theadvantages$of$complementingmt$profilesin$36d...
TRANSCRIPT
1
The advantages of complementing MT profiles in 3-‐D environments with 2
geomagnetic transfer function and inter-‐station horizontal magnetic transfer 3
function data: Results from a synthetic case study 4
5
6
Joan Campanyà1, Xènia Ogaya1, Alan G. Jones1,2, Volker Rath1, Jan Vozar1 & Naser 7
Meqbel3 8
9
1Dublin Institute for Advanced Studies (DIAS), Ireland 10
2Now at: Complete MT Solutions, Ottawa, Canada 11
3GFZ German Research Centre For Geosciences, Potsdam, Germany 12
13
14
15
16
E-mail: [email protected] 17
18
19
20
21
22
23
24
25
26
Page 1 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
SUMMARY 27
As a consequence of measuring time variations of the electric and the magnetic field, 28
which are related to current flow and charge distribution, magnetotelluric (MT) data in 29
two dimensional (2-D) and three dimensional (3-D) environments are not only sensitive 30
to the geoelectrical structures below the measuring points but also to any lateral 31
anomalies surrounding the acquisition site. This behaviour complicates the 32
characterisation of the electrical resistivity distribution of the subsurface, particularly in 33
complex areas. In this manuscript we assess the main advantages of complementing the 34
standard MT impedance tensor (Z) data with inter-station horizontal magnetic tensor (H) 35
and geomagnetic transfer function (T) data in constraining the subsurface in a 3-D 36
environment beneath a MT profile. Our analysis was performed using synthetic 37
responses with added normally distributed and scattered random noise. The sensitivity of 38
each type of data to different resistivity anomalies is evaluated, showing that the degree 39
to which each site and each period is affected by the same anomaly depends on the type 40
of data. A dimensionality analysis, using Z, H and T data, identifies the presence of the 41
3-D anomalies close to the profile, suggesting a 3-D approach for recovering the 42
electrical resistivity values of the subsurface. Finally, the capacity for recovering the 43
geoelectrical structures of the subsurface is evaluated by performing joint inversion 44
using different data combinations, quantifying the differences between the true synthetic 45
model and the models from inversion process. Four main improvements are observed 46
when performing joint inversion of Z, H and T data: (1) superior precision and accuracy 47
at characterising the electrical resistivity values of the anomalies below and outside the 48
profile; (2) the potential to recover high electrical resistivity anomalies that are poorly 49
recovered using Z data alone; (3) improvement in the characterization of the bottom and 50
lateral boundaries of the anomalies with low electrical resistivity; and (4) superior 51
Page 2 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
imaging of the horizontal continuity of structures with low electrical resistivity. These 52
advantages offer new opportunities for the MT method by making the results from an 53
MT profile in a 3-D environment more convincing, supporting the possibility of high-54
resolution studies in 3-D areas without expending a large amount of economical and 55
computational resources, and also offering better resolution of targets with high 56
electrical resistivity. 57
58
1. INTRODUCTION 59
Over the last thirty years theoretical and technological developments have justified and 60
established the use of the magnetotelluric (MT) method in many regions with widely 61
differing objectives, by adjusting it to the specific requirements of each area of study 62
(e.g., Chave & Jones, 2012). Thanks to this relatively rapid progress, MT studies have 63
been undertaken from the deepest seas (e.g. Heinson et al., 2000; Baba et al., 2006; 64
Seama et al., 2007) to the highest mountain ranges on Earth (e.g., Le Pape et al., 2012; 65
Kühn et al., 2014), characterizing the geoelectrical subsurface in a large variety of 66
regions and for both academic and commercial applications (e.g., Heinson et al., 2006; 67
Tuncer et al., 2006; Falgàs et al., 2009; Jones et al., 2009; Campanyà et al., 2011; La 68
Terra & Menezes, 2012; Ogaya et al., 2014; Piña-Varas et al., 2014). However, 69
limitations of the MT method still exist and there is ample room for improvement. 70
One of the main issues hindering high resolution of the subsurface using MT is that in 3-71
D environments the data can be strongly affected by geoelectrical structures located 72
some distance away from the sites where the data are acquired, affecting most data sets 73
to at least some degree (e.g. Brasse et al., 2002), and potentially leading to inaccurate 74
characterization of the subsurface below the study area (e.g. Jones & Garcia, 2003). In 75
recent years, the availability of 3-D forward and inversion codes (e.g. Farquharson et al. 76
Page 3 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
2002; Siripunvaraporn et al. 2005a; Avdeev & Avdeeva, 2009; Egbert & Kelbert 2012; 77
Kelbert et al., 2014; Grayver, 2015; Usui, 2015) has improved the characterization of the 78
subsurface in these situations, and 3-D inversion has become a more routine procedure 79
for the EM community, although more experience is needed before it becomes as firmly 80
established as is 2-D inversion (e.g., Miensopust et al., 2013; Tietze & Ritter, 2013; 81
Kiyan et al., 2014). 82
However, difficulties associated with logistics, acquisition costs and instrumentation 83
availability still often require MT field specialists to acquire data predominantly along 2-84
D profiles, even when the study is performed in a 3-D environment. Jones (1983), 85
Wanamaker et al. (1984), Berdichevsky et al. (1998), Park & Mackie (2000), Ledo et al. 86
(2002) and Ledo (2005) studied the limitations of 2-D interpretation of 3-D MT data and 87
made various recommendations, showing some situations where 2-D inversion of 3-D 88
data can properly reproduce the geoelectrical subsurface when the content in the 2-D 89
modes is appropriately chosen for inversion. Siripunvaraporn et al. (2005b) suggest that 90
if the data are 3-D in nature but collected along a profile rather than on a grid, the results 91
from the 3-D inversion of the MT profile are superior to the results obtained by 92
performing 2-D inversion of the same data, even with the far coarser mesh employed in 93
3-D as compared to 2-D inversion. Following this idea, several publications recommend 94
inverting all four components of the MT impedance tensor (Z), e.g. Siripunvaraporn et 95
al. (2005b), Tietze & Ritter (2013) and Kiyan et al. (2014), instead of the approach of 96
inverting only the off-diagonal components (Zo) (e.g. Sasaki, 2004; Sasaki & Meju, 97
2006). However, the presence of anthropogenic noise can strongly reduce the quality of 98
the Z data, in particular the quality of the diagonal components, making it sometimes 99
worthless to use all components of the Z tensor for the inversion process because of 100
scatter and/or extremely large errors in the diagonal components of Z. 101
Page 4 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Inter-station horizontal magnetic transfer function (H) and the geomagnetic transfer 102
function (T) data, which are not affected by electric field effects caused by galvanic 103
distortion from charges on conductivity boundaries or gradients on small-scale 104
structures, have been used to improve MT results and constrain the subsurface using 2-D 105
or 3-D inversion (e.g. Soyer & Brasse, 2001, 2-D H only of field data; Gabàs & 106
Marcuello, 2003, 2-D Z + T theoretical study; Siripunvaraporn & Egbert, 2009, 3-D Z, T 107
and Z+T of synthetic and field data; Berdichevsky et al., 2010a, 2-D Z + T of field data; 108
Habibian et al., 2010, 2-D Z + T of field data; Tietze & Ritter, 2013), 3-D Z, T and Z+T 109
of synthetic and field data, Habibian & Oskooi, 2014, 2-D H cf. T theoretical study, 110
Meqbel et al., 2014, Z, T and Z+T of field data, Rao et al., 2014, Z and Z+T of field 111
data; Yang et al., 2015, Z and Z+T of field data; Varentsov, 2015a,b, performing 2D, 112
2D+ and quasi 3-D inversion using Z, T and H with theoretical and field data), but no-113
one has yet undertaken an examination of the advantages of including H and T , 114
simultaneously, when performing 3-D inversion of Z data (Z+H+T in 3-D). It should be 115
noted that the magnetic effects caused by galvanic distortion charges do affect Z, H and 116
T (e.g., Chave & Smith, 1994; Chave & Jones, 1997), however, these effects rapidly 117
decrease with increasing period due to the square root dependence of their sensitivity on 118
frequency, and can effectively be ignored over broad frequency bands. This rapid 119
decrease is true for pure galvanic distortion but not in other situations as for example in 120
the case of presence of current channelling caused by thin-long structures (e.g. Lezaeta 121
and Haak, 2003; Campanyà et al., 2012). 122
Another area for improvement for the MT geophysical technique is its issue with 123
recovering high electrical resistivity anomalies in areas with contrasting low electrical 124
resistivity values, or areas beneath strong low resistivity anomalies (e.g. Vozoff, 1991; 125
Berdichevsky et al., 1998; Bedrosian, 2007; Berdichevsky & Dmitriev, 2008; Chave & 126
Page 5 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Jones, 2012; Miensopust et al., 2013). The capacity at recovering low electrical 127
resistivity structures is a particular strength of the MT method as a geophysical 128
technique when characterizing geoelectrical structures from the first few metres to the 129
asthenosphere (e.g. Jones, 1999; Ledo et al., 2011; Rosell et al., 2011; Campanyà et al., 130
2012; Ogaya et al., 2013; Dong et al., 2014; Muñoz, 2014). However, issues in 131
recovering high electrical resistivity anomalies are somewhat of a limitation for the MT 132
method in certain resource exploration applications, for example where resistive 133
hydrocarbons are present (e.g. Simpson & Bahr, 2005; Unsworth et al., 2005) and very 134
high quality data are required. 135
136
The aim of this paper is to evaluate and highlight, from numerical experiments, the 137
advantages in combining Z (full tensor) or Zo (off-diagonal components only) responses 138
with H and T responses when characterizing and modelling the subsurface below a MT 139
profile in a 3-D environment, and to evaluate if the sensitivity and resolution issues 140
discussed above can be addressed. 141
142
2. EXPERIMENTAL DESIGN 143
The numerical experiment consisted of calculating the forward responses of a defined 144
synthetic model with embedded 3-D anomalies. These responses were used for a 145
sensitivity test, evaluating the effects of each anomaly on the different types of data, Z or 146
Zo, H and T. Dimensionality analyses and inversions were also performed characterising 147
the 3-D behaviour of the responses and recovering the electrical resistivity structures of 148
the subsurface by performing 3-D joint inversion with different combinations of Z, H 149
and T. The fit of the data and the accuracy for recovering the electrical resistivity 150
Page 6 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
distribution of the subsurface was quantified with the aim of providing a robust support 151
to the conclusions extracted from the experiment. 152
153
2.1 Synthetic Model 154
The background structure of the synthetic model (Fig. 1) is a 1-D layered Earth 155
comprised of four layers: (1) a resistivity of 100 Ωm between 0 m (the surface) and 90 m 156
depth; (2) a resistivity of 400 Ωm between 90 m and 1500 m depth; (3) a low-resistive 157
layer of 10 Ωm between 1500 m and 2000 m depth; and (4) a layer from 2000 m to 90 158
km depth with a resistivity of 200 Ωm. A half space of 20 Ωm at 90 km depth was added 159
at the bottom of the model to ensure that the EM fields at the adopted periods do not 160
penetrate away from the utilized mesh, also creating a more common response for the 161
longest periods simulating the existence of a hypothetical asthenosphere. This 1-D 162
layered background was used to simulate a more realistic and complex situation than a 163
homogeneous background, and at the same time to highlight the inherent deficiencies of 164
H and T data for resolving laterally uniform structures. 165
Within this layered background structure there are three 3-D resistivity bodies; two low 166
resistivity structures (L1 and L2) and a high resistivity body (H1) (Fig. 1). Both low and 167
high resistivity anomalies have been added to highlight the resolution differences 168
between these two types of anomalies when characterising the subsurface. L1 and L2 are 169
located between 500 m and 1000 m depth and have a resistivity value of 3 Ωm. H1 is 170
located between 1500 m and 2000 m depth, with a resistivity value of 400 Ωm. L1 and 171
H1 cross the single north-south MT profile whereas the L2 structure lies completely 172
outside the simulated linear acquisition profile to its west; this geometry was chosen 173
with the aim of evaluating the capacity of the various inversions in determining 174
structures outside the profile and to appraise if unexpected artefacts associated with L2 175
Page 7 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
may be artificially created below the profile. Due to the importance of the role of H and 176
T responses in characterizing the conducting 10 Ωm layer between 1500 m and 2000 m 177
depth, this layer has been labelled as target body L3. 178
179
The forward response of the synthetic model at the 28 sites located along the north-south 180
profile, covering a range of periods between 0.0001 and 1000 s (frequencies of 10000 – 181
0.001 Hz), was calculated using the ModEM code (Egbert & Kelbert 2012, Kelbert et 182
al., 2014), modified for inclusion on H responses (see supplementary material, S1, for 183
more details), with a large 3-D mesh that comprised 222 x 182 x 132 cells in the north-184
south, east-west and vertical directions, respectively. The core of the mesh comprised 185
162 x 112 x 132 cells with a horizontal cell size of 120 m x 120 m, while the distance 186
between sites was 600 m. The lateral size of the padding cells increased by a factor of 187
1.3 from the edges of the core outward to the boundaries. The thickness of the top layer 188
was 10 m, with an increasing factor of 1.048 for the subsequent layers in the z-direction, 189
ensuring that the thickness of each cell within the depth of interest, namely the first 3 190
km, is smaller than the horizontal cell size. 191
192
2.2 Response types 193
As mentioned above, three different EM tensor relationships were studied in our 194
numerical experiment: (1) the standard MT impedance tensor (Z) (Eq. 1); (2) the 195
geomagnetic transfer function (T) (Eq. 2); and (3) the inter-station horizontal magnetic 196
transfer function (H) (Eq. 3). 197
𝒆! = 𝒁𝒉! (1)
ℎ!! = 𝑻𝒉! (2)
𝒉! = 𝑯!"𝒉! (3)
Page 8 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
198
where 𝒉! and 𝒉! are two component vectors comprising the horizontal magnetic 199
components at the local site l ℎ!! , ℎ!! and the neighbouring reference site n ℎ!!, ℎ!! , 200
respectively. The neighbouring reference site is the site used to relate the horizontal 201
magnetic fields of all the studied sites (local sites) with the horizontal magnetic fields of 202
this site (neighbouring reference site). 𝒆! is a two component vector comprising the 203
horizontal electric components at the local site l 𝑒!! , 𝑒!! and ℎ!! is the vertical component 204
of the magnetic field recorded at the local site l. 𝒁 and 𝑯!" are 2x2 complex matrices 205
and 𝑻 is a 1x2 complex vector. In all cases dependence on frequency is assumed. 206
The H transfer function relates the horizontal magnetic field between two different sites 207
(one of the local sites and the neighbouring reference site), characterizing the differences 208
associated with the electrical resistivity distribution below the two sites. This requires 209
that the structures below the neighbouring reference site should also be characterised 210
(e.g. Soyer, 2002), which can be achieved by having the neighbouring reference site 211
within the area of study. 212
213
Note that, in case the neighbouring reference site used to acquire and process the data is 214
outside the study area, for the inversion process the H data can be rearranged using as a 215
neighbouring reference site any of the sites in the study area, even if the data at this site 216
were not acquired during the whole survey. This can be achieved following steps 217
analogous to the ones used in the ELICIT methodology to process MT data (Campanyà 218
et al., 2014), in which inputs and outputs related by the transfer functions do not need to 219
be acquired simultaneously. 220
221
Page 9 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For the experiment presented in this paper, site 26 (identified with a green arrow in Fig. 222
1) has been chosen as a neighbouring reference site for all of the H responses as there 223
are no prescribed anomalies below it and thereby facilitating simple evaluation of the 224
properties of the H responses. 225
226
2.3. Sensitivity Analysis 227
A non-linear sensitivity test was performed to identify the sites and periods responsive to 228
the geoelectrical anomalies and to characterize and quantify the effects of those 229
anomalies on the Z, H and T responses, individually and collectively. The sensitivity test 230
consisted of calculating the differences between the forward responses of the 1-D 231
layered model with the studied anomalies included (L1, L2 and H1) and the forward 232
responses of the background 1-D layered model without any anomalies. The differences 233
between the model responses were divided by the adopted error, thus highlighting how 234
sensitive the responses are to the anomalies with respect to the errors in the data. 235
Impedance error bounds were set to 5 % of |𝑍𝑖𝑗| for diagonal and off-diagonal 236
impedance tensor elements, in combination with an error floor of 5 % of 𝑍!" ∗ 𝑍!"!! 237
for the diagonal elements. Error floor was added to avoid the extremely small error 238
values associated with the periods of the Zxx and Zyy components not affected by the 3-239
D anomalies. For the T and the H responses we used constant error bounds of 0.02. 240
These errors were chosen consistent with the errors assumed in other publications with 241
both real and synthetic data (e.g. Soyer, 2002; Habibian et al., 2010; Egbert & Kelbert, 242
2012; Tietze & Ritter, 2013). Figures 2, 3, 4 and 5 show the effects of the L1, L2, H1 243
and L1+L2+H1 anomalies, respectively, individually and collectively as contoured 244
pseudo-sections, with distance along the profile on the abscissa and period, on a decadic 245
logarithm scale, along the ordinate. Differences within the assumed errors, values 246
Page 10 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
between -1 and 1, are white in colour and show the periods for the various components 247
at the sites that are insensitive to the anomalies. Therefore, for these periods the 248
anomalies cannot be resolved within the assumed precision. Dark blue and dark red 249
areas show periods at the sites that are highly sensitive to the anomaly, with anomalous 250
responses of four or more standard deviations of the assumed errors. Note the somewhat 251
simplicity of anomalous response for single features (Figs. 2-4), but the complexity of 252
the response when all features are included (Fig. 5). The total summed effect of each 253
anomaly on the responses is quantified in Fig. 6. In this figure, the sensitivities of Zo 254
(off-diagonal components of Z) are also shown, as the off-diagonal components have 255
been often used as a data type, although most practitioners now invert the full impedance 256
tensor Z. These pseudo-sections show the sensitivity of each period at each site, with 257
respect to the adopted error, between the responses of the 1-D layered model with the 258
studied anomaly and the responses of the 1-D layered model without any anomaly. The 259
sensitivity values for each period and site were calculated following equation 5: 260
261
𝑆 = 1+ (𝑎𝑏𝑠 !"#$% !"#$%&#"!!"#"!""#"
− 1) ∀ !"#$% !"#$%&#"!!"#"!""#"
> 1 (5) 262
263
where S is the sensitivity per each period and site. Note that this equation has the 264
condition that each component of the analysed tensor relationship, for each period and 265
site, is included in the summand only if difference between model response and data is 266
greater than the assumed error, which means is sensitive to the analysed anomaly. 267
268
Four main conclusions can be drawn from the sensitivity tests: 269
(1) The influence of each anomaly at a given site and at a given period depends upon the 270
types of responses used. Whereas the Z and H responses are more sensitive at the sites 271
Page 11 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
directly above the anomalies, the T responses are more sensitive at the sites located at 272
the edge of the anomalies. This can be understood as a consequence of the physics of 273
induction. For a uniform source field there is no vertical field above a laterally uniform 274
Earth, the greatest effects in T are off to the side of the anomaly (see, e.g., Jones, 1986). 275
(2) The anomalies affect a narrower range of periods for H and T responses than for Z 276
responses. For the Z responses, the anomalous effect extends to longer periods without 277
seeing the end of the effect of the anomalies for the analysed periods. This is because the 278
galvanic charges on the boundaries of conductivity contrasts, once imposed, remain in 279
place at all longer periods. There are both magnetic and electric effects of the charges 280
(see, e.g. Chave & Smith, 1994; Chave and Jones, 1997; Jones 2012) but the magnetic 281
effects drop off very quickly with increasing period for isolated anomalies. 282
(3) For the H and T responses the influence of anomaly L1 is stronger than for 283
anomalies L2 and H1, though the difference is not as large as for Z and Zo responses 284
(Fig. 6). This is because MT is primarily sensitive to elongated conductors, and in our 285
numerical experiment L1 is the longest and most conducting of the three anomalies. This 286
could result in the inversion modelling of the Zo and Z responses naturally focussing 287
more on fitting data responding to the L1 structure than to the other anomalies, as this 288
will lead to the greatest overall misfit reduction, whereas for the H and T responses the 289
L1, L2 and H1 anomalies will have similar weight during the inversion process. 290
(4) With the 3-D geoelectrical structures (L1, L2 and H1) orientated perpendicular to the 291
MT profile, the presence of sensitive periods for the Zxx, Zyy, Hxy, Hyx, Hyy and Ty 292
components implies that the analysed responses are sensitive to the 3-D nature of the 293
anomalies (Berdichevsky & Dmitriev, 2008). However, in this numerical experiment, 294
the effect of the L1 anomaly on the y component of the T responses is smaller than the 295
Page 12 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
assumed error (Fig. 2c), which means that in this situation the T responses “see” the L1 296
anomaly as a 2-D structure within the precision of the responses. 297
298
The different influences of each anomaly in relation to the examined types of responses 299
suggest that the use of H and T responses as an accompaniment to the commonly used 300
Zo and Z responses will introduce complementary information that enhances resolution 301
and restricts acceptable model space, whilst simultaneously increasing the weight to the 302
L2 and H1 anomalies during the inversion process. 303
304
2.4. Dimensionality analysis 305
The dimensionality of the synthetic model was analysed using the Z, H and T responses 306
separately (Fig. 7). Normally distributed scattered random noise was added to the 307
responses in a manner consistent with the errors assumed for the sensitivity test. In 308
particular, the added noise at each site and frequency was derived by generating random 309
numbers from the normal distribution with mean parameter (0) and standard deviation 310
(1) and multiplying this number by the data error of the responses. The same errors as 311
those assumed for the sensitivity test were assumed for the distorted responses. 312
For the Z responses, dimensionality analysis was performed using the phase tensor 313
(Caldwell et al., 2004), as this method is routinely used for evaluating the dimensionality 314
of the study area without being affected by galvanic distortion. To assess the presence of 315
3-D structures, a variant of the β criterion was used. Typically β is used to justify 2-D/3-316
D interpretation based on a threshold value (see e.g. Caldwell et al., 2004; Thiel et al. 317
2009; Booker, 2014). In this case we took into account the errors of the data, as 318
suggested by Booker (2014), interpreting all deviations from zero that are above the 319
error in determining β as a 3-D. The ellipses are each filled with a colour associated with 320
Page 13 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
the magnitude of β divided by this error. Values greater than one require the presence of 321
3-D structures (Fig. 7a). The error in β was calculated by computing a thousand β values 322
using different values of the Z components (Zxx, Zxy, Zyx and Zyy, real and imaginary) 323
randomly distributed within the total range of the assumed error. 324
For the H responses dimensionality analysis was performed following similar criteria to 325
that used by Berdichevsky et al. (2010b), plotting the variations of the H responses: H-I, 326
where I is the identity matrix. In this case 1-D structures are represented by dots 327
(Phi_max = Phi_min = 0), 2-D structures by lines (Phi_max > Phi_min = 0) and 3-D 328
structures by ellipses (Phi_max ≥ Phi_min ≠ 0). The colour represents the magnitude of 329
Phi_min divided by the error of Phi_min. Only values greater than one require the 330
presence of 3-D structures to explain the observed responses (Fig. 7b). Error in Phi_min 331
values were derived following equivalent steps to that the calculation of the error in β. 332
333
Finally, induction arrows (Schmucker, 1970), following the Parkinson criteria (i.e., the 334
real arrows generally point towards current concentrations in conducting anomalies, 335
Jones, 1986), were used to study the dimensionality of the area using the T responses 336
(Fig. 7c). For the T responses, the presence of real and imaginary arrows pointing in 337
non-parallel directions at the same period and site is associated with a 3-D structure, as 338
the imaginary induction arrow is approximately parallel to the difference between two 339
real induction arrows at consecutive periods at the same site (Marcuello et al., 2005). 340
The coloured circles show the differences between the angle of the real and the 341
imaginary arrows, for each period and site, divided by the error at constraining the angle 342
difference. Note that to define the angle difference between real and imaginary arrows 343
we computed the differences using the same and opposite directions of the real and 344
imaginary arrows, and selected the minimum angle difference value. Values greater than 345
Page 14 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
one at dividing the angle difference by the error at determining the angle difference 346
suggest the presence of 3-D structures (Fig. 7c). Errors at defining the angle difference 347
between real and imaginary arrows were derived following equivalent steps to that the 348
calculation of the error in β. 349
350
Comparing the results in Fig. 7 between all sites and periods, differences between the 351
angles of the phase tensor ellipses (removing the 90º ambiguity), the directions of the 352
arrows (T responses) and the angles of Phi_max (H responses), all of them are also 353
related to a 3-D environment. Note that for structures crossing the MT profile (L1 and 354
H1), the phase tensor and the induction arrows appear to characterize the 3-D behaviour 355
at the sites located at the edges of the anomalies, whereas looking at the variations of H, 356
sites located above of the anomalies seems to be more affected by the 3-D effect of the 357
anomalies. 358
359
2.5. 3-D Inversion of the MT Profile 360
Nine inversions were performed using different combinations of Z, Zo, H and T 361
responses employing the ModEM algorithm (Egbert & Kelbert 2012, Kelbert et al., 362
2014) modified for inclusion on H responses, (see supplementary material, S1, for more 363
details). In all cases, the same inversion parameters were used to avoid differences 364
associated with inversion set-up parameters and thus focusing only on the differences 365
related to the investigated data types. 366
367
2.5.1 Inversion parameters 368
The 3-D solution mesh, purposely chosen with a lower resolution than the forward mesh 369
described above, comprised 137 x 120 x 99 cells in the north-south, east-west and 370
Page 15 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
vertical directions, respectively. The centre of the mesh comprised 99 x 62 x 99 cells 371
with a horizontal cell size of 200 m x 200 m, while the distance between sites was 600 372
m, thus ensuring the presence of free cells between the sites. The lateral extent of the 373
padding cells increased by a factor of 1.3. The thickness of the top layer was 10 m, 374
increasing by a factor of 1.069 for the subsequent layers in the z-direction. The same 375
responses studied in the dimensionality analysis, with added noise, were used for the 376
inversion process, but only for periods between 0.0001 s (frequency of 10000 Hz) and 377
31.62 s (frequency of 0.0316 Hz), focusing on the periods affected by the anomalies and 378
avoiding the periods affected by the assumed hypothetical asthenosphere at 90 km depth. 379
380
Several tests were executed to choose the initial electrical resistivity model and the 381
smoothing parameters. The starting model was chosen after inverting the Z responses, 382
which are sensitive to lateral-vertical variations and to the actual resistivity of the 383
subsurface, with seven different half-space starting models of 80, 100, 120, 150, 200 and 384
400 Ωm, respectively. Inversion results starting with a half-space model of 120 Ωm 385
recovered the most similar electrical resistivity values to the synthetic model used to 386
generate the synthetic data, having a lower average distance between the final inversion 387
result and the original synthetic model. For all the inversions a similar nRMS was 388
obtained regardless of the different starting models. The smoothing values used for this 389
test were 0.3 in all directions. As shown below, these smoothing values do not differ 390
significantly from the smoothing values finally used during the inversion problem. 391
392
It is well known that regularization, necessary in any ill-posed inverse problem, has a 393
large influence on the results of the inversion and particularly on resolution. In the case 394
of the ModEM code we have to choose horizontal and vertical smoothing parameters. 395
Page 16 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
To take care of that, we performed additional tests to determine the appropriate 396
regularization parameters, being aware that different data types may react differently to 397
the same smoothing parameters. Inverting Z and Z+T+H several combinations of 398
horizontal and vertical smoothing parameters were executed using smoothing values 399
between 0.1 and 0.6 (always assuming the same smoothing values for north-south and 400
east-west directions). Searching for the simplest model that can fit the data, the largest 401
smoothing parameters were chosen that adequately recover the synthetic model while 402
still allowing the inversion processes to fit the data within the error bars. Based on these 403
tests, the chosen starting model for all the inversions presented in this manuscript was a 404
homogeneous half-space of 120 Ωm, and the smoothing values for all the inversions 405
were 0.4 for north-south and east-west directions, and 0.3 for the vertical direction. Note 406
that despite a smoothing of 0.5 for the north-south and east-west direction was also 407
appropriate to fit Z data, it was not appropriate when inverting the combined Z+T+H 408
responses, suggesting that the joint inversion process including all possible data requires 409
more detail in the model to fit the data. No outstanding differences were observed when 410
using 0.5 or 0.4 smoothing parameter for north-south and east-west directions when 411
performing the inversion process with Z data. In the presented inversions, the a priori 412
model was the same as the initial model (both are homogeneous), assuming that there is 413
no prior information about the electrical resistivity values of the subsurface.” 414
415
An additional test was performed to evaluate if the choose neighbouring reference site 416
for the H responses in the sensitivity test was adequate for the inversion process. Four 417
inversions with Z and H responses were performed using four different reference sites 418
for the H responses: (1) using a site close to the L2 anomaly; (2) on top of the L1 419
anomaly; (3) on top of the H1 anomaly; (4) as in the sensitivity test, on top of a non-420
Page 17 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
complex environment (almost 1-D). Results from the inversion process (Fig. 8a) show 421
that locating the reference site on top of a non-complex environment, such as the one 422
used for the sensitivity test, facilitate convergence of the inversion process (lower 423
nRMS) also obtaining the model that better recovers the original synthetic model. Figure 424
8b shows the nRMS of each inversion for each site and data type. From the results of 425
this test, it is strongly recommended not to use a reference site on top of anomalies (L1 426
and H1 in our case), in particular on top of a strong conductive anomaly such as L1. In 427
this experiment, the inversion process using a reference site on top of L1 was not able to 428
fit the data and the obtained model was the one that recovered the original synthetic 429
model worst of all (Fig. 8). The inversion process using a neighbouring reference site 430
close to L2 also does not converge to an nRMS lower than 1.2 but the model from the 431
inversion process is superior to the two cases where the neighbouring reference site was 432
on top of L1 and H1 anomalies. Note that the nRMS for the H response associated with 433
the neighbouring reference site is always close to 1 in all cases. This is because the H 434
responses compare the signal between two sites, and the signal between the 435
neighbouring site with itself is the same. The fact that it is one and not zero is because of 436
the noise we added in the responses. For the following inversions presented in this 437
manuscript, the same site as the one used in the sensitivity test was used as a reference 438
site for the H responses. 439
440
For the numerical experiment, four inversions were performed to examine and highlight 441
the improvements that can be made by complementing the Zo and Z responses with the T 442
and H responses, whilst also evaluating the necessity of using diagonal components of Z 443
in these situations. The final inversions were (1) Zo, (2) Z, (3) Zo+H+T, and (4) Z+H+T. 444
Four additional inversions, (5) Zo+H, (6) Zo+T, (7) Z+H and (8) Z+T were performed to 445
Page 18 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
differentiate the contributions of the H and T responses independently. Finally, the 446
capacity of the H and T responses to characterize the subsurface by themselves was 447
considered in inversion (9) H+T. This latter one is interesting as no electric fields are 448
involved, thus there should be primary sensitivity to lateral conductivity variations rather 449
than to the absolute values of the conductivities themselves. 450
451
2.5.2. Results 452
The results are displayed showing a north-south section of the obtained electrical 453
resistivity models beneath the MT profile and with horizontal slices of the models at 650 454
and 1550 m depths. Figure 9 shows the results of inversion (1) Zo, (2) Z, (3) Zo+H+T 455
and (4) Z+H+T, highlighting the advantages of complementing Z or Zo with H and T. 456
Figure 10 displays the results of inversions (5) Zo+T, (6) Zo+H, (7) Z+T, (8) Z+H and 457
(9) H+T demonstrating the consequences of different contributions of H and T 458
responses. The same colour scale was used in all figures. White dashed lines indicate the 459
positions of the main resistivity structures, L1, L2, L3 and H1, to facilitate the 460
comparison of the obtained inversion results with the original synthetic model. 461
2.5.2.1 Fit of the data 462
The fit of the data of each inversion process is shown in Fig. 11 plotting the summed 463
normalized root mean square (nRMS) as calculated in the ModEM code (Egbert & 464
Kelbert 2012, Kelbert et al., 2014) for each site, type of response and for each type of 465
inversion, providing valuable information on how data misfit distributes along the 466
profile for different inversions and different response types. Different colours were used 467
for each type of response: red for Zo responses, black for Z responses, blue for T 468
responses, and green for H responses. No systematic problems related to over-fitting 469
were observed with a particular response type or site locations, although some of the 470
Page 19 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
sites positioned at the edges of the profile have slightly larger nRMS than the ones 471
located within the profile. Note that the nRMSes associated with Z and Zo responses are 472
similar when only Z or Zo were used in the inversion process or when they were 473
combined with T and H responses during the inversion process. This result shows that 474
all datasets are fit to a similar level and that no remarkable decrease in resolution of a 475
particular type of response is observed when other responses are used during the 476
inversion process. 477
478
More detailed information about the fit of the data can be found in Fig. S2 479
(Supplementary Material), where differences between data and model responses divided 480
by the error were plotted for each type of inversion, component of the data, site and 481
period in a pseudo-section format. Grey areas are periods with differences lower than the 482
error; red and blue areas are periods with differences larger than the errors. 483
484
2.5.2.2 Electrical resistivity model beneath the MT profile 485
Models resulting from inversion of the Zo or Z responses (Figs. 9a1, 9a2) show the 486
existence and presence of the main anomalies but do not resolve or characterize them 487
well; in particular the resistivity values directly below L1 are poorly resolved due to the 488
shielding effects of L1. These two models appear to be unable to differentiate between 489
anomaly L1 and the conducting anomalous layer L3 that underlies L1, implying a 490
continuous horst-like structure with the L3 getting shallow to the depths of L1 in the 491
middle of the profile. Additionally, the high electrical resistivity anomaly H1 is inferred 492
but not well recovered, showing resistivity values around 80 Ωm, which are almost an 493
order of magnitude lower than the correct value (400 Ωm). Note that these values (∼80 494
Ωm) are also lower than the initial half-space model used for the inversion process (120 495
Page 20 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Ωm). The use of the H and T responses as complements to the Zo or Z responses (Figs. 496
9a3, 9a4) significantly improves resolution of all targets, conductive and resistive, 497
showing L1 and L3 structures as disconnected (Z+H+T) or almost disconnected 498
(Zo+H+T), thus better constraining the bottom of anomaly L1 and clearly recovering 499
anomaly H1 in the northern part of the profile. The use of the H and T responses also 500
better defines the north and south boundaries of anomalies L1 and H1. By 501
complementing the Zo or Z responses with the H responses, the continuity of the low 502
resistivity layer (L3) is better defined, although in none of the models is the continuity of 503
L3 correctly recovered below L1 due to the electromagnetic shielding effects of L1 to 504
any structure immediately below it. 505
506
2.5.2.3 Horizontal slices at 650 m and 1550 m depth 507
The horizontal slices through the various 3-D resistivity models at 650 m depth are 508
shown in Fig. 9b. If only the Zo responses are inverted, the model is symmetrical on both 509
sides of the profile, thus characterizing the anomalies to extend outside of the profile but 510
without indicating on which side the anomalies are located. This illustrates and 511
emphasizes that the off-diagonal components contain no directional information for off-512
profile structures. The use of diagonal components in inversion of the full Z responses 513
characterizes the dimensionality of the structures whilst locating L2 to be west of the 514
MT profile and inferring the westerly extension of L1 (compare Fig. 9b1 with Fig. 9b2). 515
From Figs. 9b3 and 9b4 it is clear that by complementing the Zo or Z responses with H 516
and T responses the correct dimensionality of the structures is recovered in both cases, 517
showing in particular a remarkable improvement when Zo responses are complemented 518
with H and T responses. Fairly equivalent models are obtained from the Zo+H+T and 519
Z+H+T responses, but with slightly improved resolution, closer to the original synthetic 520
Page 21 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
model, when using Z+H+T. The use of the H and T responses also improves the 521
characterization of L2 by showing more reliable low resistivity values than when only 522
using Zo or Z responses, even if the resistivity values are still higher than in the original 523
synthetic model. Additionally, the north and south boundaries and the east and west 524
propagation of L1 are better defined when complementing the Zo or Z responses with the 525
H and T ones. 526
527
Figure 9c shows the horizontal slices of the final 3-D models at 1550 m depth. At this 528
depth the synthetic model is a uniform 10 Ωm conducting layer (defined as anomaly L3) 529
within which a 400 Ωm resistive anomaly (H1) is embedded. Models from the 530
conventional MT, Zo and Z, responses (Figs. 9c1, 9c2), exhibit a more complex situation 531
than might be expected from the original synthetic model, somewhat suggesting the 532
presence of the H1 anomaly with the highest electrical resistivity values (∼ 100 Ωm) 533
either directly below or outside of the profile, which is still lower than the 120 Ωm of 534
the starting model used for the inversion process. By carrying out joint inversion 535
including the H and T responses, the models better define the L3 layer and more 536
accurately recover H1 below and outside the MT profile, with electrical resistivity 537
values of around 250 Ωm, which is greater than the 120 Ωm of the starting model used 538
for the inversion process (Figs. 9c3, 9c4) and closer to the 400 Ωm of the original 539
synthetic model (Fig. 9c5). 540
541
2.5.2.4 Contribution of H and T 542
Figure 10 shows the results of the inversion process combining Zo and Z responses with 543
either T or H responses separately. From Fig. 10a, results below the profile, it can be 544
seen that the inclusion of the H responses is responsible for superior differentiation 545
Page 22 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
between L1 and L3. This differentiation also occurs when inverting Z+T but not when 546
Zo is complemented with T responses (Zo+T). i.e., the use of diagonal terms of the 547
impedance tensor is key to resolving the lack of interconnection between L1 and L3. In 548
addition, the H responses contribute to better defining the continuity of the low electrical 549
resistivity L3 layer. The main improvement gained by using the T responses is superior 550
definition of the north and south boundaries of L1. 551
Figure 10b shows the results at 650 m depth when inverting the H or T responses as a 552
complement to the Zo and Z responses. In all cases, comparing with the results in Figs. 553
9b1 and 9b2, the resolution and characterization of the anomalies are improved, 554
especially in defining the L2 anomaly. The main differences between complementing 555
with either the H or T responses are observed when characterizing the propagation of 556
anomaly L1 to the west, which is constrained by the H and Z responses but not by the T 557
responses, as expected from our sensitivity tests above. Figure 10c displays the model 558
slices at 1550 m depth, showing the contribution of including the H and T responses 559
separately from each other (cf. Figs. 9c). In all cases, the use of either or both the H and 560
T responses lead to superior recovery of the H1 anomaly and show its propagation to the 561
east. Figure 10c also shows that by using the H responses as a complement to the Z and 562
Zo responses, the models better reproduce the electrical resistivity values of the original 563
synthetic model off-profile. 564
565
Figures 10a5, 10b5 and 10c5 show that the magnetic-field only H and T responses alone 566
are insufficient to characterize the subsurface by themselves. The associated models 567
locate the anomalies along the north-south and east-west directions, but have no 568
resolution in the vertical direction and do not recover the electrical resistivity values. 569
This is because the magnetic field is sensitive to lateral resistivity contrasts (both 570
Page 23 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
amplitude and position), but is insensitive to 1-D layering and to the actual resistivity 571
values of the synthetic model 572
573
2.5.2.5 Accuracy of the models 574
One way to quantify how well the models obtained from the inversion process reproduce 575
the synthetic model is to calculate the differences between them in model space. We 576
have calculated these differences with a focus on the structures below the profile, where 577
we expect to obtain higher resolution. Figure 12 shows the differences between the 578
synthetic model and the models obtained from the inversion process, using decadic 579
logarithm electrical resistivity values. Blue colours indicate results from the inversion 580
process that are more resistive than the synthetic model, red colours indicate results 581
more conductive than the synthetic model, and white colours indicate an almost-perfect 582
fit between the inversion process model and the synthetic model. Note that the colour 583
scale indicates differences between the two models that are greater than ±0.023; this 584
value is equivalent to a difference of approx. ±0.5 Ωm for electrical resistivity values of 585
10 Ωm, and a difference of approx. ±50 Ωm for electrical resistivity values of 1000 Ωm. 586
The average differences for each type of inversion are shown in Table 1. 587
588
Table 1. Average differences between the final inversion models and the original 589 synthetic model. Differences calculated using log10 electrical resistivity values. 590 Inverted
Data Zo Z Zo+H+T Z+H+T Zo+T Zo+H Z+T Z+H H+T
Difference 0.151 0.149 0.133 0.134 0.144 0.141 0.134 0.138 0.214
591
These results indicate the superior performance when reproducing the electrical 592
resistivity values of the synthetic model when the Z or Zo responses are combined with T 593
Page 24 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
and H responses. The results also corroborate that the electrical resistivity values 594
obtained when using only the T and H responses are not correctly reproduced. 595
596
By conducting a more detailed analysis of Fig. 12, the following important points 597
associated with the recovering of each anomaly become apparent: 598
1) Structure L1: The edges and bottom are better recovered when T and H responses are 599
each or both involved in the inversion process. Despite the electrical resistivity values 600
always remaining more conductive than in the synthetic true model, they are better 601
recovered when the H responses are included in the inversion process; 602
2) Structure H1: In all the inversion processes the electrical resistivity values associated 603
with H1 are less resistive than in the synthetic model, however, recovery of these 604
resistivity values, and the geometry of the anomaly, is always improved when the H and 605
T responses are each or both involved in the inversion process. 606
3) Structure L3: The electrical resistivity values are improved when the H responses are 607
included in the inversion process. Note that when the T responses are used in the 608
inversion process, the top of the anomaly is also better characterized, particularly in the 609
southern part of the profile. 610
4) Background 1-D model: Results from the inversion processes complementing Z or Zo 611
with the H responses are the ones that best recover the electrical resistivity values of the 612
background 1-D resistivity model; not only do the results show smaller differences with 613
respect to the original synthetic model, but they also show a more homogeneous (less 614
blobby) resistivity model than the other inversion results. 615
616
3. DISCUSSION 617
Focusing on the recovery of electrical resistivity structures in 3-D areas, the use of the 618
full impedance tensor responses (Z), instead of using only the off-diagonal elements 619
Page 25 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
(Zo), is highly recommended for all 3-D inversions because of the far superior capacity 620
to reproduce the dimensionality of the geoelectrical structures in the subsurface. It 621
reaffirms and further validates the conclusions drawn and emphasized previously by 622
Siripunvaraporn et al. (2005b), Tietze & Ritter (2013), and Kiyan et al. (2014). 623
However, it is not always possible to use all of the components of the MT impedance 624
tensor because diagonal components, being of lower amplitude, are far more affected by 625
the presence of noise. This requires the necessity of including robust complementary 626
information in these situations. Siripunvaraporn & Egbert (2009) highlighted the 627
advantages of using Z+T data during the 3-D inversion process, suggesting that 628
resistivity values, geometry of the anomalies, and depths are all better characterized. 629
However, not all of the studies show major advantages when performing joint inversion 630
of Z+T responses in a 3-D environment. Tietze & Ritter (2013) present a situation where 631
results from Z responses alone are more reliable than the results from Z+T responses. In 632
previous synthetic studies working with T and H responses, Habibian and Oskooi (2014) 633
show that inversion processes using either or both T and H responses have vertical 634
resolution and are able to recover the correct electrical resistivity values of the 635
subsurface, which does not happen in the experiment shown in this manuscript (Fig. 10). 636
However, in the previous study the background structure was homogeneous with 637
uniform resistivity and the same as the one used for the starting model for the inversion. 638
This configuration masks the incapacity of the H and T responses on their own to 639
recover vertical anomalies or the correct electrical resistivity values. The numerical 640
experiment design in this manuscript, using a 1-D layered model as a background, 641
highlight this deficiency of the H and T responses, showing that vertical resolution or 642
the obtained electrical resistivity values from an inversion using either or both T and H 643
responses are only reliable if additional information defining the background structures 644
Page 26 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
of the study area is added. However, complementing Z and Zo responses, T and H aids 645
superior determination of the electrical resistivity values of the subsurface, including the 646
characterisation of the absolute electrical resistivity values (see Table 1 and Fig. 12). 647
Results from Varentsov (2015a,b) combining Z, H and T responses using 2-D, 2-D+ and 648
a quasi 3-D approaches encourage the use of H responses for the inversion process, 649
improving the accuracy of the results, and better separating close spaced conductors in 650
complex geoelectrical situations. 651
652
In this work, the advantages of using T and H responses as a complement to Z or Zo 653
responses in a 3-D environment are evaluated from the results of a numerical 654
experiment. Starting with the sensitivity of each response type to the studied anomalies, 655
the results of our numerical experiment exhibit that both the H and T responses are also 656
sensitive to the dimensionality of the structures and can be used to successfully 657
reproduce the dimensionality of the anomalies when complementing Zo or Z responses 658
(Figs. 9 and 10). The sensitivity tests performed (Fig. 6) demonstrate that the Zo and Z 659
responses are intrinsically more sensitive to the existence of the anomalies than are the 660
H and T responses. However, the dimensionality analysis (Fig. 7) and the capacity for 661
recovering the detail of the geoelectrical structures (Figs. 9, 10 and 12) are remarkably 662
improved when the Zo or Z responses are complemented with either, and especially both, 663
the H and T responses. Each response type has a specific behaviour when characterizing 664
the dimensionality of the structures and consistency between all of them will support the 665
ultimately adopted dimensionality of the study area. The use of the T and H responses 666
can also improve the dimensionality analysis in situations where the phase tensor is not a 667
strong tool; see for example the results close to the L2 and H1 anomalies (Fig. 7). In that 668
case the phase tensors do not correctly define the presence of 3-D structures having no 669
Page 27 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
periods showing absolute values of the β angle divided by the associated error greater 670
than 1. However, by using the H and T responses the characterisation of 3-D structures 671
of these areas is significantly improved. Comparing the dimensionality analysis from the 672
H and T responses close to resistive anomaly H1, induction arrows clearly point in 673
different directions than the direction of Phi_max from the H responses, which is also 674
suggestive of a 3-D situation. 675
676
Focusing now on the ability of recovering the correct electrical resistivity distribution of 677
the subsurface, the observed improvements are associated with the use of different types 678
of responses that better restrict the model space than if any are used individually. More 679
specifically, this phenomenon occurs because the Z responses are far more sensitive to 680
the L1 anomaly than to the L2 and H1 anomalies, whereas for our model the T and H 681
responses have similar sensitivity to all of the anomalies (Fig. 6). Such sensitivity means 682
that the inversion of the Zo or Z responses focuses mainly on L1, with less importance 683
placed upon the L2 and H1 anomalies. Results from the sensitivity tests (Figs. 2, 3, 4 684
and 5) can also be used to understand how inclusion of the H and T responses improves 685
the characterization of the subsurface. 686
687
Inspecting the resolved structures beneath the MT profile (Figs. 9, 10 and 12), four main 688
improvements are apparent when performing joint inversion of the Zo or Z responses 689
with H and T ones: 690
(1) Characterization of the bottom of the L1 anomaly with better differentiation between 691
the L1 and L3 structures; 692
(2) The possibility of recovering the high resistivity anomaly, H1, surrounded by low 693
resistivity values, L3, that are poorly recovered when using only Zo or Z responses; 694
Page 28 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
(3) Better resolution in determining the resistivity values of the subsurface, obtaining 695
results more similar to the original synthetic model; 696
(4) Superior definition of the shape of the anomalies in the north-south direction, i.e. 697
along profile, with a better fit inside the white dashed lines in the figures. 698
699
Despite all these advantages, we would like to point out that in all models the location of 700
the L3 layer at the edges of the profile is somewhat shallower than what it should be. 701
This problem is not properly corrected when complementing Z or Zo with T and H 702
responses, although results with T responses seem to slightly improve the depth of the 703
conductive layer L3 (Figs. 9, 10 and 12). 704
705
Looking outside the line of the profile (Figs. 9b, 9c and 10b, 10c), L2 and H1 are better 706
recovered when the inversion is complemented with H and T responses. However, 707
although the characterization of the off-profile anomalies is improved when 708
complementing the Z responses with the H and T ones, the shape of the anomalies and 709
the electrical resistivity values are still not properly constrained, probably as a 710
consequence of a combination of (i) the non-uniqueness of the MT method, and (ii) 711
scatter and error. 712
713
Inspecting the results obtained from the T responses, the main signal associated with the 714
anomalies is located at the edge of the structures (Figs. 2, 3, 4, 5, 6), making the T 715
responses a good type of data to define boundaries of the structures along the MT 716
profile. Note that from Fig. 6d this edge-detection is more precise for low resistivity 717
anomalies (L1) than for high resistivity anomalies (H1). This can also be observed when 718
recovering the anomalies from the inversion process (Figs. 9 and 10) where the north 719
Page 29 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
and south boundaries of L1 and H1 are improved when the T responses are included. 720
This result, highlighting the capacity of the T responses characterising the edges of the 721
anomalies, is consistent with previous results, such as Siripunvaraporn & Egbert (2009) 722
and Habibian and Oskooi (2014). In addition, the top of the low electrical resistivity 723
layer L3 is improved when the inversion of Z or Zo data is complemented with T data 724
(Figs. 10a1, 10a3, 12c and 12d). This makes T data a good complement to improve the 725
definition in recovering the edges and shape of the anomalies below a profile located in 726
a 3-D environment, in agreement with results suggested by Siripunvaraporn & Egbert 727
(2009). Note that, apart from the H1 anomaly, no major improvement associated with 728
the use of T responses has been observed in defining the electrical resistivity values of 729
the anomalies or the background 1-D model (Figs. 12c, 12d). Results from Fig. 10 730
shows that, although the use of T data as a complement of Zo provides an improvement 731
in characterizing the subsurface, the results are remarkably better when Z and T 732
responses are both used in the inversion process. 733
734
Examining the results from the H responses, the major advantage is that their use makes 735
the resistivity model more consistent as a whole as all the sites are related to the 736
neighbouring reference site and the intensity of the anomalies tend to be consistent along 737
the whole profile. This effect is also observed in the sensitivity test (Figs. 4b and 6c). In 738
this case the neighbouring reference site is located close to H1 and is affected by its 739
presence, which is why the effect of H1 can be seen at sites south of the profile. This 740
property of H responses increases the consistency of the model as a whole for shallow 741
and deep structures independently of the extension of the area of the study. This level of 742
consistency is not required when using the Zo, Z or T responses, where each site is 743
mostly focused on the anomalies nearby but without any explicit link to anomalies that 744
Page 30 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
are located far away along the study area, where the periods of the site are insensitive. 745
The consequences of this property in the inversion process can be seen in Fig. 12, where 746
the models including the H responses obtain electrical resistivity values of the anomalies 747
and the background 1-D resistivity model closer to the original synthetic model. Note 748
that from Fig. 10, although results are better quality when Z+H are used in the inversion 749
process, instead of Zo+H, differences are not as remarkable as the differences observed 750
between Z+T and Zo+T, having similar results from both types of inversion. Note also 751
that, from the performed test shown in Fig. 8, to facilitate and improve the results from 752
the inversion process using H responses it is recommended to use a neighbouring 753
reference site located on top of a non-complex geoelectrical area. Results from the 754
dimensionality analysis or from prior inversions using Z alone or Z+T can be used to 755
choose the optimum neighbouring reference site for Z+H or Z+T+H inversion. 756
757
4. CONCLUSIONS 758
759
The use of the H and T responses is demonstrated to be a crucial complement to the Zo 760
or Z ones in order to better resolve and characterize the structures beneath MT profiles 761
in 3-D environments. For the synthetic model considered here, large differences were 762
seen whether the diagonal components of Z were used or not, having far better results 763
when the diagonal components were included. These differences were reduced when 764
including the H and T responses in the inversion process, although results using the 765
diagonal components of Z still recover the subsurface better than when the diagonal 766
components are not used. From the results of this experiment, four main improvements 767
are observed when carrying out joint inversion of the Zo or Z responses with the H and T 768
ones: 769
Page 31 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
(1) More accurate characterization of the electrical resistivity values of the studied 770
anomalies and also of the background resistivity model; 771
(2) Superior capability for recovering high electrical resistivity anomalies than by using 772
the Zo or Z responses alone; 773
(3) Improved definition of the bottom of low electrical resistivity anomalies and the 774
north, south, east and west boundaries of the anomalies beneath and outside the MT 775
profile; and 776
(4) Superior imaging of the horizontal continuity of low electrical resistivity anomalies. 777
Additionally, note that the use of the H responses increases the reliability of the 778
resistivity model as a whole, reducing the possibility for the inverse problem to generate 779
spurious anomalies if the anomaly is not consistent with the data acquired at the 780
neighbouring reference site. Although any site could be used as a neighbouring reference 781
site, as far as the structures below the neighbouring reference site are properly 782
constrained, it is recommended to use a site located on top of a simple geoelectrical area 783
and it is particularly strongly recommended not to use a site on top of a strong anomaly. 784
785
All of these advantages incentivise the use of H and T responses as complements Z or Zo 786
responses, in particular when characterizing the subsurface beneath a MT profile in a 3-787
D environment. This supports the possibility of high-resolution studies in 3-D 788
environments without expending large amounts of economical and computational 789
resources, and also opening the door for the achievement of targets with high electrical 790
resistivity values. 791
792
5. ACKNOWLEDGMENTS 793
Page 32 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
We would like to acknowledge the financial support to JC and XO from the IRECCSEM 794
project (www.ireccsem.ie) funded by a Science Foundation of Ireland Investigator 795
Project Award (SFI: 12/IP/1313) to AGJ. Gary Egbert, Anna Kelbert and Naser Meqbel 796
are very gratefully thanked for making their ModEM code available to the community. 797
The authors wish to acknowledge Irish Centre for High-End Computing (ICHEC) for the 798
computing facilities provided. Finally, the authors would like to thank Stephan Thiel, the 799
two anonymous reviewers, and the Associate Editors Mark Everett and Ute Weckmann 800
for their very helpful suggestions and comments on our submitted versions that 801
improved our final paper. 802
803
804
6. REFERENCES 805
1. Avdeev, D.B. & Avdeeva, A., 2009. 3D magnetotelluric inversion using a 806
limited-memory quasi-Newton optimization, Geophysics, 74(3), F45– F57. 807
2. Baba, K., Chave, A.D., Evans, R.L., Hirth, G. & Mackie, R.L., 2006. Mantle 808
dynamics beneath the East Pacific Rise at 17S: insights from the Mantle 809
Electomagnetic and Tomography (MELT) experiment, J. Geophys. Res., 111, 810
B02101, doi:10.1029/2004JB003598. 811
3. Bedrosian, P.A., 2007. MT+, Integrating Magnetotellurics to Determine Earth 812
Structure, Physical State, and Processes. Surv. Geophys, 28: 121-167. doi: 813
10.1007/s10712-007-9019-6 814
4. Berdichevsky, M.N., Dmitriev, V. & Pozdnjakova, E.E., 1998. On two-815
dimensional interpretation of magnetotelluric soundings, Geophys. J. Int., 133, 816
585–606. 817
5. Berdichevsky M.N. & Dmitriev V.I., 2008. Models and Methods of 818
Magnetotellurics.Springer-Verlag Berlin Heidelberg. 819
6. Berdichevsky, M.N., Golubtsova, N.S., Varentsov, Iv. M., Pushkarev, P.Yu., 820
Rybin, A.K. & Sokolova, E. Yu., 2010a. Geoelectric section of the Central Tien 821
Shan: Sequential inversion of the magnetovariational and magnetotelluric data 822
Page 33 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
along the Naryn Line, Physics of the Solid Earth, 46(8), 698-706. 823
doi:10.1134/S1069351310080069. 824
7. Berdichevsky, M.N., Sokolova, E.Yu., Varentsov, Iv.,M., Rybin, A.K., 825
Baglaenko, N.V., Batalev, V.Yu., Golubtsova, N.S., Matayukov, V.E. & 826
Pushkarev, P.Yu., 2010b. Geoelectric section of the Central Tien Shan: Analysis 827
of magnetotelluric and magnetovariational Responses along the Naryn 828
Geotransverse, Physics of the Solid Earth, 46(8), 679-697. 829
doi:10.1134/S1069351310080057. 830
8. Booker, J.R., 2014. The Magnetotelluric Phase Tensor: A Critical Review. Surv. 831
Geophys. 35,7-40. doi:10.1007/s10712-013-9234-2 832
9. Brasse, H., P. Lezaeta, V. Rath, K. Schwalenberg, W. Soyer, & V. Haak, 2002. 833
The Bolivian Altiplano conductivity anomaly, J. Geophys. Res., 107(B5), 2096, 834
doi:10.1029/2001JB000391 835
10. Caldwell, G.T., Bibby, H.M., Brown, C., 2004. The Magnetotelluric Phase 836
Tensor, Geophys. J. Int., 158, 457–469 837
11. Campanyà, J., Ledo, J., Queralt, P., Marcuello, A., Liesa, M. & Muñoz, J.A. 838
2011. Lithospheric characterization of the Central Pyrenees based on new 839
magnetotelluric data. Terra Nova. 23, 213 - 219. doi: 10.1111/j.1365-840
3121.2011.01001.x . 841
12. Campanyà, J., Ledo, J., Queralt, P., Marcuello, A., Liesa, M. & Muñoz, J.A. 842
2012. New geoelectrical characterization of a continental collision zone in the 843
West- Central Pyrenees: Constraints from long period and broadband 844
magnetotellurics. Earth Planet. Sci. Lett., 333-334, 112-845
121.doi:10.1016/j.epsl.2012.018 846
13. Campanyà, J., J. Ledo, P. Queralt, A. Marcuello, & A.G. Jones, 2014. A new 847
methodology to estimate magnetotelluric (MT) tensor relationships: Estimation 848
of Local transfer-functIons by Combining Inter-station Transfer-functions 849
(ELICIT), Geophy. J. Int., 198, 484-494, doi: 10.1093/gji/ggu147. 850
14. Chave, A.D. & Jones, A.G., 1997. Electric and magnetic field galvanic distortion 851
decomposition of BC87 data, Journal of Geomagnetism and Geoelectricity, 49, 852
767-789. 853
15. Chave, A.D. & Jones, A.G., 2012. The Magnetotelluric Method. Cambridge 854
University press. United Kingdom. 855
Page 34 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
16. Chave, A.D. & Smith, J.T., 1994. On electric and magnetic galvanic distortion 856
tensor decompositions, Journal of Geophysical Research-Solid Earth, 99, 4669-857
4682. 858
17. Dong, H., Wei, W., Gaofeng, Y., Jin, S., Jones, A.G., Jing, J., Zhang, L., Xie, C., 859
Zhang, F. & Wang, H., 2014. Three-dimensional electrical structure of the crust 860
and upper mantle in Ordos Block and adjacent area: Evidence of regional 861
lithospheric modifications, Geochem. Geophys. Geosyst., 15, 862
doi:10.1002/2014GC005270. 863
18. Egbert, G. D., & Kelbert, A., 2012. Computational recipes for electromagnetic 864
inverse problems, Geophys. J. Intern., 189(1), 251-267. 865
19. Falgàs, E., Ledo, J., Marcuello, A. & Queralt., P., 2009. Monitoring freshwater-866
seawater interface dynamics with audiomagnetotellurics data, Near Surface 867
Geophysics,7 (5-6), 391-399. 868
20. Farquharson, C., Oldenburg, D., Haber, E. & Shekhtman, R., 2002. An al- 869
gorithm for the three-dimensional inversion of magnetotelluric data, SEG 870
Technical Program Expanded Abstracts, Society of Exploration Geo- physicists. 871
21. Gabàs, A. & Marcuello, A., 2003. The relative influence of different types of 872
magnetotelluric data on joint inversions. Earth, Planets and Space. 55 (5), 243- 873
248. 874
22. Grayver, A.V., 2015. Parallel three-dimensional magnetotelluric inversion using 875
adaptive finite-element method. Part I: theory and synthetic study, Geophysical 876
J. Int., 202, 584-603. 877
23. Habibian, B.D.; Brasse, H.; Oskooi, B.; Ernst, T.; Sokolova, E.; Varentsov, Iv. & 878
EMTESZ Working Group, 2010. The conductivity structures across the Trans-879
European Suture Zone from magnetotelluric and magnetovariational data 880
modeling. Phys. Earth Planet.Int., 183, 377-386. 881
24. Habibian, B.D. & Oskooi, B., 2014. A resolution comparison of horizontal and 882
vertical magnetic transfer functions. Journal of the Earth and Space Physics, 40 883
(3), 47-53. 884
25. Heinson, G.S.; Constable, S.C. & White, A., 2000. Episodic melt transport at 885
mid-ocean ridges inferred from magnetotelluric sounding. Geophy. Res. Lett., 27, 886
2317- 2320. 887
Page 35 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
26. Heinson, G.S., Direen, N.G., Gill, R.M., 2006. Magnetotelluric evidence for a 888
deep crustal mineralizing system beneath the Olympic Dam iron oxide copper-889
gold deposit, southern Australia. Geology, 34, 573-576, doi:10.1130/G22222.1. 890
27. Jones, A.G., 1983. The problem of current channelling: A critical review, 891
Surveys in Geophysics, 6, 79-122. 892
28. Jones, A.G., 1986. Parkinson's pointers' potential perfidy!, Geophysical Journal 893
of the Royal Astronomical Society, 87, 1215-1224. 894
29. Jones, A.G., 1999. Imaging the continental upper mantle using electromagnetic 895
methods, Lithos, 48, 57-80. 896
30. Jones, A. G. 2012, Distortion of magnetotelluric data: its identification and 897
removal, in A. D. Chave and A. G. Jones, eds., The Magnetotelluric Method: 898
Theory and Practice: Cambridge University Press. 899
31. Jones, A.G., Evans, R.L., Muller, M.R., Hamilton, M.P., Miensopust, M.P., 900
Garcia, X., Cole, P., Ngwisanyi, T., Hutchins, D., Fourie, C.J.S., Jelsma, H., 901
Evans, S., Aravanis, T., Pettit, W., Webb, S., Wasborg, J. & Team, S., 2009. 902
Area selection for diamonds using magnetotellurics: Examples from southern 903
Africa. in Lithos, pp. 83-92. 904
32. Jones, A.G. & Garcia, X., 2003. Okak Bay AMT data-set case study: Lessons in 905
dimensionality and scale, Geophysics, 68, 70-91. 906
33. Kelbert, A., Meqbel, N., Egbert, G. D., & Tandon, K., 2014. ModEM: A 907
Modular System for Inversion of Electromagnetic Geophysical Data. Computers 908
& Geosciences. 66,40-53. http://dx.doi.org/10.1016/j.cageo.2014.01.010. 909
34. Kiyan, D., A.G. Jones, & J. Vozar, 2014. The inability of magnetotelluric off-910
diagonal impedance tensor elements to sense oblique conductors in 3-D 911
inversion. Geophys. J. Int., 196, 1351-1364, doi: 10.1093/gji/ggt470. 912
913 35. Kühn, C., Küster, J. & Brasse, H., 2014. Three-dimensional inversion of 914
magnetotelluric data from Central Andean continental margin, Earth, Planets 915
and Space, 66:112. 916
36. La Terra, E.F. & Menezes, P.T.L., 2012. Audio-magnetotelluric 3D imaging of 917
the Regis kimberlite pipe, Minas Gerais, Brazil. J. of Appl. Geophys., 77, 30-38. 918
37. Le Pape, F.; Jones, A.G.; Vozar, J. & Wenbo, W.; 2012. Penetration of crustal 919
melt beyond the Kunlun Fault into northern Tibet. Nature Geoscience, 5, 330-920
335. 921
Page 36 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
38. Ledo, J., Queralt, P., Marti, A. & Jones, A.G., 2002. Two-dimensional inter-922
pretation of three-dimensional magnetotelluric data: an example of limitations 923
and resolution, Geophys. J. Int., 150, 127–139. 924
39. Ledo, J., 2005. 2-D versus 3-D magnetotelluric data interpretation, Surv. 925
Geophys., 26, 511–543. 926
40. Ledo, J.; Jones, A.G.; Siniscalchi, A.; Campanyà, J.; Kiyan, D.; Romano, G.; 927
Rouai, M.; TopoMed MT Team, 2011. Electrical signature of modern and 928
ancient tectonic processes in the crust of the Atlas mountains of 929
Morocco. Physics of the Earth and Planetary Interiors, 185, 82-930
88,doi:10.1016/j.pepi.2011.01.008. 931
41. Lezaeta, P., Haak, V. ,2003. Beyond magnetotelluric decomposition: induction, 932
current channelling, and magnetotelluric phases over 90. J.Geophys.Res. 108 933
(B6), 2305 ,http://dx.doi.org/10.1029/2001JB000990. 934
42. Marcuello, A., Queralt, P., Ledo, J., 2005. Applications of dispersion relations to 935
the geomagnetic transfer function. Physics of the Earth and Planetary Interiors, 936
150, 85-91. 937
43. Miensopust, M.P., P. Queralt, A.G. Jones, & the 3D MT modellers, 2013. 938
Magnetotelluric 3D inversion -a review of two successful workshops on forward 939
and inversion code testing and comparison. Geophys. J. Int., 193, 1216-1238, 940
doi: 10.1093/gji/ggt066. 941
44. Meqbel, N.M; Egbert, G.D., Wannamaker, P.E., Kelbert, A., Schultz, A., 2014. 942
Deep electrical resistivity structure of the northwestern U.S. derived from 3-D 943
inversion of USArray magnetotelluric data. Earth and Planetary Science Letters, 944
402, 290-304. doi:10.1016/j.epsl.2013.12.026. 945
45. Muñoz, G., 2014. Exploring for geothermal Resources with Electromagnetic 946
Methods, Surveys in Geophysics, 35 (1), 101-122. 947
46. Ogaya, X., Ledo, J., Queralt, P., Marcuello, A. & Quintà, A., 2013.First 948
geoelectrical image of the subsurface of the Hontomín site (Spain) for CO 2 949
geological storage: a magnetotelluric 2D characterization. International Journal 950
of Greehouse Gas Control, 13, 168-179. 951
47. Ogaya, X., Queralt, P., Ledo, J., Marcuello, A. & Jones, A.G., 2014. 952
Geoelectrical baseline model of the subsurface of the Hontomin site (Spain) for 953
CO 2 geological storage in a deep saline aquifer: A 3D magnetotelluric 954
characterization. International Journal of Greenhouse Gas Control, 27, 120-138. 955
Page 37 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
48. Park, S.K. & Mackie, R.L., 2000. Resistive (dry?) lower crust in an active 956
orogen, Nanga Parbat, northern Pakistan, Tectonophysics, 316, 359-380. 957
49. Piña-Varas, P., Ledo, J., Queralt, P., Marcuello, A., Bellmunt, F., Hidalgo, R. & 958
Messeiller., 2014. 3-D Magnetotelluric Exploration of Tenerife Geothermal 959
System (Canary Islands, Spain). Surveys in Geophysics, 35 (4), 1045-1064. 960
50. Rao, C.K., Jones, A.G., Moorkamp, M. & Weckmann, U. 2014, Implications for 961
the lithospheric geometry of the Iapetus suture beneath Ireland based on 962
electrical resistivity models from deep- probing magnetotellurics. Geophysical 963
Journal International, 198, 737-759, doi: 10.1093/gji/ggu136. 964
51. Rosell, O., Martí, A., Marcuello, À., Ledo, J., Queralt, P., Roca, E. & Campanyà, 965
J., 2011. Deep electrical resistivity structure of the northern Gibraltar Arc 966
(western Mediterranean): evidence of lithospheric slab break-off. Terra 967
Nova. 23, 179 - 186. doi: 10.1111/j.1365-3121.2011.00996.x. 968
52. Sasaki, Y., 2004. Three-dimensional inversion of static-shifted magnetotelluric 969
data, Earth Planets and Space, 56, 239-248. 970
53. Sasaki, Y. & Meju, M.A., 2006. Three-dimensional joint inversion for 971
magnetotelluric resistivity and static shift distributions in complex media, 972
Journal of Geophysical Research-Solid Earth, 111, 11. 973
54. Schmucker, U., 1970. Anomalies of geomagnetic variations in the Southwestern 974
United States. J. Geomagn. Oceanogr., Univ. of California Press, Berkeley. 975
55. Seama, N., Baba, K., Utada, H., Toh, H., Tada., N., Ichiki, M. & Matsuno, T., 976
2007. 1-D electrical conductivity structure beneath the Philippine Sea: Results 977
from an ocean bottom magnetotelluric survey. Physics of the Earth and 978
planetary interiors, 162, 2-12. 979
56. Siripunvaraporn, W., Egbert, G., Lenbury, Y. & Uyeshima, M., 2005a. Three- 980
dimensional magnetotelluric inversion: data-space method, Phys. Earth planet. 981
Inter., 150, 3–14. 982
57. Siripunvaraporn, W. and Egbert, G., 2009. WSINV3DMT: vertical magnetic 983
field transfer function inversion and parallel implementation, Phys. Earth Planet. 984
Inter., 173, 317—329. 985
58. Siripunvaraporn, W., Egbert, G. & Uyeshima, M., 2005b. Interpretation of two-986
dimensional magnetotelluric profile data with three-dimensional inversion: 987
synthetic examples, Geophys. J. Int., 160, 804–814. 988
Page 38 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
59. Simpson, F. & Bahr, K., 2005. Practical Magnetotellurics. Cambridge University 989
press. United Kingdom. 990
60. Soyer , W & Brasse, H., 2001. A magneto-variational array study in the Central 991
Andes of N Chile and SW Bolivia, Geophysical Research Letters, 28, 3023-992
3026. 993
61. Soyer, W., 2002. Analysis of geomagnetic variations in the Central and Southern 994
Andes. PhD Thesis, Freie Universität Berlin, Fachbereich Geowissenschaften, 995
Berlin. 996
62. Thiel, S., Heinson, G., Gray D.R., and Gregory. R.T., 2009. Ophiolite 997
emplacement in NE Oman: constraints from magnetotelluric sounding. 998
Geophysical Journal International, 176(3):753-766. 999
63. Tietze, K. & Ritter, O., 2013. Three-dimensional magnetotelluric inversion in 1000
practice-the electrical conductivity structure of the San Andreas Fault in Central 1001
California, Geophys. J. Int., 195, 130-147. 1002
64. Tuncer, V.M., Unsworth, E.M. & Pana, D., 2009. Deep electrical structure of 1003
northern Alberta (Canada): implications for diamond exploration, Can. J. Earth 1004
Sci., 46, 139-154. 1005
65. Unsworth, M.J., 2005. New developments in conventional hydrocarbon 1006
exploration with EM methods. Canadian Society of Exploration Geophysicists 1007
Recorder, pp. 34–38. [April.] 1008
66. Usui, Y., 2015. 3-D inversion of magnetotelluric data using unstructured 1009
tetrahedral elements: applicability to data affected by topography, Geophys. J. 1010
Int., 202, 828-849. 1011
67. Varentsov Iv.M., 2015a. Methods of joint robust inversion in MT and MV 1012
studies with application to synthetic datasets // Electromagnetic Sounding of the 1013
Earth's Interior: Theory, Modeling, Practice (Ed. V.V. Spichak). Amsterdam: 1014
Elsevier. P. 191-229. 1015
68. Varentsov Iv.M., 2015b. Arrays of simultaneous EM soundings: design, data 1016
processing, analysis, and inversion // Electromagnetic Sounding of the Earth's 1017
Interior: Theory, Modeling, Practice (Ed. V.V. Spichak). 1018
Amsterdam: Elsevier. P. 271-299. 1019
69. Vozoff, K., 1991. The magnetotelluric method, Electromagneitc methods in 1020
applied geophysics –Aplications, Chapter 8, 1991. 1021
Page 39 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
70. Wanamaker, P.E., Hohmann, G.W. & Ward, S.H., 1984. Magnetotelluric 1022
responses of three-dimensional bodies in layered earths, Geophysics, 49, 1517–1023
1533. 1024
71. Yang, B., Egbert, G.D., Kelbert, A., Meqbel, N., 2015. Three-dimensional 1025
electrical resistivity of the north-central USA from EarthScope long period 1026
magnetotelluric data. Earth and Planetary Science Letters, 422, 87–93. 1027
1028
1029
FIGURE CAPTIONS: 1030
1031
Figure 1: Synthetic geoelectrical model composed of four layers: (1) Between 0 m and 1032
90 m depth with an electrical resistivity of 100 Ωm; (2) Between 90 m and 1500 m depth 1033
with an electrical resistivity of 400 Ωm; (3) Between 1500 m and 2000 m depth with an 1034
electrical resistivity of 10 Ωm; (4) Layer from 2000 m to 90 km depth with an electrical 1035
resistivity of 200 Ωm. Two low electrical resistivity structures (L1 and L2) and a high 1036
electrical resistivity body (H1) are also present. L1 and L2 are located between 500 m 1037
and 1000 m depth and have an electrical resistivity value of 3 Ωm. H1 is located 1038
between 1500 m and 2000 m depth with electrical resistivity values of 400 Ωm. The 1039
green arrows points to site 26, which is used as a neighbouring reference site for H 1040
responses. 1041
1042
Figure 2: Results of the sensitivity test for the L1 anomaly. Values obtained by 1043
calculating the differences between the forward responses of the 1-D layered model 1044
containing L1 anomaly and the forward responses of the background 1-D layered model 1045
without anomalies. The differences are divided by the assumed error, thus showing how 1046
sensitive the responses are to the anomaly with respect to the errors. White areas are 1047
Page 40 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
periods that are non-sensitive to L1 (differences smaller than the assumed error). Red 1048
and blue areas are periods that are sensitive to L1. 1049
1050
Figure 3: Results of the sensitivity test for the L2 anomaly. Values obtained by 1051
calculating the differences between the forward responses of the 1-D layered model 1052
containing L2 anomaly and the forward responses of the background 1-D layered model 1053
without anomalies. The differences are divided by the assumed error, thus showing how 1054
sensitive the responses are to the anomaly with respect to the errors. White areas are 1055
periods that are non-sensitive to L2 (differences smaller than the assumed error). Red 1056
and blue areas are periods that are sensitive to L2. 1057
1058
Figure 4: Results of the sensitivity test for the H1 anomaly. Values obtained by 1059
calculating the differences between the forward responses of the 1-D layered model 1060
containing H1 anomaly and the forward responses of the background 1-D layered model 1061
without anomalies. The differences are divided by the assumed error, thus showing how 1062
sensitive the responses are to the anomaly with respect to the errors. White areas are 1063
periods that are non-sensitive to H1 (differences smaller than the assumed error). Red 1064
and blue areas are periods that are sensitive to H1. 1065
1066
Figure 5: Results of the sensitivity test for the combined L1+L2+H1 anomalies. Values 1067
obtained by calculating the differences between the forward responses of the 1-D 1068
layered model containing L1+L2+H1 anomalies and the forward responses of the 1069
background 1-D layered model without anomalies. The differences are divided by the 1070
assumed error, thus showing how sensitive the responses are to the anomalies with 1071
respect to the errors. White areas are periods that are non-sensitive to L1+L2+H1 1072
Page 41 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
anomalies (differences smaller than the assumed error). Red and blue areas are periods 1073
that are sensitive to L1+L2+H1. 1074
1075
Figure 6: Results from the sensitivity test to quantify the effect of each anomaly on the 1076
Z, Zo, H and T responses. White areas are periods that are not sensitive to the anomalies. 1077
Note that all plots have different colour scales. Grey thick lines at the top of each 1078
pseudo-section show the location of the studied anomalies along the profile. 1079
1080
Figure 7: a) Phase tensor dimensionality analysis using Z responses. White-grey colours 1081
are periods affected by signal associated with 1-D or 2-D structures. Rainbow colours 1082
are periods affected by signal associated with 3-D anomalies. b) Results from the 1083
dimensionality analysis using variations of the H responses (H-I), where I is the 1084
identity. Dots are related to periods affected by signal associated with 1-D structures. 1085
Lines are related to periods affected by signal associated with 2-D structures. Ellipses 1086
are related to periods affected by signal associated with 3-D anomalies. White-grey 1087
colours are periods that, because of the assumed errors, do not require the presence of 3-1088
D anomalies. Rainbow colours are periods affected by signal that, even with the 1089
assumed errors, requires the presence of 3-D anomalies. c) Induction arrows from T 1090
responses, following the Parkinson criteria, used to characterize the dimensionality of 1091
the area. White-grey colours are periods that, because of the assumed errors, do not 1092
require the presence of 3-D anomalies. Rainbow colours are periods affected by signal 1093
that, even with the assumed errors, requires the presence of 3-D anomalies. 1094
1095
Figure 8: a) Results of the different inversion process below the MT profile using 1096
neighbouring reference sites located at different places along the MT profile. White 1097
Page 42 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
dashed lines are plotted to better compare the results with the original synthetic model. 1098
b) nRMS of each inversion processes for each site and for each type of data. Circles are 1099
the nRMS of the corresponding site. 1100
1101
1102
Figure 9: a) Results of the different inversion process below the MT profile. b) 1103
Horizontal slices showing the results of the inversion processes at 650 m depth. c) 1104
Horizontal slices showing the results of the inversion processes at 1550 m depth. a5), 1105
b5) and c5) are the original synthetic model. White dashed lines are plotted to better 1106
compare the results with the original synthetic model. 1107
1108
Figure 10: a) Results of the different inversion processes beneath the MT profile. b) 1109
Horizontal slices showing the results of the inversion processes at 650 m depth. c) 1110
Horizontal slices showing the results of the inversion processes at 1550 m depth. White 1111
dashed lines are plotted to better compare the results with the original synthetic model. 1112
1113
Figure 11: nRMS of each inversion processes for each site and for each type of data. 1114
Circles are the nRMS of the corresponding site. Excellent nRMS misfits of between 1.0 1115
and 1.05 were achieved for all inversions using different types of responses. 1116
1117
Figure 12: Differences, using decadic logarithm electrical resistivity values, between 1118
models obtained from the inversion process and the synthetic model. 1119
Page 43 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
NS
De
pth
(m
)
1000
2000
3000
0
0
4000
-4000
0-6000 6000
No
rth
ing
(m
)
Easting (m)
0
4000
-4000
0-6000 6000
No
rth
ing
(m
)
Easting (m)
8000
-8000 -8000
8000
b) Horizontal slice (700 m depth) c) Horizontal slice (1500 m depth)
0 8000-8000 -4000 4000
Distance Profile (m)
Figure 1 (Campanyà et al. 2016)
400 ohm·m
200 ohm·m
(L3) 10 ohm·m
3 ohm·m
400 ohm·m
(L1)3 ohm·m
(L2)3 ohm·m
(H1)400 ohm·m
(L3)10 ohm·m
(H1)
(L1)
1 14 217 28
a) MT profile
Electrical resistivity [log10(ohm·m)]
0 0.5 1.5 2.51 2
Synthetic model
Page 44 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Figure 2 (Campanyà et al. 2016)
L1 anomaly
Difference / Errorfloor
a)
Hxx real Hxy real Hyx real Hyy real
Hxx imag Hxy imag Hyx imag Hyy imag
Zxx real Zxy real Zyx real Zyy real
Zxx imag Zxy imag Zyx imag Zyy imag
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Sites1 26Sites1 26 Sites1 26 Sites1 26
b)
c) Tx real Tx imag Ty real Ty imag
3-3-4 -2 0 2 41-1
Page 45 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Figure 3 (Campanyà et al. 2016)
a) Zxx real Zxy real Zyx real Zyy real
Zxx imag Zxy imag Zyx imag Zyy imag
Pe
rio
ds
(lo
g1
0(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g1
0(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g1
0(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g1
0(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g1
0(s
))
-4-3
-2-10123
Sites1 26Sites1 26 Sites1 26 Sites1 26
b)
c)
Difference / Errorfloor
L2 anomaly
3-3-4 -2 0 2 41-1
Hxx real Hxy real Hyx real Hyy real
Hxx imag Hxy imag Hyx imag Hyy imag
Tx real Tx imag Ty real Ty imag
Page 46 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Figure 4 (Campanyà et al. 2016)
Difference / Errorfloor
a) Zxx real Zxy real Zyx real Zyy real
Zxx imag Zxy imag Zyx imag Zyy imag
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Pe
rio
ds
(lo
g10(s
))
-4-3
-2-10123
Sites1 26Sites1 26 Sites1 26 Sites1 26
b)
c)
H1 anomaly
3-3-4 -2 0 2 41-1
Hxx real Hxy real Hyx real Hyy real
Hxx imag Hxy imag Hyx imag Hyy imag
Tx real Tx imag Ty real Ty imag
Page 47 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Figure 5 (Campanyà et al. 2016)
Difference / Errorfloor
3-3
Hxx real Hxy real Hyx real Hyy real
Hxx imag Hxy imag Hyx imag Hyy imag
Tx real Tx imag Ty real Ty imag
a) Zxx real Zxy real Zyx real Zyy real
Zxx imag Zxy imag Zyx imag Zyy imag
Periods
(log
10(s
))
-4-3
-2-10123
Periods
(log
10(s
))
-4-3
-2-10123
Periods
(log
10(s
))
-4-3
-2-10123
Periods
(log
10(s
))
-4-3
-2-10123
Periods
(log
10(s
))
-4-3
-2-10123
Sites1 26Sites1 26 Sites1 26 Sites1 26
b)
c)
-4 -2 0 2 41-1
L1, L2 and H1 anomalies
Page 48 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
42.51 63.51
71 4
Periods
(log
10(s
))
-4-3
-2-10123
Periods
(log
10(s
))
-4-3
-2-10123
HT
Periods
(log
10(s
))-4-3
-2-10123
Periods
(log
10(s
))
-4-3
-2-10123
Z_all
Zo
Sites1 26Sites1 26 Sites1 26 Sites1 26
L1 L2 H1 L1, L2, H1
Sites1 26Sites1 26 Sites1 26 Sites1 26
Sites1 26Sites1 26 Sites1 26 Sites1 26
Sites1 26Sites1 26 Sites1 26 Sites1 26
1891 1261
96481 1681 126 10011 50
126.51
1051 51 101 5.53
a)
b)
c)
d)
Figure 6 (Campanyà et al. 2016)
1001 50 1001 50
1161
Page 49 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Periods
(log
10(s
))-4
-3
-2
-1
0
1
2
3
a)
Figure 7 (Campanyà et al. 2016)
41 6 11 16 21 26
Phase Tensor (Z)
abs(Skew angle beta /error Skew angle beta)sites
Pe
rio
ds
(lo
g10(s
))
-4
-3
-2
-1
0
1
2
3
c)
41 6 11 16 21 26
Induction Arrows (T)
sites
Pe
rio
ds
(lo
g10(s
))
-4
-3
-2
-1
0
1
2
3
b)
41 6 11 16 21 26
variation of H
0 1 2 3
abs(Phi_min /error Phi_min)sites
0 1 2 3
0 1 2 3
abs( angle difference / error angle difference)
L2 L1 H1
L2 L1 H1
L2 L1 H1
real imag0.3 0.3
Page 50 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
0 0.5 1.5 2.51 2
Electrical resistivity [log10(ohm·m)]
RMS:1.09
RMS:1.98RMS:1.25
Depth
(km
)
1
2
3
0
0-4 4Distance Profile (km)
8-8
2 431
b1 b2 b3 b4
De
pth
(km
)
1
2
3
0
a1) Z+H, close L2 a2) Z+H, top L1S N S N
0-4 4Distance Profile (m)
RMS:1.35De
pth
(km
)
1
2
3
0
0-4 4Distance Profile (m)
Figure 8 (Campanyà et al. 2016)
a3) Z+H, top H1 a4) Z+H
1.5
1
0.5
RM
S
2
2.5
1.5
1
0.5
2
2.5
Z H
Sites1 28 Sites1 28Sites1 28 Sites1 28
b)
a)
Page 51 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Depth
(km
)
1
2
3
0
0-4 4Distance Profile (km)
0 0.5 1.5 2.51 2
Electrical resistivity [log10(ohm·m)]
b)
c)
0
4
-4Nort
hin
g (
km)
0-4 4Easting (km)
0-4 4Easting (km)
0-4 4Easting (km)
0-4 4Easting (km)
0-4 4Easting (km)
c1) Zo c3) Zo+H+T c2) Z c4) Z+H+T c5) S. Model
0
4
-4Nort
hin
g (
km)
0-4 4Easting (km)
0-4 4Easting (km)
0-4 4Easting (km)
0-4 4Easting (km)
0Easting (km)
-4 4
b1) Zo b3) Zo+H+T b2) Z b4) Z+H+T b5) S. Model
a)
Depth
(km
)
1
2
3
0
Depth
(km
)
1
2
3
0
0-4 4
Distance Profile (km)
a1) Zo a2) Z
a3) Zo+H+T a4) Z+H+T
a5) Synthetic Model
8-8
-8 8
S N S N
Figure 9 (Campanyà et al. 2016)
Page 52 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
b)
c)
a)
0 0.5 1.5 2.51 2
Electrical resistivity [log10(ohm·m)]
De
pth
(km
)
1
2
3
0
a1) Zo+T a2) Zo+H S N S N
0-4 4Distance Profile (m)
0
4
-4
0-4 4
No
rth
ing
(km
)
Easting (km)0-4 4
Easting (km)
b1) Zo+T b2) Zo+H
0-4 4Easting (km)
0-4 4Easting (km)
c1) Zo+T c2) Zo+H
De
pth
(km
)
1
2
3
0
De
pth
(km
)
1
2
3
0
0-4 4Distance Profile (m)
Figure 10 (Campanyà et al. 2016)
a3) Z+T a4) Z+H
a5) T+H
0-4 4Easting (km)
0-4 4Easting (km)
0-4 4Easting (km)
b3) Z+T b5) H+T b4) Z+H
0
4
-4
0-4 4
No
rth
ing
(km
)
Easting (km)0-4 4
Easting (km)0-4 4
Easting (km)
c3) Z+T c5) H+T c4) Z+H
Page 53 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Figure 11 (Campanyà et al. 2016)
1.5
1
0.5
RM
S
1.5
1
0.5
RM
S
1.5
1
0.5
RM
S
Sites1 28 Sites1 28 Sites1 28
Zo inversion Zo+T inversion Zo+H inversion
Z inversion Z+T inversion Z+H inversion
Zo+T+H inversion Z+T+H inversion T+H inversion
Zo Z T H
1.5
1
0.5
1.5
1
0.5
1.5
1
0.5
Page 54 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Figure 12 (Campanyà et al. 2016)
De
pth
(km
)
1
2
3
0
e) Zo+H f) Z+H
0-4 4Distance Profile (m)
De
pth
(km
)
1
2
3
0
De
pth
(km
)
1
2
3
0
0-4 4Distance Profile (m)
g) Zo+T+H h) Z+T+H
i) T+H
b) Z S N
De
pth
(km
)
1
2
3
0
c) Zo+T d) Z+T
a) Zo S N
De
pth
(km
)
1
2
3
0
0 0.5-0.5-0.75 0.75-0.25 0.25
Difference electrical resistivity [ohm·m]log10(Synth. mod) - log10(Inv. mod)
Page 55 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Supplementary material S1 Campanyà et al., 2016 GJI Supplementary 1: Implementation of the H responses
Following the mathematical recipes (Egbert and Kelbert, 2012) and the practical
implementation of ModEM (Kelbert et al., 2014) we modified the code accordingly to allow
inverting for horizontal magnetic inter-station transfer functions. The modifications are
implemented mainly in the data functional module within the interface layer (Kelbert et al.,
2014, Fig. 1). In the data functional module (data responses routine), all kinds of transfer
functions are computed from the electric field solution (primary field) obtained from solving
the induction problem in 3-D. For the horizontal inter-station transfer functions we use the
magnetic field (dual field) derived from the electric field using Faraday's law. To simulate the
process of measuring EM components (e.g., inter-station horizontal magnetic transfers
functions) the magnetic field is interpolated at specific locations (local and neighbouring
reference sites). Using the interpolated horizontal magnetic field components (ℎ! and ℎ!) at
the local (𝒉!) and neighbouring reference site (𝒉!), we construct the inter-station horizontal
magnetic transfer functions (𝑯) equation S1.
𝒉! = 𝐇 !"𝒉! (𝑆1)
For the inversion process in ModEM we use the None Linear Conjugate Gradients algorithm
(NLCG), which is based on a direct minimization of the penalty functional (Egbert and
Kelbert, 2012). All gradient-based optimization algorithm (including NLCG) require the
sensitivity of the data at each iteration. The sensitivity matrix in ModEM is computed by
multiplying several sub matrices (Egbert and Kelbert, 2012; Eqs. 14 and 15). The matrix that
depends on the inverted data type (e.g., H) is the so-called L matrix that represents the
linearized data functional. Mimicking the computation of the L matrix for the impedances
transfer functions (Z; Egbert and Kelbert, 2012; Eqs. 42 and 43) we computed the sensitivity
matrix for the inter-station horizontal magnetic transfer functions (𝑯). Hence, most of the
modifications required to compute the sensitivity matrix in case of inverting for H are
implemented in the none linearized data functional routine within the data functional module.
Page 56 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
a) Zo inversionlo
g10 f
req. (H
z)Zxy real Zxy imag Zyx real Zyx imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
b) Z inversion
log1
0 fr
eq. (
Hz)
Zxx real Zxy real Zyx real Zyy real
log1
0 fr
eq. (
Hz)
Zxx imag Zxy imag Zyx imag Zyy imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
Figure S2a,b (Campanyà et al. 2016)
(data - model_response) / error
Page 57 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
c) Zo+T inversionlo
g10 f
req. (H
z)Zxy real Zxy imag Zyx real Zyx imag
d) Z+T inversion
log1
0 fr
eq. (
Hz)
Zxx real Zxy real Zyx real Zyy real
log1
0 fr
eq. (
Hz)
Zxx imag Zxy imag Zyx imag Zyy imag
Figure S2c,d (Campanyà et al. 2016)
log1
0 fr
eq. (H
z)
Tx real Tx imag Ty real Ty imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
log1
0 fr
eq. (
Hz)
Tx real Tx imag Ty real Ty imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
(data - model_response) / error
Page 58 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
e) Zo+H inversionlo
g10 f
req. (H
z)
Zxy real Zxy imag Zyx real Zyx imag
log1
0 fr
eq. (
Hz)
Hxx real Hxy real Hyx real Hyy real
Figure S2e (Campanyà et al. 2016)
log1
0 fr
eq. (
Hz)
Hxx imag Hxy imag Hyx imag Hyy imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
(data - model_response) / error
Page 59 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
f) Z+H inversionlo
g10 f
req. (H
z)
Zxx real Zxy real Zyx real Zyy real
log1
0 fr
eq. (
Hz)
Zxx imag Zxy imag Zyx imag Zyy imag
Figure S2f (Campanyà et al. 2016)
log1
0 fr
eq. (
Hz)
Hxx real Hxy real Hyx real Hyy real
log1
0 fr
eq. (
Hz)
Hxx imag Hxy imag Hyx imag Hyy imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
(data - model_response) / error
Page 60 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
g) Zo+T+H inversionlo
g10 f
req. (H
z)
Zxy real Zxy imag Zyx real Zyx imag
log1
0 fr
eq. (
Hz)
Tx real Tx imag Ty real Ty imag
Figure S2g (Campanyà et al. 2016)
log1
0 fr
eq. (
Hz)
Hxx real Hxy real Hyx real Hyy real
log1
0 fr
eq. (
Hz)
Hxx imag Hxy imag Hyx imag Hyy imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
(data - model_response) / error
Page 61 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
h) Z+T+H inversionlo
g10 f
req. (H
z)
Zxx real Zxy real Zyx real Zyy real
log1
0 fr
eq. (
Hz)
Tx real Tx imag Ty real Ty imag
Figure S2h (Campanyà et al. 2016)
log1
0 fr
eq. (
Hz)
Hxx real Hxy real Hyx real Hyy real
log1
0 fr
eq. (
Hz)
Hxx imag Hxy imag Hyx imag Hyy imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
log1
0 fr
eq. (
Hz)
Zxx imag Zxy imag Zyx imag Zyy imag
(data - model_response) / error
Page 62 of 63Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
i) T+H inversionlo
g10 f
req. (H
z)
Tx real Tx imag Ty real Ty imag
Figure S2i (Campanyà et al. 2016)
log1
0 fr
eq. (
Hz)
Hxx real Hxy real Hyx real Hyy real
log1
0 fr
eq. (
Hz)
Hxx imag Hxy imag Hyx imag Hyy imag
Sites1 28
Sites1 28
Sites1 28
Sites1 28
(data - model_response) / error
Page 63 of 63 Geophysical Journal International
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960