2-D and 3-D Electrical Imaging Surveys

Tutorial : 2-D and 3-D electrical imaging surveys
By
Dr. M.H.Loke
Copyright (1996-2018)
email : [email protected], [email protected]

Web : www.geotomosoft.com
(All rights reserved)
(Revision date : 6th March 2018)

Copyright and disclaimer notice
The author, M.H.Loke, retains the copyright to this set of notes. Users may print a copy
of the notes, but may not alter the contents in any way. The copyright notices must be
retained. For public distribution, prior approval by the author is required.
It is hoped that the information provided will prove useful for those carrying out 2-D
and 3-D field surveys, but the author will not assume responsibility for any damage or
loss caused by any errors in the information provided. If you find any errors, please
inform me by email and I will make every effort to correct it in the next edition.
You can download the programs mentioned in the text (RES2DMOD, RES2DINV,
RES3DMOD, RES3DINV) from the following Web site
www.geotomosoft.com
M.H.Loke
Table of Contents
1 Introduction to resistivity surveys ....................................................................................... 1
1.1 Introduction and basic resistivity theory ...................................................................... 1
1.2 Electrical properties of earth materials ......................................................................... 5
1.3 1-D resistivity surveys and inversions – applications, limitations and pitfalls ............ 6
1.4 Basic Inverse Theory .................................................................................................. 11
1.5 2-D model discretization methods .............................................................................. 15
2 2-D electrical surveys – Data acquisition, presentation and arrays ................................... 18
2.1 Introduction ................................................................................................................ 18
2.2 Field survey method - instrumentation and measurement procedure ......................... 18
2.3 Available field instruments ......................................................................................... 21
2.4 Pseudosection data plotting method ........................................................................... 24
2.5 A comparison of the different electrode arrays .......................................................... 26
2.5.1 The Frechet derivative for a homogeneous half-space ........................................ 26
2.5.2 A 1-D view of the sensitivity function - depth of investigation .......................... 27
2.5.3 A 2-D view of the sensitivity function ................................................................. 30
2.5.4 Wenner array........................................................................................................ 31
2.5.5 Dipole-dipole array .............................................................................................. 31
2.5.6 Wenner-Schlumberger array ................................................................................ 35
2.5.7 Pole-pole array ..................................................................................................... 37
2.5.8 Pole-dipole array .................................................................................................. 38
2.5.9 Multiple gradient array ........................................................................................ 41
2.5.10 High-resolution electrical surveys with overlapping data levels ...................... 44
2.5.11 Summary of array types .................................................................................... 45
2.5.12 Use of the sensitivity values for multi-channel measurements or streamers .... 46
3 A 2-D forward modeling program .................................................................................... 48
3.1 Finite-difference and finite-element methods............................................................. 48
3.2 Using the forward modeling program RES2DMOD .................................................. 49
3.3 Forward modeling exercises ....................................................................................... 50
4 A 2-D inversion program .................................................................................................. 52
4.1 Introduction ................................................................................................................ 52
4.2 Pre-inversion and post-inversion methods to remove bad data points ....................... 52
4.3 Selecting the proper inversion settings ....................................................................... 55
4.4 Using the model sensitivity and uncertainty values ................................................ 61
4.5 Methods to handle topography ................................................................................... 66
4.6 Incorporating information from borehole logs and seismic surveys .......................... 68
4.7 Model refinement ....................................................................................................... 72
4.8 Fast inversion of long 2-D survey lines ...................................................................... 76
4.8.1 Preprocessing steps .............................................................................................. 76
4.8.2 Data set and computer system used for tests ....................................................... 76
4.8.3 Finite-element mesh size...................................................................................... 77
4.8.4 Limit calculation of Jacobian matrix values ........................................................ 77
4.8.5 Use sparse inversion techniques .......................................................................... 78
4.8.6 Use wider cells ..................................................................................................... 78
4.8.7 Overview of methods to reduce the calculation time........................................... 79
4.9 Model resolution and automatic array optimization methods .................................... 81
4.9.1 Concept of model resolution ................................................................................ 81
4.9.2 A heuristic explanation of model resolution – through a glass darkly ................ 81
4.9.3 Examples of model resolution with standard arrays ............................................ 82
4.9.4 Array optimization methods ................................................................................ 83
4.9.5 DOI versus model resolution? ............................................................................. 86
4.9.6 The streamer design problem using a model resolution approach. ...................... 87
4.10 Pitfalls in 2-D resistivity surveys and inversion...................................................... 89
4.11 The pole-pole inversion paradox ............................................................................. 94
5 I.P. inversion ..................................................................................................................... 96
5.1 Introduction ................................................................................................................ 96
5.2 The IP effect ............................................................................................................... 96
5.3 IP data types................................................................................................................ 98
5.4 IP mathematical models............................................................................................ 100
5.5 I.P. surveys with multi-electrode systems ................................................................ 101
6 Cross-borehole imaging .................................................................................................. 103
6.1 Introduction .............................................................................................................. 103
6.2 Electrode configurations for cross-borehole surveys ............................................... 103
6.2.1 Two electrodes array – the pole-pole ................................................................. 103
6.2.2 Three electrodes array – the pole-bipole ............................................................ 105
6.2.3 Four electrodes array – the bipole-bipole .......................................................... 107
6.3 Single borehole surveys ............................................................................................ 110
6.4 Cross-borehole optimized arrays .............................................................................. 112
6.5 Optimized arrays with subsurface electrodes ........................................................... 112
7 2-D field examples .......................................................................................................... 116
7.1 Introduction .............................................................................................................. 116
7.2 Landslide - Cangkat Jering, Malaysia ...................................................................... 116
7.3 Old Tar Works - U.K. ............................................................................................... 116
7.4 Holes in clay layer - U.S.A. ...................................................................................... 117
7.5 Time-lapse water infiltration survey - U.K. ............................................................. 119
7.6 Pumping test, U.K. ................................................................................................... 121
7.7 Wenner Gamma array survey - Nigeria .................................................................... 122
7.8 Mobile underwater survey - Belgium ....................................................................... 123
7.9 Floating electrodes survey – U.S.A. ......................................................................... 125
7.10 Oil Sands, Canada ................................................................................................. 125
8 3-D electrical imaging surveys ........................................................................................ 127
8.1 Introduction to 3-D surveys ...................................................................................... 127
8.2 Array types for 3-D surveys ..................................................................................... 127
8.2.1 The pole-pole array ............................................................................................ 127
8.2.2 The pole-dipole array ......................................................................................... 130
8.2.3 The dipole-dipole, Wenner and Schlumberger arrays ....................................... 130
8.2.4 Summary of array types ..................................................................................... 132
8.3 3-D roll-along techniques ......................................................................................... 135
8.4 A 3-D forward modeling program ............................................................................ 137
8.5 3-D inversion algorithms and 3-D data sets ............................................................. 139
8.6 A 3-D inversion program .......................................................................................... 140
8.7 Banding effects in 3-D inversion models ................................................................. 141
8.8 The use of long electrodes in 3-D surveys ............................................................... 144
8.9 Data grid formats and model discretizations ............................................................ 146
8.9.1 Types of surveys and model grids ..................................................................... 146
8.9.2 Model grid optimization for large surveys with arbitrary electrodes positions . 147
8.10 3-D array optimization – grids and perimeters...................................................... 151
8.11 Unstable arrays and the geometric factor relative error ........................................ 158
8.12 Not on firm foundations – inversion with shifting electrodes in 2-D and 3-D ..... 160
8.13 Examples of 3-D field surveys .............................................................................. 165
8.13.1 Birmingham field test survey - U.K. .............................................................. 165
8.13.2 Sludge deposit - Sweden................................................................................. 166
8.13.4 Copper Hill - Australia ................................................................................... 169
8.13.5 Athabasca Basin – Canada ............................................................................. 171
8.13.6 Cripple Creek and Victor Gold Mine, Colorado – USA ................................ 173
8.13.7 Burra copper deposit, South Australia : Model reliability determination ...... 175
8.14 Closing remarks on the 3-D method ..................................................................... 178
Acknowledgments.................................................................................................................. 179
References .............................................................................................................................. 180
Appendix A The smoothness constraint and resolving deeper structures in I.P. surveys 189
A.1 A problem with I.P. inversion with a conductive overburden .................................. 189
A.2 2-D synthetic model test ........................................................................................... 189
A.3 3-D field data set test ................................................................................................ 190
A.4 Software implementation.......................................................................................... 192
Appendix B Modeling long electrodes ............................................................................ 193
B.1 Two methods to model a long electrode ................................................................... 193
Appendix C New I.P. systems and negative apparent resistivity values ......................... 194
C.1 The distributed receivers I.P. systems and negative apparent resistivity values. ..... 194
List of Figures
Figure 1. The flow of current from a point current source and the resulting potential
distribution. ......................................................................................................................... 3
Figure 2. The potential distribution caused by a pair of current electrodes. The electrodes are
1 m apart with a current of 1 ampere and a homogeneous half-space with resistivity of 1
m...................................................................................................................................... 3
Figure 3. A conventional array with four electrodes to measure the subsurface resistivity. .... 4
Figure 4. Common arrays used in resistivity surveys and their geometric factors. Note that
the dipole-dipole, pole-dipole and Wenner-Schlumberger arrays have two parameters, the
dipole length “a” and the dipole separation factor “n”. While the “n” factor is commonly
an integer value, non-integer values can also be used......................................................... 4
Figure 5. The resistivity of rocks, soils and minerals. ............................................................... 6
Figure 6. The three different models used in the interpretation of resistivity measurements. ... 7
Figure 7. A typical 1-D model used in the interpretation of resistivity sounding data for the
Wenner array. ...................................................................................................................... 7
Figure 8. A 2-D two-layer model with a low resistivity prism in the upper layer. The
calculated apparent resistivity pseudosections for the (a) Wenner and (b) Schlumberger
arrays. (c) The 2D model. The mid-point for a conventional sounding survey is also
shown. ................................................................................................................................. 9
Figure 9. Apparent resistivity sounding curves for a 2-D model with a lateral inhomogeneity.
(a) The apparent resistivity curve extracted from the 2D pseudosection for the Wenner
array. The sounding curve for a two-layer model without the low resistivity prism is also
shown by the black line curve. (b) The apparent resistivity curves extracted from the 2-D
pseudosection for the Schlumberger array with a spacing of 1.0 meter (black crosses) and
3.0 meters (red crosses) between the potential electrodes. The sounding curve for a two-
layer model without the low resistivity prism is also shown. ........................................... 10
Figure 10. Coupling between neighboring model cells through the roughness filter in a 2-D
model. (a) In the horizontal and vertical diretcions only, and (b) in diagonal diretcions as
well. ................................................................................................................................... 13
Figure 11. (a) Coupling between corresponding model blocks in two time-lapse models using
a cross-model time-lapse smoothness constraint. (b) Example Jacobian matrix structure
for five time series data sets and models. Each grey rectangle represents the Jacobian
matrix associated with a single set of measurements. ....................................................... 15
Figure 12. The different models for the subsurface used in the interpretation of data from 2-D
electrical imaging surveys. (a) A purely cell based model. (b) A purely boundary based
model. (c) The laterally constrained model. (d) A combined cell based and boundary
based model with rectangular cells, and (e) with boundary conforming trapezoidal cells.
........................................................................................................................................... 16
Figure 13. Example of a cell based model with a variable boundary. (a) The test model. (b)
The apparent resistivity pseudosection. (c) The inversion model with a variable sharp
boundary that is marked by a black line. ........................................................................... 17
Figure 14. The arrangement of electrodes for a 2-D electrical survey and the sequence of
measurements used to build up a pseudosection. .............................................................. 19
Figure 15. The use of the roll-along method to extend the area covered by a 2-D survey. ..... 20
Figure 16. Sketch outline of the ABEM Lund Imaging System. Each mark on the cables
indicates an electrode position (Dahlin, 1996). The cables are placed along a single line
(the sideways shift in the figure is only for clarity). This figure also shows the principle
of moving cables when using the roll-along technique. .................................................... 22
Figure 17. The Aarhus Pulled Array System. The system shown has two current (C)
electrodes and six potential electrodes (Christensen and Sørensen 1998, Bernstone and
Dahlin 1999)...................................................................................................................... 23
Figure 18. The Geometrics OhmMapper system using capacitive coupled electrodes.
(Courtesy of Geometrics Inc.). .......................................................................................... 23
Figure 19. Schematic diagram of a possible mobile underwater survey system. The cable has
two fixed current electrodes and a number of potential electrodes so that measurements
can be made at different spacings. The above arrangement uses the Wenner-
Schlumberger type of configuration. Other configurations, such as the gradient array, can
also be used. ...................................................................................................................... 24
Figure 20. The apparent resistivity pseudosections from 2-D imaging surveys with different
arrays over a rectangular prism. ........................................................................................ 25
Figure 21. The parameters for the sensitivity function calculation at a point (x,y,z) within a
half-space. A pole-pole array with the current electrode at the origin and the potential
electrode “a” meters away is shown. ................................................................................ 26
Figure 22. A plot of the 1-D sensitivity function. (a) The sensitivity function for the pole-pole
array. Note that the median depth of investigation (red arrow) is more than twice the
depth of maximum sensitivity (blue arrow). (b) The sensitivity function and median
depth of investigation for the Wenner array...................................................................... 28
Figure 23. 2-D sensitivity sections for the Wenner array. The sensitivity sections for the (a)
alpha, (b) beta and (c) gamma configurations. .................................................................. 32
Figure 24. 2-D sensitivity sections for the dipole-dipole array. The sections with (a) n=1, (b)
n=2, (c) n=4 and (d) n=6. ................................................................................................. 33
Figure 25. Two possible different arrangements for a dipole-dipole array measurement. The
two arrangements have the same array length but different “a” and “n” factors resulting
in very different signal strengths. ...................................................................................... 35
Figure 26. 2-D sensitivity sections for the Wenner-Schlumberger array. The sensitivity
sections with (a) n=1, (b) n=2, (c) n=4 and (d) n=6. ........................................................ 36
Figure 27. A comparison of the (i) electrode arrangement and (ii) pseudosection data pattern
for the Wenner and Wenner-Schlumberger arrays. ........................................................... 37
Figure 28. The pole-pole array 2-D sensitivity section............................................................ 38
Figure 29. The forward and reverse pole-dipole arrays. .......................................................... 39
Figure 30. The pole-dipole array 2-D sensitivity sections. The sensitivity sections with (a)
n=1, (b) n=2, (c) n=4 and (d) n=6. .................................................................................... 40
Figure 31. Sensitivity sections for the gradient array. The current electrodes are fixed at x=0
and 1.0 meters, and the distance between the potential electrodes is 0.1 m. The position
of the P1-P2 potential electrodes at (a) 0.45 and 0.55 m., (b) 0.55 and 0.65 m., (c) 0.65
and 0.75 m, (d) 0.75 and 0.85 m. and (e) 0.80 and 0.90 m. .............................................. 42
Figure 32. (a) Example of multiple gradient array data set and inversion model. (b) Profile
plot using exact pseudodepths. (c) Profile plot using approximate pseudodepths. ........... 43
Figure 33. The apparent resistivity pseudosection for the dipole–dipole array using
overlapping data levels over a rectangular prism. Values of 1 to 3 meters are used for the
dipole length ‘a’, and the dipole separation factor ‘n’ varies from 1 to 5. Compare this
with Figure 20c for the same model but with ‘a’ fixed at 1 meter, and “n” varying from 1
to 10. In practice, a ‘n’ value greater than 8 would result in very noisy apparent resistivity
values. ................................................................................................................................ 45
Figure 34. Cumulative sensitivity sections for different measurement configurations using (a)
a dipole-dipole sequence, (b) a moving gradient array and (c) an expanding Wenner-
Schlumberger array. .......................................................................................................... 47
Figure 35. The output from the RES2DMOD software for the SINGLE_BLOCK.MOD 2-D
model file. The individual cells in the model are shown in the lower figure, while the
upper figure shows the pseudosection for the Wenner Beta (dipole-dipole with n=1)
array. .................................................................................................................................. 49
Figure 36. An example of a field data set with a few bad data points. The most obvious bad
datum points are located below the 300 meters and 470 meters marks. The apparent
resistivity data in (a) pseudosection form and in (b) profile form. ................................... 53
Figure 37. Selecting the menu option to remove bad data points manually. ........................... 54
Figure 38. Error distribution bar chart from a trial inversion of the Grundfor Line 1 data set
with five bad data points. .................................................................................................. 55
Figure 39. Different options to modify the inversion process. ................................................ 56
Figure 40. Example of inversion results using the l 2-norm smooth inversion and l1-norm
blocky inversion model constrains. (a) Apparent resistivity pseudosection (Wenner array)
for a synthetic test model with a faulted block (100 m) in the bottom-left side and a
small rectangular block (2 m) on the right side with a surrounding medium of 10 m.
The inversion models produced by (b) the conventional least-squares smoothness-
constrained or l2-norm inversion method and (c) the robust or l1-norm inversion method.
........................................................................................................................................... 57
Figure 41. Different methods to subdivide the subsurface into rectangular prisms in a 2-D
model. Models obtained with (a) the default algorithm, (b) by allowing the number of
model cells to exceed the number of data points, (c) a model which extends to the edges
of the survey line and (d) using the sensitivity values for a homogeneous earth model. .. 58
Figure 42. The options to change the thickness of the model layers. ...................................... 59
Figure 43. The options under the ‘Change Settings’ menu selection. ..................................... 60
Figure 44. The dialog box to limit the model resistivity values. ............................................. 61
Figure 45. Landfill survey example (Wenner array). (a) Apparent resistivity pseudosection ,
(b) model section, (c) model sensitivity section (d) model uncertainty section, (e)
minimum and maximum resistivity sections. .................................................................... 62
Figure 46. Landfill survey depth of investigation determination. (a) Model section with
extended depths and (b) the normalized DOI index section. ............................................ 65
Figure 47. Beach survey example in Denmark. The figure shows the model section with
extended depths and the normalized DOI index section. .................................................. 65
Figure 48. Different methods to incorporate topography into a 2-D inversion model. (a)
Schematic diagram of a typical 2-D inversion model with no topography. A finite-
element mesh with four nodes in the horizontal direction between adjacent electrodes is
normally used. The near surface layers are also subdivided vertically by several mesh
lines. Models with a distorted grid to match the actual topography where (b) the
subsurface nodes are shifted vertically by the same amount as the surface nodes, (c) the
shift in the subsurface nodes are gradually reduced with depth or (d) rapidly reduced with
depth, and (e) the model obtained with the inverse Schwartz-Christoffel transformation
method. .............................................................................................................................. 67
Figure 49. Fixing the resistivity of rectangular and triangular regions of the inversion model.
........................................................................................................................................... 70
Figure 50. The inversion model cells with fixed regions. The fixed regions are drawn in
purple. Note that the triangular region extends beyond the survey line. ........................... 71
Figure 51. Example of an inversion model with specified sharp boundaries. (a) The
boundaries in the Clifton survey (Scott et al., 2000) data set is shown by the blue lines.
(b) The measured apparent resistivity pseudosection and the inversion model section. ... 71
Figure 52. The effect of cell size on the model misfit for near surface inhomogeneities. (a)
Model with a cell width of one unit electrode spacing. (b) A finer model with a cell width
of half the unit electrode spacing. The near surface inhomogeneities are represented by
coloured ovals. .................................................................................................................. 72
Figure 53. Synthetic model (c) used to generate test apparent resistivity data for the pole-
dipole (a) and Wenner (b) arrays....................................................................................... 73
Figure 54. The effect of cell size on the pole-dipole array inversion model. Pole-dipole array.
(a) The apparent resistivity pseudosection. The inversion models obtained using cells
widths of (b) one, (b) one-half and (c) one-quarter the unit electrode spacing. ................ 74
Figure 55. Example of the use of narrower model cells with the Wenner-Schlumberger array.
(a) The apparent resistivity pseudosection for the PIPESCHL.DAT data set. The
inversion models using (b) cells with a width of 1.0 meter that is the same as the actual
unit electrode, and (c) using narrower cells with a width of 0.5 meter. ............................ 75
Figure 56. Example of reduction of near-surface 'ripples' in inversion model. (a) Apparent
resistivity pseudosection for the BLUERIDGE.DAT data set. (b) Normal inversion using
the robust inversion norm and model refinement. (c) Inversion using higher damping
factor for first layer to reduce the 'ripple' effect in the top layer. ...................................... 75
Figure 57. Inversion models for a long survey line using different settings to reduce the
calculation time. (a) The measured apparent resistivity pseudosection for the Redas
underwater survey for the first 2600 meters. Inversion models using (b) standard
inversion settings with 4 nodes between adjacent electrode positions, (c) standard
inversion settings with 2 nodes between adjacent electrode positions, (d) with calculation
of Jacobian matrix values for selected model cells, (e) with sparse inversion technique,
(f) with 2 meter model cell width and (g) with 3 meter model cell width. ....................... 80
Figure 58. Part of finite-element mesh used to model a survey with submerged electrodes.
The resistivity of mesh cells in the water layer are fixed at the known water resistivity. 80
Figure 59. Part of a finite-element mesh for a long survey line. The electrodes used in the
same array are 3 meters apart, while the data unit electrode spacing is 1 meter. Using 2
nodes between adjacent electrodes in the mesh will actually result in 6 nodes between
electrodes in the same array. ............................................................................................. 81
Figure 60. Model resolution sections for the (a) Wenner, (b) Wenner-Schlumberger and (c)
simple dipole-dipole array (d) dipole-dipole array with overlapping data levels. ............ 83
Figure 61. (a) Model resolution section for comprehensive data set. (b) Model resolution
section for optimized data set generated by the ‘Compare R’ method. ............................ 84
Figure 62. Test inversion model for the different arrays. (a) Apparent resistivity
pseudosection for the simple dipole-dipole array with a dipole length ‘a’ of 1.0 meter
with the dipole separation factor ‘n’ of 1 to 6. (b) Apparent resistivity pseudosection for
the dipole-dipole array with overlapping data levels. (c) The synthetic model with 4
rectangular blocks of 100 .m embedded in a medium of 10 .m. as used by Wilkinson
et al. (2006b). .................................................................................................................... 85
Figure 63. Inversion results with the (a) Wenner array, (b) Wenner-Schlumberger array, (c)
simple dipole-dipole array and (d) dipole-dipole array with overlapping data levels. ..... 86
Figure 64. Inversions models with the (a) optimized data set with 4462 data points and (b) a
truncated optimized data set with 413 data points. ........................................................... 86
Figure 65. Comparison between the (a) model resolution, (b) model resolution index and (c)
DOI index sections for the Landfill data set. .................................................................... 87
Figure 66. The model resolution plots for the three streamer configurations. ......................... 88
Figure 67. An example of 3-D effects on a 2-D survey. (a) Apparent resistivity
pseudosections (Wenner array) along lines at different y-locations over (b) a 3-D
structure shown in the form of horizontal slices. .............................................................. 91
Figure 68. The 2-D sensitivity sections for the pole-dipole array with a dipole length of 1
meter and with (a) n=6, (b) n=12 and (c) n=18. Note that as the ‘n’ factor increases, the
zone of high positive sensitive values becomes increasingly concentrated in a shallower
zone below the P1-P2 dipole. ............................................................................................ 92
Figure 69. Example of apparent resistivity pseudosection with pole-dipole array with large
‘n’ values. Note that the anomaly due to a small near-surface high resistivity block
becomes greater as the ‘n’ factor increases. This means that the sensitivity of the array to
the near-surface region between the P1-P2 potential dipole becomes greater as the ‘n’
factor increases. ................................................................................................................. 93
Figure 70. Diagrammatic illustration of differences in objective function shapes for the pole-
pole array and dipole-dipole array data sets leading to different models obtained from
optimization routine. ......................................................................................................... 95
Figure 71. The I.P. values for some rocks and minerals. ......................................................... 97
Figure 72. The Cole-Cole model. (a) Simplified electrical analogue circuit model (after
Pelton et al. 1978).  = resistivity, m = chargeability,  = time constant, c = relaxation
constant. (b) Amplitude and phase response to sine wave excitation (frequency domain).
(c) Transient response to square wave current pulse (time domain). Most I.P. receivers
measure the integral of the decay voltage signal over a fixed interval, mt, as a measure of
the I.P. effect. .................................................................................................................... 97
Figure 73. Magusi River massive sulphide ore body inversion models. (a) Apparent resistivity
pseudosection. Resistivity inversion models obtained using the (b) perturbation and (c)
complex resistivity methods. (d) Apparent I.P. (metal factor) pseudosection. I.P. models
obtained using the (e) perturbation and (f) complex resistivity methods. ....................... 101
Figure 74. Sketch of separated cable spreads setup used (after Dahlin and Loke, 2015). ..... 102
Figure 75. Resistivity and chargeability pseudosection from field demo at 3rd IP workshop at
Ile d’Oleron (after Dahlin and Loke, 2015). .................................................................. 102
Figure 76. The possible arrangements of the electrodes for the pole-pole array in the cross-
borehole survey and the 2-D sensitivity sections. The locations of the two boreholes are
shown by the vertical black lines. ................................................................................... 104
Figure 77. A schematic diagram of two electrodes below the surface. The potential measured
at P can be considered as the sum of the contribution from the current source C and its
image C’ above the ground surface. ................................................................................ 105
Figure 78. The 2-D sensitivity patterns for various arrangements with the pole-bipole array.
The arrangement with (a) C1 and P1 in first borehole and P2 in second borehole, (b) C1
in the first borehole and both P1 and P2 in the second borehole, (c) all three electrodes in
the first borehole and (d) the current electrode on the ground surface. .......................... 106
Figure 79. The 2-D sensitivity patterns for various arrangements of the bipole-bipole array.
(a) C1 and P1 are in the first borehole, and C2 and P2 are in second borehole. (b) C1 and
C2 are in the first borehole, and P1 and P2 are in second borehole. In both cases, the
distance between the electrodes in the same borehole is equal to the separation between
the boreholes. The arrangements in (c) and (d) are similar to (a) and (b) except that the
distance between the electrodes in the same borehole is half the spacing between the
boreholes. ........................................................................................................................ 107
Figure 80. Possible measurement sequences using the bipole-bipole array. Other possible
measurements sequences are described in the paper by Zhou and Greenhalgh (1997). . 109
Figure 81. Several possible bipole-bipole configurations with a single borehole. (a) The C1
and C2 electrodes at depths of 3 and 2 meters respectively below the 0 meter mark. (b)
The C1 and P1 electrodes at depths of 3 and 2 meters respectively below the 0 meter
mark. (c). The C1 electrode is at a depth of 3 meters below the 0 meter mark while the
C2 electrode is on the surface. (d). The C1 electrode is at a depth of 3 meters below the 0
meter mark while the P1 electrode is on the surface. ...................................................... 111
Figure 82. A pole-bipole survey with a single borehole. The C1 electrode is at a depth of 3
meters below the 0 meter mark. ...................................................................................... 111
Figure 83. Test of optimized cross-borehole arrays with a synthetic model. (a) Two-layer test
model with conductive and resistive anomalies. Inversion models for (b) optimized data
set with all arrays, (c) 'standard' data set and (d) the reduced optimized data set that
excludes arrays with both current (or potential) electrodes in the same borehole. All the
data sets have 1875 data points. The outlines of the rectangular blocks showing their true
positions are also shown. ................................................................................................. 113
Figure 84. Schematic diagram of the MERIT method with the electrodes are planted along
the surface and directly below using the direct push technology (after Harro and Kruse,
2013)................................................................................................................................ 114
Figure 85. Inversion models for the different data sets for the data collected with electrodes at
surface and 7.62 m depth with 4 m horizontal spacing. Models for the (a) standard arrays
(405 data points), optimized arrays with (b) 403 and (c) 514 data points....................... 115
Figure 86. Landslide field example, Malaysia. (a) The apparent resistivity pseudosection for a
survey across a landslide in Cangkat Jering and (b) the interpretation model for the
subsurface. ....................................................................................................................... 116
Figure 87. Industrial pollution example, U.K. (a) The apparent resistivity pseudosection from
a survey over a derelict industrial site, and the (b) computer model for the subsurface. 117
Figure 88. Mapping of holes in a clay layer, U.S.A. (a) Apparent resistivity pseudosection for
the survey to map holes in the lower clay layer. (b) Inversion model and (c) sensitivity
values of the model cells used by the inversion program. .............................................. 118
Figure 89. Water infiltration mapping, U.K. (a) The apparent resistivity and (b) inversion
model sections from the survey conducted at the beginning of the Birmingham
infiltration study. This shows the results from the initial data set that forms the base
model in the joint inversion with the later time data sets. As a comparison, the model
obtained from the inversion of the data set collected after 10 hours of irrigation is shown
in (c). ............................................................................................................................... 119
Figure 90. Time-lapse sections from the infiltration study. The sections show the change in
the subsurface resistivity values with time obtained from the inversion of the data sets
collected during the irrigation and recovery phases of the study. ................................... 120
Figure 91. Hoveringham pumping test, U.K. (a) The apparent resistivity pseudosection at the
beginning of the test. The inversion model sections at the (b) beginning and (v) after 220
minutes of pumping. ........................................................................................................ 121
Figure 92. Percentage relative change in the subsurface resistivity values for the
Hoveringham pumping test. To highlight the changes in the subsurface resistivity, the
changes in the model resistivity are shown. Note the increase in the model resistivity
below the borehole with time. ......................................................................................... 121
Figure 93. Use of Archie’s Law for the Hoveringham pumping test. Sections showing the
relative desaturation values obtained from the inversion models of the data sets collected
during the different stages of the Hoveringham pumping test. Archie’s Law probably
gives a lower limit for the actual change in the aquifer saturation. ................................ 122
Figure 94. Groundwater survey, Nigeria. (a). Apparent resistivity pseudosection. (b) The
inversion model with topography. Note the location of the borehole at the 175 meters
mark................................................................................................................................. 123
Figure 95. The inversion model after 4 iterations from an underwater riverbed survey by
Sage Engineering, Belgium. ............................................................................................ 124
Figure 96. Thames River (CT, USA) survey with floating electrodes. (a) The measured
apparent resistivity pseudosection. Inversion models obtained (b) without constraints on
the water layer, and (c) with a fixed water layer. ............................................................ 125
Figure 97. Survey to map oil sands, Alberta, Canada. (a) Location of major tar sands deposits
in Alberta, Canada. (b) Example of resistivity log and geologic column of Athabasca oil
sands. (c) 2-D resistivity model from imaging survey (Kellett and Bauman, 1999). ..... 126
Figure 98. A simple arrangement of the electrodes for a 3-D survey. ................................... 128
Figure 99. Two possible measurement sequences for a 3-D survey. The location of potential
electrodes corresponding to a single current electrode in the arrangement used by (a) a
survey to measure the complete data set and (b) a cross-diagonal survey. ..................... 128
Figure 100. 3-D sensitivity plots for the pole-pole array. The plots are in the form of
horizontal slices through the earth at different depths. ................................................... 129
Figure 101. 3-D sensitivity plots for the pole-dipole array with n=1 in the form of horizontal
slices through the earth at different depths. The C1 electrode is the leftmost white cross.
......................................................................................................................................... 131
slices through the earth at different depths...................................................................... 131
Figure 103. 3-D sensitivity plots for the dipole-dipole array with n=1 in the form of
horizontal slices through the earth at different depths. The C2 electrode is the leftmost
white cross. ...................................................................................................................... 133
horizontal slices through the earth at different depths. The C2 electrode is the leftmost
white cross. ...................................................................................................................... 133
Figure 105. The 3-D sensitivity plots for the Wenner alpha array at different depths. ......... 134
Figure 106. The 3-D sensitivity plots for the Wenner-Schlumberger array with the n=4 at
different depths................................................................................................................ 134
Figure 107. The 3-D sensitivity plots for the Wenner gamma array at different depths. ...... 135
Figure 108. Using the roll-along method to survey a 10 by 10 grid with a multi-electrode
system with 50 nodes. (a) Surveys using a 10 by 5 grid with the lines orientated in the x-
direction. (b) Surveys with the lines orientated in the y-direction. ................................. 136
Figure 109. A 3-D model with 4 rectangular prisms in a 15 by 15 survey grid. (a) The finite-
difference grid. (b) Horizontal apparent resistivity pseudosections for the pole-pole array
with the electrodes aligned in the x-direction.................................................................. 138
Figure 110. The model used in 3-D inversion. ...................................................................... 140
Figure 111. Inversion models for the Vetlanda landfill survey data set. (a) Using standard
inversion settings. (b) With a higher damping factor for the topmost layer. (c) Using
diagonal roughness filter in the horizontal (x-y) directions. (d) Using diagonal roughness
filters in the vertical (x-z and y-z) directions as well. (e) Using the roughness filter in all
directions. ........................................................................................................................ 142
Figure 112. Types of 3-D roughness filters. (a) With components in the x- and y- directions
only for the horizontal filter. (b) With components in the diagonal directions in the x-y
plane for the horizontal filter. (c) Applying the roughness filter with the corner model
cells as well. Only two (out of eight) corner cells are shown. ........................................ 143
Figure 113. Inversion models for the Vetlanda landfill survey data set using model cells of
equal lengths in the x- and y- directions. (a) Using standard inversion settings. (b) With a
higher damping factor for the topmost layer. (c) Using diagonal roughness filter in the
horizontal (x-y) directions. (d) Using diagonal roughness filters in the vertical (x-z and y-
z) directions as well. ........................................................................................................ 144
Figure 114. Synthetic model for long electrodes survey. 3-D model using cells of low
resistivity (0.01 .m) that are marked in red to simulate cased wells. ........................... 145
Figure 115. Comparison of inversion models using point electrodes with and without long
electrodes. (a) Inversion model for pole-pole data set using only surface point electrodes.
(b) Inversion model for pole-pole data set using 121 point and 3 long electrodes. ........ 146
Figure 116. 3-D data grid formats.......................................................................................... 147
Figure 117. Methods to model the effect of an electrode using the finite-difference and finite-
element methods.............................................................................................................. 148
Figure 118. The use of an appropriate mesh spacing to obtain sufficient accuracy for
electrodes in the same array that close together. ............................................................. 148
Figure 119. Map with survey lines and infrastructure at the Hanford site. ........................... 149
Figure 120. Types of model grids for the Hanford survey data set. (a) Using a 5 meters
spacing model for the entire area. (b) Using a 4 meters spacing model for the entire area.
(c) Using a mixture of 4 and 5 meters spacing model grid. ............................................ 150
Figure 121. Inversion model for Hanford survey site. ........................................................... 151
Figure 122. Arrangement of survey lines using a 3 cable system with the Abem SAS
instrument. ....................................................................................................................... 152
Figure 123. Inversions model for (a) combined Wenner-Schlumberger and dipole-dipole data
set, (b) optimized data set. The actual positions of the blocks are marked by black
rectangles......................................................................................................................... 152
Figure 124. Horizontal sections showing the model resolution for the (a) comprehensive data
set, (b) standard arrays and (c) optimized arrays. The electrode positions are marked by
small green crosses in the top layer in (a). ...................................................................... 153
Figure 125. Vertical cross-sections showing the model resolution for the comprehensive,
‘standard’, small and large optimized data sets............................................................... 154
Figure 126. The synthetic test model with two 1000 ohm.m rectangular blocks (marked by
black rectangles) embedded in a 100 ohm.m background medium. ............................... 154
Figure 127. The inversion models for the synthetic model with two 1000 ohm.m rectangular
blocks (marked by black rectangles) embedded in a 100 ohm.m background medium. 156
Figure 128. Comprehensive data set point-spread-function plots on a vertical x-z plane
located at y=4.5 m for a model cell at different depths with centre at x=4.5m and y=4.5m.
......................................................................................................................................... 156
Figure 129. Model resolution sections with circular perimeter for (a) comprehensive data set
with 180300 arrays, optimized data sets with (b) 946 and (c) 2000 arrays. .................... 157
Figure 130. Inversion results for survey with a circular perimeter using optimized arrays with
(a) 946 and (b) 2000 data points...................................................................................... 157
Figure 131. The (a) initial and (b) perturbed synthetic test models with apparent resistivity
pseudosections and inversion models assuming a constant electrode spacing and flat
surface. (c) The inversion model obtained with the algorithm that allows the electrodes
to shift.............................................................................................................................. 162
Figure 132. (a) 3-D synthetic test model with a rectangular survey grid. In the perturbed
model, the resistivities of the smaller blocks were changed from 400 and 20 ohm.m to
350 and 25 ohm.m. (b) The survey grid for the perturbed model. The four electrodes
shifted are marked by red circles. Electrode 1 was shifted 0.3 m in the x-direction,
electrode 2 moved 0.3 m in the y-direction, electrode 3 moved -0.2 m in both x and y-
directions while electrode 4 was shifted vertically upwards by 0.4 m. ........................... 163
Figure 133. Inversion models for the (a) initial and (b) perturbed data sets with fixed
electrodes in a rectangular grid. The true positions of the bands and prisms are marked by
black lines. The positions of the shifted electrodes are marked by small crosses in the top
layer in (b). ...................................................................................................................... 163
Figure 134. Inversion models for the perturbed data set using different relative damping
factors for the electrodes positions vector with a homogenous half-space starting model.
......................................................................................................................................... 164
Figure 135. Surface x-z profiles along the line y=10 m for inversions using a (a) homogenous
half-space and (b) initial data set starting models. .......................................................... 164
Figure 136. Inversion models for the perturbed data set using different relative damping
factors for the electrodes positions vector with the inversion model from the initial data
set as the starting model. ................................................................................................. 164
Figure 137. (a) Arrangement of electrodes in the Birmingham 3-D field survey. (b)
Horizontal and (c) vertical cross-sections of the model obtained from the inversion of the
Birmingham field survey data set. The locations of observed tree roots on the ground
surface are also shown..................................................................................................... 165
Figure 138. Example L-curve plots. (a) A plot of the model roughness versus the data misfit
for the Birmingham survey data set for a number of damping factor values. (b) A plot of
the curvature of the curve in (a) for the different damping factor values. The ‘optimum’
damping factor is at the maximum curvature point......................................................... 166
Figure 139. The 3-D model obtained from the inversion of the Lernacken Sludge deposit
survey data set displayed as horizontal slices through the earth. .................................... 167
Figure 140. A 3-D view of the model obtained from the inversion of the Lernacken Sludge
deposit survey data set displayed with the Slicer/Dicer program. A vertical exaggeration
factor of 2 is used in the display to highlight the sludge ponds. Note that the color contour
intervals are arranged in a logarithmic manner with respect to the resistivity. ............... 167
Figure 141. Map of the Panama Canal region with the survey area marked (Noonan and
Rucker, 2011). ................................................................................................................. 168
Figure 142. The model grid used for the Panama Canal floating electrodes survey with 20 by
20 meters cells. The location of the electrodes along the surveys lines are shown as
colored points. ................................................................................................................. 169
Figure 143. The inversion model for the Panama Canal floating electrodes survey. The first 4
layers correspond to the water column. ........................................................................... 169
Figure 144. Geological map of the Copper Hill area (White et al. 2001). ............................. 170
Figure 145. Electrodes layout used for the 3-D survey of the Copper Hill area.................... 170
Figure 146. The I.P. model obtained from the inversion of the Copper Hill survey data set.
Yellow areas have chargeability values of greater than 35 mV/V, while red areas have
chargeability values of greater than 45 mV/V (White et al., 2001). ............................... 171
Figure 147. Location of uranium mines in the Athabasca basin, Saskatchewan (Bingham et
al., 2006). ........................................................................................................................ 172
Figure 148. Geological model of uranium deposit (Bingham et al., 2006). .......................... 172
Figure 149. Example of inversion model from the Midwest deposit area showing the
resistivity at a depth of about 200 meters (Bingham et al., 2006). ................................. 173
Figure 150. A complex time-lapse field survey example. (a) Map of Cripple Creek survey
site. (b) Overhead view of the inversion model grid with electrodes layout. (c) Iso-surface
contours for the -4% resistivity change at different times after the injection of the sodium
cyanide solution (that started at 2.8 hours from the first data set in snapshots used). t1=
1.1 hours, t2= 2.4 hours, t3= 3.7 hours, t4= 4.9 hours. (d) Overhead view of iso-surfaces.
......................................................................................................................................... 174
Figure 151. Geological map of (a) south-west South Australia, (b) the Burra area, and (c) a
plot of survey electrodes and model cells layout. ........................................................... 175
Figure 152. Burra survey (a) resistivity and (b) I.P. inversion model layers. ........................ 176
Figure 153. The model (a) resistivity and (b) I.P. resolution index, and (c) VOI values. The
red arrows at the left side of (c) shows the position of the vertical slice shown in Figure
154. .................................................................................................................................. 177
Figure 154. Vertical cross-sections of the (a) resistivity model resolution index, (b) I.P
resolution index and (c) VOI in the X-Z plane 0.8 km north of the origin. .................... 178
Figure 155. 2-D test model with conductive overburden. ..................................................... 189
Figure 156. Models obtained with different relative damping weights αs. ............................ 190
Figure 157. Models for the Burra data set with a relative damping weight of 0.5 for αs. ..... 191
Figure 158. I.P. vertical sections along the y-direction (at x=1650 to 1700 m) Burra data set
(a) without (αs=0.0) and (b) with (αs=0.5) a reference model constraint. ....................... 191
Figure 159. A long electrode in a (a) homogeneous medium, (b) two-layer medium with a
low resistivity lower layer and (c) partly in air. .............................................................. 193
Figure 160. Example of non-conventional electrodes arrangements. (a) Offset pole-dipole
arrangement, (b) distributed pole-dipole arrangement. ................................................... 194
List of Tables
Table 1. 1-D inversion examples using the RES1D.EXE program. ........................................ 11
Table 2. The median depth of investigation (ze) for the different arrays (Edwards, 1977). L is
the total length of the array. Note identical values of ze/a for the Wenner-Schlumberger
and pole-dipole arrays. Please refer to Figure 4 for the arrangement of the electrodes for
the different arrays. The geometric factor is for an "a" value of 1.0 meter. For the pole-
dipole array, the array length ‘L’ only takes into account the active electrodes C1, P1 and
P2 (i.e. it does not take into account the remote C2 electrode). ........................................ 29
Table 3. Forward modeling examples. ..................................................................................... 50
Table 4. Methods to remove bad data points ........................................................................... 54
Table 5. Tests with different inversion options ....................................................................... 63
Table 6. Tests with different topographic modeling options ................................................... 68
Table 7. Tests with option to fix the model resistivity ............................................................ 70
Table 8. Inversion times for the long Redas survey data set using different settings. ............. 79
Table 9. Inversion times for different line lengths. .................................................................. 79
Table 10. Tests with 2-D I.P. inversion ................................................................................. 102
Table 11. A few borehole inversion tests............................................................................... 109
Table 12. 3-D forward modeling examples ........................................................................... 137
Table 13. 3-D inversion examples ......................................................................................... 141
1
1 Introduction to resistivity surveys
1.1 Introduction and basic resistivity theory

The resistivity method is one of the oldest geophysical survey techniques (Loke, 2011).
The purpose of electrical surveys is to determine the subsurface resistivity distribution by
making measurements on the ground surface. From these measurements, the true resistivity of
the subsurface can be estimated. The ground resistivity is related to various geological
parameters such as the mineral and fluid content, porosity and degree of water saturation in the
rock. Electrical resistivity surveys have been used for many decades in hydrogeological,
mining, geotechnical, environmental and even hydrocarbon exploration (Loke et al., 2013a).
The fundamental physical law used in resistivity surveys is Ohm’s Law that governs
the flow of current in the ground. The equation for Ohm’s Law in vector form for current flow
in a continuous medium is given by
J=E (1.1)
where  is the conductivity of the medium, J is the current density and E is the electric field
intensity. In practice, what is measured is the electric field potential. We note that in
geophysical surveys the medium resistivity , which is equals to the reciprocal of the
conductivity (=1/), is more commonly used. The relationship between the electric potential
and the field intensity is given by
E    (1.2)
Combining equations (1.1) and (1.2), we get
J    (1.3)
In almost all surveys, the current sources are in the form of point sources. In this case, over an
elemental volume V surrounding the a current source I, located at x s , y s , z s  the relationship
between the current density and the current (Dey and Morrison, 1979a) is given by
 I 
.J    ( x  x s )( y  y s )( z  z s ) (1.4)
 V 
where  is the Dirac delta function. Equation (3) can then be rewritten as
 I 
  • x, y, z x, y, z     ( x  x s )( y  y s )( z  z s ) (1.5)
 V 
This is the basic equation that gives the potential distribution in the ground due to a
point current source. A large number of techniques have been developed to solve this equation.
This is the “forward” modeling problem, i.e. to determine the potential that would be observed
over a given subsurface structure. Fully analytical methods have been used for simple cases,
such as a sphere in a homogenous medium or a vertical fault between two areas each with a
constant resistivity. For an arbitrary resistivity distribution, numerical techniques are more
commonly used. For the 1-D case, where the subsurface is restricted to a number of horizontal
layers, the linear filter method is commonly used (Koefoed, 1979). For 2-D and 3-D cases, the
finite-difference and finite-element methods are the most versatile. In Chapter 2, we will look
at the use of a forward modeling computer program for 2-D structures.
The more complicated cases will be examined in the later sections. First, we start with
the simplest case with a homogeneous subsurface and a single point current source on the
ground surface (Figure 1). In this case, the current flows radially away from the source, and the
potential varies inversely with distance from the current source. The equipotential surfaces
have a hemisphere shape, and the current flow is perpendicular to the equipotential surface.
The potential in this case is given by
I
 (1.6)
2r
Copyright (1996-2018) M.H.Loke

2
where r is the distance of a point in the medium (including the ground surface) from the
electrode. In practice, all resistivity surveys use at least two current electrodes, a positive
current and a negative current source. Figure 2 show the potential distribution caused by a pair
of electrodes. The potential values have a symmetrical pattern about the vertical place at the
mid-point between the two electrodes. The potential value in the medium from such a pair is
given by
I  1 1 
    (1.7)
2  rC1 rC 2 
where rC1 and rC2 are distances of the point from the first and second current electrodes.
In practically all surveys, the potential difference between two points (normally on the
ground surface) is measured. A typical arrangement with 4 electrodes is shown in Figure 3.
The potential difference is then given by
I  1 1 1 1  (1.8)
      
2  rC1P1 rC 2 P1 rC1P 2 rC 2 P 2 
The above equation gives the potential that would be measured over a homogenous half space
with a 4 electrodes array.
Actual field surveys are conducted over an inhomogenous medium where the
subsurface resistivity has a 3-D distribution. The resistivity measurements are still made by
injecting current into the ground through the two current electrodes (C1 and C2 in Figure 3),
and measuring the resulting voltage difference at two potential electrodes (P1 and P2). From
the current (I) and potential (  ) values, an apparent resistivity ( a) value is calculated.

a  k (1.9)
I
where 2
k
 1 1 1 1 
    
 rC1P1 rC 2 P1 rC1P 2 rC 2 P 2 
k is a geometric factor that depends on the arrangement of the four electrodes. Resistivity
measuring instruments normally give a resistance value, R = /I, so in practice the apparent
resistivity value is calculated by
a = k R (1.10)
The calculated resistivity value is not the true resistivity of the subsurface, but an
“apparent” value that is the resistivity of a homogeneous ground that will give the same
resistance value for the same electrode arrangement. The relationship between the “apparent”
resistivity and the “true” resistivity is a complex relationship. To determine the true subsurface
resistivity from the apparent resistivity values is the “inversion” problem. Methods to carry out
such an inversion will be discussed in more detail at the end of this chapter.
Figure 4 shows the common arrays used in resistivity surveys together with their
geometric factors. In a later section, we will examine the advantages and disadvantages of some
of these arrays.
There are two more electrical based methods that are closely related to the resistivity
method. They are the Induced Polarization (IP) method, and the Spectral Induced Polarization
(SIP) (also known as Complex Resistivity (CR)) method. Both methods require measuring
instruments that are more sensitive than the normal resistivity method, and with significantly
higher currents. IP surveys are comparatively more common, particularly in mineral
exploration surveys. It is able to detect conductive minerals of very low concentrations that

3
might otherwise be missed by resistivity or EM surveys. Commercial SIP surveys are

comparatively rare, although it is a popular research subject. Both IP and SIP surveys use
alternating currents (in the frequency domain) of much higher frequencies than standard
resistivity surveys. Electromagnetic coupling is a serious problem in both methods. To
minimize the electromagnetic coupling, the dipole-dipole (or pole-dipole) array is commonly
used.
Figure 1. The flow of current from a point current source and the resulting potential
distribution.
Figure 2. The potential distribution caused by a pair of current electrodes. The electrodes are
1 m apart with a current of 1 ampere and a homogeneous half-space with resistivity of 1 m.

4
Figure 3. A conventional array with four electrodes to measure the subsurface resistivity.
Figure 4. Common arrays used in resistivity surveys and their geometric factors. Note that the
dipole-dipole, pole-dipole and Wenner-Schlumberger arrays have two parameters, the dipole
length “a” and the dipole separation factor “n”. While the “n” factor is commonly an integer
value, non-integer values can also be used.

5
1.2 Electrical properties of earth materials

Electric current flows in earth materials at shallow depths through two main methods.
They are electronic conduction and electrolytic conduction. In electronic conduction, the
current flow is via free electrons, such as in metals. In electrolytic conduction, the current flow
is via the movement of ions in groundwater. In environmental and engineering surveys,
electrolytic conduction is probably the more common mechanism. Electronic conduction is
important when conductive minerals are present, such metal sulfides and graphite in mineral
surveys.
The resistivity of common rocks, soil materials and chemicals (Keller and Frischknecht,
1966; Daniels and Alberty, 1966; Telford et al., 1990) is shown in Figure 5. Igneous and
metamorphic rocks typically have high resistivity values. The resistivity of these rocks is
greatly dependent on the degree of fracturing, and the percentage of the fractures filled with
ground water. Thus a given rock type can have a large range of resistivity, from about 1000 to
10 million m, depending on whether it is wet or dry. This characteristic is useful in the
detection of fracture zones and other weathering features, such as in engineering and
groundwater surveys.
Sedimentary rocks, which are usually more porous and have higher water content,
normally have lower resistivity values compared to igneous and metamorphic rocks. The
resistivity values range from 10 to about 10000 m, with most values below 1000 m. The
resistivity values are largely dependent on the porosity of the rocks, and the salinity of the
contained water.
Unconsolidated sediments generally have even lower resistivity values than
sedimentary rocks, with values ranging from about 10 to less than 1000 m. The resistivity
value is dependent on the porosity (assuming all the pores are saturated) as well as the clay
content. Clayey soil normally has a lower resistivity value than sandy soil. However, note the
overlap in the resistivity values of the different classes of rocks and soils. This is because the
resistivity of a particular rock or soil sample depends on a number of factors such as the
porosity, the degree of water saturation and the concentration of dissolved salts.
The resistivity of groundwater varies from 10 to 100 m. depending on the
concentration of dissolved salts. Note the low resistivity (about 0.2 m) of seawater due to
the relatively high salt content. This makes the resistivity method an ideal technique for
mapping the saline and fresh water interface in coastal areas. One simple equation that gives
the relationship between the resistivity of a porous rock and the fluid saturation factor is
Archie’s Law. It is only applicable for certain types of rocks and sediments, particularly those
that have a low clay content. The electrical conduction is assumed to be through the fluids
filling the pores of the rock. Archie's Law is given by
  a w  m (1.11)
where  is the rock resistivity, w is fluid resistivity,  is the fraction of the rock filled with the
fluid, while a and m are two empirical parameters (Keller and Frischknecht, 1966). For most
rocks, a is about 1 while m is about 2. For sediments with a significant clay content, other more
complex equations have been proposed (Olivar et al., 1990).
The resistivities of several types of ores are also shown. Metallic sulfides (such as
pyrrhotite, galena and pyrite) have typically low resistivity values of less than 1 m. Note that
the resistivity value of a particular ore body can differ greatly from the resistivity of the
individual crystals. Other factors, such as the nature of the ore body (massive or disseminated)
have a significant effect. Note that graphitic slate has a low resistivity value, similar to the
metallic sulfides, which can give rise to problems in mineral surveys. Most oxides, such as
hematite, do not have a significantly low resistivity value. One exception is magnetite.
The resistivity values of several industrial contaminants are also given in Figure 5.

6
Metals, such as iron, have extremely low resistivity values. Chemicals that are strong
electrolytes, such as potassium chloride and sodium chloride, can greatly reduce the resistivity
of ground water to less than 1 m even at fairly low concentrations. The effect of weak
electrolytes, such as acetic acid, is comparatively smaller. Hydrocarbons, such as xylene
(6.998x1016 m), typically have very high resistivity values. However, in practice the
percentage of hydrocarbons in a rock or soil is usually quite small, and might not have a
significant effect on the bulk resistivity. However when the concentration of the hydrocarbon
is high, such as the commercial oil sands deposits in Canada, the resistivity method has proved
to be a useful exploration method for such deposits (see section 7.10 for an example).
Figure 5. The resistivity of rocks, soils and minerals.
1.3 1-D resistivity surveys and inversions – applications, limitations and pitfalls
The resistivity method has its origin in the 1920’s due to the work of the Schlumberger
brothers. For approximately the next 60 years, for quantitative interpretation, conventional
sounding surveys (Koefoed, 1979) were normally used. In this method, the center point of the
electrode array remains fixed, but the spacing between the electrodes is increased to obtain
more information about the deeper sections of the subsurface.
The measured apparent resistivity values are normally plotted on a log-log graph paper.
To interpret the data from such a survey, it is normally assumed that the subsurface consists of
horizontal layers. In this case, the subsurface resistivity changes only with depth, but does not
change in the horizontal direction. A one-dimensional model of the subsurface is used to
interpret the measurements (Figure 6a). Figure 7 shows an example of the data from a sounding
survey and a possible interpretation model. This method has given useful results for geological

7
situations (such the water-table) where the one-dimensional model is approximately true.
The software provided, RES1D.EXE, is a simple inversion and forward modeling
program for 1-D models that consists of horizontal layers. In the software package, several files
with extensions of DAT are example data files with resistivity sounding data. Files with the
MOD extension are model files that can be used to generate synthetic data for the inversion
part of the program. As a first try, read in the file WENNER3.DAT that contains the Wenner
array sounding data for a simple 3-layer model.
Figure 6. The three different models used in the interpretation of resistivity measurements.
Figure 7. A typical 1-D model used in the interpretation of resistivity sounding data for the
Wenner array.
The greatest limitation of the resistivity sounding method is that it does not take into
account lateral changes in the layer resistivity. Such changes are probably the rule rather than
the exception. The failure to include the effect of such lateral changes can results in errors in
the interpreted layer resistivity and/or thickness. As an example, Figure 8 shows a 2-D model
where the main structure is a two-layer model with a resistivity of 10 m and a thickness of 5
meters for the upper layer, while the lower layer has a resistivity of 100 m. To the left of the

8
center point of the survey line, a low resistivity prism of 1 m is added in the upper layer to
simulate a lateral inhomogeneity. The 2-D model has 144 electrodes that are 1 meter apart. The
apparent resistivity pseudosections for the Wenner and Schlumberger array are also shown. For
the Schlumberger array, the spacing between the potential electrodes is fixed at 1.0 meter for
the apparent resistivity values shown in the pseudosection. The sounding curves that are
obtained with conventional Wenner and Schlumberger array sounding surveys with the mid-
point at the center of the line are shown in Figure 9. In the 2-D model, the low resistivity
rectangular prism extends from 5.5 to 18.5 meters to the left of the sounding mid-point. The
ideal sounding curves for both arrays for a two-layer model (i.e. without the low resistivity
prism) are also shown for comparison. For the Wenner array, the low resistivity prism causes
the apparent resistivity values in the sounding curve (Figure 9a) to be too low for spacing
values of 2 to 9 meters and for spacings larger than 15 meters. At spacings of between 9 to 15
meters, the second potential electrode P2 crosses over the low resistivity prism. This causes the
apparent resistivity values to approach the two-layer model sounding curve. If the apparent
resistivity values from this model are interpreted using a conventional 1-D model, the resulting
model could be misleading. In this case, the sounding data will most likely to be interpreted as
a three-layer model.
The effect of the low resistivity prism on the Schlumberger array sounding curve is
slightly different. The apparent resistivity values measured with a spacing of 1 meter between
the central potential electrodes are shown by black crosses in Figure 9b. For electrode spacings
(which is defined as half the total length of the array for the Schlumberger array) of less than
15 meters, the apparent resistivity values are less than that of the two-layer sounding curve. For
spacings greater than 17 meters, the apparent resistivity values tend to be too high. This is
probably because the low resistivity prism lies to the right of the C2 electrode (i.e. outside the
array) for spacings of less than 15 meters. For spacings of greater than 17 meters, it lies between
the P2 and C2 electrodes. Again, if the data is interpreted using a 1-D model, the results could
be misleading. One method that has been frequently recommended to “remove” the effect of
lateral variations with the Schlumberger array is by shifting curve segments measured with
different spacings between the central potential electrodes. The apparent resistivity values
measured with a spacing of 3 meters between the potential electrodes are also shown in Figure
9b. The difference in the sounding curves with the spacings of 1 meter and 3 meters between
the potential electrodes is small, particularly for large electrode spacings. Thus any shifting in
the curve segments would not remove the distortion in the sounding curve due to the low
resistivity prism. The method of shifting the curve segments is probably more applicable if the
inhomogeneity lies between the central potential electrodes, and probably ineffective if the
inhomogeneity is beyond the largest potential electrodes spacing used (which is the case in
Figure 8). However, note that the effect of the prism on the Schlumberger array sounding curve
is smaller at the larger electrode spacings compared with the Wenner array (Figure 9). The
main reason is probably the larger distance between the P2 and C2 electrodes in the
Schlumberger array.
A more reliable method to reduce the effect of lateral variations on the sounding data
is the offset Wenner method (Barker, 1978). It makes use of the property that the effect of an
inhomogeneity on the apparent resistivity value is of opposite sign if it lies between the two
potential electrodes or if it is between a potential and a current electrode. For the example
shown in Figure 8, if the low resistivity body lies in between a current and potential electrode
(the P2 and C2 electrodes in this case), the measured apparent resistivity value would be lower.
If the low resistivity body lies in between the P1 and P2 electrodes, it will cause the apparent
resistivity value to be higher. The reason for this phenomenon can be found in the sensitivity
pattern for the Wenner array (see Figure 23a). By taking measurements with different positions
for the mid-point of the array, the effect of the low resistivity body can be reduced.

9
Figure 8. A 2-D two-layer model with a low resistivity prism in the upper layer. The calculated
apparent resistivity pseudosections for the (a) Wenner and (b) Schlumberger arrays. (c) The
2D model. The mid-point for a conventional sounding survey is also shown.
Another classical survey technique is the profiling method. In this case, the spacing
between the electrodes remains fixed, but the entire array is moved along a straight line. This
gives some information about lateral changes in the subsurface resistivity, but it cannot detect
vertical changes in the resistivity. Interpretation of data from profiling surveys is mainly
qualitative.
The most severe limitation of the resistivity sounding method is that horizontal (or
lateral) changes in the subsurface resistivity are commonly found. The ideal situation shown in
Figure 6a is rarely found in practice. As shown by the examples in Figure 8 and Figure 9,
lateral changes in the subsurface resistivity will cause changes in the apparent resistivity values
that might be, and frequently are, misinterpreted as changes with depth in the subsurface
resistivity. In many engineering and environmental studies, the subsurface geology is very
complex where the resistivity can change rapidly over short distances. The 1-D resistivity

10
sounding method would not be sufficiently accurate for such situations.
Figure 9. Apparent resistivity sounding curves for a 2-D model with a lateral inhomogeneity.
(a) The apparent resistivity curve extracted from the 2D pseudosection for the Wenner array.
The sounding curve for a two-layer model without the low resistivity prism is also shown by
the black line curve. (b) The apparent resistivity curves extracted from the 2-D pseudosection
for the Schlumberger array with a spacing of 1.0 meter (black crosses) and 3.0 meters (red
crosses) between the potential electrodes. The sounding curve for a two-layer model without
the low resistivity prism is also shown.
To use the RES1D.EXE program for the exercises in the table below, as well as the
other programs that we shall use in the later sections, follows the usual sequence used by
Windows XP/Vista/7/8/10. Click the ‘Start’ button, followed by ‘Programs’ and the look for

11
the RES1D folder in the list of installed programs. Alternatively, you can create a shortcut icon
on the Windows Desktop.
Table 1. 1-D inversion examples using the RES1D.EXE program.

Data set and purpose Things to try
WENNER3.DAT – A simple (1). Read in the file, and then run the “Carry out inversion”
synthetic data file for a 3 layer step.
model.
WENN_LATERAL.DAT and (1). Read in the files, and then invert the data sets.
SCHL_LATER.DAT – (2). Compare the results with the true two-layer model
Wenner and Schlumberger (that has resistivities of 10 m and 100 m for the first
array sounding data shown in and second layers, and thickness of 5 meters for the first
Figure 9 that are extracted layer).
from the 2-D pseudosections.
WENOFFSET.DAT – A field (1). Read in the files, and then invert the data set.
data set collected using the
offset Wenner method.
IPTESTM.DAT – A 1-D (1). Read in the files, and then invert the data set.
sounding data file with IP
measurements as well to round
things up.
To obtain a more accurate subsurface model than is possible with a simple 1-D model,
a more complex model must be used. In a 2-D model (Figure 6b), the resistivity values are
allowed to vary in one horizontal direction (usually referred to as the x direction) but assumed
to be constant in the other horizontal (the y) direction. This approximation is reasonable for
survey lines that are perpendicular to the strike of an elongated structure. The most realistic
model would be a fully 3-D model (Figure 6c) where the resistivity values are allowed to
change in all 3 directions. The use of 2-D and 3-D surveys and interpretation techniques will
be examined in detail in the following chapters.
1.4 Basic Inverse Theory

In geophysical inversion, we seek to find a model that gives a response that is similar
to the actual measured values. The model is an idealized mathematical representation of a
section of the earth. The model has a set of model parameters that are the physical quantities
we want to estimate from the observed data. The model response is the synthetic data that can
be calculated from the mathematical relationships defining the model for a given set of model
parameters. All inversion methods essentially try to determine a model for the subsurface
whose response agrees with the measured data subject to certain restrictions and within
acceptable limits. In the cell-based method used by the RES2DINV and RES3DINV programs,
the model parameters are the resistivity values of the model cells, while the data is the measured
apparent resistivity values. The mathematical link between the model parameters and the model
response for the 2-D and 3-D resistivity models is provided by the finite-difference (Dey and
Morrison, 1979a, 1979b) or finite-element methods (Silvester and Ferrari 1990).
In all optimization methods, an initial model is modified in an iterative manner so that
the difference between the model response and the observed data values is reduced. The set of
observed data can be written as a column vector y given by
y  col( y1 , y 2 ,....., y m ) (1.12)

12
where m is the number of measurements. The model response f can be written in a similar form.
f  col( f1 , f 2 ,....., f m ) (1.13)
For resistivity problems, it is a common practice to use the logarithm of the apparent resistivity
values for the observed data and model response, and the logarithm of the model values as the
model parameters. The model parameters can be represented by the following vector
q  col(q1 , q2 ,....., qn ) (1.14)
where n is the number of model parameters. The difference between the observed data and the
model response is given by the discrepancy vector g that is defined by
g=y-f (1.15)
In the least-squares optimization method, the initial model is modified such that the
sum of squares error E of the difference between the model response and the observed data
values is minimized.
n
E  g Tg   g i2 (1.16)
i 1
To reduce the above error value, the following Gauss-Newton equation is used to
determine the change in the model parameters that should reduce the sum of squares error
(Lines and Treitel 1984).
J T J Δqi  J T g (1.17)
where q is the model parameter change vector, and J is the Jacobian matrix (of size m by n)
of partial derivatives. The elements of the Jacobian matrix are given by
f
J ij  i (1.18)
q j
that is the change in the ith model response due to a change in the jth model parameter. After
calculating the parameter change vector, a new model is obtained by
q k 1  q k  Δqk (1.19)
In practice, the simple least-squares equation (1.17) is rarely used by itself in
geophysical inversion. In some situations the matrix product J T J might be singular, and thus
the least-squares equation does not have a solution for q. Another common problem is that
the matrix product J T J is nearly singular. This can occur if a poor initial model that is very
different from the optimum model is used. The parameter change vector calculated using
equation (1.17) can have components that are too large such that the new model calculated with
(1.19) might have values that are not realistic. One common method to avoid this problem is
the Marquardt-Levenberg modification (Lines and Treitel, 1984) to the Gauss-Newton
equation that is given by
J T

J  I Δqk  J T g (1.20)
where I is the identity matrix. The factor  is known as the Marquardt or damping factor, and
this method is also known as the ridge regression method (Inman 1975) or damped least-
squares method. The damping factor effectively constrains the range of values that the
components of parameter change vector can q take. While the Gauss-Newton method in
equation (1.17) attempts to minimize the sum of squares of the discrepancy vector only, the
Marquardt-Levenberg method also minimizes a combination of the magnitude of the
discrepancy vector and the parameter change vector. This method has been successfully used
in the inversion of resistivity sounding data where the model consists of a small number of
layers. For example, it was used in the inversion of the resistivity sounding example in Figure
7 with three layers (i.e. five model parameters). However when the number of model
parameters is large, such as in 2-D and 3-D inversion models that consist of a large number of

13
small cells, the model produced by this method can have an erratic resistivity distribution with
spurious high or low resistivity zones (Constable et al., 1987). To overcome this problem, the
Gauss-Newton least-squares equation is further modified so as to minimize the spatial
variations in the model parameters (i.e. the model resistivity values change in a smooth or
gradual manner). This smoothness-constrained least-squares method (Ellis and Oldenburg
1994a, Loke 2011) has the following mathematical form.
 
J T J F Δqk  J T g  Fq k , (1.21)
where F   x C x C x   y C y C y   z C z C z
T T T
and Cx, Cy and Cz are the roughness filter matrices in the x-, y- and z-directions that couples
the model blocks in those directions (Figure 10, Figure 112). x, y and z are the relative
weights given to the roughness filters in the x-, y- and z-directions. One common form of the
roughness filter matrix is the first-order difference matrix (deGroot-Hedlin and Constable
1990) that is given by
 1 1 0 0 .. .. .. 0
 0 1 1 0 .. .. .. 0

0 0 1 1 0 .. .. 0
 
..
C  (1.22)
 .. 
 
 .. 
 .. 
 
 0
Figure 10. Coupling between neighboring model cells through the roughness filter in a 2-D
model. (a) In the horizontal and vertical diretcions only, and (b) in diagonal diretcions as well.
Equation 1.21 also tries to minimize the square of the spatial changes, or roughness, of
the model resistivity values. It is in fact an l2 norm smoothness-constrained optimization
method. This tends to produce a model with a smooth variation of resistivity values. This
approach is acceptable if the actual subsurface resistivity varies in a smooth or gradational
manner. In some cases, the subsurface geology consists of a number of regions that are
internally almost homogeneous but with sharp boundaries between different regions. For such

14
cases, the inversion formulation in (1.21) can be modified so that it minimizes the absolute
changes in the model resistivity values (Claerbout and Muir, 1973). This can sometimes give
significantly better results. Technically this is referred to as an l1 norm smoothness-constrained
optimization method, or more commonly known as the blocky inversion method. A number of
techniques can be used for such a modification. One simple method to implement an l1 norm
based optimization method using the standard least-squares formulation is the iteratively
reweighted least-squares method (Wolke and Schwetlick, 1988; Farquharson and Oldenburg,
1998). The optimization equation in (1.21) is modified to
J T

R d J FR Δqk  J T R d g  FR q k , (1.23)
with FR   x C Tx R m C x   y C Ty R m C y   z C Tz R m C z
where Rd and Rm are weighting matrices introduced so that different elements of the data misfit
and model roughness vectors are given equal weights in the inversion process.
Equation (1.23) provides a general method that can be further modified if necessary to
include known information about the subsurface geology. As an example, if it is known that
the variations in the subsurface resistivity are likely to be confined to a limited zone, the
damping factor values can modified (Ellis and Oldenburg 1994a) such that greater changes are
allowed in that zone. If the errors in the data points are known, a diagonal weighting matrix
can be used to give greater weights to data points with smaller errors.
One important modification to the least-squares optimization method is in time-lapse
inversion using the following equation (Kim et al. 2009, Loke et al. 2014a).
J T
i    
R d J i  λ i FR  α M T R t M Δri  J Ti R d g i  λ i FR  α i M T R t M r i 1 (1.24)
M is the difference matrix applied across the time models with only the diagonal and one sub
diagonal elements having values of 1 and -1, respectively, while r is the combined resistivity
model for all the time series.  is the temporal damping factor that gives the weight for
minimizing the temporal changes in the resistivity compared to the model roughness and data
misfit. Note the smoothness-constraint is not only applied in space through the F matrix, but
also across the different time models through the M matrix. This is illustrated in Figure 11.
One common variation is to apply a constraint such that the model is ‘close’ to a
reference model, qR. Equation (1.24) then then modified to the following form.
J T J FR Δqk  J T R dg  (FR  I) q k  q R  (1.25)
qR is usually a homogeneous background model. This equation imposes and additional
constraint the new model is close to the reference model with the damping factor weight α. Its
effect is similar to the Marquardt constraint in equation 1.20, and it prevents very large
deviations from the reference model.
The smoothness-constrained least-squares optimization method involves the damping
factor term . This term balances the need to reduce the data misfit (so that the calculated
apparent resistivity values are as close as possible to the measured values) while producing a
model that is ‘reasonably’ smooth and perhaps more geologically realistic. A smaller value of
 term will generally produce a model with a lower data misfit but usually at the expense of
larger variations in the model resistivity values. If the error of the measured apparent resistivity
values is known, a prudent approach might be to select a model where the data misfit is similar
to the known measurement errors. However, for most field data sets, the measurement error is
not known. There are generally two methods to automatically select the ‘optimum’ damping
factor for such cases, the GCV and L-curve methods (Farquharson and Oldenburg, 2004). For

15
the inversion of a single data set, either method can be used. However, for time-lapse data sets
(Loke et al. 2014a), the L-curve method has a clear advantage due to the sparse block structure
of the Jacobian matrix (Figure 11b). The GCV method requires a matrix inversion. However,
the inverse of a sparse matrix is usually a full matrix. This makes it impractical to use for time-
lapse models with many time-series measurements such that the combined model might have
hundreds of thousands of model parameters (i.e. the number of model parameters for a single
data set multiplied by the number of time series data sets).
Figure 11. (a) Coupling between corresponding model blocks in two time-lapse models using
a cross-model time-lapse smoothness constraint. (b) Example Jacobian matrix structure for five
time series data sets and models. Each grey rectangle represents the Jacobian matrix associated
with a single set of measurements.
1.5 2-D model discretization methods

In the previous section, we have seen that the least-squares method is used to calculate
certain physical characteristics of the subsurface (the “model parameters”) from the apparent
resistivity measurements. The “model parameters” are set by the way we slice and dice the
subsurface into different regions. Figure 12 shows various possibilities that can be used. The
method most commonly used in 2-D (and 3-D) interpretation is a purely cell based model where
the subsurface is subdivided into rectangular cells. The positions of the cells are fixed and only
the resistivities of cells are allowed to vary during the inversion process. The model parameters
are the resistivities the cells. In the example shown in Figure 12, the model parameters are the
seventy-two cell resistivity values 1 to 72.
A radically different approach is a boundary based inversion method. This method
subdivides the subsurface into different regions. The resistivity is assumed to be homogenous
within each region. The resistivity is allowed to change in an arbitrary manner across the
boundaries, and thus it useful in areas with abrupt transitions in the geology. The resistivity of
each region and the depths to the boundaries are changed by the least-squares optimization
method so that the calculated apparent resistivity values match the observed values. The “model
parameters” for the example shown in Figure 12 are the two resistivity values (1 and 2) and
the depths at five points (z1 to z5) along the boundary that gives a total of seven parameters.
While this method works well for synthetic data from numerical models, for many field data
sets it can lead to unstable results with highly oscillating boundaries (Olayinka and Yaramanci,
2000). Its greatest limitation is probably the assumption of a constant resistivity within each
region. In particular, lateral changes in the resistivity near the surface have a very large effect
on the measured apparent resistivity values. Since this model does not take into account such

16
lateral changes, they are often mistakenly modeled as changes in the depths of the boundaries.
More recent efforts have been in combining the cell based and boundary based
inversion methods (Smith et al., 1999). One such method is the laterally constrained inversion
method (Auken and Christiansen, 2004). In this method, lateral changes (but not vertical
changes) are allowed in each region (Figure 12c), and abrupt transitions across the boundaries
are also allowed. The “model parameters” for the example in Figure 12 are then the twenty-
four resistivity values (1 and 24) and the depths at thirteen points (z1 to z13) along the boundary
giving a total of thirty-seven parameters. Information from other sources, such as borehole or
seismic data, can be used to provide an initial estimate of the depth to the boundary. A common
situation is when the depth information is available at only one borehole. In this case, the initial
boundary is usually set a constant depth. The inversion method then adjusts the depths at a
number of points along the boundary during the inversion process. A smoothness-constraint is
applied to minimize changes in the depths between adjacent points on the same boundary
(Smith et al., 1999). This method works particularly well where the subsurface consists of
several sedimentary zones.
A further generalization of this concept is to allow both vertical and lateral changes
within each region (as in a pure cell based model) while also allowing sharp changes across the
boundaries (Figure 12d). The model shown in Figure 12d has seventy-two resistivity values
and five depth values, giving a total of seventy-seven model parameters. This type of
discretization is particularly useful where near-surface inhomogeneities that occur at different
depths within the top layer have a large effect on the measured apparent resistivity values
(Figure 13). The model cells used so far has rectangular shapes (Figure 12). This is partly due
to the use of the finite-difference method in calculating the model apparent resistivity values.
A slight disadvantage is that the boundary is approximated by a series of rectangular steps.
Figure 12e shows a possible variation using the finite-element method with trapezoidal cells
where the edges of the cells adjacent to the boundary are adjusted so as to conform to the true
shape of the boundary.
Figure 12. The different models for the subsurface used in the interpretation of data from 2-D
electrical imaging surveys. (a) A purely cell based model. (b) A purely boundary based model.
(c) The laterally constrained model. (d) A combined cell based and boundary based model with
rectangular cells, and (e) with boundary conforming trapezoidal cells.

17
Figure 13. Example of a cell based model with a variable boundary. (a) The test model. (b)
The apparent resistivity pseudosection. (c) The inversion model with a variable sharp boundary
that is marked by a black line.

18
2 2-D electrical surveys – Data acquisition, presentation and

arrays
2.1 Introduction
We have seen that the greatest limitation of the resistivity sounding method is that it
does not take into account horizontal changes in the subsurface resistivity. A more accurate
model of the subsurface is a two-dimensional (2-D) model where the resistivity changes in the
vertical direction, as well as in the horizontal direction along the survey line. It is assumed that
resistivity does not change in the direction that is perpendicular to the survey line. In many
situations, particularly for surveys over elongated geological bodies, this is a reasonable
assumption. In theory, a 3-D resistivity survey and interpretation model should be even more
accurate. However, at the present time, 2-D surveys are the most practical economic
compromise between obtaining very accurate results and keeping the survey costs down
(Dahlin, 1996). Typical 1-D resistivity sounding surveys usually involve about 10 to 20
readings, while 2-D imaging surveys involve about 100 to 1000 measurements. In comparison,
a 3-D survey might involve several thousand measurements.
The cost of a typical 2-D survey could be several times the cost of a 1-D sounding
survey, and is probably comparable with a seismic refraction survey. In many geological
situations, 2-D electrical imaging surveys can give useful results that are complementary to the
information obtained by other geophysical methods. For example, seismic methods can map
undulating interfaces well, but will have difficulty (without using advanced data processing
techniques) in mapping discrete bodies such as boulders, cavities and pollution plumes. Ground
radar surveys can provide more detailed pictures but have very limited depth penetration in
areas with conductive unconsolidated sediments, such as clayey soils. Two-dimensional
electrical surveys can be used in conjunction with seismic or GPR surveys as they provide
complementary information about the subsurface.
2.2 Field survey method - instrumentation and measurement procedure

Two-dimensional electrical imaging/tomography surveys are usually carried out using
a large number of electrodes, 25 or more, connected to a multi-core cable (Griffiths and Barker,
1993). A resistivity meter system with an internal microprocessor controlled circuitry together
with an electronic switching unit is commonly used to automatically select the relevant four
electrodes for each measurement (Figure 14). At present, field techniques and equipment to
carry out 2-D resistivity surveys are fairly well developed. The necessary field equipment is
commercially available from a number of companies. These systems typically costs from about
US$15,000 upwards. Some institutions have even constructed “home-made” manually
operated switching units at a nominal cost by using a seismic cable as the multi-core cable!
Figure 14 shows the typical setup for a 2-D survey with a number of electrodes along a
straight line attached to a multi-core cable. Normally a constant spacing between adjacent
electrodes is used. The multi-core cable is attached to an integrated resistivity meter system
that includes an electronic switching unit. The sequence of measurements to take, the type of
array used and other survey parameters (such the current to use) is normally transferred to an
internal microprocessor system within the resistivity meter from a personal computer. After
reading the control file, the control program then automatically selects the appropriate
electrodes for each measurement.
In a typical survey, most of the fieldwork is in laying out the cable and electrodes. After
that, the measurements are taken automatically and stored in the resistivity meter system. Most
of the survey time is spent waiting for the resistivity meter to complete the set of measurements!
To obtain a good 2-D picture of the subsurface, the coverage of the measurements must
be 2-D as well. As an example, Figure 14 shows a possible sequence of measurements for the

19
Wenner electrode array for a system with 20 electrodes. In this example, the spacing between
adjacent electrodes is “a”. The first step is to make all the possible measurements with the
Wenner array with an electrode spacing of “1a”. For the first measurement, electrodes number
1, 2, 3 and 4 are used. Notice that electrode 1 is used as the first current electrode C1, electrode
2 as the first potential electrode P1, electrode 3 as the second potential electrode P2 and
electrode 4 as the second current electrode C2. For the second measurement, electrodes number
2, 3, 4 and 5 are used for C1, P1, P2 and C2 respectively. This is repeated down the line of
electrodes until electrodes 17, 18, 19 and 20 are used for the last measurement with “1a”
spacing. For a system with 20 electrodes, note that there are 17 (20 - 3) possible measurements
with “1a” spacing for the Wenner array.
After completing the sequence of measurements with “1a” spacing, the next sequence
of measurements with “2a” electrode spacing is made. First electrodes 1, 3, 5 and 7 are used
for the first measurement. The electrodes are chosen so that the spacing between adjacent
electrodes is “2a”. For the second measurement, electrodes 2, 4, 6 and 8 are used. This process
is repeated down the line until electrodes 14, 16, 18 and 20 are used for the last measurement
with spacing “2a”. For a system with 20 electrodes, note that there are 14 (20 - 2x3) possible
measurements with “2a” spacing.
Figure 14. The arrangement of electrodes for a 2-D electrical survey and the sequence of
measurements used to build up a pseudosection.
The same process is repeated for measurements with “3a”, “4a”, “5a” and “6a”
spacings. To get the best results, the measurements in a field survey should be carried out in a
systematic manner so that, as far as possible, all the possible measurements are made. This will
affect the quality of the interpretation model obtained from the inversion of the apparent
resistivity measurements (Dahlin and Loke, 1998).
Note that as the electrode spacing increases, the number of measurements decreases.
The number of measurements that can be obtained for each electrode spacing, for a given

20
number of electrodes along the survey line, depends on the type of array used. The Wenner
array gives the smallest number of possible measurements compared to the other common
arrays that are used in 2-D surveys.
The survey procedure with the pole-pole array is similar to that used for the Wenner
array. For a system with 20 electrodes, firstly 19 of measurements with a spacing of “1a” are
made, followed by 18 measurements with “2a” spacing, followed by 17 measurements with
“3a” spacing, and so on.
For the dipole-dipole, Wenner-Schlumberger and pole-dipole arrays (Figure 1.4), the
survey procedure is slightly different. As an example, for the dipole-dipole array, the
measurement usually starts with a spacing of “1a” between the C1-C2 (and also the P1-P2)
electrodes. The first sequence of measurements is made with a value of 1 for the “n” factor
(which is the ratio of the distance between the C1-P1 electrodes to the C1-C2 dipole length),
followed by “n” equals to 2 while keeping the C1-C2 dipole pair spacing fixed at “1a”. When
“n” is equals to 2, the distance of the C1 electrode from the P1 electrode is twice the C1-C2
dipole length. For subsequent measurements, the “n” spacing factor is usually increased to a
maximum value of about 6, after which accurate measurements of the potential are difficult
due to very low potential values. To increase the depth of investigation, the spacing between
the C1-C2 dipole pair is increased to “2a”, and another series of measurements with different
values of “n” is made. If necessary, this can be repeated with larger values of the spacing of
the C1-C2 (and P1-P2) dipole pairs. A similar survey technique can be used for the Wenner-
Schlumberger and pole-dipole arrays where different combinations of the “a” spacing and “n”
factor can be used.
It should be noted that in practice the sequence of measurements to take should be
arranged so that electrode polarization effects do not occur (Dahlin, 2000). This happens when
an electrode that was used as a current electrode is used as potential electrode in a following
measurement within a short time.
One technique used to extend horizontally the area covered by the survey, particularly
for a system with a limited number of electrodes, is the roll-along method. After completing
the sequence of measurements, the cable is moved past one end of the line by several unit
electrode spacings. All the measurements that involve the electrodes on part of the cable that
do not overlap the original end of the survey line are repeated (Figure 15).
Figure 15. The use of the roll-along method to extend the area covered by a 2-D survey.

21
2.3 Available field instruments

Over the last 15 years, there has been a steady growth in the number of commercial
companies that offer systems for resistivity imaging surveys. The ones that I have come across
are listed below in alphabetical order together with the web site address where available.
Manufacturer Instrument Type

Static Dynamic IP
Abem Instruments, Sweden (www.abem.se) x Y
Advanced Geohysical Instruments, USA (www.agiusa.com) x Y
Allied Associates, UK (www.allied-associates.co.uk) x Y
Geofyzika., Czech Republic (www.geofyzika.com) x
GF Instruments, Czech Republic (www.gfinstruments.cz) x Y
Geometrics, USA (www.geometrics.com) x
Geolog, Germany (www.geolog2000.de) x
IDS Scintrex, Canada (www.idsdetection.com) x Y
Iris Instruments, France (www.iris-instruments.com) x x Y
OYO, Japan (www.oyo.co.jp) x
Pasi Geophysics, Italy (www.pasigeophysics.com) x Y
MAE srl, Italy (www.mae-srl.it) x Y
ST Geomative Co.,Ltd, China (www.geomative.com) x Y
Landviser LLC, USA/Russia (www.landviser.net) x Y
Quantec Geoscience, Canada (www.quantecgeoscience.com) x Y
Zonge International, USA (zonge.com) x Y
Most of the above manufacturers have sub-agents in different countries, and there are
probably a few others that are not on the list. An interesting recent development is the
availability of lower cost Russian and Chinese systems. A few academic and research
institutions have designed their own systems, for example the British Geological Survey has
built an electrostatic based mobile system as well as an automatic system for time-lapse
surveys.
The instrument type can be divided into two broad categories, static and dynamic
systems. Most instruments are of the static type where many electrodes are connected to a
multi-electrode cable and planted into the ground during the survey. A typical static system is
the Abem Lund system shown in Figure 16. One common configuration is a split spread type
of cable connection to the switching unit at the center to reduce the individual cable length and
weight. The weight of a cable roll is directly proportional to the number of nodes and the
spacing between the nodes! A common spacing used for most engineering and environmental
surveys is 5 meters. Most systems come with a minimum of 28 nodes, with some system having
up to 128 nodes or more! The Lund system is a little unusual in that there are 4 individual
cables. Most systems use a 2 cables arrangement. The static systems can be further divided into
two sub-categories depending on the arrangement for the switching of the electrodes. Most of
the systems house the switches in a single unit and uses a cable with many individual wires
connected to each node. Typical examples are the Abem Lund and Campus systems. Another
arrangement is to have a small switching unit at each electrode and a cable with the minimum
number of wires. One early example is the Campus MRT system (Griffiths and Turnbull,

22
1985). More recent examples are the PASI and Iris Syscal systems.
There have been two new and interesting developments in the resistivity meter systems.
One is the addition of I.P. capability. However, the usefulness of the I.P. data from most multi-
electrode systems is of limited use. Most systems use a battery power source that cannot deliver
enough current (usually less than 1 ampere) for reliable I.P. signals, so the I.P. data is often
extremely noisy. However, there have been recent developments to reduce the noise in the I.P.
measurements (Dahlin and Leroux, 2012).
The second new development is multi-channel measuring systems. In such a system, a
number of potential measurements can be simultaneously made for a single pair of current
electrodes. This could significantly reduce the survey time. With certain array configurations,
a single 2-D survey line could involve thousands of measurements. The major part of the survey
time is waiting for a single channel instrument to complete the measurements that could take
more than several hours! The IP and multi-channel capability are relatively new developments,
so you will need to check the manufacturer's web site to get the latest information.
Figure 16. Sketch outline of the ABEM Lund Imaging System. Each mark on the cables
indicates an electrode position (Dahlin, 1996). The cables are placed along a single line (the
sideways shift in the figure is only for clarity). This figure also shows the principle of moving
cables when using the roll-along technique.
Static systems use a large number of nodes to get a wide data coverage. In contrast,
dynamic systems use a small number of nodes but move the entire system to obtain a wide
coverage. An example of such a system designed by Aarhus University in Denmark (Sorenson,
1996) is shown in Figure 17. A 100 meters cable with nine heavy cylindrical electrodes is
pulled by a small vehicle. Two of the electrodes are used as current electrodes, while six of
them are used for the potential measurements and one is used as a ground electrode. This
system relies on the current being injected into the ground by direct contact, so it can only be
used in open ground, such as farmlands in Northern Europe. A Wenner-Schlumberger type of
arrangement (Figure 4) is used but with non-integer “n” values for some of the measurements.
Another mobile system that does not require direct contact with the ground but uses capacitive
coupling (Gerard and Tabbagh, 1991; Shima et al., 1996, Panissod et al., 1998; Loke et al.,
2011a) to induce the flow of current in the ground. This system can be used in areas that are
paved, such as roads and city areas. One such system shown in Figure 18 is the Geometrics
OhmMapper system where a cable with 4 to 6 electrodes is attached to a measuring unit is
pulled by a single operator. The dipole-dipole type of arrangement is used but with non-integer
“n” values for some measurements.

23
One of main problems faced by mobile systems on land is to get sufficient current to
flow into the ground. Direct contact systems such as the Aarhus Pulled Array System can only
be used in areas with open ground. The capacitive coupling type does not require direct ground
contact and thus can be used in many areas where normal resistivity surveying systems cannot
be used (for example in built-up areas) but has the problem of a more limited depth of
penetration due to the limited amount of current that can be induced into the ground compared
to direct contact systems. An underwater environment provides an almost ideal situation for a
direct contact type of mobile system since there is no problem in obtaining good electrode
contact! Figure 19 shows a possible arrangement for an underwater mobile surveying system
where a cable with a number of nodes is pulled along the river/lake/sea bottom by a boat. Two
of the nodes are used as current electrodes, while the rest are used as potential electrodes. An
example of such an underwater survey is described in section 7.9. If this system is coupled with
a multi-channel resistivity meter, the survey can be carried out very rapidly. Shallow seismic
reflection surveys are frequently used in rivers/lakes/marine environments for engineering site
surveys. A mobile resistivity survey might be a useful addition in some situations, such as in
seismically opaque areas. In theory, both surveys can be carried out simultaneously to reduce
costs. The arrangement in Figure 19 has the cable dragged along the sea or river bed. This
places the electrodes close to the targets of interest but has the disadvantage that the cable can
get snagged be obstructions on the bottom. Another system uses a cable floating on the water
surface (section 7.9) but this has the disadvantage that most of the current flows within the
water layer, and very little goes into the sub-bottom materials which are being mapped. To
avoid both problems, some recent surveys use a submerged cable that is suspended about a
meter above the water bottom.
pulling vehicle
c c
c
current elec.
potential elec.
Figure 17. The Aarhus Pulled Array System. The system shown has two current (C) electrodes
and six potential electrodes (Christensen and Sørensen 1998, Bernstone and Dahlin 1999).
Figure 18. The Geometrics OhmMapper system using capacitive coupled electrodes.
(Courtesy of Geometrics Inc.).

24
Figure 19. Schematic diagram of a possible mobile underwater survey system. The cable has
two fixed current electrodes and a number of potential electrodes so that measurements can be
made at different spacings. The above arrangement uses the Wenner-Schlumberger type of
configuration. Other configurations, such as the gradient array, can also be used.
2.4 Pseudosection data plotting method

To plot the data from a 2-D imaging survey, the pseudosection contouring method is
normally used. In this case, the horizontal location of the data point is placed at the mid-point
of the set of electrodes used to make that measurement. The vertical location of the plotting
point is placed at a distance that is proportional to the separation between the electrodes. For
I.P. surveys using the dipole-dipole array, one common method is to place the plotting point at
the intersection of two lines starting from the mid-point of the C1-C2 and P1-P2 dipole pairs
with a 45 angle to the horizontal. It is important to emphasize that this is merely a plotting
convention, and it does not imply that the depth of investigation is given by the point of
intersection of the two 45 angle lines (it certainly does not imply the current flow or
isopotential lines have a 45 angle with the surface). Surprisingly, this is still a common
misconception, particularly in North America!
Another method is to place the vertical position of the plotting point at the median depth
of investigation (Edwards, 1977), or pseudodepth, of the electrode array used. This
pseudodepth value is based on the sensitivity values or Frechet derivative for a homogeneous
half space. Since it appears to have some mathematical basis, this method that is used in plotting
the pseudosections in the later parts of these notes. The pseudosection plot obtained by
contouring the apparent resistivity values is a convenient means to display the data.
The pseudosection gives a very approximate picture of the true subsurface resistivity
distribution. However the pseudosection gives a distorted picture of the subsurface because the
shapes of the contours depend on the type of array used as well as the true subsurface resistivity
(Figure 20). The pseudosection is useful as a means to present the measured apparent resistivity
values in a pictorial form, and as an initial guide for further quantitative interpretation. One
common mistake made is to try to use the pseudosection as a final picture of the true subsurface
resistivity. As Figure 20 shows, different arrays used to map the same region can give rise to
very different contour shapes in the pseudosection plot. Figure 20 also gives you an idea of the
data coverage that can be obtained with different arrays. Note that the pole-pole array gives the
widest horizontal coverage, while the coverage obtained by the Wenner array decreases much
more rapidly with increasing electrode spacing.
One useful practical application of the pseudosection plot is for picking out bad
apparent resistivity measurements. Such bad measurements usually stand out as points with
unusually high or low values.

25
Figure 20. The apparent resistivity pseudosections from 2-D imaging surveys with different
arrays over a rectangular prism.

26
2.5 A comparison of the different electrode arrays

As shown earlier in Figure 20, the shape of the contours in the pseudosection produced
by the different arrays over the same structure can be very different. The arrays most commonly
used for resistivity surveys were shown in Figure 4. The choice of the “best” array for a field
survey depends on the type of structure to be mapped, the sensitivity of the resistivity meter
and the background noise level. In practice, the arrays that are most commonly used for 2-D
imaging surveys are the (a) Wenner, (b) dipole-dipole (c) Wenner-Schlumberger (d) pole-pole
and (d) pole-dipole. Among the characteristics of an array that should be considered are (i) the
depth of investigation, (ii) the sensitivity of the array to vertical and horizontal changes in the
subsurface resistivity, (iii) the horizontal data coverage and (iv) the signal strength.
2.5.1 The Frechet derivative for a homogeneous half-space

The first two characteristics can be determined from the sensitivity function of the array
for a homogeneous earth model. The sensitivity function basically tells us the degree to which
a change in the resistivity of a section of the subsurface will influence the potential measured
by the array. The higher the value of the sensitivity function, the greater is the influence of the
subsurface region on the measurement. Mathematically, the sensitivity function is given by the
Frechet derivative (McGillivray and Oldenburg, 1990). Consider the simplest possible array
configuration shown in Figure 21 with just one current located at the origin (0,0,0) and one
potential electrode located at (a,0,0), i.e. both electrodes are on the ground surface and they are
“a” meters apart.. We inject 1 ampere of current into the ground through the C1 current
electrode that results in a potential  observed at the potential P1 electrode. Suppose we were
to change the resistivity within a small volume of the ground located at (x,y,z) by a small
amount, say . What would be the corresponding change in the potential, , measured at P1?
It can be shown (Loke and Barker, 1995) that this is given by
δρ
2 V
δ     ' d (2.1)
ρ
where the change in the resistivity has a constant value in a small volume element d and zero
elsewhere. The parameter ’ is the potential resulting from a current electrode located at the
position of the P1 potential electrode.
Figure 21. The parameters for the sensitivity function calculation at a point (x,y,z) within a half-
space. A pole-pole array with the current electrode at the origin and the potential electrode “a”
meters away is shown.
For the special case of a homogeneous half-space, the potential  at a point in the half-
space due to a unit current source on the surface has a relatively simple form, which is

27

 , and similarly

2 x 2  y 2  z 2 
0.5

'  (2.2)

2  x  a   y 2  z 2
2

0.5
After differentiating the above equations to obtain the divergence, and substituting into (2.1)
we get
 1 x x  a   y 2  z 2
  dxdydz (2.3)
  
 V 4 2 x 2  y 2  z 2 1.5 x  a 2  y 2  z 2 
1.5
The 3-D Frechet derivative is then given by the term within the integral, i.e.
1 x x  a   y 2  z 2
F3D x, y, z    (2.4)
  
4 2 x 2  y 2  z 2 1.5 x  a 2  y 2  z 2 
1.5
This gives the Frechet derivative or sensitivity function for the pole-pole array consisting of
just one current and one potential electrode. To obtain the Frechet derivative for a general four
electrodes array, we need to just add up the contributions from the four current-potential pairs,
just as we have done earlier for the potential in equation (1.8).
2.5.2 A 1-D view of the sensitivity function - depth of investigation

In resistivity sounding surveys, it is well known as the separation between the
electrodes is increased, the array senses the resistivity of increasingly deeper layers. One
common question is – What is a depth of investigation of an array? One quantitative means to
put a numerical value for the depth of investigation is by using the sensitivity function or
Frechet derivative of the array. In resistivity sounding surveys, the subsurface is assumed to
consist of horizontal layers. What we want to determine is the change in the potential as
measured by the array on the surface if the resistivity of a thin horizontal is changed. For a
horizontal layer, the x and y limits of the layer extends from - to +. Thus the sensitivity
function for a thin horizontal layer is obtained by integrating the 3D sensitivity function given
in equation (2.4) in the x and y directions, i.e.
x x  a   y 2  z 2

1
F1D z  
4 2  x 2  y 2  z 2 1.5 x  a 2  y 2  z 2 1.5
dxdy
  
The above equation has a simple analytical solution (Roy and Apparao, 1971), which is given
by
2 z
F1D z    . (2.5)

 a 2  4z 2 1.5 
The above function is also known as the depth investigation characteristic and has been used
by many authors to determine the properties of various arrays in resistivity sounding surveys
(Edwards, 1977; Barker, 1989; Merrick, 1997). Figure 22a shows a plot of this function. Note
that it starts from zero and then increases to a maximum value at a depth of about 0.35a and
then decreases asymptotically to zero. Some authors have used the maximum point as the depth
of investigation of the array. However, Edwards (1977) and Barker (1991) have shown that a
more robust estimate is the "median depth of investigation". It is the depth above which the
area under the curve is equal to half the total area under the curve. In layman's terms, the upper

28
section of the earth above the "median depth of investigation" has the same influence on the
measured potential as the lower section. This tells us roughly how deep we can see with an
array. This depth does not depend on the measured apparent resistivity or the resistivity of the
homogeneous earth model. It should be noted that the depths are strictly only valid for a
homogeneous earth model, but they are probably good enough for planning field surveys. If
there are large resistivity contrasts near the surface, the actual depth of investigation could be
somewhat different.
The sensitivity function for other arrays can be determined by adding up the
contributions from the appropriate four pairs of current-potential electrodes. Figure 22b shows
the sensitivity function plot for the Wenner (alpha) array. Note that the curve around the
maximum is narrower for the Wenner array compared with the pole-pole array. This implies
that the Wenner array has a better vertical resolution than the pole-pole array.
Table 2 gives the median depth of investigation for the different arrays. To determine
the maximum depth mapped by a particular survey, multiply the maximum “a” electrode
spacing, or maximum array length “L“, by the appropriate depth factor given in Table 2. For
example, if the maximum electrode “a” spacing used by the Wenner array is 100 meters (or
maximum L 300 meters), then the maximum depth mapped is about 51 meters. For the dipole-
dipole, pole-dipole and Wenner-Schlumberger arrays, the “n” factor (Figure 4) must also be
taken into consideration. For the arrays with four active electrodes (such as the dipole-dipole,
Wenner and Wenner-Schlumberger arrays), it is probably easier to use the total array length
“L”. As an example, if a dipole-dipole survey uses a maximum value of 10 meters for “a” and
a corresponding maximum value of 6 for n, then the maximum “L” value is 80 meters. This
gives a maximum depth of investigation of 80x0.216 or about 17 meters.
Table 2 also includes the geometric factor for the various arrays for an "a" spacing of
1.0 meter. The inverse of the geometric factor gives an indication of the voltage that would be
measured between the P1 and P2 potential electrodes. The ratio of this potential compared to
the Wenner alpha array is also given, for example a value of 0.01 means that the potential is
1% of the potential measured by the Wenner alpha array with the same "a" spacing.
Figure 22. A plot of the 1-D sensitivity function. (a) The sensitivity function for the pole-
pole array. Note that the median depth of investigation (red arrow) is more than twice the
depth of maximum sensitivity (blue arrow). (b) The sensitivity function and median depth of
investigation for the Wenner array.

29
Table 2. The median depth of investigation (ze) for the different arrays (Edwards, 1977). L is
the total length of the array. Note identical values of ze/a for the Wenner-Schlumberger and
pole-dipole arrays. Please refer to Figure 4 for the arrangement of the electrodes for the
different arrays. The geometric factor is for an "a" value of 1.0 meter. For the pole-dipole array,
the array length ‘L’ only takes into account the active electrodes C1, P1 and P2 (i.e. it does not
take into account the remote C2 electrode).
z e/a z e/L Geometric Inverse Geometric
Factor Factor (Ratio)
Wenner Alpha 0.519 0.173 6.2832 0.15915 (1.0000)
Wenner Beta 0.416 0.139 18.850 0.05305 (0.3333)
Wenner Gamma 0.594 0.198 9.4248 0.10610 (0.6667)
Dipole-dipole n=1 0.416 0.139 18.850 0.05305 (0.3333)

n=2 0.697 0.174 75.398 0.01326 (0.0833)
n=3 0.962 0.192 188.50 0.00531 (0.0333)
n=4 1.220 0.203 376.99 0.00265 (0.0166)
n=5 1.476 0.211 659.73 0.00152 (0.0096)
n=6 1.730 0.216 1055.6 0.00095 (0.0060)
n=7 1.983 0.220 1583.4 0.00063 (0.0040)
n=8 2.236 0.224 2261.9 0.00044 (0.0028)
Equatorial dipole-dipole
n=1 0.451 0.319 21.452 0.04662 (0.2929)
n=2 0.809 0.362 119.03 0.00840 (0.0528)
n=3 1.180 0.373 367.31 0.00272 (0.0171)
n=4 1.556 0.377 841.75 0.00119 (0.0075)
Wenner - Schlumberger
n=1 0.519 0.173 6.2832 0.15915 (1.0000)
n=2 0.925 0.186 18.850 0.05305 (0.3333)
n=3 1.318 0.189 37.699 0.02653 (0.1667)
n=4 1.706 0.190 62.832 0.01592 (0.1000)
n=5 2.093 0.190 94.248 0.01061 (0.0667)
n=6 2.478 0.191 131.95 0.00758 (0.0476)
n=7 2.863 0.191 175.93 0.00568 (0.0357)
n=8 3.247 0.191 226.19 0.00442 (0.0278)
n=9 3.632 0.191 282.74 0.00354 (0.0222)
n = 10 4.015 0.191 345.58 0.00289 (0.0182)
Pole-dipole n=1 0.519 0.260 12.566 0.07958 (0.5000)

n=2 0.925 0.308 37.699 0.02653 (0.1667)
n=3 1.318 0.330 75.398 0.01326 (0.0833)
n=4 1.706 0.341 125.66 0.00796 (0.0500)
n=5 2.093 0.349 188.50 0.00531 (0.0334)
n=6 2.478 0.354 263.89 0.00379 (0.0238)
n=7 2.863 0.358 351.86 0.00284 (0.0178)
n=8 3.247 0.361 452.39 0.00221 (0.0139)
Pole-Pole 0.867 6.28319 0.15915 (1.0000)

30
2.5.2b The median depth of investigation and the Gamma array problem
The use of the median depth of investigation runs into an interesting problem with the
Gamma array type (Carpenter and Habberjam, 1956) for certain configurations. It gives
ridiculously large and even infinite values for certain electrode arrangements (Barker, 1989).
This is because the sensitivity function changes from positive to negative with the depth.
2.5.3 A 2-D view of the sensitivity function

The plot of the 1-D sensitivity function in Figure 22 suggests that the sensitivity of an
array to the topmost layer is very small. The plot actually gives the net contribution calculated
by summing up the contribution for all x- and y-values at the same depth, and it hides a
multitude of effects. The net contribution for the topmost strip is small only if the ground is
completely homogeneous. If it is not homogeneous, the results can be very different.
To study the suitability of the different arrays for 2-D surveys, we need to go one step
beyond the simple 1-D sensitivity function, i.e. the 2-D sensitivity function. In this case, for a
particular (x,z) location, we add up the contribution from all points for y-values ranging from
+ to -. This involves the integration of the 3-D sensitivity function in equation (2.4) with
respect to y, which is
x x  a   y 2  z 2

1
F2 D x, z   2  dy
   
4  x 2  y 2  z 2 1.5 x  a 2  y 2  z 2 1.5
This integral has an analytic solution (Loke and Barker 1995) that is given in terms of elliptic
integrals. The complete solution is
1
F2 D x, z   2 2  
  
 2 E k    2 K k    2   2 E k   2 2 K k  

2   
2 2  
2 2
2
 
(2.6)
where
k
 2
 2 
0.5

for x>0.5a,
 2  x 2  z 2 ,  2  x  a   z 2 ,   xa
2
and for x<0.5a,

 2  x 2  z 2 ,  2   x  a   z 2 ,   a x  a 
2
and for x=0.5a,

1  1 3a 2 
F2 D x, z   3  5
, with  2  0.25a 2  z 2
4  2 16 
As an example, Figure 23a shows the contour pattern for the sensitivity function of the
Wenner array. The sensitivity function shows the degree to which a change in the resistivity of
a section of the subsurface will influence the potential measured by the array. The higher the
value of the sensitivity function, the greater is the influence of the subsurface region on the
measurement. Note that for all the three arrays, the highest sensitivity values are found near the
electrodes. At larger distances from the electrodes, the contour patterns are different for the
different arrays. The difference in the contour pattern of the sensitivity function plot helps to
explain the response of the different arrays to different types of structures.
In the following plots of the sensitivity sections, the distance between the first electrode
and the last electrode (for example the C1 and C2 in the case of the Wenner alpha array in
Figure 23a) is normalized to 1.0 meter. To avoid the singularities at the electrodes, the

31
sensitivity values are shown from a depth of 0.025 meter downwards to 1.0 meter. In all the
sensitivity section diagrams, the location of the plotting point used in the pseudosection is
marked by a small black cross.
2.5.4 Wenner array

This is a robust array that was popularized by the pioneering work carried by The
University of Birmingham research group (Griffiths and Turnbull, 1985; Griffiths, Turnbull
and Olayinka, 1990). Many of the early 2-D surveys were carried out with this array. The
"normal" Wenner array is technically the Wenner Alpha array. For a four-electrode array, there
are three possible permutations of the positions of the electrodes (Carpenter and Habberjam,
1956). In Figure 23a, the sensitivity plot for the Wenner Alpha array has almost horizontal
contours beneath the center of the array. Because of this property, the Wenner array is
relatively sensitive to vertical changes in the subsurface resistivity below the center of the array.
However, it is less sensitive to horizontal changes in the subsurface resistivity. In general, the
Wenner array is good in resolving vertical changes (i.e. horizontal structures), but relatively
poor in detecting horizontal changes (i.e. narrow vertical structures). In Table 2, the median
depth of investigation for the Wenner Alpha array is approximately 0.5 times the “a” spacing
used. Compared to other arrays, the Wenner Alpha array has a moderate depth of investigation.
The signal strength is inversely proportional to the geometric factor used to calculate the
apparent resistivity value for the array (Table 2). The geometric factor for the Wenner array is
2a. This is smaller than the geometric factor for other arrays. Among the common arrays, the
Wenner array has the strongest signal strength. This can be an important factor if the survey is
carried in areas with high background noise. One disadvantage of this array for 2-D surveys is
the relatively poor horizontal coverage as the electrode spacing is increased (Figure 20). This
could be a problem if you use a system with a relatively small number of electrodes.
Note that the sensitivity section shows large negative values near the surface between
the C1 and P1 electrodes, as well as between the C2 and P2 electrodes. This means that if a
small body with a higher resistivity than the background medium is placed in these negative
zones, the measured apparent resistivity value will decrease. This phenomenon is also known
as an "anomaly inversion". In comparison, if the high resistivity body is placed between the P1
and P2 electrodes where there are large positive sensitivity values, the measured apparent
resistivity will increase. This is the basis of the offset Wenner method by Barker (1992) to
reduce the effects of lateral variations in resistivity sounding surveys.
The other two permutations of the Wenner array are the Wenner Beta and the Wenner
Gamma arrays. The Wenner Beta array is in fact a special case of the dipole-dipole array where
the spacings between the electrodes are the same. Thus this array will be discussed in the
following section under the dipole-dipole array. The Wenner Gamma array has a relatively
unusual arrangement where the current and potential electrodes are interleaved. The sensitivity
section shows that the deepest regions mapped by this array are below the two outer electrodes
(C1 and P2 in Figure 23c), and not below the center of the array.
2.5.5 Dipole-dipole array

This array has been, and is still, widely used in I.P. surveys because of the low EM
coupling between the current and potential circuits. The arrangement of the electrodes is shown
in Figure 4. The spacing between the current electrodes pair, C2-C1, is given as “a” which is
the same as the distance between the potential electrodes pair P1-P2. This array has another
factor marked as “n” in Figure 4. This is the ratio of the distance between the C1 and P1
electrodes to the C2-C1 (or P1-P2) dipole length “a”. For surveys with this array, the “a”
spacing is initially kept fixed at the smallest unit electrode spacing and the “n” factor is
increased from 1 to 2 to 3 until up to about 6 in order to increase the depth of investigation.

32
Figure 23. 2-D sensitivity sections for the Wenner array. The sensitivity sections for the (a)
alpha, (b) beta and (c) gamma configurations.
Figure 24 shows the sensitivity sections for this array for "n" values ranging from 1 to
6. The largest sensitivity values are generally located between the C2-C1 dipole pair, as well
as between the P1-P2 pair. This means that this array is most sensitive to resistivity changes
below the electrodes in each dipole pair. As the "n" factor is increased, the high sensitivity
values become increasingly more concentrated beneath the C1-C2 and P1-P2 dipoles, while
the sensitivity values beneath the center of the array between the C1-P1 electrodes decreases.
For "n" values of greater than 2, the sensitivity values at the pseudosection plotting point
become negligible. The sensitivity contour pattern becomes almost vertical for "n" values
greater than 2. Thus the dipole-dipole array is very sensitive to horizontal changes in resistivity,
but relatively insensitive to vertical changes in the resistivity. Thus it is good in mapping
vertical structures, such as dykes and cavities, but relatively poor in mapping horizontal
structures such as sills or sedimentary layers. The median depth of investigation of this array
depends on both the “a” spacing and the “n” factor (Table 2). In general, this array has a

33
shallower depth of investigation compared to the Wenner array, for example at n=1 the depth
of investigation is 0.416a compared to 0.512a for the Wenner Alpha array.
Figure 24. 2-D sensitivity sections for the dipole-dipole array. The sections with (a) n=1, (b)
n=2, (c) n=4 and (d) n=6.

34
Due to the almost vertical pattern of the sensitivity contours, the depth of investigation
(which a 1-D horizontal average of the sensitivity values) is not particularly meaningful for the
dipole-dipole array for "n" values greater than 2. From experience with synthetic modeling and
field data, the median depth of investigation might underestimate the depth of structures sensed
by this array by about 20% to 30% for the large “n” factors. For 2-D surveys, this array has
better horizontal data coverage than the Wenner (Figure 20). This can be an important
advantage when the number of nodes available with the multi-electrode system is small.
One possible disadvantage of this array is the very low signal strength for large values
of the “n” factor. The voltage is inversely proportional to the cube of the “n” factor. For the
same current, the voltage measured by the resistivity meter drops by about 56 times when “n”
is increased from 1 to 6 (Table 2). One method to overcome this problem is to increase the “a”
spacing between the C1-C2 (and P1-P2) dipole pair to reduce the drop in the potential when
the overall length of the array is increased to increase the depth of investigation. Figure 25
shows two different arrangements for the dipole-dipole array with the same array length but
with different “a” and “n” factors. The signal strength of the array with the smaller “n” factor
is about 28 times stronger than the one with the larger “n” factor.
To use this array effectively, the resistivity meter should have comparatively high
sensitivity and very good noise rejection circuitry, and there should be good contact between
the electrodes and the ground. With the proper field equipment and survey techniques, this
array has been successfully used in many areas to detect structures such as cavities where the
good horizontal resolution of this array is a major advantage.
The plotting location of the corresponding data point (based on the median depth of
investigation) used in drawing the apparent resistivity pseudosection is also shown in Figure
24. Note that the pseudosection plotting point falls in an area with very low sensitivity values
for “n” values of 4 and above. For the dipole-dipole array, the regions with the high sensitivity
values are concentrated below the C1-C2 electrodes pair and below the P1-P2 electrodes pair.
In effect, the dipole-dipole array gives minimal information about the resistivity of the region
surrounding the plotting point, and the distribution of the data points in the pseudosection plot
does not reflect the subsurface area mapped by the apparent resistivity measurements. Note
that if the data point is plotted at the point of intersection of the two 45 angle lines drawn from
the center of the two dipoles, it would be located at a depth of 0.7 units in Figure 24d (compared
with 0.19 units given by the median depth of investigation method) where the sensitivity values
are almost zero!
Loke and Barker (1996a) used an inversion model where the arrangement of the model
blocks directly follows the arrangement of the pseudosection plotting points. This approach
gives satisfactory results for the Wenner and Wenner-Schlumberger arrays where the
pseudosection point falls in an area with high sensitivity values (Figure 23a and Figure 26).
However, it is not suitable for arrays such as the dipole-dipole and pole-dipole where the
pseudosection point falls in an area with very low sensitivity values. The RES2DINV program
uses a more sophisticated method to generate the inversion model where the arrangement the
model blocks is not tightly bound to the pseudosection.
A final minor note. In most textbooks, the electrodes for this array are arranged in a C1-
C2-P1-P2 order that will in fact give a negative geometric factor. The arrangement assumed in
these notes is the C2-C1-P1-P2 arrangement.

35
Figure 25. Two possible different arrangements for a dipole-dipole array measurement. The
two arrangements have the same array length but different “a” and “n” factors resulting in
very different signal strengths.
2.5.6 Wenner-Schlumberger array

This is a hybrid between the Wenner and Schlumberger arrays (Pazdirek and Blaha
1996) arising out electrical imaging surveys with multi-electrode systems. The classical
Schlumberger array is one of the most commonly used arrays for resistivity sounding surveys.
A digitized form of this array so that it can be used on a system with the electrodes arranged
with a constant spacing is shown in Figure 27b. The “n” factor for this array is the ratio of the
distance between the C1-P1 (or P2-C2) electrodes to the spacing between the P1-P2 potential
pair. Note that the Wenner array is a special case of this array where the “n” factor is equals to
1.
Figure 26 shows the sensitivity pattern for this array as the "n" factor is increased from
1 (Wenner array) to 6 (the classical Schlumberger array). The area of highest positive
sensitivity below the center of the array becomes more concentrated beneath central P1-P2
electrodes as the "n" factor is increased. Near the location of the plotting point at the median
depth of investigation, the sensitivity contours has a slight vertical curvature below the center
of the array. At n=6, the high positive sensitivity lobe beneath the P1-P2 electrodes becomes
more separated from the high positive sensitivity values near the C1 and C2 electrodes. This
means that this array is moderately sensitive to both horizontal (for low "n" values) and vertical
structures (for high "n" values). In areas where both types of geological structures are expected,
this array might be a good compromise between the Wenner and the dipole-dipole array. The
median depth of investigation for this array is about 10% larger than that for the Wenner array
for the same distance between the outer (C1 and C2) electrodes for "n" values greater than 3.
The signal strength for this array is approximately inversely proportional to the square of the
"n" value. The signal strength is weaker than that for the Wenner array, but it is higher than the
dipole-dipole array and twice that of the pole-dipole.
Figure 27 shows the pattern of the data points in the pseudosections for the Wenner and
Wenner-Schlumberger arrays. The Wenner-Schlumberger array has a slightly better horizontal
coverage compared with the Wenner array. For the Wenner array each deeper data level has 3
data points less than the previous data level, while for the Wenner-Schlumberger array there is
a loss of 2 data points with each deeper data level. The horizontal data coverage is slightly
wider than the Wenner array, but narrower than that obtained with the dipole-dipole array.

36
Figure 26. 2-D sensitivity sections for the Wenner-Schlumberger array. The sensitivity
sections with (a) n=1, (b) n=2, (c) n=4 and (d) n=6.

37
Figure 27. A comparison of the (i) electrode arrangement and (ii) pseudosection data pattern
for the Wenner and Wenner-Schlumberger arrays.
2.5.7 Pole-pole array

This array is not as commonly used as the Wenner, dipole-dipole and Schlumberger
arrays. In practice the ideal pole-pole array, with only one current and one potential electrode
(Figure 4d), does not exist. To approximate the pole-pole array, the second current and potential
electrodes (C2 and P2) must be placed at a distance that is more than 20 times the maximum
separation between C1 and P1 electrodes used in the survey. The effect of the C2 (and similarly
for the P2) electrode is approximately proportional to the ratio of the C1-P1 distance to the C2-
P1 distance. If the effects of the C2 and P2 electrodes are not taken into account, the distance
of these electrodes from the survey line must be at least 20 times the largest C1-P1 spacing
used to ensure that the error is less than 5%. In surveys where the inter-electrode spacing along
the survey line is more than a few meters, there might be practical problems in finding suitable
locations for the C2 and P2 electrodes to satisfy this requirement. Another disadvantage of this
array is that because of the large distance between the P1 and P2 electrodes, it is can pick up a
large amount of telluric noise that can severely degrade the quality of the measurements. Thus
this array is mainly used in surveys where relatively small electrode spacings (less than a few
meters) are used. It is popular in some applications such as archaeological surveys where small
electrode spacings are used. It has also been used for 3-D surveys (Li and Oldenburg, 1992).
This array has the widest horizontal coverage and the deepest depth of investigation.
However, it has the poorest resolution, which is reflected by the comparatively large spacing
between the contours in the sensitivity function plot (Figure 28).

38
Figure 28. The pole-pole array 2-D sensitivity section.
2.5.8 Pole-dipole array

The pole-dipole array also has relatively good horizontal coverage, but it has a
significantly higher signal strength compared with the dipole-dipole array and it is not as
sensitive to telluric noise as the pole-pole array. Unlike the other common arrays, the pole-
dipole array is an asymmetrical array (Figure 4f). Over symmetrical structures the apparent
resistivity anomalies in the pseudosection are asymmetrical (Figure 20d). In some situations,
the asymmetry in the measured apparent resistivity values could influence the model obtained
after inversion. One method to eliminate the effect of this asymmetry is to repeat the
measurements with the electrodes arranged in the reverse manner (Figure 29b). By combining
the measurements with the “forward” and “reverse” pole-dipole arrays, any bias in the model
due to the asymmetrical nature of this array would be removed. This procedure will double the
number of data points and consequently the survey time, but it is not a severe problem with
multi-channel systems.
The sensitivity section shows that area with the greatest sensitivity lies beneath P1-P2
dipole pair, particularly for large “n” factors. For “n” values of 4 and higher, the high positive
sensitive lobe beneath the P1-P2 dipole becomes increasingly vertical. Thus, similar to the
dipole-dipole array, this array is probably more sensitive to vertical structures. Note also the
zone with negative sensitivity values between the C1 and P1 electrodes, as well as the smaller
zone of high positive values to the left of the C1 electrode.
The pole-dipole array requires a remote electrode, the C2 electrode, which must be
placed sufficiently far from the survey line. We can estimate the effect of the remote electrode
by calculating the potential for a simple homogeneous half-space. For the pole-dipole array,
the potential due to the C1 electrode is as follows.
𝐼𝜌
𝑉𝐶1 = 2𝜋𝑛(𝑛+1)𝑎
Similarly if the C2 electrode is at ka distance from the P1-P2 dipole, and assuming it is on the
same side as the C1 electrode, the potential due to this remote electrode is

39
𝐼𝜌
𝑉𝐶2 = 2𝜋𝑘(𝑘+1)𝑎.
The ratio of the potentials due to the C2 and the C1 electrodes is then given by
𝑉𝐶2 𝑛(𝑛+1) 𝑛 2
= ≈ (𝑘 )
𝑉𝐶1 𝑘(𝑘+1)
if n is reasonably large (example 10). Thus the effect of the C2 electrodes decreases with square
of its distance from the survey line. The pole-dipole array measurements are less affected by
the C2 remote electrode compared to the pole-pole array. If the distance of the C2 electrode is
more than 5 times the largest C1-P1 distance used, the error caused by neglecting the effect of
the C2 electrode is less than 5% (the exact error also depends on the location of the P2 electrode
for the particular measurement and the subsurface resistivity distribution).
Due to its good horizontal coverage, this is an attractive array for multi-electrode
resistivity meter systems with a relatively small number of nodes. The signal strength is lower
compared with the Wenner and Wenner-Schlumberger arrays but higher than the dipole-dipole
array. For I.P. surveys, the higher signal strength (compared with the dipole-dipole array)
combined with the lower EM coupling (compared with the Wenner and Wenner-Schlumberger
arrays) due to the separation of the circuitry of the current and potential electrodes makes this
array an attractive alternative.
The signal strength for the pole-dipole array decreases with the square of the “n” factor.
While this effect is not as severe as the dipole-dipole array, it is usually not advisable to use
“n” values of greater than 8 to 10. Beyond this, the “a” spacing between the P1-P2 dipole pair
should be increased to obtain a stronger signal strength. The ratio of the depth of investigation
to the array length does not increase significantly for ‘n’ values beyond 4 or 5 (Table 2). Thus
to increase the depth of investigation, it is better to increase the ‘a’ spacing rather than to keep
increasing the ‘n’ value when it reaches 5. There is another interesting effect when the ‘n’ factor
is increased that is frequently not appreciated, and leads to an interesting pitfall in field surveys.
This is discussed in section 4.10.
Figure 29. The forward and reverse pole-dipole arrays.

40
Figure 30. The pole-dipole array 2-D sensitivity sections. The sensitivity sections with (a)
n=1, (b) n=2, (c) n=4 and (d) n=6.

41
2.5.9 Multiple gradient array

This is a relatively new array developed primary for multi-channel resistivity meter
systems (Dahlin and Zhou, 2006). A multi-channel system can make several measurements
simultaneously with the potential electrode pairs at different location but with the same
positions of the current electrodes. The dipole-dipole and pole-dipole arrays are the obvious
arrays that are suitable for such measurements. However, the dipole-dipole array is relatively
sensitive to noise while the pole-dipole array requires a remote electrode at a sufficiently large
distance from the survey line. For some surveys, an ‘alpha’ type of electrode arrangement with
the potential electrodes located between the current electrodes is desirable. One possibility is
the ‘reverse’ Schlumberger array with the current electrodes at the centre and the potential
electrodes as the outer electrodes. However, this arrangement will be more sensitive to telluric
noise due to the large distance between the potential electrodes. The current electrodes in the
traditional gradient array are fixed at the end of the line, while measurements are made with
the potential electrodes at different positions along the line.
In the multiple gradient array, different sets of measurements are made with the current
electrodes at different locations. As an example, for a system with 32 electrodes, one set of
measurements can be made with the current electrodes at nodes 1 and 32. Next, another series
of measurements are made with the current electrodes at nodes 1 and 16, as well as another
with the current electrodes at 16 and 32. A similar set of measurements can be made with the
C1-C2 electrodes at 1-8, 8-16, 16-24 and 24-32. This can be repeated using smaller distances
between the current electrodes. The sensitivity sections with the same positions of the C1-C2
current electrodes, but with the potential dipole P1-P2 being moved from the center to one end
of the array is shown in Figure 31. Not surprisingly, the sensitivity contour pattern slowly
changes from a Wenner-Schlumberger pattern (Figure 26) towards the pole-dipole pattern
(Figure 30) as the potential dipole moves closer to the current electrode at one end of the array.
The results obtained by this array is comparable to those obtained by the Wenner-Schlumberger
and pole-dipole arrays (Dahlin and Zhou, 2004) while it has more favourable noise
characteristics (Zhou and Dahlin, 2003).
One interesting minor problem associated with this array is the plotting method for the
pseudosection and profiles. The vertical position can be easily determined by using the median
depth of investigation which can be numerically determined. The main issue is the horizontal
position of the plotting point.
Two possibilities are using the average of the locations of all the four electrodes (C1-
P1-P2-C2), or just the midpoint of the P1-P2 electrodes (Dahlin and Zhou, 2006). However,
for some data sets, both methods might result in a situation where two data points measured
with different array configurations end up at the same plotting location. There might be no
single method that works satisfactory in all cases. The RES2DINV program has two options
for plotting the data as profiles are shown in Figure 32. The profile plot is commonly used to
pick out outliers in the data.
One possible method for selecting the x-location for the pseudosection plotting point
for non-sysmmetrical arrays such as the multiple gradient and pole-dipole is to carry out an
integration of the Frechet derivative from -∞ to +∞ in the y and z directions, and from -∞ to
xp in the x direction, such as
x x  a   y 2  z 2
 x p
F1D x p  
1
4 2
0   
   x
 y 2
 z 2 1.5
2
 
x  a 2
 y 2
 z 
2 1.5
dxdydz .
The value of xp where this integral reaches half the total value can then be used.

42
Figure 31. Sensitivity sections for the gradient array. The current electrodes are fixed at x=0
and 1.0 meters, and the distance between the potential electrodes is 0.1 m. The position of the
P1-P2 potential electrodes at (a) 0.45 and 0.55 m., (b) 0.55 and 0.65 m., (c) 0.65 and 0.75 m,
(d) 0.75 and 0.85 m. and (e) 0.80 and 0.90 m.

43
Figure 32. (a) Example of multiple gradient array data set and inversion model. (b) Profile
plot using exact pseudodepths. (c) Profile plot using approximate pseudodepths.

44
2.5.10 High-resolution electrical surveys with overlapping data levels

In seismic reflection surveys, the common depth point method is frequently used to
improve the quality of the signals from subsurface reflectors. A similar technique can be used
to improve the data quality for resistivity/I.P. surveys, particularly in noisy areas. This is by
using overlapping data levels with different combinations of “a” and “n” values for the
Wenner-Schlumberger, dipole-dipole and pole-dipole arrays.
To simplify matters, let us consider the case for the Wenner-Schlumberger array with
an inter-electrode spacing of 1 meter along the survey line. A high-resolution Wenner-
Schlumberger survey will start with the “a” spacing (which is the distance between the P1-P2
potential dipole) equals to 1 meter and repeat the measurements with “n” values of 1, 2, 3 and
4. Next the “a” spacing is increased to 2 meter, and measurements with “n” equals to 1, 2, 3
and 4 are made. This process is repeated for all possible values of the “a” spacing. To be on
the safe side, the data set should contain all the possible data points for the Wenner array. The
number of data points produced by such a survey is more than twice that obtained with a normal
Wenner array survey. Thus the price of better horizontal data coverage and resolution is an
increase in the field survey time.
A Wenner array with “a” equals to 2 meters (Figure 27) will have a total array length
of 6 meters and a median depth of investigation of about 1.04 meters. In comparison, a
measurement made with “a” equals to 1 meter and “n” equals to 2 using the Wenner-
Schlumberger array will have a total array length of 5 meters and a slightly smaller depth of
investigation of 0.93 meter (Figure 27). While the depth of investigation of the two
arrangements are similar, the section of the subsurface mapped by the two arrays will be
slightly different due to the different sensitivity patterns (Figure 26a,b). So the two
measurements will give slightly different information about the subsurface. A measurement
with “a” equals to 1 meter and “n” equals to 3 (Figure 27) will have a depth of investigation of
1.32 meters. If all the 3 combinations are used, the data set will have measurements with
pseudodepths of 0.93, 1.02 and 1.32 metres. This results in a pseudosection with overlapping
data levels.
A similar “high-resolution” survey technique can also be used with the dipole-dipole
and pole-dipole arrays by combining measurements with different “a” and “n” values to give
overlapping data levels. In particular, this technique might be useful for the dipole-dipole array
since the signal strength decreases rapidly with increasing "n" values (section 2.5.5). A typical
high-resolution dipole-dipole survey might use the following arrangement; start with a dipole
of "1a" and "n" values of 1, 2, 3, 4, 5; followed by a dipole of "2a" and "n" values of 1, 2, 3,
4, 5; and if necessary another series of measurements with a dipole of "3a" and "n" values of
1, 2, 3, 4, 5. Measurements with the higher "n" values of over 4 would have higher noise levels.
However, by having such redundant measurements using the overlapping data levels, the effect
of the more noisy data points will be reduced. Figure 33 shows the apparent resistivity
pseudosection for single prism model using this arrangement. This arrangement has been
widely used for I.P. surveys with the dipole-dipole array (Edwards, 1977).
In theory, it should be possible to combine measurements made with different arrays to
take advantage of the different properties of the various arrays. Although this is not a common
practice, it could conceivably give useful results in some situations. The RES2DINV program
supports the use of such mixed data sets.
Besides the standard arrays discussed, there is now interest in automatically computer
generated optimized non-standard arrays. This is discussed later in section 4.9.4.

45
2.5.11 Summary of array types

The Wenner array is an attractive choice for a survey carried out in a noisy area (due to
its high signal strength) and also if good vertical resolution is required. The dipole-dipole array
might be a more suitable choice if good horizontal resolution and data coverage is important
(assuming your resistivity meter is sufficiently sensitive and there is good ground contact). The
Wenner-Schlumberger array (with overlapping data levels) is a reasonable all-round alternative
if both good and vertical resolutions are needed, particularly if good signal strength is also
required. If you have a system with a limited number of electrodes, the pole-dipole array with
measurements in both the forward and reverse directions might be a viable choice. For surveys
with small electrode spacings and require a good horizontal coverage, the pole-pole array might
be a suitable choice. For multi-channel arrays, the possible arrays are the multiple gradient, the
dipole-dipole and pole-dipole arrays.
The papers by Dahlin and Zhou (2004) and Zhou and Dahlin (2003) have interesting
discussions on the properties of different arrays.
Figure 33. The apparent resistivity pseudosection for the dipole–dipole array using
overlapping data levels over a rectangular prism. Values of 1 to 3 meters are used for the dipole
length ‘a’, and the dipole separation factor ‘n’ varies from 1 to 5. Compare this with Figure 20c
for the same model but with ‘a’ fixed at 1 meter, and “n” varying from 1 to 10. In practice, a
‘n’ value greater than 8 would result in very noisy apparent resistivity values.

46
2.5.12 Use of the sensitivity values for multi-channel measurements or streamers

Here we will look at one example of the use of the sensitivity values. The problem faced
is at follows.
a). We have a multi-channel resistivity meter that can make 8 measurements at a single time.
b). The cable used (possibly in the form of a streamer for mobile surveys on land or in water
covered areas) has 10 nodes. Two of the nodes are fixed as the current electrodes, while the
potential measurements are made using any 2 of the remaining 8 nodes.
c). We want to get the most information possible in terms of depth and a uniform horizontal
coverage below the streamer for a single set of 8 measurements.
To study the characteristics of the array configurations, we add up the sensitivity values
for all the 8 measurements. The formula used for the cumulative absolute sensitivity, FI, is
1 i 8
FI   F2 Di x, z 
8 i 1
(2.7)
Note that the absolute value of the sensitivity value is used, since negative sensitivity values
also give information about the subsurface. This is also to avoid a situation where the negative
sensitivity value for one measurement cancels out the positive value for another measurement.
Figure 34a shows a dipole-dipole array based measurement sequence. The two current
electrodes are set at one end of the line and the potential measurements are made using
successive pairs of the potential nodes. Note the area with the highest cumulative sensitivity is
around the first 4 electrodes. The area of deepest penetration is between the second and third
electrodes, i.e. between the current dipole and the first potential electrode.
The second configuration places the current electrodes at the ends of the line (Figure
34b). Almost all the measurements are made using a fixed separation of one unit electrode
spacing between the potential electrodes. The potential dipole is moved from one end of the
survey line to the other end, i.e. a gradient array type of arrangement. One advantage of this
configuration over the dipole-dipole arrangement is the larger voltage (i.e. lower noise level)
measured by the potential electrodes. The cumulative sensitivity section shows a more uniform
pattern compared to that produced by the dipole-dipole configuration. The depth of
investigation is slightly deeper below current electrodes, and slightly less below the center of
the line.
The third configuration also places the current electrodes at the ends of the lines, but
keeps the center of the potential dipoles at or near the center of the line. The first 4
measurements use a symmetrical Wenner-Schlumberger arrangement and increase the
separation between the potential electrodes until they reach next to the current electrodes. The
second set of 4 measurements uses separations of 2 and 3 times the unit spacing for the potential
dipole pair but with a mid-point slightly to one side of the center of the line. The cumulative
sensitivity pattern has an even more uniform pattern below the line compared with the gradient
array configuration.
To get an idea of the relative depths of investigation for the 3 configurations, we use
the light blue contour (with a value of 8 units) as a guide. The expanding Wenner-Schlumberger
configuration has the deepest depth with high sensitivity values while the dipole-dipole
configuration appears to have the shallowest. Note the current electrodes for dipole-dipole
array is limited to the left end of the line, and for this reason the region with the deepest depth
of investigation is skewed towards the left.
One problem with using the cumulative sensitivity sections is that it is a rather crude
measure of the resolution that the data set has. It probably overestimates the depth of
investigation for the gradient and Wenner-Schlumberger configurations (Figure 34b, c). This
is because it does not take into account the orthogonality of data, i.e. the information from two

47
measurements for the same region might not be independent. A better method to evaluate the
information content from a group of array configurations is the model resolution (section
4.9.6).
Figure 34. Cumulative sensitivity sections for different measurement configurations using (a)
a dipole-dipole sequence, (b) a moving gradient array and (c) an expanding Wenner-
Schlumberger array.

48
3 A 2-D forward modeling program
3.1 Finite-difference and finite-element methods

In the forward modeling problem, the subsurface resistivity distribution is specified and
the purpose is to calculate the apparent resistivity that would be measured by a survey over
such a structure. A forward modeling subroutine is in fact also an integral part of any inversion
program since it is necessary to calculate the theoretical apparent resistivity values for the
model produced by the inversion routine to see whether is agrees with the measured values.
There are three main methods to calculate the apparent resistivity values for a specified model.
They are (i) analytical methods, (ii) boundary element methods and (iii) the finite-difference
and finite-element methods. Analytical methods are probably the most accurate methods, but
they are restricted to relative simple geometries (such as a sphere or cylinder). Boundary
element methods are more flexible, but the number of regions with different resistivity values
that is allowed is somewhat limited (usually less than 10). In engineering and environmental
surveys the subsurface can have an arbitrary resistivity distribution, so the finite-difference and
finite-element methods are usually the only viable choice. These methods can subdivide the
subsurface into thousands of cells with different resistivity values. However, the analytical and
boundary element methods are useful independent methods that can be used to check the
accuracy of the finite-difference and finite-element methods.
In the RES2DMOD software, the user can choose the finite-difference or the finite-
element method. The subsurface is subdivided into a large number of rectangular cells (Figure
35) and the user can specify the resistivity value of each cell. The finite-difference method is
based on a method described by Dey and Morrison (1979a) but with a modification by Loke
(1994) to correct for a minor inconsistency in the Dey and Morrison discretization by area
method. The finite-element method uses the standard first-order triangular elements (Silvester
and Ferrari, 1990).
While our main interest is in the inversion of field data, the forward modeling program
is also useful, particularly in the planning stage of the survey. In the previous chapter we have
seen that different arrays can have sensitivity sections that are radically different. In theory,
from the sensitivity sections we can get an idea of the type of array that will give a reasonably
good response over a particular class of structures (for example a vertical fracture zone).
However, there is no substitute for a hands-on direct calculation of the expected apparent
resistivity pseudosection. Before carrying out a field survey, some information about the shape
and size of expected targets is frequently known. By trying different arrays digitally on the
computer screen, we can avoid using an array that is unsuitable for the detection of the
structures of interest. We can also have an idea of a suitable spacing between adjacent
electrodes to use, and the maximum electrode separation needed.
In the program, the subsurface is divided into a large number of small rectangular cells.
This program is largely intended for teaching about the use of the 2-D electrical imaging
method. Hopefully, it will assist the user in choosing the appropriate array for different
geological situations or surveys. The arrays supported by this program are the Wenner (Alpha,
Beta and Gamma configurations), Wenner-Schlumberger, pole-pole, inline dipole-dipole, pole-
dipole and equatorial dipole-dipole (Edwards, 1977). Each type of array has its advantages and
disadvantages. This program will hopefully help you in choosing the "best" array for a
particular survey area after carefully balancing factors such as the cost, depth of investigation,
resolution and practicality.

49
Figure 35. The output from the RES2DMOD software for the SINGLE_BLOCK.MOD 2-D
model file. The individual cells in the model are shown in the lower figure, while the upper
figure shows the pseudosection for the Wenner Beta (dipole-dipole with n=1) array.
3.2 Using the forward modeling program RES2DMOD

The program requires the resistivity model values to be typed in separately in a text file.
The model data format, and other details about the use of this program, can be found in the
RES2DMOD.PDF manual file. In this course, we will use several model files that are already
present in the program package to take a look at the shapes of the contours in the pseudosections
for different geological structures. By playing around with this program, you can get a feel of
the effects of array type over the amplitude, size and shape of the contours in the pseudosection.
After starting up the RES2DMOD program, click the ‘File’ menu option on the main
menu bar. Next select the “Read data file with forward model” suboption to read in one of the
example input model files provided. The files with the forward model have a MOD extension.
As an example, read in the file SINGLE_BLOCK.MOD that has a simple model with a
rectangular prism. After reading in this file, select the “Edit/Display Model” option followed
by the “Edit model” suboption to take a look at the model. This will show you the actual cells
that make up the model.
The program also allows you to change the model interactively using the left and right
mouse buttons. To change a single cell, click it with the left mouse button. Then move the

50
cursor to one of the color boxes in the legend above the model, and click the resistivity value
you want. Press the F1 key to get information about the keys used by the program to edit the
model. Note that clicking the cells with the mouse buttons will only change the resistivity of
the cells displayed on the screen, but will not change the resistivity of the buffer cells towards
the left, right and bottom edges of the mesh. To change the resistivity of the buffer cells, you
need to use the “[“, “]” and “D” keys.
Next select the “Model Computation” option to calculate the potential values for this
model. The calculations will probably only take a few seconds, after which you should go back
to “Edit/Display Model” option. In this option, select the “Edit model” sub-option again to see
the apparent resistivity pseudosection for this model. The program will ask you to select the
type of contour intervals you wish to use. For most resistivity pseudosections choose the
‘Logarithmic contour intervals’, while for I.P. pseudosections choose the linear contour
intervals.
To change the type of array, use the “Change Settings” sub-option. Select another array,
such as the pole-pole or dipole-dipole, and see what happens to the shape of the contours in the
pseudosection.
3.3 Forward modeling exercises

Here we will try out a few model files that are provided to get a feel of what
pseudosections look like, and the effect of different choices on the results. The RES2DMOD
program also has an option to save the calculated apparent resistivity values into the format
used by the RES2DINV inversion program. You might like to save some of the results from
this exercise into this format to use with the RES2DINV program later on. This is useful in
studying the model resolution that can be obtained over different structures using various
arrays.
Now that we had some experience in creating pseudosections, it is time to have a look
the inversion program RES2DINV that will do the reverse, i.e. creating a model from the
pseudosection.
Table 3. Forward modeling examples.

Model file and purpose Things to try
BLOCK_ONE.MOD (1). The default array type is the Wenner Alpha. Run the ‘Model
To see the effect of selecting computation’ option to get things started. Next use the ‘Edit model’
the array type on the shape of option to take a look at the pseudosection. Compare the shape of
the contours in the the anomaly in the pseudosection with the actual shape of the
pseudosection. The model is a prism. What is the maximum apparent resistivity value (compare it
simple rectangular prism. with the prism resistivity of 100 m)?
(2). Next change the array type to Wenner Beta, and take a look at
the shape of the contours. Also, what is the maximum apparent
resistivity value? Repeat with the Wenner Gamma array. Among
the 3 versions of the Wenner array, which do you think is the
‘best’?
(4). Next try the inline dipole-dipole. What is the maximum ‘n’
value you need to use in order to fully map this prism? For a current
of 10mA, what would the voltage be for this ‘n’ value when the
dipole length is 1 meter? The program also allows you to change
the ‘a’ dipole length. Try changing it to 2 and see what happens.
(5). To complete things, try out the other arrays, such as the pole-
pole, pole-dipole, equatorial dipole-dipole and Wenner-
Schlumberger.
(6). Edit the model using the ‘Edit Model’ option. Try increasing

51
the thickness and/or width of the prism and see what happens to the
anomaly in the pseudosection. Also try changing the resistivity of
the prism to 500 m.
BLOCK_TWO2.MOD (1). The default array type is the Wenner Alpha. Run the ‘Model
In this case, we have two computation’ option to get things started.
prisms at a mean depth of 1 (2). Next, take a look at the pseudosection. Is it possible to separate
meter, and 2 meters apart. The the highs due to each prism?
two prisms are identical and at (3). Try the whole range of arrays, i.e. the 3 different versions of
the same depth. We will try the Wenner, pole-dipole, dipole-dipole, Schlumberger etc. Which
different arrays to which ones arrays are more likely to be able to resolve the two prisms?
are more likely to be able to
resolve the prisms in the
pseudosection.
BLOCKS_UP.MOD (1). The default array type is the Wenner Alpha. Run the ‘Model
Now we have two prisms, one computation’ option to get things started.
on top of another. Is it possible (2). Next, take a look at the pseudosection. Is it possible to separate
to tell them apart? the highs due to each prism?
(3). Try the whole range of arrays, i.e. the 3 different versions of
the Wenner, pole-dipole, dipole-dipole, Schlumberger etc. Which
arrays are more likely to be able to resolve the two prisms?
THICK_DIKE.MOD (1). The default array type is the Wenner Beta. Calculate the
This model has a wide low potentials and take a look at the pseudosection. Is it possible to infer
resistivity dike in high the existence of the dike from the pseudosection?
resistivity bedrock. (2). Change to the Wenner Alpha, and see what happens.
(3). Try other arrays such as the Wenner Gamma, Wenner-
Schlumberger and pole-dipole. Which arrays show a significant
low resistivity anomaly?
THIN_DIKE.MOD (1). The default array type is the Wenner Beta. Calculate the
This model has a thin dike in potentials and take a look at the pseudosection. Is it possible to infer
high resistivity bedrock. This is the existence of the dike for the pseudosection?
a more difficult target than the (2). Change to the Wenner Alpha, and see what happens.
thick dike. (3). Try other arrays such as the Wenner Gamma, Wenner-
Schlumberger and pole-dipole. Which arrays show a significant
low resistivity anomaly?
MODEILIP.MOD (1). The default array type is the dipole-dipole, which is probably
An I.P. model just to round the most commonly used array for I.P. surveys. Calculate the
things up. potential values, and take a look at the pseudosections.
(2). Another array that is sometimes used for I.P. surveys is the
pole-dipole that has the advantage of a stronger signal strength.
Check the pseudosections obtained with this array.

52
4 A 2-D inversion program
4.1 Introduction
After the field survey, the resistance measurements are usually reduced to apparent
resistivity values. Practically all commercial multi-electrode systems come with the computer
software to carry out this conversion. In this section, we will look at the steps involved in
converting the apparent resistivity values into a resistivity model section that can be used for
geological interpretation. I will assume that the data is already in the RES2DINV format. The
conversion program is provided together with many commercial systems. So far, the ones that
have the conversion program include Abem, AGI, Campus, Geofysika, Geometrics, Iris, OYO,
Pasi and Scintrex. If your equipment manufacturer is not on the list, please contact them about
the conversion software. The data format used by the RES2DINV program is described in detail
in the RES2DINV.PDF manual provided with the program. Please refer to the manual for the
details.
In the next section, we will look at two methods to handle bad data points. Such bad
data points should be removed before a final interpretation is made.
Due to the large variety of data sets collected over various geological environments, no
single inversion method will give the optimum results in all cases. Thus the RES2DINV
program has a number of settings that can be changed by the user to obtain results that are
closer to the known geology. The various options are also discussed in the following sections.
4.2 Pre-inversion and post-inversion methods to remove bad data points

To get a good model, the data must be of equally good quality. Bad data points fall into
two broad categories, i.e. “systematic” and “random” noise. Systematic noise is usually caused
by some sort of failure during the survey such that the reading does not represent a true
resistivity measurement. Examples include breaks in the cable, very poor ground contact at an
electrode such that sufficient current cannot be injected into the ground, forgetting to attach the
clip to the electrode, connecting the cables in the wrong direction etc. Systematic noise is fairly
easy to detect in a data set as it is usually present in limited number of readings, and the bad
values usually stick out like sore thumbs. Random noise include effects such telluric currents
that affects all the readings, and the noise can cause the readings to be lower or higher than the
equivalent noise-free readings. This noise is usually more common with arrays such as the
dipole-dipole and pole-dipole that have very large geometric factors, and thus very small
potentials for the same current compared to other arrays such as the Wenner. It is also common
with the pole-pole array due to the large distance between the P1 electrode and the remote (and
fixed) P2 electrode. This array tends tend to pick up a large amount of telluric noise due to the
large distance between the two potential electrodes.
As a general rule, before carrying out the inversion of a data set, you should first take a
look at the data as a pseudosection plot (Figure 36a) as well as a profile plot (Figure 36b). The
bad data points with “systematic” noise show up as spots with unusually low or high values
(Figure 36a). In profile form, they stand out from the rest and can be easily removed manually
for the data set. In the RES2DINV program, choose the ‘Edit data’ on the top menu bar
followed by the ‘Exterminate bad data points’ option (Figure 37). The bad data points can be
removed by clicking them with the mouse.
When the noise is of a more random nature, the noisy data points are not as obvious, so
it might not be practical to remove them manually. Also, manually picking out the bad data
points becomes impractical if there are a large number of bad data points, particularly if the
data set contains more than a thousand data points. In some cases, it is not possible to display
the data as pseudosections or profiles, such as in 3D data sets. RES2DINV (and RES3DINV)
has a general technique to remove the bad data points with minimal input from the user, and

53
can be used for practically any array and any distribution of the data points. The main
disadvantage of the method is the larger amount of computer time needed. In this method, a
preliminary inversion of the data set is first carried with all the data points. After carrying out
the trial inversion, switch to the 'Display' window in RES2DINV, and read in the INV file
containing the inversion results. After that, select the 'RMS error statistics' option that displays
the distribution of the percentage difference between the logarithms of the measured and
calculated apparent resistivity values. The error distribution is shown in the form of a bar chart,
such as in Figure 38. Normally, the highest bar is the one with the smallest errors, and the
heights of the bars should decrease gradually with increasing error values. The bad data points,
caused by problems such as poor ground contact at a small number of electrodes, should have
significantly higher errors than the "good" data points.
Figure 36. An example of a field data set with a few bad data points. The most obvious bad
datum points are located below the 300 meters and 470 meters marks. The apparent resistivity
data in (a) pseudosection form and in (b) profile form.

54
Figure 37. Selecting the menu option to remove bad data points manually.
Figure 38 shows the error distribution bar chart for the data set shown in Figure 36 that
has a few bad data points. In the bar chart, almost all the data points have errors of 20 percent
or less. The bad data points show up data points with errors of 60 percent and above, which
can be easily removed from the data set by moving the green cursor line to the left of the 60%
error bar. In this way the 5 bad data points are removed from this data set. For some data sets,
the error distribution might show a more complicated pattern. As a general rule, data points
with misfits of 100 percent and above can usually be removed.
Table 4. Methods to remove bad data points

GRUNDF1.DAT – An (1) Use the ‘File’ and then the ‘Read data file’ options to
example of a field data set with read in this data file. Go to the ‘Display’ windows, and then
bad data points. the ‘Display data and model sections’ option. The bad data
points should be quite obvious.
(2) Next, leave the ‘Display’ window, and then choose
‘Edit data’ on the top menu bar followed by the
‘Exterminate bad data points’ option. Pick out the bad data
points. After that save the edited data in a file. Read in this
edited data file, and then go back to the ‘Display’ window
and check the pseudosection again.
(3) After that, leave the ‘Display’ window, and then run an
inversion of the data set using the ‘Inversion’ and then the
‘Least-squares inversion’ menu options.
(4) After the inversion has finished, go the ‘Display’
window to take a look at the model. After that choose the
‘Edit data’ and then the ‘RMS error statistics’ options.
Take a look at the bar chart. Is it possible to remove more
bad data points?
(5). Try running an inversion of the data set without first
manually removing the bad data points. Then use the ‘RMS
error statistics’ option to remove them. Does this get rid of
the bad data points also?

55
Figure 38. Error distribution bar chart from a trial inversion of the Grundfor Line 1 data set
with five bad data points.
4.3 Selecting the proper inversion settings

Many professionals carrying out resistivity imaging surveys will likely to be field
engineers, geologists or geophysicists who might not familiar with geophysical inversion
theory. The RES2DINV program is designed to operate, as far as possible, in an automatic and
robust manner with minimal input from the user. It has a set of default parameters that guides
the inversion process. In most cases the default parameters give reasonable results. This section
describes some of the parameters the user can modify to fine-tune the inversion process. In the
program, the different options are divided into six major groups. The groups are placed under
the ‘Change Settings’ or ‘Inversion’ choices on the main menu bar. Here, we will look at a few
of the more important settings that can be changed.
a). Inversion methods
The problem of non-uniqueness is well known in the inversion of resistivity sounding
and other geophysical data. For the same measured data set, there is wide range of models that
can give rise to the same calculated apparent resistivity values. To narrow down the range of
possible models, normally some assumptions are made concerning the nature of the subsurface
that can be incorporated into the inversion subroutine. In almost all surveys, something is
known about the geology of the subsurface, for example whether the subsurface bodies are
expected to have gradational or sharp boundaries.
The default smoothness-constrained inversion formulation used by the RES2DINV
program is given by (please refer to section 1.4 for the details)
 
J T J F Δq J T g . (4.1)
This formulation constrains the change in the model resistivity values, Δq , to be smooth but
does not guarantee that the resistivity values change in a smooth manner. However, this
formulation has been quite popular and used by a number of researchers (deGroot-Hedlin and
Constable, 1990; Sasaki, 1992).
The first option ‘Include smoothing of model resistivity’ uses a formulation that will
apply the smoothness constrain directly on the model resistivity values. This formulation is

56
given by
 
J T J W T W ΔqJ T g W T Wq , (4.2)
so that the model resistivity values, q, changes in a smooth manner. The next option ‘Use
combined inversion method’ attempts to combined the smoothness-constrained method as
given in (4.1) with the Marquardt-Levemberg as given in (1.20). However, the result obtained
by this combination has not been very impressive and will not be examined.
The ‘Select robust inversion’ option has proved to be much more useful. It modifies the
formulation in (4.2) so that different elements of the model parameter change and data misfit
vectors have the same magnitudes. It is given by
 
J T R d J W T R m W Δqk J T R d g  W T R m Wq k . (4.3)
The details are described in section 1.4. This method is also known as an l 1-norm or
robust or blocky inversion method (Loke et al. 2003), whereas the conventional smoothness-
constrained least-squares method as given in equation (4.2) is an l 2-norm inversion method.
The l2-norm inversion method gives optimal results where the subsurface geology exhibits a
smooth variation, such as the diffusion boundary of a chemical plume. However, in cases where
the subsurface consists of bodies that internally homogeneous with sharp boundaries (such as
an igneous dyke), this method tends to smear out the boundaries. The l 1-norm or blocky
optimization method tends to produce models that are piecewise constant (Ellis and Oldenburg,
1994a). This might be more consistent with the known geology in some situations.
Figure 39. Different options to modify the inversion process.
Figure 40 shows the inversion results for data from a synthetic model with sharp
boundaries. In this case, the robust inversion method gives significantly better results since the
true model consists of three homogenous regions with sharp boundaries. Many synthetic test
models are of a similar nature with sharp boundaries, so not surprisingly, results obtained with
the l2-norm smooth inversion method are not optimal for such data sets (Olayinka and
Yaramanci, 2000).
Resistivity values have a logarithmic range, possibly ranging from less than 1 to over
1000 m in a single data set. By using the logarithm of the resistivity values as the parameters
in the inversion process, the numerical range of the parameters can be reduced to a linear range.
However, in some situations, it is not possible to make use of the logarithm if there are negative
or zero values. This does not usually occur for normal surface surveys with the standard arrays,
but could occur in borehole surveys or if non-standard arrays are used. The program also allows
the use to use the apparent resistivity value directly as the inversion parameter in the “Choose

57
logarithm of apparent resistivity” option.
Figure 40. Example of inversion results using the l2-norm smooth inversion and l1-norm blocky
inversion model constrains. (a) Apparent resistivity pseudosection (Wenner array) for a
synthetic test model with a faulted block (100 m) in the bottom-left side and a small
rectangular block (2 m) on the right side with a surrounding medium of 10 m. The
inversion models produced by (b) the conventional least-squares smoothness-constrained or l2-
norm inversion method and (c) the robust or l1-norm inversion method.
The setting ‘Type of method to solve least-squares equation’ provides two numerical methods
to solve the least-squares equation (as in equations 4.1 to 4.3). The default method, particularly
when the number of model parameters n is small, is the standard (or complete) Gauss-Newton
method where a direct method (Golub and van Loan, 1989) is used to solve the equation. This
method produces an exact solution but as the time taken is proportional to n3, it could take a
very long time for large data set with over 5000 model cells. An alternative method, the
incomplete Gauss-Newton method, can be used for such cases. In the incomplete Gauss-
Newton method, an approximate solution of the least method is determined by using an
iterative method (Golub and van Loan, 1989). The final solution obtained with this method has
a difference of about 1 to 2 percent compared to the complete Gauss-Newton method solution.
By sacrificing a small amount of accuracy in the solution to the least-squares equation, the
computer time required could be reduced by a factor of 5 to 10 times for very large models.
b). Model discretization

This set of options control the way the program subdivides the subsurface into
rectangular cells. By default, the program uses a heuristic algorithm partly based on the position
of the data points to generate the size and position of the model blocks. In the default algorithm
used, the depth to the deepest layer in the model is set to be about the same as the largest depth
of investigation of the data points, and the number of model cells does not exceed the number

58
of data points. The thickness of each deeper layer is increased to reflect the decreasing
resolution of the resistivity method with increasing depth. In general, this produces a model
where the thickness of the layers increases with depth, and with thicker cells at the sides and
in the deeper layers (Figure 41a). For most cases, this gives an acceptable compromise. The
distribution of the data points in the pseudosection is used as a rough guide in allocating the
model cells, but the model section does not rigidly follow the pseudosection. However, the user
can change the width and thickness of the cells using a variety of options.
Figure 41. Different methods to subdivide the subsurface into rectangular prisms in a 2-D
model. Models obtained with (a) the default algorithm, (b) by allowing the number of model
cells to exceed the number of data points, (c) a model which extends to the edges of the
survey line and (d) using the sensitivity values for a homogeneous earth model.

59
After reading a data file, clicking the ‘Display model blocks’ option will show the
distribution of the model cells currently used by the program, such as in Figure 41. Clicking
the ‘Change thickness of layers’ option will bring up the following dialog box.
Figure 42. The options to change the thickness of the model layers.
When the program reads in a data file, it will normally set the first layer thickness using
the minimum pseudodepth of the data points. For surface surveys, since the resolution
decreases with depth, the thickness of the layers is normally increased by between 5 to 15 %
with each deeper layer. The program normally uses a model where the depth to the deepest
layer does not exceed the maximum pseudodepth in the data set. To use a model that spans a
deeper depth range, you can change the factor to increase model depth range, for example from
1.0 to 1.30 to increase the model depth range by 30%. In this dialog box, you can also allow
the program to use a model where the number of model cells exceeds the number of data points.
This is useful to avoid having a model with very wide cells near the bottom for data sets with
very sparse data sets. Figure 41b shows such an example with this option enabled.
The ‘Modify depth to layers’ option allows you to set the depth to each layer
individually. This is useful if you want a layer boundary to coincide exactly with a known
geological boundary.
In ‘Use extended model’ option, model cells of uniform thickness right up to the left
and right edges of the survey line are used (Figure 41c). This is probably an extreme case. As
the number of model parameters increase, the computer time needed to carry out the inversion
also increases. This can be an important consideration for very large data sets with several
hundred electrodes. The ‘Make sure model blocks have same widths’ option is probably more
useful. It uses a base model such as in Figure 41b, but avoids thicker cells at the sides. It ensures
that the cells at the sides to have the equal widths.
The ‘Reduce effect of side blocks’ option affects the calculation of the Jacobian matrix
values for the model cells located at the sides and bottom of the model section. Normally, for
a cell located at the side, the contributions by all the mesh elements associated with the model
cell are added up right to the edge of the mesh. This gives a greater weight to the side cell
compared to the interior cells. In some cases, particularly when the robust inversion option is
used, this can result in unusually a high or low resistivity value for the side cell. This option

60
leaves out the contribution of the mesh elements outside the limits of the survey line to the
Jacobian matrix values for the side cells.
The last two options, ’Change width of blocks’ and ‘Type of cross-borehole model’,
are only used for certain special cases, and will not be described here. Please refer to the
RES2DINV.PDF manual for the details.
c). Model sensitivity options
This set of options relate to the model sensitivity (Jacobian matrix) values. The ‘Display
model blocks sensitivity’ option will display the sum of the absolute values of the sensitivity
values associated with the model cell. Figure 41d shows an example. Note the blocks at the
sides and bottom has greater sensitivity values due to the larger sizes. To avoid this effect, the
‘Display subsurface sensitivity’ option divides the subsurface into cells of equal size and shows
the sensitivity values. This is useful to get an idea of the regions of the subsurface “scanned”
by the survey configuration used.
All the techniques used to subdivide the subsurface described in the earlier section are
based on heuristic algorithms. Figure 41d shows the block distribution generated by a more
quantitative approach based the sensitivity values of the model cells. This method is selected
by the ‘Generate model block’ option. This technique takes into account the information
contained in the data set concerning the resistivity of the subsurface for a homogeneous earth
model. It tries to ensure that the data sensitivity of any cell does not become too small (in which
case the data set does not have much information about the resistivity of the cell).
The next set of inversion options are grouped below the ‘Change Settings’ choice on
the main menu bar. Clicking this will bring up the list of options shown below.
Figure 43. The options under the ‘Change Settings’ menu selection.
d). INVERSION DAMPING PARAMETERS

This set of options is related to the damping factor  used in the least-squares equations
(4.1) to (4.3). By default, the program uses a value of 0.16 for the initial damping factor value
for the first iteration. The damping factor is decreased by about half after each iteration.
However, to avoid instability in the model values due to a damping factor value that is too
small, a minimum limit of 0.015 is used by the program.
The ‘Damping factors’ option allows you to set the initial damping factor and the
minimum limit. If the data set appears to be very noisy, and unusually high or low model
resistivity values are obtained with the default values, try using larger values for the damping
factors.

61
Since the model resolution decreases with depth, the program increases the damping
factor value by about 5% for each deeper layer. You can change this factor using the “Change
damping factor with depth’ option. Instead of automatically decreasing the damping factor by
half after each iteration, the ‘Optimize damping factor’ option allows the program to look for
an optimum damping factor. The time taken per iteration will be more, but fewer iterations
might be needed for the program to converge.
The ‘Limit range of model resistivity values’ option is intended for cases where the
default settings (without limits) produces a model with resistivity values that are too high or
too low. Selecting this option will bring up the dialog box shown in Figure 44. In this example,
the upper limit for is 20 times the average model resistivity value for the previous iteration
while the lower limit is 0.05 times (i.e. 1/20 times). The program uses “soft” limits that allow
the actual resistivity model values to exceed the limits to a certain degree. However, this option
will avoid extremely small or large model resistivity values that are physically unrealistic.
The ‘Vertical/Horizontal flatness filter ratio’ option allows the user to fine-tune the
smoothness-constrain to emphasize vertical or horizontal structures in the inversion model.
Some geological bodies have a predominantly horizontal orientation (for example sedimentary
layers and sills) while others might have a vertical orientation (such as dykes and faults). This
information can be incorporated into the inversion process by setting the relative weights given
to the horizontal and vertical flatness filters. If for example the structure has a predominantly
vertical orientation, such as a dyke, the vertical flatness filter is given a greater weight than the
horizontal filter.
Figure 44. The dialog box to limit the model resistivity values.
e). Forward modeling method settings

This set of options controls the finite-element or finite-difference mesh used by the
forward modeling subroutine. The ‘Finite mesh grid size’ option changes the mesh size in the
horizontal direction, while the ‘Mesh refinement’ option changes the mesh settings in the
vertical direction. Using a finer mesh normally increases the accuracy of the forward modeling
subroutine, particularly if very large resistivity contrasts are present.
4.4 Using the model sensitivity and uncertainty values

The depth of investigation and sensitivity sections described in section 2.5 gives an idea
of the depth of the regions sensed by a single electrode configuration. A 2-D survey typically
has hundreds of data points collected with electrodes at different locations and spacings. A
question that frequently arises in 2-D interpretation is as follows. What are the regions of the
subsurface sensed by the survey, and what is the reliability of the results? The first question
can be easily determined, but at present there is no simple answer to the second.
The “Display blocks sensitivity” option under the “Inversion” menu will show a plot of
the sensitivity of the cells used in the inversion model. The cumulative sensitivity value (see

62
equation 2.17 in section 2.5.12) is a measure of the amount of information about the resistivity
of a model block cell in the measured data set. The higher the sensitivity value, the more reliable
is the model resistivity value. In general, the cells near the surface usually have higher
sensitivity values because the sensitivity function has very large values near the electrodes. The
cells at the sides and bottom can also have high sensitivity values due to the much larger size
of these cells that are extended to the edges of the finite-difference or finite-element mesh (the
program has an option to reduce this effect which might produce artifacts at the edges of the
model). If you had carried out an inversion of the data set before calling this option, the program
will make use of the Jacobian matrix of the last iteration. Otherwise, it will calculate the
Jacobian matrix for a homogenous earth model. Figure 41d shows an example of a plot of the
sensitivity section for a model.
Figure 45b shows the model section obtained from the inversion of a data set for a
survey to map leakage of pollutants from a landfill site (Niederleithinger, 1994). The model
sensitivity section in Figure 45c shows high sensitivity values near the surface with decreasing
values with depth. This is to be expected as the near surface materials have a larger influence
on the measured apparent resistivity values.
Figure 45. Landfill survey example (Wenner array). (a) Apparent resistivity pseudosection ,
(b) model section, (c) model sensitivity section (d) model uncertainty section, (e) minimum
and maximum resistivity sections.
In order to assess the accuracy of the inversion model, an estimate of the reliability of the model
values is required. One possible approach is by using the model covariance matrix (Menke,
1984). This is commonly used for models that consist of a small number of parameters (such
as the 1-D model in Figure 7). Figure 45d shows the model uncertainty values obtained from
the covariance matrix method as described by Alumbaugh and Newman (2000) where the
smoothness constraint is included in the model uncertainty estimate. In this way, the model
uncertainty values are less sensitive to size of the model cells. However, the uncertainty values
are only meaningful if the subsurface resistivity varies in a smooth manner, as assumed by the
smoothness constraint. If the subsurface resistivity does not vary in a smooth manner, this
method is likely to underestimate the actual uncertainty. Figure 45e shows the maximum and
minimum resistivity values of each cell at the limits of the model uncertainty range. Features

63
that are common to both model sections can be considered more reliable.
Table 5. Tests with different inversion options

GRUNDFOR.DAT – An (1). Read in the data set using the ‘File’ and then carry out
example of a field data with the inversion with the default model discretization. You
smooth variation of the can take a look at the way the subsurface is divided into
resistivity values. The purpose cells by selecting the ‘Display model blocks’ option.
is to see the effect of the (2). Now choose the option to ‘Allow number of model
different model discretizations. blocks to exceed data points’, and run the inversion again.
The purpose of this field survey Make sure to use a different name for the inversion results
was to map the soil lithology file, for example GRUNFOR2.INV. Check out the
(Christensen and Sorenson arrangement of the cells again using the ‘Display model
1994). blocks’ option.
(3). Now we will reduce the width of the side cells as well.
Select the ‘Make sure model blocks have same widths’
option, and check out the arrangement of the cells. Next
run the inversion again.
(4). Finally we will use an arrangement with even more
model cells. Select the ‘Use extended model’ option, and
then run the inversion.
As the number of model cells increase, the computer time
increases. Which do you think is the best compromise
between getting a finer model and reducing the computer
time?
BLOCK_ONE.DAT – A (1). After reading in the data file, choose the ‘Include
synthetic test model. To show smoothing of model resistivity’ option to ensure inversion
advantage of robust l1 norm model is smooth. Run the inversion and see what you get.
inversion for models with sharp You might like to display the inversion model in the form
boundaries. of blocks rather than contours. This shows up the smearing
out of the boundaries better. Compare the maximum and
minimum model resistivity values with the true values.
(2). Next select the ‘Select robust inversion’ option, and
the enable both the robust model and data constrains.
Check out the resulting inversion model.
ODARSLOV.DAT – A field (1). To get the best results, enable both the option to
data set with a vertical dike. “Allow number of model blocks to exceed datum points”.
Make sure the option ‘Include smoothing of model
resistivity’ is on to get a maximally smooth model.
(2). Next use the robust inversion method. You should see
a dramatic change in the model.
(3). One of the undesirable side effects of the robust
inversion option is the model resistivity values at the
bottom-left and bottom-right corners can take extreme
values. To reduce this effect, select the ‘Reduce effect of
side blocks’ option, and then run the inversion again.
A different approach is to empirically determine the depth of investigation of the data
set is by carrying out at least two inversions of the data set using different constraints.
Oldenburg and Li (1999) used the following least-squares formulation to carry out the 2-D
inversion.

64
J 
J FR Δqk  J T R d g  FR (q k  q o ) ,
T
(4.4)
with FR   s   x C x R m C x   z C z R m C z .
T T
qo is a homogeneous half-space reference model and s is an additional “self” damping factor

that has a value of about 0.0001 to 0.01 times the x and z damping factors. In the proposed
method, two inversions are carried out using different resistivity values for the reference model.
Typically, the second reference model has a resistivity of 10 to 100 times the first reference
model. From the model resistivity values, the following depth of investigation (DOI) index is
calculated.
m x, z   m2 x, z 
R  x, z   1 (4.5)
m1r  m2 r
m1r and m2r are the resistivity of first and second reference models, m1(x,z) and m2(x,z) are
model cell resistivity obtained from the first and second inversions. R will approach a value of
zero where the inversion will produce the same cell resistivity regardless of the reference model
resistivity. In such areas, the cell resistivity is well constrained by the data. In areas where the
data do not have much information about the cell resistivity, R will approach a value of one as
the cell resistivity will be similar to the reference resistivity. Thus the model resistivity in areas
where R has small values is considered to be “reliable”, while in areas with high R values are
not reliable.
The model used to calculate the DOI index use cells that extend to the ends of the survey
line, and a depth range of about three to five times the median depth of investigation of the
largest array spacing used. This ensures that data has minimal information about the resistivity
of the cells near the bottom of the model, i.e. in theory the bottom cells have DOI values of
almost 1.0.
Figure 46a shows the inversion model of the landfill survey data set with a depth range
of about 3.5 times the maximum pseudodepth in the apparent resistivity pseudosection in
Figure 45a. The resistivity of the reference model used is obtained from the average of the
logarithms of the apparent resistivity values. Figure 46b shows the DOI plot calculated after
carrying out a second inversion using a reference model with 100 times the resistivity of the
first reference model. Oldenburg and Li (1999) recommended using a value of 0.1 as the cut-
off limit for the effective depth of investigation of the data set. The depth to the contour with a
DOI value of 0.1 is about 28 m. along most of the survey line. This is close to the maximum
median depth of investigation of about 25 m. for this data set. There are shallower regions with
high DOI value, particularly below the 50 m. mark. This is probably partly due to the low
resistivity plume that limits the amount of current flowing into the deeper sections below it.
The resistivity method uses the electrical current as a probing tool, and where the current flow
is limited, the amount of information provided is less. Note the regions with high DOI values
at the sides of the section. This is to be expected as the sides of the survey line have less data
points compared to the center.
Figure 47 shows the DOI plot from a survey to map the boundary between the salt and
fresh water zones across a beach in Denmark (Marescot and Loke, 2003). The profile is
perpendicular to the seashore and the electrodes 1 to 16 are under the seawater, while the rest
of the profile is above the water. Note the strong increase in DOI values on the left part of the
profile that is covered with seawater. As this section is covered with seawater, most of the
current actually flows within the seawater and does not penetrate into the subsurface. Thus the
depth of investigation there is much shallower.

65
Figure 46. Landfill survey depth of investigation determination. (a) Model section with
extended depths and (b) the normalized DOI index section.
Figure 47. Beach survey example in Denmark. The figure shows the model section with
extended depths and the normalized DOI index section.

66
The DOI method is useful in marking the regions where the model values are well
constrained by the data set, and thus greater confidence can be placed on the model resistivity
values at such regions. The DOI method may be considered an empirical method to determine
the regions where we can reasonably resolve the subsurface. Another method based on the
model resolution values is discussed in section 4.9.
4.5 Methods to handle topography

In surveys over areas with significant changes in the elevation of the ground surface,
the effect of the topography must be taken into account when carrying out an inversion of the
data set. It is now generally recognized that the traditional method of using the “correction
factors” for a homogeneous earth model (Fox et al., 1980) does not give sufficiently accurate
results if there are large resistivity variations near the surface (Tong and Yang, 1990). Instead
of trying to “correct” for the effect of the topography on the measurements, the preferred
method now is to incorporate the topography into the inversion model. The RES2DINV
program has three different methods that can be used to incorporate the topography into the
inversion model (Loke, 2000).
The three methods are similar in that they use a distorted finite-element mesh. In all
these methods, the surface nodes of the mesh are shifted up or down so that they match the
actual topography. In this case, the topography becomes part of the mesh and is automatically
incorporated into the inversion model. The difference between these three methods is the way
the subsurface nodes are shifted. The simplest approach, used by the first finite-element
method, is to shift all the subsurface nodes by the same amount as the surface node along the
same vertical mesh line. This is probably acceptable for cases with a small to moderate
topographic variation (Figure 48b).
In the second approach, the amount the subsurface nodes are shifted is reduced in an
exponential manner with depth (Figure 48c) such that at a sufficiently great depth the nodes
are not shifted. This comes from the expectation that the effect of the topography is reduced or
damped with depth. This produces a more pleasing section than the first finite-element method
in that every kink in the surface topography is not reproduced in all the layers. For data sets
where the topography has moderate curvature, this is probably a good and simple method. One
possible disadvantage of this method is that it sometimes produces a model with unusually
thick layers below sections where the topography curves upwards. Thus in Figure 48d, the
model is probably slightly too thick near the middle of the line where the topography curves
upwards and too thin towards the right end of the line where the topography curves downwards.
The resulting model is partly dependent on the degree of damping chosen by the user. A value
of 0.5 to 1.0 is usually used for the topography damping factor in the RES2DINV program.
One main advantage of this method is that it can be easily implemented, particularly for 3-D
models (Holcombe and Jirack, 1984).
In the third method, the inverse Schwartz-Christoffel transformation method (Spiegel
et al., 1980) is used to calculate the amount to shift the subsurface nodes (Loke, 2000). Since
this method takes into account the curvature of the surface topography it can, for certain cases,
avoid some of the pitfalls of the second finite-element method and produces a more “natural”
looking model section (Figure 48e). For this data set, this method avoids the bulge near the
middle of the line produced by the second finite-element method with a damped distorted mesh.
However, in the middle part of the line, the model produced by this method is slightly thicker
that that produced by the first finite-element method with a uniform distorted mesh.

67
Figure 48. Different methods to incorporate topography into a 2-D inversion model. (a)
Schematic diagram of a typical 2-D inversion model with no topography. A finite-element
mesh with four nodes in the horizontal direction between adjacent electrodes is normally used.
The near surface layers are also subdivided vertically by several mesh lines. Models with a
distorted grid to match the actual topography where (b) the subsurface nodes are shifted
vertically by the same amount as the surface nodes, (c) the shift in the subsurface nodes are
gradually reduced with depth or (d) rapidly reduced with depth, and (e) the model obtained
with the inverse Schwartz-Christoffel transformation method.

68
Table 6. Tests with different topographic modeling options

RATCHRO.DAT – An (1). After reading the data file, select the robust inversion
example field data with option as the burial chamber probably has a sharp contrast
significant topography. This with the soil. After that, click the ‘Display topography’
survey was conducted over a option under the ‘Topography Options’ in the Main Menu
possible ancient burial mound bar.
in Ireland (Waddell and Barton (2). Next click the ‘Type of topographic modeling’ option.
1995). Select the uniformly distorted grid method. Run the
inversion, and then switch to the ‘Display’ window to take
a look at the results. In the ‘Display’ window, choose the
‘Display sections’ followed by ‘Include topography in
model display’ options.
(3). Now, select the damped distorted grid option in the
‘Type of topographic modeling’ dialog box. Run the
inversion again, and then take a look at the model section
with the topography. Now we will reduce the width of the
side cells as well. Select the ‘Make sure model blocks have
same widths’ option, and check out the arrangement of the
cells. Next run the inversion again.
(4). Next select the inverse Schwartz-Christoffel
transformation method in the ‘Type of topographic
modeling’ dialog box. Run the inversion again, and take a
look at the model with the topography.
Which topographic modeling method do you prefer?
GLADOE2.DAT – A data set (1). If you have the time, run the same tests as you did
taken to check for leakage from earlier for the RATHCRO.DAT data set.
a dam (Dahlin pers. comm.). This is another example where the option to reduce the
The survey area has effect of the side cells makes a significant difference when
topography. the robust inversion option is used.
4.6 Incorporating information from borehole logs and seismic surveys

In some areas, information from borehole logs is available concerning the resistivity of
part of the subsurface. This program allows you to fix the resistivity of up to 1000 sections of
the subsurface. The shape of the section to be fixed must be rectangular or triangular. Borehole
logs give the resistivity of the formations along the borehole. However, some caution must be
used when incorporating the borehole log information into an inversion model for surface
measurements. Borehole measurements usually only give the resistivity of a very limited zone
near the borehole. Depending on the type of instrument used, the borehole log generally
samples the subsurface within about 1 meter from the borehole. In contrast, the inversion model
gives the average resistivity of a much larger region of the subsurface. Thus the RES2DINV
program uses a flexible method to incorporate the borehole log information.
The format used by the input data file for RES2DINV to fix the model resistivity values
is described in detail in the RES2DINV manual. Here, we will only look at a fragment of it to
illustrate the general ideas involved. As an example, part of the example data file
MODELFIX.DAT with the resistivity fixing option is listed below.

69
2 | Number if regions to fix, put 0 if none

R | Type of first region, R for rectangular
24,0.7 | X and Z coordinates of top-left corner of rectangle
28,2.3 | X and Z coordinates of bottom-right corner of rectangle
2.0 | Resistivity value of rectangular region
2.0 | Damping factor weight
T | Type of second region, T for triangular
30,0.0 | X and Z coordinates of first corner of triangle
30,3.0 | X and Z coordinates of second corner of triangle
45,3.0 | Coordinates of third corner of triangle
10.0 | Resistivity value of triangular region
2.0 | Damping factor weight
The first item is the number of regions where the resistivity is to be specified. In the example
above, 2 regions are specified. If a value of 0 is given (default value), then there is no region
where the resistivity is specified by the user. Next, the shape of the region is given, R for
rectangular or T for triangular. If a rectangular region is specified, then the X and Z coordinates
of the top-left and bottom-right corners of the rectangle are given, as shown in the Figure 49.
If a triangular region is chosen, the X and Z coordinates of the 3 vertices of the triangle must
be given in an anti-clockwise order. After the coordinates of the region to be fixed are given,
the next data item is the resistivity of the region. After that, the damping factor weight for the
resistivity of the region is needed. This parameter allows you control the degree in which the
inversion subroutine can change the resistivity of the region. There is usually some degree of
uncertainty in resistivity of the region. Thus, it is advisable that the program should be allowed
(within limits) to change the resistivity of the region. If a damping factor weight of 1.0 is used,
the resistivity of the region is allowed to change to the same extent as other regions of the
subsurface model. The larger the damping factor weight is used, the smaller is the change that
is allowed in the resistivity of the "fixed" region. Normally, a value of about 1.5 to 2.5 is used.
If a relatively large value is used, for example 10.0, the change in the resistivity of the region
would be very small during the inversion process. Such a large value should only be used if the
resistivity and shape of the region is accurately known. Figure 50 shows the allocation of the
cells in the subsurface model together with the fixed regions for the MODELFIX.DAT data
set.
Seismic refraction and seismic reflection surveys are commonly used in engineering
surveys. Both methods can give accurate and detailed profiles of the subsurface interfaces. In
some cases, a distinct and sharp transition between two layers can be mapped by the seismic
survey. This information can be used to improve the results from the inversion of a 2-D
resistivity imaging survey along the same line. The subsurface in the inversion model can be
divided into two zones, one above and one below the interface calculated from the seismic
survey. The resistivity values are constrained to vary in a smooth manner within each zone, but
an abrupt transition across the zone boundary is allowed by removing any constrain between
the resistivity values below and above the zone boundary (Smith et al. 1999).
Figure 51 shows an example where the boundaries of a clay and a gravel layer were
known from a seismic refraction survey (Scott et al., 2000). The known boundaries were then
incorporated into the resistivity inversion model (Figure 51a) that allows the model resistivity
to change abruptly across the boundaries. Note the sharp contrast across the gravel layer and
the underlying low resistivity clay layer (marked by the bottom blue region in lower section in
Figure 51b).

70
Figure 49. Fixing the resistivity of rectangular and triangular regions of the inversion model.
Table 7. Tests with option to fix the model resistivity

MODELFIX.DAT – A (1). After reading the data file, select the robust inversion
synthetic data file with two option as the synthetic model has sharp boundaries. Next
fixed regions. choose the ‘Display model blocks’ option to take a look at
the locations of the fixed regions. Run the inversion of the
data set.
CLIF4_NORMAL.DAT – A (1). After reading in the CLIF4_NORMAL.DAT data file,

data set from the Clifton, select the robust inversion option, and then run an
Birmingham area (Scott et al. inversion of the data set. In this case, we have not included
2000) to map layers within the additional constrains on the inversion. You might have to
unconsolidated sediments. use the option to “Reduce effect of the side blocks” to
avoid distortions at the bottom-left and bottom-right
CLIF4_LAYERS.DAT – The corners of the inversion model.
same data set but with the
layers specified. The depths to (2). Do the same for the CLIF4_LAYERS.DAT data file.
the layers were determined In this file, the depths to two layers determined from a
from a seismic refraction seismic refraction survey have been included. Take a look
survey. the distribution of the model cells, as well as the layers
specified, using the ‘Display model blocks’ option. Run
the inversion again, and compare the results with that
obtained earlier without the layers.

71
Figure 50. The inversion model cells with fixed regions. The fixed regions are drawn in purple.
Note that the triangular region extends beyond the survey line.
Figure 51. Example of an inversion model with specified sharp boundaries. (a) The boundaries
in the Clifton survey (Scott et al., 2000) data set is shown by the blue lines. (b) The measured
apparent resistivity pseudosection and the inversion model section.

72
4.7 Model refinement

Normally RES2DINV uses a model where the width of the interior model cells is the
same as the unit electrode spacing (for example as in Figure 51a). In some situations with large
resistivity variations near the ground surface, this might not be sufficiently accurate. The model
with a cell width of one unit electrode spacing has a maximum possible misfit of one-half the
electrode spacing for a near-surface inhomogeneity (Figure 52a). In some cases, this misfit can
cause significant distortions in the lower sections of the inversion model. The cell size is too
coarse to accurately model the anomalies due to the small near-surface inhomogeneities. This
forces the inversion program to distort the lower sections of the model in an attempt to reduce
the data misfit.
A finer model with a cell width of half the electrode spacing has a maximum misfit of
one-quarter the unit electrode spacing (Figure 52b), so the effect of the model cell boundary
misfit should be much less. In theory, it is possible to reduce the cell width further, but the error
due to the misfit becomes increasingly less significant. Reducing the cell increases the number
of model parameters, thus increasing the computer time and memory required.
Figure 52. The effect of cell size on the model misfit for near surface inhomogeneities. (a)
Model with a cell width of one unit electrode spacing. (b) A finer model with a cell width of
half the unit electrode spacing. The near surface inhomogeneities are represented by coloured
ovals.
Figure 53 shows a synthetic model used to illustrate the effect of the model cell misfit.
The main structure is a faulted block of 100 m and a rectangular prism of 1 m in a medium
of 10 m. A series of small near-surface high resistivity blocks with widths of 1.0, 0.75, 0.50
and 0.25 m. and resistivity of 300 m are placed above the faulted block. A similar series of
low resistivity blocks of 1.0 m are located to the left.The pole-dipole array has the P1-P2
dipole length (“a”) fixed at 1.0 m., but with “n” factor ranging from 1 to 16. Note the strong
anomalies produced by the near-surface inhomogeneities. Note also that the Wenner array is
much less affected by the near-surface inhomogeneities. The reason lies in the sensitivity
patterns of the two arrays (compare Figure 23 and Figure 30). For the pole-dipole array with
large “n” values, the region with the highest positive sensitivity values is concentrated below
the P1-P2 dipole pair.
Figure 54 shows the inversion results for the pole-dipole data set with different cell
widths. The model with a cell width of 1.0 m. (i.e. one unit electrode spacing) shows significant
distortions near the top of the faulted block as well as in the low resistivity rectangular block
(Figure 54b). Most of the distortions are removed in the model with a cell width of half the unit
electrode spacing (Figure 54c). This means the residual misfits with widths of up to one-quarter
the unit electrode spacing does not have a significant effect on the calculated apparent
resistivity values. The model with a cell width of one-quarter the unit electrode spacing (Figure
54d) does not show any major improvement over the half-cell width model although in theory
it should more accurately model the near surface inhomogeneities. In fact, there is a poorer
agreement with the true model in the lower part of the faulted block (Figure 54d). Experiments

73
with a number of data sets show that using a cell width of one-quarter the unit electrode spacing
can sometimes lead to oscillating model resistivity values. This is probably because the data
does not have sufficient information to accurately resolve such small cells.
Figure 53. Synthetic model (c) used to generate test apparent resistivity data for the pole-dipole
(a) and Wenner (b) arrays.
Figure 55 shows an example from a survey over an underground pipe using the Wenner-
Schlumberger array. There are large resistivity variations near the surface, probably due to
stones in the topmost layer of the soil. If the effect of the near surface variations are not
accurately accounted for by the inversion model, it can lead to distortions in the lower portions
of the model as the programs attempts to reduce the data misfit by distorting the lower part of
the model. The model with a cell width of half the unit electrode spacing in Figure 55c shows
a slightly better fit with the measured data (i.e. lower RMS error) and a more circular shape for
the low resistivity anomaly below the 12 meters mark.
In conclusion, for most cases, using a cell width of half the unit electrode spacing seems
to give the optimum results. Using a cell width of one-third the unit spacing seems to be
beneficial only a certain cases with the pole-dipole and dipole-dipole arrays with very large ‘n’
values (see section 4.10). A cell width of one-quarter the unit spacing sometimes leads to
instability with oscillating model values, particularly in the first few layers. Thus the use of a
cell width of less than one-quarter the true unit electrode spacing is not advisable.

74
Note that using finer cells will lead to longer inversion times, so using a width of the
half the unit electrode spacing seems to provide the best trade-off. These examples present a
strong case for using a model with a cell width of half the unit electrode spacing as the default
choice in the inversion of most data sets. It avoids the problem caused by model cells boundary
misfits if the cell is too coarse, and the increase in computer time is tolerable.
One effect of using finer model cells for some data set is the emergence of 'ripples' in
the first couple of layers in the inversion model. Figure 56a shows the pseudosection from a
survey in the Blue Ridge region in north-east USA (Seaton and Burbey, 2000). The inversion
model using the blocky model inversion norm in Figure 56b show some rapid near-surface
alternating resistivity values between the 160 and 240 meter marks. These artifacts can be
significantly reduced by using a higher damping factor for the top layer as shown in Figure
56c.
As a final note, it appears that the effect of using narrower model cells is less dramatic
in 3-D inversion. This is probably because in the 2-D model each cell extends to infinity in the
y-direction, whereas in the 3-D model the same cell is divided into a large number of much
smaller cells. Thus the effect of a single near-surface cell in the 3-D model on the calculated
apparent resistivity values is much smaller than in the equivalent 2-D model.
Figure 54. The effect of cell size on the pole-dipole array inversion model. Pole-dipole array.
(a) The apparent resistivity pseudosection. The inversion models obtained using cells widths
of (b) one, (b) one-half and (c) one-quarter the unit electrode spacing.

75
Figure 55. Example of the use of narrower model cells with the Wenner-Schlumberger array.
(a) The apparent resistivity pseudosection for the PIPESCHL.DAT data set. The inversion
models using (b) cells with a width of 1.0 meter that is the same as the actual unit electrode,
and (c) using narrower cells with a width of 0.5 meter.
Figure 56. Example of reduction of near-surface 'ripples' in inversion model. (a) Apparent
resistivity pseudosection for the BLUERIDGE.DAT data set. (b) Normal inversion using the
robust inversion norm and model refinement. (c) Inversion using higher damping factor for
first layer to reduce the 'ripple' effect in the top layer.

76
4.8 Fast inversion of long 2-D survey lines

Mobile resistivity systems such as the capacitive coupled systems or aquatic surveys
(section 2.3) can produce very long survey lines. For example, a line can easily cover 10
kilometers with readings made at every 1 meter. Frequently only a very limited number of
different array configurations are used, so the data density and maximum depth of investigation
tends to be low compared with conventional systems. There are a number of techniques that
can be used to greatly reduce the inversion time required for such data sets.
4.8.1 Preprocessing steps

While data measured using a standard multi-core cable system have electrode positions that are
equally spaced, measurements made with a mobile system normally have the electrode
positions that are not regularly spaced. Data collected with a multi-core cable system will have
electrode positions with a minimum spacing equal to the spacing between adjacent takeouts on
the cable. In contrast, data measured with a mobile system can have measurement positions
that are much less than the minimum spacing between electrodes positions in the towed cable.
To overcome the problem of measurement positions that are not equally spaced, the true
electrode positions are rounded up to the nearest convenient chosen unit electrode spacing. A
guide to an appropriate value to select for the 'unit electrode spacing' is the minimum takeout
spacing in the cable. As shown in the previous section, the measured data is unlikely to resolve
structures that are much less than the half the minimum spacing between electrodes in the cable.
Thus a value of between half to one-quarter the minimum cable takeout spacing is usually used.
As an example, if the minimum takeout spacing is 3 meters, a convenient value might be 1
meter. The program can automatically shift the positions of the electrodes for the data points
using the 'File - Round up positions of electrode' option shown below.
This will round up the positions of the electrodes for all the data points to the nearest unit
electrode spacing specified in the data file. Once the data in the correct format, we next look at
ways to reduce the computer time needed to invert the data set.
4.8.2 Data set and computer system used for tests

The data set from an underwater survey where the cable was dragged along the bottom of a
river will be used as an example. A section of the data set with a line length of nearly 2600
meters with a nominal unit electrode spacing of 1 meter is used (Figure 57a). The Res2dinvx64
program is used for the data inversion on a PC with a hex-core Intel i7 970 CPU.

77
4.8.3 Finite-element mesh size

For the underwater survey data set, the Res2dinv program will automatically use the finite-
element method. Figure 58 shows part of the finite-element mesh used in calculating the
potential values. The first setting we will look at is the horizontal mesh size. This setting is
accessed via the 'Change Settings - Forward modeling methods settings' option that brings up
the following dialog box.
The program has options to use 2 or 4 nodes between adjacent electrodes. For a long survey
line, the program automatically selects the 2 nodes option. Here we will examine the effect of
using 2 or 4 nodes on the resulting inversion model. Normally selecting the 4 nodes option
increases the accuracy of the finite element routine used. However, due to the nature of the data
from a mobile survey, the effect might be minimal. There should be at least 4 horizontal nodes
between two electrodes in an array used in a measurement. In mobile surveys, the distance
between adjacent electrode positions is frequently much smaller than the electrode spacing in
an array. The example in Figure 59 has a spacing of 1 meter between adjacent electrode
positions but has an electrode spacing of 3 meter in an array. Thus there are 6 nodes between
two electrodes in the same array although a 2 nodes spacing is used between adjacent electrode
positions. As shown in Table 8, using 2 nodes instead of 4 nodes reduces the inversion time by
almost half (from 2093 to 1074 seconds) for the test data set. The effect on the resulting
inversion model is minimal as shown by Figure 57b,c. The difference in the data misfit is only
about 0.1% and there are no obvious differences in the model sections.
4.8.4 Limit calculation of Jacobian matrix values

The program normally calculates the Jacobian matrix values for all the model cells for every
data point. The data set in Figure 57 has 2293 data points while the inversion model has 11673
cells. The Jacobian matrix value at a model cell far from the electrodes used in a measurement
is likely to be very small, and can be neglected without significantly affecting the resulting
inversion model. This option is accessed vai the 'Inversion - Inversion methods and settings -
Fast inversion of long survey lines'. Selecting the 'Fast calculation of Jacobian matrix' option
will limit the calculation for model cells that are 'close' to the electrodes in the array. This

78
further reduces the calculation time by about 14% (Table 8), while the inversion model (Figure
57d) is essentially identical to the one obtained with the full Jacobian matrix (Figure 57c).
4.8.5 Use sparse inversion techniques

Mobile electrical resistivity imaging surveys can have thousands of electrode positions, data
points, and model cells. Because long survey lines results in a finite-element grid where the
number of nodes in the horizontal direction is very much greater than the number of nodes in
the vertical direction, the calculation time can be greatly reduced by splitting up the finite-
element grid into a number of overlapping sub-grids. This also automatically reduces the size
of the Jacobian matrix as well. This method is enabled using the 'sparse inversion techniques'
option the in above dialog box. This method will still produce a contiguous model along the
entire survey line, but takes advantage of the segmented finite-element grid and sparse nature
of the Jacobian matrix to reduce the number of calculations needed. Using this option makes a
major reduction in the calculation time (by nearly a factor of 5 times) to 210 seconds, while the
resulting inversion model (Figure 57e) does not show significant differences from previous
models.
4.8.6 Use wider cells

The smallest spacing between two electrodes used in the same array for the data set is 3 meters
while the width of the model cells is 1 meter. Since we have seen that most arrays are unlikely
to resolve structures less than half the minimum array electrode spacing, we next examine the
use of coarser model cells. This is accessed via the 'Inversion - Model discretization - Change
width of model blocks' option that brings up the following dialog box. Selecting a value of '2'
will create a model where the cells have widths of twice the nominal unit electrode spacing (1
meter in this example). The resulting models with cells widths of 2 and 3 meters are shown in
Figure 57f,g. Using a model cell width of 2 meters gives a significant reduction in the
calculation time if about 15% with very small changes in the resulting inversion model. This
probably provides the best compromise for this data set. As a general guide, using a slight
smaller cells width of about one-third to two-thirds the smallest electrode spacing in the arrays
used seems to provide the best results.

79
4.8.7 Overview of methods to reduce the calculation time

The methods described have reduced the calculation time for the example data set by more than
a factor of 10, from 2093 to 181 seconds. Table 9 shows the inversion times for data sets from
the same survey with different line lengths. Note the largest data set with nearly 7500 electrode
positions takes less than 17 minutes on an inexpensive desktop PC. The inversion of a 10
kilometers survey line with a nominal unit electrode spacing of 1 meter will take less than 30
minutes. The field survey time will certainly be much longer than the data inversion time.
Table 8. Inversion times for the long Redas survey data set using different settings.
Trial Setting used Time (s)
1 Standard settings with 4 nodes between adjacent electrodes. 2093
2 Standard settings with 2 nodes between adjacent electrodes. 1074
3 Use fast calculation of Jacobian matrix values 934
4 Use sparse inversion techniques 210
5 Use 2 meter model cells 181
6 Use 3 meter model cells 171
Table 9. Inversion times for different line lengths.

Length No. of electrodes No. of data No. of model cells Time (s)
(m.) points
1589 1413 1244 6354 78
2912 2595 2293 11673 181
4099 3672 3252 16524 290
8329 7479 6636 33651 995

80
Figure 57. Inversion models for a long survey line using different settings to reduce the
calculation time. (a) The measured apparent resistivity pseudosection for the Redas underwater
survey for the first 2600 meters. Inversion models using (b) standard inversion settings with 4
nodes between adjacent electrode positions, (c) standard inversion settings with 2 nodes
between adjacent electrode positions, (d) with calculation of Jacobian matrix values for
selected model cells, (e) with sparse inversion technique, (f) with 2 meter model cell width and
(g) with 3 meter model cell width.
Figure 58. Part of finite-element mesh used to model a survey with submerged electrodes. The
resistivity of mesh cells in the water layer are fixed at the known water resistivity.

81
Figure 59. Part of a finite-element mesh for a long survey line. The electrodes used in the same
array are 3 meters apart, while the data unit electrode spacing is 1 meter. Using 2 nodes between
adjacent electrodes in the mesh will actually result in 6 nodes between electrodes in the same
array.
4.9 Model resolution and automatic array optimization methods

4.9.1 Concept of model resolution
The sensitivity functions plots are useful explaining the behavior of common arrays,
but they only tell us the characteristics of each individual array. This will help in selecting an
array for certain structures, eg. for predominantly horizontal or vertical structures. However, a
survey consists of many measurements with different spacings and horizontal positions along
the line. The cumulative sensitivity values described in the previous section is a simple method
to combine the sensitivity values from different measurements. It can be calculated rapidly but
it is not mathematically rigorous. Our purpose in carrying out the survey is to detect or resolve
structures below the survey line. How do we quantify the model ‘resolution’?
The model resolution equation (Day-Lewis et al., 2005) is related to the least-squares
equation used for the inversion for resistivity data described earlier in section 1.4.
 
J T J F Δqk  J T g  Fq k , (4.6)
Using linear approximations, it can be shown that the relation between the calculated model
resistivity and the true resistivity is approximately given by
q Model  Rq True ,
where

R  J T J F  1
JTJ (4.7)
The model resolution matrix R may be viewed as a filter through which the inversion
method attempts to resolve the subsurface resistivity. In the ideal case with perfect resolution,
the elements of the main diagonal (Rjj) are 1 while the off-diagonal elements are 0.
4.9.2 A heuristic explanation of model resolution – through a glass darkly

This section attempts to give an illustration of model resolution in a less mathematical
manner. The effect of the model resolution matrix can be shown diagrammatically below.
Consider an original optical image used in testing eyesight, such as
A person with less than perfect eyesight might see it as
Someone with very poor eyesight might see it as

82
The matrix R might be considered as a ‘blurring’ matrix which contaminates the

calculated value with values from nearby cells. To illustrate it in terms of the model used in
resistivity inversion, consider a simple model with only 4 cells.
This can be written in matrix form as
(4.8)
If the cells are perfectly resolved, the model resolution matrix will have the form below
where the diagonal elements are 1.0 and elsewhere it is 0, so the model resolution matrix can
be written as
This means the calculated value for each cell only depends on the true value. In the
case with imperfect resolution, we might have something like
The diagonal elements give the ‘degree’ of resolution, while the off diagonal elements
give the degree of ‘contamination’ or cross-correlation with the neighboring model cells. One
way to illustrate the resolution is to plot the values of the diagonal elements of the R matrix.
This shows the degree at which the calculated model value depends on the true value. Some
authors choose a value of about 0.05 (5%) as the ‘cutoff’ value.
4.9.3 Examples of model resolution with standard arrays

The diagonal elements of the resolution matrix give a quantitative means to compare
the resolving power of different arrays. In this section, plots of the model resolution values of
three common arrays are shown. As a simple example, we consider the case for a multi-

83
electrode system with 30 independent nodes and a spacing of 1.0 meter between adjacent
electrodes. First we look at the resolution for the widely used Wenner array. Assume we make
all the possible measurements, with the “a” spacing ranging from 1.0 to 9.0 meters giving a
total of 135 data points. The model resolution section (Figure 60a) shows that the resolution is
greatest near the surface, decreases rapidly with depth, and is very small below a depth of about
2.0 meters. Next we look at the Wenner-Schlumberger array with the ‘a’ spacing ranging from
1.0 to 3.0 m. and the ‘n’ factor ranging from 1 to 8 giving a total of 292 data points. It performs
significantly better than the Wenner array (Figure 60b) with significant resolution values up to
about 3.0 m. The third array we will look at is the dipole-dipole array with the dipole length
‘a’ fixed at 1.0 m. and the ‘n’ factor ranging from 1 to 6, giving a total of 147 data points. Its
performance is surprisingly good and comparable to the Wenner-Schlumberger array with
significant model resolution values to about 3.5 meters, although it only has about half the
number of data points (Figure 60c). It is much better than the Wenner array although it only
has 12 more data points.
In an attempt to improve the resolution for the dipole-dipole array, we next attempt to
use overlapping data levels with it. The ‘a’ dipole length ranges from 1.0 to 3.0 m. while the
‘n’ factor ranges from 1 to 6. The region with significant resolution values increases to about
5 meters (Figure 60d), an improvement from the 3.5 meters limit with a single ‘a’ dipole length
(although it is mainly concentrated near the center). It only has about 50 data points more than
the Wenner-Schlumberger (342 compared to 292), but it is a significant improvement.
It is possible to further improve the resolution we can get for a limited number of
independent nodes and data points?
Figure 60. Model resolution sections for the (a) Wenner, (b) Wenner-Schlumberger and (c)
simple dipole-dipole array (d) dipole-dipole array with overlapping data levels.
4.9.4 Array optimization methods

The model resolution section gives an insight on the section of the subsurface we can resolve.
Is there a way to automatically select the arrays that will give the maximum resolution for a
given number of electrodes? In recent years, there have been some significant developments in
this effort (Stummer et al., 2004; Wilkinson et al., 2006b; Loke et al., 2010a, 2010b, 2015a).
Consider a survey system with N independent electrodes. The number of possible array
combinations (M) with 4 electrodes is
M=N(N-1)(N-2)(N-3)/8
As an example, for a system with 30 electrodes, there are 82215 possible
configurations. We can reduce the number by rejecting all configurations with the Gamma

84
arrangement (with interleaved current and potential electrodes), and those with very high
geometric factors (i.e. low potentials) of greater than the dipole-dipole with a=1 and n=6. This
reduces the total number of possible array configurations to 51283, which is called the
‘comprehensive’ data set (Stummer et al., 2004; Wilkinson et al., 2006b). If we make
measurements with all these possible array configurations, we should get the best possible
resolution. The model resolution section for this ‘comprehensive’ data set is shown in Figure
61a. The region with significant resolution values extends to about 8 meters, significantly more
than the dipole-dipole (5 meters) array.
Figure 61. (a) Model resolution section for comprehensive data set. (b) Model resolution
section for optimized data set generated by the ‘Compare R’ method.
Although taking all the 51283 possible readings will give the maximum possible resolution,
this is not practical in an actual field survey. We want to take a small subset of the possible
readings that gives almost the same resolution. The main steps taken by the array optimization
routine is as follows.
1. Start with a small set, the ‘base’ data set, such as the dipole-dipole readings with a=1
and n=1 to 6 (147 readings).
2. Calculate the increase in the model resolution as each new array is added to the ‘base’
set (there are several alternatives for this step with differences in speed and accuracy).
Add the arrays that give the highest increase to the model resolution (and are
sufficiently independent of each other) to the base set. After each iteration, increase the
base set by a set percentage (usually about 1% to 5%).
3. Use the new set (base plus new arrays) as the base set, and repeat the procedure. Stop
when the desired maximum number of arrays (eg. 5000) is reached.
Different Methods have been proposed to carry out step (2) in selecting the arrays that
will give the maximum increase in the model resolution (Wilkinson et al., 2006b). Here, only
the results of one method, the ‘Compare R’ method, will be shown. It gives the most accurate
results in terms of optimized arrays, but it also takes the longest computer time to generate the
arrays (Wilkinson et al., 2006b). However the time needed has been reduced by more than
three orders of magnitude using optimized computer code and parallel programming techniques
(Loke et al., 2010a, 2010b, 2015a). The model resolution section for optimized data set (Figure
61b) is almost identical with that for the ‘comprehensive’ data set (Figure 61a). The optimized
data set has only 4462 array configurations, which is about 9% of the ‘comprehensive’ data set.
To illustrate the performance of the different array configurations in detecting
subsurface structures, a test model with four rectangular blocks at different depths (Wilkinson
et al., 2006b) is used (Figure 62). Synthetic data sets were generated for the Wenner, Wenner-
Schlumberger, dipole-dipole and optimized arrays. The data sets were then inverted to recover
back the subsurface resistivity.
Figure 63a shows the resulting inversion model for the Wenner array data set. The two
topmost blocks are well resolved while the third block is poorly resolved. The deepest block is
completely unresolved. The Wenner-Schlumberger array data set (Figure 63b) performs slight

85
better in resolving the third block but still cannot resolve the deepest block. The simple dipole-
dipole array data set with a single dipole length (a=1.0m.) performs significantly better than
even the Wenner-Schlumberger array data set where the third block is well resolved (Figure
63a), even though it has fewer data points (147 compared with 292). However, it still cannot
resolve the deepest block. The dipole-dipole array data set with multiple dipole lengths (all
possible data points with a=1.0 to 3.0 meters and n=1 to 6, with the restriction that the geometric
factor cannot exceed that for a=1.0 meter and n=6) shows higher resistivity values at the
location of the deepest block but it is not able to resolve the block (Figure 63d).
Figure 62. Test inversion model for the different arrays. (a) Apparent resistivity pseudosection
for the simple dipole-dipole array with a dipole length ‘a’ of 1.0 meter with the dipole
separation factor ‘n’ of 1 to 6. (b) Apparent resistivity pseudosection for the dipole-dipole array
with overlapping data levels. (c) The synthetic model with 4 rectangular blocks of 100 .m
embedded in a medium of 10 .m. as used by Wilkinson et al. (2006b).
The deepest block is fairly well resolved in the inversion model for the optimized data
set with 4462 data points (Figure 64a) where the shape and dimensions are close to the true
structure. It might be argued that the better performance of the optimized data set compared to
the dipole-dipole array with multiple dipole lengths is due to the much larger number of data
points (4462 compared to 342). Next we use a truncated subset of the optimized data set with
only 413 data points that is comparable to the dipole-dipole array data set. The inversion model
with the truncated optimized data set (Figure 64b) still shows significant better resolution for
the deepest block compared to the dipole-dipole array model (Figure 63d). While the shape of
the deepest block is significantly less sharp compared to the large optimized data set, it still is

86
better defined compared to the dipole-dipole array model. The third deepest block is also much
better resolved by both optimized data sets with resistivity values of over 50 Ohm.m at the
center compared to less than 30 Ohm.m for the dipole-dipole array model.
This simple example shows that there can be significant improvements in the depth
resolution by using an optimized data set. An old rule of thumb in resistivity sounding is that
the maximum depth of investigation is about one-sixth (17%) the line length, i.e. slightly less
than 6 meters near the center of the line in the above examples. However, using an optimized
data set has pushed this limit close to 8 meters or slightly more than one-fifth (about 28%) the
line length.
The array optimization problem for 2-D surveys has been quite well studied (Wilkinson
et al., 2012). Loke et al. (2014) describes the use of array optimization for cross-borehole
surveys. With recent advances in computer software and hardware, it can be used for long 2-D
survey lines (Loke et al., 2015a) and even 3-D surveys (Loke et al., 2014c). It is interesting to
note since the initial work by Wilkinson et al. (2006b), the calculation time required by the
‘Compare R’ method to generate the optimized arrays for a survey line with 30 electrodes has
been reduced from about 6 hours to 3 seconds, an improvement of about 7,000 times (Loke et
al., 2015a) over 9 years!
Figure 63. Inversion results with the (a) Wenner array, (b) Wenner-Schlumberger array, (c)
simple dipole-dipole array and (d) dipole-dipole array with overlapping data levels.
Figure 64. Inversions models with the (a) optimized data set with 4462 data points and (b) a
truncated optimized data set with 413 data points.
4.9.5 DOI versus model resolution?
In section 4.4 we have looked at the Depth of Investigation (DOI) method to estimate
the practical depth of investigation of field data sets. How does the model resolution section
compare with the DOI? The model sections and DOI section for the Landfill survey data set

87
were shown in Figure 45 and Figure 46. Figure 65 shows a comparison between the model
resolution and DOI sections for this data set. Using a value of 0.05 as the cutoff point for the
model resolution values might be arbitrary as it does not take into account the model
discretization used. If a finer model discretization (with more model cells) is used, we would
expect on the average the model resolution for a cell at the same location will be reduced since
the cell size is smaller. Theoretically the sum of the elements in a column of the model
resolution matrix (equation 4.8) is equals to 1.0. The average value of the array elements in the
column would then be equals to 1.0/m, where m is the number of model cells. A more useful
measure to judge whether a model cell is resolved is the ratio the cell resolution value (i.e. the
diagonal element of the resolution matrix) to this average value. If a value of about 10 is used
for this ratio (model resolution index), then the average maximum depth of survey is probably
about 12 to 14 meters. It is slight deeper between the 72 and 120 meters marks and slightly
shallower between the 24 and 72 meters marks (Figure 65b). This is similar to that shown by
the DOI index section using the 0.1 contour as the cut-off value (Figure 65c) The model
resolution sections shows a more gradual change with depth (and also laterally) in the
resolution values, compared to the DOI index section that has localized regions with high DOI
index values.
One advantage of the model resolution section is that it avoids the localized regions
with high DOI index values (that is sometimes caused by local high resistivity regions, or noise
in the data). It has a theoretical appeal in that it is less empirical in nature compared to the DOI
index method. One disadvantage of the model resolution method is that it requires an inversion
of a matrix with a computer time that is proportional to n3 (where n is the number of model
cells). This makes it impractical for models with more than about 100000 cells that could be a
significant limit for very large 3-D models with present PC technology. Research is being
carried out to extend the practical limit to beyond about 100000 cells. One possible solution is
to use GPU technology (Loke et al., 2010b).
Figure 65. Comparison between the (a) model resolution, (b) model resolution index and (c)
DOI index sections for the Landfill data set.
4.9.6 The streamer design problem using a model resolution approach.

In section 2.5.12 we compared the sensitivity values for several possible streamer
configurations. Here we revisit the streamer design problem using a model resolution approach.
To reduce the edge effects, we plot the resolution sections for 26 consecutive sets of
measurements for the three streamer configurations. This gives 208 (26 x 8) measurements for
each streamer. Figure 66 shows the resulting model resolution sections for the dipole-dipole,
gradient and Wenner-Schlumberger array streamers.

88
Figure 66. The model resolution plots for the three streamer configurations.
The resolution plot for the dipole-dipole array streamer (Figure 66a) shows the
maximum depth of investigation is about 4.5 meters. For long survey lines that consists of
hundreds of measurements the end effects are negligible so the resolution values with almost
flat contours near the center of the section gives an estimate of the maximum depth with

89
significant resolution values. The performance of gradient array streamer (Figure 66b) is much
poorer. The resolution plot shows the maximum depth of investigation is about 3 meters that is
significantly less than the dipole-dipole array streamer. The maximum depth of investigation
of the Wenner-Schlumberger array streamer (Figure 66c) is slightly less than the gradient array
streamer, between 2.5 and 3 meters, and much less than the dipole-dipole array streamer.
Overall, the dipole-dipole streamer has the best depth penetration but it has the lowest signal
strength that has to be compensated for in a practical design. The results from the model
resolution plots are significantly different from the cumulative sensitivity plots (Figure 34), so
some care must be exercised in using the results from the sensitivity plots.
4.10 Pitfalls in 2-D resistivity surveys and inversion

While 2-D resistivity surveys have made the mapping of many complex structures
possible, caution must still be exercised in interpreting the results from the data. Below are
some of the common pitfalls.
(a). Incorrect use of the dipole-dipole array. This is still a surprisingly common problem.
There are two common mistakes in the use of this array. The first is to assume that the depth
of investigation is at the point of intersection of the two 45 diagonals projected from the
dipoles. This greatly overestimates the depth of investigation. For example, for the case where
the dipole separation factor “n” is equals to 6, the point of intersection is about 3 times the
median depth of investigation (see Table 2). The second common mistake is to monotonically
increase the “n” factor, while keeping the dipole length “a” fixed, in an effort to increase the
depth of investigation. This usually results in very noisy and unusable data, with negative
apparent resistivity values in some cases, for “n” values of greater than 8. To solve this problem
the “n” value should not exceed 6, and the method of overlapping data levels (section 2.5.6)
with different “a” dipole lengths can be used.
(b). Poor electrode ground contact. This problem arises in stony or dry soils where it is not
possible to plant the electrodes to a sufficient depth, and/or the soil is too dry such that it is not
possible to pass enough current into the ground. In the pseudosection, this is seen as an inverted
“V” shaped pattern of bad data points with the two legs originating from an electrode. This
problem is more severe when the electrode is used as a current electrode. The potential
electrode is less sensitive to poor ground contact, so this problem in certain situations can be
overcome by swapping the current and potential electrodes.
(c). Poor current penetration. The success of the resistivity method depends on a current
flowing through the areas to be mapped. If the top layer has a very high resistivity, it might be
very difficult to get enough current to flow through the ground at all. The opposite problem
occurs if the top layer has an extremely low resistivity. The current might be trapped in the top
layer, so not much information is expected from the lower layers.
(d). Not letting the current charge decay. When an electrode is used as a current electrode,
charges tend to build up around the electrode. When the current is no longer flowing through
the electrode, it still takes a finite amount of time for the charges to disperse. If the same
electrode is used as a potential electrode immediately after it has been used as a current
electrode, this could result in an erroneous reading (Dahlin, 2000).
(e). Mistakes in the field. This can arise from a variety of sources. It could be caused by
instrumentation errors during the field survey, poor electrode contact in dry, sandy or stony
ground, shorting of electrodes due to very wet conditions or metal objects (such as fences, pipes
etc.) or mistakes such as attaching electrodes to the wrong connectors. Fortunately, it is usually
very easy to pick out such bad data points by viewing the pseudosection or the data in the form
of profiles. Before inverting a data set, take a look at it!
(f). 3-D geology. It is assumed that the subsurface is 2-D when interpreting the data from a
single line. This assumption is valid if the survey is carried out across the strike of an elongated

90
structure. If there are significant variations in the subsurface resistivity in a direction

perpendicular to the survey line (i.e. the geology is 3-D), this could cause distortions in the
lower sections of the model obtained. Measurements made with the larger electrode spacings
are not only affected by the deeper sections of the subsurface, they are also affected by
structures at a larger horizontal distance from the survey line. This effect is most pronounced
when the survey line is placed near a steep contact with the line parallel to the contact. Figure
67 shows the apparent resistivity pseudosection for the Wenner array over a 3-D model that
has two structures of 100 m that have the same widths within a 10 .m medium. One
structure is truly 2-D while the second is truncated near the middle of the area at the y-axis
coordinate of 3 meters. At a large distance from the edge of the second structure, for example
at the y-axis coordinate of 8 meters, the anomaly due to the right structure is almost as high as
that due to the left structure. However, close to the edge, for example at y-axis coordinate of 5
meters, the anomaly due to the right structure is significantly weaker. If a survey was carried
out along this line, and the subsurface was assumed to be 2-D, then the dimensions of the right
structure obtained from an inversion of the data set would be wrong. Dahlin and Loke (1997)
did a comparison of the sensitivity of different arrays to structures off the axis of a 2-D survey
line. In general, it was found that the dipole-dipole array was the most sensitive (i.e. suffered
the greatest distortion) due to off-axis structures. The best way to handle 3-D structures would
be a full 3-D survey and data inversion. We will look at 3-D surveys in the later part of these
notes.
It must be emphasized that a 2-D survey with very good quality data and very dense
data coverage, and inverted with a good inversion algorithm, can still give the wrong results if
the assumption of a 2-D geology on which the model is based is seriously wrong. This is a
particularly a problem in mineral surveys (which commonly also involve I.P. measurements)
where very complex geological structures and mineralization patterns are usually encountered.
In such situations, the results from a 2-D resistivity and I.P. model should be treated with some
reservations unless it is confirmed by a 3-D survey and model. There have been many cases
where expensive drill-holes have passed through barren zones in areas where the 2-D model
shows a strong IP anomaly.
(g). Limits of the physics of the resistivity method. While 2-D and 3-D surveys and data
inversion has greatly extended the range of field problems that can be solved using the
resistivity method, there are still basic laws of physics that place certain limitations on these
techniques. The resolution of the resistivity method decreases exponentially with depth. We
see the subsurface “as through a glass darkly” and the image becomes increasingly fuzzier with
depth. It is unlikely to be able to map a structure with a size of 1 meter at a depth of 10 meters
using the resistivity method. The resistivity phenomenon is based on the diffusion equations,
so its resolution is inherently poorer than the seismic or ground radar method at depths greater
than one wavelength.
(h). Non-uniqueness. It is well known that more than one model can produce the same
response that agrees with the observed data within the limits of the data accuracy. In 1-D
resistivity sounding modeling, the problems of equivalence and suppression are well known.
The problems, in different forms, also occur in 2-D and 3-D modeling. A good example was
shown in the paper by Oldenburg and Li (1999). In 2-D and 3-D modeling, constrains are used
so that a stable solution can be obtained. The use of a smooth or blocky constraint results in the
production of models that look more reasonable, but it is no guarantee that they are indeed
correct. The accuracy of the result is only as good as the accuracy of the assumptions made.
The resulting model thus depends to a significant extent on the constraint used, and will closely
approximate the true subsurface resistivity only if the constrains correspond to the real
situation.
(i). Optimization versus inversion. The RES2DINV program, like most non-linear inversion

91
programs, actually carries out an optimization (i.e. not a direct one-to-one inversion in the sense
it must have only one solution) in that it tries to reduce the difference between the calculated
and measured apparent resistivity values. If there is infinite data and a perfect fit between the
calculated and measured values, how the data is measured should not have an effect on the
results. However with real, noisy and limited data, how the data is measured does have an
effect. A model with 5% RMS error in the fit between the measured and calculated apparent
resistivity values with one data set might not give the same model as a 5% RMS error with
another data set although both might be from the same place. For this reason, the dipole-dipole
array gives an inversion model with much better resolution than the pole-pole array although
in theory the dipole-dipole values can be extracted from the pole-pole values.
Figure 67. An example of 3-D effects on a 2-D survey. (a) Apparent resistivity pseudosections
(Wenner array) along lines at different y-locations over (b) a 3-D structure shown in the form
of horizontal slices.

92
Figure 68. The 2-D sensitivity sections for the pole-dipole array with a dipole length of 1 meter
and with (a) n=6, (b) n=12 and (c) n=18. Note that as the ‘n’ factor increases, the zone of high
positive sensitive values becomes increasingly concentrated in a shallower zone below the P1-
P2 dipole.
(j). Increasing the electrode separation does not always increase the survey depth. It is
generally assumed that as the separation between the electrodes is increased, the region of the
subsurface that is ‘sensed’ by the array also increases. While this is true of most arrays, there
are certain important exceptions. In particular, this is not true of the pole-dipole array under
certain circumstances. In some surveys with the pole-dipole array, the separation between the
C1 current electrode and the P1-P2 dipole is increased in an effort to increase the depth of
survey by the array. However, if this is done with the P1-P2 dipole length (the ‘a’ factor in

93
Figure 4) kept at a constant spacing, certain interesting effects come into play. In section 2.5.8
this practice was strongly discouraged on the basis that the potential will decrease with the
square of the ‘n’ factor. This problem can be overcome by a combination of using higher
currents and more sensitive receivers. However, the problem caused by the change in the array
sensitivity pattern as the ‘n’ factor is monotonically increased is usually not taken into account.
The change in the sensitivity pattern when the ‘n’ factor changes from 1 to 6 was shown earlier
in Figure 30. Figure 68 shows what happens when the ‘n’ factor jumps from 6 to 12 to 18.
Here, the dipole length is kept constant at 1 meter. When ‘n’ is equal to 6 there are reasonably
high sensitivity values to a depth of about 3 to 4 meters between the C1 current and the P1
potential electrode. When ‘n’ is increased to 12, the zone of high sensitivity values becomes
increasingly more concentrated below the P1-P2 dipole in an even shallower region. This
means that the array with ‘n’ equals to 12 is in fact less sensitive to deeper structures than the
array with ‘n’ equals to 6. This effect is even more pronounced when ‘n’ is increased to 18.
Thus increasing the separation between the current electrode and the potential dipole,
while keeping the dipole length fixed, does not increase the survey depth of the array. It, in
fact, effectively decreases the depth of the region sensed by the array!
Figure 69 shows the apparent resistivity anomaly due to a small near-surface high
resistivity block for the pole-dipole array for ‘n’ values of up to 28. Note that the amplitude of
the high resistivity anomaly due to the near-surface block increases with the ‘n’ value, i.e. the
array becomes increasingly more sensitive to the near-surface block as the separation between
the electrodes increase. In field surveys with the pole-dipole array where the ‘n’ factor is
monotonically increased in the belief that this increases the survey depth, the pseudosection is
frequently dominated by a series of parallel slanting high-amplitude anomalies due to near-
surface inhomogeneities. The anomalies due to the near-surface structures frequently mask the
anomalies due to deeper structures that are of interest.
Figure 69. Example of apparent resistivity pseudosection with pole-dipole array with large ‘n’
values. Note that the anomaly due to a small near-surface high resistivity block becomes greater
as the ‘n’ factor increases. This means that the sensitivity of the array to the near-surface region
between the P1-P2 potential dipole becomes greater as the ‘n’ factor increases.

94
To increase the depth of penetration with the pole-dipole array, the ‘a’ dipole length
should be increased when the ‘n’ factor exceeds 6 to 8. This method was discussed in detail in
section 2.5.9. For example, instead of fixing the ‘a’ dipole length to 1 meter and increasing the
‘n’ factor to 28 in Figure 69, a more prudent approach is increase the ‘a’ spacing from 1 to 4
meters while ensuring the ‘n’ factor does not exceed 8.
4.11 The pole-pole inversion paradox

In theory, if we were to take all the possible pole-pole measurements, we can
reconstruct the measurements for any array such as the dipole-dipole array. The maximum
number of possible measurements (complete data set) for a survey line with N electrodes for
the pole-pole array is only N(N-1)/2, compared to N(N-1)(N-2)(N-3)/8 for four-electrode arrays.
Since the measurements for any four-electrodes arrays can be reconstructed from the pole-pole
measurements, the pole-pole complete data set should have all information that can be obtained
from a resistivity survey. However, we have seen that the data for some arrays such as the
dipole-dipole has much higher resolution compared to the pole-pole array, which seems
inconsistent if the complete pole-pole array data set has the same information. This is an
interesting paradox that has come up occasionally, for example there is a mention of it in Blome
et al. (2011) concerning the reconstruction of 4-electrodes arrays data from pole-dipole
measurements.
My own personal opinion is that when we carry out an 'inversion', it is not a direct
inversion where we go from the data to a unique in one step but rather a non-linear optimization
is carried out where we try to find the minimum of an objective function by using the slope or
gradient of the function. A pole-pole measurement is relatively insensitive to changes in the
resistivity of a model cell, i.e. the Jacobian matrix values have smaller amplitudes compared to
a 4-electrodes array such as the dipole-dipole where the apparent resistivity changes by a larger
amount for the same change in the cell resistivity. The objective function for the pole-pole data
set has a relatively broad minimum with gentle slopes, and within the limits of the measurement
error (eg. 1%) it cannot distinguish between the different models in the vicinity of the global
minimum. The inversion method tries to minimize the model roughness as well, so it ends up
with the model with the minimum structure possibly in a local minimum. The objective
function for a dipole-dipole data set has steeper slopes, so for the same data misfit it ends up
closer to the global minimum. The situation is shown diagrammatically in Figure 70.
A similar principle applies in cases where four-electrode arrays measurements are
reconstructed from pole-bipole measurements (Blome et al., 2011).

95
Figure 70. Diagrammatic illustration of differences in objective function shapes for the pole-
pole array and dipole-dipole array data sets leading to different models obtained from
optimization routine.

96
5 I.P. inversion
5.1 Introduction
One of the more recent developments in the instrumentation for electrical imaging
surveys has been the addition of Induced Polarization (I.P.) capability in the multi-electrode
resistivity meter system. Many of the early 2-D surveys were resistivity and I.P. surveys carried
out using conventional 4-electrode systems in the 1950's onwards for mineral exploration,
particularly for conductive sulfide ore bodies. Quantitative interpretation of such historical data
was rather limited due to the limited computing facilities available at that time. Such historical
data provides an interesting source of data for testing modern 2-D and 3-D inversion software.
Re-interpretation of such old data to produce quantitative models sometimes has shed new light
on the geological structures.
One of the distinctive characteristics of the I.P. method has been the different
parameters in the time and frequency domains used to represent the I.P. effect. The following
section briefly discusses the I.P. phenomena and the different I.P. parameters. This is followed
by a few exercises in the inversion of IP data with the RES2DINV program. The data format
used by the RES2DINV and RES3DINV programs is described in their respective manuals. As
such, we will not cover it here.
5.2 The IP effect

A very brief description of the IP effect will be given here. Further details can be found
in many fine textbooks, such as by Keller and Frischknecht (1966), Summer (1976), Telford et
al. (1990) and Zhdanov and Keller (1994).
The I.P. effect is caused by two main mechanisms, the membrane polarization and the
electrode polarization effects. The membrane polarization effect is largely caused by clay
minerals present in the rock or sediment. This is particularly relevant in engineering and
environmental surveys. The chargeability of clays are in the 10 to 50 mV/V range (Sumi, 1965)
is much smaller than that due to conductive minerals. Interestingly, negative I.P. effects have
been reported for certain types of clays (Brandes and Acworth, 2003). Another application of
I.P. in environmental surveys in detecting decomposing organic matter (Weller et al., 2000).
The electrode polarization effect is caused by conductive minerals in rocks such that the current
flow is partly electrolytic (through groundwater) and partly electronic (through the conductive
mineral). This effect is of particular interest in surveys for metallic minerals, such as
disseminated sulfides.
I.P. measurements are made in the time-domain or frequency domain. In the time-
domain, the I.P. effect is measured by the residual decay voltage after the current is switched
off (Figure 72c). The time domain I.P. unit (chargeability) is usually given in millivolt per volt
(mV/V) or in milliseconds. Figure 71 shows the I.P. values (in terms of mV/V) for several
mineralized rocks and common rocks. Note that the I.P. effect due to sulfide mineralization
(the electrode polarization effect with values in the 100 to 200 mV/V range) is much larger
than that due to clay minerals (membrane polarization) in sandstone and siltstones. The high
I.P. effect for disseminated metallic sulfides makes I.P. surveys a standard tool for exploration
of such minerals. The resistivity contrast for such deposits is frequently low due to its
disseminated nature which makes detection by standard resistivity and EM surveys difficult.
In the frequency domain (Figure 72b), the I.P. effect is measured by the change in
apparent resistivity value from low to high frequencies (typically 1 to 10 Hz) where the unit
used is the percent frequency effect. Another measure of the I.P. effect in the frequency domain
is the phase shift between the potential signal and the input current, where the unit used is in
milliradians.

97
Figure 71. The I.P. values for some rocks and minerals.
Figure 72. The Cole-Cole model. (a) Simplified electrical analogue circuit model (after Pelton
et al. 1978).  = resistivity, m = chargeability,  = time constant, c = relaxation constant. (b)
Amplitude and phase response to sine wave excitation (frequency domain). (c) Transient
response to square wave current pulse (time domain). Most I.P. receivers measure the integral
of the decay voltage signal over a fixed interval, mt, as a measure of the I.P. effect.

98
One mathematical model that attempts to explain variation of resistivity with frequency
observed in the IP method is the Cole-Cole mode (Pelton et al., 1978), which is defined by
  1 
 ( )   0 1  m1  
c  (5.1)
  1  i  
where  0 is the DC resistivity, m is the chargeability,  is the angular frequency (2f),  is a
time constant and c is the exponent or relaxation constant. While the DC resistivity and
chargeability determine the behavior of the material at very low and very high frequencies, the
variation of the amplitude and phase curves at intermediate frequencies are also affected by the
time and relaxation constants.
The time constant factor  has a large range, from 0.01 second to several thousand
seconds. The relaxation constant factor c is bounded by 0.0 to 1.0, with values frequently
between 0.2 and 0.7. Much of the earlier work was on the use of spectral I.P. (SIP)
measurements to differentiate between different types of conductive minerals for mining
purposes (Van Voorhis et al.,1973; Zonge and Wynn, 1975; Pelton et al., 1978; Vanhalla and
Peltoniemi, 1992). More recently, attempts have been made to use the SIP method for
environmental surveys, such as in the detection contaminants (Vanhala et al., 1992). Figure 72
shows a simplified electrical analogue circuit for the Cole-Cole model, together with typical
response curves in the frequency and time domain.
5.3 IP data types

Although the chargeability is defined as the ratio of amplitude of the residual voltage
to the DC potential (Figure 72c) immediately after the current cut-off, this is not used in actual
field measurements. Field measurements of the I.P. effect may be divided into two main groups,
the time-domain and frequency domain methods.
(a) Time domain IP measurements
In the time-domain method, the residual voltage after the current cut-off is measured.
Some instruments measure the amplitude of the residual voltage at discrete time intervals after
the current cut-off. A common method is to integrate the voltage electronically for a standard
time interval. In the Newmont M(331) standard (Van Voorhis et al., 1973), the chargeability, mt,
is defined as
1.1
mt  1870
 0.15
,
Vs dt
(5.2)
VDC
where the integration is carried out from 0.15 to 1.1 seconds after the current cut-off. The
chargeability value is given in milliseconds. The chargeability value obtained by this method
is calibrated (Summer 1976) so that the chargeability value in msec. has the same numerical
value as the chargeability given in mV/V. In theory, the chargeability in mV/V has a maximum
possible value of 1000.
I.P. surveys have traditionally been used in the mineral exploration industry,
particularly for metal sulfides, where heavy electrical generators producing high currents of the
order of 10 Amperes are used. The apparent I.P. values from such surveys are usually less than
100 msec. (or mV/V). One recent development is the addition of I.P. capability to battery based
systems used in engineering and environmental surveys where currents of 1 Ampere or less are
normally used. An accompanying phenomenon is the observation of I.P. values of over 1000
msec. (or less than -1000 msec.) in some data sets. Such values are almost certainly caused by
noise due to a very weak IP signal. To check whether such high I.P. values are real, first check
the apparent resistivity pseudosection. If it shows unusually high and low values that vary in
an erratic manner, the data is noisy. If the apparent resistivity values are noisy, then the apparent

99
I.P. values are almost certainly unreliable. Next check the apparent I.P. pseudosection. If the
apparent I.P. values show an erratic pattern (frequently with anomalous values lined up
diagonally with an apex at a doubtful electrode), then the I.P. values are too noisy to be
interpretable. There has been some recent work on improving the reliability of I.P.
measurements made with the multi-electrode type of systems (Dahlin and Leroux, 2012).
(b) Frequency domain I.P. measurements
In the frequency-domain methods, the apparent resistivity is measured at least two
frequencies, for example at 1 Hz. and 10 Hz. The higher frequency is usually set at 10 times
time the lower frequency. One commonly used frequency domain IP unit is the percent
frequency effect, PFE, which is defined by
  L     H 
PFE  .100 (5.3)
  H 
Another closely related unit that is also commonly used is the metal factor MF which is defined
by
  L     H 
MF  .2 .10 5 (5.4)
  L .  H 
Another common frequency domain IP unit is the phase angle, . It is the phase shift between
the transmitter current and the measured voltage, and the unit commonly used is milliradians.
This has the advantage that only measurement at a single frequency is need, but the current
circuit must be coupled with the potential measuring circuit in some way so that the phase shift
between the measured potential and the input current can be determined. This is not a problem
in the multi-electrode type of system, such as the Pasi Polares system, where both circuits are
in the same unit.
(c) Relationship between the time and frequency domain IP units
From the measurement of the amplitude and phase spectrum of porphyry copper
mineralization, Van Voorhis et al. (1973) proposed the following equation to describe the
observed spectra.
    K  j b (5.5)
K is constant and b is a measure of the IP effect. It is a positive number of more than 0 and less
than 0.1. This is also known as the constant phase model (Weller et al., 1996). By using the
above model, the following relationship between the different IP units and the b parameter
were derived.
PFE  10b  1.100
 = 1571 b (5.6)
mt  1320b
These relationships provide a numerical link between the different IP units (Van Voorhis et al.,
1973; Nelson and Van Voorhis, 1973). An alternative relationship is given in the paper by
Kemna et al. (1997).
There have been a large number of 2-D I.P. surveys published over the years, so there
is certainly no lack of data to test the IP inversion with the RES2DINV program. Most of the
newer multi-electrode systems now come with an IP option. However, the data from many of
these systems is rather noisy due to the limited current that can be injected into the ground. For
reasonable IP data quality, a current of at least 0.5 Amperes is probably necessary. For
environmental and engineering surveys, the most useful application of the IP data is probably
in differentiating between sand and clay sediments.

100
5.4 IP mathematical models

There are generally two methods to calculate I.P. effects in a forward modeling
program, the perturbation and the complex resistivity methods. The first approach considers
the I.P. model as a small perturbation of the base resistivity model. Consider a base model that
has a conductivity DC that is measured by a normal resistivity survey. The effect of the
chargeability m is to decrease the effective conductivity to IP = (1 - m) DC. The apparent I.P.
is then calculated by two forward models using the original and perturbed conductivities, such
as
ma = [ (IP ) -  (DC ) ] /  (DC) (5.7)
Where  is the calculated potential. This approach works well in most cases. The inversion of
I.P. data is a two step process (Oldenburg and Li, 1994).
(1). A resistivity inversion step using the apparent resistivity data alone is first carried out.
This produces a resistivity model ( DC) that is independent of the measured IP data.
(2). Using the using the resistivity model as a base, the apparent I.P. values are calculated
using equation 5.7. An inversion of the apparent I.P. data is then carried out to obtain a
chargeability model (mIP). The resistivity model ( DC) is kept fixed during this inversion step.
There are two main problems with the perturbation approach. The apparent
chargeability is calculated from the difference of two DC potentials. The difference is usually
less than 1% of the DC potential values, so this tends to magnify numerical errors in the finite-
difference or finite-element methods used to calculate the DC potentials. For most traditional
arrays, this is usually not a serious problem. However, the increasing popularity of the offset
pole-dipole and dipole-dipole arrays (White et al., 2001) can produce array configurations with
extreme geometrical factors. In some situations, with rugged topography and large resistivity
contrasts, this can sometimes lead to negative apparent resistivity values (Jung et al., 2009). In
such situations, it is found that the apparent chargeability values calculated using equation (5.7)
is not reliable, particularly for 3-D models. The second problem is that it does not take into
consideration the apparent chargeability values when calculating the resistivity model,
although for a non-homogeneous model the resistivity values have a significant effect on the
calculated apparent chargeability values. The resistivity model is optimal for the apparent
resistivity data but it might not be optimal for the apparent chargeability data. This can lead to
some distortions in the I.P. inversion model as it is also based on the resistivity model.
The second approach for I.P. model calculation is to treat the conductivity as a complex
quantity with real and imaginary components (Kenma et al., 2000). The complex conductivity
is given by
 = DC – i mDC (5.8)
The DC conductivity DC forms real part, while mDC forms the imaginary part. A complex
potential is then calculated for this complex model.
 = r + i i (5.9)
The complex potential has two components, r and i. The apparent chargeability is calculated
using the ratio of the imaginary component to the real component, ma = i /r. The inversion
process using a complex resistivity approach is as follows.
(1). A joint inversion where both the apparent resistivity and apparent I.P. data (a, ma) are
used together to generate a resistivity and chargeability model ( DC, mIP) is carried out.
(2). The first inversion tends to reduce the apparent resistivity data misfit at the expense of
the apparent I.P. misfit, i.e. it tends to produce a more accurate resistivity model at the expense
of the I.P. model. A second inversion is commonly carried out to refine the I.P. model. The
resistivity model ( DC) is kept fixed, and only the I.P. model (mIP) is changed to further reduce
the apparent IP (ma) data misfit.

101
Figure 73 below shows the inversion results from both I.P. inversion methods for the
data from the Magusi River ore body (Edwards, 1977) survey. The models obtained by both
methods are generally similar. The ore body shows up as a low resistivity body with high IP
values near the middle of the survey line in the model sections. The blocky inversion method
was used to sharpen the boundary between the ore body and the surrounding country rocks in
the model. The massive sulfide ore body shows as a prominent low resistivity and high I.P.
structure. The complex resistivity method gives a model with slightly higher I.P. values.
Figure 73. Magusi River massive sulphide ore body inversion models. (a) Apparent resistivity
pseudosection. Resistivity inversion models obtained using the (b) perturbation and (c)
complex resistivity methods. (d) Apparent I.P. (metal factor) pseudosection. I.P. models
obtained using the (e) perturbation and (f) complex resistivity methods.
5.5 I.P. surveys with multi-electrode systems

Many multi-electrode systems now offer an I.P. measurement option. The maximum
current from battery-based systems is usually 1 Amp or less. This is usually too low to give
reliable I.P. data when the electrode spacing is more than a few meters. However there have
been recent improvements in the electronic circuitry and field survey procedure (Dahlin and
Leroux, 2012.) to improve the data quality. One method that can be used with conventional
multi-electrode systems that has two separate cables is by using different cables for the current
and potential electrodes (Figure 74). This reduces the EM coupling between the current and
potential cables. The possible current and potential electrodes positions are reduced and special
control files are needed for this configuration. However, this method can be used with any
multi-electrode system that uses a two cable arrangement. The two cables are separated at a
distance that is twice the electrode to cable jumper distance, which is about as far as apart as
possible. The results from a multi gradient array survey are shown in Figure 75. A
demonstration of the data acquisition method with separated parallel cable spreads for current
transmission and potential measurement (Dahlin and Leroux, 2012) was made at the 3rd
International Workshop on Induced Polarization held on Ile d’Oleron on 7-9 April 2014. The
site is geologically characterized by sedimentary deposits including marine clay, locally known
as “bri”, and river alluvium (Gouet, 2007). The resistivity section (Figure 75a) shows an upper
3 to 4 meters sandy layer underlain by lower resistivity saline mud sediments (Dahlin and Loke,
2015). The I.P. section (Figure 75b) shows a top 1 to 2 meters layer with chargeability values
of 4 to 8 mV/V which is probably sandy sediments with some organic content. The low I.P.
values below this layer is probably due to high salinity that tends to reduce the I.P. effect
(Weller et al., 2011). Note the apparent resistivity pseudosection shows fairly regular contour
patterns, whereas the bottom part of the I.P. pseudosection has noisier data due to larger
electrode spacings and weaker signals.

102
Figure 74. Sketch of separated cable spreads setup used (after Dahlin and Loke, 2015).
Figure 75. Resistivity and chargeability pseudosection from field demo at 3rd IP workshop at
Ile d’Oleron (after Dahlin and Loke, 2015).
Table 10. Tests with 2-D I.P. inversion

IPMODEL.DAT – A time- (1). After reading the data file, run the inversion with the
domain synthetic data set. default inversion options.
(2). Next try with the robust inversion option.
(3). The program also has an option to carry out the IP
inversion sequentially after the resistivity inversion (under
the “Type of IP inversion method” option in the
“Inversion” menu). Try this and see if there are any
differences in the results.
IPSHAN_PFE.DAT – A field (1). After reading the data file, carry out the inversion with
data set from Burma (Edwards, the default inversion options. Try again using the option to
1977) with measurements in ‘Limit the range of model resistivity’ option.
PFE. (2). Try reducing the unit electrode spacing in the data file
with a text editor to half the given value. Does it improve
the inversion results?
IPMAGUSI_MF.DAT – A (1). After reading the data file, carry out the inversion with
field data set from Canada the default inversion options. Try again using the option to
(Edwards, 1977) with ‘Limit the range of model resistivity’ option.
measurements in metal factor (2). Run the inversion again with the ‘Select robust
values. inversion’ option. This should make a significant
difference in the results.
(3). Try reducing the unit electrode spacing in the data file
with a text editor to half the given value. Does it improve
the inversion results?
IPKENN_PA.DAT – A field (1). After reading the data file, carry out the inversion with
data set with the IP values in the default inversion options.
phase angles (Hallof, 1990).

103
6 Cross-borehole imaging
6.1 Introduction
One of the most severe limitations of 2-D imaging surveys carried out along the ground
surface is the reduction in the resolution with depth. This is a fundamental physical limitation
that no amount of reconfiguration of the surface arrays or computer modeling can overcome.
In theory, the only way to improve the resolution at depth is to place the sensors (i.e. the
electrodes) closer to the structures of interest. This is not always possible, but when such
boreholes are present, cross-borehole surveys can give more accurate results than is possible
with surface surveys alone.
That has been many fine publications on such surveys; such as by Zhao et al. (1986),
Daily and Owen (1991), Sasaki (1992), LaBrecque et al. (1996), Slater et al. (1997, 2000),
Zhou and Greenhalgh (1997, 2000) and Wilkinson et al. (2006a, 2008). In the following
section, I will attempt to summarize the main results with regards to the choice of array
configurations for cross-borehole surveys. However, please refer to the original papers for the
details.
6.2 Electrode configurations for cross-borehole surveys

In theory, any array that is used for normal surface surveys can be adapted for cross-
borehole surveys. We can have arrays with two electrodes, three electrodes and four electrodes.
6.2.1 Two electrodes array – the pole-pole
There are two possible configurations, both electrodes can be in the same borehole
(Figure 76a), or the electrodes can be in different boreholes (Figure 76b). Figure 76c considers
a third possibility that is sometimes not considered, with one electrode on the surface. In all
cases, the areas of highest sensitivities are concentrated near the electrodes, particularly if the
two electrodes are far apart in different boreholes as in Figure 76c. Note that the electrodes do
not actually scan the area between them, as would be expected for a seismic survey with the
source and receiver in different boreholes. Note that the region between the two electrodes
generally has negative sensitivity values.
If there are n electrodes (including surface electrodes, if any), there are a total of n(n-
1)/2 possible independent measurements. Most authors recommend measuring all the possible
readings at the other electrodes for a current electrode, i.e. measure all the possible
combinations. One problem with this array is the physical location of the two remote electrodes,
C2 and P2. They must be sufficiently far so that the pole-pole approximation is sufficiently
accurate. This means they must be located at a distance of at least 20 times the maximum
separation used by the active C1 and P1 electrodes in the boreholes. The large distance between
the P1 and P2 electrodes leads to the problem of contamination by telluric noise. These
problems are exactly the same as that faced with the pole-pole array for normal surface arrays.
While many earlier researchers have used this array for cross-borehole surveys (Dailey
and Owen, 1991; Shima, 1992; Spies and Ellis, 1995), more recent work tends to be less
enthusiastic about it. Sasaki (1992) and Zhou and Greenhalgh (2000) found that this array has
a significantly poorer resolution than the bipole-bipole and pole-bipole arrays.
The subsurface pole-pole array gives a useful illustration of the change in the geometric
factor for subsurface arrays. We have seen that for a pole-pole array with both electrodes
located on the surface of a half-space, the geometric factor k is given by
k  2 a
where a is the spacing between the electrodes. For the case where both electrodes are within
an infinite medium, the geometric factor is given by
k  4 a
where r is the distance between the current and potential electrodes. For the case where both

104
electrodes are below the surface of a half-space, the geometric factor is given by
 rr ' 
k  4  
 r  r' 
where r’ is the distance of the reflected image of the current (Figure 77) from the potential
electrode.
Figure 76. The possible arrangements of the electrodes for the pole-pole array in the cross-
borehole survey and the 2-D sensitivity sections. The locations of the two boreholes are shown
by the vertical black lines.

105
Figure 77. A schematic diagram of two electrodes below the surface. The potential measured
at P can be considered as the sum of the contribution from the current source C and its image
C’ above the ground surface.
6.2.2 Three electrodes array – the pole-bipole

According to Zhou and Greenhalgh (1997), there are six possible independent
configurations for the pole-bipole type of array (ignoring the surface electrodes). Of these six
configurations there are two basic combinations, with the current electrode and one potential
electrode in one borehole and the other potential electrode in the second borehole (Figure 78a),
and with the current electrode in one borehole and both potential electrodes in the other
borehole (Figure 78b). Zhou and Greenhalgh (1997) recommend the configuration with one
current and potential electrode in the same borehole, and the second potential electrode in the
other borehole (the C1P1-P2 or AM-N configuration). In Figure 78a, it can be seen that his
configuration has high positive sensitivity values in between the two boreholes. This means
that it provides significant information about the resistivity of the material between the two
boreholes. The zones with the negative sensitivity values are confined to between the C1 and
P1 electrodes in the first borehole and to the left of the P2 electrode in the second borehole.
The second configuration, with the current electrode in one borehole and both potential
electrodes in the other borehole, has large negative sensitivity values between the C1 and P1
electrode and large positive values between the C1 and P2 electrodes. In between these two
bands, there is a zone with small sensitivity values, i.e. the array does not give significant
information about the resistivity in this zone. Another possible disadvantage of this array is
that for some positions of the P1-P2 bipole, the potential value measured is very small or zero.
This causes the signal to noise ratio to be small.
Figure 78c shows the sensitivity pattern when all three electrodes are in the same
borehole. There are very high negative and positive sensitivity values in the vicinity of the
borehole. Sugimoto (1999) recommends that such measurements also be made as they give
valuable information about the dip of structures between the boreholes. Figure 78d shows the
sensitivity pattern when the current electrode is on the surface. The sensitivity values between
the C1 and P1 electrodes are relatively small (probably due to the large distance between these
two electrodes), while the sensitivity values between the P1 and P2 electrodes have moderate

106
positive values.
Overall, many authors have positive remarks about this array. It provides better
resolution and is less sensitive to telluric noise (since the two potential electrodes are kept
within the survey area) compared to the pole-pole array. While in theory the resolution of the
array is slightly poorer than the bipole-bipole array, the potential values measured are
significantly higher.
Figure 78. The 2-D sensitivity patterns for various arrangements with the pole-bipole array.
The arrangement with (a) C1 and P1 in first borehole and P2 in second borehole, (b) C1 in the
first borehole and both P1 and P2 in the second borehole, (c) all three electrodes in the first
borehole and (d) the current electrode on the ground surface.

107
6.2.3 Four electrodes array – the bipole-bipole

Zhou and Greenhalgh (1997) list four possible independent configurations for this
array, of which there two basic combinations. In the first arrangement, the positive current and
potential electrodes C1 and P1 are located in one borehole, while the negative current and
potential electrodes C2 and P2 are located in the second borehole (Figure 79a).
Figure 79. The 2-D sensitivity patterns for various arrangements of the bipole-bipole array. (a)
C1 and P1 are in the first borehole, and C2 and P2 are in second borehole. (b) C1 and C2 are
in the first borehole, and P1 and P2 are in second borehole. In both cases, the distance between
the electrodes in the same borehole is equal to the separation between the boreholes. The
arrangements in (c) and (d) are similar to (a) and (b) except that the distance between the
electrodes in the same borehole is half the spacing between the boreholes.

108
In this arrangement, the C1P1-C2P2 configuration (i.e. the AM-BN arrangement of

Zhou and Greenhalgh (1997), there are large positive sensitivity values in the area between the
two boreholes. This is a desirable property for a cross-borehole array since the intention is to
map the material between the two boreholes. The large negative sensitivity values are confined
to the region along the boreholes between the C1 and P1 (and C2 and P2 electrodes). Figure
79c shows a similar arrangement, but with the spacing between the electrodes in the same
borehole reduced to half the spacing between the boreholes. Note again the large positive
sensitivity values between the two boreholes.
In the second basic configuration (Figure 79b), the C1C2 current bipole is located in
one borehole while the P1P2 potential bipole is located in the other borehole. There is also a
large region with positive sensitivity values between the two boreholes. However it is flanked
by two zones with large negative sensitivity values. Thus the response of this C1C2-P1P2
arrangement to inhomogeneities between the boreholes is more complicated than the first
arrangement. When the bipole length is reduced, the positive region is significantly reduced.
These features make this arrangement less desirable for cross-borehole surveys. Another
disadvantage is that the potential signal strength is weaker in the C1C2-P1P2 arrangement
compared to C1P1-C2P2 configuration.
Overall, Zhou and Greenhalgh (1997, 2000) recommended the C1P1-C2P2
configuration. Sasaki (1992) has found the bipole-bipole array to have better resolution
compared to the pole-pole and pole-dipole arrays. Figure 80 shows one possible field survey
measurement sequence. A series of measurements is first made with a short spacing (for
example half the distance between the boreholes) starting from the top. Next, measurements
are repeated with a larger spacing between the electrodes. Due to the symmetry in the
arrangements shown in Figure 80a,b measurements should also be made with other less
symmetrical arrangements. Figure 80c,d show two other possible measurement sequences.
Note that the length of the boreholes must be comparable to the distance between the
boreholes. Otherwise, if the spacing between the electrodes in the same borehole is much
smaller than the distance between the boreholes, the readings are likely to be more influenced
by the materials in the immediate vicinity of the boreholes, rather than the material in between
the boreholes. In such a situation, the only alternative is probably to include the surface-surface
and surface-borehole measurements.
Some more recent research in cross-borehole imaging may be found in the papers by
Oldenborger et al. (2005), Chambers et al (2007), Wilkinson et al (2008) and Nimmer et al
(2008).

109
Table 11. A few borehole inversion tests.

BOREHOLE.DAT – A (1). Read in the file, and try an inversion with the default
synthetic data set with the pole- inversion parameters.
bipole array. (2). Next you can try with the robust inversion option.
BORELUND.DAT – A pole- (1). Read in the file, and try an inversion with the default
pole array field data set from inversion parameters.
Lund University, Sweden. The (2). Next you can try different settings, such as the robust
survey was conducted to map inversion option.
fractures in a limestone-marl (3). The program also has an option to reduce the size of
formation (Dahlin pers. the model cells by half, which you might like to try.
comm.).
BORELANC.DAT – A bipole- (1). Read in the file, and run the inversion. The path of the
bipole array field data set from saline tracer is represented by regions with low resistivity
Lancaster University, U.K values. Can you identify the tracer in the model sections?
(Slater et al. 1997). The survey
was conducted to map the flow
of a saline tracer from the
ground surface through the
unsaturated zone.
Figure 80. Possible measurement sequences using the bipole-bipole array. Other possible
measurements sequences are described in the paper by Zhou and Greenhalgh (1997).

110
6.3 Single borehole surveys

When only a single borehole is available, only the borehole to surface measurements
are possible. Figure 81a and Figure 81b shows two possible arrangements with two electrodes
in the borehole and two electrodes on the surface. The first arrangement (Figure 81a) has the
C1-C2 current bipole in the borehole while the P1-P2 bipole is on the surface. This arrangement
has large positive sensitivity values between the C1-C2 bipole, and also between the P1-P2
bipole. The region of the subsurface between the two bipoles have moderate positive sensitivity
values, while an approximately parallel zone stretching from the C2 current electrode (which
is higher in the borehole) to the left of the P2 potential electrode has moderate negative
sensitivity values. Figure 81b shows the alternative arrangement with the C1 and P1 electrodes
in the borehole and the C2 and P2 pair on the surface. There is a small region of large negative
sensitivity values along the borehole between the C1 and P1 electrodes. There is also a small
near-surface zone between the C2 and P2 electrodes with large negative sensitivity values.
However, note that most of the region between the two pairs of electrodes has relatively high
positive sensitivity values. Since the intention of the survey is to map the subsurface between
the two pairs of electrodes, this might be a better arrangement than the one given in Figure 81a.
In some situations, it might only be possible to have only one subsurface electrode, for
example at the end of a drill bit or penetrometer (Sorensen, 1994). Figure 81c and Figure 81d
shows the arrangements where the upper electrode that was formerly in the borehole (in Figure
81a and Figure 81b) is now placed on the surface near the borehole. The arrangement with the
C1 electrode in the borehole and the C2 electrode on the surface (Figure 81c) has a near-surface
zone of large negative sensitive values between the C2 electrode and the P2 electrode.
However, there are moderately high positive sensitivity values in the region between the C1
electrode in the borehole and the P1-P2 bipole on the surface. The alternative arrangement with
the C1 electrode in the borehole and the P1 electrode near the top of the borehole has a zone of
large negative sensitivity values along the borehole. The region between the borehole and the
C2-P2 electrodes has generally high positive sensitivity values, so this arrangement again might
be better.
Figure 82 shows a pole-bipole configuration with the C1 current electrode in the
borehole and the P1-P2 potential bipole on the surface. The sensitivity pattern is fairly similar
to that obtained in Figure 81a. There is a region with moderate positive sensitivity values
between the C1 electrode and the P1-P2 bipole, together with an approximately parallel zone
of moderate negative sensitivity values between the C1 electrode and the region to the left of
the P1 electrode. Overall, the arrangement shown in Figure 81d appears to be best due to the
large zone of high positive sensitivity values between the borehole and the surface electrode
pair. For all the possible arrangements, it is assumed that the normal surface-to-surface
measurements are also made to fill in the gaps not covered by the borehole to surface
measurements. The surface-to-surface measurements are particularly important to accurately
map the resistivity distribution in the near-surface zone.

111
Figure 81. Several possible bipole-bipole configurations with a single borehole. (a) The C1
and C2 electrodes at depths of 3 and 2 meters respectively below the 0 meter mark. (b) The C1
and P1 electrodes at depths of 3 and 2 meters respectively below the 0 meter mark. (c). The C1
electrode is at a depth of 3 meters below the 0 meter mark while the C2 electrode is on the
surface. (d). The C1 electrode is at a depth of 3 meters below the 0 meter mark while the P1
electrode is on the surface.
Figure 82. A pole-bipole survey with a single borehole. The C1 electrode is at a depth of 3
meters below the 0 meter mark.

112
6.4 Cross-borehole optimized arrays

Generating possible measurement sequences such as that described in section 6.2.3
involves some degree of subjective guesswork. A possibly more objective approach is to use
the automatic array optimization method described earlier in section 4.9.4, which is described
in detail in Loke et al. (2014b). As the details can be found in this paper, one of the results is
shown here to illustrate the improvements that can be obtained. The test model consists of
conductive and resistive blocks in a two-layer medium (Figure 83a). A block is also placed
beyond the right column of borehole electrodes to assess the resolution of the different arrays
for structures outside the borehole region. Three test arrays are used, (i) a ‘standard’ cross-
borehole measurement sequence with 1875 data points that was used by Wilkinson et. al.
(2006a) (ii) an optimized data set generated using the ‘Compare R’ method and (iii) a ‘reduced’
optimized data sets that excludes arrays with both current or both potential electrodes in the
same borehole to avoid ‘current channelling effects’ in cases where the resistivity of the fluid
filling the boreholes is much lower than the surrounding subsurface materials (Wilkinson et
al., 2006a). Measurements are made using 20 electrodes in each borehole together with 9
electrodes on the surface that are 1 meter apart. Gaussian random noise with a maximum
amplitude of 1.0 milliohm was added to the resistance values before they were converted to
apparent resistivity values. This resulted in average noise levels of 2.6%, 0.1% and 1.0% in the
final apparent resistivity values for the full optimized, 'standard' and reduced optimized data
sets. The higher noise levels of the optimized data sets are due to the higher average geometric
factor of the arrays used.
The inversion models (Figure 83) show that the boundary between the two layers is
well resolved by all the data sets. The background resistivity values, away from the embedded
rectangular blocks, in the models are also generally within a few percent of the true values of
100 and 30 ohm-m in the two layers. The topmost high resistivity block is well resolved by all
three data sets. The models with the full and reduced optimized data sets give maximum values
of 888 and 980 ohm-m respectively (that are closer to the true value of 1000 ohm-m), while
the 'standard' data set achieved a maximum value of only 653 ohm-m. The second (low
resistivity) block within the borehole region is also detected by the three data sets. The full
optimized data set model gives a minimum resistivity value of 50 ohm-m (true value 30 ohm-
m), while the reduced optimized and 'standard' data sets give values of 56 and 65 ohm-m
respectively. The third deepest block between the boreholes is well resolve by all the three data
sets, with lowest model values of 15, 16 and 22 ohm-m (true value 10 ohm-m) respectively for
the full optimized, reduced optimized and 'standard' data sets. The deepest block is resolved
only in the full optimized data set (Figure 83b). The high resistivity block to the right of the
borehole region is resolved by the full and reduced optimized data sets giving values of about
138 and 135 ohm-m (true value 1000 ohm-m) that are well above the background value of 100
ohm-m, while the 'standard' data set model fails to detect it. Overall, the optimized data sets
perform better than the standard measurement sequence. One advantage of using the array
optimization method that it is completely automatic taking just minutes on modern PCs.
6.5 Optimized arrays with subsurface electrodes

The cost involved in drilling conventional vertical boreholes and placement of the
subsurface electrodes is much higher compared to conventional surveys with electrodes only
on the ground surface. This in practice the number of cross-borehole field surveys is relatively
small compared to conventional surface only surveys. An inexpensive and innovative method
to overcome the high costs of drilling conventional boreholes is the MERIT (Multi-Electrode
Resistivity Implant Technique) system using a direct-push installation technique (Harro and
Kruse, 2013) where a matching set of subsurface electrodes is installed directly below a line of
surface electrodes (Figure 84). The cost to install the subsurface electrodes is about half that of

113
a conventional vertical borehole. This system is useful when it is necessary to image a region
of a long horizontal extent over a limited depth range. In a typical survey, the length of the line
is about 5 times the depth of the subsurface electrodes.
Figure 83. Test of optimized cross-borehole arrays with a synthetic model. (a) Two-layer test
model with conductive and resistive anomalies. Inversion models for (b) optimized data set
with all arrays, (c) 'standard' data set and (d) the reduced optimized data set that excludes arrays
with both current (or potential) electrodes in the same borehole. All the data sets have 1875
data points. The outlines of the rectangular blocks showing their true positions are also shown.
A method to generate optimized arrays for this type of survey configuration is described in
Loke et al. (2015b). An example of the improvement that can be obtained compared to
‘standard’ arrays generated manually using heuristic rules (Harro and Kruse, 2013) is shown
in Figure 85. The survey was carried out at the Geopark research site on University of South
Florida campus in west-central Florida, United States. The site is characterized by karstified
limestone bedrock overlain by about 5 meters of overburden soils consisting of granular sands
over more cohesive sandy clay and clay soil with clay content generally increasing with depth
(Loke et al., 2015b). Depths to contacts were available for standard penetration tests (SPTs),
cone penetration tests (CPTs) and GPR data (Stewart and Parker, 1992). A deep array of 14

114
electrodes was implanted at 7.62 m below ground surface with an electrode spacing of 4 m
with a matching set of 14 electrodes on the surface directly above the implanted electrodes.
The models for the optimized data sets (Figure 85b and c) shows better agreement with the
known geology from the geotechnical and GPR data, in terms of the lower boundary of the top
sandy layer and the positions of the cavities, compared to the ‘standard’ arrays model (Figure
85a).
Figure 84. Schematic diagram of the MERIT method with the electrodes are planted along the
surface and directly below using the direct push technology (after Harro and Kruse, 2013).

115
Figure 85. Inversion models for the different data sets for the data collected with electrodes at
surface and 7.62 m depth with 4 m horizontal spacing. Models for the (a) standard arrays (405
data points), optimized arrays with (b) 403 and (c) 514 data points.

116
7 2-D field examples
7.1 Introduction
Here we will look at a number of examples from various parts of the world to give
you an idea of the range of practical survey problems in which the electrical imaging method
has been successfully used.
7.2 Landslide - Cangkat Jering, Malaysia

A common problem faced in Malaysia is landslides on hill slopes. The landslides are often
triggered by water accumulation within part of the slope that leads to weakening of a section
of the slope. Figure 86 shows the results from a survey conducted on the upper part of a slope
where a landslide had occurred in the lower section. Weathering of the granite bedrock
produces a clayey sandy soil mixed with core boulders and other partially weathered material.
The image obtained from this survey shows a prominent low resistivity zone below the center
of the survey line. This is probably caused by water accumulation in this region that reduces
the resistivity to less than 600 m. To stabilize the slope, it would be necessary to pump the
excess water from this zone. Thus, it is important to accurately map the zone of ground water
accumulation. This data set also shows an example with topography in the model section.
Figure 86. Landslide field example, Malaysia. (a) The apparent resistivity pseudosection for
a survey across a landslide in Cangkat Jering and (b) the interpretation model for the
subsurface.
7.3 Old Tar Works - U.K.

A common environmental problem in industrial countries is derelict industrial land.
Before such land can be rehabilitated, it is necessary to map old industrial materials (such as
metals, concrete blocks and chemical pollutants) that are left buried in the ground. Another
problem in such areas is chemical wastes that had been stored within the factory grounds.

117
Due to the nature of such sites, the subsurface is often very complex and is a challenging
target for most geophysical methods. The survey for this example was carried out on a
derelict industrial site where leachate was known, from a small number of exploratory wells,
to be moving from a surface waste lagoon into the underlying sandstones (Barker, 1996).
Eventually the leachate was seen seeping into a nearby stream. However, the extent of the
subsurface contamination was not known.
An electrical imaging survey was carried out along an old railway bed between the
lagoon and the stream. The metal railway lines had been removed except for short lengths
embedded in asphalt below a large metal loading bay. In the apparent resistivity pseudosection
(Figure 87a), the area with contaminated ground water shows up as a low resistivity zone to
the right of the 140 meters mark. The metal loading bay causes a prominent inverted V shaped
low resistivity anomaly at about the 90 meters mark. In the inversion model (Figure 87b), the
computer program has managed to reconstruct the correct shape of the metal loading bay near
the ground surface. There is an area of low resistivity at the right half of the section that agrees
with what is known from wells about the occurrence of the contaminated ground water. The
plume is clearly defined with a sharp boundary at 140 meters along the profile. The
contaminated zone appears to extend to a depth of about 30 meters.
Figure 87. Industrial pollution example, U.K. (a) The apparent resistivity pseudosection from
a survey over a derelict industrial site, and the (b) computer model for the subsurface.
7.4 Holes in clay layer - U.S.A.

This survey was carried out for the purpose of mapping holes in a clay layer that
underlies 8 to 20 feet (2.5 to 6.2 meters) of clean sand (Cromwell pers. comm.). The results
from the electrical imaging survey were subsequently confirmed by boreholes.
The pseudosection from one line from this survey is shown in Figure 88a. The data in
the pseudosection was built up using data from horizontally overlapping survey lines. One
interesting feature of this survey is that it demonstrates the misleading nature of the
pseudosection, particularly for the dipole-dipole array. In the inversion model, a high resistivity

118
anomaly is detected below the 200 ft. mark, which is probably a hole in the lower clay layer
(Figure 88b). This feature falls in an area in the pseudosection where there is an apparent gap
in the data.
However, a plot of the sensitivity value of the cells used in the inversion model shows
that the model cells in the area of the high resistivity body have higher sensitivity values (i.e.
more reliable model resistivity values) than adjacent areas at the same depth with more data
points in the pseudosection plot (Figure 88c). This phenomenon is due to the shape of the
contours in the sensitivity function of the dipole-dipole array (Figure 24), where the areas with
the highest sensitivity values are beneath the C1C2 and P1P2 dipoles, and not at the plotting
point below the center of the array. This example illustrates the danger of only using the
distribution of the data points in the pseudosection to constrain the position of the model cells
(Barker, 1992; Loke and Barker, 1996a). If the model cells are placed only at the location of
the data points, the high resistivity body will be missing from the inversion model, and an
important subsurface feature would not be detected!
Figure 88. Mapping of holes in a clay layer, U.S.A. (a) Apparent resistivity pseudosection for
the survey to map holes in the lower clay layer. (b) Inversion model and (c) sensitivity values
of the model cells used by the inversion program.

119
7.5 Time-lapse water infiltration survey - U.K.

Resistivity imaging surveys have not only been carried out in space, but also in time!
In some studies, the change of the subsurface resistivity with time has important applications.
Such studies include the flow of water through the vadose (unsaturated) zone, changes in the
water table due to water extraction (Barker and Moore, 1998), flow of chemical pollutants and
leakage from dams (Johansson and Dahlin, 1996). The modifications to the inversion method
to minimize variations across the different time models are described in Loke et al. (2014a).
A simple, but very interesting, experiment to map the flow of water from the ground
surface downwards through the unsaturated zone and into the water table was described by
Barker and Moore (1998). In this section, only some of the highlights of this experiment are
described as an illustration of a time-lapse survey. This experiment was carried out in the
Birmingham (England) area where forty thousand litres of water was poured on the ground
surface using a garden hose over a period of 10 hours. Measurements were made before and
during the irrigation of the ground surface, and after that for a period of about two weeks.
Figure 89 shows the results of a survey carried out at the beginning of the experiment before
the irrigation started. The inversion model (Figure 89b) shows that the subsurface, that consists
of sand and gravel, is highly inhomogeneous. The water was poured out near the 24 meters
mark on this line, and Figure 89c shows the inversion model for the data set collected after 10
hours of continuous irrigation. While the model resistivity values in the vicinity of the 24
meters mark are generally lower than the initial data set model in Figure 89b, the subsurface
distribution of the water is not very clear from a direct comparison of the inversion models
alone.
Figure 89. Water infiltration mapping, U.K. (a) The apparent resistivity and (b) inversion
model sections from the survey conducted at the beginning of the Birmingham infiltration
study. This shows the results from the initial data set that forms the base model in the joint
inversion with the later time data sets. As a comparison, the model obtained from the inversion
of the data set collected after 10 hours of irrigation is shown in (c).

120
The water distribution is more easily determined by plotting the percentage change in
the subsurface resistivity of the inversion models for the data sets taken at different times
(Figure 90) when compared with the initial data set model. The data set collected at 5 hours
after the pumping began shows a reduction in the resistivity (of up to over 50 percent) near the
ground surface in the vicinity of the 24 meters mark. The near-surface low resistivity zone
reaches its maximum amplitude after about 10 hours when the pumping was stopped (Figure
90b). Twelve hours after the pumping was stopped, the low resistivity plume has spread
downwards and slightly outwards due to infiltration of the water through the unsaturated zone.
There is a decrease in the maximum percentage reduction in the resistivity values near the
surface due to migration of the water from the near surface zone. This effect of the spreading
of the plume becomes increasingly more pronounced after 24 hours (Figure 90d) and 36 hours
(Figure 90e) due to further migration of the water. Note that the bottom boundary of the zone
with approximately 20 percent reduction in the resistivity values tends to flatten out at a depth
of about 3 meters (Figure 90e) where the plume from the surface meets the water table.
Figure 90. Time-lapse sections from the infiltration study. The sections show the change in the
subsurface resistivity values with time obtained from the inversion of the data sets collected
during the irrigation and recovery phases of the study.

121
7.6 Pumping test, U.K.

Resistivity imaging measurements were made during a pumping test in the Hoveringham area
of East Central England. The aquifer is a sand and gravel layer overlying mudstone. Figure 91
shows the initial apparent resistivity pseudosection, and model sections before pumping and
after 220 minutes of pumping. Figure 92 that shows the relative change in the resistivity at 40,
120 and 220 minutes after the start of the pumping test.
Figure 91. Hoveringham pumping test, U.K. (a) The apparent resistivity pseudosection at the
beginning of the test. The inversion model sections at the (b) beginning and (v) after 220
minutes of pumping.
Figure 92. Percentage relative change in the subsurface resistivity values for the Hoveringham
pumping test. To highlight the changes in the subsurface resistivity, the changes in the model
resistivity are shown. Note the increase in the model resistivity below the borehole with time.

122
Figure 92 clearly shows the increase in the zone with higher resistivity values with time due to
the extraction of the water. By using Archie’s Law, and assuming the water resistivity does not
change with time, we can estimate the change in the water saturation values. The decrease in
the water saturation level within the aquifer, or desaturation values, is shown in Figure 93. As
Archie’s Law assumes that the conduction is due to the water content alone, the desaturation
values are likely to be lower than the true values if there is significant clay content.
Figure 93. Use of Archie’s Law for the Hoveringham pumping test. Sections showing the
relative desaturation values obtained from the inversion models of the data sets collected during
the different stages of the Hoveringham pumping test. Archie’s Law probably gives a lower
limit for the actual change in the aquifer saturation.
7.7 Wenner Gamma array survey - Nigeria

The Wenner Gamma array (Figure 4c) has a relatively unusual arrangement where the
current and potential electrodes are interleaved. Compared to the Wenner Alpha and Beta
arrays, the Wenner Gamma array is much less frequently used in field surveys. However, in
some situations, there might be some advantage in using this array. The depth of investigation
is significantly deeper than the Wenner Alpha array (0.59a compared to 0.52a, see Table 2),
but the potential measured between the potential electrodes is only about 33% less than the
Alpha array. In comparison, the voltage that would be measured by the Wenner Beta array is
one-third that of the Alpha array which could be a serious disadvantage in noisy environments.
Figure 94a shows the Wenner Gamma array pseudosection from a groundwater survey
in the Bauchi area of Nigeria (Acworth, 1981). In this region, groundwater is frequently found
in the weathered layer above the crystalline bedrock. The weathered layer is thicker in areas
with fractures in the bedrock, and thus such fractures are good targets for groundwater. In this
area, the surveys were carried out with the Wenner Alpha, Beta and Gamma arrays, together
with electromagnetic profiling measurements using a Geonics EM34-3 system (Acworth,
1987). Here, only the result from the Wenner Gamma array data set is shown as an example.

123
To emphasize the boundary between the soil layer and the bedrock, the robust inversion
option was used (section 4.3). The inversion model is shown in Figure 94b. The thickness of
the lower resistivity weathered layer is generally about 10 to 20 meters. There is a narrow
vertical low resistivity zone with a width of less than 20 meters below the 190 meters mark that
is probably a fracture zone in the bedrock. A borehole well that was placed at the 175 meters
mark that lies just at the edge of the fracture zone. It had yields that were somewhat lower than
expected (Acworth, 1987). In such a situation, the 2D resistivity model would be useful to
pinpoint the exact location of the center of the fracture zone to improve the yield from the
borehole. The placement of the well was largely based on resistivity and EM profiling data,
and many years before 2D resistivity inversion software and fast microcomputers were widely
available.
Figure 94. Groundwater survey, Nigeria. (a). Apparent resistivity pseudosection. (b) The
inversion model with topography. Note the location of the borehole at the 175 meters mark.
As a final note, it is possible to invert data collected with the Wenner Alpha, Beta and
Gamma arrays along the same line simultaneously with the RES2DINV program as a single
data set. This can be done by using the "non-conventional array" option in the program where
the positions of all the four electrodes in an array are explicitly specified. This might be an
interesting method to combine the advantages of the different variations of the Wenner array.
7.8 Mobile underwater survey - Belgium

Contrary to popular belief, it is actually possible to carry out resistivity surveys
underwater, even in marine environments. This example is one of the most unusual data sets
that I have come across, and an interesting challenge for any resistivity imaging inversion
software. It is not only the longest in physical length and number of electrode positions, but
also uses an unusual highly asymmetrical non-conventional electrode arrangement collected
by an underwater mobile surveying system. Mobile surveying systems have an advantage of
faster surveying speed, but on land they suffer from the problem of poor ground contact (for
the direct contact type) or low signal strength (for the electrostatic type). Please refer to section
2.3 for the details. An underwater environment provides an almost ideal situation for a direct
contact type of mobile system since there is no problem in obtaining good electrode contact!
Figure 95a shows the data from an eight kilometers survey line along a river. This
survey was carried out by Sage Engineering of Belgium. The purpose of the survey was to map

124
the near surface lithology of the riverbed where there were plans to lay a cable. The data set
has a total of 7479 electrode positions and 6636 data points, whereas the inversion model used
has 19936 cells. On a 3.2 Ghz Pentium 4 computer, it took slightly less than 2 hours to process
this data set. On a newer PC, and using the appropriate software settings, it will probably takes
less than 15 minutes (see section 4.8)!
Figure 95. The inversion model after 4 iterations from an underwater riverbed survey by Sage
Engineering, Belgium.
In the inversion model (Figure 95b), most of the riverbed materials have a resistivity of
less than 120 m. There are several areas where the near-surface materials have significantly
higher resistivities of over 150 m. Unfortunately, geological information in this area is rather
limited. In the high resistivity areas, the divers faced problems in obtaining sediment samples.
The lower resistivity materials are possibly more coherent sediments (possibly sand with
silt/clay), whereas the higher resistivity areas might be coarser and less coherent materials.

125
7.9 Floating electrodes survey – U.S.A.

This survey was carried out along the Thames River in Connecticut, USA using a streamer
floating on the water surface that was towed behind a boat. The steamer has 2 fixed current
electrodes and 9 potential electrodes. Measurements were made with an 8-channel resistivity
meter system. The water depth and conductivity were also measured during the survey. A
dipole-dipole type of array configuration was used but some of the measurements used a non-
symmetrical arrangement where the potential dipole length was different from the current
dipole length. The apparent resistivity pseudosection from a survey line is shown in Figure 96a.
The inversion model (Figure 96b) obtained when no special constraints were placed on the part
of the model that falls within the water layer shows artefacts near the surface due to noise in
the data. This model is fairly accurate when the material just below the river bottom has a high
resistivity, but poor in areas with low resistivities. These problems do not occur in the model
where the water layer resistivity was fixed during the inversion process (Figure 96c).
Figure 96. Thames River (CT, USA) survey with floating electrodes. (a) The measured
apparent resistivity pseudosection. Inversion models obtained (b) without constraints on the
water layer, and (c) with a fixed water layer.
7.10 Oil Sands, Canada

A new application for electrical and electromagnetic exploration methods is in the
direct detection of hydrocarbons. It is well known that hydrocarbons have much higher
resistivity values as compared to the surrounding sediments or sedimentary rocks, and thus
ideal targets for electrical and EM based methods. One special form of hydrocarbon deposit
that is ideal for the electrical imaging method is the oil sands in Canada (Kellett and Bauman,
1999). It is estimated that Canada has over 400 billion cubic meters (2.5 trillion barrels) of oil
sands and heavy oil deposits. Commercial oil sands deposits are located at depths much
shallower than conventional hydrocarbon reservoirs, so they can often be accurately mapped
by conventional multi-electrode systems.
Figure 97a show the location of three major deposits in the Province of Alberta. The

126
largest deposits are the Athabasca Oil Sands with over 200 Gm 3 of reserves. The resistivity log
across one of the oil sands bodies (Figure 97b) shows high resistivity values approaching 1000
ohm.m for the tar sands as compared to less than 30 ohm.m for the sediments, making the
resistivity method a potentially good exploration approach. An example of a resistivity section
obtained from a 2D survey (Figure 97c) shows prominent high resistivity zones associated with
the oil sands.
Figure 97. Survey to map oil sands, Alberta, Canada. (a) Location of major tar sands deposits
in Alberta, Canada. (b) Example of resistivity log and geologic column of Athabasca oil sands.
(c) 2-D resistivity model from imaging survey (Kellett and Bauman, 1999).
Besides these examples, 2-D imaging surveys have been carried for many other
purposes such as the detection of leakage of pollutants from landfill sites, areas with undulating
limestone bedrock, mapping of the overburden thickness over bedrock (Ritz et al., 1999),
leakage of water from dams, saline water intrusion in coastal aquifers, freshwater aquifers
(Dahlin and Owen, 1998), monitoring of groundwater tracers (Nyquist et al., 1999) and
mapping of unconsolidated sediments (Christensen and Sorensen, 1994). The resistivity
imaging method has also been used in underwater surveys in lakes and dams.

127
8 3-D electrical imaging surveys
8.1 Introduction to 3-D surveys

Since all geological structures are 3-D in nature, a fully 3-D resistivity survey using a
3-D interpretation model (Figure 6c) should in theory give the most accurate results. At the
present time 3-D surveys is a subject of active research. However it has not reached the level
where, like 2-D surveys, it is routinely used. The main reason is that the survey cost is
comparatively higher for a 3-D survey of an area that is sufficiently large. There are two current
developments that should make 3-D surveys a more cost-effective option in the near future.
One is the development of multi-channel resistivity meters that enables more than one reading
to be taken at a single time. This is important to reduce the survey time. The second
development is faster microcomputers to enable the inversion of very large data sets (with more
than 10,000 data points and survey grids with more than 1000 electrode positions) to be
completed within a reasonable time.
8.2 Array types for 3-D surveys

The pole-pole, pole-dipole and dipole-dipole arrays are frequently used for 3-D surveys.
This is because other arrays have poorer data coverage near the edges of the survey grid. The
advantages and disadvantages of the pole-pole, pole-dipole and dipole-dipole arrays that were
discussed in section 2.5 with regards to 2-D surveys are also valid for 3-D surveys.
8.2.1 The pole-pole array

Figure 98 shows one possible arrangement of the electrodes for a 3-D survey using a
multi-electrode system with 25 nodes. For convenience the electrodes are usually arranged in
a square grid with the same unit electrode spacing in the x and y directions. To map slightly
elongated bodies, a rectangular grid with different numbers of electrodes and spacings in the x
and y directions could be used. The pole-pole electrode configuration was commonly used for
3-D surveys, such as the E-SCAN method (Li and Oldenburg, 1992; Ellis and Oldenburg,
1994b). The maximum number of independent measurements, nmax, that can be made with ne
electrodes is given by
nmax = ne (ne -1) / 2
In this case, each electrode is in turn used as a current electrode and the potential at all
the other electrodes are measured. Note that because of reciprocity, it is only necessary to
measure the potentials at the electrodes with a higher index number than the current electrode
in Figure 99a. For a 5 by 5 electrodes grid, there are 300 possible measurements. For 7 by 7
and 10 by 10 electrodes grids, a survey to measure the complete data set would have 1176 and
4500 data points respectively. For commercial surveys, grids of less than 10 by 10 are probably
not practical as the area covered would be too small.
It is can be very time-consuming (at least several hours) to make such a large number
of measurements, particularly with a single-channel resistivity meters. To reduce the number
of measurements required without seriously degrading the quality of the model obtained, an
alternative measurement sequence is shown in Figure 99b. In this proposed "cross-diagonal
survey" method, the potential measurements are only made at the electrodes along the x-
direction, the y-direction and the 45 degrees diagonal lines passing through the current
electrode. The number of data points with this arrangement for a 7 by 7 grid is reduced to 476
which is about one-third of that required by a complete data set survey (Loke and Barker,
1996b).

128
Figure 98. A simple arrangement of the electrodes for a 3-D survey.
Figure 99. Two possible measurement sequences for a 3-D survey. The location of potential
electrodes corresponding to a single current electrode in the arrangement used by (a) a survey
to measure the complete data set and (b) a cross-diagonal survey.

129
In some cases, 3-D data sets are constructed from a number of parallel 2-D survey lines
(section 8.3). Ideally there should be a set of survey lines with measurements in the x-direction,
followed by another series of lines in the y-direction. The use of measurements in two
perpendicular directions helps to reduce any directional bias in the data.
However, in some cases, only the data from a series of survey lines in one direction is
available. This is particularly common if the surveys were originally conducted to provide 2-
D images. Sometimes the spacing between the “in-line” electrodes is significantly smaller than
the spacing between the lines. One important question is the maximum spacing between the
lines that can be used for the data to be still considered “3-D”. A useful guide is the 3-D
sensitivity plot. Figure 100 shows the sensitivity values on horizontal slices through the earth.
The electrodes are arranged along the 0 and 1 meter marks along the x-axis. Near the surface,
there is an approximately circular region with negative sensitivity values in the top two slices
at depths of 0.07 and 0.25 meter. The zone with the largest sensitivity (using the 4 units
sensitivity contour line as a guide) extends in the y-direction to slightly over half the electrode
spacing. This means to get a complete 3-D coverage, if the measurements are only made in the
x-direction, the spacing between the lines should not be much more than the smallest electrode
spacing used.
The papers by Bentley and Gharibi (2004) and Gharibi and Bentley (2005) give a fairly
comprehensive discussion on constructing 3-D data sets from 2-D survey lines.
Figure 100. 3-D sensitivity plots for the pole-pole array. The plots are in the form of horizontal
slices through the earth at different depths.

130
The pole-pole array has two main disadvantages. Firstly it has a much poorer resolution
compared to other arrays. Subsurface structures tend to be smeared out in the final inversion
model. The second disadvantage, particularly for large electrode spacings, is that the second
current electrode and potential electrode must be placed at a sufficiently large distance from
the survey grid. Both disadvantages have been discussed in detail in Section 2.5.7. Park and
Van (1991) who used this array for a field experiment found that about 15% of the
measurements did not satisfy reciprocity because the contributions from the remote electrodes
were significant. In general, this probably affects the readings with the larger spacings.
8.2.2 The pole-dipole array

This array is an attractive alternative to the pole-pole array for surveys with medium
and large survey grids (12 by 12 and above). It has a better resolving power than the pole-pole
array (Sasaki, 1992), and is less susceptible to telluric noise since both potential electrodes are
kept within the survey grid. Compared to the dipole-dipole array, it has a significantly stronger
signal strength. Although it has one “remote” electrode (the C2 electrode), the effect of this
electrode on the measurements is much smaller compared to the pole-pole array (section 2.5.8).
As the pole-dipole array is an asymmetrical array, measurements should be made with the
“forward” and “reverse” arrangements of the electrodes (Figure 29). To overcome the problem
of low signal strength for large values of the “n” factor (exceeding 8), the “a” spacing between
the P1-P2 dipole pair should be increased to get a deeper depth of investigation with a smaller
“n” factor. The use of redundant measurements with overlapping data levels to increase the
data density can in some cases help to improve the resolution of the resulting inversion model
(section 2.5.9).
Figure 101 and Figure 102 show the sensitivity patterns for this array with the dipole
separation factor “n” equal to 1 and 4 respectively. There is prominent area with negative
sensitivity values between the C1 and P1 electrodes (located at 0.0 and 0.5 meter along the x-
axis). The plots are arranged such that the array length (in this case the distance between the
C1 and P2 electrodes) is set at 1.0 meter for both “n” factors. For a larger “n” factor, the area
of negative sensitivities between the C1 and P1 electrodes becomes larger and extends to a
greater depth. The array is more sensitive to structures off the array axis (i.e. in the y-direction)
when “n” is equals to 1. The area with the higher sensitivity values extends to about 0.8 times
the array length (Figure 101), or 1.6 times the unit electrode spacing. When the “n” factor is
larger (Figure 102), the array is more sensitive to off-axis structures near the P1-P2 dipole.
Note also the negative sensitivity values to the right of the P2 electrode.
If the 3-D survey is carried out with a series of parallel lines, and the cross-line
measurements are not made, the distance between the lines should preferably be within two to
three times the inline unit electrode spacing. This is to ensure that the subsurface material
between the lines is adequately mapped by the in-line measurements.
8.2.3 The dipole-dipole, Wenner and Schlumberger arrays

The dipole-dipole array can be recommended only for grids that are larger than 12 by
12 due to the poorer horizontal data coverage at the sides. The main problem that is likely to
be faced with this array is the comparatively low signal strength. Similar to 2-D surveys, this
problem can be overcome by increasing the “a” spacing between the P1-P2 dipole to get a
deeper depth of investigation as the distance between the C1-C2 and P1-P2 dipoles is increased.
Also, the use of overlapping data levels is recommended (section 2.5.9).

131
slices through the earth at different depths. The C1 electrode is the leftmost white cross.
slices through the earth at different depths.

132
Figure 103 and Figure 104 shows the sensitivity patterns for the dipole-dipole array
when the “n” factor is equal to 1 and 4 respectively. The sensitivity values have small but
negative values outside the immediate vicinity of the array. Another interesting feature is that
the sensitivity contours tend to be elongated in the y-direction, particularly for the larger “n”
value. The 4 unit sensitivity contour extends to about 0.6 times the array length in the y-
direction, or about 1.8 times the unit electrode spacing. This means that the array is more
sensitive to structures off the array axis compared to the pole-pole and pole-dipole arrays. This
feature is troublesome in 2-D surveys, but might be advantageous in 3-D surveys.
The off-axis elongation of the sensitivity contours agrees with the observation by
Dahlin and Loke (1997) that the dipole-dipole array is more sensitive to 3-D effects compared
to other common arrays. This factor is important when the dipole-dipole array is used in 2-D
imaging surveys where it is assumed that the subsurface geology is 2-D. In many cases, 3-D
data sets for the dipole-dipole arrays are constructed from a number of parallel 2-D survey
lines, particularly from previous surveys. Due to the elongated sensitivity pattern, the dipole-
dipole array can probably tolerated a larger spacing between the survey lines (to about three
times the inline unit electrode spacing) and still contain significant 3-D information.
In closing this section on four electrodes arrays, the sensitivity patterns for the Wenner
alpha (Figure 105), Schlumberger (Figure 106) and Wenner gamma arrays (Figure 107, the
Wenner beta is the dipole-dipole with a “n” value of 1 as shown in Figure 101) are shown. The
sensitivity contours for the Wenner alpha array, outside of the immediate vicinity of the
electrodes, are elongated in the direction of the line of electrodes. This means that the Wenner
alpha array is less sensitive to off-line structures than the dipole-dipole array, i.e. it is less
sensitive to 3-D. This agrees with empirical observations by Dahlin and Loke (1997). The
sensitivity pattern for the Wenner-Schlumberger array (Figure 106) is similar to that for the
Wenner alpha array except for a slight bulge near the center of the array. The sensitivity
patterns for the Wenner gamma array (Figure 107) show characteristic bulges near the C1 and
C2 electrodes that were observed earlier in the 2-D sensitivity sections (Figure 23). Thus it is
expected to be more sensitive to 3-D structures near the C1 and C2 electrodes.
8.2.4 Summary of array types

For relatively small grids of less than 12 by 12 electrodes, the pole-pole array has a
substantially larger number of possible independent measurements compared to other arrays.
The loss of data points near the sides of the grid is kept to a minimum, and it provides better
horizontal data coverage compared to other arrays. This is an attractive array for small survey
grids with relatively small spacings (less than 5 meters) between the electrodes. However, it
has the disadvantage of requiring two “remote” electrodes that must be placed at a sufficiently
large distance from the survey grid. The pole-dipole array is an attractive option for medium
size grids. It has a higher resolution than the pole-pole array and it requires only one remote
electrode and is much less sensitive to telluric noise. For surveys with large grids, particularly
when there is no convenient location for a remote electrode, the dipole-dipole array can be
used. The electrodes for 3-D surveys are frequently arranged in a rectangular grid. However,
the RES3DINV resistivity and I.P. inversion program can also handle survey layouts with an
arbitrary distribution of the electrodes.
In large 3-D I.P. surveys, it has become popular to use the offset version of the pole-
dipole and dipole-dipole arrays to cover a large area rapidly. The C1 current electrode (or C1-
C2 current dipole) is placed on a separate line parallel to the line with the potential receivers.
An example of this type of survey is described in section 8.12.4.

133
horizontal slices through the earth at different depths. The C2 electrode is the leftmost white
cross.
horizontal slices through the earth at different depths. The C2 electrode is the leftmost white
cross.

134
Figure 105. The 3-D sensitivity plots for the Wenner alpha array at different depths.
Figure 106. The 3-D sensitivity plots for the Wenner-Schlumberger array with the n=4 at
different depths.

135
Figure 107. The 3-D sensitivity plots for the Wenner gamma array at different depths.
8.3 3-D roll-along techniques

Most commercial 3-D surveys will probably involve grids of at least 16 by 16 in order
to cover a reasonably large area. A 16 by 16 grid will require 256 electrodes which is more
than that available on many multi-electrode resistivity meter systems. One method to survey
such large grids with a limited number of electrodes is to extend the roll-along technique used
in 2-D surveys to 3-D surveys (Dahlin and Bernstone, 1997). Figure 108 shows an example of
a survey using a multi-electrode resistivity-meter system with 50 electrodes to survey a 10 by
10 grid. Initially the electrodes are arranged in a 10 by 5 grid with the longer lines orientated
in the x-direction (Figure 108a). Measurements are made primary in the x-direction, with some
possible measurements in the diagonal directions. Next the entire grid is moved in the y-
direction so that the 10 by 5 grid now covers the second half of the 10 by 10 grid area. The 10
by 5 grid of electrodes is next orientated in the y-direction (Figure 108b).
The example data file PIPE3D.DAT was obtained from a survey using such a roll-along
technique. It was carried out with a resistivity-meter system with only 25 nodes with the
electrodes arranged in an 8 by 3 grid. The long axis of this grid was orientated perpendicularly
to two known subsurface pipes. The measurements were made using three such 8 by 3 sub-
grids so that the entire survey covers an 8 by 9 grid. For each 8 by 3 sub-grid, all the possible
measurements (including a limited number in the y-direction) for the pole-pole array were
made. In this survey, the second set of measurements in the y-direction (as in Figure 108b) was
not carried out to reduce the survey time and also because the pipes have an almost two-
dimensional structure.
For practical reasons, the number of field measurements in some surveys might be even
less than the cross-diagonal technique. Another common approach is to just make the

136
measurements in the x- and y- directions only, without the diagonal measurements. This is
particularly common if the survey is made with a system with a limited number of independent
electrodes, but a relatively large grid is needed. A roll-along procedure using only 3 parallel
cables is described in Dahlin and Bernstone (1997).
In some cases, measurements are made only in one direction. The 3-D data set consists
of a number of parallel 2-D lines. The data from each 2-D survey line is initially inverted
independently to give a series of 2-D cross-sections. The measured apparent resistivity values
from all the lines can also be combined into a 3-D data set and inverted with RES3DINV to
give a 3-D picture. While the quality of the 3-D model is expected to be poorer than that
produced with a complete 3-D survey, such a “poor man’s” 3-D data set could reveal major
resistivity variations across the survey lines (see also section 8.5). Until multi-channel
resistivity instruments are widely used, this might be the most cost-effective solution to extract
some 3-D information from 2-D surveys. This arrangement might be particularly useful for
surveys with the dipole-dipole array that can tolerate larger spacings between the survey lines.
Figure 108. Using the roll-along method to survey a 10 by 10 grid with a multi-electrode
system with 50 nodes. (a) Surveys using a 10 by 5 grid with the lines orientated in the x-
direction. (b) Surveys with the lines orientated in the y-direction.

137
8.4 A 3-D forward modeling program

In the interpretation of data from 2-D resistivity imaging surveys, it is assumed that the
subsurface geology does not change significantly in the direction that is perpendicular to the
survey line. In areas with very complex geology, there are could be significant variations in the
subsurface resistivity in this direction (i.e. the geology is 3-D), which could cause distortions
in the lower sections of the 2-D model obtained (please refer to section 4.7).
The 3-D resistivity forward modeling program, RES3DMOD.EXE, enables you to
calculate the apparent resistivity values for a survey with a rectangular grid of electrodes over
a 3-D structure. This is a Windows based program that can be used from within Windows
XP/Vista/7/8/10. To take a look the operation of the program, use the “File” option followed
by “Read model data” to read in the file BLOCK11.MOD, which has a 11 by 11 survey grid.
After that, click the “Edit/Display” option.
To modify the 3-D model, click the “Edit resistivity model” option. In this option, you
can change the resistivity of the 3-D cells in the mesh used by the finite-difference method
(Dey and Morrison, 1979b) to calculate the apparent resistivity values. To quit from the “Edit”
mode, press the Q or the Esc key. To calculate the apparent resistivity values, click the
“Calculate” option. To take a look at the apparent resistivity pseudosections, click the “Display
apparent resistivity” option. You can choose to display the apparent resistivity values in the
form of horizontal pseudosections, or as vertical pseudosections as used in 2-D surveys.
Displaying the vertical pseudosections will give you an idea of the effect of a 3-D structure on
the measurements in a 2-D survey. In a study made by Dahlin and Loke (1997), the dipole-
dipole array was found to be the most sensitive to 3-D effects while the Wenner array was the
least sensitive.
The RES3DMOD program also has an option to save the apparent resistivity values
into a format that can be accepted by the RES3DINV inversion program. As an exercise, save
the apparent resistivity values as a RES3DINV data file for one of the models, and later carry
out an inversion of this synthetic data set.
Table 12. 3-D forward modeling examples

PRISM2.MOD – This is a 3-D (1). Read in the file, and then run the “Calculate” step to
model used to generate the calculate the potential values. This is a relatively large
pseudosections shown in model file, so you might to collect some coffee while the
Figure 67. computer is running.
(2). Next choose the “Display apparent resistivity” option
under the “Edit/Display” menu. Try first with pole-pole
array, and take a look at the pseudosections.
(3).Next try with other arrays.
BLOCK15.MOD – A 15 by 15 (1). Read in the file, and then run the “Calculate” step to
grid model with several calculate the potential values.
rectangular prisms. (2). Next choose the “Display apparent resistivity” option
under the “Edit/Display” menu. Take a look at the
pseudosections for a few arrays.
(3). Try using the “Edit model” option to change the model,
and then recalculate the potential values. Check out the
effect of your changes on the apparent resistivity
pseudosections.

138
Figure 109a show an example of a 3-D model with a 15 by 15 survey grid (i.e. 255
electrodes). The model, which consists of four rectangular prisms embedded in a medium with
a resistivity of 50 m, is shown in the form of horizontal slices through the earth. The apparent
resistivity values for the pole-pole array (with the electrodes aligned in the x-direction) are
shown in the form of horizontal pseudosections in Figure 109b. Note the low resistivity block
with a resistivity of 10 m near the centre of the grid that extends from a depth of 1.0 to 3.2
meters. For measurements with the shorter electrode spacings of less than 4 meters this block
causes a low resistivity anomaly. However, for electrode spacings of greater than 6 meters, this
low resistivity prism causes a high resistivity anomaly! This is an example of “anomaly
inversion” which is caused by the near-surface zone of negative sensitivity values between the
C1 and P1 electrodes (Figure 100).
Figure 109. A 3-D model with 4 rectangular prisms in a 15 by 15 survey grid. (a) The finite-
difference grid. (b) Horizontal apparent resistivity pseudosections for the pole-pole array with
the electrodes aligned in the x-direction.

139
8.5 3-D inversion algorithms and 3-D data sets

Two interesting questions that sometimes arise, particularly in non-academic circles, is
the definition of a 3-D inversion algorithm and what constitutes a 3-D data set. The first
question can be easily answered, but the second is less clear.
A defining feature of a 3-D algorithm inversion algorithm is that it allows the model
resistivity values to vary in all three directions, i.e. in the x-, y- and z-directions. This is in
contrast to 2-D inversion where the subsurface resistivity is assumed to vary only in the x- and
z-directions but constant in the y-direction; and in 1-D inversion where the resistivity is only
allowed to change in the z-direction. The inversion model used by the RES3DINV program
consists of independent rectangular cells (Figure 110) where the model values are allowed to
vary in all three directions simultaneously, so it is uses a true 3-D inversion algorithm. Note
that a model constructed from a series of 2-D inversions along parallel lines is not a true 3-D
inversion model.
Another defining characteristic of a 3-D inversion algorithm is the use of a 3-D forward
modeling subroutine, such as the 3-D finite-difference and finite-element methods (Dey and
Morrison, 1979b; Silvester and Ferrari, 1990), to calculate the model apparent resistivity and
Jacobian matrix values.
A more difficult question is what constitutes a “3-D” data set. At present there is no
generally accepted definition of a 3-D data set. While the inversion algorithm used to invert
the data set is 3-D, whether the data set contains significant 3-D information is another matter.
It was stated in section 2.2 that to get a good 2-D model for the subsurface the data coverage
must be 2-D as well. For a 2-D survey, measurements are made with different electrode
spacings and at different horizontal locations to obtain such a 2-D coverage. However, the
degree of data coverage needed before a data set can be considered “3-D” is less clear. Using
the pole-pole survey as an example, the following classification system is proposed for 3-D
data sets. They are listed in decreasing order of 3-D information content.
Category 1 - An ideal 3-D survey with the electrodes arranged in a rectangular grid, and with
measurements in all possible directions (such as in Figure 99a), i.e. along the grid lines as well
as at different angles to the grid lines.
Category 2 – The electrodes are arranged in a rectangular grid. All the measurements along
the grid lines (i.e. in the x- and y-directions) but only a limited number of measurements at an
angle to the grid lines( such as along the 45 degree diagonals for square grids as shown in
Figure 8.2b) are made. The ROOTS7.DAT data file is such an example of a survey with limited
measurements in the angular directions. In this case, the survey was carried out 49 electrodes
in a 7 by 7 grid and 468 measurements were made. A “complete” 3-D data set would have 1176
measurements.
Category 3 – Measurements are only made in the two directions along the grid lines, i.e. in the
x- and y-directions, and no measurements at an angle to the grid lines are made. This
measurement sequence is frequently used when there are insufficient nodes in the multi-
electrode system to cover the entire survey area at a single time. One possible measurement
sequence is shown in Figure 108. The data for the sludge deposit field example (Dahlin and
Bernstone 1997) in section 8.7.2 falls under this category.
Category 4 – Measurements in only one direction (for example the x-direction) along a series
of parallel 2-D survey lines. This situation is common for data from old surveys, particularly
I.P. surveys for mineral deposits. For this type of data set, a series of 2-D inversions is usually
first carried out. The 3-D inversion is then used on a combined data set with the data from all
the survey lines in an attempt to gleam new information out of old data, and to see whether 3-
D effects are significant (i.e. whether the results from the 2-D inversions are valid). The success
of the 3-D inversion partly depends on the spacing between the lines and the type of array used
(section 8.2). As a general rule, the spacing between the lines should not be more than twice

140
the unit electrode spacing along the lines.

While ideally the data should be of Category 1 or at least Category 2 so that some
angular data is available, the RES3DINV program will also accept data that falls under
Categories 3 and 4. The accuracy of the 3-D models obtained from the Categories 3 and 4 data
types will be lower than the Categories 1 and 2 data types, and will greatly depend on the
spacing between the lines and the type of array used. However, even for the Category 4 data
type, the results from the 3-D inversion should provide a useful indicator on whether 3-D
effects are significant. This can provide a check on the validity of the results obtained from
independent 2-D inversions of the different survey lines.
Figure 110. The model used in 3-D inversion.
8.6 A 3-D inversion program

3-D inversion of field data set can be carried out in a similar way using the smoothness-
constrained least-squares method used for the 2-D inversion. The model used to interpret the
3-D data set is shown in Figure 110. The subsurface is divided into several layers and each
layer is further subdivided into a number of rectangular cells. A 3-D resistivity inversion
program, RES3DINV, is used to interpret the data from 3-D surveys. This program attempts to
determine the resistivity of the cells in the inversion model that will most closely reproduce the
measured apparent resistivity values from the field survey. Within the RES3DINV program,
the thickness of the layers can be modified by the user.
Please refer to the instruction manual (RES3DINV.PDF) for the RES3DINV program
for the data format. The set of files that comes with the RES3DINV program package has a
number of field and synthetic data files. You can carry out an inversion of some of these files
to get a feel of how the program works.

141
Table 13. 3-D inversion examples

PIPE3D.DAT – A field survey data (1). Read in the file, and then run the “Carry out
file where a roll-along technique inversion” step.
was used. The total grid size is 9 by (2). After that, take a look at the results using the
8. “Display” window.
ROOTS7.DAT – A “cross-diagonal (1). Read in the file, and then invert the data set.
survey” data set using a 7 by 7 (2). Try using the “Robust inversion” option, and see
survey. whether there is a significant change in the results.
ROOT7_REMOTE.DAT – The Try inverting this data file and see whether the remote
same data set with the location of electrodes have a significant effect of the results.
the C2 and P2 electrodes included.
BLOCKS11T.DAT – A synthetic (1). Read in the file, and then invert the data set.
data set with topography. It has a 11 (2). Invert again using the “Robust inversion” option.
by 11 grid.
8.7 Banding effects in 3-D inversion models

Many 3-D field data sets are constructed by combining a series of parallel 2-D survey
lines. Ideally the distance between the lines should be about the same as the distance between
adjacent electrodes in each line. However, in most cases, the distance between the lines can be
two or more times larger than the inter-electrode spacing along the lines. This can lead to a
‘banding’ or ‘herring-bone’ effect with the structures aligned along the two axes of the survey
grid, particularly if the l1-norm or blocky inversion method is used. Figure 111 shows several
alternative models for a survey over the Vetlanda landfill site in Sweden that consists of 11
parallel 2-D lines with 63 electrodes positions along each line (Rosqvist et al., 2009). The pole-
dipole array was used. The spacing between the lines is twice the unit electrode spacing along
the lines. Only the first 6 layers (that are most affected by the banding effect) of the inversion
models are shown. The model using the standard inversion settings (Figure 111a) shows severe
banding effects, particularly in the low resistivity zone in the top two layers. This is reduced,
particularly in the topmost layer, by using a higher damping factor for the topmost layer (Figure
111b).
The normal horizontal roughness filter used has components in the x and y directions
only (Figure 112b). Thus it has a tendency to produce structures aligned along the x and y
directions. Since the model cells have sides that are aligned in the same direction as the survey
grid lines, the structures in the inversion models tend to have edges that are aligned in the same
directions as the survey lines. To reduce this effect, the roughness filter used is modified so
that it has components in the diagonal directions (Farquharson, 2008) as well in the x-y plane
(Figure 112d). Figure 111c shows the inversion model produced when this modification is
made to the horizontal roughness filter, while the vertical roughness filter is still only applied
between the model cell and the cells immediately above and below it. The prominent banding
effect in the low resistivity region in the top two layers of the model (Figure 111c) is greatly
reduced using this horizontal diagonal filter.
Applying the diagonal filter components in the horizontal direction but not in the
vertical direction can lead to a bias (in the horizontal direction) in the shape of the structures
produced. To remove this bias, the diagonal components are also applied to the vertical
roughness filter as well in the x-z and y-z planes. The model produced with the diagonal
components in the horizontal and vertical directions is shown in Figure 111d. The banding
effect is no longer discernable in this model.

142
Figure 111. Inversion models for the Vetlanda landfill survey data set. (a) Using standard
inversion settings. (b) With a higher damping factor for the topmost layer. (c) Using diagonal
roughness filter in the horizontal (x-y) directions. (d) Using diagonal roughness filters in the
vertical (x-z and y-z) directions as well. (e) Using the roughness filter in all directions.

143
Figure 112. Types of 3-D roughness filters. (a) With components in the x- and y- directions
only for the horizontal filter. (b) With components in the diagonal directions in the x-y plane
for the horizontal filter. (c) Applying the roughness filter with the corner model cells as well.
Only two (out of eight) corner cells are shown.
In the final model, the diagonal components for the roughness filter are also applied
between the model cell and the cells adjacent to the corners of the cell (Figure 112e). Except
for the cells at the surface, bottom and sides of the model grid, there are eight such corner
neighboring cells surrounding each cell. The roughness filter is now applied between each
interior model cell and all the 26 cells surrounding it. The resulting inversion model shown is
in Figure 111e. In theory, this should reduce any bias in shape of the structures from the
alignment of model cells grid (Farquharson, 2008; Loke and Dahlin, 2010).
The model in Figure 111 has model cells with widths of 1 meter in the x direction and
2 meters in the y direction, i.e. elongated in the y direction. This is because the model
discretization follows the spacing between the electrode positions. It is possible the banding in
the y direction is partly caused by the model cells being elongated in the y direction. A possible
alternative method to reduce the banding is to use model cells that are of the same widths in
both directions, i.e. a selective model refinement in the y direction. This doubles the number of
model cells as well as the number of nodes in the finite-difference grid. Figure 113 shows the
inversion models using cells of 1 meter length in both the x and y directions. The inversion
model with the standard inversion settings (Figure 113a) does not exhibit the prominent
banding in the y direction (Figure 111a). Instead there is a slight banding in the x direction in
the top two layers with clustering of very low or high resistivity values along the survey lines.
This feature is removed when a higher damping factor is used for the first layer (Figure 113b).
The boundary of the low resistivity zone (marked by the light blue color) between x distances
of 15 to 27 meters in the top three layers in this model shows a 'stepped' pattern aligned in the
x or y direction. This is probably due to the use of a roughness filter with components in these
directions (Figure 112b). This feature is removed when the roughness filter with diagonal
components is used (Figure 112d) that does not bias the structure in the x and y directions
(Figure 113c). The inversion model with the diagonal filter components in the z direction as
well (Figure 113d) is also free of the 'stepped' boundary pattern.

144
Figure 113. Inversion models for the Vetlanda landfill survey data set using model cells of
equal lengths in the x- and y- directions. (a) Using standard inversion settings. (b) With a higher
damping factor for the topmost layer. (c) Using diagonal roughness filter in the horizontal (x-
y) directions. (d) Using diagonal roughness filters in the vertical (x-z and y-z) directions as well.
8.8 The use of long electrodes in 3-D surveys

It is normally assumed that the dimensions of the electrodes used are much smaller than
the spacing between them so that they can be considered as ‘point’ electrodes. In some
situations, the length of the electrode is sufficiently long such that it has to be taken into
account. One situation where this arises is when a metallic well casing is used as an electrode.

145
The resistivity of the metal well casing is much lower than the surrounding medium, so the
potential on the surface of the casing is essentially constant. The use of such ‘long’ electrodes
has been the subject of several papers (Schenkel and Morrison, 1994; Singer and Strack, 1998;
Ramirez et al., 2003; Daily et al., 2004). Many of these studies involve deep surveys where the
measurements are made using the metal cased wells as electrodes. One situation of interest for
environmental and engineering surveys is the use of both point and long electrodes in a survey
where existing metal wells can be used as the long electrodes (Rucker at al., 2007, 2010). It is
well known that well-to-well measurements involving vertical ‘long’ electrodes only have poor
vertical resolution. However, measurements between a ‘point’ and a ‘long’ electrode have
better vertical resolution. It might even help when the long electrode extends below the target
of interest.
Figure 114. Synthetic model for long electrodes survey. 3-D model using cells of low
resistivity (0.01 .m) that are marked in red to simulate cased wells.
Figure 114 shows a model with a low resistivity block of 20 .m at a depth of 2.9 to
6.0 meters. The maximum length of the block is 3 meter that is less than its average depth, thus
it will be a difficult target for normal surveys with point electrodes on the surface. Figure 115a
shows the inversion model for a pole-pole survey data set using the cross-diagonal
measurement sequence (Figure 99) using only point electrodes on the surface. Another data set
was generated for the same model with three of the point electrodes converted into long
electrodes. This was done by assigning a very low resistivity value to the cells around the point
electrode, as marked by the red cells in Figure 114. Two of the long electrodes extend below
the low resistivity block. The inversion model for the data set using the combination of point
and long electrodes (Figure 115b) shows slightly better resolution for the low resistivity block.
The lowest resistivity achieved in model is about 77 .m (Figure 115b) which is lower than
the value of 83 .m achieved by the model for the data set using only surface point electrodes

146
(Figure 115a).
Figure 115. Comparison of inversion models using point electrodes with and without long
electrodes. (a) Inversion model for pole-pole data set using only surface point electrodes. (b)
Inversion model for pole-pole data set using 121 point and 3 long electrodes.
8.9 Data grid formats and model discretizations

8.9.1 Types of surveys and model grids
There has been a trend over the years to reconcile theory with practice. Initially, 3-D
models assume a simple situation where the electrodes used in the survey have uniform spacing
in both the x and y directions (Figure 116a). This situation was sufficient in simple research
setups, and easy for computer programming. However, over the years, more complex data
formats were developed so that more complex field situations could be accommodated. The
next step is to allow for a situation where the electrodes are still arranged in a rectangular grid,
but the distance between the survey lines is not constant (Figure 116b). However, due to
physical obstructions, it is sometimes not possible to run straight survey lines. The trapezoidal
data grid format allows for small deviations from straight survey lines (Figure 116c). However,
it still assumes each line has the same number of electrodes in the x or y directions. To convert
the 3-D of field data set into a resistivity model for the subsurface, we divide the subsurface
into a number of blocks. The three data grid formats tie the (x,y) positions of the corners of the
model blocks to the electrode positions. In more complex surveys, the surveys lines might be

147
in different directions with different numbers of electrodes. A common situation encountered

in mineral surveys is data collated from different survey campaigns at different times over the
same prospect. This results in a data set with survey lines running in different directions. The
fourth grid format (Figure 116d) is so far the most general and designed to handle almost any
electrode distribution. The (x,y) positions of the model blocks are now independent of the
electrode positions. A non-uniform rectangular grid is used to set up the (x,y) corners of the
model blocks so that a finer grid can be used where there are more electrode positions. In the
future, more complex mathematical models might have to be used so that it more closely
matches the field situation. One modification is to handle surveys where the lines are not
orientated north-south or east-west. This requires a simple data coordinate reorientation.
Another possible situation is where the length of the survey lines vary across the survey area.
Figure 116. 3-D data grid formats.
8.9.2 Model grid optimization for large surveys with arbitrary electrodes positions
The data format that separates the model grid from the survey electrodes positions
(Figure 116d) has made it possible to model data from very complex survey layouts. This
flexibility comes at the price of higher computer memory requirements. Thus this option is
only available in the 64-bit RES3DINVx64 program. In this section, we will take a closer look
at the finite-difference and finite-element mesh used and ways to optimize it so as to reduce
the memory required.
Firstly, the program subdivides each model cell by four mesh lines in both the x and y
directions (Figure 117). When the electrode falls on a node location (at the intersection of the
x and y mesh lines), it can be directly modeled by that node. There are two alternatives to model
the effect of an electrode at an arbitrary position with the finite-difference and finite-element
methods when it does not fall on a mesh node. The first is to calculate the potential at the
electrode by interpolating the potentials (Figure 117a) at the four nearest nodes in the mesh
(and similarly replace a single current electrode node by four equivalent current sources). The
second method moves the nearest node to the location of the electrode using a distorted finite-

148
element mesh (Figure 117b). For both methods, it is important that two electrodes used in the
same array are sufficiently far apart in the mesh setup used. Figure 118a shows a problem when
a current electrode is close to a potential electrode in the same array when the interpolation
method is used. Both the current and potential electrodes will share a common node if they are
too close together. To avoid this problem, there should be at least two mesh lines between the
two electrodes so that they do not share a common node (Figure 118b).
It is possible to avoid the situation in Figure 118a by using smaller model cells.
However, using small model cells will increase the number of nodes in the finite-difference or
finite-element mesh. It will also increase the total number of model cells in the inversion model.
Both of this will increase the computer memory and time required for the inversion of the data
set. It might be necessary to make some adjustments to the model grid specified to obtain the
optimum balance.
Figure 117. Methods to model the effect of an electrode using the finite-difference and finite-
element methods.
Figure 118. The use of an appropriate mesh spacing to obtain sufficient accuracy for electrodes
in the same array that close together.
To illustrate this, we used the data from a survey at the Hanford site in Washington state, USA
where the waste material was stored in trenches and concrete cribs (Rucker et al., 2007).
Different resistivity survey phases were carried out using 2-D lines (Figure 119). The

149
distribution of the electrodes does not fit into a simple rectangular grid. The first attempt to
model the data uses a model grid with cell widths of 5 meters. This produced a model grid with
125x98 lines in the x-y directions. The inversion model had 16 layers that gave a total of 192448
model cells. While the inversion of the data set could be carried out on a PC with 24 GB RAM,
there were a number of electrode that were too close in the mesh, as marked by the light blue
points in the grid display (Figure 120a). The next model grid has cell widths of 4 meters (Figure
8.23b). This eliminated electrodes that were less than 2 mesh lines apart, but this produced a
model grid with 156x122 lines in the x-y directions. The inversion model has 300080 cells, and
the finite-difference mesh had more than 1.5 million nodes and the PC with 24 GB RAM was
not able to run the inversion. The third model grid with cell widths of 4 meters in the x-direction
where the electrodes were close together (on the right side of the mesh) and 5 meters elsewhere.
This gave a model grid with 136x98 lines. The model has 209520 cells that could be handled
by the PC with 24 GB RAM. The final inversion model obtained using this model grid setup
is shown in Figure 121. The linear features in the second to fifth layers are due to the trenches
and concrete cribs. The leakage zones are marked by the prominent low resistivity zones in the
fifth and deeper layers. Note that the bottom left corner of all the layers show relatively less
resistivity variation compared to the other sections. This is basically due to the fact it has no
data coverage as the survey lines do not cross this section.
Figure 119. Map with survey lines and infrastructure at the Hanford site.

150
Figure 120. Types of model grids for the Hanford survey data set. (a) Using a 5 meters spacing
model for the entire area. (b) Using a 4 meters spacing model for the entire area. (c) Using a
mixture of 4 and 5 meters spacing model grid.

151
Figure 121. Inversion model for Hanford survey site.
8.10 3-D array optimization – grids and perimeters

In a previous section (section 4.9.4), a method to generate optimized arrays for 2-D
surveys using the model resolution was discussed. As 3-D surveys might involve hundreds to
thousands of electrode positions, the use of similar array optimization techniques might appear
to be an intractable problem. However by using a combination of improvements in the array
optimization algorithm, advances in computer hardware and improved numerical algorithms
(Loke et al., 2015a) together with a novel method to generate the test arrays (Loke et al.,
2013b), this is now possible. The key to the use of the array optimization method is to take
advantage of the fact that practical field survey have limitations in the number of electrodes
that be accessed at the same time due to equipment limitations. Most field equipment have less
than 100 nodes, certainly less than the thousands of possible electrode positions in a 3-D field
survey. One example is shown in Figure 122 with 3 cables with 21 electrodes each (Dahlin et
al., 2002). A larger area can be surveyed by moving the setup in the y direction using a roll-
along method, such as the 21 by 17 survey grid used by Dahlin et al. (2002). Although the final
survey grid has 357 electrodes, not all possible combinations of the electrodes can be used.
This greatly reduces the size of the possible array combinations. As an example of the
improvements possible using optimized arrays, the results for a synthetic test model used in
Loke et al. (2013b) is shown in Figure 123. The model consists of three rectangular blocks with
resistivity of 1000 ohm.m at different depths embedded in a homogeneous medium of 100
ohm.m. The survey grid consists of 8 lines with 21 electrodes each. The inline electrode spacing
is 1 meter while the distance between the lines is 2 meters. The first data set consists of the
inline Wenner-Schlumberger and dipole-dipole arrays where the geometric factor is less than
2262 m. The second data set was created by the ‘Compare R’ array optimization method.
Voltage dependent Gaussian random noise (Zhou and Dahlin, 2003) with an amplitude of 1
milli-ohm was added to the resistance values before they were converted into apparent
resistivity values. The L1-norm method was used for both the data misfit and model roughness

152
(Loke et al., 2003) in the inversion of this data. The L-curve method was used to estimate the
optimum inversion damping factor. Figure 123 shows the inversion models for the
conventional arrays (with 3152 data points) and the optimized data (3154 data points) sets. The
optimized data set has slightly more points to maintain symmetry in the array configurations
used. The top two blocks are poorly resolved by the conventional arrays data set (Figure 123a)
compared to the optimized data set (Figure 123b). The width of the topmost block anomaly is
twice the actual size while its maximum resistivity is less than 240 ohm.m. In comparison, the
optimized arrays inversion model has a maximum resistivity of about 640 ohm.m and the
correct width. The optimized arrays also achieve significantly higher resistivity values at the
locations of the second and the third blocks. Note that although the data misfit for the optimized
data set (0.66%) is slightly higher than that for the conventional data set (0.34%) due to the
higher average geometric factor of the optimized arrays, the model resolution is better.
Figure 122. Arrangement of survey lines using a 3 cable system with the Abem SAS
instrument.
Figure 123. Inversions model for (a) combined Wenner-Schlumberger and dipole-dipole data
set, (b) optimized data set. The actual positions of the blocks are marked by black rectangles.
An interesting special case of a 3-D survey is when the electrodes are confined to the

153
perimeter of the survey area (Tejero-Andrade et al., 2015). The arrays used for such surveys
are commonly based on heuristic rules, and very often designed for a perimeter with sharp
corners such as a rectangle (Baker et al., 2001), and might not be applicable for perimeters with
smooth shapes such as a circle. An algorithm to generate optimized arrays for perimeters of
any shape is described in Loke et al. (2015d). As an example, for a survey with 40 electrodes
arranged in a square perimeter, the comprehensive data set has nearly 180 thousand arrays.
Figure 124a shows model resolution plots for the first 6 layers for the comprehensive data set
that shows the maximum possible resolution for this survey setup. The highest resolution
values are in the first layer near the electrodes and decrease rapidly with depth and towards the
centre of the survey region. The resolution sections for the ‘conventional’ arrays consisting of
the ‘L and Corner’ arrays (Tejero-Andrade et al., 2015) with 946 data points have much lower
resolution values, particularly in the deeper layers (Figure 124b). The optimized arrays
generated by the ‘Compare R’ method with the same number of data points have significantly
higher resolution values (particularly in the 5 th layer), although the resolution values are well
below that of the comprehensive data set (Figure 124c). Figure 125 shows model resolution
sections in a vertical plane across the middle of the area. It more clearly shows that the
optimized arrays have higher resolution values than the conventional arrays particularly at the
middle and at depth. This is clearly shown by the relative model resolution sections calculated
using the ratio of the resolution of the data set with that of the comprehensive data set (lower
row of sections in Figure 125). The ‘conventional’ arrays clearly has lower resolution values
near the center of the survey region and at depth. The larger optimized data set with 2000 data
points has an average resolution of 0.091 compared to 0.111 for the comprehensive data set
with almost 100 times more arrays.
Figure 124. Horizontal sections showing the model resolution for the (a) comprehensive data
set, (b) standard arrays and (c) optimized arrays. The electrode positions are marked by small
green crosses in the top layer in (a).

154
Figure 125. Vertical cross-sections showing the model resolution for the comprehensive,
‘standard’, small and large optimized data sets.
Figure 126. The synthetic test model with two 1000 ohm.m rectangular blocks (marked by
black rectangles) embedded in a 100 ohm.m background medium.
To illustrate the performance of the different arrays, a test model consisting of two high
resistivity rectangular blocks of 1000 ohm.m embedded within a 100 ohm.m medium ( Figure 126)
was used. The first test data set consists of the conventional 'L and Corner' arrays (Tejero-

155
Andrade et al., 2015) with the maximum geometric factor set at 4147 m that gives 946 data
points. Gaussian random noise (Zhou and Dahlin, 2003) with a mean amplitude of 1 milliohm
was added to the calculated resistance values before they were converted to apparent resistivity.
The 'L and Corner' arrays model (Figure 4a) detects the blocks with maximum resistivity values
of about 110 ohm.m but the shape of the deeper block is not clearly defined. The smaller
optimized data set model has a higher maximum value of about 119 ohm.m for the deeper block
(Figure 4b) with the maximum anomaly value that is closer to the true position in the upper
part of the block. However the top edge of the block is not well resolved with a small artifact
that extends to the surface. The poor vertical resolution is expected from the PSF plot (Figure
3e). The model for the larger optimized set with 2000 data points (Figure 4c) shows higher
values of 115 and 124 ohm.m at the positions of the top and bottom high resistivity blocks.
Although the optimized data sets have higher models misfits (1.5%) compared to the 'L and
Corner' arrays (0.5%) due to higher average geometric factors of the arrays used, the deeper
block is better resolved in the inversion models. The array optimization program took 616
seconds to generate the data set with 2000 arrays.
Despite being able to detect the larger block near the center of the survey grid, the
optimized data sets cannot clearly resolve the top of this structure. It was also observed by Tejero-
Andrade et al. (2015) that it was not possible to resolve the top of the structures used in their synthetic
models tests. Figure 128 shows plots of the point-spread-function (PSF) values (Oldenborger
and Routh, 2009) for a model cell at different depths located near the center of the square for
the comprehensive data set. All the PSF plots have a predominantly vertical pattern. The
regions with significant PSF values are spread out about 2 m laterally but about 4 m vertically.
As the comprehensive data set contains all the possible arrays, this indicates any survey with
the electrodes confined to the perimeter will be able to resolve horizontal boundaries better
than vertical boundaries. When the depth of the model cell is less than 2 m (or less than half
the distance from the nearest electrode) the highest PSF values occur near the surface. This is
the reason the anomalies in the inversion models reach the surface although the actual top
boundary of the structure is deeper. The maximum value of the PSF plot for the cell at 3.08 m
depth (Figure 128e) is located below the surface closer to the actual depth of the cell. Thus for
depths of more than half the distance from the nearest electrode the maximum model anomaly
value is expected to be below the surface although detecting the structure will more difficult
due to the loss of resolution with depth.
An example with a survey on a circular perimeter is next used to illustrate the general
nature of the array optimization algorithm. The array optimization algorithm has the advantage
over heuristic rules (based on a particular perimeter shape) as the same technique can be used
regardless of the shape of the perimeter. Figure 129 shows the model resolution sections for a
survey perimeter with 40 electrodes in a circle. As expected, the resolution decreases towards
the center of the survey area. The test model has two rectangular blocks of 10 ohm.m in a 100
ohm.m medium (Figure 130). The two blocks are detected in the inversion model for the
smaller optimized data set with 946 data points (Figure 130a). However, the top and bottom
boundaries of the deeper block are now well resolved. There is a slight improvement with the
larger optimized data set (Figure 130b). The left boundary of the top block and the bottom
boundary of the lower block are better resolved.

156
Figure 127. The inversion models for the synthetic model with two 1000 ohm.m rectangular
blocks (marked by black rectangles) embedded in a 100 ohm.m background medium.
Figure 128. Comprehensive data set point-spread-function plots on a vertical x-z plane located
at y=4.5 m for a model cell at different depths with centre at x=4.5m and y=4.5m.

157
Figure 129. Model resolution sections with circular perimeter for (a) comprehensive data set
with 180300 arrays, optimized data sets with (b) 946 and (c) 2000 arrays.
Figure 130. Inversion results for survey with a circular perimeter using optimized arrays with
(a) 946 and (b) 2000 data points.

158
8.11 Unstable arrays and the geometric factor relative error

It has become increasingly popular in large scale mineral exploration surveys to use the
offset pole-bipole or bipole-bipole type of arrays that greatly reduces the survey time and cost
(White et al., 2001). The current and potential electrodes are laid out on separate lines. This
not only reduces EM coupling in I.P. surveys, but reduces the number of times the current
electrodes (with associated heavy current generator) have to be moved. One problem that has
arisen with this type of layout that is gives rise to 'unstable' arrays where the two potential
electrodes almost lie on an equipotential contour line. Usually it is sufficient to use the
geometric factor to identify 'unstable' arrays that are likely to have very low potentials. Arrays
with very large geometric factors will have very low potentials, and thus likely to be very noisy.
However it has been found that arrays with acceptable geometric factors but with electrodes on
different lines can also be 'unstable' (Loke et al., 2014c). A similar situation occurs for cross-
borehole surveys where the current and potential electrodes are located in different boreholes
(Wilkinson et al., 2008). In mineral exploration surveys, the current transmitter is usually
not linked to the potential measuring units, thus the sign of the measured potential cannot be
determined. It is usually assumed the measured potential (and thus apparent resistivity) is
positive. So a situation can arise where the field array has a positive apparent resistivity value,
but the calculated apparent resistivity value might be negative. Numerical methods, such as the
finite-difference and finite-element methods, have finite accuracy. For some array
configurations, the theoretical potential difference might be smaller than the error in the finite-
difference or finite-element forward modeling method. Thus the forward modeling routine
might give a negative apparent resistivity value. Although it is possible to carry out an inversion
using the apparent resistivity instead of the logarithm of the apparent resistivity, using the
logarithm is preferable as it scales the apparent resistivity values that can have a very large
linear range.
The geometric factor relative error was proposed by Wilkinson et al. (2008) to identify
cross-borehole arrays that are likely to be unstable. This gives the sensitivity of the array to
errors in the position of the electrodes. The geometric factor K for any four electrode array
located on the ground surface is given by
 1 1 1 1 
K  2 / H , H      . (8.1)
 rAM rAN rBM rBN 
The current electrodes are denoted by A and B, while the potential electrodes are M and N. rAM
is the distance between A and M. The sensitivity (s) of the geometric factor to errors in the
positions of the electrodes (Loke et al., 2014c) is given by
2 2 2 2
 K   K   K   K 
s2         . (8.2)
 A   B   M   N 
The geometric factor relative error is then defined to be
RE  s / K . (8.3)
Note while s is independent of the unit electrode spacing (such as the ‘a’ spacing for
the Wenner array) RE depends on the unit electrode spacing. To avoid this dependency, it is
proposed that a normalized geometric factor is used in the calculation of RE. In this case, the
geometric is calculated where the electrode distances are divided by the unit electrode spacing.
If a finite-difference or finite-element mesh with 4 nodes between adjacent electrode positions
is used, it is found that using an upper limit of about 5 for RE is usually sufficient to filter out
the offset type of arrays that are likely to give negative calculated apparent resistivity values.
If an even finer mesh is used, such as 8 nodes between adjacent electrode spacings, a higher
upper limit can be used at the expense of longer calculation times and greater computer memory
requirements.

159
In terms of use in actual field data inversion, the user defined model grid option
(probably with the arbitrary electrodes format) is likely to be used with the Res3dinvx64
program. This has the flexibility to adjust the x and y model grid spacings to fit the unit
electrode spacing used in the field survey. The model grid spacing should not be larger than
the unit electrode spacing used in the survey, and 4 nodes between adjacent model grid lines
should be used for the maximum accuracy. To effectively use more than 4 nodes between
adjacent electrode positions in the mesh, the user defined grid option allows using a grid
spacing that is smaller the unit electrode spacing. As an example, if a 50 meters electrode
spacing was used in the survey, and a model grid with 25 meter spacing is used, the finite-
element mesh will effectively have 8 nodes between adjacent electrode positions.
In designing the array configurations for a field survey, the geometric factor as well as
geometric factor relative error can be used to avoid arrays that are likely to be unstable. The
potential values measured by such unstable arrays are likely to be very small (and thus
overwhelmed by telluric noise), or very sensitive to errors in the electrode positions (this could
be a significant issue in field surveys in rugged and forested terrains), such that the data
measured is not meaningful and does not contribute significantly to resolving the subsurface
geology. This should be used to curb extreme field survey designs in an attempt to maximize
the data collection speed or coverage. In one data set I have come across, the arrays used have
geometric factors ranging from about 800 to 300 million m. Obviously measurements using
arrays with very large geometric factors are likely to be too noisy to be useful.
In selecting arrays to be used for field surveys, the following filtering steps are
suggested to eliminate arrays that are likely to be unstable.
1). Set an upper geometric factor limit, and remove arrays that exceed this limit. Arrays with
very large geometric factors will have very low potentials, and thus data from such arrays might
be too noisy to be useful. The upper geometric factor will depend on the characteristics of
equipment used (maximum current, signal processing features to improve signal-to-noise ratio
etc.) and the survey environment. A variation of this is to set a limit of the ratio of the maximum
geometric factor to the minimum geometric factor of the arrays in the data set to reflect the
dynamic range of signals that can be detected by the instrument.
2). Set an upper limit for the geometric factor relative error. This is to eliminate the offset type
of arrays that might have a reasonable geometric factors but likely to be unstable. A value of 4
to 10 can be used for most data sets.
In the Res3dinvx64 program, the above filtering methods are set using the ‘File-Cutoff
factor to filter data’ option. The following dialog box is shown on selecting this option.
Setting a value of 0.0001 for the geometric factor cutoff value allows the maximum geometric

160
factor to be 10,000 times the minimum geometric factor in the data points selected. This is
probably higher than the dynamic range of most resistivity-meter systems. For the geometric
factor relative error, a cutoff value of about 5 seems to work for most data sets.
3). Conduct a homogeneous earth model test using a finite-difference modeling program, such
as RES3DMODx64. In this case, calculate the apparent resistivity values for the proposed array
configurations for a homogeneous numerical model, such as with a resistivity of 100 ohm.m.
In theory, the calculated apparent resistivity values should be the same as the model resistivity
value. If the calculated apparent resistivity value differs by more than 20% from the model
value, the array is likely to be unstable. In any case, the forward modeling routine used in the
inversion cannot provide reasonably accurate apparent resistivity values for such arrays, so a
numerical interpretation is not viable using such data points.
In relation to this, some of the offset type of arrays used in 3-D surveys also use very
long distances between the current electrodes and potential dipole. This results in arrays with
very low potentials which are particularly sensitive to the boundary conditions used in the
finite-difference or finite-element forward modeling routine. In some cases, the small
inaccuracies in the boundary conditions can cause a net negative potential to be calculated
which in turn results in negative apparent resistivity values. One method to reduce the small
errors in the boundary conditions is to use more buffer nodes near the edges of the mesh used.
In the Res3dinvx64 program, this is selected using the ‘Change Settings – Forward Modeling
Settings – Change number of nodes’ option that brings up the dialog box shown below.
Using an extended boundary increases the number of nodes in the mesh, and thus increases the
computer time and memory required for the inversion of the data set. To satisfy these
requirements, a multi-core CPU and 64-bit system is required.
8.12 Not on firm foundations – inversion with shifting electrodes in 2-D and 3-D
It is implicitly assumed that the positions of the electrodes used in 2-D or 3-D surveys
are accurately known when interpreting the data set. In recent years, there has been significant
developments in geoelectrical monitoring surveys to detect temporal changes in the subsurface
(Chambers et al., 2014; Perrone et al., 2014), such as mapping of gas flows from landfills (Loke
et al., 2014a) and the monitoring of unstable slopes (Supper et al., 2014). The measurements

161
are repeated a number of times using the same set of electrodes. The electrode positions are
measured at the start of the campaign and possibly also at regular intervals. However, ground
movements sometimes occur between the times of electrode positions measurements. Thus the
electrode positions might not be accurately known for some data sets, making it necessary to
determine the shifts in the electrode positions from the resistivity data itself (Wilkinson et al.,
2015). A method was presented by Kim (2014) where both the subsurface resistivity and
electrodes positions are unknown variables to be determined by the least-squares optimization
method. The least-squares optimization equation (for a 2-D problem) is modified to the
following form.
G T
i 
R d G i  λ i V T R m V Δqi  G Ti R d g i  λ i V T R m V q i 1 (8.4)
r  J  W 
     
where q   x , G   X , V   αWx  .
z Z   βW 
     z
The combined Jacobian matrix G consists of J (the model resistivity sensitivity values), X and
Z (the sensitivity values for the changes in the x and z electrode positions) matrices.  is a
damping factor vector and g is the data misfit vector. q is the model parameters vector
consisting of r (model resistivity) and x and z (electrode position) vectors. qi is the change
in the model parameters. W is the resistivity spatial roughness filter term. Wx and Wy are the
roughness filters for the electrode position vectors. Rd and Rm are weighting matrices used by
the L1-norm inversion method (Loke et al., 2003).  and  are the relative damping factor
weights for the shifts in the electrode positions.
Kim (2014) used the perturbation method (McGillivray and Oldenburg, 1990) to
calculate the spatial Jacobian matrices X and Z. For a survey line with e electrodes it will be
necessary to recalculate the potentials by resolving the finite-element capacitance matrix
equation 2e times. While this is possible for 2-D problems, it becomes impractical for 3-D
problems where the forward modeling routine is about two to three orders of magnitude slower.
Furthermore, it is necessary to calculate the partial derivatives in 3 directions (x, y, z). A much
faster method using a modification of the adjoint-equation approach (McGillivray and
Oldenburg, 1990) to calculate the X and Z Jacobian matrices is described in Loke et al. (2015c).
For 3-D problems, the adjoint-equation method is one to two orders of magnitude faster than
the perturbation method. For 3-D problems, it is probably three orders of magnitude faster. A
very large value of the spatial damping factors (  and  ) of about 100 was also used by Kim
(2014) to reduce distortions in the electrode positions caused by near-surface resistivity
variations. However, it was found that using such large values for the  and  damping factors
tend to reduce the accuracy of the recovered electrode positions as well (Loke et al., 2015c).
A homogeneous half space is commonly used as the starting model for the optimization
algorithm. In a time-lapse survey the initial positions of the electrodes are usually accurately
measured, and can be treated as fixed parameters in the inversion of the initial data set. The
temporal changes in the subsurface resistivity are usually much less than the spatial variations
(Loke et al, 2014a). Thus the inversion model of the initial data set provides a good starting
model for the later time data sets. Loke et al. (2015c) modified the inversion algorithm to use
the model from a previous survey as the starting model which essentially eliminated distortions
in recovered positions of the x and z electrodes caused by near-surface resistivity variations.
Figure 131 shows a synthetic 2-D model used to illustrate the effect of shifts in the electrode
positions without the need to use large values for the spatial damping factors. The initial model
has 31 electrode positions with a uniform 1 m spacing on a flat surface (Figure 131a). Figure
131b shows the perturbed model with four changes. The electrode at the 5.0 m mark is shifted
0.3 m. to the right. The electrode at the 17.0 m mark is shifted 0.4 m. upwards. A 70 ohm.m

162
prism (depth of 1.0 m to the top) is added between the two existing prisms, and the 20 ohm.m
low resistivity prism is extended downwards by 0.73 m. The measured data consists of dipole-
dipole arrays with the 'a' dipole lengths ranging from 1 to 4 m., and the 'n' factor ranging from
1 to 6. This gives a total of 415 data points. Gaussian random noise (Zhou and Dahlin, 2003)
with a mean amplitude of 2.5 milliohm was added to the data before they were converted to
apparent resistivity values. The average noise level for the data set is about 1.0%. Figure 131b
shows the inversion models obtained assuming the electrodes are equally spaced on a flat
surface. The model for the perturbed data set has significant distortions (Figure 131b) below
the electrode at the 5m mark which has shifted horizontally, as well as below the 17 m mark
that has shifted vertically. Figure 131c shows the inversion model obtained when the electrodes
are allowed to shift. The model obtained from the inversion of the initial data set shown in
Figure 131a was used as the starting model. The distortions in the model have been removed
and the data misfit of 1.0% is close to the noise added. Note the upwards shift in the electrode
at the 17 m mark that matches shift in the synthetic model. In this inversion the spatial relative
damping factors  and  were set at 1.0.
Figure 132 shows a 3-D example with shifted electrodes. The initial model has the
electrodes arranged in a 29 by 13 rectangular grid with electrodes 1 meter apart on a flat surface.
The background medium has a resistivity of 100 ohm.m. There are two low (30 ohm.m) and
high (300 ohm.m) resistivity bands to demonstrate the effect of large near surface anomalies
on the inversion. There is a small near surface high resistivity rectangular prism (400 ohm.m)
and a deeper low resistivity prism (20 ohm.m). In the perturbed model, 3 of the electrodes are
shifted horizontally and 1 electrode vertically (Figure 132b). The resistivities of the small
rectangular prims are also changed to 25 and 350 ohm.m to simulate changes in the subsurface
resistivity. Voltage dependent random noise was added to the data which gave an average noise
level of about 1.1%.
Figure 131. The (a) initial and (b) perturbed synthetic test models with apparent resistivity
pseudosections and inversion models assuming a constant electrode spacing and flat surface.
(c) The inversion model obtained with the algorithm that allows the electrodes to shift.
Figure 133 shows the inversion results with fixed electrodes. The model for the initial
data set closely matches the true structure (Figure 133a) with a data misfit of 1.2% that is

163
slightly above the noise level added. The model for the perturbed data set show significant
distortions at the positions of the 4 shifted electrodes (Figure 133b) and a higher data misfit of
1.8%.
Figure 132. (a) 3-D synthetic test model with a rectangular survey grid. In the perturbed model,
the resistivities of the smaller blocks were changed from 400 and 20 ohm.m to 350 and 25
ohm.m. (b) The survey grid for the perturbed model. The four electrodes shifted are marked by
red circles. Electrode 1 was shifted 0.3 m in the x-direction, electrode 2 moved 0.3 m in the y-
direction, electrode 3 moved -0.2 m in both x and y-directions while electrode 4 was shifted
vertically upwards by 0.4 m.
Figure 133. Inversion models for the (a) initial and (b) perturbed data sets with fixed electrodes
in a rectangular grid. The true positions of the bands and prisms are marked by black lines. The
positions of the shifted electrodes are marked by small crosses in the top layer in (b).
Figure 134 shows the inversion models obtained when the electrodes positions are
allowed to change during the inversion using different values for electrodes relative damping
factor. The data misfit is similar to the added noise level with damping factors of 1.0 and 5.0
and the distortions in the resistivity structure seen in the model with fixed electrodes (Figure
134b) are removed. Using a high damping factor of 50 essentially fixes the electrode positions
(Figure 134c) giving a model similar to that with fixed electrodes (Figure 133b). However, the
recovered electrodes grid shows significant distortions with a damping factor of 1.0. It is
expanded outwards along the y-direction over the low resistivity band and compressed inwards
over the high resistivity band. This is probably because an increase in the spacing between the
electrodes reduces the measured resistance value in a similar way as a decrease in the
resistivity. The distortions are greatly reduced with a damping factor of 5.0. Figure 135a shows
the x-z surface profile along the y=10 m line that crosses electrode 4 (Figure 132b) that was
shifted upwards. The profile with a damping factor of 1.0 shows significant distortions that is
greatly reduce when it is increased to 2.5, and almost completely eliminated with a value of
5.0. Figure 136 shows the results obtained when the model obtained for the initial data set
(Figure 136a) is used as the starting model for the inversion of the perturbed data set. The

164
distortions with the lowest damping factor of 1.0 is greatly reduced (Figure 135b and Figure
136a) and almost eliminated with a damping value of 2.5 (Figure 135b). Using the initial data
set model essentially carries out the inversion using the change in the apparent resistivity values
which removes the effect of the common background resistivity structures.
Figure 134. Inversion models for the perturbed data set using different relative damping factors
for the electrodes positions vector with a homogenous half-space starting model.
Figure 135. Surface x-z profiles along the line y=10 m for inversions using a (a) homogenous
half-space and (b) initial data set starting models.
Figure 136. Inversion models for the perturbed data set using different relative damping factors
for the electrodes positions vector with the inversion model from the initial data set as the
starting model.

165
8.13 Examples of 3-D field surveys

In this section, we will take a look at the results from a couple of 3-D field surveys over
areas with complex geology.
8.13.1 Birmingham field test survey - U.K.

This field test was carried out using a multi-electrode system with 50 electrodes
commonly used for 2-D resistivity surveys. The electrodes are arranged in a 7 by 7 grid with a
unit spacing of 0.5 meter between adjacent electrodes (Figure 137a). The two remote electrodes
were placed at more than 25 meters from the grid to reduce their effects on the measured
apparent resistivity values. To reduce the survey time, the cross-diagonal survey technique was
used. The subsurface is known to be highly inhomogenous consisting of sands and gravels.
Figure 137b shows the horizontal sections of the model obtained at the 6th iteration. The two
high resistivity zones in the upper left quadrant and the lower right corner of Layer 2 are
probably gravel beds. The two low resistivity linear features at the lower edge of Layer 1 are
due to roots from a large sycamore tree just outside the survey area. The vertical extent of the
gravel bed is more clearly shown in the vertical cross-sections across the model (Figure 137c).
The inverse model shows that the subsurface resistivity distribution in this area is highly
inhomogenous and can change rapidly within a short distance. In such a situation a simpler 2-
D resistivity model (and certainly a 1-D model from conventional sounding surveys) would
probably not be sufficiently accurate. This is a rather small data set with 468 data points and
324 model cells where the inversion on a modern PC with a hex-core CPU took just 10 seconds!
Figure 137. (a) Arrangement of electrodes in the Birmingham 3-D field survey. (b) Horizontal
and (c) vertical cross-sections of the model obtained from the inversion of the Birmingham
field survey data set. The locations of observed tree roots on the ground surface are also shown.
Figure 138 shows the use of the L-curve method (Farquharson and Oldenburg, 2004)
to estimate the optimum damping factor (section 1.4) in the inversion of this data set. The L-

166
curve method estimates the data misfit and model roughness for a large range of damping factor
values such as starting from one-hundredth to 100 times a trial value. The shape of the curve
obtained by plotting the model roughness versus the data misfit usually has an ‘L’ shape with
a distinct corner (Figure 138a). The damping factor value that corresponds to the maximum
curvature point of this curve is then chosen as the optimum damping factor value (Figure 138b).
Figure 138. Example L-curve plots. (a) A plot of the model roughness versus the data misfit
for the Birmingham survey data set for a number of damping factor values. (b) A plot of the
curvature of the curve in (a) for the different damping factor values. The ‘optimum’ damping
factor is at the maximum curvature point.
8.13.2 Sludge deposit - Sweden

This survey covers a 21 by 17 grid by using a 3-D roll-along method (Dahlin and
Bernstone, 1997). To reduce the survey time, a number of parallel multi-electrode cables were
used (Figure 122). This survey was carried out at Lernacken in Southern Sweden over a closed
sludge deposit. Seven parallel multi-electrode cables were used to cover a 21 by 17 grid with a
5 meters spacing between adjacent electrodes. There were a total number of 3840 data points
in this data set.
The maximum spacing is chosen so that the survey will map structures to the maximum
depth of interest (section 2.5.2). In this case, the maximum spacing was 40 meters compared
to the total length of 100 meters along a line in the x-direction. In this survey, the cables were
initially laid out in the x-direction, and measurements were made in the x-direction. After each
set of measurements, the cables were shifted step by step in the y-direction until the end of the
grid. In surveys with large grids, such as in this example, it is common to limit the maximum
spacing for the measurements.
The model obtained from the inversion of this data set is shown in Figure 139. The
former sludge ponds containing highly contaminated ground water show up as low resistivity
zones in the top two layers (Dahlin and Bernstone, 1997). This was confirmed by chemical
analysis of samples. The low resistivity areas in the bottom two layers are due to saline water
from a nearby sea. Figure 140 shows a 3-D view of the inversion model using the Slicer-Dicer
3-D contouring program. An interesting more recent 3-D survey over a landfill site is described

167
in Chambers et al (2006).
Figure 139. The 3-D model obtained from the inversion of the Lernacken Sludge deposit
survey data set displayed as horizontal slices through the earth.
Figure 140. A 3-D view of the model obtained from the inversion of the Lernacken Sludge
deposit survey data set displayed with the Slicer/Dicer program. A vertical exaggeration factor
of 2 is used in the display to highlight the sludge ponds. Note that the color contour intervals
are arranged in a logarithmic manner with respect to the resistivity.

168
8.13.3 Aquatic survey - Panama

This is an unusual example from an aquatic survey along the Panama Canal (Figure
141). Dredging of the canal is necessary to expand it to accommodate larger vessels. The cost
of the dredging greatly depends on the nature of sub-bottom sediments and bedrock. A detailed
aquatic survey was thus conducted to determine the nature of the sediments and bedrock. This
survey was conducted using a streamer cable pulled behind a boat (Noonan and Rucker, 2011;
Rucker and Noonan, 2013). A series of quasi-parallel 2-D surveys lines were used that were
nominally spaced 25 meters apart. The dipole-dipole array with an electrode spacing of 15
meters and a total cable length of 170 meters was used. This arrangement gave a depth of
investigation of up to about 25 meters. The water depth was about 15 to 17 meters. The survey
was conducted along several lines running along the axis of the canal. Auxiliary data such as
GPS, water bathymetry and resistivity were also measured during the survey. Unlike land
surveys, it is not possible to keep the lines straight in a mobile aquatic survey. One section of
the survey area with the 2-D lines is shown in Figure 142. Thus conventional 2-D inversion of
each individual line did not give sufficiently accurate results, as a 2-D model assumes that the
line is straight. Furthermore, a 2-D model assumes that the geological structures are 2-D and
the line is perpendicular to the strike of the structures. The data from the different 2-D lines
were then collated into a single 3-D data file. The arbitrary electrodes data format (section 8.9)
was used to accommodate the non-regular arrangement of lines. As there is usually an abrupt
change in the resistivity between the water and the sub-bottom materials, a sharp boundary
corresponding to the water bottom at a depth of about 15 to 17 meters was added and the
resistivity of the first layer (corresponding to the water column) was set at 67 ohm.m. The
inversion model shows significant variations in the sub-bottom materials (the last 3 sections in
Figure 143). The lowest resistivity material generally corresponds to moderately soft tuff,
intermediate resistivity to moderately hard agglomerate and high resistivity to hard andesite
(Noonan and Rucker, 2011). Overall, the results of the 3-D inversion model agreed more
closely with the known geology from drill-cores than individual 2-D inversion models. The
example in Figure 143 has 13655 data points and 6006 model cells. The inversion of this
medium sized data set took about 13 minutes on a PC with an Intel i7 970 hex-core CPU and
24GB RAM.
Figure 141. Map of the Panama Canal region with the survey area marked (Noonan and
Rucker, 2011).

169
Figure 142. The model grid used for the Panama Canal floating electrodes survey with 20 by
20 meters cells. The location of the electrodes along the surveys lines are shown as colored
points.
Figure 143. The inversion model for the Panama Canal floating electrodes survey. The first 4
layers correspond to the water column.
8.13.4 Copper Hill - Australia

This is an interesting example of a 3-D resistivity and IP survey provided by Arctan
Services Pty. and Golden Cross Resources, Australia. Copper Hill is the oldest copper mine in
NSW, Australia. An earlier survey was conducted in 1966 using mapping, rock chip sampling,
an I.P. survey and 7 drill-holes (Chivas and Nutter, 1975; White et al., 2001). Copper porphyry
with minor gold and palladium mineralization were found to occur in structurally controlled
fractures and quartz veins. However, due to the very complex geology (Figure 144), large
differences in ore grades were found in drill-holes that were less than 200 meters apart. To map
the ore deposit more accurately, a new 3-D resistivity and I.P. survey using the offset pole-
dipole array was used. The survey covered a large (1.6 x 1.1km) area using a series of 1.6 km
lines with a spacing of 25m between adjacent electrodes. Figure 145 shows the arrangement of
the transmitter and receiver lines. Currents of up to 7 Amps were used. The entire survey took
10 days giving a total of over 7000 measurements. Further details about the survey layout and
procedures used to improve the data quality as well as to reduce the survey time are described
in the paper by White et al. (2001). Other interesting information about the mineralization at
the Copper Hill area is available at the Golden Cross Resources Ltd. web site
www.reflections.com.au/GoldenCross/.
The data was inverted with the RES3DINV program that produced a 3-D resistivity and
I.P. model for the area. The 3-D I.P. model that shows the location of the mineralized zones
more clearly is shown in Figure 146. The inversion model output from the RES3DINV program
was rearranged into a VRML format that could be read by a 3-D visualization program (please
contact Arctan Services for the details) that enables the user to display the model from any
direction.

170
Figure 144. Geological map of the Copper Hill area (White et al. 2001).
The 3-D I.P. model in Figure 146 shows two en-echelon north-south trends and two
approximately east-west trends forming an annular zone of high chargeability. The results from
existing drill-holes that had targeted the shallower part of the western zone agree well with the
resistivity and IP model. A drill-hole, CHRC58, intersected a 217m zone with 1.7 g/t gold and
0.72% copper coincided well an IP zone of greater than 35mV/V. The lower boundary of the
western zone with high chargeability coincides well with low assay results from existing drill-
holes. The eastern zone with high chargeability and resistivity values do not outcrop on the
surface and very little drilling has penetrated it. Further surveys, including drilling, is presently
being carried out.
Figure 145. Electrodes layout used for the 3-D survey of the Copper Hill area.

171
Figure 146. The I.P. model obtained from the inversion of the Copper Hill survey data set.
Yellow areas have chargeability values of greater than 35 mV/V, while red areas have
chargeability values of greater than 45 mV/V (White et al., 2001).
8.13.5 Athabasca Basin – Canada

The Athabasca basin in northern Saskatchewan (Figure 147) has over 30 deposits and
it is the richest uranium region in the world. It produces about 21% of the world’s uranium
(Raemakers et al., 2006). The uranium ore is typically found in sandstones overlying
metamorphic basement where it fills reactivated fault zones. The objective of the survey is to
detect the low resistivity alteration zone associated with the uranium deposit (Figure 148). The
survey area is a challenging environment with dry surface sand with a very thin humus layer
that makes good electrical contact difficult. With temperatures of minus 40ºC in winter, laying
out long survey lines to achieve the required depth penetration of more than 500 meters require
tremendous efforts. The example in this section comes from Bingham et al. (2006) where some
interesting choices were made in the survey design. The pole-pole array was used to reach the
required depth of penetration with the limited survey line lengths that were available. While
the pole-pole array has the poorest resolution (for depths within reach of the other arrays), here
it is choice between POOR resolution or NO resolution at depths of up to 700 m. Figure 149
shows an example of one of the survey results for the resistivity at a depth of about 200 meters.
A 3-D data set was constructed by collating data from parallel 2-D survey lines. The inversion
model shows a prominent narrow low resistivity zone that corresponds to the alteration zone.
The advantages and disadvantages of the pole-pole array is a matter of some debate,
particularly for deep surveys. Besides deep uranium surveys, it has also been used for other
mineral surveys by some companies (see for example the www.geofisicos.com.pe web site).
While there are many advocates of this array, others point out to its poor resolution compared

172
to other arrays (assuming these arrays used a sufficient long survey line to get the same
penetration depth). The pole-pole array offers field advantages of much shorter measurement
arrays and the ability to penetrate deeper than any other array for the same survey line length,
although there is the extra effort of maintaining two infinites (David Bingham, pers. comm.,
2009). The pole-dipole array has also been used for uranium exploration in the Athabasca basin,
and the 2-D and 3-D resistivity surveys played a significant role in the discovery of a very large
deposit (Donald Carriere, pers. comm., 2009) at Wheeler River near the existing McArthur
River deposit.
Figure 147. Location of uranium mines in the Athabasca basin, Saskatchewan (Bingham et al.,
2006).
Figure 148. Geological model of uranium deposit (Bingham et al., 2006).

173
Figure 149. Example of inversion model from the Midwest deposit area showing the resistivity
at a depth of about 200 meters (Bingham et al., 2006).
8.13.6 Cripple Creek and Victor Gold Mine, Colorado – USA

This example illustrates the tremendous progress made in the resistivity method over
the last 25 years, for simple 1-D models to 3-D models with complex variations in both space
and time (Loke et al., 2014a) using an unusual electrode layout. The survey was made during
a secondary recovery of gold operation by injecting sodium cyanide solution into ore rock piles
after surface leaching had ceased at the Cripple Creek and Victor Gold Mine in Colorado, USA
(Figure 150a) in September 2011 (Rucker et al., 2014). Resistivity measurements were made
to monitor the flow of the solution so as to optimize gold recovery. The measurements were
made using 48 electrodes on the ground surface arranged in a radial pattern (Figure 150b), 94
electrodes along six boreholes and 8 long electrodes using steel-case injection wells (Rucker et
al., 2014). Each snapshot took 14 minutes to complete, and a total of 780 snapshots were
acquired. The positions of the electrodes together with the surface topography are shown in
Figure 150c. Due to the use of a radial layout, the data coverage is very sparse towards the
edges of the model grid (Figure 150b). As the resistivity distribution within the ore heap is
highly inhomogeneous, the change in the resistivity is used to monitor the flow of the solution.
Figure 150c,d shows the results from one series of measurements in the form of iso-
surface contours for the -4% change in the resistivity at different times. Note the area with the
largest change is located to the north of the well. This is probably due to differences in the
subsurface permeability and structural nonuniformities within the heap created during end-
dump construction (Rucker et al., 2014). The heap was built up over the past 20 years by trucks
dumping fresh ore over the edge of older ore.

174
Figure 150. A complex time-lapse field survey example. (a) Map of Cripple Creek survey site.
(b) Overhead view of the inversion model grid with electrodes layout. (c) Iso-surface contours
for the -4% resistivity change at different times after the injection of the sodium cyanide
solution (that started at 2.8 hours from the first data set in snapshots used). t1= 1.1 hours, t2=
2.4 hours, t3= 3.7 hours, t4= 4.9 hours. (d) Overhead view of iso-surfaces.

175
8.13.7 Burra copper deposit, South Australia : Model reliability determination

This example is used to illustrate different methods of determining model reliability
(Loke, 2015). The Burra copper deposit in South Australia (Figure 151a) was discovered in
1845 and mining started in 1848 and ceased in 1877. It was at one time the largest copper mine
in Australia. It was reopened in 1971 and closed again in 1981. The primary ore is probably
chalcopyrite, with significant amounts of malachite, azurite, cuprite, chrysocolla and native
copper. Figure 151c shows some of the lines from a 1966 I.P. survey. Because of urban
development and consequent restricted access for any new survey lines, a re-interpretation of
the data was carried using modern 3-D inversion methods to glean more information from it
(Loke et al., 2013b). There is highly uneven data coverage. A model with a uniform grid using
50x50 meter cells (in the x-y directions) is used (Figure 151c).
Figure 151. Geological map of (a) south-west South Australia, (b) the Burra area, and (c) a
plot of survey electrodes and model cells layout.
The resistivity model (Figure 152a) shows a prominent north-south low resistivity
linear feature near the 1.8 km mark (x-axis) that corresponds to the Kingston Fault. The I.P.
anomaly (Figure 152b) in the northern part of the fault zone at depths of 100 to 200 meters
corresponds to the Eagle deposit prospect. The nature of the I.P. anomaly towards the bottom-
left edge of the deeper layers is uncertain as there is not much data coverage there. However it
lies in the Kingston Fault zone (Figure 151b) with reports of pyrites in a nearby bore.
Figure 153 show plots of the model resolution calculated using the resistivity and I.P.
Jacobian matrices, and the VOI using the model resistivity values. If a cutoff value of about 50
is used for the resistivity resolution index (Figure 153a), the maximum depth of investigation
is about 200 m. Not surprisingly, the highest resolution values are concentrated near the survey
lines, particularly around the group of shorter spacing lines in the northern third of the survey
area. This pattern is more pronounced in the I.P. resolution plots (Figure 153b). The I.P.
resolution sections have a shallower maximum depth of investigation than the resistivity
sections. This was confirmed by similar calculations for synthetic models with more uniform
data coverage. The VOI sections give a maximum depth of investigation of about 200 m. in the
southern half of the area below the longer survey lines (Figure 153c). The VOI plot show a
more complex pattern with local artefacts at several places and does not always increase

176
monotonically with depth (Figure 154c). The I.P. anomaly in the northern half of the survey
area lie in a region with higher resolution (and generally low VOI) values, so it is likely to be
real. The southern I.P. anomaly lie in a region with low resolution (and high VOI) values, so
its nature from the data alone is uncertain without independent confirmation.
Figure 152. Burra survey (a) resistivity and (b) I.P. inversion model layers.

177
Figure 153. The model (a) resistivity and (b) I.P. resolution index, and (c) VOI values. The red
arrows at the left side of (c) shows the position of the vertical slice shown in Figure 154.

178
Figure 154. Vertical cross-sections of the (a) resistivity model resolution index, (b) I.P
resolution index and (c) VOI in the X-Z plane 0.8 km north of the origin.
Some remarks about model reliability estimation - Some method of assessing the reliability
of the inversion results should be used. The sensitivity method (section 4.4) is the simplest but
it is difficult to pick a consistent cutoff value. The DOI/VOI method can be used for any data
set where it is possible to carry out an inversion. This plot normally has sharp boundaries that
make it easier to pick out the region with ‘reliable’ model values. A cut-off value of about 0.1
is usually used. However it is susceptible to local artefacts possibly caused by the local
optimization inversion method used. The model resolution method is more robust and less
affected by the inversion settings used. However it is computationally more demanding which
limits the size of the problem it can handle. Unlike the DOI/VOI plot, it shows a gradational
change in the resolution values that makes it more difficult to select a cut-off value.
8.14 Closing remarks on the 3-D method

A characteristic feature of 3-D surveys is the large number of electrode positions and
measurements. To achieve a sufficiently wide coverage at a lower cost, commercial surveys
are usually carried out using a series of 2-D parallel lines using multi-channel systems. For fast
computer inversion of very large survey grids, a multi-core CPU with at least 8 GBs of RAM
is needed. Recent advances in both software and hardware have now made 3-D surveys and
computer modeling within reach of small service companies. The future probably lies in
parallel survey techniques (using multi-channel systems) and parallel computing techniques
(using multi-core CPUs). The new 64-bit versions of Windows have greatly increased the
amount of usable RAM (in theory up to 192 GB, but in practice 16 to 64 GB on most off-the-
shelf modern PCs) and consequently the size of the data sets that can be processed. The extra
memory space is essential for the new inversion techniques for 3-D time-lapse surveys where
up to 100 data sets measured at different times at the same survey site using the same equipment
setup (Loke et al., 2011b, 2014a). The limiting factor now is the computer processing time.
This challenge will be met by future 8-core and 16-core CPUs (and GPUs with hundreds of
cores) together with innovative programming techniques to extract the maximum performance
from the available computer hardware (Loke et al., 2010b).

179
Acknowledgments
Dr. Torleif Dahlin of Lund University in Sweden kindly provided the Odarslov Dyke,
Lernacken and Vetlanda data sets. The Grundfor data set was provided by Dr. Niels B.
Christensen of the University of Aarhus in Denmark and Dr. Torleif Dahlin. The Rathcroghan
data set was kindly provided by Dr. Kevin Barton and Dr. Colin Brown from data collected by
the Applied Geophysics Unit of University College Galway, Ireland. Many thanks to Richard
Cromwell and Rory Retzlaff of Golder Assoc. (Seattle) for the survey example to map holes in
a clay layer. Dr. Andrew Binley of Lancaster University kindly provided the interesting cross-
borehole data set. The Bauchi data set was provided by Dr. Ian Acworth of the University of
New South Wales, Australia. The Redas river survey data set was kindly provided by Jef
Bucknix of Sage Engineering, Belgium. Mr. Julian Scott and Dr. R.D. Barker of the School of
Earth Sciences, University of Birmingham provided the Clifton, the Tar Works and the 3-D
Birmingham survey data sets. Slicer/Dicer is a registered trademark of Visualogic Inc. I would
like to thank Paul Bauman and WorleyParsons Canada for permission to use the figures for the
Athabasca oil sands case history. I would like to thank David Bingham and Grant Nimeck of
Areva Resources Canada Inc. for permission to use the figures for the Athabasca uranium
deposit case history. The challenging Hanford, Panama Canal and Cripple Creek surveys data
set were provided by Dale Rucker of HydroGeophysics Inc. (USA). I would like to thank Kim
Frankcombe and Phoenix Copper for the Burra data set. Last, but not least, permission from
Arctan Services Pty. and Gold Cross Resources to use the Copper Hill 3-D example is
gratefully acknowledged.

180
References
Acworth, R.I., 1981. The evaluation of groundwater resouces in the crystalline basement of
Northen Nigeria. Ph.D. thesis, Univ. of Birmingham.
Acworth, R. I., 1987, The development of crystalline basement aquifers in a tropical
environment: Quarterly Journal Engineering Geology, 20, 265-272.
Alumbaugh, D.L. and Newman, G.A., 2000. Image appraisal for 2-D and 3-D electromagnetic
inversion. Geophysics, 65, 1455-1467.
Tejero-Andrade, A., Cifuentes-Nava, G., Chávez, R., Lopez-Gonzalez, A. and C. Delgado-Solorzano ,
2015. L- and CORNER-arrays for 3D electric resistivity tomography : an alternative for
geophysical surveys in urban zones. Near Surface Geophysics, 13, 335-367.
Aristodemou, E and Thomas-Betts, A, 2000. DC resistivity and induced polarisation
investigations at a waste disposal site and its environments. Journal of Applied
Geophysics, 44 (2-3), 275-302
Auken, E., and Christiansen, AV., 2004. Layered and laterally constrained 2D inversion of
resistivity data. Geophysics, 69, 752-761.
Baker, H.A., Djeddi, M., Boudjadja, A.G. and K. Benhamam, 2001. A different approach in
delineating near surface buried structures. EAGE 63rd Conference & Technical
Exhibition.
Barker, R.D., 1978. The offset system of electrical resistivity sounding and its use with a
multicore cable. Geophysical Prospecting, 29, 128-143.
Barker, R.D., 1989. Depth of investigation of collinear symmetrical four-electrode arrays.
Geophysics, 54, 1031-1037.
Barker R.D., 1992. A simple algorithm for electrical imaging of the subsurface. First Break 10,
53-62.
Barker R.D., 1996. The application of electrical tomography in groundwater contamination
studies. EAGE 58th Conference and Technical Exhibition Extended Abstracts, P082.
Barker, R. and Moore, J., 1998. The application of time-lapse electrical tomography in
groundwater studies. The Leading Edge, 17, 1454-1458.
Bauman, P., 2005. 2-D resistivity surveying for hydrocarbons – A primer. CSEG Recorder,
April, 25-33.
Bentley, L.R. and Gharibi, M., 2004. Two- and three-dimensional electrical resistivity
imaging at a heterogeneous remediation site. Geophysics, 69, 674-680.
Bernstone, C. and Dahlin, T., 1999. Assessment of two automated electrical resistivity data
acquisition systems for landfill location surveys : Two case histories. Journal of
Environmental and Engineering Geophysics, 4, 113-122.
Bingham, D., Nimeck, G., Wood, G. and Mathieson, T., 2006. 3D resistivity in the Athabasca
basin with the pole-pole array. 1 day workshop - Geophysical methods and techniques
applied to uranium exploration. SEG Annual General Meeting 2006, New Orleans.
Blome, M., Maurer, H. and Greenhalgh, S., 2011. Geoelectric experimental design — Efficient
acquisition and exploitation of complete pole-bipole data sets. Geophysics, 76, F15-F26.
Carpenter, E.W. and Habberjam, G.M., 1956. A tri-potential method of resistivity prospecting.
Geophysical Prospecting, 29, 128-143.
Chambers , J.E., Kuras, O., Meldrum, P.I., Ogilvy, R.O. and Hollands, J., 2006. Electrical
resistivity tomography applied to geologic, hydrogeologic, and engineering
investigations at a former waste-disposal site. Geophysics, 71, B231-B239.
Chambers J E, Wilkinson P B, Weller A L, Ogilvy R D, Meldrum P I, and Caunt, S, 2007.
Mineshaft imaging using surface and crosshole 3D electrical resistivity tomography: A
case history from the East Pennine Coalfield, UK. Journal of Applied Geophysics, 62,
324-337.
Chambers, J.E., Gunn, D.A., Wilkinson, P.B., Meldrum, P.I., Haslam, E., Holyoake, S.,

181
Kirkham, M., Kuras, O., Merritt, A. and Wragg, J., 2014. 4D electrical resistivity
tomography monitoring of soil moisture dynamics in an operational railway
embankment, Near Surface Geophysics, 12, 61-72.
Chivas, A.R. and Nutter, A.H., 1975. Copper Hill porphyry-copper prospect. Knight, S.L. (ed.),
Economic geology of Australia and Papua New Guinea vol. I. Metals: Australasian
Institute of Mining and Meallurgy Monograph No. 5, p. 716-720.
Christensen N.B. and Sorensen K.I., 1994. Integrated use of electromagnetic methods for
hydrogeological investigations. Proceedings of the Symposium on the Application of
Geophysics to Engineering and Environmental Problems, March 1994, Boston,
Massachusetts, 163-176.
Claerbout, J.F. and Muir, F., 1973. Robust modeling with erratic data. Geophysics, 38, 826-
844.
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, 289-300.
Dahlin, T., 1996. 2D resistivity surveying for environmental and engineering applications. First
Break, 14, 275-284.
Dahlin, T., 2000. Short note on electrode charge-up effects in DC resistivity data acquisition
using multi electrode arrays. Geophysical Prospecting, 48, 181-187.
Dahlin, T. and Bernstone, C., 1997. A roll-along technique for 3D resistivity data acquisition
with multi-electrode arrays, Procs. SAGEEP’97 (Symposium on the Application of
Geophysics to Engineering and Environmental Problems), Reno, Nevada, March 23-26
1997, vol 2, 927-935.
Dahlin, T. and Loke, M.H., 1997. Quasi-3D resistivity imaging-mapping of three dimensional
structures using two dimensional DC resistivity techniques. Proceedings of the 3rd
Meeting of the Environmental and Engineering Geophysical Society. 143-146.
Dahlin,T. and Loke, M.H., 1998. Resolution of 2D Wenner resistivity imaging as assessed by
numerical modelling, Journal of Applied Geophysics, 38, 237-249.
Dahlin, T. and Owen, R., 1998. Geophysical investigations of alluvial aquifers in Zimbabwe.
Proceedings of the IV Meeting of the Environmental and Engineering Geophysical
Society (European Section), Sept. 1998, Barcelona, Spain, 151-154.
Dahlin, T., Bernstone, C. and Loke, M.H., 2002, A 3D resistivity investigation of a
contaminated site at Lernacken in Sweden. Geophysics, 60, 1682-1690.
Dahlin, T. and Zhou, B., 2004, A numerical comparison of 2D resistivity imaging with ten
electrode arrays, Geophysical prospecting, 52, 379-398.
Dahlin, T. and Zhou, B., 2006, Gradient array measurements for multi-channel 2D resistivity
imaging, Near Surface Geophysics, 4, 113-123.
Dahlin, T. and Leroux, V., 2012. Improvement in time-domain induced polarisation data
quality with multi-electrode systems by separating current and potential cables. Near
Surface Geophysics, 10, 545-565.
Dahlin, T. and Loke, M.H., 2015. Negative Apparent Chargeability in Time-domain Induced
Polarisation Data. Journal of Applied Geophysics, in press.
Daily, W. and Owen, E, 1991. Cross-borehole resistivity tomography. Geophysics, 56, 1228-
1235.
Daily, W., Ramirez, A., Newmark, R. and Masica, K., 2004. Low-cost reservoir tomographs
of electrical resistivity. The Leading Edge, May 2007, 272-280.
Daniels F. and Alberty R.A., 1966. Physical Chemistry. John Wiley and Sons, Inc.
deGroot-Hedlin, C. and Constable, S., 1990. Occam's inversion to generate smooth, two-
dimensional models form magnetotelluric data. Geophysics, 55, 1613-1624.

182
Day-Lewis, F.D., Singha K. , and Binley A., 2005. The application of petrophysical models to
radar and electrical resistivity tomograms: resolution dependent limitations. Journal of
Geophysical Research, 110, B08206, doi: 10.1029/2004JB003569, 17 p.
Dey A. and Morrison H.F. 1979a. Resistivity modelling for arbitrary shaped two-
dimensional structures. Geophysical Prospecting 27, 106-136.
Dey A. and Morrison H.F., 1979b. Resistivity modeling for arbitrarily shaped three-
dimensional shaped structures. Geophysics 44, 753-780.
Edwards L.S., 1977. A modified pseudosection for resistivity and induced-polarization.
Geophysics, 42, 1020-1036.
Ellis, R.G. and Oldenburg, D.W., 1994a, Applied geophysical inversion: Geophysical Journal
International, 116, 5-11.
Ellis, R.G. and Oldenburg, D.W., 1994b. The pole-pole 3-D DC-resistivity inverse problem : a
conjugate gradient approach. Geophys. J. Int., 119, 187-194.
Farquharson, C.G., and D.W. Oldenburg, 1998. Nonlinear inversion using general measures of
data misfit and model structure, Geophysical Journal International, 134, 213-227.
Farquharson, C.G., and D.W. Oldenburg, 2004. A comparison of automatic techniques for
estimating the regularization parameter in non-linear inverse problems. Geophysical
Journal International, 156, 411-425.
Farquharson, C.G., 2008. Constructing piecewise-constant models in multidimensional
minimum-structure inversions. Geophysics, 73, K1-K9.
Fox, R.C., Hohmann, G.W., Killpack,T.J. and Rijo, L., 1980, Topographic effects in resistivity
and induced polarization surveys. Geophysics, 45, 75-93.
Gerard, R. and Tabbagh, A., 1991. A mobile four electrodes array and its application to
electrical survey of planetary grounds at shallow depths. J. Geophys. Res., 96 (B-3),
4117-4123.
Gharibi, M. and Bentley, L.R., 2005. Resolution of 3-D electrical resistivity images from
inversions of 2-D orthogonal lines. Journal of Environmental and Engineering
Golub, G.H. and van Loan, C.F., 1989. Matrix computations. The John Hopkins Un. Press.
Gouet, O., 2007. Projet de classement au titre des articles L.341-1 et suivants du Code de
l’Environnement De L’ÎLE D’OLÉRON. Report, Ministère de l’Écologie et du
Developpement Durable, Département de la Charente-Maritime, Poitiers, 84p.
Griffiths, D.H. and Turnbull, J., 1985. A multi-electrode array for resistivity surveying. First
Break 3 (No. 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, 121-129.
Griffiths D.H. and Barker R.D.,1993. Two-dimensional resistivity imaging and modelling in
areas of complex geology. Journal of Applied Geophysics, 29, 211-226.
Hallof, P.G., 1990. Reconnaissance and detailed geophysical results, Granite Mountain Area
Pershing County, Nevada. in Fink, J.B., McAlister, E.O., Sternberg, B.K., Wieduwilt,
W.G. and Ward, S.H. (Eds), 1990, Induced polarization : Applications and case
histories : Investigations in Geophysics No. 4, SEG, 325-353.
Harro D. and Kruse S. 2013. Improved imaging of covered karst with the multi-electrode
resistivity implant technique. 13th Sinkhole Conference, NCKRI Symposium 2,
Carlsbad, New Mexico, 7 pp.
Holcombe, J. and Jirack, G., 1984. 3-D terrain corrections in resistivity surveys. Geophysics,
49, 439-452.
Inman, J.R., 1975. Resistivity inversion with ridge regression. Geophysics, 40, 798-817.
Johansson, S. and Dahlin, T., 1996. Seepage monitoring in an earth embankment dam by
repeated resistivity measurements. European Journal of Engineering and Geophysics,

183
1, 229-247.
Jung, H.K., Min, D.J., Lee, H.S., Oh, S. and Chung, H., 2009. Negative apparent resistivity in
dipole–dipole electrical surveys. Exploration Geophysics, 40, 33–40.
Keller G.V. and Frischknecht F.C.,1966. Electrical methods in geophysical prospecting.
Pergamon Press Inc., Oxford.
Kellett, R.L. and Bauman, P.D., 1999. Electrical resistivity imaging of oil sand horizons : A
modern slant on an old technique. Komex International Ltd.
Kemna, A., Rakers, E. and Binley, A.,1997. Application of Complex Resistivity Tomography
to Field Data from a Kerosene Contaminated Site, In: Proc. 3rd Meeting of the
Environmental and Engineering Geophysics Society, Aarhus, 8-11, September, 1997,
p151-154.
Kenma, A., Binley, A., Ramirez, A. and Daily, W., 2000. Complex resistivity tomography for
environmental applications. Chemical Engineering Journal, 77, 11-18.
Kim, J H., Yi, M. J., Park, S G., Kim, J.G., 2009. 4D inversion of DC resistivity monitoring
data acquired over a dynamically changing earth model. Journal of Applied
Koefoed O.,1979. Geosounding Principles 1 : Resistivity sounding measurements. Elsevier
Science Publishing Company, Amsterdam.
LaBrecque, D.J., Miletto, M., Daily, W., Ramirez, A. and Owen, E., 1996. The effects of noise
on Occam’s inversion of resistivity tomography data. Geophysics, 61, 538-548.
Lee, H., Jung, H.K., Cho, S.H. and Min, D.J. 2014. Negative Apparent Resistivity in 2.5D
Dipole-dipole Electrical Survey for a Very Simple Model. Near Surface Geoscience
2014 - 20th European Meeting of Environmental and Engineering Geophysics.
Li Y. and Oldenburg D.W. 1992. Approximate inverse mappings in DC resistivity problems.
Geophysical Journal International 109, 343-362.
Lines L.R. and Treitel S. 1984. Tutorial : A review of least-squares inversion and its application
to geophysical problems. Geophysical Prospecting, 32, 159-186.
Loke, M.H., 1994. The inversion of two-dimensional resistivity data. Unpubl. PhD thesis, Un.
Of Birmingham.
Loke, M.H., 1999. Time-lapse resistivity imaging inversion. Proceedings of the 5th Meeting of
the Environmental and Engineering Geophysical Society European Section, Em1.
Loke, M.H., 2000. Topographic modelling in resistivity imaging inversion. 62nd EAGE
Conference & Technical Exhibition Extended Abstracts, D-2.
Loke, M.H. and Barker, R.D., 1995. Least-squares deconvolution of apparent resistivity
pseudosections. Geophysics, 60, 1682-1690.
Loke, M.M., 2015. Determination of model reliability in 3-D resistivity and I.P. inversion. 24th
International Geophysical Conference and Exhibition, 15-18 February 2015 - Perth,
Australia.
Loke M.H. and Barker R.D.,1996a. Rapid least-squares inversion of apparent resistivity
pseudosections using a quasi-Newton method. Geophysical Prospecting, 44, 131-152.
Loke M.H. and Barker R.D.,1996b. Practical techniques for 3D resistivity surveys and data
inversion. Geophysical Prospecting, 44, 499-523.
Loke, M.H. and Dahlin, T., 2002. A comparison of the Gauss-Newton and quasi-Newton
methods in resistivity imaging inversion. Journal of Applied Geophysics, 49, 149-162.
Loke, M.H., Acworth, I. and Dahlin, T., 2003. A comparison of smooth and blocky inversion
methods in 2D electrical imaging surveys. Exploration Geophysics, 34, 182-187.
Loke, M.H. and Dahlin, T, 2010. Methods to reduce banding effects in 3-D resistivity
inversion. Procs. 16th European Meeting of Environmental and Engineering
Geophysics, 6 - 8 September 2010, Zurich, Switzerland, A16.
Loke, M.H., Wilkinson, P. and Chambers, J., 2010a. Fast computation of optimized electrode

184
arrays for 2D resistivity surveys. Computers & Geosciences, 36, 1414-1426.

Loke, M.H., Wilkinson, P. and Chambers, J., 2010b. Parallel computation of optimized arrays
for 2-D electrical imaging. Geophysical Journal International, 183, 1202-1315.
Loke, M.H., 2011. Electrical resistivity surveys and data interpretation. in Gupta, H (ed.), Solid
Earth Geophysics Encyclopaedia (2nd Edition) “Electrical & Electromagnetic”
Springer-Verlag, 276-283.
Loke, M.H., Chambers, J.E. and Kuras, O., 2011a. Instrumentation, electrical resistivity. in
Gupta, H (ed.), Solid Earth Geophysics Encyclopaedia (2nd Edition) “Electrical &
Electromagnetic”, Springer-Verlag, 599-604.
Loke, M.H., Dahlin, T. and Leroux , V., 2011b. Constrained time-lapse inversion of 3-D
resistivity surveys data. Near Surface 2011 - 17th European Meeting of Environmental
and Engineering Geophysics Leicester, UK, 12-14 September 2011, F06.
Loke, M.H., Chambers, J.E., Rucker, D. F., Kuras, O. and Wilkinson, P. B., 2013a. Recent
developments in the direct-current geoelectrical imaging method. Journal of Applied
Geophysics , 95, 135-156.
Loke, M.H., Frankcombe, K. and Rucker, D.F., 2013b. The inversion of data from complex 3-
D resistivity and I.P. surveys. 23rd International Geophysical Conference and
Exhibition, 11-14 August 2013 - Melbourne, Australia, 79.
Loke, M.H., Dahlin, T., Rucker, D.F., 2014a. Smoothness-constrained time-lapse inversion of
data from 3-D resistivity surveys. Near Surface Geophysics, 12, 5-24.
Loke, M.H., Wilkinson, P. B., Chambers, P. B. and Strutt, M., 2014b. Optimized arrays for 2-
D cross-borehole electrical tomography surveys. Geophysical Prospecting, 62, 172-
189.
Loke, M.H., P. B. Wilkinson, P.B., Uhlemann, S.S., Chambers, J.E. and Oxby, L. S., 2014c.
Computation of optimized arrays for 3-D electrical imaging surveys. Geophysical
Journal International, 199, 1751-1764.
Loke, M.H., P. B. Wilkinson, P.B., Chambers, J.E., Uhlemann, S.S. and Sorensen, J.P.R,
2015a. Optimized arrays for 2-D resistivity survey lines with a large number of
electrodes. Journal of Applied Geophysics, 112, 136-146.
Loke, M.H., Kiflu, H., Wilkinson, P.B., Harro, D. and Kruse, S., 2015b. Optimized arrays for
2-D resistivity surveys with combined surface and buried arrays. Near Surface
Loke, M.H., Wilkinson, P.B. and J.E. Chambers, 2015c. Rapid Inversion of Data from 2-D and
from 3-D Resistivity Surveys with Shifted Electrodes. Near Surface Geoscience
Conference, Turin, Italy, 6-10 September 2015,We 21 B11.
Loke, M.H., Wilkinson, P.B.,Tejero-Andrade, A. and Kruse, S., 2015d. Optimized Arrays for
Resistivity Measurements Confined to the Perimeter of a Survey Area. Near Surface
Geoscience Conference, Turin, Italy, 6-10 September 2015,We 21 B16.
Masne, D.L., Poirmeur, C., 1988. Three-dimensional model results for an electrical holeto-
surface method: application to the interpretation of a field survey. Geophysics, 53, 85-
103.
Marescot, L. and Loke, M.H., 2003.Using the depth of investigation index method in 2D
resistivity imaging for civil engineering surveys. SAGEEP 2003. Denver, USA.
McGillivray, P.R. and Oldenburg, D.W., 1990. Methods for calculating Frechet derivatives and
sensitivities for the non-linear inverse problem : A comparative study. Geophysical
Prospecting, 38, 499-524.
Menke, W., 1984. Geophysical data analysis : Discrete inverse theory. Academic Press Inc.
Merrick, N.P., 1997. A new resolution index for resistivity electrode arrays. Exploration
Nelson, G.D. and Van Voorhis, G.D., 1973, Letter to the editor regarding the paper “Complex

185
resistivity spectra of porphyry copper mineralization”. Geophysics, 38, 984.

Niederleithinger, E., 1994, Use of high resolution geomagnetics, 2D-DC-geoelectrics and
induced polarisation in environmental investigations. Paper presented at SAGEEP '94,
Boston, USA.
Nimmer, R. E., Osiensky, J. L., Binley, A. M. & Williams, B. C. (2008) Three-dimensional

effects causing artifacts in two-dimensional, cross-borehole, electrical imaging. Journal
of Hydrology, 359, 59-70.
Noonan, G. and D.F., Rucker, 2011. Panama Canal Expansion Project: How Marine Electrical
Resistivity was Used in Support of Canal Dredging. SAGEEP 2011, Annual meeting of
the Environmental and Engineering Geophysical Society, Charleston, SC. April 10-14,
2011.
Nyquist, J.E., Bradley, J.C. and Davis, R.K., 1999. DC resistivity monitoring of potassium
permanganate injected to oxidize TCE in situ. Journal of Environmental & Engineering
Olivar, A., de Lime, L. and Sharma, M.M., 1990. A grain conductivity approach to shaly
sandstones. Geophysics, 55, 1347-1356.
Olayinka, A.I. and Yaramanci, U., 2000, Use of block inversion in the 2-D interpretation of
apparent resistivity data and its comparison with smooth inversion: Journal of Applied
Oldenborger, G.A., Routh, P.S. and and Knoll, M.D., 2005. Sensitivity of electrical resistivity
tomography data to electrode position errors. Geophys. J. Int., 163, 1–9.
Oldenburg D.W and Li Y. 1994. Inversion of induced polarization data. Geophysics, 59, 1327-
1341.
Oldenburg, D.W. and Li., Y., 1999. Estimating depth of investigation in dc resistivity and IP
surveys. Geophysics, 64, 403-416.
Park, S.K. and Van, G.P. 1991. Inversion of pole-pole data for 3-D resistivity structures beneath
arrays of electrodes. Geophysics, 56, 951-960.
Pazdirek, O. and Blaha, V., 1996. Examples of resistivity imaging using ME-100 resistivity
field acquisition system. EAGE 58th Conference and Technical Exhibition Extended
Abstracts, Amsterdam.
Panissod, C., Dabas, M., Hesse, A., Jolivet, A., Tabbagh, J. and Tabbagh, A., 1998. Recent
developments in shallow depth electrical and electrostatic prospecting using mobile
arrays. Geophysics, 65, 1542-1550.
Pelton, W.H., Ward, S.H., Hallof, P.G., Sill, W.R. and Nelson, P.H., 1978. Mineral
discrimination and the removal of inductive coupling with multifrequency IP.
Perrone, A., Lapenna, V. and Piscitelli, S., 2014. Electrical resistivity tomography technique
for landslide investigation: A review. Earth-Science Reviews, 135, 65–82.
Ramaekers, P., Thomas, D., Cutts, C., Gandhi, S., Delaney, G. and Olson, R., 2006.
Unconformity assosciated uranoim deposits. 1 day workshop - Geophysical methods
and techniques applied to uranium exploration. SEG Annual General Meeting 2006,
New Orleans.
Ramirez, A., Newmark, R. and Daily, W., 2003. Monitoring carbon dioxide floods using
electrical resistance tomography (ERT) : Sensitivity studies. Journal of Environmental
and Engineering Geophysics, 8, 187-208.
Ritz, M., Parisot, J.-C., Diouf, S., Beauvais, A. and Dione, F., 1999. Electrical imaging of
lateritic weathering mantles over granitic and metamorphic basement of eastern
Senegal, West Africa. Journal of Applied Geophysics, 41, 335-344.
Rosqvist, R., Leroux, V., Dahlin, T., Svensson, M., Månsson, C-H. and Lindsjö, M., 2009.

186
Detektering av gas i deponier med resistivitet (in Swedish), Rapport SGC 208·1102-
7371, ISRN SGC-R-208-SE, Svenskt Gastekniskt Center, Malmö, 66p.
Roy, A. and Apparao, A, 1971. Depth of investigation in direct current methods. Geophysics,
36, 943-959.
Rucker, D.F., Levin, M.T. and Myers, D.A., 2007. Imaging beneath Hanford's Tank Farms
with electrical resistivity geophysics - An innovative approach. WM '07 Conference,
Feb. 25-March 1, 2007, Tuscon, AZ.
Rucker, D., Loke, M.H., Levitt, M.T. and Noonan, G.E., 2010. Electrical Resistivity
Characterization of an Industrial Site using Long Electrodes. Geophysics, 75, WA95-
WA104.
Rucker, D., Fink, J.B. and Loke, M.H., 2011. Environmental Monitoring of Leaks using Time
Lapsed Long Electrode Electrical Resistivity. Journal of Applied Geophysics, 74, 242-
254.
Rucker, D.F., Crook, N., Glaser, D. and Loke, M.H., 2012a. Pilot-scale field validation of the
long electrode electrical resistivity tomography method. Geophysical Prospecting, 60,
1150-1166.
Rucker, D.F. and Loke, M.H., 2012b. Optimized geoelectrical monitoring with three-
dimensional arrays, Proceedings SEG-AGU Hydrogeophysics Workshop, 8–11 July
2012, Boise, Idaho, USA.
Rucker, D.F. and Noonan, G.E., 2013. Using marine resistivity to map geotechnical properties:
a case study in support of dredging the Panama Canal. Near Surface Geophysics, 11,
625-637.
Rucker, D.F, Crook, N., Winterton, J., McNeill, M., Baldyga, C. A., Noonan, G. and Fink, J.
B., 2014. Real-time electrical monitoring of reagent delivery during a subsurface
amendment experiment. Near Surface Geophysics, 12, 151-163.
Sasaki, Y. 1992. Resolution of resistivity tomography inferred from numerical simulation.
Geophysical Prospecting, 40, 453-464.
Scott, J.B.T., Barker, R.D. Peacock, S., 2000. Combined seismic refraction and electrical
imaging. Procs. 6th Meeting of the European Association for Environmental and
Engineering Geophysics, 3-7 Sept. 1997, Bochum, Germany, EL05.
Seaton, W.J. and Burbey, T.J., 2000. Aquifer characterization in the Blue Ridge physiographic
province using resistivity profiling and borehole geophysics : Geologic analysis.
Journal of Environmental & Engineering Geophysics, 5, no. 3, 45-58.
Schenkel, C.J. and Morrison, H.F., 1994. Electrical resistivity measurement through metal
casing. Geophysics, 59, 1072-1082.
Shima, H., 1992. 2-D and 3-D resistivity imaging reconstruction using crosshole data.
Shima, H., Sakashita, S. and Kobayashi, T., 1996. Developments of non-contact data
acquisition techniques in electrical and electromagnetic explorations. Journal of
Applied Geophysics, 35, 167-173.
Silvester P.P. and Ferrari R.L., 1990. Finite elements for electrical engineers (2nd. ed.).
Cambridge University Press.
Smith, T., Hoversten, M., Gasperikova, E. and Morrison, F., 1999. Sharp boundary inversion
of 2D magnetotelluric data. Geophysical Prospecting, 47, 469-486.
Singer, B.S. and Strack, K.M.,1998. New aspects of through-casing resistivity theory,
Slater, L., Binley, A.M., Zaidman, M.D. and West, L.J., 1997, Investigation of vadose zone
flow mechanisms in unsaturated chalk using cross-borehole ERT. Proceedings of the
EEGS European Section 3rd Meeting, Aarhus, Denmark, 17-20.
Slater, L., Binley, A.M., Daily, W. and Johnson, R., 2000. Cross-hole electrical imaging of a

187
controlled saline tracer injection. Journal of Applied Geophysics, 44, 85-102.

Sorensen, K.I., 1994: The Ellog auger drilling method. Proceedings of the symposium on the
application of geophysics to engineering and environmental problems, Boston,
Massachusetts. - Environmental and Engineering Geophysical Society, 985-994.
Sorenson, K., 1996. Pulled Array Continuous Profiling. First Break, 14, 85-90.
Spiegel, R.J., Sturdivant, V.R. and Owen, T.E., 1980, Modeling resistivity anomalies from
localized voids under irregular terrain. Geophysics, 45, 1164-1183.
Spies, B.R. and Ellis, R.G., 1995. Cross-borehole resistivity tomography of a pilot-scale, in-
situ vitrification test. Geophysics, 60, 1313-1325.
Stewart M. and Parker J. 1992. Localization and seasonal variation of recharge in a covered
karst aquifer system, Florida, USA, in Hydrogeology of Selected Karst Regions, H.
Palos and W. Back, eds. International Contributions to Hydrogeology 13, International
Association of Hydrogeologists.
Stummer, P., Maurer, H. and Green, A., 2004. Experimental design:Electrical resistivity data
sets that provide optimum subsurface information. Geophysics, 69, 120-129.
Sugimoto, Y., 1999. Shallow high-resolution 2-D and 3-D electrical crosshole imaging. The
Leading Edge, 18, 1425-1428.
Sumi, F., 1965. Prospecting for non-metallic minerals by Induced Polarization. Geophysical
Prospecting, 14, 603-616.
Summer, J.S., 1976, Principles of induced polarization for geophysical exploration. Elsevier
Scientific Publishing Company.
Supper, R., Ottowitz, D., Jochum, B., Römer, A., Pfeiler, S., Kauer, S., Keuschnig, M. and Ita,
A., 2014. Geoelectrical monitoring of frozen ground and permafrost in alpine areas:
field studies and considerations towards an improved measuring technology. Near
Surface Geophysics, 12, 93-115.
Telford, W.M., Geldart, L.P. and Sheriff, R.E., 1990. Applied Geophysics (second edition).
Cambridge University Press.
Tejero-Andrade, A., Cifuentes, G., Chávez, R.E., López-González, A.E and Delgado-
Solórzano, C., 2015. L- and CORNER-arrays for 3D electric resistivity tomography: an
alternative for geophysical surveys in urban zones. Near Surface Geophysics, 13, 355-
367.
Tong, L. and Yang, C., 1990, Incorporation of topography into two-dimensional resistivity
inversion. Geophysics, 55, 354-361.
Vanhala, H., Soininen, H. and Kukkonen, I., 1992. Detecting organic chemical contaminants
by spectral-induced polarization method in glacial till environment. Geophysics, 57,
1014-1017.
Vanhalla, H. and Peltoniemi, M., 1992. Spectral IP studies of Finnish ore propects. Geophysics,
57, 1545-1555.
Van Voorhis, G.D. Nelson, P.H. and Drake, T.L., 1973, Complex resistivity spectra of
porphyry copper mineralization. Geophysics, 38, 49-60.
Waddell, J. and Barton, K, 1995, Seeing beneath Rathcroghan. Archaeology Ireland, Vol. 9,
No. 1, 38-41.
Weller, A., Seichter, M. and Kampke, A., 1996. Induced-polarization modelling using complex
electrical conductivities. Geophys. J. Int., 127, 387-398.
Weller, A., W. Frangos, and M. Seichter, 2000. Three-dimensional inversion of induced
polarization data from simulated waste, J. Appl. Geophys., 44, 67–83.
Weller, A., Breede, K., Slater, L. and Nordsiek, S., 2011. Effect of changing water salinity on
complex conductivity spectra of sandstones. Geophysics, 76, F315–F327.
White, R.M.S., Collins, S., Denne, R., Hee, R. and Brown, P., 2001. A new survey design for
3D IP modelling at Copper hill. Exploration Geophysics, 32, 152-155.

188
Wilkinson P.B., Chambers J.E., Meldrum P.I., Ogilvy R.D. and Caunt S., 2006a. Optimization
of array configurations and panel combinations for the detection and imaging of
abandoned mineshafts using 3D cross-hole electrical resistivity tomography. Journal of
Environmental and Engineering Geophysics 11, 213-221.
Wilkinson, P.B., Meldrum, P.I., Chambers, J.C., Kuras, O. and Ogilvy, R.D., 2006b. Improved
strategies for the automatic selection of optimized sets of electrical resistivity
tomography measurement configurations. Geophys. J. Int., 167, 1119-1126.
Wilkinson, P. B., Chambers, J. E., Lelliot, M., Wealthall, G. P. and Ogilvy, R. D., 2008.
Extreme Sensitivity Of Crosshole Electrical Resistivity Tomography Measurements To
Geometric Errors. Geophysical Journal International, 173, 49-62.
Wilkinson, P.B., Loke, M.H., Meldrum, P.I., Chambers, J.E., Kuras, O., Gunn, D.A., Ogilvy,
R.D., 2012, Practical aspects of applied optimized survey design for electrical
resistivity tomography, Geophysical Journal International, 189, 428-440.
Wilkinson, P.B., Uhlemann, S., Chambers, J.C., Meldrum, P. I. and Loke, M.H., 2015.
Development and testing of displacement inversion to track electrode movements on
3D Electrical Resistivity Tomography monitoring grids. Geophysical Journal
International, 200, 1566-1581.
Witherly, K.E. and Vyselaar, J, 1990. A geophysical case history of the Poplar Lake Copper-
Molybdenum deposit, Houston Area, British Columbia. in Fink, J.B., McAlister, E.O.,
Sternberg, B.K., Wieduwilt, W.G. and Ward, S.H. (Eds), 1990, Induced polarization :
Applications and case histories : Investigations in Geophysics No. 4, Soc. Expl.
Geophys.
Wolke, R. and Schwetlick, H., 1988, Iteratively reweighted least squares algorithms,
convergence analysis, and numerical comparisons: SIAM Journal of Scientific and
Statistical Computations, 9, 907-921.
Wynn, J.C., Grosz, A.E. and Carlson, V.L., 1990. Induced-polarization response of some
titanium-bearing placer deposits in Southeastern United States. In Fink, J.B., Sternberg,
B.K., McAlister, E.O., Wieduwilt, W.G. and Ward, S.H. (Eds.), Induced polarization –
Applications and case histories. Investigations in Geophysics No. 4, SEG, 280-303.
Zhao, J.X., Rijo, L. and Ward, S.H., 1986. Effects of geoelogic noise on cross-borehole
electrical surveys. Geophysics, 51, 1978-1991.
Zhdanov, M.S. and Keller, G.V., 1994. The geoelectrical methods in geophysical exploration.
Elseiver, Amsterdam.
Zhou, B. and Greenhalgh, S.A., 1997. A synthetic study on cross-hole resistivity imaging with
different electrode arrays. Exploration Geophysics, 28, 1-5.
Zhou, B. and Greenhalgh, S.A., 2000. Cross-hole resistivity tomography using different
electrode configurations. Geophysical Prospecting, 48, 887-912.
Zhou B. and Dahlin, T., 2003, Properties and effects of measurement errors on 2D resistivity
imaging surveying, Near Surface Geophysics, 1, 105-117.
Zonge, K.L. and Wynn, J.C., 1975. Recent advances and applications in complex resistivity
measurements. Geophysics, 40, 851-864.
Zhu, T. and Feng, R., 2011. Resistivity tomography with a vertical line current source and its
applications to the evaluation of residual oil saturation. Journal of Applied Geophysics,
73, 155–163.

189
Appendix A The smoothness constraint and resolving deeper

structures in I.P. surveys
A.1 A problem with I.P. inversion with a conductive overburden

A common problem in interpretation of I.P. models from field surveys is trying to find
a bottom for the deeper structures seen. This is caused by a combination of the geology, arrays
used, generally poorer resolution of I.P. data (section 8.13.7) and the nature of the smoothness
constraint. A common problem faced in some countries is a conductive overburden that limits
the amount of current that can penetrate below this layer. This by itself makes mapping of
structures in the bedrock below the overburden a challenging problem. Figure 155 shows a
synthetic model to illustrate this problem. It has a low resistivity top layer of 10 ohm.m
overlying a bedrock of 100 ohm.m. The overburden and bedrock has chargeabilities of 2 and 4
mV/V respectively. The target is a low resistivity block within the bedrock of 5 ohm.m with
chargeability of 40 mV/V. The dipole-dipole and pole-dipole arrays are widely used in I.P.
surveys to reduce the E.M. coupling effect although these arrays have relatively poor vertical
resolution. Another common practice is to use potential dipoles of the same length, i.e. a single
‘a’ spacing value with increasing ‘n’ values to increase the depth of investigation. This is done
by placing a series of potential electrodes that are equally spaced which is more convenient in
a field survey.
For the data inversion the following form of the least-squares method is used.
 
J T J  (FR   s I) Δqk  J T R d g   (FR   s I)(q k  q o ) , (A.1)
with FR  C R m C x  C R m C z .
T
x
T
z
In this equation αs is the relative weight given to the constraint such that the inversion model q
is close to a reference model q0. Note it is basically the damping factor used in the Marquardt-
Levenburg method (Line and Treitel, 1984). Normally a small value (such as 0.05) is used for
the relative damping factor weight αs. In this case the smoothness constraint in the FR roughness
matrix is the dominant constraint.
Figure 155. 2-D test model with conductive overburden.
A.2 2-D synthetic model test

For the test data set, the pole-dipole array is used with the ‘a’ spacing fixed at 10 m and
with the ‘n’ values ranging from 1 to 16. The apparent resistivity and I.P. pseudosections are
shown in Figure 155. Figure 156 shows the models obtained using different values for αs. When
a small value of 0.05 is used the block within the second layer is poorly resolved in the I.P.
model (Figure 156a). While the top of the region with high I.P. values is close to the top surface
of the block the bottom boundary is not resolved. In fact the region of high I.P. values continue
to the bottom of the model section. This is partly because of the poor vertical resolution of the

190
pole-dipole array with a constant ‘a’ spacing as well as the generally poorer resolution of the
I.P. data. The other reason is that the use of the smoothness constraint tends to produce a model
with slowly varying model values. The I.P. data does not have much information in the lower
part of the model. The high I.P. values towards the bottom are artifacts due to the use of the
smoothness constraint. Increasing the relative damping factor weight αs to 0.20 cause slight
charges in the I.P. model (Figure 156b). Using a higher value of 0.50 starts to close off the
bottom of the I.P. anomaly (Figure 156c). The constraint associated with αs forces model to be
‘close’ to a homogenous background model with a value of close to 0 mV/V. Increasing αs
further to 1.0 further improves the shape of I.P. anomaly that is closer to the true structure
(Figure 156d). In this case, equal weights are given to the model smoothness and deviations
from the constant background model. The model I.P. values below the block are reduced as
there is not much data to support higher values there and the effect of the smoothness constraint
is reduced. The same value for damping factor λ was used in all the inversions. It should be
possible to improve the results, particularly for the larger values of αs, by using an adaptive
method such as the L-curve (Farquharson and Oldenburg 2004) to automatically set this
damping factor.
Figure 156. Models obtained with different relative damping weights αs.
A.3 3-D field data set test

The Burra copper deposit data set was previously analyzed in section 8.13.7. One
prominent feature in the I.P. model (Figure 152b) was the anomaly near the lower edge of the
sections that extends from a depth of about 100 m until the deepest layer at nearly 400 m.
Figure 157 shows the results when a value of 0.5 was used for the relative damping weight αs.
The I.P. anomaly is now largely confined to a depth range of about 100 to 200 m. This shows
that the extension of I.P. anomaly into the deeper layers is largely a product of the smoothness
constraint used and the data set does not have much information on the anomaly below 200 m.
This is more clearly shown in the vertical cross-sections for the I.P. models (Figure 158) with
and without the reference model constraint.

191
Figure 157. Models for the Burra data set with a relative damping weight of 0.5 for αs.
Figure 158. I.P. vertical sections along the y-direction (at x=1650 to 1700 m) Burra data set
(a) without (αs=0.0) and (b) with (αs=0.5) a reference model constraint.

192
A.4 Software implementation

The menu options to set the weight for the reference model relative damping factor in
Res2idnvx64 is shown below.
The equivalent menu options in Res3dinvx64 are shown below.

193
Appendix B Modeling long electrodes
B.1 Two methods to model a long electrode

It is normally assumed that the length of the electrode is much smaller (less than 10%)
than the distance between the electrodes used in a survey. As such, they can be considered as
‘point’ electrodes in calculating the apparent resistivity values. In some cases, the length of the
electrode is significant and cannot be ignored. In industrial sites with metal pipes, the pipes can
be conveniently used as ‘long’ electrodes to provide extra resolution (Rucker et al., 2010, 2011,
2012a) of areas between and below the pipes. There are two approaches used in calculating the
potentials due to long electrodes. One approach is to break up the long electrodes into a series
of points electrodes, or actually an integral of the point electrode sources along the long
electrode (Masne and Poirmeur, 1988; Zhu and Feng, 2011). The advantage of this approach
is that the normal method of calculating the potentials due to a point electrode can be used, so
only a very simple change to the computer software needs to be made. The second approach is
to assign a very low resistivity value (eg. 0.1 ohm.m) to a series of the finite-difference or
finite-element mesh cells that coincide with the position of the low electrode (Ramirez et al.,
2003). This requires more changes in the computer software to identify the cells used to model
the long electrode. The main weakness of the first approach is the assumption that the current
distribution long the entire length of the electrode is constant (Zhu and Feng, 2011). Even if
the contact resistance is constant along the length of the electrode, it can be easily shown that
the method fails when the resistivity of the medium next the electrode is not constant. Figure
159 shows three situations with a long electrode in a two-layer medium, with half of the
electrode in each medium. In the first situation (Figure 159a) the two layers have the same
resistivity, so the method works well. In the second situation (Figure 159b), the upper layer has
a resistivity of 100 ohm.m, while the second layer has a lower resistivity of 10 ohm.m. This is
a fairly common near-surface situation with a drier upper soil layer above the water table. From
physical principles, we would expect more of the current to flow through the lower half of the
electrode through the low resistivity medium compared to the upper half. Thus, the assumption
of a constant current flow along the entire length of the electrode is wrong, and the resulting
calculated potentials would also be wrong. This argument can be pushed to an extreme situation
with the upper layer having a resistivity of 1,000,000 million ohm.m (essentially air) and the
lower layer having a ground resistivity of 100 ohm.m (Figure 159c). A practical situation would
be a metal pipe with half of it in air and half buried in the ground. It is known from practical
experience that practically all the current will flow into the ground through the lower half of
the pipe only, which clearly shows that the assumption of a uniform current distribution along
the entire length of the electrode is clearly wrong. The second method of simulating the long
electrode by using low resistivity mesh cells will work correctly in all three situations. The
finite-difference and finite-element methods will automatically adjust the current flow
according to resistivity of the mesh cells.
Figure 159. A long electrode in a (a) homogeneous medium, (b) two-layer medium with a low
resistivity lower layer and (c) partly in air.

194
Appendix C New I.P. systems and negative apparent resistivity

values
C.1 The distributed receivers I.P. systems and negative apparent resistivity values.
In most surveys, the measured apparent resistivity values (geometric factor multiplied
by the measured resistance) are always positive. Negative apparent resistivity values are
usually caused by noise in the measurements. However, in some situations, negative apparent
resistivity values have been observed which are not due to noise (Lee at el., 2014). It has
become more common with the use of the offset type of arrays (White et al., 2001) and non-
conventional arrangements (Figure 160) used in large scale 3-D I.P. surveys, particularly with
near-surface large resistivity contrasts and topography. In I.P. surveys, the circuit for the
current electrodes at the transmitter is always separate from the receiver dipoles. In the older
I.P. systems, it is basically impossible to determine whether the measured potential has the
same sign as the transmitted current. Thus, a crude approach has been to assume the apparent
resistivity value is always positive. This causes a problem in modeling the data when the finite-
difference or finite-element program correctly predicts that some measurements should be
negative. It is impossible to determine whether the input apparent resistivity is wrong, or it is
due to numerical errors in the forward modeling method! Fortunately, the newer I.P. systems
(such as the Iris Instruments Full-Waver system) now has synchronization features so that it
possible to determine the correct sign of the apparent resistivity values. It has been found that
5 to 10% of the measurements might have negative apparent resistivity values, particularly with
the distributed type of system (Figure 160b), and it is important to correctly identify the
negative apparent resistivity values for a more accurate resistivity inverse model. Note the sign
of the apparent I.P. values can always be correctly determined as it is measured relative to the
primary resistivity potential.
Figure 160. Example of non-conventional electrodes arrangements. (a) Offset pole-dipole

arrangement, (b) distributed pole-dipole arrangement.

2-D and 3-D Electrical Imaging Surveys

Uploaded by

Copyright:

Available Formats

2-D and 3-D Electrical Imaging Surveys

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

2-D and 3-D Electrical Imaging Surveys

Uploaded by

Copyright:

Available Formats

Tutorial : 2-D and 3-D electrical imaging surveys

email : [email protected], [email protected]

(All rights reserved)

(Revision date : 6th March 2018)

1 Introduction to resistivity surveys

1.1 Introduction and basic resistivity theory

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

might otherwise be missed by resistivity or EM surveys. Commercial SIP surveys are

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

1.2 Electrical properties of earth materials

Copyright (1996-2018) M.H.Loke

Figure 5. The resistivity of rocks, soils and minerals.

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

sounding method would not be sufficiently accurate for such situations.

Copyright (1996-2018) M.H.Loke

Table 1. 1-D inversion examples using the RES1D.EXE program.

1.4 Basic Inverse Theory

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

1.5 2-D model discretization methods

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

2 2-D electrical surveys – Data acquisition, presentation and

2.2 Field survey method - instrumentation and measurement procedure

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

2.3 Available field instruments

Manufacturer Instrument Type

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

2.4 Pseudosection data plotting method

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

2.5 A comparison of the different electrode arrays

2.5.1 The Frechet derivative for a homogeneous half-space

Copyright (1996-2018) M.H.Loke

2.5.2 A 1-D view of the sensitivity function - depth of investigation

Copyright (1996-2018) M.H.Loke

Copyright (1996-2018) M.H.Loke

Dipole-dipole n=1 0.416 0.139 18.850 0.05305 (0.3333)

Pole-dipole n=1 0.519 0.260 12.566 0.07958 (0.5000)

Pole-Pole 0.867 6.28319 0.15915 (1.0000)

Copyright (1996-2018) M.H.Loke

2.5.3 A 2-D view of the sensitivity function

and for x<0.5a,

and for x=0.5a,

Copyright (1996-2018) M.H.Loke

2.5.4 Wenner array

2.5.5 Dipole-dipole array