Figure 3.3b. Additional details on
the deviations from the calculated data are revealed by plotting the differences as a histogram
(Figure 3.3c). Differences range between about ± 5 %, which is the standard tolerance of
the resistors (delimited by dashed red lines in all histogram plots) used to build the test
network. Better agreement would only be achieved by using precision resistors with lower
tolerance levels.
Resistance [W]
(a)
Difference [%]
Dipole Separation Factor n
(b) (c)
Difference [%]
(d) (e)
Electrode No. Deviation [%]
Figure 3.3: Test measurements on a resistor network using 25 electrodes: (a) calculated dipole-
dipole pseudosection; (b) difference between calculated and ETH-DCMES measured pseudosection
in percent; (c) histogram representation of (b); (d) difference between calculated and ABEM
measured pseudosection; (e) histogram representation of (d). Dashed red lines represent the
tolerance limits (± 5 %) of the resistors employed in the test network.
To characterize the result by a single number, a data RMS misfit erms in percent is
calculated as follows:
v
u N calc
1uX Rj − Rjobs 2
erms = t · 100, (3.7)
N j=1 Rjcalc
where N is the total number of data points (N = 162 in this study) and Rcalc and Robs are
the calculated and observed resistances, respectively. A RMS misfit of 0.15 % was obtained
for the data measured with the ETH-DCMES system (Figures 3.3b and 3.3c).
The experiment was repeated with the ABEM system (Figures 3.3d and 3.3e). Inter-
estingly, the percent-difference plot (Figure 3.3d) shows a symmetric distribution of the
differences, being highest in areas where low resistances are expected (Figure 3.3a). This
effect was likely caused by the ABEM instrument’s limited resolution. It produced erroneous
results of up to 14 % (Figure 3.3e). About 29 % of the ABEM data set were outside the 5
% resistor tolerance level. The RMS-misfit was 0.38 %, which was more than twice that of
the ETH-DCMES data set.
Effects of noise
In a second step, different types of noise were fed into the network. The objective was to
compare noise suppression capabilities of the two systems.
Before interpreting the results, it is useful to review the processing sequences used by most
commercial resistivity systems. Commonly, square-waves of a few Hz are employed as source
signals, primarily because they are simple to generate and suitable processing sequences are
well established. The analogue signal is usually pre-amplified and passed through notch filters
to suppress typical urban noise (e.g., power line (50 - 60 Hz) and railway interference (15 - 20
Hz)) before digitization. To reduce DC-offsets, the mean value is then subtracted from the
digitized signal. Subsequent stacking of signals is a simple effective technique for reducing
statistical noise and increasing the SNR. At this stage, the processed signal (voltage) is
divided by the injected current value to yield the resistance.
In contrast to most commercial resistivity systems, the ETH-DCMES system stores the
full waveform of the recovered signals and subsequently determines the resistance in the
frequency domain (see Chapter 2 for details). Consequently, analogue filters at the front end
of the DAU’s, which would decrease the input impedance, are not employed, because noise
can be discriminated from the known input signal in the corresponding amplitude spectrum.
Two different types of noise were used to test the suppression capabilities of the ETH-
DCMES and ABEM system: (i) a 16.67 Hz wavetrain, which simulates interfering noise
from the Swiss railway system, and (ii) a wavetrain containing frequencies between 5 and
500 Hz. For both types of noise, the amplitude was gradually increased until the ETH-
DCMES produced a significant number of values outside the 5 % deviation defined by the
tolerance limits of the resistors employed in the test network. The measurements were then
repeated with the ABEM system. Comparison of the distribution of these high noise values
provides information on the “reliability”of the data processing sequences employed by each
system.
Figures 3.4a - 3.4d show the results for the 16.67 Hz noise, whereas Figures 3.4e - 3.4h
present the results for the 5 - 500 Hz noise. In both cases, the ETH-DCMES system delivers
resistance readings closer to the true values than the ABEM system. Approximately 17 %
of the ETH-DCMES values for both noise types are outside the 5 % limits (Figures 3.4b
and 3.4f). For these values, the resolving limit of the ADC’s used in the DAU’s is reached;
the amplitude of the noise approaches or reaches the maximum input voltage range of the
DAU’s.
The ABEM system produces pseudosections (Figure 3.4c and 3.4g) noticeably inferior to
those of the ETH-DCMES system (Figure 3.4a and 3.4e); the sharp boundaries in the lower
parts of the difference pseudosections are artifacts associated with the limited resolution
and inadequate input filters of the ABEM system; increasing the input current and thus
increasing the voltage drops across the network’s resistors eliminated the artifacts. The
corresponding histograms reveal that only 37 % of the ABEM observations with the 16.67
Hz noise superimposed and 29 % of the data with the broad-band noise superimposed are
within the 5 % limits (Figure 3.4d and 3.4h). For the mono-frequency and the broad-band
noise, deviations of up to -100 and -300 % from the true resistance were encountered.
Difference [%]
(a) (b)
Difference [%]
Dipole Separation Factor n
(c) (d)
Difference [%]
(e) (f)
Difference [%]
(g) (h)
Electrode No. Deviation [%]
Figure 3.4: Test measurements on a resistor network using 25 electrodes: (a) difference between
calculated and ETH-DCMES measured pseudosection in percent with 16.67 Hz noise superim-
posed; (b) histogram representation of (a); (c) difference between calculated and ABEM measured
pseudosection in percent with 16.67 Hz noise superimposed; (d) histogram representation of (c);
(e) difference between calculated and ETH-DCMES measured pseudosection in percent with broad-
band noise superimposed; (f) histogram representation of (e); (g) difference between calculated
and ABEM measured pseudosection in percent with broad-band noise superimposed; (h) histogram
representation of (g). Dashed red lines represent the tolerance limits (± 5 %) of the resistors
employed in the test network.
Table 3.2 shows the RMS misfits for the four data sets. The ETH-DCMES misfits are
about ten times smaller than the ABEM ones. Although the RMS errors for the ABEM
system are not large, the difference pseudosections are unacceptably poor (Figures 3.4c
and 3.4g).
16.67 Hz noise swept noise

ABEM system 2.21 % 4.62 %
ETH-DCMES 0.29 % 0.33 %
Table 3.2: Comparison of the different processing schemes: RMS misfits of the data sets with
noise superimposed(note that the ABEM system uses a square wave, whereas the ETH-DCMES
system was tested with a sine wave).
One reason for the superior performance of the ETH-DCMES relative to the ABEM
system could be the data processing sequence. On the typical single recording presented in
Figure 3.5 (taken from the broad-band noise data set represented in Figures 3.4e and 3.4f),
the clear separation of noise and primary signal can be seen. Noise peaks have varying
amplitudes due to the transfer function of the amplifier employed to boost the noise signal
in the test. For this example, the calculated resistance (equation 3.6) is 0.044 Ω. Even
though some noise peaks in the amplitude spectrum (Figure 3.5b) are up to 50 times higher
than that of the 1 Hz signal, the 0.043 Ω resistance produced by the ETH-DCMES has
an error of only 2.5 %. The corresponding 0.100 Ω value produced by the ABEM data
set, which was recorded using a 0.4 Hz square-wave input signal, is in error by more than
120 %. This erroneous result may be explained by lack of resolution in combination with
poor performance of the input filters. The misfit may also be partly caused by the ABEM’s
processing sequence, which fails due to the highly distorted square wave.
These tests demonstrate the superiority of the high resolution ADC’s and the processing
sequence employed by the ETH-DCMES system relative to those used in the ABEM system.
0.01
0.005
Amplitude [V]
0
0.005
0.01
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
(a)
Time [s]
2
10
Amplitude [V]
4
10
6
10
8
10
1 0 1 2 3
10 10 10 10 10
(b) Frequency [Hz]
Figure 3.5: Data recorded with a single DAU of the ETH-DCMES attached to the resistor network.
Superimposed on the 1 Hz input signal is “noise”with frequencies ranging from 5 - 500 Hz: (a)
acquired time-series; (b) corresponding amplitude spectrum, in which the peak at 1Hz represents
the primary signal.
Chapter 4
Experimental Design: Electrical

Resistivity Data Sets that Provide
Optimum Subsurface Information
Peter Stummer, Hansruedi Maurer, and Alan G. Green
Geophysics, submitted
4.1 Abstract
Although multi-electrode electrical resistivity systems have been commercially available for
more than a decade, resistivity imaging of the subsurface continues to be based on data sets
recorded using one or more of the standard electrode arrays (e.g., the Wenner or conventional
dipole-dipole array). In an attempt to exploit better the full capabilities of multi-electrode
acquisition systems, we have developed an experimental design procedure to identify suites
of electrode configurations that provide subsurface information according to predefined opti-
mization criteria. The experimental design algorithm includes a goodness function that ranks
the sensitivity of every possible electrode configuration to changes in the subsurface param-
eters. To examine the potential and limitations of the new algorithm, comprehensive data
66 4. Experimental Design
sets that included data from all standard and non-standard electrode configurations were (a)
generated for a complex 2D resistivity model and (b) recorded across a well-studied test site
in Switzerland. Images determined from the resultant comprehensive data sets were used as
benchmarks against which the images derived from the optimized data sets were assessed.
Images from relatively small optimized data sets containing 265 - 282 data points provided
more information than those from standard data sets of equal size. By far the best images,
comparable to those determined from the much larger comprehensive data sets, were ob-
tained from optimized data sets with 1000 - 6000 data points. These images supplied reliable
information over depth ranges that were three times greater than the depth ranges resolved
by the standard images. The first ∼600 electrode configurations selected by the experimental
design procedure were non-standard dipole-dipole-type arrays, whereas the following ∼4800
electrode configurations were an approximately equal mix of non-standard dipole-dipole-type
arrays and nested configurations (i.e., mostly gradient and other non-standard arrays).
4.2 Introduction
Bulk electrical resistivities in the shallow subsurface are largely controlled by electrolytic
conduction in aqueous fluids that are either distributed across grain boundaries or contained
in pores, fractures, and faults (Ward, 1990a). This close connection between resistivity and
the presence of aqueous fluids is one reason why electrical resistivity techniques are employed
in diverse exploration projects and various environmental, hydrogeological, geotechnical, and
civil engineering investigations. Conventional electrical resistivity techniques have the added
benefit of being relatively inexpensive.
Until the late-1980’s, only the simple resistivity sounding and mapping methods existed
(Telford et al., 1990; Ward, 1990b,a). Data were usually acquired with one of the five
standard electrode arrays: Wenner, Schlumberger, dipole-dipole, pole-dipole, or pole-pole.
Soundings supplied one-dimensional (1D) resistivity-depth information on the condition that
lateral variations were insignificant, whereas mapping allowed lateral changes in resistivity
over narrow depth ranges to be delineated provided that vertical changes in resistivity could
be ignored. The assumptions required for meaningful resistivity sounding and mapping pre-
cluded their application at locations where complex subsurface structures were expected.
With the introduction of versatile multi-electrode acquisition systems (e.g., Griffiths and
4. Experimental Design 67
Turnbull, 1985; Griffiths et al., 1990) and efficient two- and three-dimensional (2- and 3D)
inversion schemes (e.g., Loke and Barker, 1996b,a), resistivity imaging (tomography) of rel-
atively complex structures became possible. As a result, electrical resistivity techniques have
experienced a renaissance over the past six to seven years, such that they are now the first
choice for many near-surface investigations.
Based on a wide variety of synthetic tests and comparisons of field data, the principal
characteristics of the standard electrode arrays for resistivity sounding, mapping, and primi-
tive sounding-mapping are now relatively well established (Ward, 1990a). The Wenner and
Schlumberger arrays provide good vertical resolution and generally high signal-to-noise (S/N)
data, but they have only moderate lateral resolution capabilities and are susceptible to elec-
tromagnetic coupling between cables connecting the current and potential electrodes. By
comparison, the standard pole-dipole and dipole-dipole arrays have good lateral resolution
properties and are distinguished by low levels of electromagnetic coupling. For conventional
dipole-dipole surveying (i.e., a single fixed separation of the current electrodes and potential
electrodes), vertical resolution may be quite poor and data quality decreases markedly with
increasing offset, thus limiting depth penetration. The pole-pole array provides the widest
horizontal coverage and deepest penetration, but it usually yields low-resolution information
and low S/N data.
Although resistivity data can be recorded with arbitrary two-, three-, and four-point
electrode configurations, the vast majority of recent model-based studies have been concerned
with determining the efficacy of the five standard electrode arrays for use with multi-electrode
systems and 2- and 3D inversion schemes (Friedel, 1997; Spitzer, 1998; Dahlin and Loke,
1998; Oldenburg and Li, 1999; Olayinka and Yaramanci, 2000; Dahlin and Zhou, 2001; Zhou
et al., 2002). Furthermore, most reported field applications of 2D imaging have been based
on either the Wenner array (Loke and Barker, 1996b; Dahlin, 1996; Barker, 1997; Friedel,
2000; Barker et al., 2002; Embi et al., 2002; Tejero et al., 2002; Guérin et al., 2002) or the
dipole-dipole array (Oldenburg and Li, 1999; Zhou et al., 1999; Roth et al., 1999; Friedel,
2000; Tejero et al., 2002; Slob et al., 2002a). Despite its low resolving power and high
susceptibility to noise, easy implementation has made the pole-pole array the natural choice
for full 3D resistivity surveying (Loke and Barker, 1996b; Dahlin et al., 2002; Guérin et al.,
2002; Slob et al., 2002b). Nevertheless, Chambers et al. (1999) constructed a 3D data
set from co-linear pole-dipole data recorded along perpendicular sets of closely spaced lines,
and Knödel et al. (1997) and Friedel (1997) have described dipole-dipole data sets recorded
with circular electrode arrays.
Assuming that an efficient multi-electrode acquisition system is available, which 2- or 3D
electrode array should be used at any given investigation site ? Numerous authors recommend
that the geometry of the electrode array should be chosen to meet the survey objectives. As
examples, the Wenner array might be an appropriate option for 2D surveying at locations
where good vertical resolution is required and the dipole-dipole array might be a good choice
at sites where good horizontal resolution is needed. In contrast, it is not obvious which
electrode array or combination of arrays should be used for surveying completely unknown
areas or if high resolution is required in all directions. One possibility would be to record
data with a variety of electrode configurations and invert the combined data set.
For n equally spaced co-linear electrodes, there exist a total of nd non-reciprocal four-
point electrode configurations (Noel and Xu, 1991):
nd = n · (n − 1) · (n − 2) · (n − 3)/8. (4.1)
These configurations include every possible standard and non-standard electrode array.
If measurements were to be made with the entire suite of non-reciprocal configurations,
the resulting comprehensivea data set would contain all subsurface information that the n-
electrode system is capable of gathering. Even for a modest electrode spread, however, there
are a very large number of possible configurations. For example, there are more than 690
000 configurations for a fifty-electrode spread. Since many commercial acquisition systems
can now handle >100 electrodes (Stummer et al., 2002) and future 3D investigations will
likely require the deployment of hundreds of electrodes, the acquisition of comprehensive
data sets is economically impractical for production surveying. Consequently, it is necessary
to determine if limited combinations of electrode configurations are able to supply resistivity
images that are comparable in quality to those provided by the comprehensive data sets.
a
Our definition of a comprehensive data set should not be confused with the concept of a complete data
set introduced by Xu and Noel (1993). These authors describe a complete data set as a suite of noise-free
observations from which the response of any four-point configuration can be constructed via superposition.
Pole-pole arrays and several non-standard electrode configurations provide complete data sets (Xu and Noel,
1993; Lehmann, 1995; Friedel, 2000). Unfortunately, data completeness does not ensure optimum imaging
of the subsurface, especially when ambient noise and the inherent limitations of recording equipment are
considered; the general disadvantages of the pole-pole array are well known (Beard and Tripp, 1995), and
the non-standard electrode configuration of Xu and Noel (1993) and Lehmann (1995) is characterized by
only moderate resolution and moderate depth penetration (Friedel, 2000).
In this contribution, we introduce an experimental design procedure that allows us to

estimate which electrode configurations provide the optimum information in a cost-effective
manner, whereby cost is assumed to be proportional to the number of measurements made.
Although we only consider 2D imaging, it is a trivial matter to extend the method to 3D
situations. We begin with the results of a series of model studies, in which we compare re-
sistivity images derived from synthetic comprehensive, Wenner, dipole-dipole, and combined
Wenner - dipole-dipole data sets. The synthetic data sets are derived from the 2D resistivity
model displayed in Figure 4.1a. A measure of the reliability of the different parts of each
derived resistivity image is provided by the formal model resolutions. The resistivity image
derived from the comprehensive data set is found to be markedly superior to the others. It
is used as the basis for our experimental design algorithm.
After describing key elements of the experimental design algorithm, we apply it to syn-
thetic data derived from the 2D model in Figure 4.1a. A relatively small optimized data set
(i.e., 282 values) is shown to yield an image with far more information than that produced
by the combined Wenner - dipole-dipole data set of the same size. Much better images are
obtained from larger optimized data sets (e.g., >1000 values). The problem of complex cur-
rent patterns due to heterogeneous resistivity distributions is then briefly assessed. Finally,
the new experimental design procedure is tested on a field data set recorded across a buried
landfill. Using the image derived from the comprehensive data set as a yardstick, we show
how optimized data sets provide substantially more information than the standard ones.
4.3 Images Determined from Synthetic Comprehensive,

Wenner, Dipole-Dipole, and Combined Wenner -
Dipole-Dipole Data Sets
The 2D resistivity model in Figure 4.1a includes a thin 100 Ωm surface layer A and an
underlying 1000 Ωm layer B. These layers could correspond to a moderately conductive
humus/silt unit overlying a resistive dry gravel sequence. Centered at 18.6 m depth on
the left side of the model is a conductive block C with a minimum resistivity of 10 Ωm.
Transecting the boundary between the two layers on the right side is a 10 000 Ωm resistive
block D. The conductive and resistive blocks could represent a clay-rich sedimentary feature
and a concrete foundation, respectively.
To simulate a typical field survey, thirty electrodes are spread across the top of the model
at regular 5 m intervals. For this electrode spread, an unconstrained comprehensive data
set would comprise data recorded with 82 215 non-reciprocal configurations. We exclude all
configurations with crossed current and potential electrode dipoles (i.e., Wenner-γ configura-
tions) and all configurations with geometrical factors larger than 5500 m. Although modern
multi-electrode acquisition systems are capable of measuring very low potential differences,
experience gained from numerous standard dipole-dipole surveys suggests that the maximum
dipole separation should be no more than about six times the current/potential electrode
spacing (Sharma, 1997; Loke, 1999), which roughly corresponds to a geometrical factor
threshold of 5500 m for our electrode spread. The exclusions reduce our comprehensive data
set to 51 373 data points.
All synthetic data sets were generated with a modeling code based on an algorithm
proposed by Dey and Morrison (1979a), suitably modified to account for singularities (e.g.,
Lowry et al., 1989; Zhao and Yedlin, 1996). We employed this same modified code for the
forward computations required by our inversion routine. The inverse problems were solved
in a least-squares sense with appropriate damping and smoothing constraints applied (e.g.,
Marquardt, 1970; Constable et al., 1987). Sensitivities were computed after each iteration
using the reciprocity method (e.g., Park and Van, 1991). Even though convergence rates
differed for the various synthetic data sets, twenty-five iterations were used for all inversions.
Figures 4.1b - 4.1e show the inversion results for the comprehensive, Wenner, dipole-
dipole, and combined Wenner - dipole-dipole data setsa . The input model for each inversion
was a homogeneous half-space with resistivity set to the average apparent resistivity of the
respective synthetic data set (210 - 360 Ωm; see Figure 4.1 caption for details). Tests
using various input resistivities demonstrated that, except for deeper parts of the dipole-
dipole resistivity image, the inversion results did not depend on the input model. Synthetic
measurements from all 135 possible configurations were included in the Wenner data set,
whereas only 147 of the possible 378 measurements were included in the conventional dipole-
dipole data set (the excluded conventional dipole-dipole data had geometrical factors >5500
m). The combined Wenner - dipole-dipole data set was the simple sum of the Wenner and
dipole-dipole data sets (i.e., 282 values).
To identify the well-resolved parts of the models, we consider the diagonal elements of
the model resolution matrix, a quantitative measure of the independence of the inverted
values (e.g., Menke, 1984). Figure 4.2 shows initial estimates of these formal model resolu-
tions for the data sets used to generate Figures 4.1b - 4.1e. Prior to an experiment, there
exists little or no knowledge about the subsurface. Therefore, the formal model resolutions
in Figure 4.2 are computed on the basis of a homogeneous half-space model. As expected,
the highest model resolutions are provided by the comprehensive data set in Figure 4.2a.
Generally, model resolutions less than about 0.05 are indicative of poorly resolved regions
of an inverted model. Since this value occurs at 30 m depth in Figure 4.2a, we adopt the
depth interval 0 - 30 m as the focus of our investigation. It is within this depth range that
the most interesting features are located in the resistivity model of Figure 4.1a.
The thickness and resistivity of the near-surface layer A and the resistivity of the top ∼10
m of the underlying layer B are well produced in all resistivity images (Figures 4.1b - 4.1e).
Surprisingly, quite good estimates for the resistivity of the second layer (i.e., values >400
Ωm, which compare favorably to the true value of 1000 Ωm) extend to near the base of three
of the four resistivity images (Figures 4.1b, 4.1c, and 4.1e), even though the diagrams in
Figure 4.2 demonstrate that the deeper regions of all images are poorly resolved. Only for the
conventional dipole-dipole data set is the resistivity of the input model preserved at depths
>20 m (Figure 4.1d), highlighting the limited depth penetration characteristics of this data
set (Oldenburg and Li, 1999). For the dipole-dipole data set, resistivities below 20 m depth
a
The image based on Lehmann‘s (1995) non-standard electrode configuration (total number of data
points = 405 is similar to that shown for the combined Wenner - dipole-dipole data set of Figure 4.1e (total
number of data points = 282).
do not change, regardless of the resistivity of the input model. Evidence for the shallow
resistive block D is observed in all images, but its shape and resistivity differ in each image.
By far the best representation of block D is seen in the comprehensive image (Figure 4.1b)
and the poorest is observed in the Wenner image (Figure 4.1c). The conductive body C is
only recognized in the comprehensive image.
Comparable results to those shown in Figures 4.1c - 4.1e have been obtained by others
for similarly complex models (e.g., Loke and Barker, 1996b; Oldenburg and Li, 1999). As a
consequence of the imposed damping and smoothing constraints, it is common for resistivity
contrasts determined from inversions of synthetic data to be anomalously low and boundaries
between regions of significantly different resistivity to appear gradational, even if in reality
they are sharp.
4.4 Identifying Optimal Data Sets

Statistical experimental design can be used to identify data sets that constrain the subsurface
image in a near-optimal fashion. The first attempt to optimize data used for electrical re-
sistivity imaging was probably made in the biomedical sciences (Isaacson, 1986). It involved
adjusting the intensity distribution of injected currents to maximize the response of a given
target. Cherkaeva and Tripp (1996) adapted this approach for application to geophysical
surveying. An alternative method of identifying optimal data sets was described by Maurer
and Boerner (1998). They determined optimum survey layouts for electromagnetic exper-
iments using global optimization techniques. Maurer et al. (2000) demonstrated that the
same techniques could also be used to optimally design simple electrical resistivity surveys.
Unfortunately, the very high computational costs of global optimization algorithms preclude
them from being employed for realistic tomography problems that involve many electrode
configurations and numerous inversion cells. Alternative strategies that provide reliable re-
sults at affordable computational costs are required. It is our goal to design an experiment
that will yield an image that approaches the quality of the comprehensive image, but which
requires far fewer measurements.
4.4.1 A fast approximate experimental design algorithm
We begin by assuming that an initial data set has been recorded using one of the standard
electrode arrays (e.g., the dipole-dipole array). Our experimental design algorithm then
identifies in an objective fashion those electrode configurations that provide the most new
information. The algorithm requires as input the number of available electrodes, the spacing
between the electrodes, and the discretization of the subsurface model. By excluding nexcl
configurations from consideration (e.g., those with unacceptably high geometrical factors),
equation (4.1) becomes:
nd0 = (n · (n − 1) · (n − 2) · (n − 3)/8) − nexcl . (4.2)
A quantitative measure of the information that a particular electrode configuration can

provide is its sensitivity, which represents the change in potential difference due to a small
resistivity perturbation in a particular cell of the subsurface model. Sensitivities are contained
in the Jacobian matrix S̄ (e.g., Park and Van, 1991):
∂ln(Ui )
Sij = , (4.3)
∂ln(ρj )
where Ui is the potential difference measured with the i-th configuration and ρj is the
resistivity of the j-th cell. The Jacobian matrix is a key component of our (and any other)
tomographic inversion scheme, which can be formulated as follows:
−1 −1 T
ρ est ) = (S̄ T S̄ + C̄m
ln(~ ) S̄ ln(Ū ), (4.4)
−1
where ρ~ est are the updated resistivities and the matrix C̄m represents the regularization con-
ditions that are typically included in the form of damping and smoothing constraints (Mar-
quardt, 1970; Constable et al., 1987; Maurer et al., 1998). Taking the logarithm of Ū and ρ~
improves numerical accuracy and ensures positive values of ρ~ est (e.g., Park and Van, 1991).
The quality of ρ~ est can be quantified with the model resolution matrix R̄, which is defined
by setting:
ρ true ),
ln(Ū ) ≈ S̄ ln(~ (4.5)
so that equation (4.4) becomes

−1 −1 T
ρ est ) = (S̄ T S̄ + C̄m
ln(~ ρ true ) = R̄ ln(~
) S̄ S̄ ln(~ ρ true ), (4.6)
where
−1 −1 T
R̄ = (S̄ T S̄ + C̄m ) S̄ S̄. (4.7)
A diagonal element Rjj = 1 indicates perfect resolution of ρj , whereas Rjj = 0 proves that
ρj is completely unresolved.
High values in the diagonal elements of R̄ result when the elements of S̄ are large in mag-
p
nitude (compared with those of C̄m −1 ) and the rows of S̄ are linearly independent (Menke,
1984). These two requirements form the basis of our experimental design algorithm. Assume
that there are nm inversion cells and the Jacobian matrix of the comprehensive data set is
defined as S̄ compr . From the nd0 configurations of the comprehensive data set we select nbase
configurations that allow us to subdivide the (nd0 · nm) matrix S̄ compr into a (nbase · nm)
matrix S̄ base and a ((nd0 − nbase ) · nm) matrix S̄ add . At the beginning of the optimization
process, the nbase configurations would correspond to the array (e.g., dipole-dipole) used to
collect the initial data set (i.e., to start the process, the base data set corresponds to the
initial data set). In the simple version of our experimental design algorithm, a homogeneous
half-space model is used as the basis for all computations, whereas in the more advanced ver-
sion the successively updated inverted models are used for the calculations. Equation (4.7)
allows the model resolution matrices R̄compr and R̄base to be computed.
In the next step, we rank the configurations contained in S̄ add according to the goodness
¯ :
function GF
nm  add 
  !
base
X Sij Rjj
GF (i) = 1 − compr i = 1 · · · (nd0 − nbase ), (4.8)
j=1
Sjsum Rjj
where
nd 0
1 X S compr  .

Sjsum = 0 ij (4.9)
nd i=1
Sjsum is a normalization factor that counterbalances the generally high sensitivities of the near-
surface cells relative to the lower sensitivities of the deeper cells. Without this normalization
base compr
factor, the near-surface cells would dominate the goodness function. The (1 − Rjj /Rjj )
term forces the goodness function to favor configurations that constrain the resistivities of
cells for which the base data set exploits only a small portion of the resolution offered by the
comprehensive data set. Since we are primarily interested in resolving features shallower than
30 m depth, we exclude cells below this level from the optimization process. The goodness
function defined by equation (4.8) is one of many that could be used to search for optimal
data sets. For example, a goodness function could be designed to target specific regions
that are of particular importance or interest to an investigation. From the (nd0 − nbase )
possible add-on configurations, we choose the one with the highest goodness according to
equation (4.8). Because the information content of the base data set is not considered by
the goodness function, the linear independence of the candidate configuration from the base
data set needs to be checked. Assume that the k-th row of S̄ base has the best goodness
¯
function, then we need to compute the inner products LI:
 nm 
 P base add 
 Sij Skj 
 j=1  base
 S base  S add   i = 1 · · · n .
LI(i) =  (4.10)
    
 i k 
 
¯ are below a given threshold will S̄ add be added to S̄ base , nbase

Only if all values of LI
incremented, and the procedure repeated with the next best configuration of S̄ add . After
adding a number of new configurations to S̄ base , the model resolution matrix R̄base needs to
be recomputed to account for the improved resolution capabilities of S̄ base .
4.4.2 Application of the simple experimental design procedure to

synthetic data
base compr
Figure 4.3 shows the relative model resolutions (RMR’s) Rjj /Rjj in percent for the
Wenner, dipole-dipole, and combined Wenner - dipole-dipole arrays overlying a homoge-
neous half-space. The Wenner data set (Figure 4.3a) provides good resolution (i.e., model
resolutions approaching those of the comprehensive data set) only in the top three layers
of cells. At ∼18 m depth, the model resolution of the Wenner data set decreases to 20 -
30 % of that of the comprehensive data set. It declines further with increasing depth. The
dipole-dipole data set (Figure 4.3b) yields good resolution in the upper 8 m of the model. If
this depth range was the primary objective of an investigation, a conventional dipole-dipole
survey with an electrode spacing of ∼5 m would be a good cost-effective choice. Below 8 -
10 m depth, the RMR’s of the dipole-dipole data set drop abruptly to very small values. The
Wenner - dipole-dipole data set (Figure 4.3c) combines the advantages of the two individual
data sets, but below ∼10 m depth the model resolutions are noticeably inferior to those of
the comprehensive data. It is in the 10 - 30 m depth range that we expect the optimal data
sets to have the greatest impact.
In the following, we use a standard dipole-dipole array overlying a homogeneous half-space

for the computation of S̄ base and R̄base . It provides good resolution in shallow parts of the
model. Nevertheless, any electrical resistivity data set could be used to seed the optimization
process. Indeed, our experimental design algorithm functions when only a single observation
is used to initiate the procedure. Once the various S̄ and R̄ matrices are established for
the homogeneous half-space model, the optimization process can begin. The critical R̄base
matrix is recomputed after each 5 % increase in S̄ base .
Figures 4.4b, 4.4d, and 4.4f display as functions of data set size the summed RM R0 s =
Pncd base compr
j=1 Rjj /Rjj , where ncd is the number of cells in the selected depth intervals indicated
in Figures 4.4a, 4.4c, and 4.4e. The squares, stars, and open circles in these figures are the
respective sums of the Wenner, dipole-dipole, and combined Wenner - dipole-dipole RMR’s
shown in Figure 4.3. The curves show the summed RMR’s for each depth interval as the
optimal data sets are progressively added to the base dipole-dipole data set to form the
steadily changing optimized data set (i.e., optimized data set = initial dipole-dipole data set
plus the add-on optimal data sets).
Close to the surface (Figure 4.4b), the RMR’s show only minor improvements as the
number of data points in the optimized data sets grows. This is anticipated, since the
dipole-dipole data set already offers good resolution at shallow depths. Because of the
base compr
(1 − Rjj /Rjj ) term in the goodness function of equation (4.8), our procedure is biased
against add-on configurations that improve the information content of relatively well-resolved
regions of the model. At shallow depths (Figure 4.4b), the small Wenner - dipole-dipole data
set provides better RMR’s than the optimal data sets with as many as 2450 data points.
Between 15.4 and 18.4 m depth, a region that encompasses part of the conductive block
C, the advantages of the simple experimental design procedure start to become obvious
(Figures 4.4c and 4.4d). Here, the Wenner, dipole-dipole, and Wenner - dipole-dipole data
sets have RMR’s of 17, 21, and 32 %, respectively. For an optimized data set of 282 values
(first arrow in Figure 4.4d), equal in size to the Wenner - dipole-dipole data set, the RMR
is 40 %. This increases to 53, 59, 76, and 82 % for data sets with 670, 1050, 5740, and 10
310 values (see other arrows in Figure 4.4d), respectively. Note the steep rise in RMR values
as the number of data points increases to ∼1000, after which the curve flattens. Between
25.7 and 30 m depth, where the lower part of the conductive block C is located, the benefits
of the experimental design algorithm are unmistakable (Figure 4.4f). The combined Wenner
- dipole-dipole data set and equivalent size optimized data set have RMR’s of 14 and 34 %,
respectively. As for Figure 4.4d, the rate of increase of RMR’s is high in Figure 4.4f until
the number of data points reaches ∼1000.
Images derived from the optimized data sets with 282, 670, 1050, 5740, and 10 310 points
are displayed in Figures 4.5a - 4.5e and the associated RMR plots are presented in Figures
6a - 6e. As expected, the shallow part of the resistivity image determined from the optimized
282-point data set in Figure 4.5a is very similar to that derived from the base dipole-dipole
data set in Figure 4.1d. Unlike the standard array images of Figures 4.1c - 4.1e, the presence
of the conductive block C is readily apparent in the left third of Figure 4.5a. Furthermore,
RMR’s at depths between 12 and 30 m for the optimized 282-point data set (Figure 4.6a) are
uniformly higher than for the combined Wenner - dipole-dipole data set (Figure 4.3c), which
has the same number of data points. Consequently, we can have much more confidence in
this deeper region of Figure 4.5a than in the equivalent region of Figure 4.1e.
As the number of data points in the optimized data set is increased to 670, evidence
for the limited size of conductive block C appears (Figure 4.5b), and with 1050 optimized
data points the lateral and depth extent of conductive block C is quite well delineated
(Figure 4.5c). It is noteworthy that with only 1050 data points, 50 - 70 % of the model
resolution of the comprehensive data set with 51 373 data points is achieved in the 10 - 30
m depth range (Figure 4.6c). Relatively large increases in the number of data points to 5740
and 10 310 result in marginally improved images of the conductive block C and resistive
block D, but otherwise the changes to the models are cosmetic (Figures 4.5d and 4.5e).
Simultaneously, the RMR values increase to 80 - 90 % (Figures 4.6d and 4.6e). On the basis
of information contained in Figures 4.5 and 4.6, the 1050-point optimized data set would be
a reasonable compromise between the need to limit data acquisition costs and the goal of
obtaining high quality resistivity images, whereas the 5740-point optimized data set would
yield an image that is very close to the comprehensive image for about one-tenth of the field
effort.
Figure 4.7 identifies the various optimal electrode configurations that contribute data to
the optimized 282-, 670- and 1050-point data sets. The current and potential electrode sep-
arations are plotted in Figure 4.7a and their midpoint coordinates along the recording profile
are plotted in Figure 4.7b. Marked diagonals represent the 50, 75, and 100 m equidistant
lines between the midpoints of the current and potential electrodes. Red dots correspond to
the base dipole-dipole data set, whereas diamonds and squares represent double-dipole-type
and nested configurations, respectively. Dipole-dipole-type configurations have no overlap
between the current and voltage electrode pairs. Nested configurations have the voltage
electrode pairs located between the current electrodes (e.g., Wenner, Schlumberger and
gradient arrays) or vice versa. Symbol size increases as the number of configurations with
similar electrode separations (Figure 4.7a) or locations (Figure 4.7b) increases. The three
data sets are identified by the colors of the symbols: magenta, green, and blue correspond
to the 135, 388, and 380 incremental data points that are successively added to the base
147-point dipole-dipole data set to make the optimized 282, 670, and 1050 data sets.
It is noteworthy that the vast majority of electrode configurations that contribute to the
optimized 282- and 670-point data sets are dipole-dipole-type arrays (Figure 4.7). These
configurations are distinguished by a wide variety of current and potential electrode separa-
tions (i.e., the majority are non-standard dipole-dipole configurations) and distances between
the midpoints of the electrode pairs. Nevertheless, distances between the midpoints of the
current and potential electrode pairs generally decrease as more data points are added (e.g.,
note how the center of the pattern of magenta symbols lies close to the 100 m equidistant
line, whereas that for the green symbols lies close to the 75 m equidistant line).
The nested configurations start to be chosen by the experimental design algorithm once
the total number of data points exceeds ∼600. As the size of the optimized data set
is increased, the number of selected dipole-dipole-type and nested configurations becomes
approximately equal until there are roughly 5400 data points. For the nested configurations,
the midpoints of the current electrode pairs are mostly located near the center of the recording
line between distances 60 and 90 m, whereas the midpoints of the potential electrode pairs
are scattered between distances 20 and 120 m. These correspond to gradient arrays and other
non-standard electrode configurations (Ward, 1990a; Sharma, 1997; Dahlin and Zhou, 2002).
Only a few Wenner or Schlumberger arrays, which would lie along a diagonal line defining
zero offset between the midpoints of the current and potential electrodes in Figure 4.7b, are
chosen. Increasing the number of data points above 5400 results in the nested configurations
being chosen roughly twice as often as the dipole-dipole-type arrays.
4.4.3 First attempts to account for the influence of heterogeneous

resistivity distributions in the experimental design procedure
Our selection of optimal electrode configurations has so far been based on a homogenous half-
space model, such that the associated experimental design could be performed prior to a field
survey. However, subsurface heterogeneities may cause the injected current patterns to be
highly complex, which in turn may lead to quite different model resolution patterns to those
shown in Figures 4.2, 4.3, 4.4, and 4.6. Fortunately, it is a relatively simple matter to replace
the half-space model in the optimization process with more realistic models based on a priori
knowledge of the subsurface. For a completely new investigation site, one strategy would be
to acquire a limited initial data set using one of the standard electrode arrays. Once these data
are inverted, the resultant preliminary resistivity model could be used for the computation
of the sensitivities (equation (4.3)) and formal model resolutions (equation (4.7)) necessary
for starting the optimization procedure. After acquiring additional data with a suite of
optimal electrode configurations, the model could be updated and the optimization procedure
repeated. A series of “data acquisition - data inversion - update experimental design”cycles
may be required to achieve model convergence (Stummer et al., 2002). If the inversions
and experimental design algorithm are sufficiently fast, the calculations could eventually be
performed under field conditions in near real-time.
We explore briefly the effects of subsurface heterogeneities using the 2D resistivity model
in Figure 1a. Formal model resolutions for the comprehensive data set are high for the
shallow portions of this model (i.e., to ∼10 m depth in Figure 4.8a). They are particularly
high within the resistive block D. The pattern of decreasing model resolutions at depths
>10 m is similar to that determined for the homogeneous half-space model, except for the
anomalously low values within the conductive block C (compare Figures 4.2a and 4.8a).
Although currents are undoubtedly channeled through the conductive block, such that it
is well sampled, the natural logarithm of resistivity in equations (4.3) to (4.7) results in a
weighting factor that is approximately proportional to the square of resistivity being included
in the model resolution matrix R̄. This can be shown by rewriting equation (4.3) as:
∂ln(Ui ) ρj ∂Ui
Sij = = . (4.11)
∂ln(ρj ) Ui ∂ρj
Since R̄ depends primarily on S̄ T S̄ (equation (4.7)), the model resolutions scale roughly
as ρ2 . Furthermore, analysis of the complete resolution matrix within the conductivity block
C reveals relatively small values in the diagonal elements. In some areas, they are much
smaller than other elements. We conclude that the present experimental design algorithm
has a natural bias that favors regions of high resistivity at the expense of regions of high
conductivity. This point is highlighted in Figures 4.8b - 4.8d, which show moderately high
RMR’s within the resistive block C and anomalously low RMR’s within the conductive block
C for the Wenner, dipole-dipole and Wenner - dipole-dipole data sets. One way to resolve
this problem might be to design the goodness function (equation (4.8)) based on the model
covariance matrix (Menke, 1984; Alumbaugh and Newman, 2000). This possibility is the
subject of ongoing research.
Using the true sensitivities and model resolutions derived from the 2D resistivity model
in Figure 4.1a, optimized data sets containing 282, 670, 1050, 5740, and 10 310 data
points have been generated. The image in Figure 4.9a was derived from the optimized data
set containing 1050 data points. Visual comparison of Figure 4.9a with the corresponding
image based on the sensitivities and model resolutions derived from the homogeneous half-
space model in Figure 4.5c showed no significant differences. Subtraction of the image in
Figure 4.5c from that of Figure 4.9a revealed differences of only a few percent (Figure 4.9b).
Comparison of results for the other optimized data sets yielded similarly small differences.
Although the RMR plots derived from the homogeneous half-space and 2-D models differ
in several regions (compare Figures 4.3 and 4.8b - 4.8d), the rates at which the RMR’s decay
with depth are remarkably similar. As a consequence, for a 2D model distinguished by three
orders-of-magnitude resistivity variation (i.e., Figure 4.1a), the ρ2 bias in the model resolu-
tions used to compute our goodness function does not affect significantly, either positively
or negatively, the images derived from the various optimized data sets. Since our field site
is characterized by resistivity variations of less than two orders-of-magnitude, we select the
optimal data sets based on the homogeneous half-space model, but we also present formal
model resolution plots for the various derived models.
4.5 Application to a Field Data Set
4.5.1 The Stetten landfill test site
To test further our experimental design procedure, resistivity data were collected across
the well-studied Stetten landfill and surrounding sediments in northern Switzerland (Green
et al., 1999). The landfill contains heterogeneous industrial waste that was dumped into a
disused gravel pit before being covered by a 0.8 m layer of soil. Vertical-gradient magnetic
and frequency-domain electromagnetic data were used to define the lateral boundaries of
the landfill and the location of buried metallic objects (Lanz et al., 1999). Tomographic
inversions of densely spaced seismic refraction data allowed the base of the landfill to be
delineated over a wide area (Lanz et al., 1998). Detailed characteristics of the shallow host
sediments were provided by the results of 2- and 3D high-resolution seismic reflection and
georadar surveys (Lanz et al., 1994, 1996; Green et al., 1999), electrical resistivity mapping,
and electrical resistivity sounding (Lanz et al., 1998; Lanz, 1998), and information supplied
by several shallow drillholes (Lanz, 1998). These sediments comprised a ubiquitous unit of
lacustrine clayey fine sand that was intersected by numerous channels and lenses of fluvial
gravel and coarse sand. The groundwater table was encountered at ∼3 m depth.
A novel distributed acquisition system (Stummer et al., 2002) with thirty electrodes
spaced at 5 m intervals (total line length 145 m) was used to collect a comprehensive
resistivity data set along a line that crossed the Stetten landfill and adjacent sediments.
This acquisition system recorded data from multiple channels connected in parallel. A single
measurement sequence involved choosing two current electrodes and then simultaneously
recording 27 potential differences with respect to a selected reference potential electrode.
From the 12 180 (= 30 · 29 · 28/2) measurements, data for 328 860 four-point configurations
were reconstructed, of which 82 215 were non-reciprocal. After excluding data from all
configurations with crossed current and potential electrode dipoles and geometrical factors
larger than 5500 m and eliminating low S/N data, a comprehensive data set containing
49 542 measurements was obtained. Averages and standard deviations were determined
from the suites of remaining reciprocal values. These averages, inversely weighted by the
respective standard deviations, were the basis for our calculations. As for the synthetic
examples, a homogeneous half-space with resistivity equal to the average of the measured
apparent resistivities was used as the initial input model for each inversion.
Figures 4.10a and 4.10b show the resistivity image determined from the comprehensive
data set. The associated formal model resolutions in Figure 4.11a are comparable to those
determined for the synthetic comprehensive data set (Figure 4.8a). Given that a reliable
representation of the input 2D resistivity model was determined from the synthetic compre-
hensive data set (compare Figures 4.1a and 4.1b), this observation indicates that features
above ∼30 m depth in Figures 4.10a and 4.10b should be quite well resolved. We adopt the
comprehensive resistivity image of Figures 4.10a and 4.10b as the benchmark against which
other images will be evaluated.
Our interpretation of the resistivity images is constrained by the results of previous geo-
physical surveys at this site (Lanz et al., 1994, 1996, 1998, 1999; Green et al., 1999) and
borehole information (Lanz, 1998). Electrical resistivity units (i) in Figure 4.10a correspond
to near-surface fluvial gravels and coarse sands with high resistivities mostly in the 150 -
1000 Ωm range, whereas electrical unit (ii) is a sequence of lacustrine clayey fine sands with
uniformly low resistivities of 10 - 60 Ωm. Based on their high resistivities, we suggest that
electrical units (iv) are deep occurrences of fluvial gravels and coarse sands. Identified as
electrical unit (iii) in Figure 4.10a, the heterogeneous landfill is characterized by diverse but
generally low resistivities in the 10 - 100 Ωm range. The origin of the high resistive surface
feature at distance 60 - 75 m is unknown. It is the highly variable nature of the resistivi-
ties within the landfill that distinguishes it from adjacent units. Finally, the 80 - 110 Ωm
resistivities of electrical unit (v) likely represent a sandy till.
The Wenner, dipole-dipole, and combined Wenner - dipole-dipole data sets are extracted
from the comprehensive data set. Images determined from these standard data sets in Fig-
ures 4.10c - 4.10e, which are noticeably inferior to the comprehensive image in Figures 4.10a
and 4.10b, outline the correct position of the eastern landfill boundary, the high resistivities
within shallow parts of the gravels and coarse sands (resistivity unit (i)), and the generally
low resistivities within the landfill (resistivity unit (iii)). The anomalous surface feature at
distance 60 - 75 m is also reproduced in all images. There is a hint in Figures 4.10d and 4.10e
that moderate to high resistivities may underlie the eastern region of the landfill. Otherwise,
the standard arrays yield no useful knowledge below ∼10 m. The formal model resolution
plots in Figures 4.11b - 4.11d confirm the shallow nature of information provided by the
standard arrays.
Using the optimal configurations selected by the simple experimental design procedure,
optimized data sets containing 265 (same size as the combined Wenner - dipole-dipole data
set), 600, 1020, 4080, and 5670 data points are generated from the comprehensive data
set. The corresponding resistivity images and formal model resolution plots are shown in
Figures 4.12 and 4.13, respectively. When compared to Figures 4.10a, 4.10b and 4.11a, these
figures demonstrate the significant improvements that result from progressively increasing
the number of data points in the optimized data set. The 265-point image (Figure 4.12a) is
similar to the Wenner - dipole-dipole image (Figure 4.10e), except that evidence for moderate
to high resistivities below the eastern part of the landfill is much better in the former. The first
suggestion that the gravels and coarse sands (resistivity unit (i)) extend uninterrupted from
the surface to below the eastern part of the landfill and the first hint of variable resistivities
within the landfill are provided by the 1020-point image (Figure 4.12c). Except for the
eastern occurrence of the possible deep gravels and sands (resistivity unit (iv)), all important
features are reproduced in the 4080- and 5670-point images (Figure 4.12d and 4.12e). The
associated formal model resolution plots in Figures 4.13d and 4.13e are very similar to that
of the compressive data set in Figure 4.11a.
4.6 Conclusions
Using synthetic and field data examples, the imaging potential of data acquired with modern
multi-electrode resistivity systems has been assessed. For a given electrode spread deployed
along a survey line of fixed length, we have demonstrated that comprehensive data sets
recorded with large numbers of four-point electrode configurations provided significantly
more information than data sets collected with standard Wenner, dipole-dipole, and combined
Wenner - dipole-dipole arrays. As an example, using a thirty-electrode spread of 145 m length,
the comprehensive data sets supplied well-resolved information from the surface to depths
of ∼30 m, whereas the equivalent standard data sets only yielded details in the uppermost
∼10 m (Figures 4.1 and 4.10). To quantify the resolution of the derived images, we analyzed
the formal model resolutions (Figures 4.2 and 4.11).
Unfortunately, the recording of comprehensive data sets is not an option for routine
commercial surveying, even when the most versatile multi-electrode acquisition systems are
employed (Stummer et al., 2002); too many measurements are required. Depending on the
number of excluded configurations, a comprehensive data set recorded with a 30-electrode

spread may contain >50 000 values.
To define a limited suite of electrode configurations that supply resistivity images com-
parable in quality to comprehensive images, a new experimental design procedure has been
developed. This is probably the first attempt to take full advantage of the imaging capabili-
ties of modern multi-electrode acquisition systems. To initiate the procedure, it is necessary
to generate or record a small base data set. For the synthetic and field data examples, we
use a standard dipole-dipole data set. The core of the optimization procedure is a goodness
function (equation (4.8)) that ranks the sensitivities (Jacobian defined by equation (4.3))
of all possible electrode configurations. This goodness function includes weighting terms to
counterbalance the high sensitivities of shallow parts of the model relative to deeper parts
and to minimize the influence of well-resolved regions of the base model on the experimental
design procedure. Data generated or recorded with electrode configurations that provide
large amounts of new information according to their high sensitivities and depth of influence
are incorporated into the successively increasing optimal data sets. A final optimized data
set is the sum of the base dipole-dipole and add-on optimal data sets.
Using our simple experimental design procedure, optimized synthetic and field data sets
with progressively more values have been generated. Relatively small optimized data sets
(265 - 282 data points) were shown to supply resistivity images superior to those provided by
the combined Wenner - dipole-dipole data sets with equal numbers of data points (compare
Figure 4.5a with Figure 4.1e and Figure 4.12a with Figure 4.10e). Nevertheless, optimized
data sets containing an order of magnitude more data points yielded substantially more
information (Figures 4.5 and 4.12). For the 30-electrode spread, relative model resolution
plots (Figure 4.4) demonstrated that the rate of increase of information was high until
the number of optimized data points exceeded ∼1000. Although the ∼1000-point images
revealed many key details, images that approached the quality of the comprehensive image
required 4000 - 6000 data points. For example, the 4080-point image in Figure 4.12d
(compare with Figures 4.10a and 4.10b):
1. provided well-resolved information from the surface to a depth of ∼30 m;
2. highlighted the heterogeneous nature of resistivities within the Stetten landfill;
3. delineated its lateral boundaries and base;

4. demonstrated the continuation of high-resistivity fluvial gravels and coarse sands be-
neath its eastern boundary;
5. outlined the extent of low-resistivity lacustrine clayey fine sands;
6. suggested the existence of the deep fluvial gravel and coarse sand unit beneath the
western end of the survey line;
7. provided evidence for the intermediate-resistivity deep till unit.
Our simple experimental design algorithm is based on a homogeneous half-space model.

No assumptions about the subsurface resistivity distribution are required to determine the
optimized electrode configurations. This makes the approach generally applicable, such that
the optimized electrode configurations may be established prior to a field campaign. Only
the number and spacing of electrodes and discretization of the subsurface model are required
as input parameters. The first ∼600 electrode configurations selected by the simple exper-
imental design procedure are non-standard dipole-dipole-type arrays (Figure 4.7), whereas
the next ∼4800 are an equal number of non-standard dipole-dipole-type arrays and nested
configurations (i.e., mostly gradient and non-standard arrays).
We have also briefly examined experimental design procedures based on complex resistivity
models, such as those resulting from the inversion of preliminary field data sets. Initial
results from applying this scheme were surprisingly similar to those obtained using the simple
experimental design algorithm (Figure 4.9). To improve this scheme, we intend to test
other possible goodness functions. One promising approach might be to base the goodness
function on the model covariance matrix (Alumbaugh and Newman, 2000).
Application of the experimental design procedure to delineate specific targets is another
avenue of research that is being pursued. This involves including in the goodness function
special weighting factors that favor parts of the model space where the targets are expected.
Finally, we are expanding the algorithm to include the design of 3D surveys. Considering
the large number of electrodes required for 3D resistivity methods and the associated wide
variety of possible electrode configurations, we expect optimized experimental design to be
very important for such surveys.
4.7 Acknowledgements
Financial support for this project was provided by the Research Commission of ETH Zurich
and the Swiss National Science Foundation.
log Resistivity [Ω m]
10
0 1 2 3 4
0 A
10 D
Depth [m]
20
30 C
40 B
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(e)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.1: (a) Resistivity model employed for the analysis of data incompleteness and equivalence: A -
100 Ωm upper layer (green); B - 1000 Ωm lower layer (yellow); C - conductive block that grades from 100
Ωm to 10 Ωm in the center (blue); D - 10 000 Ωm resistive block (black). Simulation of all synthetic data
sets and other computations are based on a thirty-electrode spread with 5 m electrode spacing (145 m total
length). Recovered resistivity models after inverting: (b) 51 373-point comprehensive data set (reciprocal
measurements, values simulated using Wenner γ-arrays, and values simulated using electrode configurations
with geometrical factors >5500 m are not included in the calculations) using a 360 Ωm homogeneous input
model, (c) 135-point Wenner data set using a 360 Ωm homogeneous input model, (d) 147-point dipole-dipole
data set using a 210 Ωm homogeneous input model, and (e) 282-point combined Wenner - dipole-dipole
data set using a 280 Ωm homogeneous input model. We seek to optimize resistivity information at depths
630 m.
Formal Model Resolution
0 0.2 0.4 0.6 0.8 1

0
10
Depth [m]
20
30
40
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.2: For a homogeneous half-space model, absolute values of the formal model resolutions
for: (a) 51 373-point comprehensive, (b) 135-point Wenner, (c) 147-point dipole-dipole, and (d)
282-point combined Wenner - dipole-dipole synthetic data sets.
Relative Model Resolution [%]
0 20 40 60 80 100
0
10
Depth [m]
20
30
40
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.3: For a homogeneous half-space model, relative model resolutions (RMR’s) in percent
for: (a) 135-point Wenner (Figure 4.2b / Figure 4.2a), (b) 147-point dipole-dipole (Figure 4.2c /
Figure 4.2a), and (c) 282-point combined Wenner - dipole-dipole (Figure 4.2d / Figure 4.2a)
synthetic data sets.
100
↓ ↓
Average Depth: 2.8 m ↓↓↓
Depth Range: 2.1 to 3.4 m 80
Rel. Model Resol. [%]

Formal Model Resolution: 0.453
0
Depth [m]
60
20
40 40
(a)
0 50 100
Distance [m] 20
(b) 0
0 2000 4000 6000 8000 10000
No. of Data
100
Average Depth: 16.9 m
Depth Range: 15.4 to 18.4 m
↓
Rel. Model Resol. [%] 80 ↓
0
↓
Depth [m]
60
20 ↓
40 40
↓
(c)
0 50 100
Distance [m] 20
(d) 0
0 2000 4000 6000 8000 10000
No. of Data
100
Average Depth: 27.8 m ↓
Depth Range: 25.7 to 30.0 m 80 ↓
Rel. Model Resol. [%]

0
↓
Depth [m]
60
20
↓
40 40 ↓
(e)
0 50 100
Distance [m] 20
(f) 0
0 2000 4000 6000 8000 10000
No. of Data
Figure 4.4: For experimental design based on a homogenous half-space model, diagrams demon-
strate the dependency of relative model resolutions (RMR’s) on choice and size of data sets. Plots
(a), (c), and (e) show the locations and thicknesses of the analyzed horizontal sections superim-
posed on the inversion grid, together with the absolute values of the formal model resolutions for
the respective comprehensive data sets. Plots (b), (d), and (f) show RMR’s for 135-point Wenner
(squares), 147-point dipole-dipole (stars) and 282-point combined Wenner - dipole-dipole (circles)
data sets. Using the dipole-dipole data set as a basis, the solid lines show RMR’s for optimized
data sets (see text) as the number of data points is increased. Arrows indicate data subsets (282,
670, 1050, 5740, and 10 310 data points) chosen for inversion runs.
10
0 1 2 3 4
0
10
Depth [m]
20
30
40
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(e)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.5: Recovered resistivity models after inverting five optimized synthetic data sets: (a)
282-point data set using a 250 Ωm homogeneous input model, (b) 670-point data set using a 330
Ωm homogeneous input model, (c) 1050-point data set using a 380 Ωm homogeneous input model,
(d) 5740-point data set using a 400 Ωm homogeneous input model, and (e) 10 310-point data
set using a 430 Ωm homogeneous input model. In these examples, the electrode configurations
optimized for a homogenous half-space are employed. We seek to optimize resistivity information
at depths 630 m.
0 20 40 60 80 100
0
10
Depth [m]
20
30
40
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(e)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.6: For a homogeneous half-space model, relative model resolutions (RMR’s) in percent
for the five optimized data sets of Figure 5: (a) 282 data points (147 dipole-dipole + 135 add-on
optimal data points), (b) 670 data points (147 dipole-dipole + 523 add-on optimal data points), (c)
1050 data points (147 dipole-dipole + 903 add-on optimal data points), (d) 5740 data points (147
dipole-dipole + 5593 add-on optimal data points), and (e) 10 310 data points (147 dipole-dipole
+ 10 163 add-on optimal data points). Compare these diagrams with those in Figure 4.3.
150
Potential Electrode Separation [m]

100
50
0
0 50 100 150
(a) Current Electrode Separation [m]
150
Potential Electrodes Midpoint Coord. [m]
100
50
0
0 50 100 150
(b) Current Electrodes Midpoint Coord. [m]
Figure 4.7: Electrode configurations of the add-on optimal data sets using experimental design
procedures based on a homogenous half-space model. (a) Current electrode separations versus po-
tential electrode separations. (b) Midpoint locations of the current electrode pairs versus midpoint
locations of the potential electrode pairs. Red dots represent the initial dipole-dipole data set.
Diamonds represent double-dipole-type configurations with separate potential and current dipoles.
Squares represent configurations for which the potential electrode pairs are nested between the
current electrodes or vice versa. Symbol sizes are a function of the number of configurations with
similar electrode (a) separations or (b) locations. In (a) and (b), magenta symbols identify the
first 135 incremental data points, green symbols identify the next 388 incremental data points and
blue symbols identify the next 380 incremental data points. Lines of equal distance (50, 75, and
100 m) between the midpoints of the current and potential electrodes are indicated.
0 0.2 0.4 0.6 0.8 1

0
Depth [m] 10
20
30
40
50
(a)
0 20 40 60 80 100 120 140
0 20 40 60 80 100
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.8: For the resistivity model in Figure 4.1a: (a) absolute values of the formal model
resolutions for the 51 373-point comprehensive data set (compare this diagram with Figure 4.2a,
which shows same information for a homogenous half-space model); relative model resolutions
(RMR’s) in percent for the (b) 135-point Wenner, (c) 147-point dipole-dipole, and (d) 282-point
combined Wenner and dipole-dipole synthetic data sets (compare these diagrams with those of
Figure 4.3, which show the same information for a homogenous half-space model).
10
0 1 2 3 4
0
10
Depth [m]
20
30
40
50
(a)
0 20 40 60 80 100 120 140
Difference [%]
−10 0 10 20
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.9: (a) Recovered resistivity model after inverting the initial 147 dipole-dipole resistivity
values plus an optimized suite of 903 resistivity values determined for the resistivity model in
Figure 4.1a using a 360 Ωm homogeneous input model. We seek to optimize resistivity information
at depths 630 m. (b) Percent-difference of the model in (a) relative to the model depicted in
Figure 4.5c.
W log
10
Resistivity [ Wm ] E
1 1.5 2 2.5
(i) (iii) (i)
0
(ii)
10
(ii)
Depth [m]
20
30
(iv)
(iv)
40
(v)
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(e)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.10: Models that result from inverting field data sets acquired along a line of thirty
electrodes with 5 m spacing (145 m total length): (a) and (b) 49 542-point comprehensive data
set using a 70 Ωm homogeneous input model, (c) 134-point Wenner data set using a 70 Ωm
homogeneous input model, (d) 131-point dipole-dipole data set using a 80 Ωm homogeneous input
model, (e) 265-point combined Wenner - dipole-dipole data set using a 70 Ωm homogeneous input
model. We seek to optimize resistivity information at depths 630 m. Landfill and geological units
identified in (a): (i) - fluvial gravels and coarse sands, (ii) - lacustrine clayey fine sands, (iii) -
landfill, (iv) - probable fluvial gravels and coarse sands, (v) - probable sandy till.

W E
0 0.2 0.4 0.6 0.8 1
0
10
Depth [m]
20
30
40
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.11: Absolute values of the formal model resolutions for the corresponding models of
Figure 4.10: (a) 49 542-point comprehensive, (b) 134-point Wenner, (c) 131-point dipole-dipole,
and (d) 265-point combined Wenner - dipole-dipole (265 values) data sets.
W log
10
Resistivity [W m] E
1 1.5 2 2.5
0
10
Depth [m]
20
30
40
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(e)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.12: Models that result from inverting five optimized field data sets acquired along a line
of thirty electrodes with 5 m spacing (145 m total length): (a) 265-point data set using a 90 Ωm
homogeneous input model, (b) 600-point data set using a 80 Ωm homogeneous input model, (c)
1020-point data set using a 70 Ωm homogeneous input model, (d) 4080-point data set using a 70
Ωm homogeneous input model, (e) 5470-point data set using a 70 Ωm homogeneous input model.
We seek to optimize resistivity information at depths 630 m. Compare these results with those
shown in Figure 4.10.

W E
0 0.2 0.4 0.6 0.8 1
0
10
Depth [m] 20
30
40
50
(a)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(b)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(c)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(d)
0 20 40 60 80 100 120 140
0
10
Depth [m]
20
30
40
50
(e)
0 20 40 60 80 100 120 140
Distance [m]
Figure 4.13: Absolute values of the formal model resolutions for the corresponding models of
Figure 4.12: (a) 265-point, (b) 600-point, (c) 1020-point, (d) 4080-point, and (e) 5470-point data
sets.
Chapter 5
Outlook
The goals of this thesis were to design and construct of a flexible DC-resistivity multi-
electrode system and develop an adaptive data acquisition strategy.
Comparison of various instrument configurations has shown that a fully distributed system
provides the flexibility required for implementing adaptive experimental design (Chapter 2).
My new digital resistivity system, the ETH DC Multi-Electrode System (ETH-DCMES),
meets this need. Though expensive, it offers the following advantages over conventional
equipment:
• capable of being expanded to include a very large number of electrodes (necessary for
efficient 3D recording),
• software-controlled selection of arbitrary electrode configurations,
• parallel recording of many configurations,
• low susceptibility to ambient electromagnetic noise,
• full-waveform recording,
• high overall accuracy,
• automated field-operation,
• arbitrary source signals.
Laboratory tests have proven the high accuracy of the ETH-DCMES (Chapter 3), whereas
its ruggedness and reliability have been established in an extensive field campaign conducted
102 5. Outlook
under harsh weather conditions. This field campaign involved the acquisition of the first-ever
comprehensive data set (i.e., all possible current - voltage electrode configurations for a
given number of electrodes [30 electrodes for this example]). Assessment of the data quality
and noise levels was possible through the recording of full-waveform data. The acquired
data were processed using an innovative frequency-domain scheme, which delivered reliable
resistivity values at locations where IP-effects had only limited influence on the data.
Analyses of synthetic and field examples demonstrated that data recorded with standard
electrode arrays (e.g., Wenner, dipole-dipole, etc.) provided only limited knowledge about
the subsurface. Using the same spread of electrodes, much more information was obtained
by using a combination of standard and non-standard electrode configurations (Chapter 4).
Originally, I planned to develop a real-time experimental design procedure that would allow
me to define, during the course of a field survey, those electrode configurations that would
provide optimum subsurface information at minimum cost (Chapter 2). The ETH-DCMES
was developed with this objective in mind.
Tests on synthetic data suggested that the optimization procedure based on a homoge-
nous half-space model worked well in areas characterized by relatively heterogeneous resis-
tivity distributions. Indeed, results obtained using this simple approach were very similar to
those obtained using an optimization procedure based on the true subsurface model. Con-
sequently, the simple optimization procedure was used for many tests based on the Stetten
field data.
At this stage of experimental design development, it remains unclear as to whether the
simple approach will be sufficient in all situations or whether a new cost function needs to be
defined. I suspect that a new cost function applied in a non-linear fashion may lead to results
that are superior to those presented in Chapter 4. Ultimately, a quasi-real-time experimental
design algorithm may lead to a procedure that can be applied in the field.
Several concrete ideas for improving the ETH-DCMES and the experimental design
scheme are discussed in the following sections.
5. Outlook 103
5.1 Improvements to the ETH-DCMES
5.1.1 Weight reduction and mobility
The ETH-DCMES is designed primarily to meet the needs of adaptive experimental design
and the acquisition of dense 3D data sets. For regular 2D resistivity surveying, the system
is somewhat bulky for effective handling in the field. The main cause of this problem is the
combined weight of the generator and signal transmitter. The signal transmitter comprises
laboratory-standard equipment, with the amplifier and UPS requiring a 220 V power supply
and special protection (e.g., from rain and dirt) for field use (Figure 2.6). Replacing the
amplifier and UPS with a lightweight battery-powered unit would not only decrease markedly
the weight of the signal transmitter, but it would also remove the requirement for the
generator. Unfortunately, extensive market research has not been successful in tracking
down a rugged low-frequency battery-powered transmitter that allows signals with arbitrary
waveforms to be generated. Such a special purpose transmitter could, however, be designed
and constructed in a modern technical laboratory. Considering the modular design of the
ETH-DCMES, it would be a relatively simple matter to replace the present heavy transmitter
with a new light-weight one. The primary source signal would still be generated on a DA-
board attached to the PC (Figure 2.3).
5.1.2 Reducing data acquisition time by distributing the data

analysis
The ETH-DCMES records full-waveforms, which are necessary for a variety of analysis pro-
cedures (e.g., studies of induced polarization) and research. However, the recording of
full waveforms is time-consuming. For resistivity surveying of the type described in Chap-
ter 4, the time to acquire the data could be decreased substantially by sending only a small
amount of information (e.g., the averages of the potentials measured at the frequency of the
source-signal and estimates of the average background noise) from the DAU’s to the cen-
tral recording PC. Basic processing steps required to compute the desired parameters could
be performed on the small CPU’s contained in the DAU’s. To acquire the comprehensive
full-waveform data set at the Stetten site required thirteen days of recording. By only trans-
ferring the average resistivity and noise estimates, the acquisition time would be reduced to
104 5. Outlook
two days. It is informative to compare the recording capabilities of the ETH-DCMES with
those of the GeoTom system, arguably one of the best commercial systems on the market
(Chapter 2). To acquire a 3D pole-pole data set using a 15 x 15 electrode spread and a 1
Hz signal would require fifty-six hours with the GeoTom system and a little less than one
hour with the ETH-DCMES, assuming that only the average resistivity and noise estimates
were transferred from the DAU’s.
With an upgraded ETH-DCMES, the user should be able to choose between a full-
waveform mode and a resistivity surveying mode. This concept could be extended further
by implementing a software download feature, such that users could pre-define a processing
sequence (e.g., an algorithm tailored to process responses of a specific source signal) on the
PC and then download it to the DAU’s before initiating the field campaign. The processing
sequence would be executed on the internal CPU’s of the DAU’s and only the processed
information transferred to the field PC.
5.1.3 Surveying with long electrode spreads
Generally, the length of an electrode spread is limited by the maximum extent of the digital
data bus, which is defined by the RS-485 standard of 1.2 km. Although the ETH-DCMES
was not designed for surveying with long electrode spreads, by using four shielded four-wire
extension cables it would be possible to construct a spread that included 120 DAU’s spaced
at regular 10 m intervals (Figure 5.1). Only short reference potential cables (solid green
lines in Figure 5.1) would be required to connect the reference lines between the DAUs at
the ends of the buses. The length of the electrode spread could be doubled by deploying
electrodes and cables on either side of the interface box.
When planning the deployment of long electrode spreads, it is necessary to account for
the relatively low data transfer rates that accompany the long bus lengths. Extending the bus
length beyond 10 m causes data transfer rates to drop as a result of cable capacitance and
when they exceed several hundred meters the cable resistance becomes a limiting factor (e.g.,
Stanek et al., 2002). For the ETH-DCMES, a data transfer rate of 38.5 kBit/s allows error-
free transfer of full-waveform data over the maximum 1.2 km electrode spread even under
very noisy field conditions.
5. Outlook 105
to Field PC
#1 RS-485 / Source Signal

(max. 124 DAU's)
Interface Box
#2 Ref. Pot.
91 120
#3
61 90
#4
31 DAU No. 60
1 30
10 m 290 m 10 m 290 m 10 m 290 m 10 m 290 m
1200 m
from
Signal
Source
Figure 5.1: Schematic layout of the ETH-DCMES for acquiring long profiles.
5.1.4 System expandability
When this project was initiated, the number of RS-485 channels and thus the number of
electrodes (DAU’s) could be increased by attaching additional multi-channel RS-485 interface
cards to the field PC. The number of PC extension slots, therefore, limited the number
of electrodes (Chapter 2). Presently, Universal Serial Bus (USB) technology provides the
opportunity to expand greatly the maximum number of electrodes in a cost-effective manner.
This possibility will be particularly important for efficient 3D surveying with large numbers
of electrodes.
A single USB-to-RS-485 interface can receive data from a single RS-485 interface con-
nected to four RS-485 channels, each of which can accommodate 31 electrodes. Information
from seven USB-to-RS-485 interfaces can be transferred through a single low-cost USB hub.
Furthermore, data from seven such low-cost USB hubs can be received by a single USB hub
further down the data acquisition line, and so on. As an example, Figure 5.2 shows how
seven USB hubs each serving seven USB-to-RS-485 interfaces are connected via a master
USB hub to the controlling PC. Such a setup would result in a spread consisting of 4·31·7·7
= 6076 electrodes.
106 5. Outlook
from
Signal
Source
DAU No.
1 31
#1
Ref. Pot.
Interface Box 1
(max. 124 DAU's)
USB #2
to
RS-485 #3
#1 94 124
#4
#1
RS-485 / Source Signal
#2
USB
HUB DAU No.
#1 #1 745 775
#7
#1
to Field
#2 Ref. Pot.
Interface Box 7
(max. 124 DAU's)
Notebook Master
USB #2
USB
HUB to
RS-485 #3
#7 #7 838 868
#4
RS-485 / Source Signal
to from
USB-HUB Signal
#7 Source
Figure 5.2: Extension of the ETH-DCMES to include a larger number of DAU’s using USB-
technology.
Theoretically, data exchange rates between USB devices can reach 12 Mbit/s over dis-
tances up to 5 m. Such short distances between the USB devices would be acceptable for
3D surveying. This USB-based approach would allow the costly and somewhat fragile PC
(i.e., one with numerous full-size PCI extension slots) that is presently at the heart of the
ETH-DCMES to be replaced by a conventional field-rugged PC containing the now-standard
USB port and equipped with a card-size DA-board to generate the source signal. Besides
smaller volume and lower weight, the overall power consumption of the system would be
reduced by implementing this modification, which would require no hardware alterations to
the existing interface boxes and DAU’s. Even though this modification would speed-up data
transfer between USB devices, the data transfer rate along the entire acquisition line would
continue to be limited by the maximum bus length.
5. Outlook 107
5.1.5 Acquisition of induced polarization data
One important characteristic of the ETH-DCMES has yet to be exploited: the possibility
to acquire induced polarization (IP) data (e.g., Fink, 1990). Since any source signal can
be employed (subject to limitations imposed by the amplifiers frequency range and output
power), conventional IP and spectral IP (SIP) measurements (e.g., Luo and Zhang, 1998)
can be made with the ETH-DCMES.
Time-domain IP usually requires a square-wave source signal, whereas the SIP method
involves making a series of measurements over a range of signal frequencies (e.g., a swept
signal). SIP would be my method of choice, because phase-shifts can be determined directly
by simply cross-correlating the measured voltages with the injected current. It will be im-
portant to ensure that the DAU’s are properly synchronized; additional phase-lags would be
introduced if the DAU’s were not triggered at exactly the same time. Synchronization tests
with the DAU’s set to their highest sampling frequency of 1 kHz revealed no measurable
timing errors. However, if the number of DAU’s would be increased significantly, synchro-
nization errors might become an important issue. In such a case, the correction scheme
described by Bigalke et al. (1999) could be applied.
5.2 Influence of Noise
Here, I consider the influence of various types of noise on the acquired data and the subse-
quent data inversion. Model resolution is generally limited by the quality of the field data,
which to some extent is dependent on the data acquisition system. This is especially critical
for resistivity data, since subsurface electrical properties can vary over several decades (Chap-
ter 1). As a consequence, the recording of resistivity data requires instruments with large
dynamic ranges. The introduction of inexpensive high-resolution ADC’s helped to resolve
this problem and eventually led to the development of modern multi-electrode resistivity
systems (Chapter 2).
One major advantage of the ETH-DCMES is its full-waveform recording capability. Trans-
formation of the measured time series into the frequency domain allows the signals (i.e., the
averages of potentials measured at the frequency of the source-signal) to be distinguished
from the ambient EM noise (i.e., the averages of potentials measured well away from the
source-signal frequency).
108 5. Outlook
0
10
Std. Dev. of four Measurements; Mean = 2.91 mV
Estimated Noise; Mean = 0.05 mV
Computational Accuracy; Mean = 0.03 mV
−1
10
−2
10
Pot. Diff. [V]
−3
10
−4
10
−5
10
Average Device Noise Level: ~1.3 µV

−6
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5
No. of Data 4
x 10
Figure 5.3: Result of 52 100 resistivity observations at the Stetten field site. Standard deviations
of suites of four measurements (solid blue line) made at the 1 Hz source-signal frequency (see
Figure 2.7) and corresponding mean value of all standard deviations (dashed blue line); noise levels
(solid red line) with corresponding mean value (dashed red line) based on values recorded well away
from the source-signal frequency (i.e., 12 - 30 Hz); computational accuracy (solid green line) and
corresponding mean value (dashed green line) achievable with the newly developed forward and
inverse algorithms; average device noise level controlled by the resolution of the ADC (dashed black
line).
A subset of the comprehensive data set acquired at the Stetten test site (Chapter 4)
has been analyzed with respect to data quality (Figure 5.3). The subset under investigation
consisted of 52 100 suites of voltages, each comprising four reciprocal values (i.e., a total
of 208 400 values; note, that reciprocal values that result from interchanging voltage and
current electrodes were not measured). The mean of these voltages was ∼0.5 V and the
maximum was ∼2.4 V. Standard deviations of these 52 100 suites of voltages (solid blue
line in Figure 5.3) revealed a distinct pattern influenced by changing weather conditions and
different levels of noise that were a function of electrode geometry. The mean value of
these standard deviations, which was ∼2.9 mV (dashed blue line in Figure 5.3), was inserted
in equation 1.20 to determine the maximum allowable geometric factor Kmax . Using the
average ∼80 Ωm resistivity of the Stetten site and the 200 mA injected current resulted in
5. Outlook 109
Kmax ∼
= 5500 m. Only data acquired with configurations with K-values smaller than ∼5500
m were included in the inversions described in Chapter 4.
The mean of all 52 100 noise estimates (i.e., solid red line in Figure 5.3 shows noise values
determined from the suites of four reciprocal measurements) is ∼0.05 mV (dashed red line),
whereas the average noise level of the ETH-DCMES, which is defined by the ADC’s, is ∼1.3
µV (dashed black line). The orders-of-magnitude differences between the blue, red, and black
lines demonstrates that the signal and ambient noise are well resolved by the ETH-DCMES.
Typical surveys in applied and environmental geophysics are conducted in urban areas, such
that high levels of EM noise may affect the measured potential differences (Chapter 2). In
these areas, the instruments electronics do not control limit the achievable resolution, but
rather the ambient EM noise is the dominant factor.
In addition to being dependent on data quality, the inversion runs reported in Chapter 4
are also limited by computational precision. The following simple numerical experiment,
which only involves the forward calculations of the inversion scheme, provides a rough esti-
mate of the achievable computational accuracy. Using the comprehensive Stetten resistivity
model (Figures 4.10a and 4.10b), the voltages for all electrode configurations of the 30-
electrode spread are numerically simulated. For perfect numerical precision, the voltages
computed for the reciprocal electrode configurations should be identical to each other, such
that the standard deviations between the reciprocal values would be zero. In fact, the stan-
dard deviations (solid green line in Figure 5.3) follow closely the measured noise estimates
(solid red line). The mean value of the computed standard deviations is 0.03 mV (dashed
green line). The computational errors, which result from approximating the governing partial-
differential equations, are of the same order-of-magnitude as the ambient EM noise levels at
Stetten. The parallel nature of the blue, red and green lines in Figure 5.3 demonstrate the
strong influence of the electrode configuration on the voltage estimates.
From Figure 5.3 and the preceding discussion, I conclude that: (i) higher resolution ADC’s
will not necessarily lead to higher quality data, primarily because of the inherent background
noise that is present at most survey sites; (ii) redundant (i.e., reciprocal) measurements can
provide useful guidelines for choosing maximum site-dependent K-factors, which in turn can
be used to select data to be included in the inversions; (iii) noise measurements made in the
field supply useful estimates of data quality; (iv) future research efforts should focus on the
development of more accurate numerical forward algorithms and matrix equation solvers.
110 5. Outlook
5.3 Future Developments of the Experimental Design

Strategy
In Chapter 4, a fast approximate experimental design algorithm based on sensitivities com-

puted for a homogenous half-space model was introduced. Application of this algorithm
resulted in synthetic and field data sets that yielded resistivity images with higher model
resolutions than images derived from the standard Wenner, dipole-dipole, and combined
Wenner - dipole-dipole data sets. This simple experimental design scheme could easily be
extended to take advantage of the parallel recording capabilities of the ETH-DCMES. In-
stead of searching for individual electrode configurations with the highest goodness values
(equation 4.8), the optimization process would seek suites of electrode configurations that
could be recorded in parallel and together have optimally high goodness values.
Using realistic heterogeneous models in the experimental design scheme yielded images
that were surprisingly similar to those based on the homogeneous half-space model (Chap-
ter 4). Yet, the model resolution plots and consideration of how subsurface currents are
distributed in heterogeneous models suggest that the anomalous blocks of low and high
resistivity should have an important impact on the experimental design procedure. This
point is highlighted by comparing the relative model resolutions (RMR’s) of the Wenner,
dipole-dipole and combined Wenner - dipole-dipole data sets derived from the homogenous
half-space model (Figure 4.3) with the corresponding RMR’s derived from the synthetic
model (Figures 4.8b - 4.8d). The regions of anomalously high and low resistivity influence
significantly the RMR’s. The similarity of the images derived by employing the homogeneous
and the heterogeneous models in the experimental design scheme may be a consequence of
the electrode geometry playing a greater role than the subsurface resistivity distribution in
determining the current patterns.
I suggest two strategies for enhancing the experimental design procedure as applied to
heterogeneous earth models: (i) attempt to improve the goodness function and (ii) find
or develop fast new methods for approximating the matrices so that more comprehensive
search procedures can be employed. As mentioned briefly in Chapter 4, one way to improve
the goodness function might be to use the model covariance matrix rather than the model
resolution matrix in equation 4.8. The model covariance matrix describes how data errors are
mapped into model errors (Menke, 1984), such that the square roots of its diagonal elements
5. Outlook 111
represent the standard errors of the model parameters. For the particular case of logarithmic
inversions the model covariance matrix provides relative errors, which for experimental design
purposes may be superior to the formal model resolutions (Alumbaugh and Newman, 2000).
One would simply replace Rjj in all equations with 1/Cmjj , where C̄m is the model covariance
matrix.
The ultimate goal of experimental design is to maximize the overall model resolution
(i.e., the sum of the diagonal elements of the model resolution matrix) or to minimize the
model covariance (i.e., the sum of the diagonal elements of the model covariance matrix).
In the experimental design schemes described by Maurer and Boerner (1998) and Maurer
et al. (2000), this is achieved by testing many different configurations and choosing the one
that provides the best resolution. Such an approach requires numerous computations of the
respective matrices, which is computationally prohibitive for realistic problems of the type
discussed in Chapter 4. For this reason, the algorithm described in Chapter 4 assesses the
sensitivities of the individual measurements (equation 4.8) and estimates their independence
(i.e., their complimentary characteristics) using the inner products (equation 4.10). The
combination of equations 4.8 and 4.10 enables the information contained in the resolution
matrix to be estimated approximately. A possible extension of this methodology would be to
find clever new approximations to the model covariance matrix that can be computed swiftly.
Then, genetic algorithms could then be used to solve the experimental design problem.
Appendix A
Resolution Capabilities of Standard

Electrode Configurations
To appreciate better the resolution capabilities of data acquired with a typical standard
electrode configuration, we analyze the noise-free Wenner and dipole-dipole pseudo-sections
shown in Figures A.1 and A.2 respectively. Figure A.1a depicts a Wenner pseudo-section
computed from the synthetic resistivity model shown in Figure 4.1a, whereas Figure A.1b
displays the Wenner pseudo-section computed from the Wenner model in Figure 4.1c. The
percent-differences of the two sections are shown in Figure A.1c. The average RMS-misfit
is 0.12 % and the individual differences are mostly less than 1 %.
In a next step, we simulate Wenner pseudo-sections using the dipole-dipole model shown
in Figure 4.1d and the complete model shown in Figure 4.1b (Figures A.1d and A.1f). The
simulation completely fails in this case, because it is based on the dipole-dipole model as
shown in Figure A.1d. The associated percent-difference plot reveals individual errors of >10
%, which could be explained by the limited penetration depth of the dipole-dipole array (Fig-
ure A.11e). Missing information of deeper sections in the dipole-dipole model (Figure 4.1d)
prevent perfect reconstruction of a Wenner pseudo-section. The opposite is true for the
synthetic Wenner pseudo-section based on the comprehensive model (Figure 4.1b). Here,
nearly perfect match between the original (Figure A.1a) and the synthetic (Figure A.11f)
pseudo-section is achieved (Figure A.1g). Similar conclusions to those derived from our study
of the Wenner model and various pseudo-sections (Figure A.1) apply to the dipole-dipole
models and associated pseudo-sections (Figure A.2). The simulations reveal the limited
depth resolution of both standard data set.
114 A. Resolution Capabilities
Only the comprehensive data set contains information on deeper parts of the subsurface
sections.
log10 Resistivity [Wm]
Difference [%]
(a)
Pseudo Depth [m]
(b) (c)
Pseudo Depth [m]
(d) (e)
(f) (g)
Distance [m] Distance [m]
Figure A.1: Simulated Wenner data example. (a) Wenner pseudosection computed from resistivity
model shown in Figure 4.1a (i.e., effective noise-free Wenner data set that would be recorded across
the model of Figure 4.1a). (b) Pseudosection computed from Wenner model shown in Figure 4.1c.
(c) Percent difference between (a) and (b). (d) Simulated Wenner pseudosection derived from
the dipole-dipole model shown in Figure 4.1d. (e) Percent difference between (a) and (d). (f)
Simulated Wenner pseudosection derived from the comprehensive model shown in Figure 4.1b. (g)
Percent difference between (a) and (f).
A. Resolution Capabilities 115
log10 Resistivity [Wm]
Difference [%]
(a)
Pseudo Depth [m]
(b) (c)
Pseudo Depth [m]
(d) (e)
(f) Distance [m] (g) Distance [m]
Figure A.2: Simulated dipole-dipole data example. (a) Dipole-dipole pseudosection computed from
resistivity model shown in Figure 4.1a (i.e., effective noise-free dipole-dipole data set that would
be recorded across the model of Figure 4.1a). (b) Pseudosection computed from dipole-dipole
model shown in Figure 4.1d. (c) Percent difference between (a) and (b). (d) Simulated dipole-
dipole pseudosection derived from the Wenner model shown in Figure 4.1c. (e) Percent difference
between (a) and (d). (f) Simulated dipole-dipole pseudosection derived from the comprehensive
model shown in Figure 4.1b. (g) Percent difference between (a) and (f).
Appendix B
ETH-DCMES System Set-Up
B.1 Field Set-Up

The following components are needed to set up the ETH-DCMES system for surveying:
• motor-generator;
• uninterruptible power supply (UPS);
• amplifier;
• 24 VDC power supply;
• n DAU’s;
• n four wire cables;
• n − 1 coaxial cables;
• n electrode cables;
• n metal stakes.
Figure B.1 shows a block diagram of the entire system set-up for surveying with the maximum
number of DAU’s possible (i.e., 124 DAU’s with the current hardware). It is crucial to
follow the steps for the field set-up exactly in the order given here !
This allows for maximum re-charging time of the DAU’s internal batteries before the
measuring sequence is initiated. The electrodes (metal stakes) should be deployed across
the area of interest in such a way, that groups of up to 31 DAU’s can be formed and
118 B. ETH-DCMES Set-Up
connected to one of the four field bus channels (Field Bus #1 - #4) of the interface box
(IFB). Connecting more than 31 DAU’s to one channel will demand too much power from
the corresponding RS-485 port and thus cause damage to the RS-485 interface card. The
maximum spacing between the electrodes is limited by the pre-configured cables (four-pole
and coaxial) to 10 m.
Once the electrodes are deployed in the field, DAU’s have to be snapped on to the metal
stakes (Figure B.2). The DAU’s should be arranged according to their ID’s (printed on the
DAU’s casing next to the standard input (Figure B.2)) in increasing order to keep the effort
for setting up an array configuration file at a minimum. Every DAU needs to be connected to
the following unit with a four wire cable (i.e., source signal and data bus) till the maximum
of 31 units are daisy-chained together. The first DAU in this chain has then to be connected
to one of the four field bus channel sockets denoted as X11 - X14 on the IFB (Figures B.1
and B.3). Arrays comprising more than 31 DAU’s have to be set-up in a similar way. The
remaining set-up should be done according to figure B.1 in the following order:
1. The 24 VDC power supply should be plugged into the IFB’s X10 jack and to one of
the UPS’ 220 VAC outlets (see Figure B.3 - from 24VDC Power Supply and blue lines
on Figure B.1). After attaching the UPS to a 220 VAC motor-generator or the mains
and starting the UPS by pressing the I-button, the LCD display on the IFB should now
either show the input voltage (24 - 28 VDC) or the current drawn by the DAU’s in µA
depending on the setting of the display switch. The DAU’s internal batteries are now
being recharged.
2. To enable arbitrary measurements between all electrodes, all DAU’s need to be inter-
connected with each other by a coaxial reference cable (see Figure B.2 and green lines
in Figure B.1).
3. Every DAU needs to be connected to its corresponding electrode via the standard
input for surface surveys using a red electrode cable (red lines in Figures B.1 and B.2).
If for some reason greater distances (>1 m) between electrodes and DAU’s have to
be covered, coaxial electrode cables should be employed and thus connected to the
optional coax inputs instead of the standard (Banana) inputs (Figure B.2).
4. The amplifier has two cables attached (input = thin cable with small connector, output
= thick cable with bigger connector), which need to be connected to the IFB’s X7 and
B. ETH-DCMES Set-Up 119
X8 jacks (see Figures B.1 and B.3 - to / from Amplifier - Source Signal).
5. Depending on the field bus channels (jacks X11 - X14 in Figures B.1 and B.3) in use,
appropriate RS-485 connections have to be made between the IFB (RS-485 jacks X1
- X4 correspond to field bus jacks X11 - X14) and the PC (Figure B.4 - to Interface
Box - RS-485 Channel 1 to 4 ).
6. Control signals are transferred between the PC (i.e., the built-in DAC-board) and
the IFB via a 50-wire shielded cable. 50-pin connectors on both ends should thus
be plugged into the appropriate jacks - see Figure B.3 from PC - DAC-Board and
Figure B.4 to Interface Board - DAC-Board. Care should be taken not to confuse
the DAC-board connector with the SCSI-port connector (see Figure B.4) !
7. The amplifier and the PC need to be connected to the UPS’ 220 VAC outlets (blue
lines in Figure B.1).
8. Finally, the amplifier’s front panel settings should be checked: MODE switch - CUR-
RENT ; VOLTAGE CONTROL switch - OFF ; CURRENT CONTROL switch - OFF
(see Figure B.5).
Starting-up the PC (main switch on the back of the PC-casing) and the amplifier by actuating
the POWER switch on the front panel are the last tasks before a measuring sequence can
be initiated.
In case the UPS will not be employed to power the system it is crucial to use
a multiple socket outlet which has the internal connection between the protected
earth contacts (i.e., yellow-green wire) removed !
B.2 Software Configuration

As the hardware of the ETH-DCMES system was designed to offer greatest possible flexibility,
also the software was written to offer a maximum degree of control of all relevant acquisition
parameters.
B.2.1 Array configuration
Prior to the field campaign, the user has to set up an array configuration file. This ASCII-
text file consists of a very simple header (line 1 and 2 in Table B.1) and the measurement
sequence. The first line sets the operation mode flag, where 0 stands for reference-type
operation and 1 represents standard four-point mode. The second line contains the number
Line No. Content Ref. Mode Four-Point Mode

1 op. mode flag 0 1
2 no. of meas. 100 100
3 1st config. 123 1234
4 2nd config. 234 2345
.. .. .. ..
. . . .
102 last config. 1 2 30 1 2 29 30
Table B.1: A typical array configuration file for reference and standard four-point mode.
of measurements which must be equal to the number of configuration lines to follow. From
line three onwards a single line represents one measurement with a certain configuration given
by the ID’s of the DAU’s employed. In case of the reference mode operation (operation mode
flag equal to 0) three numbers have to be given separated by a single space character, where
the first and the second one represent the injecting electrodes A and B respectively and the
third one is the ID of the DAU acting as reference electrode M (column 3 in Table B.1). In
this mode all remaining (n − 3) DAU’s in the array measure with respect to the reference-
DAU. Therefore, all (n − 3) DAU’s will send back the acquired data on completion of the
entire sequence. Classical four-point measurements are achieved by setting the operation
mode flag to 1 and by specifying the only measuring electrode N with a fourth ID in every
configuration line (column 4 in Table B.1).
Before the configuration file can be processed, several system parameters have to be
entered via a graphical user interface (GUI), which consists of a main panel (Figure B.6) and
two auxiliary panels (Figures B.7 and B.8).
After starting the ETH-DCMES application by double-clicking on its icon, the main panel
appears as shown in Figure B.6. All settings on the main panel are mandatory ! The
following descriptions of the GUI-panels are given in the order from top to bottom.
B.2.2 System preferences
By clicking the Set Preferences button on the main panel the System Preferences auxiliary
panel opens (Figure B.7). Here, the location for the acquired data and the system log file
(i.e., all system activities are recorded in a text file) have to be set by pushing the according
Set buttons on top of this window. Subsequently, a file select pop-up dialog will open where
the operator can navigate to the desired directory on the hard-disk.
Setting the ID’s of the connected DAU’s
Depending on the number of DAU’s and thus the number of RS-485 links employed, one or
more of the four radio buttons marked Port 1 - 4, which relate to field bus channels 1 - 4,
have to be activated. It then is very important to correctly set the lowest ID (i.e., commonly
the first ID on a particular field bus channel) and the highest ID (i.e., commonly the last ID
on a particular field bus channel) in the appropriate fields underneath the port buttons (refer
to Figure B.1). By clicking on the Test button, the system opens a terminal window and
displays a brief status report from every DAU it could establish a connection with.
At this stage, the operator can check if all DAU’s were properly connected together and
attached to the interface box. In case a field bus cable is missing between two DAU’s, the
status report of all following DAU’s will say that the devices are not connected. All missing
connections should then be tracked down before the test routine is initiated again.
Setting the ID’s of the DAU’s to be read back
After successful completion, the operator can now select the DAU’s which should transfer
data back to the PC after a measurement finished in the frame marked Read DAU on the
GUI (Figure B.7). Usually, those ranges of DAU’s will be equal to the ones set in the previous
step (i.e., the user wants to read back all data that were acquired). Finally, the Preferences
windows is closed through clicking the DONE button and the interface (and thus the DAU’s
connected) will be declared operational by the READY light on the main panel (Figure B.6).
B.2.3 Acquisition parameters
Back on the main panel (Figure B.6), the user has to set all relevant acquisition parameters.
Settings only relevant to four-point measurements
Only for the collection of a conventional resistivity surveys using four-point configurations
(i.e., operation mode flag set to 1in the configuration file), the operator has to set a battery
recovery interval in the field marked 4-Point Measurements Only. The time needed to transfer
the acquired data back to the PC is required to re-charge the DAU’s internal batteries. The
minimum charging time depends on the length of the logging window (i.e., during this time
the DAU’s work independently powered by their internal battery) and can be estimated by
the following rule of thumb: re-charge time [s] > 2.5 · logging window [s]. Unfortunately,
this criterion can not be fulfilled in the case of four-point measurements, when only a single
DAU transfers data back to the PC. By setting a Recharging Time of at least 15 minutes
after 150 four-point measurements have been completed, save operations can be ensured
assuming all batteries are in good condition and ambient temperatures are not too low.
With the Configuration switch, the operator can choose the way the array configuration
file is read: either the first two numbers of each configuration line are interpreted as current
electrodes (A B M N) then the switch should be set to Dipole-Dipole or the first and the
last number are represent the current electrodes (A M N B) then the switch should be set
to Wenner.
Setting the source signal
Differently to most commercial DC resistivity systems the source waveform can be altered
by creating arbitrary wavetrains that are subsequently amplified and fed into the ground. If
one of the pre-set signals (sine wave, square wave, pulsed DC or pure DC) is chosen, the
current amplitude (in Amps) and the frequency (in Hz) have to be set in the frame marked
Source Signal on the GUI (Figure B.6).
Setting parameters for active DAU’s
Parameters of the active DAU’s (i.e., those which are neither acting as injecting nor as
reference electrodes) have to be set before an automatic acquisition sequence can be initiated
(see first step in Figure B.9) in the frame marked Data Acquisition Unit (DAU). According
to the selected source signal frequency, the sampling frequency of the active DAU’s has to
be set to avoid aliasing effects (Nyquist theorem). Depending on the subsurface resistivity
distribution, the expected ambient noise-level and the electrode coupling resistances, the
amount of data to record should be entered in the field Buffer Size. Finally, the duration
of the so called Logging Window has to be entered. The entire preparation process of the
system and the data acquisition itself have to take place within this time frame while all
DAU’s are set on autonomous mode (i.e., the DAU’s are self-powered from their internal
batteries). To avoid timing conflicts, the logging window has to be chosen according to the
buffer length, the sampling frequency and some minimum overhead time (i.e., overhead time
>4 s)for preparations (Figure B.6).
Now that all settings are done, clicking the RUN button will open a file open pop-up
dialog where the operator can select the prepared configuration file. Subsequent loading the
configuration file triggers the measurement sequence. Progress and successful completion of
the entire acquisition sequence can be monitored on the terminal window.
B.2.4 Optional parameters
The system offers a number of optional settings, which can be accessed by clicking the
Options button on the main panel (Figure B.6). An option panel will open as shown in
Figure B.8.
Field tests have shown that the AGC radio button (Automatic Gain Control) should
always be set, otherwise if a fixed gain setting for the DAU’s is used some ADC’s might
be saturated and the data is clipped and therefore lost for processing. The switch Self
Potential disables the source signal transmitter if set to enabled, thus the user can estimate
the ambient noise level which will be recorded with the electrode configurations defined in
the configuration file. By default, this option is disabled.
Two more options can be set, which might give better results in some situations. Firstly,
the shielding of the coaxial reference cables can be connected to the system electronics
ground level by setting a relay in every DAU. Due to the high input impedance of the ADC’s
front ends, noise interference can become a severe problem especially in urban areas. For
this reason the reference cable shielding is pre-set to be connected to the system ground
potential and thus the according switch is disabled on the GUI (see Figure B.8).
Secondly, the ADC’s used in the DAU’s and in the interface box (i.e., to measure the
input current) comprise an input buffer stage. Activating this buffer increases the ADC’s
input impedance yielding lower measurable currents. However, this advantage comes at the
cost of a reduced measuring range (i.e., small gain values result in distorted acquired signals).
This feature is not recommended to be used in day-to-day field work as it might cause more
trouble than it will enhance signal quality. For this reason, both switches in the frame Data
Acquisition Unit (DAU) on the options GUI (Figure B.8) to activate the input buffer of the
current DAU in the interface box and all measuring DAU’s are set to Off by default.
B.3 Measurement Sequence and Data Preparation
B.3.1 Measurement sequence
Figure B.9 shows the entire measuring process as flow diagram. The input parameters and
the set-up of an array configuration file have been described in the previous section. After the
information from the GUI’s and the configuration file have been read in (second step from
the top in Figure B.9), the DAC and thus the source transmitter is initialized. Subsequently,
the internal switch-matrix of every DAU is set according to the desired array configuration
(i.e., as read from the configuration line). Thereafter all DAU’s are disconnected from the
24 VDC source (i.e., re-charging DAU’s internal batteries) and from the RS-485 data bus
for the pre-defined time frame given by the logging window (see previous section). Now the
DAU’s work autonomous, meaning they either inject current to the ground or they record
the user-defined amount of data without any data communication taking place. Short before
the logging window is elapsed, the source signal is disconnected from the field bus and with
the end of the logging time all DAU’s automatically re-connect to the field bus again.
Data is then transferred back to the PC from the current DAU (i.e., built into the IFB)
and one or all other measuring DAU’s (depending on the operation mode - see configuration
file set-up). The binary data is stored on the PC’s hard-disk at the user-defined location.
If more configurations are defined, the sequence is repeated accordingly. All details of the
triggering process of a measurement are given in the timing diagram shown in Figure B.10.
B.3.2 Data preparation
After the data has been dumped on the hard-disk is has to be converted into a format
which can be read by the ETH inversion code DCINV. The following Matlab script has to be
stored in the same directory where the data is located. The script is called by typing dcin-
vin(ID,LOW,HIGH) at the Matlab command prompt. The argument ID has to be replaced
with the lowest ID of the employed DAU’s, whereas LOW and HIGH denote the lower and
upper frequency boundaries in Hz for the noise estimation. The script will ask for a filename
for the output file and where the configuration file is located.
If this information is entered correctly, the script automatically creates an input-file for
the inversion code using the frequency domain method described earlier (see Chapter 2 for
details). Every line in this ASCII-file has the format A B M N v dv i, where A and B represent
the ID’s of the current injecting DAU’s, M and N represent the ID’s of the measuring DAU’s,
v is the recorded voltage value, dv is the estimated noise, and i is the corresponding current.
The corner frequencies for the noise estimation have to be set by the user according to
the employed sampling frequency and source signal frequency. For example, if the source
signal was a 1 Hz sine wave and the DAU’s sampled time series with 60 samples/s, the lower
noise corner could be set to 5 Hz and the higher one to 12 Hz.
B.4 Resetting ID’s

In case that a DAU fails during field operation it can be replaced by any other unit to continue
surveying. SetID, a simple application program allows to assign an arbitrary ID to any DAU.
Before using this program it very important to disconnect all DAU’s from the
IFB - otherwise, all DAU’s connected will get the same ID !
The ETH-DCMES operation program has to be shut down by clicking the QUIT button
on the main panel (Figure B.6). The DAU, which should get a new ID has to be connected
to one of the IFB’s field bus cannels X11 - X14 (Figure B.1). Now, the SetID application
should be started by double-clicking on its icon. Figure B.11 shows the simple GUI, where
only the port number (according to IFB-port used) and the new ID can be set. Pressing the
Set new ID button changes the DAU’s ID.
As the current DAU is built into the IFB, its ID was also changed to the new value !
Thus, after disconnecting the just changed DAU, the current DAU’s ID has to be
set back to 0 using the SetID application a second time.
B.5 DCINVIN Matlab Script

function dcinvin(firstid,lowf,hif);
%
% Computes voltage values from time series acquired with ETH-DCMES
% and generates input file for ETH inversion code DCINV.
%
% Arguments:
% firstid = smallest ID of DAU’s used
% lowf = lower corner frequency for noise estimation
% hif = upper corner frequency for noise estimation
%
cc = 0;
[outfile, outpath] = uiputfile(’*.dat’,’OUTPUT Filename’);

outfile1 = sprintf(’%s.dat’,outfile);
fid11 = fopen([outpath outfile1],’wt’);
outfile2 = sprintf(’%s_reject.dat’,outfile);
fid10 = fopen([outpath outfile2],’wt’);
conpathn = pwd;
[confilen, conpathn] = uigetfile(’*.dat’,’Electrode CONFIGURATION Filename’);
fid1 = fopen([conpathn confilen],’r’);
mode = fscanf(fid1,’%d’,1); % operation mode: 0 = reference mode

% 1 = four electrode mode
all_conf = fscanf(fid1,’%d’,1); % number of measurements

if mode == 0
config = fscanf(fid1,’%d %d %d’,[3 inf]);
else
config = fscanf(fid1,’%d %d %d %d’,[4 inf]);
end
config = config’;
fclose (fid1);
for nofile = 1 : all_conf
A = config (nofile,1,1);
B = config (nofile,2,1);
N = config (nofile,3,1);
if mode ~= 0
M = config (nofile,4,1);
binfilen = sprintf(’out%2d%2d%2d%d.bin’, A,B,N,M);
else
binfilen = sprintf(’out%2d%2d%2d.bin’, A,B,N);
end
binpathn = pwd;
fid = fopen([binpathn binfilen],’r’,’l’);
all = fread(fid,1,’unsigned short’);
buff_size = fread(fid,1,’unsigned short’);
sample_freq = fread(fid,1,’unsigned short’);
set_gain(:,1) = fread(fid,[all,1],’unsigned char’);
adc_buffer = fread(fid,1,’unsigned char’);
adc_shield = fread(fid,1,’unsigned char’);
adc_buffer_i = fread(fid,1,’unsigned char’);
form = fread(fid,1,’int’);
ampl = fread(fid,1,’double’);
frequ = fread(fid,1,’double’);
samples = fread(fid,1,’int’);
gain_ii = power(2,set_gain(1,1));
for a = 1:all
rawdata(:,a) = fread(fid,[buff_size,1],’double’);
if a ~= 1
gain_uu(a) = power(2,set_gain(a,1));
end
if a == 1
raw_i = rawdata(:,1);
mi = mean(raw_i);
corri = raw_i - mi;

fti = abs(fft (corri, buff_size));
ampspec_i = fti/(buff_size/2);
idx = find(ampspec_i(1:buff_size/2+1) > (0.4*max(ampspec_i)));
fi = ((idx-1) * smple_freq) / buff_size;
im = ampspec_i(idx);
if (im < 1e-6)
errmess = sprintf(’No current signal acquired, data set %s lost !’, binfilen);
disp (errmess);
cc = cc + 1;
fprintf (fid10,’%d %s no current signal\n’,cc, binfilen);
end
else
if (mode ~= 1)
ru = a + firstid - 2;
if (ru ~= A) & (ru ~= B) & (ru ~= N)
raw_u(:,a) = rawdata(:,a);
mu(:,a) = mean(raw_u(:,a));
corru(:,a) = raw_u(:,a) - mu(:,a);
ftu(:,a) = abs(fft (corru(:,a), buff_size));
ampspec_u(:,a) = ftu(:,a)/(buff_size/2);
vm = ampspec_u(idx,a);
low = mean(ampspec_u(2:idx-1,a));
start_noise = (lowf*buff_size/smple_freq)+1;
end_noise = (hif*buff_size/smple_freq)+1;
noise = mean(ampspec_u(start_noise:end_noise,a));
M = ru;
if low >= (0.70 * vm)

errmess1 = sprintf(’Remote unit no. %d of data set %s failed !’,a-1, binfilen);
disp (errmess1);
cc = cc + 1;
fprintf (fid10,’%d RU no. %d of data set %s failed - low = %f ...
u = %f\n’,cc, a-1, binfilen, low, vm);
else
fprintf (fid11,’%d %d %d %d %f %f %f\n’, A, B, N, M, vm, noise, im);
end
end
else
raw_u(:,a) = rawdata(:,a);
mu(:,a) = mean(raw_u(:,a));
corru(:,a) = raw_u(:,a) - mu(:,a);
ftu(:,a) = abs(fft (corru(:,a), buff_size));
ampspec_u(:,a) = ftu(:,a)/(buff_size/2);
vm = ampspec_u(idx,a);
low = mean(ampspec_u(2:idx-1,a));
start_noise = (lowf*buff_size/smple_freq)+1;
end_noise = (hif*buff_size/smple_freq)+1;
noise = mean(ampspec_u(start_noise:end_noise,a));
if low >= (0.70 * vm)

errmess1 = sprintf(’Remote unit no. %d of data set %s failed !’,a-1, binfilen);
disp (errmess1);
cc = cc + 1;
fprintf (fid10,’%d RU no. %d of data set %s failed - low = %f ...
u = %f\n’,cc, a-1, binfilen, low, vm);
else
fprintf (fid11,’%d %d %d %d %f %f %f\n’, A, B, N, M, vm, noise, im);
end
end
end
end
fclose(fid);
end
if (mode ~= 1)
fprintf (fid10,’\n%d recordings of total %d rejected.\n’, ...
cc, all_conf*(all-4));
disp(sprintf (’%d recordings of total %d for inversion.’, ...
(all_conf*(all-4))-cc, all_conf*(all-4)));
else
fprintf (fid10,’\n%d recordings of total %d rejected.\n’, ...
cc, all_conf);
disp(sprintf (’%d recordings of total %d for inversion.’, ...

all_conf-cc, all_conf));
end
fclose(fid10);
fclose(fid11);
}
24 VDC 220 VAC In
Power Supply
220 VAC
24 VDC Out
Out
X10 X7 In
Amplifier Uninterruptible
X6 Out Power
Supply (UPS)
Interface Box
X5 DAC-Board
Field Bus Field PC to 220 VAC
}
#1 Generator
X11 X1 #1
#2
X12 X2 #2 RS-485
#3 Ports
X13 X3 #3
#4
X14 X4 #4
Reference Cable
DAU DAU DAU DAU DAU
1 2 3 30 31
Electrode Cables
Electrodes
DAU DAU DAU DAU DAU

32 33 34 61 62
DAU DAU DAU DAU DAU

63 64 65 92 93
DAU DAU DAU DAU DAU

94 95 96 123 124
Figure B.1: Schematic block-diagram of the ETH-DCMES system showing all necessary cable
connections.
Electrode Cable
Reference Pot. Data Acquisition

(Coaxial Cable) Unit (DAU) Source Signal &
Data Bus
(4 Wire Cable)
Connector
Protection Caps
(plugged together Standard Input
for field use) (Surface Surveys)
Coax Input Electrode

(e.g., Borhole Surveys)
Figure B.2: A single DAU ready for field-use equipped with electrode (metal stake) and all
necessary cables.
to Amplifier from Amplifier

Source Signal Source Signal
from PC
RS-485 Channel 1
from PC
DAC-Board
Interface Box
from 24 VDC
Power Supply to DAU 1 - 31
Source Signal &
Display Data Bus
Switch Channel 1
24 VDC
Power Supply
Figure B.3: Interface Box (IFB) and 24 VDC power supply with cables attached to operate a
single field bus channel (max. 31 DAU’s). Red text denotes the location where the cable has to
be connected to; black text describes the function of the specific item.
to Interface Box
DAC-Board
to Interface Box
RS-485 Channel 1
RS-485 Port 1 - 4
SCSI Port
External Display Network Port

(VGA) (Ethernet)
Field Computer
220 VAC
Figure B.4: Side-view of the PC showing all interface connectors. Red text denotes the location
where the cable has to be connected to; black text describes the function of the specific item.
Remote
Manual Current Mode: Manual
Voltage Control: CURRENT Current Control:
OFF OFF
Main Switch
Figure B.5: Front panel of the source signal amplifier: Settings for constant current mode opera-
tion.
Figure B.6: Main panel of the ETH-DCMES graphical user interface (GUI).
Figure B.7: Auxiliary panel of the ETH-DCMES GUI to set the system preferences.
Figure B.8: Auxiliary panel of the ETH-DCMES GUI for optional settings.
set input parameters

source: data acquisition:
- waveform - sample frequ.
- frequency - buffer size
- amplitude - logging window
array configuration file:

read array configuration:
- ref. mode - A B M
- initialize source
- 4-pole mode - A B M N
- set acq. time
- set array configuration

- disconnect DAU's
(logging window)
record data
logging window no
terminated ?
yes
- re-connect to field-bus
- download data to PC
store
time series data
on PC harddisk
further config. yes

to measure?
no
transform time series to frequency domain &

evaluate apparent resistivity values
Figure B.9: Principal flow diagram of the controlling software.

prepare all DAU's (1)

< 500 ms 300 ms
DAU's disconn.
from 24 VDC (2)
start source signal

generation (3)
test: 24 VDC disconn. ? (4)
24 VDC discon. ? (5)

(DI0=0)
set switch matrix (6)

(A, B, N)
enable start (7)
(K9, K4 - K7) 100 ms 500 ms
disconn. RS-485 (8)

(DO3=0)
100 ms
set op. mode (int.) (9)
DAU's disconn. (10)

RS-485 (int.)
50 ms
(11)
initialize DAU's (int.)
20 ms
disconn. RS-485 (int.) (12)
DAU's autonomous ? (DI1=0) (13)

>5s
source signal on (14)

field-bus (DO1=1)
acquire data (15)
RECORDING
battery test (int.) (16)
100 ms
logging window < 20 s
int. = internal DAU command
Figure B.10: Timing diagram of a measuring sequence.

Figure B.11: GUI of the tool for setting a new ID of a DAU.

Appendix C
Technical Details of the ETH-DCMES

System
This chapter will provide an overview of the technical specifications of the DC-resistivity
measurement system developed by the ETH Institute of Geophysics (ETH-DCMES).
C.1 Data Acquisition Unit (DAU)
The core part of the DAU is its ADC. All other components are only needed to support the
ADC. We decided to employ the so called sigma-delta type of analogue-to-digital conversion
principle due to the high resolution capabilities at affordable price. The sigma-delta ADC
architecture is capable of much higher resolution than most other conversion concepts. The
typical range is 16 to 24 bits. The tradeoff comes in the form of slower sampling speeds, which
for our application is no disadvantage. Sampling speeds are typically limited to approximately
100 ks/s.
Sigma-delta converters use a technique called oversampling that is based on the theory
that a very fast, low-resolution converter can be used to take a large number of samples
on an analog signal. The outputs from this oversampling process are then combined into
groups, and the groups are averaged using a digital filter like an accumulator-adder. This
averaging operation can increase the accuracy of the conversion as long as the input signal
does not vary any faster than the sampling rate. This process also eliminates any in-band
noise present in the ADC by spreading the noise across the entire sampling frequency band.
Figure C.1 shows the wiring diagram of the digital section of a DAU, which comprises the
controller (CPU), 32 kbyte memory (RAM), clock signal generator and the RS-485 interface.
142 C. ETH-DCMES Technical Details
Figure C.2 shows the wiring diagram of the analogue section of a DAU, which comprises the
24 bit sigma-delta ADC, a precision voltage reference, an automatic charging circuit for the
internal battery and the front end input switching matrix. Figure C.3 shows the connection
scheme for the analogue and the digital section of the DAU.
C.2 Interface Box (IFB)

The interface box provides the electrical link between the field cabling (i.e., the daisy-chained
DAU’s) the source signal transmitter and the controlling PC. The unit routes the source signal
to the four field bus channels and switches the current signal on and off controlled by the
PC. A built in DAU measures the injected current wavetrain via the voltage drop across a
precision resistor (Figure C.4). The second most important task of the IFB is to monitor
activity on the RS-485 buses and to provide synchronized activation / de-activation of the
digital buses Figure C.5. Additionally, this part of the circuit checks that the source signal
is not routed through to the field buses while the DAU’s are in re-charging mode.
1 2 3 4
VCC
D2 74HC573
RR1 1 D3 32K*8SRAM
GND OC
8x33k 11 19 A0 10 11 AD0
C 1Q A0 DQ0
D1 87C520 18 A1 9 12 AD1
2Q A1 DQ1
1 39 AD0 AD0 2 17 A2 8 13 AD2
D DATA_AD P10/T P00 1D 3Q A2 DQ2 D
2 38 AD1 AD1 3 16 A3 7 15 AD3
SCLK_AD P11/T P01 2D 4Q A3 DQ3
3 37 AD2 AD2 4 15 A4 6 16 AD4
/SYNC P12 P02 3D 5Q A4 DQ4
/TXENA 4 36 AD3 AD3 5 14 A5 5 17 AD5
P13 P03 4D 6Q A5 DQ5
5 35 AD4 AD4 6 13 A6 4 18 AD6
P14 P04 5D 7Q A6 DQ6
D7 6 34 AD5 AD5 7 12 A7 3 19 AD7
P15 P05 6D 8Q A7 DQ7
15 13 7 33 AD6 AD6 8 A8 25
BATT0 QA G P16 P06 7D A8
1 12 8 32 AD7 AD7 9 A9 24
BATT1 QB RCK P17 P07 8D A9
2 VCC A10 21
BATT2 QC A10
3 10 13 21 A8 A11 23
BATT3 QD SRCLR /DRDY INT1 P20 A11
4 11 RxD 12 22 A9 A12 2 27 /WR
C. ETH-DCMES Technical Details
generator, and RS-485 interface.

BUF QE SRCK INT0 P21 A12 /WE
5 23 A10 A13 26 22 /RD
P_HLD QF P22 A13 /OE
6 SDA 15 24 A11 A14 1 20
K1A QG T1 P23 A14 /CE GND
7 14 SLC 14 25 A12
K1B QH SER T0 P24
9 26 A13
QH1 P25
31 27 A14
VCC EA/VP P26
74HC595 28 A15 VCC VCC
P27 VCC VCC
19
X1
XTAL1 14.7456 18
X2
C C19 C10 C
RESET 9 10 RxD C20 C11
RESET RXD
11 TxD 100n
C2 33p C1 33p TXD 10u/25V 100n
/RD 17 30 ALE 10u/25V
RD ALE/P
D6 /WR 16 29
WR PSEN
15 13 GND GND
K2A QA G GND GND
1 12 GND GND
K2B QB RCK
2 VCC
K3A QC
3 10
K3B QD SRCLR
4 11
K4A QE SRCK
5 D4 VCC
K4B QF
6 7 5 SDA
REL_S QG GND (/WC) SDA
7 14 3 6 SLC
REL_L QH SER GND A2 SCL
9 2
QH1 GND A2
1 R1
GND A1
74HC595 2k2
X24C02
D5
rt TxD isable 5
GND GND
VCC V1
B B
VCC VCC VCC VCC N5 LED1 TxD 4
D
VCC 6
VCC A RxTx+
3
DE
C17 C7 RST RESET
C6 RST
C18
10u/25V 100n 100n GND 2
10u/25V GND GND R5 T1 GND /RE
/TXENA 7
BC337 B RxTx-
DS1812 10k RxD 1
R
GND GND GND GND
8
VCC VCC
LTC1487
GND
VCC VCC VCC VCC
C16
10u/25V C9 C8 C15 Title
A A
100n 100n 100n
Size Number Revision
GND GND GND GND REV 01
A4
Date: 25-Sep-2000
File:
1 2 3 4
Figure C.1: Digital section of a data acquisition unit (DAU) comprising CPU, RAM, clock signal
143
144
1 2 3 4
1N5819
C30
V6
N4
D GND D
7 Source A N2 LM2931Z5.0
1 8 F2 LM2931CT 47uF/50V T2
X1 V3 R19 V5 BC327 V4
1 9 4 5
X2 VIN OUT IN OUT VCC
1 4 Source B 1N4007 22E 1N5819
X3 1N5819
GND
1 3 PTC C850
X4 F1 C29 R6
2 1 C25 C26
ADJ
GND
ON/OFF
ZNR 47uF/50V U-BATT R10 100n
TF2 10k 10u/25V
C31
3
2
t 10k 47u/50V
10 + 1 R11 R7 R8
1
K5 VCC 10k 10k 10k
BATTERY X14
GND GND GND GND GND GND
8,4V/150mAh
GND GND
V7 T4
1N4148 R9
T3 BC337
REL_S BC337 P_HLD
1 10k
X5 X15
1 7 V9
1
X6 RxTx+
1 8 5V6 VCC VCC
X7
9
4 I-BATT GND C13 100n C12 100n
C RxTx- GND GND C
1 3 GND
X8 R23 R25 R24
2
BATT0 1E 1E
68k C22 C23
R16 GND GND
TF2 BATT1 10u/25V
10 + 1 R17 10u/25V
VCC 100k R21 GND
K6
BATT2 100K 100K
330k GND GND R18
12
23
R22 N1
V8 BATT3
1N4148 470k
2 11
REL_L CLKIN STBY VCC
XTAL2 2.4576
AVDD
DVDD
U-BATT 3 21
R12 R13 CLKOUT DOUT
VCC 10k 10k R20 A_IN_1 7 22
1M C4 33p C3 33p AIN 1 DIN DATA_AD
Reference Reference 8 20
R2 100n AIN 2 DRDY /DRDY
1 Reference C14
X9 GND
1 2k2 GND GND GND GND 9 1
X13 AIN 3 SCLK SCLK_AD
1 1 C5
B R15 R4 B
1 5 5 33p C24 I-BATT 10 5
X10 R3 GND 100k AIN 4 SYNC /SYNC
10 10 A_IN_1 2k2
10u/25V 16 13
2k2 AIN 5 BUF BUF
2
3 8 3 8 N3 17 4
K1B GND AIN 6 POL VCC
REL_P1C1 REL_P1C1 GND
5 6 15 6
TRIM V_OUT VREF+ RES VCC
+VIN
K2 K1 K1A
8 3 14 19
OPSEL TEMP VREF- CS GND
K2B
GND
DGND
AGND
C27 C21 C28

K2A 100n 100n
AD780 AD7714
4
24
18
1 10u/25V
X11
1 1 Source A
voltage reference, automatic battery charging circuit, and input switch matrix.
1 5 5
X12
10 10 Source B GND GND GND GND GND GND
GND
Elect r ode R14
3 8 3 8 20k Title
A K3B A
100k P1
REL_P1C1
K4 K3 Adj . Ref . : +/- 20mV Size Number Revision
K3A
A4 REV 01
REL_P1C1 K4B
GND Date: 25-Sep-2000
K4A
File:
1 2 3 4
Figure C.2: Analogue section of a data acquisition unit (DAU) comprising 24 bit ADC, precision
1 2 3 4
DI GI TAL
D D
RxTx+
RxTx-
P_HLD
/DRDY
REL_S
REL_L
SCLK_AD
DATA_AD
BATT0
BATT1
BATT2
BATT3
/SYNC
BUF
C C
ANALOG
/DRDY
RxTx+
RxTx-
REL_L
REL_S
SCLK_AD
DATA_AD
BATT0
BATT1
BATT2
BATT3
P_HLD
/SYNC
BUF
B B
Figure C.3: Connection scheme of a data acquisition unit (DAU).

Title
A A

A4 REV 01
Date: 25-Sep-2000
File:
1 2 3 4
145
146
1 2 3 4
A D C -signal f rom Source

to A mp.
X8
GND
Sig.
X7
D D
1
2
2
1
Sig.
GND
Source B
Source A
2
1
DAC0OUT
X24
X21 X23 Source A DAU
1
X5 X6 GeneratorON Source B
DO1 1 2 X9/X13 Reference
50pol Enable Start
1 1 DO2 2 R1 X10 Shield
Trigger
2 2 DO3 3 X22 1E0
Source GND +
3 3 DO4 4 BAT 1 X14
A D -Bd. Supply-Isol-Test
4 4 DO5 5 2
/Supply-Isol /Source (K1)
DI 0 6 3 X11/X12 Elect r ode
/RS485-Isol
DI 1 7 4
48 48 8 +BAT 5 X1/X5 Source A / +24V Bat t . 9 V
-
49 49 9 V? 6 X2/X6 Source B / GND X15
PC_GND (ws) +24V L1
50 50 PC_GND 10 7
RxTx+1
C 1N4148 8 X3/X7 RxTx+ C
RxTx-1
X4/X8 RxTx-
RxTx+1 X26 X25 RxTx+1
Source B
Source A
AD-Board 1 1 LED
RxTx-1 RxTx-1
DAU to sample the injected current wavetrain.

2 2
RxTx+2 RxTx+2 Source
3 3
RxTx-2 RxTx-2
4 4
RxTx+3 RxTx+3 X9
5 5 Contr.
RxTx-3 RxTx-3 +24V
6 6 1
RxTx+4 RxTx+4 Source / +24 V
7 7 2
RxTx-4 RxTx-4 GND
8 8 Exp.-Box
Moni to r i ng
2
9
8
7
5
3
4
X1 X2 X3 X4
R310k K1
9 9 9 9 R2
6
1
10
11
8 8 8 8
7 7 7 7 0E01
RS485 X15
6 6 6 6 REL_3UK
4 DMM 939 5 3 Source A / +24V
B 5 5 5 5 1 B
BUS 1 - 1 Source B / GND
4 4 4 4 V 2
BUS 4 Exp.-Box
3 3 3 3 3
A 8
2 2 2 2 4
15 16 2 7
1 1 1 1
EEG.3k
RxTx+1
RxTx-1
RxTx+2
RxTx-2
RxTx+3
RxTx-3
RxTx+4
RxTx-4
GND+11V
+24V
GND
RxTx-1
RxTx+1
RxTx-2
RxTx+2
RxTx-3
RxTx+3
RxTx-4
RxTx+4
D-Sub 9pol (w) ON = [A] Ref . Pot.
1
2
4
3
2
1
4
3
2
1
4
3
2
1
4
3
2
1
X10 X11 X12 X13 X14

EEG.3k EEG.3k EEG.3k EEG.3k
Supply DAU Bus 1 DAU Bus 2 DAU Bus 3 DAU Bus 4

24 V olt
Title
A A

A4 REV 01
Date: 25-Aug-2000
File:
1 2 3 4
Figure C.4: Main board of the interface box connecting the field buses to the PC comprising a
C. ETH-DCMES Technical Details 147
1 2 3 4
Generator
H SourceA H
SourceB Gnd Signal
2
1
X24
+BAT
Source
SourceA
SourceB
Source
X21 PIN6
D3
Supply-Isol 4 1 V6
1
LED2 1N4148
DI0 3 2 R17
TLP124 2k2 +BAT

LED gn Suppl y Moni to r i ng
Open
PC +BAT
8
+24V Moni to ri ng BAT N2A
3
+BAT 1
G G
2
R15 R16
Test
2k2 2k2 R20 LM358
4
3 470k
4 > 1W
BAT
5
N2B
10 5
1 mA
X21 PIN5 9 7
Supply-Isol-Test R13 8 R14 6
1
1k K3 2k2
DO5 R6 V8 LM358
7 1 D2 10
10k 4 12 1 1N4004
- + +BAT R21 R4
8 2 3 9 500E 10k R18
TLP127 REL_TX2 V14 10M

+BAT 1N4004
V4
PC BAT
TX2-24V
F 1N4148 F
BAT BAT BAT
10
K2
1
3
+
V5 Open t o K1
Generator
1N4148 Test X23
-
SourceA SourceA or Bat +24V

12
1
X21 PIN4
R12 2
SourceB SourceB or Bat GND
Sour ce 1
1k
DO0 R5
10k 5 1 D2 12
4
+BAT
6 2 3 11 X22
TLP127 +12V +12V RS485-BUS Moni to r i ng V15 8 +24 BAT
7
PC BAT +12V +12V
V16 1N4004 6
X21 PIN3
R9
R24
22k
R22
22k 5 Battery
Trigger 1 1N4004 4
1k V10
3 & K1
8
DO3 R3 N1A 1N4004

10k D1 2
E 1 4 StopBit = 1 = moni to r 3 F1 PTC C980 80V 1 GND BAT E
1
2 3 2
BAT
TLP124 R23 LM358 V9
4
PC GND12 22k 1N4004

+12V
GND12 -12V -12V

X21 PIN7 +12V
Low = OK! LED1
RS485 Isol. 1 TX2-24V RS485 BUS
DI1 R25
10
K10
3
LED gn 1M
+
V1
8
1N4148 R8 N1A
-
10k 3 R28 I Test = 5µA

12
4
1 47k
2 V11
R27 C1 V13 1N4004 R29
D D
PTC C980 80V
PC 47k LM358 100k

4
470n F2
R26
T1 47k 1N4004 V12
PC I/O Board BC337 R19
1N4004
R7
10M
10k GND12 GND12 GND12 GND12 GND12
-12V +BAT
V3
+BAT -12V
1N4148
K9 K7 K6 K5 K4
10
1
3
+
V7 1 + - 12 1 + - 12 1 + - 12 1 + - 12
X21 PIN2
1N4148
-
R10 TX2-24V TX2-24V TX2-24V TX2-24V

12
C Enable Start 1 C
1k TX2-24V 8 8 8 8
DO2 R1 9 9 9 9
10k 3 1 D2 4 14 10 10 10 10
GND
4 2 3 13 5 5 5 5
+BAT 4 4 4 4
TLP127
3 3 3 3
PC BAT
X25
K8
10
1
1
Bus1
2
+
V2
3
X21 PIN1 Bus2
1N4148 4
-
R11
DAU Bus
12
Generator ON 1 5
1k TX2-24V Bus3
R2 6
DO1
10k D2 16 7
1 4 Bus4
8
BAT
2 3 15
B B
TLP127 X26
1
X21 PIN10 PC BAT Bus1
2
GND 1 3
Bus2
4
PC +BAT Bus3
5 PC RS485
6
DC1
7
22 14 Bus4
+VIN +VOUT +12V 8
23
+VIN C5
C3
100n 10u/25V
16
COM
C4 GND12
100n 9
COM
C6
C2
2 100n
-VIN 10u/25V
3 11
-VIN -VOUT -12V
A Title A
TEN 4-4822
BAT
A3 REV 01
Date: 25-Aug-2000
File:
1 2 3 4
Figure C.5: Monitoring circuit of the central interface box.

Appendix D
Real-Time Experimental Design:

Application to 3D High-Resolution DC
Resistivity and IP Surveys
Peter Stummer, Hansruedi Maurer and David E. Boerner
Proceedings of the 69th Ann. Mtg. of the

Society of Exploration Geophysicists (SEG),
Houston, Texas, 1999
D.1 Abstract
Enhancing the information content expected of geophysical surveys usually comes at the
expense of increasing data acquisition costs. In order to optimize the balance between in-
formation and costs, statistical experimental design techniques can be applied. A primary
limitation of this approach results from the non-linear relationship between geophysical data
and the earth model under investigation. It is generally impossible to predict a data acqui-
sition strategy without any a priori knowledge of the subsurface structure. Therefore, we
propose an extension to experimental design that involves updating the trial model contin-
uously as data are acquired (i.e., experimental design in real-time). To illustrate this idea,
150 D. Ext. Abstract SEG 1999
we consider the implications of real-time experimental design for direct current/induced po-
larization (DC/IP) surveying of a three-dimensional subsurface structure using large-scale
electrode arrays. The experimental procedure consists of (i) acquiring an initial data set,
(ii) inversion for an earth model, (iii) selecting an improved experimental design and iter-
ating steps (i) to (iii) until the resolution of suburface structure no longer improves. Since
measurements with some electrode configurations that were not necessarily predicted before
the experiment began have to be realized quickly, implementation of real-time experimental
design demands extremely rapid data acquisition capabilities and flexibility. Ensuring suffi-
cient flexibility requires a fully digital multi-electrode system in which each electrode could
be a current source/sink or a voltage reference (to define the potential difference). Our
new digital system design also calls for parallel data acquisition in order to scan efficiently a
number of electrode configurations.
D.2 Introduction
Delineation of complex three-dimensional (3D) structures using geophysical techniques is a

challenging task, restricted primarily by the physics of the method. Additional limitations
on resolution are imposed by the design of the geophysical experiment, a key example being
the usual restriction of having sources on or above the air/earth interface. Cost is another
constraint, limiting the quantity of data available to interpret. Achieving an acceptable
balance between information content and costs requires an appreciation of how acquiring
incremental new data might benefit our understanding of the earth.
Recently, attempts have been made to optimize survey procedures by the application
of statistical experimental design (Maurer and Boerner, 1998). With this technique, it
is possible to identify survey layouts that provide maximum information for a particular
subsurface model. When the relationships between models and data are non-linear, the
choice of an optimized survey layout is necessarily model dependent. If substantial a priori
information about the earth exists, experimental design is likely to be highly beneficial in
guiding field practices. However, if the subsurface to be investigated is entirely unknown,
statistical experimental design procedures, as proposed by Maurer and Boerner (1998), are
limited and therefore need to be extended.
Some knowledge of the subsurface is required before designing an optimized survey layout.
D. Ext. Abstract SEG 1999 151
Figure D.1: Basic steps in real-time experimental design showing the iterative nature of the
process.
On the other hand, acquiring subsurface information is also the objective of the survey to
be conducted. One solution to this “chicken-and-egg”problem is to couple the processes of
data acquisition and data analysis and implement a real-time experimental design strategy.
This is outlined schematically in Figure D.1. An initial data set is recorded with a standard
configuration based on experience gained from previous experiments. Inversion of the initial
data set provides a first estimate of the subsurface structure. Statistical experimental design
based on this estimated model is then applied to identify regions in the data space that
might contribute to increased model resolution and/or decreased model variance. Collection
of the additional data and their incorporation into the inversion process should allow model
parameter estimates to improve. The cycle “data acquisition - data analysis - experimental
design” should be repeated until the relevant regions of the data space are fully characterized
or until the costs become prohibitive.
Here, we discuss application of real-time experimental design to direct current (DC)

resistivity and induced polarization (IP) data. We present an acquisition strategy amenable
to solve three-dimensional DC/IP problems. An important pre-requisite of the proposed
approach includes acquisition hardware that not only allows rapid measurements, but also
ensures a high degree of flexibility. This motivated the development of a new, sophisticated
multi-electrode array that is capable of performing paralleled scans of many electrodes and
automatically selectable measurements with arbitrary four-point configurations.
D.3 Designing DC Experiments
Schlumberger arrays are generally considered the most suitable configuration for vertical
electrical sounding of a quasi-layered earth (Oldenburg, 1978), whereas the Wenner and
dipole-dipole configurations are commonly employed for mapping lateral variations in elec-
trical resistivity (e.g., Telford et al., 1990). For delineating 3D structures, pole-pole arrays
seem to be the preferred choice, supported by the results of several synthetic studies (e.g.,
Loke and Barker, 1996a).
To outline the problem of survey optimization and to demonstrate the need for real-
time experimental design, a simple synthetic 1D experiment was conducted. Two simple
resitivity-depth functions, shown in Figures D.2a and D.2c, were chosen as hypothetical
subsurface structures. The experimental design problem includes identification of electrode
configurations that optimally resolve these models. For the sake of simplicity, we have
restricted ourselves to only Schlumberger configurations. Hence, we use the design solely
to select the distance from the current electrodes to the midpoint of the array. Calculated
apparent resistivity responses for the different models are shown in Figures D.2b and D.2d.
Most of the subsurface information is expected to be obtained from data measured in the
curved parts of the apparent resistivity functions (delimited by vertical lines in Figures D.2b
and D.2d). Since the two models in Figures D.2a and D.2c are substantally different, it
is not surprising that the distance ranges where most of the information is expected to lie
are also quite different. Designing an optimized layout for an assumed model as shown in
Figure D.2a, but measuring over an earth such as shown in Figure D.2c, would undoubtedly
lead to a severe lack of information.
The models shown in Figures D.2a and D.2c are perhaps the simplest earth models
possible. Furthermore, by restricting the array geometry to Schlumberger configurations, we
have greatly simplified the data space (i.e. the range of possible experimental designs). For
this intentionally idealized example, the experimental design problem is trivial and could be
solved easily by measuring over a sufficiently large distance range, i.e. by considering the
entire data space.
In more realistic geological scenarios, such an approach would be difficult to follow.
Particularly the shallow subsurface is expected to contain complicated 3D structures with
pronounced resistivity contrasts. Constraining an appropriate 3D resistivity model requires
the electrodes to be distributed over the entire area of interest, which results in a large
number of electrode locations. Measuring all possible configurations of current/potential

electrodes, placed on a grid, would result in a huge amount of data. Such a data set would
be very time-consuming to acquire, require massive computing power for data analysis and
include a large amount of data that provide only little subsurface information.
5 2
10
Resistivity [Ohm m]
10
Depth [m]
15
20
1
25 10
a) b)
30 −1 0 1 2 3 4 5
0 1 2 3 4 10 10 10 10 10 10 10
Log(Resistivity) [Ohm m] Log(Distance) [m]
20 2
10
Resistivity [Ohm m]
40
Depth [m]
60
80
1
100 10
c) d)
120 −1 0 1 2 3 4 5
0 1 2 3 10 10 10 10 10 10 10
Log(Resistivity) [Ohm m] Log(Distance) [m]
Figure D.2: Comparison of Schlumberger sounding results for two synthetic models. (a) K-type
model: A thin, resistive layer in homogenous half-space. (b) Calculated sounding curve for K-
type model. (b) A conductive layer in homogenous half-space. (d) Calculated sounding curve for
conductive layer model.
Consequently, attempts should be made to identify and measure only those electrode
configurations that provide most subsurface information. This is a task that can be readily
solved with real-time experimental design techniques.
D.4 Real-time Experimental Design for

Three-Dimensional DC/IP Problems
Following the general scheme of real-time experimental design (Figure D.1), an initial data
set has to be recorded with a standard electrode configuration. Pole-pole configurations have
been selected as reasonable method for acquiring the initial data set. 3D-Inversions of the
pole-pole data provide an acceptable first estimate of the resistivity model (Loke and Barker,
1996a).
Theoretically, pole-pole data can be formed into linear combinations that simulate the
measurements obtainable from any four-point array configurations. Due to the fixed dynamic
range of realizable acquisition hardware and due to the presence of noise, there is a limited
utility for such mathematical constructs (Beblo, 1997). Further limitations are introduced
by the finite distance between the reference electrodes and the grid (current and potential
reference electrodes should be theoretically at infinity). Explicit measurements of some four-
point array configurations are thus eminently desirable.
On the basis of the initial model estimate, suitable four-point configurations have to
be identified that may contribute to an increase of model resolution and/or decrease of
model variance. It should be noted that definition of “suitable four-point configurations”can
be easily expanded beyond the limits of model resolution to include, for example, data
acquisition costs as a further factor to be optimized.
Explicit measurements of selected four-point configurations and subsequent inversions
with the extended data set provide an improved 3D resistivity model. If the updated model
differs significantly from the initial estimate, other four-point configuration may become
relevant to constrain the resistivity structure. This would require further “data acquisition -
data analysis - experimental design”cycles until the model updates become sufficiently small.
D.5 A New Digital Multi-Electrode Array
Implementation of the real-time experimental design concept demands a sophisticated multi-

electrode system. The following requirements are based on the concept that a three-
dimensional DC/IP survey would involve placing electrodes in a grid over the area to be
surveyed:
• parallel data acquisition to perform fast scans of many different electrode arrays simul-
taneously;
• software-controlled selection of arbitrary four-point configurations; for any given array,

an electrode in the grid could be a current source/sink or voltage reference;
• sufficient computing power to perform rudimentary data analysis and simplified statis-
tical experimental design.
Figure D.3: Schematic layout of a fully distributed digital DC/IP data acquisition system.
Other important features for a flexible and versatile system would be a large dynamic
range, modularity and digital recordings of the full-waveform data (for IP measurements).
Since none of the commercially available multi-electrode systems fulfilled all of these needs,
we have developed a new acquisition system. Extensive examination of various system de-
signs led us to the development of a fully distributed system, such that each electrode is
equipped with its own remote digitization unit (RU) (Figure D.3). This concept offers several
advantages. It allows
• each electrode to be configured either as a current source/sink or voltage reference

unit,
• measured voltages digitized directly at the electrode to minimize noise pick-up of the
system due to induction in long cables (this feature allows full characterization of small
amplitude IP effects),
• digital data transfered rather than low level analogue signals (in fact, there is only one
analogue reference potential that needs to be distributed to each of the RUs),
• high redundancy within the system (if one RU fails only a single electrode position
within the entire grid is lost),
• simultaneous data acquisition from all RUs (to the limit allowed by the bandwidth of
the data bus),
• high degree of modularity (the number of RUs to be employed can be selected arbi-
trarily).
Each RU includes a pre-amplifier, a 24-bit A/D converter and a communication interface

to interact with the host PC (Figure D.3). An interface box routes the daisy-chained RUs
to a controlling PC and the transmitter unit. The PC controls remotely the RUs and runs
the data analysis and experimental design software.
For a transmitter unit, we have chosen a high-end amplifier that is capable of converting
arbitrary digital waveform data generated at the PC to analog signals. Having complete
control on the current waveform at the transmitter facilitates the acquisition of IP data in
either the time or frequency domains.
D.6 Conclusions
The introduction of real-time experimental design techniques has resulted in a novel approach
for collecting geophysical data. Partial data sets may be examined in the field with the intent
of improving experimental design before the experiment is completed. The overall goal is to
reduce the requirement for acquisition of data that are of marginal importance to understand-
ing the subsurface. Meeting the goal of adaptive survey design without involving additional
costs to the field program has prompted the development of a new multi-electrode system for
high-resolution DC resistivity and IP data acquisition. This system should set new standards
in terms of data accuracy and flexibility, and is expected to lead to a significant increase of
experimental efficiency by limiting the amount of data that need to be acquired. With a
multi-electrode array deployed in the field and an iterative acquisition-analysis-experimental
design cycle, our aim is to have a near optimal three-dimensional subsurface image at the
end of the day.
Appendix E
Real-Time Experimental Design

Applied to High-Resolution
Direct-Current Resistivity Surveys
Peter Stummer and Hansruedi Maurer
Proceedings of the 46th Ann. Mtg. of the

International Society for Optical Engineering (SPIE),
San Diego, California, 2001
E.1 Abstract
Designing optimized survey layouts for DC resistivity data is a difficult task. This is primarily
due to the non-linear relationship between observed data and the subsurface resistivity struc-
ture. A possible solution is provided by real-time experimental design, a novel approach for
acquiring geophysical data. Technical realization of real-time experimental design requires
versatile recording facilities. We have developed a new multi-electrode system, that offers
the necessary flexibility. Furthermore, our concept of distributed acquisition units with a
digital layout enables the recording of high quality data.
160 E. Ext. Abstract SPIE 2001
E.2 Introduction
Direct-current (DC) resistivity methods are widely used for investigating the shallow sub-
surface in a variety of environmental, hydro geological and civil engineering applications.
Resistivity contrasts between any subsurface targets and the surrounding host sediments can
be very large. Compared to other physical parameters, such as seismic velocities, densities
and relative permittivities, which may vary from a few percent to a factor of 10, electrical
resistivities of rocks extend over several decades (e.g., Keller and Frischknecht, 1966). This
makes electrical prospecting a very powerful investigative technique, but it also introduces
a high degree of non-linearity in the relationship between observed data and the subsurface
resisistivities.
The most important expression of non-linearity in DC resistivity investigations is the
dependency of current flow paths on the resistivity values. This results in data analysis
problems and complications in the survey design. For example, an electrode layout suitable
for delineating sub-horizontal boundaries may be inadequate for resolving steeply dipping
features. One solution would be to deploy a very large number of closely spaced electrodes
and to measure all possible combinations of voltage and current electrodes. Since there
exist numerous combinations, such an approach for many situations is even for a small total
number of electrodes either not desirable nor feasible from an economical point of view.
Our intention is to acquire a maximum of subsurface information with a minimum of
measurements (i.e. with minimal costs). To consider the non-linearity of the DC resistivity
method, we have developed real-time experimental design, a novel data acquisition approach.
Technical realization of this new concept requires a versatile recording system. We have
developed a new multi-electrode system, that meets these requirements. It includes other
critical features such as a parallel measurement mode and digital data transmission.
E.3 Real-Time Experimental Design
Most applications of the DC resistivity method employ such standard electrode configura-
tions as the Wenner, Schlumberger, dipole-dipole or pole-pole configurations (e.g., Telford
et al., 1990). These configurations were introduced many years ago, when DC resistivity
measurements were performed manually with four-electrode systems. With the introduction
of multi-electrode systems, arbitrary configurations may be easily implemented. Neverthe-
E. Ext. Abstract SPIE 2001 161
less, one or more of the standard configurations are still employed in virtually all studies
published in the geophysical literature.
One reason for the ongoing popularity of the standard configurations is the difficulty of
identifying alternative options that will supply significantly superior results. This problem
has been addressed by Maurer and Boerner (1998), who introduced statistical experimental
design for geoelectrical methods. In their approach, optimized survey layouts may be delin-
eated with global optimization methods. An application of statistical experimental design to
DC resistivity surveys can be found in Maurer et al. (2000).
A major shortcoming of the statistical experimental design technique described by Maurer
and Boerner (1998) is the underlying assumption that sufficient a priori information about the
earth exists prior to a survey. If there is considerable uncertainty about subsurface structures
and the non-linearity of the problem is severe, standard statistical experimental design will
likely fail.
Figure E.1: Basic steps in real-time experimental design showing the iterative nature of the process.
An alternate approach would involve integrating the data acquisition, analysis and inter-
pretation into the real-time, iterative process outlined in Figure E.1. Estimates of the loca-
tion(s) and approximate extent(s) of the subsurface target(s) and the number of electrodes
available are the only input parameters required to initiate this process. The investigation
area would be either crossed by a number of profiles or covered with a regular grid of elec-
trodes. In a first step, a limited number of measurements would be performed. Pole-pole
recording would be the obvious choice for the initial measurements, since this type of data
would provide good subsurface coverage in noise-free environments (e.g., Loke and Barker,
1996a).
Preliminary analysis of the initial data would outline areas with conspicuous anomalies.
Using arbitrary electrode configurations, additional measurements across the anomalous re-
gions would allow the causative bodies and their electrical properties to be delineated more
accurately. Alternatively, if sufficient computing power is on hand, an approximate inversion
of the initial data set could be performed. The resulting subsurface model would then be
the basis for statistical experimental design, which would identify suitable electrode config-
urations, that would complement the pole-pole data in an optimal fashion. Measurements
with the selected electrode configurations and subsequent analysis of the extended data set
would lead to an improved subsurface model. If there are significant differences between
the updated model and the initial one, other electrode configurations may be necessary to
constrain the updated model. This would require further “data acquisition - data analysis -
experimental design” cycles until the model updates become negligibly small.
The resulting data set would be optimal in the sense that its information content is
limited only by the physics of the DC resistivity method and the total number of electrodes
employed. Furthermore, the data acquisition process is cost-optimized, since the iterative
cycle shown in Figure E.1 avoids unnecessary measurements.
E.4 A Novel Digital Multi-Electrode Recording System

Implementation of the real-time experimental design concept requires a highly flexible multi-
electrode data acquisition system that includes:
• an extendable bank of electrodes;
• parallel data acquisition that allows fast scans of many different electrode arrays to be
performed simultaneously;
• software-controlled selection of arbitrary four-point configurations;
• sufficient computing power to perform rudimentary data analysis and statistical exper-
imental design.
Besides specific requirements dictated by the real-time experimental design scheme, a

modern DC recording system should meet several other technical needs:
• digital system layout - digitization of voltage differences directly at the electrodes;
• programmable waveform transmitter;
• large dynamic range of receiver unit(s);
• high input impedance of receiver unit(s).
Table E.4 provides an overview of the principal technical features of multi-electrode

systems available on the market as of April, 2001. None of the systems satisfies all of
the above requirements. The SIP-256 system from the Deutschen Arbeitsgemeinschaft für
SIP-Anwendungen (Table E.4, pos. no. 13) matches most closely our requirements, but it
lacks expandability to more than 256 electrodes and it is not possible to select separately
the operational mode of individual electrodes. As a consequence, we have developed a new
system tailored to meet our needs.
As shown in Table E.4, most commercial multi-electrode systems include a single-channel
resistivity meter connected to a multiplexer. A disadvantage of such a set-up is the low data
acquisition speed; measurements from every dipole must pass through the multiplexer unit
before analogue-to-digital conversion (ADC). This “serial” way of collecting data is not
appropriate for key aspects of real-time experimental design. Acquiring an initial pole-pole
data set, for example, would be very time consuming.
To overcome these shortcomings, we have adopted the concept of distributed data ac-
quisition units (DAU’s), each attached to an electrode (Figure E.2). The DAU’s comprise
a fully digital layout, that enables data to be digitized directly at the electrode. Noise on
analogue signal paths is therefore minimized. Data transfer from the daisy-chained DAU’s is
routed via a bi-directional serial data bus (RS-485) and an interface board to the controlling
personal computer(PC).
The PC (Figure E.2) not only serves as a control unit for the DAU’s, but it also generates
the source signal using a digital-to-analogue converter board. This allows arbitrary waveforms
to be employed. The source signal is enhanced through a power amplifier (Figure E.2) and
fed into the electrode array.
An internal switching matrix allows every DAU (Figure E.3) to be configured either as
a current source, current sink, voltage reference or measuring device (Figure E.4). In the
active mode, signals from the electrode are fed into a 24-bit ADC, that has a very high
input impedance (≈ 1 GΩ). A microcomputer controls data flow from and to the DAU’s
Figure E.2: Schematic layout of a fully distributed data acquisition system for DC resistivity
surveying.
internal random access memory (RAM), while a RS-485 controller enables communication
with the PC. A unique feature of the DAU design is the internal rechargeable battery, which
is hooked up to an automatic charging circuit. After the controlling PC has initiated a
measurement, all DAU’s disconnect from the digital data bus and operate independently
until the requested amount of data has been acquired. No data transfer takes place during
a measuring sequence, so that cross-talk between wires carrying the source signal and the
data bus is avoided.
The data bus (Figure E.3) consists of four wires within a single cable and a parallel
separate coaxial cable carrying the reference signal. One DAU always operates as a reference
unit; the associated ADC is inactive and the electrode is connected to the reference line
(coaxial cable) on the field data bus (Figure E.4). All active DAU’s in the array measure
with respect to the reference electrode. Note, that an arbitrary number of DAU’s can be
”active”(Figure E.4), allowing parallel scans to be performed.
Figure E.3: Functional block diagram of one Data Acquisition Unit (DAU) and the field data bus.
After data acquisition is completed, the DAU’s reconnect to the RS-485 bus to enable
data transfer, and the power amplifier disconnects from the source signal bus (Figure E.3).
Concurrently, a 24 V DC source connects to the same wires to recharge the DAU’s internal
batteries.
Figure E.5 shows a photograph of the entire system. To protect the electronics from poor
weather conditions, the central parts of the system are stored in a box. An uninterruptable
power supply (UPS) enhances operational security of the system.
E.5 System Performance
Initial tests have been performed across a landfill site about 25 km west of Zürich. Data
were recorded on a single line of 30 DAU’s spaced 5 m apart (total line length: 145
m). The programmable source generated a constant 200 mA sinusoidal signal with a
frequency of 1 Hz. The sampling frequency of all DAU’s was set to 64 samples/s and
4 s of data were recorded.
Figure E.4: Switching scheme of an arbitrary four-pole arrangement showing the three different
operational modes of a Data Acquisition Unit (DAU).
Figure E.5: Typical field deployment of the data acquisition system.
Figure E.6 shows the acquired time series (small upper display) and the corresponding
frequency spectrum (injecting dipole spacing: 40 m; recording dipole spacing: 60 m). As
expected, there is a peak of about 0.3 V at the source signal frequency of 1 Hz. Additional
strong signals at 13 - 14 Hz are likely caused by a sorting machine within a nearby gravel
pit. An even higher peak at 16.67 Hz is caused by the Swiss railway system.
Figure E.6: Data recorded by a single Data Acquisition Unit (DAU): Acquired time series (small
upper display) and corresponding amplitude spectrum.
Digital signal processing techniques can be used to determine apparent resistivities from
the stored full-waveform current signal and voltage responses. Extensive tests have shown
that the best results are obtained when spectral divisions of voltage and source current signals
within a narrow frequency band around the dominant source frequency are performed.
As previously outlined, the ADC used in the DAU’s offer a theoretical resolution of 24
bits. The maximum input voltage range of a DAU is 5 V (±2.5V ). Thus the smallest
detectable change in voltage is given by:
±2.5V
≈ 0.3µV (E.1)
224 bit
Due to the presence of ambient electromagnetic noise, the lowest detectable voltage shown
in Figure E.6 (solid line) is about 4 µV. Based on this value, we estimated that the achievable
resolution is about 20 bits. This demonstrates that the resolution limit is governed primarily
by the ambient noise and not by the system specifications.
E.6 Conclusions
Real-time experimental design, a novel approach for collecting geophysical data, reduces field
effort and costs, while optimizing resolution provided by a data set. Implementation of this
strategy for geoelectrical prospecting has prompted the development of a new multi-electrode
data acquisition system. Application of both the real-time experimental design concept and
the new digital data acquisition system is expected to improve the quality of DC resistivity
measurements substantially.
E.7 Acknowledgments
We thank Heinrich Horstmeyer for technical advice, which improved the quality of the new
system significantly. Furthermore, we are grateful to Alan Green for improving the clarity of
the manuscript. Finally, we acknowledge financial support from ETH (project no. 0-20394-
97).
Table E.1: Comparison of commercially available multi-electrode DC resistivity systems.
No. Name Supplier Cable Type Chann. Type Source El.

1. E-Scan 3D Premier Gph. Inc., CND multi-core 1 digital 400 Vpp > 100
2. Tomoplex Campus Ltd., UK multi-core max. 64 analog 360 Vpp , 100 mA 64
E. Ext. Abstract SPIE 2001
Geopulse-Tigre multi-core 1 analog 360 Vpp , 100 mA 256

3. McOhm-El Oyo Geoelectric, JP ? 1 digital, 24 Bit 400 Vpp , 120 mA 750
4. IPR-12 IDS (Scintrex), USA multi-core max. 8 digital external
IPR-10A multi-core max. 6 digital external
5. Super Sting R8 / Swift AGI, USA multi-core / int. el. 8 ? 800 Vpp , 500 mA 254
Sting R1 / Swift multi-core / int. el. 1 ? 800 Vpp , 500 mA 254
6. SAS 4000 Lund ABEM, S multi-core 4 digital, 24 Bit 400 Vpp , 1 A 75
SAS 1000 Lund multi-core 1 digital, 24 Bit 400 Vpp , 1 A 75
7. RESECS / RESITOMO DMT, D multi-core / int. el. 6 digital, 16 Bit 120 Vpp 960
8. GDP-32 II Zonge Inc., USA multi-core max. 16 digital, 16 Bit external 16
9. SYSCAL Switch-48 IRIS Instruments, F multi-core 1 digital (?) 800 Vpp , 1.2 A 48
SYSCAL Switch-24 multi-core 1 digital (?) 400 Vpp , 500 mA 24
Multinode / Syscal Rx multi-core 1 digital (?) max. 800 Vpp , 2.5 A 16 x 16
E-NODE / Syscal Rx multi-core 1 digital (?) max. 800 Vpp , 2.5 A 255
10. GeoTom 250 / 100 Geolog, D multi-core 1 digital, 20 Bit 360 Vpp , 50 mA 250
11. GAM 200 GeoSys, D multi-core 4 ? 400 Vpp , 16 W 125
GMS 125 multi-core 1 ? 400 Vpp, 16 W 125
12. GRM-30.2 KBFI-TRIAS Ltd., H multi-core 1 ? 200 Vpp , 250 mA 30
RESP-12 multi-core 1 ? 120 Vpp , 120 mA 60
13. SIP-256 ARGE SIP-Anw., D multi-core / int. el. 256 digital, 24 Bit 200 Vpp , 100 mA 256
14. PA-CEP Aarhus Univ., DK pulled array 6 ? 36 mA 6
169
Bibliography
Alumbaugh, D. L. and Newman, G. A. (2000). Image appraisal for 2-D and 3-D electromag-
netic inversion. Geophysics, 65(5), 1455–1467.
Apparao, A., Rao, T. G., Sastry, R. S., and Sarma, V. S. (1992). Depth of detection of
buried conductive targets with different electrode arrays in resistivity prospecting. Geophys.
Prospect., 40, 749–760.
Archie, G. E. (1942). The electrical resistivity log as an aid in determining some reservoir
characteristics. Petroleum Trans. AIME, 146, 54–62.
Aspinall, A. and Gaffney, C. F. (2001). The Schlumberger array - potential and pitfalls in
archaeological prospecting. Archaeol. Prospect., 8, 199–201.
Barker, R., Venkateswararao, T., and Thangarajan, M. (2002). Application of electrical

imaging for borehole siting in hardrock regions of India. In 8th Meeting of the Environ-
mental and Engineering Geophysical Society (European Section), pages 49–52. Environ.
and Eng. Geophys. Soc., Euro. Section.
Barker, R. D. (1989). Depth of investigation of collinear symmetrical four-electrode arrays.

Geophysics, 54(8), 1031–1037.
Barker, R. D. (1997). Modern Geophysics in Engineering Geology, chapter Electrical imaging

and its application in engineering investigations, pages 37–43. The Geological Society,
London, UK.
Beard, L. P. and Tripp, A. C. (1995). Investigating the resolution of IP arrays using inverse
theory. Geophysics, 60(5), 1326–1341.
Beard, L. P., Hohmann, G. W., and Tripp, A. C. (1996). Fast resistivity / IP inversion using
a low-contrast approximation. Geophysics, 61(1), 169–179.
172 Bibliography
Beblo, M. (1997). Umweltgeophysik. Ernst und Sohn, Berlin.
Bhattacharya, B. B. and Dutta, I. (1982). Depth of investigation studies for gradient arrays
over homogenous isotropic half-space. Geophysics, 47(8), 1198–1203.
Bhattacharya, B. B. and Sen, M. K. (1981). Depth of investigation of collinear electrode

arrays over homogenous anisotropic half-space in direct current methods. Geophysics,
46(5), 768–780.
Bhattacharyya, B. K. (1957). Propagation of an electric pulse through a homogeneous and

isotropic medium. Geophysics, 22(4), 905–921.
Bigalke, J., Schleifer, N., Kötter, M., and Junge, A. (1999). SIP-256: Eine mehrkanalige
IP-Apparatur zur schnellen Bestimmung der komplexen elektrischen Leitfähigkeit des Un-
tergrundes. DGG Mitteilungen, 4, 36–44.
Broadbent, M. and Habberjam, G. M. (1971). A solution to the dipping interface problem

using the square array resistivity technique. Geophys. Prospect., 19, 321–338.
Brunner, I. (2001). Beiträge zur geoelektrischen Potentialtomographie für die Ermittlung

von Leitfähigkeitsunterschieden im Untergrund. Ph.D. thesis, Fakultät für Physik und
Geowissenschaften, Universität Leipzig, Germany.
Chambers, J., Ogilvy, R., Meldrum, P., and Nissen, J. (1999). 3D resistivity imaging of
buried oil- and tar-contaminated waste deposits. J. Environ. Eng. Geophys., 4, 3–15.
Cheng, K., Simske, S. J., Isaacson, D., Newell, J. C., and Gisser, D. G. (1990). Errors due to
measuring voltage on current-carrying electrodes in electric current computed tomography.
IEEE Trans. Med. Imag., 37(1), 60–65.
Cherkaeva, E. and Tripp, A. C. (1996). Optimal survey design using focused resistivity arrays.
IEEE Trans. Geosci. Remote Sensing, 34(2), 358–366.
Christensen, N. B. (1989). AC resistivity sounding. First Break, 7(11), 447–452.
Coggon, J. H. (1971). Electromagnetic and electrical modeling by the finite element method.
Geophysics, 36(1), 132–155.
Coggon, J. H. (1973). A comparison of IP electrode arrays. Geophysics, 38(4), 737–761.

Bibliography 173
Constable, S. C., Parker, R. L., and Constable, C. G. (1987). Occam’s inversion: A practical
algorithm for generating smooth models from electromagnetic sounding data. Geophysics,
52(3), 289–300.
Dahlin, T. (1993). On the automation of 2D resistivity surveying for engineering and envi-
ronmental applications. Ph.D. thesis, Department of Engineering Geology, Lund Institute
of Technology, Lund University, Sweden.
Dahlin, T. (1996). 2D resistivity surveying for environmental and engineering applications.

First Break, 14(7), 275–283.
Dahlin, T. (2000). Short note on electrode charge-up effects in DC resistivity data acquisition
using multi-electrode arrays. Geophys. Prospect., 48, 181–187.
Dahlin, T. and Loke, M. H. (1998). Resolution of 2D Wenner resistivity imaging as assessed

by numerical modelling. J. Appl. Geophys., 38, 237–249.
Dahlin, T. and Zhou, B. (2001). A numerical comparison of 2D resistivity imaging with eight
electrode arrays. In 7th Meeting of the Environmental and Engineering Geophysical Society
(European Section), pages 977–983. Environ. and Eng. Geophys. Soc., Euro. Section.
Dahlin, T. and Zhou, B. (2002). Gradient and mid-point-referred measurements for multi–
channel 2D resistivity imaging. In 8th Meeting of the Environmental and Engineering
Geophysical Society (European Section), pages 157–160. Environ. and Eng. Geophys.
Soc., Euro. Section.
Dahlin, T., Bernstone, C., and Loke, M. H. (2002). A 3-D resistivity investigation of a
contaminated site at Lernacken, Sweden. Geophysics, 67(6), 1692–1700.
deGroot Hedlin, C. and Constable, S. C. (1990). Occam’s inversion to generate smooth

two-dimensional models from magnetotelluric data. Geophysics, 55(12), 1613–1624.
Dey, A. and Morrison, H. F. (1973). Electromagnetic coupling in frequency and time-domain

induced-polarization surveys over a multilayered earth. Geophysics, 38(2), 380–405.
Dey, A. and Morrison, H. F. (1979a). Resistivity modeling for arbitrarily shaped three-
dimensional structures. Geophysics, 44(4), 753–780.
174 Bibliography
Dey, A. and Morrison, H. F. (1979b). Resistivity modelling for arbitrarily shaped two-
dimensional structures. Geophys. Prospect., 27, 106–136.
Dey, A., Meyer, W. H., Morrison, H. F., and Dolan, W. M. (1975). Electric field response
of two-dimensional inhomogeneities to unipolar and bipolar electrode configurations. Geo-
physics, 40(4), 630–640.
Edwards, L. S. (1977). A modified pseudosection for resistivity and IP. Geophysics, 42(5),
1020–1036.
Ellis, R. G. and Oldenburg, D. W. (1994). The pole-pole 3-D DC-resistivity inverse problem:
a conjugate-gradient approach. Geophys. J. Internat., 119, 187–194.
Embi, A. K., Nawawi, M., and Harith, Z. Z. T. (2002). The use of 2D resistivity imag-
ing techniques for detecting sinkholes and cavities in a subsurface limestone quarry at
Rawang, Malaysia. In Symposium on the Application of Geophysics to Engineering and
Environmental Problems. Env. Eng. Geophys. Soc.
Evjen, H. M. (1938). Depth factors and resolving power of electrical measurements. Geo-
physics, 3, 78–95.
Fink, J. B., editor (1990). Induced polarization: applications and case histories. Investigations
in Geophysics No. 4. Soc. Expl. Geophys.
Friedel, S. (1997). Umweltgeophysik, chapter Hochauflösende Geoelektrik - Geoelektrische

Tomographie, pages 131–151. Ernst und Sohn, Berlin.
Friedel, S. (2000). Über die Abbildungseigenschaften der geoelektrischen Impedanztomogra-

phie unter Berücksichtigung von endlicher Anzahl und endlicher Genauigkeit der Meßdaten.
Ph.D. thesis, Fakultät für Physik und Geowissenschaften, Universität Leipzig, Germany.
Furness, P. (1993). Gradient array profiles over thin resistive veins. Geophys. Prospect., 41,
113–130.
Geselowitz, D. B. (1971). An application of electrocardiographic lead theory to impedance

plethysmography. IEEE Trans. Biomed. Eng., BME-18, 38–41.
Green, A. G., Lanz, E., Maurer, H., and Boerner, D. E. (1999). A template for geophysical
investigations of a small landfill. The Leading Edge, 18(2), 248–254.
Bibliography 175
Griffiths, D. H. and Barker, R. D. (1993). Two-dimensional resistivity imaging and modelling

in areas of complex geology. J. Appl. Geophys., 29, 211–226.
Griffiths, D. H. and Turnbull, J. (1985). A multi-electrode array for resistivity surveying.

First Break, 3(7), 16–20.
Griffiths, D. H., Turnbull, J., and Olayinka, A. I. (1990). Two-dimensional resistivity mapping
with a computer controlled array. First Break, 8(4), 121–129.
Guérin, R., Begassat, P., Benderitter, Y., David, J., Tabbagh, A., and Thiry, M. (2002).
Electrical resistivity by electromagnetic Slingram mapping, electrical 2D and 3D imaging,
and electrostatic logging: a tool for studying an old waste landfill. In 8th Meeting of the
Environmental and Engineering Geophysical Society (European Section), pages 141–144.
Environ. and Eng. Geophys. Soc., Euro. Section.
Guptasarma, D. (1981). Comments on “A theorem for direct current regimes and some of
its consequences” and some related papers. Geophys. Prospect., 29, 308–311.
Hallof, P. G. (1957). On the interpretation of resistivity and induced polarization measure-

ments. Ph.D. thesis, MIT, Cambridge, Massachusetts, USA.
Inman, J. R. (1975). Resistivity inversion with ridge regression. Geophysics, 40(5), 798–817.
Inman, J. R., Ryu, J., and Ward, S. H. (1973). Resistivity inversion. Geophysics, 38(6),
1088–1108.
Isaacson, D. (1986). Distinguishability of conductivities by electric current computed tomog-

raphy. IEEE Trans. Med. Imag., 5, 91–95.
Keller, G. V. and Frischknecht, F. C. (1966). Electrical methods in geophysical prospecting.

Pergamo Press Inc., Oxford.
Knödel, K., Krummel, H., and Lange, G. (1997). Geophysik. Springer-Verlag, Berlin.
Koefoed, O. (1979). Geosounding Principles, 1. Resistivity Sounding Measurements. Elsevier

Publishing Company, Amsterdam.
Kunetz, G. (1966). Principles of direct current resistivity prospecting. Borntraeger, Berlin.

176 Bibliography
LaBrecque, D. J., Miletto, M., Daily, W., Ramirez, A., and Owen, E. (1996). The effect of
noise on Occam’s inversion of resistivity tomography data. Geophysics, 61(2), 538–548.
Lanz, E., Jemmi, L., Müller, R., Green, A. G., Pugin, A., and Huggenberger, P. (1994).
Integrated studies of Swiss waste disposal sites: Results from georadar and other geo-
physical surveys. In 5th Internat. Conf. on Ground Penetrating Radar (GPR 1994), pages
1261–1274.
Lanz, E., Pugin, A., Green, A. G., and Horstmeyer, H. (1996). Results of 2- and 3-D high-
resolution seismic reflection surveying of surficial sediments. Geophys. Res. Lett., 23(5),
491–494.
Lanz, E., Maurer, H., and Green, A. G. (1998). Refraction tomography over a buried waste
disposal site. Geophysics, 63(4), 1414–1433.
Lanz, E., Boerner, D. E., Maurer, H., and Green, A. G. (1999). Landfill delineation and
characterization using electrical, electromagnetic and magnetic methods. J. Environ. Eng.
Geophys., 3(4), 185–196.
Lanz, E. B. (1998). Integrated geophysical studies of a composite landfill and its surround-
ings. Ph.D. thesis, Institute of Geophysics, Swiss Federal Institute of Technology, Zürich,
Switzerland.
Lehmann, H. (1995). Potential representation by independent configurations on a multi-

electrode array. Geophys. J. Internat., 120, 331–338.
Lesur, V., Cuer, M., and Straub, A. (1999a). 2-D and 3-D interpretation of electrical
tomography measurements, Part 1: The forward problem. Geophysics, 64(2), 386–395.
Lesur, V., Cuer, M., and Straub, A. (1999b). 2-D and 3-D interpretation of electrical
tomography measurements, Part 2: The inverse problem. Geophysics, 64(2), 396–402.
Li, Y. and Oldenburg, D. W. (1994). Inversion of 3-D DC resistivity data using an approximate
inverse mapping. Geophys. J. Internat., 116, 527–537.
Loke, M. H. (1999). Electrical imaging surveys for environmental and engineering studies.
http://www.abem.se.
Bibliography 177
Loke, M. H. and Barker, R. D. (1996a). Practical techniques for 3D resistivity surveys and
data inversion. Geophys. Prospect., 44, 499–523.
Loke, M. H. and Barker, R. D. (1996b). Rapid least-squares inversion of apparent resistivity

pseudosections by a quasi-Newton method. Geophys. Prospect., 44, 131–152.
Lowrie, W. (1997). Fundamentals of geophysics. Cambridge University Press.
Lowry, T., Allen, M. B., and Shive, P. N. (1989). Singularity removal: A refinement of
resistivity modeling techniques. Geophysics, 54(6), 766–774.
Luo, Y. and Zhang, G. (1998). Theory and Application of Spectral Induced Polarization.
Society of Exploration Geophysicists.
Marquardt, D. (1970). Generalized inverse, ridge regression, biased linear estimation and
nonlinear estimation. Technometrics, 12, 591–612.
Maurer, H. and Boerner, D. E. (1998). Optimized and robust experimental design: a non–
linear application to EM sounding. Geophys. J. Internat., 132, 458–468.
Maurer, H., Holliger, K., and Boerner, D. E. (1998). Stochastic regularization: Smoothness
or similarity ? Geophys. Res. Lett., 25(15), 2889–2892.
Maurer, H., Boerner, D. E., and Curtis, A. (2000). Design strategies for electromagnetic
geophysical surveys. Inverse Problems, 16, 1097–1117.
Menke, W. (1984). Geophysical Data Analysis: Discrete Inverse Theory. Academic Press,
Orlando, FL.
Møller, I., Jacobsen, B. H., and Christensen, N. B. (2001). Rapid inversion of 2-D geoelctrical
data by multichannel deconvolution. Geophysics, 66(3), 800–808.
Mooney, H. M., Orellana, E., Pickett, H., and Tornheim, L. (1966). A resistivity computation
method for layered earth models. Geophysics, 31(1), 192–203.
Mufti, I. (1976). Finite-difference resistivity modeling for arbitrarily shaped two-dimensional

structures. Geophysics, 41(1), 62–78.
Muskat, M. and Evinger, H. H. (1941). Current penetration in direct current prospecting.

Geophysics, 6, 397–427.
178 Bibliography
Noel, M. and Xu, B. (1991). Archaeological investigation by electrical resistivity tomography:

a preliminary study. Geophys. J. Internat., 107, 95–102.
Olayinka, A. I. and Yaramanci, U. (2000). Assessment of the reliability of 2D inversion of

apparent resistivity data. Geophys. Prospect., 48, 293–316.
Oldenburg, D. W. (1978). The interpretation of direct current resistivity measurements.

Geophysics, 43(3), 610–625.
Oldenburg, D. W. and Li, Y. (1994). Inversion of induced polarization data. Geophysics,

59(9), 1327–1341.
Oldenburg, D. W. and Li, Y. (1999). Estimating depth of investigation in dc resistivity and

IP surveys. Geophysics, 64(2), 403–416.
Park, S. K. and Van, G. P. (1991). Inversion of pole-pole data for 3-D resistivity structure
beneath arrays of electrodes. Geophysics, 56(7), 951–960.
Pelton, W. H., Ward, S. H., Hallof, P. G., Sill, W. R., and Nelson, P. H. (1978). Mineral
discrimination and removal of inductive coupling with multifrequency IP. Geophysics,
43(3), 588–609.
Pridmore, D. F., Hohmann, G. W., Ward, S. H., and Sill, W. R. (1981). An investigation
of finite-element modeling for electrical and electromagnetic data in three dimensions.
Geophysics, 46, 1009–1024.
Reynolds, J. M. (1997). An introduction to applied and environmental geophysics. John

Wiley & Sons Ltd.
Robinson, E. S. and Coruh, C. (1988). Basic exploration geophysics. John Wiley & Sons
Ltd.
Roth, M. J. S., Mackey, J. R., and Nyquist, J. E. (1999). A case study of the use of
multi-electrode earth resistivity in thinly mantled karst. In Symposium on the Application
of Geophysics to Engineering and Environmental Problems, pages 293–302. Env. Eng.
Geophys. Soc.
Roy, A. (1972). Depth of investigation in Wenner, three-electrode and dipole-dipole DC

resistivity methods. Geophys. Prospect., 20, 329–340.
Bibliography 179
Roy, A. (1978). A theorem for direct current regimes and some of its consequences. Geophys.
Prospect., 26, 442–463.
Roy, A. and Apparao, A. (1971). Depth of investigation in direct current methods. Geo-
physics, 36(5), 943–959.
Sasaki, Y. (1994). 3-D resistivity inversion using the finite-element method. Geophysics,
59(11), 1839–1848.
Schlumberger, C. and Schlumberger, M. (1932). Depth of investigation attainable by po-

tential methods of electrical exploration. A.I.M.E. Trans. Geophys. Prosp., 97, 127–133.
Schön, J. H. (1996). Physical properties of rocks: fundamentals and principles of petro-

physics. Pergamo Press Inc., Oxford.
Sharma, P. V. (1997). Environmental and engineering geophysics. Cambridge University

Press.
Slob, E., Bloem, E., Post, V., and Groen, K. (2002a). Integrated use of geo-electrical
methods for subsurface salinity distribution mapping. In 8th Meeting of the Environmental
and Engineering Geophysical Society (European Section), pages 65–68. Environ. and Eng.
Geophys. Soc., Euro. Section.
Slob, E., Bloem, E., van Breukelen, B., and Groen, K. (2002b). Two and three-dimensional
resistivity surveys in a landfill leachate investigation. In 8th Meeting of the Environmental
and Engineering Geophysical Society (European Section), pages 195–198. Environ. and
Eng. Geophys. Soc., Euro. Section.
Sørensen, K. I. (1994). Pulled array continuous electrical profiling. In IV Meeting of the

Environmental and Engineering Geophysical Society (European Section), pages 92–93.
Environ. and Eng. Geophys. Soc., Euro. Section.
Sørensen, K. I. (1996). Pulled array continuous electrical profiling. First Break, 14, 85–90.
Spitzer, K. (1995). A 3-D finite-difference algorithm for DC resistivity modelling using

conjugate gradient methods. Geophys. J. Internat., 123, 903–914.
Spitzer, K. (1998). The three-dimensional DC sensitivity for surface and subsurface sources.
Geophys. J. Internat., 134, 736–746.
180 Bibliography
Stanek, J., Rehak, J., and Hlavac, J. (2002). Introduction to RS-422 and RS-485.
http://www.hw.cz/english/docs/rs485/rs485.html.
Stoyer, C. H. and Greenfield, R. J. (1976). Numerical solutions of the response of a two-

dimensional earth to an oscillating magnetic dipole source. Geophysics, 41(3), 519–530.
Stummer, P., Maurer, H., Horstmeyer, H., and Green, A. G. (2002). Optimization of DC
resistivity data acquisition: real-time experimental design and a new multielectrode system.
IEEE Trans. Geosci. Remote Sensing, 40(12), 2727–2735.
Tarantola, A. (1987). Inverse Problem Theory. Elsevier.
Tejero, A., Chávez, R. E., Urbieta, J., and Flores-Márquez, E. L. (2002). Cavity detection
in the southwestern hilly portion of Mexico city by resistivity imaging. J. Environ. Eng.
Geophys., 7(3), 130–139.
Telford, W. M., Geldart, L. P., and Sheriff, R. E. (1990). Applied Geophysics. Cambridge
University Press.
Unsworth, M. J., Travis, B. J., and Chave, A. D. (1993). Electromagnetic induction by a

finite electric dipole source over a 2-D earth. Geophysics, 58(2), 198–214.
Van Nostrand, R. G. (1953). Limitations on resistivity methods as inferred from the buried
sphere problem. Geophysics, 18, 423–433.
Van Nostrand, R. G. and Cook, K. L. (1966). Interpretation of resistivity data. U.S. Geol.
Survey, Prof. Paper, 449.
Wait, J. R. (1951). Transient electromagnetic propagation in a conducting medium. Geo-

physics, 16(2).
Ward, S. H. (1990a). Geotechnical and Environmental Geophysics, chapter Resistivity and

Induced Polarization Methods, pages 147–189. Investigations in Geophysics No. 5. Soc.
Expl. Geophys.
Ward, S. H., editor (1990b). Geotechnical and Environmental Geophysics. Investigations in

Geophysics No. 5. Soc. Expl. Geophys.
Bibliography 181
Weller, A., Seichter, M., and Kampke, A. (1996). Induced-polarization modelling using
complex electrical conductivities. Geophys. J. Internat., 127.
Xu, B. and Noel, M. (1993). On the completeness of data sets with multielectrode systems
for electrical resistivity surveys. Geophys. Prospect., 41, 791–801.
Zhang, J., Mackie, R. L., and Madden, T. R. (1995). 3-D resistivity forward modeling and
inversion using conjugate gradients. Geophysics, 60(5), 1313–1325.
Zhao, S. and Yedlin, M. J. (1996). Some refinements on the finite-difference method for
3-D dc resistivity modeling. Geophysics, 61(5), 1301–1307.
Zhody, A. A. R. (1989). A new method for the automatic interpretation of Schlumberger

and Wenner sounding curves. Geophysics, 54(2), 245–253.
Zhody, A. A. R., Anderson, L. A., and Muffler, L. J. P. (1973). Resistivity, self-potential,

and induced-polarization surveys of a vapor-dominated geothermal system. Geophysics,
38(6), 1130–1144.
Zhou, W., Beck, B. F., Stephenson, J. B., LaMoreaux, P., and Associates, I. (1999). Defining
the bedrock/overburden boundary in covered karst terranes using dipole-dipole electrical
resistivity tomography. In Symposium on the Application of Geophysics to Engineering
and Environmental Problems, pages 331–340. Env. Eng. Geophys. Soc.
Zhou, W., Beck, B. F., and Adams, A. L. (2002). Selection of electrode array to map
sinkhole risk areas in karst terranes using electrical resistivity tomography. In Symposium
on the Application of Geophysics to Engineering and Environmental Problems. Env. Eng.
Geophys. Soc.
Acknowledgments
This thesis would never have been written without the generous support of many people. I
wish to thank all those who helped me during my time at ETH-Zürich:
I would foremost like to thank my direct supervisor PD Dr. Hansruedi Maurer with whom
I interacted the most and established the first contact with the AUG group. I appreciate
his belief in my abilities to undertake this work. He was always very patient and willing to
answer my questions and helped me to solve various kinds of geophysical and computational
problems.
I want to thank Prof. Dr. Alan Green for thoroughly reading my publications, recognizing
deficiencies in them and making useful recommendations. His experience and insightful
comments greatly improved the quality of my publications. I am also grateful for the financial
assistance he secured for my project.
Thanks to both, Hansruedi and Alan, for the extra effort they undertook in helping me
to meet my final deadline.
A special “thank you” goes out to Heinrich Horstmeyer for providing his tremendous
experience in nearly every field whenever there was no solution in sight.
I thank the staff of the electronics lab, especially Christoph Bärlocher, Beat Rinderknecht,
Dani Winiger and Peter Zweifel for assistance with various electronic issues I had to deal
with during the development of the multi-electrode system.
I am grateful to Sven Friedel for discussing various issues of my thesis during the late
stages of this work.
Thanks to all my colleagues of the AUG group, especially to the ones I shared the office
with: Roman Spitzer and Frank Nitsche for the first three years, Jens Tronicke and Björn
Heincke for the remaining time. I very much enjoyed the time I shared with Roman, Mike
Roth and Bernhard Lampe while riding our motorbikes. Not to forget, all the other lovely
people I have met during my time at the ETH - Regina Lippitsch, Anja and Ralf Gross,
Jaqueline Hannam, Sarah Jenny, Uli Kastrup, Sarah Signorelli - thanks for being around.
184 Acknowledgments
Also people from outside ETH have been helping at various stages of this work: I am
grateful to Herby Bettschen, Rolf Eigenheer and Franz Hüppin for their great enthusiasm of
the technical aspects of my project.
I would like to thank David Boerner for the fruitful discussions we had during the very
early stages of my doctoral project.
Last but not least, more than once I was happy for the support and encouragement I
received from my parents, my sister and her family and some very special friends back home
in Austria and in the UK.
There were a lot more people inside and outside ETH that I could not list here. Whoever
reads these words most probably is part of that crowd.
Curriculum Vitae
Personal
Name: Peter Stummer
Date of Birth: 16 / 07 / 1970
Place of Birth: St. Pölten, Austria
Citizen of: Austria
Education
1998 - 2003 Doctor of Natural Science in Geophysics, Swiss Federal Institute
of Technology (ETH) Zürich, Switzerland.
1990 - 1997 Master of Science in Applied Geophysics, University of Leoben, Austria.
1984 - 1989 Abitur in Electrical Engineering, Secondary Technical School St. Pölten, Austria.
Work Experience
1997 - 1998 Employment as processing geophysicist with
Compagnie Générale de Géophysique (CGG), London, United Kingdom.
Other
1989 - 1990 Military service.

Research Collection: New Developments in Electrical Resistivity Imaging

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Research Collection: New Developments in Electrical Resistivity Imaging

Uploaded by

Copyright:

Available Formats

Research Collection

New developments in electrical resistivity imaging

NEW DEVELOPMENTS IN ELECTRICAL

A dissertation submitted to the

Doctor of Natural Sciences

accepted on the recommendation of

Prof. Dr. Alan G. Green, examiner

2 New Multi-Electrode DC Resistivity System 33

2.4.1 Centralized systems . . . . . . . . . . . . . . . . . . . . . . . . . . 39

5.1.2 Reducing data acquisition time by distributing the data

A Resolution Capabilities 113

B ETH-DCMES Set-Up 117

C ETH-DCMES Technical Details 141

D Ext. Abstract SEG 1999 149

D.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

E Ext. Abstract SPIE 2001 159

Curriculum Vitae 185

• Entwicklung und Aufbau eines Multielektrodenmesssystems, welches die Auswahl von

• Erarbeitung eines Algorithmus zur Bestimmung von Elektrodenkonfigurationen, welche

akquisition - Datenanalyse - Aktualisierung der Experimentgeometrie” - Zyklen und erfordert

Obwohl einfache Multielektrodenmesssysteme seit mehr als einem Jahrzehnt kommerzi-

Elektrodenanordnungen waren hierbei asymmetrische Dipol-Dipol Konfigurationen, wohin-

• Development of an algorithm to identify electrode configurations that provide optimum

1.1 Electrical Properties of Rocks

Figure 1.2: Various textures of rocks (after Ward, 1990a).

1.2 Background Theory

1.3 Survey Design

  e Electrode Arrangement Geometric Factor

1.3.1 Depth of investigation

Definitions of investigation depth

A central question in the characterization of resistivity measurements is the ability of an

Forward modeling and inversion

1.3.2 Signal-to-noise ratio and geometric factor

1.3.4 Lateral resolution

is focused at a depth that is dependent on the dimensions of the array.

1.4 Instrument Design

• an ammeter to measure the injecting current;

• a sensitive voltmeter to measure the response signal;

• four metal stakes (electrodes);

Instead of commutated DC sources, low frequency AC signals (Christensen, 1989) can

Cable Drums + Switching Units

1.5 Data Analysis

• vertical electric sounding (VES)

• electric profiling or areal mapping

1.5.1 Vertical electric sounding (VES)

1.5.2 Electric profiling or areal mapping

1.5.3 2D and 3D imaging

Most applications in engineering or environmental geophysics require high-resolution resistiv-

The forward problem

U4−point = UAM − UAN − UBM + UBN , (1.21)

U (x, y, z) = Ur (x, y, z) + Us (x, y, z). (1.22)

The inverse problem

ρ ) is assumed to be linear around ρ~ 0 , such that a small perturbation

~ obs will differ

Combining equations 1.28 and 1.27 leads to:

Introduction of δ~ ρ − ρ~ 0 ) the unknown model adjustments and denoting ∂F(ρ)/∂ρj as

ρ = (S̄ T S̄)−1 S̄ T ~e = Ḡ−g~e,

ρ = (S̄ T S̄ + W̄ )−1 S̄ T ~e.

Solution appraisal of the inverse problem

1.6 Motivation of the Thesis

1.7 Outline of the Thesis

Optimization of DC Resistivity Data

Peter Stummer, Hansruedi Maurer, Heinrich Horstmeyer, and Alan G. Green

e Electrode Arrangement Geometric Factor