Geophysical Journal International

Geophys. J. Int. (2014) 198, 1016–1033 doi: 10.1093/gji/ggu180

GJI Marine geosciences and applied geophysics

Spectral induced polarization porosimetry

A. Revil,1,2 N. Florsch3,4 and C. Camerlynck4

1 Departmentof Geophysics, Colorado School of Mines, Golden, CO 80401, USA. E-mail: arevil@mines.edu
2 CNRS-UMR 5559-LGIT, Université de Savoie, Equipe volcan, F-73376 Le-Bourget-du-Lac, France
3 UMMISCO UMI n◦ 209, 32 Av. Henry Varagnat, F-93143 Bondy cedex, France
4 Sorbonne Universités, UPMC Univ Paris 06, UMR7619 Métis, F-75005 Paris, France

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

Accepted 2014 May 14. Received 2014 May 11; in original form 2014 January 24

SUMMARY
Induced polarization is a geophysical method looking to image and interpret low-frequency
polarization mechanisms occurring in porous media. Below 10 kHz, the quadrature conductiv-
ity of metal-free sandy and clayey materials exhibits a distribution of relaxation times, which
can be related to the pore size distribution of these porous materials. When the polarization
spectra are fitted with a Cole–Cole model, we first observe that the main relaxation time is
controlled by the main pore size of the material and that the Cole–Cole exponent c is never
much above 0.5, a value corresponding to a Warburg function. The complex conductivity
is then obtained through a convolution product between the pore size distribution and such
Warburg function. We also provide a way to recover the pore size distribution by performing
a deconvolution of measured spectra using the Warburg function. A new dataset of mercury
porosimetry and induced polarization data of six siliciclastic materials supports the hypothesis
that the Cole–Cole relaxation time is strongly controlled by the pore size, and especially the
characteristic pore size corresponding to the peak of the pore size distribution from mercury
porosimetry. The distribution of the pore throat sizes of these materials seems fairly well
recovered using the Warburg decomposition of the spectral induced polarization spectra but
additional data will be needed to confirm this finding.
Key words: Electrical properties; Hydrogeophysics; Permeability and porosity.

by an electrical double layer formed by a Stern layer of sorbed coun-

1 I N T RO D U C T I O N
terions and a diffuse layer of ions attached to the mineral surface
Induced polarization characterizes the ability of a porous ma- through Coulombic interaction with the net electrical charge on
terial to store reversibly electrical charges at low frequencies the surface of the minerals (Avena & de Pauli 1998). The ex-
(<10 kHz). The geophysical method named from the measurement istence of this electrical double layer and specially the Stern
of these low-frequency polarization mechanism seems very promis- layer seems to have a strong influence on electrical polarization.
ing in environmental geosciences and hydrogeology to characterize This has been proven recently by changing the type of specific
the permeability distribution of the subsurface (e.g. Kemna 2000; cations sorbed in the Stern layer and looking at the effect of these
Binley et al. 2005; Hördt et al. 2007; Revil & Florsch 2010; Attwa changes upon the magnitude of the polarization (see Vaudelet et al.
& Günther 2013), to differentiate sedimentary formations (Attwa 2011a,b).
& Günther 2012; Gazoty et al. 2012), to monitor biodegradation Induced polarization measurements can be performed either in
processes (Davis et al. 2006; Revil et al. 2012c) and to delineate the frequency domain (spectral induced polarization, SIP) or in the
contaminant plumes and to monitor their remediation (e.g. Vanhala time domain. In addition to a normalized chargeability (difference
1997; Deceuster et al. 2005; Schmutz et al. 2010; Vaudelet et al. between the low and high-frequency electrical conductivities), a
2011a,b; Chen et al. 2012; Deceuster & Kaufmann 2012; Schwartz distribution of relaxation times can be inverted from induced polar-
& Furman 2012). Reviews of recent progress in the field of induced ization spectral data by performing a deconvolution of the response
polarization applied to hydrogeophysical problems can be found in using a transfer function for the system (see Tong et al. 2006a,b;
Kemna et al. (2012) and Revil et al. (2012a). Tarasov & Titov 2007, for time-domain induced polarization and
Low-frequency polarization is usually associated with the fact Kemna 2000; Ghorbani et al. 2007; Chen et al. 2008; Nordsiek &
that the movement of ions in a porous material is governed not Weller 2008 for frequency-domain induced polarization). Using a
only by Coulombic forces but also by concentration gradients (both mechanistic model, the distribution of relaxation times can be as-
potentials can be combined to form electrochemical potentials, for sociated in turn to a distribution of polarization length scales, for
example, Marshall & Madden 1959). Minerals are generally coated instance a distribution of grain sizes or pore sizes (see Tong et al.

1016 
C The Authors 2014. Published by Oxford University Press on behalf of The Royal Astronomical Society.
 Influence of pore size on induced polarization 1017

2004, 2006a,b; Revil & Florsch 2010; Revil et al. 2012b; Bücker & (a)
Hördt 2013).
In this paper, we first use a simple Cole–Cole model to describe
the low-frequency complex conductivity of porous media. In this
case, we will also provide evidences that the Cole–Cole exponent
c is comprised between zero (for very broad pore size or parti-
cle size distributions) and 0.5 (in the case of very narrow pore
size or particle size distributions). This observation points out that
the correct transfer function that should be used to perform the de-
convolution of the spectra is not the Debye function (a Cole–Cole
model with a c-exponent of 1) but a Warburg function (a special case
of the Cole–Cole function with a c-exponent equal to 0.5). Then,
we will develop a method to perform the deconvolution of induced
(b)
polarization spectra using the Warburg function and to analyse the

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

distribution of relaxation times in terms of pore size distribution.
This approach may offer a non-intrusive, non-destructive and en-
vironmentally friendly way to get porosimetry data from SIP data
rather than using mercury porosimetry.

2 A COLE–COLE MODEL FOR THE

COMPLEX CONDUCTIVITY
The goals of this section are (1) to demonstrate that the main relax-
ation time in a Cole–Cole model of complex conductivity is related
to the main pore size of a porous material and (2) to show that for
very narrow pore size distributions (close to a delta function), the
Cole–Cole exponent is close to 0.5.

2.1 Definitions Figure 1. Polarization of a single grain. (a) Electrical double layer (Stern
plus diffuse layers) at equilibrium. (b) Polarization mechanisms of the grain
In frequency-domain induced polarization, we record the magni-
under an electrical field E. Three mechanisms occur: (i) the Falkenhagen
tude of the electrical conductivity from the current and the mea- contribution is due to the deformation of the diffuse layer and is characterized
sured voltage (corrected by a geometrical factor depending on the by a very fast relaxation time. (ii) The Stern layer polarization is due to
position of the electrodes and boundary conditions) and a phase the migration of the counterions in the Stern layer. (iii) The membrane
between the current and the voltage in response to a periodic har- polarization is due to the increase of salinity in the direction of the electrical
monic current. We note ω = 2 π f the angular frequency in rad field and the decrease of salinity on the other side of the grain (actually this
s−1 , f is the frequency in Hz and i = (−1)1/2 the pure imaginary increase/decrease of the salinity depends on the value of the transference
number, AB the current electrodes and MN the voltage electrodes. numbers for the cations and anions).
The results can be expressed as a complex conductivity σ ∗ or
the isoelectric point of the mineral surface (zeta potential equals
a complex resistivity ρ ∗ = 1/σ ∗ . The relationships between the
zero) for which there is no electrical diffuse layer (e.g. Revil et al.
modulus of this conductivity |σ | and the phase ϕ and the real
2013a,b).
and imaginary components of the conductivity,  σ  and σ  , are
∗  
A very popular complex conductivity model is the Cole–Cole
given by σ = |σ | exp(iϕ) = σ + iσ , with |σ | = σ 2 + σ 2 and function (e.g. Tarasov & Titov 2013):
tan ϕ = σ  /σ  . In Section 2.2, we provide a simple general model
Mn
to describe the complex conductivity of clayey or clean sands and σ ∗ (ω) = σ∞ − , (1)
sandstones. 1 + (iωτ0 )c
where the normalized chargeability is traditionally defined by (e.g.
Kemna 2000),
2.2 Model description
Mn = σ∞ − σ0 ≥ 0, (2)
Fig. 1 depicts three polarization mechanisms associated with the
polarization of a mineral grain coated by an electrical double layer while the chargeability is defined by M = 1 − σ0 /σ∞ (Kemna
coating the surface of the grains (e.g. Wang & Revil 2010). They in- 2000), c denotes to the Cole–Cole exponent (0 ≤ c ≤ 1), τ0 de-
clude (1) the deformation of the diffuse layer, which occurs at very notes the characteristic relaxation time (or time constant) and σ0
high frequency (>10 kHz, see the so-called Debye–Falkenhagen and σ∞ denote the low-frequency and high-frequency asymptotic
effect, Falkenhagen 1934), (2) the polarization of the Stern layer limits of the electrical conductivity, respectively. This model was
(Schwarz 1962; Leroy et al. 2008) and (3) the diffusion or mem- derived from the paper of Cole & Cole (1941) in which the Cole–
brane polarization, which is related to the difference in the trans- Cole expression was derived to describe the complex permittivity
ference number for the cations and anions during their migration of polar fluids. Florsch et al. (2012) noted that the Cole–Cole model
in the porous material (Marshall & Madden 1959; Bücker & Hördt in complex conductivity is not equivalent to the Cole–Cole model
2013). It seems that the main polarization mechanism is associated used by Pelton et al. (1978) to describe the complex resistivity of
with the Stern layer since we still observe a strong polarization at porous rocks but some relationships can be drawn between the two
1018 A. Revil, N. Florsch and C. Camerlynck

models (see Florsch et al. 2012; Tarasov & Titov 2013, for further
details). In the following, we will call ‘Warburg model’, a special
case of the Cole–Cole model described above by eq. (1) with c fixed
to 0.5. We will call ‘Debye model’, a special case of the Cole–Cole
model described above with c fixed to 1.
In the frequency dependent conductivity model obtained through
a volume-averaging approach by Revil (2013a,b), the low- and high-
frequency conductivities entering eqs (1) and (2) are given by:
 
1 1
σ0 = σw + ρ S β(+) (1 − f )CEC, (3)
F Fφ
 
1 1  
σ∞ = σw + ρ S β(+) (1 − f ) + β(+)
S
f CEC, (4)
F Fφ

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

where F denotes the electrical formation factor (dimensionless),
φ denotes the connected porosity (dimensionless), σw (in S m−1 )
corresponds to the pore water conductivity, f (dimensionless) de-
notes the fraction of counterions in the Stern layer (typically ∼0.90 Figure 2. Phase spectrum for glass beads having the same size but different
for clays, see Revil 2013a,b), ρS denotes the mass density of the roughness. The low-frequency polarization is dominated by the response
solid phase (typically 2650 kg m−3 for clay minerals and silica), of the grains (suspensions) or the pores (dense porous material). The inter-
β(+) corresponds to the mobility of the counterions in the diffuse mediate frequency range is clearly dominated by the effect associated with
layer [e.g. β(+) (Na+ , 25 ◦ C) = 5.2 × 10−8 m2 s−1 V−1 ], β(+) S
de- the roughness of the grains. The high frequency polarization is associated
notes the mobility of the counterions in the Stern layer β(+) S
(Na+ , with dielectric effects (Maxwell–Wagner polarization including the high-
frequency dielectric limit, see discussion in Revil 2013a). Data from Leroy
25 ◦ C) = 1.5 × 10−10 m2 s−1 V−1 for clay minerals and β(+) S
(25 ◦ C, et al. (2008), Cf denotes the salinity of the pore water solution (NaCl) and
Na+ ) = 5.2 × 10−8 m2 s−1 V−1 for silica grains, Revil 2013a,b) ω the pulsation frequency.
and CEC denotes the cation exchange capacity of the material (ex-
pressed in C kg−1 ). The cation exchange capacity denotes the total where |q(+) | is the charge of these counterions (e.g. |q(+) | = e
amount of cations that can get sorbed on the surface of a mineral at for Na+ where e denotes the elementary charge). The value
a given pH (generally ∼7) and the product (1 − f )CEC denotes the β(+)
S
(Na+ , 25 ◦ C) = 1.5 × 10−10 m2 s−1 V−1 yields D(+) S
(Na+ ,
◦ −12 2 −1
amount of cations (per mass of grains) located in the diffuse layer 25 C) = 3.8 × 10 m s for clay minerals (see Revil 2013a,b).
whereas f CEC denotes the amount of cations (per mass of grains) For the clean sands and sandstones, the mobility of the sodium
located in the Stern layer. in water [β(+) (Na+ , 25 ◦ C) = 5.2 × 10−8 m2 s−1 V−1 ] leads to a
S
diffusion coefficient of D(+) (Na+ , 25 ◦ C) = 1.32 × 10−9 m2 s−1 .
2.3 Relation between the -Parameter and the Cole–Cole
relaxation time 2.4 Cole–Cole model versus experimental data
Johnson et al. (1986) introduced a dynamic pore throat size
that We compare now the previous model with experimental data.
can be determined from the distribution of the electrical potential in We first plot the phase lag for a pack of glass beads in the frequency
a pore network in absence of surface conductivity (e.g. Bernabé & range 3 mHz to 45 kHz (Fig. 2). These glass beads have identical
Revil 1995). Revil et al. (2012b) found that the Cole–Cole relaxation particle sizes. The low-frequency polarization has previously asso-
time τ0 entering eq. (1) seems to be controlled by this pore size
ciated either with grain sizes (Revil & Florsch 2010) or with pore
(rather than the grain size) according to, sizes (Revil et al. 2012b; Bücker & Hördt 2013, this work). Note

2 that the phase scales as ω1/2 at low frequencies. We will show later
τ0 = S , (5) that this behaviour is that predicted by a Warburg function. Fitting
2D(+)
these data with a Cole–Cole model, we obtain a Cole–Cole expo-
S
where D(+) denotes the diffusion coefficient of the counterions in the nent of 0.48 ± 0.3 (Fig. 2). The intermediate frequency range seems
Stern layer (the layer of sorbed counterions on the mineral surface, controlled partly by the polarization of heterogeneities associated
see Revil 2012, 2013a,b and Fig. 1). with the roughness of grain surfaces (see Leroy et al. 2008). The
Katz & Thompson (1987) developed a relationship between the phase at high frequencies scales as ω and this behaviour is a char-
permeability at saturation k and the percolation length scale rc acteristic of the high frequency dielectric response −iωε∞ where
using percolation principles k = rc2 /(226F). A comparison with ε∞ denotes the high-frequency dielectric permittivity.
k =
2 /8F (Johnson et al. 1986; Avellaneda & Torquato 1991) In Figs 3(a) and (b), we plot the complex conductivity spectrum
yields rc ≈ 5.3
. Using
2 = 8Fk and
= rc /5.3 in eq. (5) of a fine sand characterized by a narrow grain size distribution with
yields: a mean of 100 µm. Despite the fact that the grain size distribution of
rc2 this sand is narrow, it is not as narrow as for the glass beads used for
τ0 ≈ S . (6) the measurements displayed in Fig. 2. Fig. 3(b) shows that a portion
56D(+)
of the spectrum is not well fitted by a Cole–Cole model because of
The diffusion coefficient entering into the expression of the the potential effect of the roughness of the grain-pore water interface
Cole–Cole time constant (eqs 5 and 6) can be related to the mo- between the main polarization frequency and the high-frequency
bility of the counterions in the Stern layer entering eqs (3) and dielectric effect (see Fig. 2). Fig. 3(c) shows a similar attempt to
(4), β(+)
S S
, by the Nernst–Einstein relationship D(+) = kb Tβ(+)
S
/|q(+) |, fit the data but without considering the data comprised between
 Influence of pore size on induced polarization 1019

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

Figure 3. In-phase and quadrature conductivities of a clean sand with a very narrow grain size distribution with a mean diameter of 100 µm (data from
Vaudelet et al. 2011a). (a, b) For the quadrature conductivity, the data comprised between the low-frequency peak of the spectrum and the high-frequency
dielectric effect is not fitted because of the potential effect of the roughness of the grain-pore water interface in this domain as shown in Fig. 2. ε 0 denotes the
dielectric constant of vacuum. (c) Fit of the quadrature conductivities data from Vaudelet et al. (2011a) not taking into account the data potentially influenced
by the roughness of the grains. Grain diameter 100 µm, formation factor 3.1 and permeability 2.85 × 10−12 m2 . The pore size of the sand is roughly 8.4 µm.
(d) Microphotograph of the sand.

the peak frequency and the high frequency dielectric behaviour. different pore size distributions. Very sharp particle size distribution
The data are pretty well fitted with a Cole–Cole model with a (PSD) corresponds to PSD distributed over less than one decade
c-exponent of c = 0.39, therefore much lower than the value c = 1 while very broad PSD corresponds to the PSDs spreading over
that would be associated with a Debye model and a bit smaller than three decades. For very well-sorted sands, c is equal to 0.40 ± 0.12
the one found for the glass beads (c = 0.48). We also observe the (11 samples with the very sharp grain size distributions, see Fig. 4).
high-frequency behaviour in ω in Fig. 3 characterizing dielectric This shows that the exponent c in the Cole–Cole model does not
behaviour. take the full range of values between 0 and 1 but is restricted to
From the previous figures, it is possible that the broadness of the the range 0 and 0.5. The upper bound, c = 0.5 is consistent with
polarization is associated with the broadness of the grain or pore a Warburg model and not with a Debye model for which c = 1.
size distributions. We test now this idea using a broader database When the distribution of pore sizes is very broad, the exponent
of experimental data. Fig. 4 shows the cementation exponent m c has a tendency to be close to zero (flat distribution) while m
entering Archie’s law F = φ −m between the formation factor F and increases to values around 2. In conclusion, for very narrow grain
the connected porosity φ (Archie 1942) and the Cole–Cole exponent size distribution (all the grains have essentially the same size),
c distribution for a variety of sands and sandstones characterized by we would expect c to be close to 1 according to the Debye model
1020 A. Revil, N. Florsch and C. Camerlynck

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

Figure 5. Cole–Cole relaxation time τ 0 versus the pore size
for clean
sands. For the data from Koch et al. (2011), the pore size is determined
from the median grain size and the formation factor using the relationship
Figure 4. Cross-plot of the cementation exponent (m > 1) entering Archie’s developed by Revil & Florsch (2010). Pore size is determined from mercury
law F = φ −m in which F is the intrinsic formation factor corrected for intrusion porosimetry (Binley et al. 2005), from rc = 5.6 (8 k /φ 2 )1/2 for the
surface conductivity (Archie 1942) and the Cole–Cole exponent c (according data of Tong et al. (2006a) and from permeability for the data of Vaudelet
to the Warburg model discussed in the main text, 0 ≤ c ≤ 0.5). There are two et al. (2011a,b). For the clean sands and sandstones, the diffusion coefficient
extreme end-regions denoted as A and B. Domain A is typical of well-sorted S
is D(+) (Na+ , 25 ◦ C) = 1.32 × 10−9 m2 s−1 .
clean sands while domain B is typical of clay-rich materials. The plain line
denotes the trend between m and c. Data from Revil et al. (2013b, saprolites
and clayey andstones) and Vinegar & Waxman (1984, samples 3477, 3336A,
3478, 101 and 102).

but we actually find that c is always smaller than ∼0.50 as shown in

Figs 3 and 4.
We discuss now the relationship between the Cole–Cole relax-
ation time and the main pore size for both clay-free and clayey
granular porous materials using a broad database of published ex-
perimental data. We plot the Cole–Cole relaxation time versus the
main pore size data for clean sands and sandstones (Fig. 5) and
for shaly sands and sandstones (Fig. 6). We note that the trends
are linear in a log–log scale with a power-law coefficient of two as
predicted by eqs (5) and (6). The second observation is that the two
trends are distinct indicating that the diffusion coefficient for the
counterions in the Stern layer is not the same on the surface of pure
silica and clays, in agreement with the observations made by Revil
(2013a,b). Figure 6. Cole–Cole relaxation time τ 0 versus the pore size for clayey
We fit the data shown in Fig. 5 with the Cole–Cole model (plain and shaly porous media. For the clayey material, the diffusion coefficient
line) calculated using a mobility of 5.2 × 10−8 m2 s−1 V−1 (hence S
is estimated to be equal to D(+) (Na+ , 25 ◦ C) = 3.8 × 10−12 m2 s−1 and
a diffusion coefficient D(+)S
(Na+ , 25 ◦ C) = 1.32 × 10−9 m2 s−1 in is determined from the mobility of the counterions in the Stern layer (see
eq. 6) for the counterions in the Stern layer equal to the mobility Revil 2013 and discussion in the main text). Data from Scott & Barker
(diffusion coefficient) in water. There is a fair agreement between the (2003), Revil et al. (2013a), Comparon (2005), Lesmes & Frye (2001) and
model and the data. In Fig. 6, we perform the same comparison for this work. The
-parameter is determined from the permeability and the
formation factor using
2 = 8Fk.
clayey sands and sandstones. Again the model fits the data very well
with the model computed with a diffusion coefficient of D(+) S
(Na+ ,
◦ −12 2 −1
25 C) = 3.8 × 10 m s determined from the ionic mobility
3 E L E C T R I C A L P O R O S I M E T RY O F
inferred by Revil (2012, 2013a) for clay minerals (see Appendix A
S A N D S T O N E S : T H E O RY
for the description of the in-phase and quadrature components of the
complex conductivity). Our model indicates that the mobility of the
3.1 Rationale for a Warburg transfer function
counterions of clays is roughly 300 times smaller on the surface of
clay minerals than the value obtained for silica sands (see discussion Warburg (1899) developed the first solution to the diffusion equation
above at the end of Section 2.3). with oscillating concentration as boundary condition. His approach
 Influence of pore size on induced polarization 1021

has been subsequently used in the literature to model a variety of the polarization shown in Fig. 1 can be seen as a leaky capacitance
polarization mechanisms. The complex conductivity of a Warburg characterized by a Warburg impedance model.
model can be written as (ii) If we adopt a Debye decomposition of spectra to obtain relax-
Mn ation times, the associated distribution of polarization length scale
σ ∗ = σ∞ − . (7) seems always much broader than the true distribution of polarization
1 + (iωτ0 )1/2
length scales (e.g. Leroy et al. 2008; Vaudelet et al. 2011a). This
Eq. (7) shows that the Warburg model is a special case of the observation is consistent with the other observation made above
Cole–Cole model with an exponent of c = 0.5. At low-frequencies, that the spectral induced polarisation data of glass beads and silica
the phase or the quadrature conductivities scales as (ω)−1/2 as shown sands fitted with a Cole–Cole model show a Cole–Cole exponent
in Fig. 2 for glass beads with a uniform grain size. When decom- always smaller or equal to 0.50. In both cases, the Debye model is
posed with a Debye model, the (normalized) distribution of relax- not the model describing the behaviour of a porous material with
ation times for the Cole–Cole model is (Cole & Cole 1941) all the grains being the same and the observations point out that the
1 sin [π (1 − c)] Warburg model is the correct model to be used in deconvolving the
g(τ ) =  
. (8)
2π τ cosh c ln τ − − complex conductivity data.

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

τ0
cos [π (1 c)]

In the special case of the Warburg model, we have therefore:

  3.2 Relation between the relaxation time distribution
1 1
g(τ, c = 0.5) =  
, (9) (RTD) and the pore size distribution
2π τ cosh 1 ln τ
2 τ0 We consider the case of a sandstone characterized by a (normal-
ized) pore size probability distribution f (r ) where r is a given
 1/2
τ pore radius. We assume that this pore size distribution f (r ) is as-
  sociated with a (normalized) distribution of relaxation times h(τ ).
1 τ0
g(τ, c = 0.5) = . (10) Assuming,
π τ + τ0
The idea we follow in the remaining part of the paper is that r2
τ= S , (14)
the Warburg function, rather than the Debye function, is the correct 2D(+)
transfer function to determine the pore size or the pore size distri- S
bution. In the case that all the pores have the same size, a single (D(+) denotes the diffusion coefficient of the counterions), the two
pore size should be reflected by a single relaxation time, that is probability density functions f (r ) and h(τ ) are related to each other
the Warburg function should be analysed in terms of a distribution by their probability distributions following the same type of analysis
of relaxation times h(τ ) given by h(τ ) = δ(τ − τ0 ) (i.e. a Dirac made by Revil & Florsch (2010). This yields,
function). f (r )
τ h(τ ) = , (15)
The relationship between the distribution of relaxation times ob- 2Fh
tained with Debye and Warburg decompositions of a given complex
with,
conductivity spectrum is discussed further in Appendix B. Using  +∞
the convolution theorem based on the superposition principle, the
Fh = f (r ) d ln r. (16)
complex conductivity is given by, 0
 ∞
h(τ ) The parameter Fh can be seen as a normalization constant. Eqs
σ ∗ = σ∞ − Mn dτ, (11)
0 1 + (iωτ )1/2 (14) and (15) can be used to obtain an explicit function of the
relaxation time using the distribution of pore sizes,
The distribution of the intrinsic relaxation times, h(τ ), corresponds  S
to a normalized probability density and therefore obeys to D(+) f (r )
 ∞ h(τ ) = . (17)
Fh r2
h(τ ) dτ = 1. (12)
0
Eq. (17) can be used to compute the distribution of relaxation
We believe that there are two reasons to choose the Warburg times from the distribution of pore sizes. Alternatively, pore size
function as transfer function rather than the Debye function: distribution can be constrained from the distribution of relaxation
(i) There is a group of theoretical transfer functions based on times inverted from complex conductivity spectra. In a second step,
mechanistic models (Wong 1979; de Lima & Sharma 1992; Dukhin this distribution can be inverted to retrieve f (r ). Such an inversion
& Shilov 2002) describing the polarization of charged colloidal par- would allow getting the pore size distribution from SIP data in the
ticles and granular materials (all with the same polarization length same way it is presently performed with nuclear magnetic resonance
scale) with a transfer function close to a Warburg transfer function. (NMR, see for instance Timur 1969; Weller et al. 2010). In the next
The Warburg model is usually used to model a leaky capacitance. section, we explain how we can perform a deconvolution of SIP
In our case, this would mean that the grains coated by the Stern spectra and use the result to determine the pore size distribution.
layer behave as leaking capacitances. If we consider for instance a
sand saturated by a NaCl solution, the controlling reaction for the
3.3 Warburg decomposition of SIP spectra
interfacial impedance corresponds to the sorption/desorption of the
sodium on the silica surface (Fig. 1b): We consider below that a porous rock is a linear system in terms of
the relationship between the pore size distribution and the resulting
> SiO− Na+ ⇔ > SiO− + Na+ , (13)
induced polarization spectra. We have seen in the previous sections
−
where ‘>’ refers to the crystalline framework and >SiO denotes that if the pore size distribution is described by a delta function,
the negative site at the surface of the silica SiO2 . It follows that the associated RTD will be described by a delta function, and the
1022 A. Revil, N. Florsch and C. Camerlynck

associated impulse function of the system is a Warburg function. L-curve to a very similar problem but based on the Debye model).
Therefore the pore size distribution can be recovered by perform- We can also use a variant of the method proposed by Florsch et al.
ing a deconvolution of the induced polarization spectra using this (2012) solving the problem in the Fourier space. A third possibil-
impulse or Green’s function. We propose below a methodology to ity is to transform the result of the Debye decomposition into a
perform this task, which is based on a generalization of the approach Warburg decomposition distribution of relaxation times. This third
proposed recently by Florsch et al. (2012) who used the Debye func- idea is explored in Appendix B. In Appendix B, we derive a formula
tion to perform the deconvolution of complex conductivity spectra. (eq. B17) that can be used to transform the distribution of relax-
We start by using a complete expression of the effective complex ation time obtained from a Debye deconvolution to a distribution
conductivity derived above in eq. (11) and adding the dielectric of relaxation time obtained from a Warburg deconvolution and vice
polarization (electric displacement) to the complex conductivity: versa. This formula can be very useful to researchers using existing
 ∞ codes based on the Debye deconvolution of the complex conduc-
∗ h(τ )
σeff = σ∞ − Mn dτ + iωε∞ , (18) tivity spectra and who want to transform their results in terms of a
0 1 + (iωτ )1/2 distribution of relaxation times using a Warburg deconvolution.
In the following, the set {Mn , h(τ )} is replaced by a non- Once we have obtained the distribution of the relaxation times,

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

normalized function h̄(τ ) (to be inverted after discretization, see we still need to determine the pore size distribution. We start with
Florsch et al. 2012, Appendix A): the relationship,
 ∞
∗ h̄(τ ) r2
σeff − iωε∞ = σ∞ − dτ. (19) τ= , (26)
0 1 + (iωτ )1/2 S
a D(+)
Using the following transformation of variables (Florsch et al. where r is a characteristic pore size, a is a constant. It follows:
2012): 
 r = τ a D(+)S
, (27)
z = − ln(ω) ⇔ ω = e−z
, (20) Actually, h(τ ) is not needed, only the logarithm of the distribution:
s = ln(τ ) ⇔ τ = es
H (s) = τ h(τ ) is required. The independent variable s is given in
Eq. (19) is written in a convolutive form: log10 (we use the base 10 by choice). Therefore the transformation
 ∞ from τ to r is just an affine transformation:
∗
σ = σ∞ − G s (s) H̄ z (z − s)ds. (21)
−∞ 1  S 
log10 r = log10 τ + log10 a D(+) . (28)
with, 2
⎧ Therefore, we can use directly the RTD in log10 τ to determine the
⎨ G s (s) = τ h(τ )
⎪ distribution of pore sizes in log10 r . This transformation corresponds
. (22) only to a translation along the x-axis and a factor 1/2 applied to the
⎪
⎩ H̄ (z, s) = 1
scaling. A comparison between this procedure and experimental
1 + iec(z−s)
data is provided in Section 4.
The discretization of eq. (21) on NG points leads to equations
that corresponds to a Riemann sum with step s. By separating the
real and imaginary parts, we obtain: 4 E L E C T R I C A L P O R O S I M E T RY:
⎧
⎪
⎪ 
NG C O M PA R I S O N W I T H E X P E R I M E N TA L
⎪
⎪ σ ∗
= σ − s k j G j
⎪
⎨ k ∞ D ATA
j=1
(23)
⎪
⎪ 
NG
4.1 Core samples
⎪
⎪ ∗
σ = − s
k j G j + Ce −z k
⎪
⎩ k
j=1 Six siliciclastic samples were collected from the Great Divide Basin
(Wyoming, USA). Five samples are sandstones with angular to
At this stage the system of equations is linear, but it should
subrounded grains and sample #439 denotes a mudstone (Table 1).
include also the fundamental property that the RTD is a probability
The samples have been chosen because they are characterized by
function and must be necessarily positive. This property is insured
a broad variability in pore sizes. Their permeability ranges from 3
by an additional change of variable:
to 1600 mD and their porosity ranges from 0.21 to 0.31 (Table 1).

∀ j, G j = e G j . (24) The formation factors were obtained through classical conductivity
measurements at different salinities (see for instance Revil 2012)
leading to the final model in G j : and were corrected for surface conductivity.
⎧
⎪
⎪ 
NG
⎪ σ ∗ = σ∞ − s
⎪

k j · e G j
⎪
⎨ k
4.2 Complex conductivity measurements
j=1
. (25)
⎪
⎪  NG
The six samples were saturated under vacuum with the natural
⎪
⎪ σ ∗ = − s

 k j · e G j + Ce−zk
⎪
⎩ k ground water saturating the formations from which they were
j=1
extracted. The pore water conductivity of this ground water is
To solve that set of non-linear equations, we use the ‘non-linear 0.048 ± 0.013 S m−1 at 25 ◦ C and its composition is reported in
generalized least-square method’ developed by Tarantola & Valette Table 2. The complex conductivity measurements were taken using
(1982). To optimize the damping factor devoted to smooth the so- the ZEL-SIP04-V02 impedance meter at 25 frequencies from 1 mHz
lution of this ill-posed problem (see Tikhonov 1977), we use the to 45 kHz (see Fig. 7 for the experimental setup). This impedance
L-curve method (see Florsch et al. 2012 for the application of the meter was built by Egon Zimmermann in Germany (Zimmermann
 Influence of pore size on induced polarization 1023

Table 1. Description of the core samples in terms of porosity φ, permeability k, formation factor F, pore
size
= (8k F)1/2 and Cole–Cole relaxation time τ0 from the measured SIP spectra. The measured grain
density is in the range 2596–2620 kg m−3 . All the samples are sandstones except Sample #439, which
is a mudstone. 1 mD–10−15 m2 . The pore sizes rc (50) and rc (peak) denote the median of the pore size
distribution and the value of the pore radius corresponding to the peak of the distribution, respectively.
Sample φ k F
τ0 rc (50) rc (peak) τ0
(−) (mD) (−) (µm) (1) (s) (2) (µm) (3) (µm) (4) (s) (5)
S499 0.265 1103 5.6 ± 0.4 7.0 2.6 ± 0.2 14.3 22 2.3
S498 0.206 35.9 9.0 ± 0.8 1.6 0.20 ± 0.01 1.98 4.5 0.10
S490 0.233 635 12.1 ± 0.4 7.8 3.2 ± 0.3 12.2 25 2.9
S493 0.232 115 18.3 ± 0.4 4.1 0.41 ± 0.04 4.4 6 0.17
S439 0.208 2.62 13.3 ± 0.7 0.53 0.023 ± 0.001 0.52 0.9 0.004
S436 0.306 1623 4.0 ± 0.3 7.2 25.5 ± 3.6 13.2 55 14.2
Notes: (1) Determined using
2 = 8Fk with k determined from the capillary entry pressure (see Fig. 15);

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

(2) using a Cole–Cole fit of the spectral induced polarization spectra; (3) median of the pore size dis-
tribution from Hg porosimetry; (4) peak of the pore size distribution and (5) predicted value using
τ0 = rc (peak)2 /(56D(+)
S S
) with D(+) (Na+ , 25 ◦ C) = 3.8 × 10−12 m2 s−1 .

Table 2. Composition of the natu- bell shapes. At the relaxation peak, the phase is in the range –6
ral groundwater for the sandstones. (mudstone, #439) to –38 mrad (Sample #499). The peak of the
TDS, total dissolved solids. phase lag occurs in the range 2 mHz (Sample #436) to 20 Hz
Parameter Units Value (mudstone, #439). The electrical conductivities are in the range
0.02–0.12 S m−1 .
TDS mg l−1 318
Conductivity µS cm−1 479
pH – 8.1
Alkalinity mg l−1 109
4.3 Pore size distribution
Na+ mg l−1 30.6 Mercury porosimetry measurements were performed on the six
K+ mg l−1 3.9 samples (Figs 9a–14a). Thin sections of the core samples are also
Ca2+ mg l−1 65.0
shown in Figs 9(b)–14(b). The permeability reported in Table 1
Mg2+ mg l−1 3.1
was determined from the mercury intrusion experiments using a
Cl− mg l−1 6.0
HCO3 − mg l−1 123 modified Swanson (1981) approach derived by Revil et al. (2014)
SO4 2− mg l−1 132 based on the capillary entry pressure pe (in Pa),
 
2γ φ
pe = √ √ , (29)
226 k
where φ denotes the connected porosity, γ represents the surface
tension between water and air (71.99 ± 0.05) × 10−3 N m−1 and k
the permeability at saturation. We test the proportionality
√ between
the capillary entry pressure and the ratio φ/ k in Fig. 15 (the
proportionally coefficient is fixed by the theory). Eq. (20) is used to
predict the permeability from the porosity and the capillary entry
pressure to Mercury for the different samples. When using use the
following formula
2 = 8Fk to determine the pore scale
we
obtain
≈ (0.38γ )/ pe , which is used to determine
using the
capillary entry pressure.

4.4 Data analysis

Figure 7. Experimental setup. (a) Position of the current (A and B) and We found that the induced polarization spectra cannot be fitted
voltage (M and N) electrodes and sensitivity map (assuming the sam-
by the Cole–Cole model. That said, the portion of the induced
ple is homogeneous), which shows that there is a good sensitivity of the
polarization data in the vicinity of the relaxation peak can be
measurement over the cross-section area of the column. The sensitivity
for the electrode array ABMN was computed in Vaudelet et al. (2011b). fitted very well by the Cole–Cole model described in Section 2.1
(b) ZEL-SIP04-V02 impedance meter build by Egon Zimmerman. and Appendix A (using eqs A4 and A5). The corresponding fits are
shown by Figs 9(c)–14(c) (the non-linear optimization is done with
et al. 2008a,b). The accuracy was ∼0.1–0.3 mrad at frequencies the least-square criterion using the Gauss–Newton method, see Chen
<1 kHz. Ag–AgCl electrodes were used for both injection and po- et al. 2008). The values of the Cole–Cole parameters [normalized
tential electrodes and the electrode array was circumferential around chargeability Mn , Cole–Cole exponent c and Cole–Cole relaxation
the cylindrical core samples (Fig. 7). Benchmark tests of this equip- time τ 0 are shown on the Figs 9(c)–14(c), see also Table 1 for the
ment can be found in Revil & Skold (2011). value of τ 0 ].
The spectra are shown in Fig. 8. The magnitude of the conductiv- In Fig. 16, we plot the Cole–Cole relaxation time with the pore
ity increases with frequency and the phase shows well-characterized scale
(determined from the formation factor and the permeability,
1024 A. Revil, N. Florsch and C. Camerlynck

(a)
-40
#499 -
-
#439-Mudstone
-35 - -
#493-Sandstone
#498-Sandstone
-30 #490 #436-Sandstone
Phase (in mrad)

#490-Sandstone
-25 #499-Sandstone

-20
#498
-15
#436
-10 #493

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

-5 #439-Mudstone

0 0.001 0.01 0.1 1.0 10 100 1000 10,000

(b) Frequency (in Hertz)

0.12
#439-Mudstone

0.10 #493
Conductivity (in S/m)

0.08
#498 Figure 9. Pore size distribution and quadrature conductivity. (a) Pore size
distribution of Sample #499 (clayey sandstone). (b) Thin section. (c) The
0.06 quadrature conductivity data (filled circles) was only fitted with a Cole–Cole
model (plain line) for the frequencies around the peak frequency (R2 = 0.99).
0.04
#490

0.02 #499
#436

0
0.001 0.01 0.1 1.0 10 100 1000 10,000
Frequency (in Hertz)
Figure 8. Phase and magnitude of the complex conductivity for the six
samples investigated in this study. (a) Phase lag angle in mrad. (b) Magnitude
of the conductivity in S m−1 . For the mudstone, the phase data at low and
high frequencies are not shown because they were characterized by a high
standard deviation. The lines are guides for the eyes.

see Johnson et al. 1986), the median of the pore size distribution
from mercury porosimetry and the value of the pore size corre-
sponding to the peak of the pore size probability distribution. We
see that the Cole–Cole relaxation time is best correlated (R2 = 0.94)
with the value of the peak of the pore size probability distribution.
The relaxation time predicted by eq. (19) in Table 1 predicts fairly
well the Cole–Cole relaxation time (R2 = 0.89 in a linear–linear
plot).
The results of the inversion of the spectra are reported in terms of
relaxation times and pore size distributions (Figs 17–19). In these
figures, we see both the results of the Debye and Warburg decompo-
sition directly in terms of pore size distribution using the transform
developed in Section 3.3. The Debye and Warburg decompositions
fit the data equally well (Figs 17–19). Figure 10. Pore size distribution and quadrature conductivity. (a) Pore size
The pore size distributions resulting from the Debye decomposi- distribution of Sample 490 (clayey sandstone). (b) Thin section. (c) The
tion (c = 1 in the equations of Section 3.3) are generally too broad quadrature conductivity data (filled circles) was only fitted with a Cole–Cole
with respect to the pore size distributions obtained from mercury model (plain line) for the frequencies around the peak frequency (R2 = 0.97).
 Influence of pore size on induced polarization 1025

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

Figure 11. Pore size distribution and quadrature conductivity. (a) Pore Figure 13. Pore size distribution and quadrature conductivity. (a) Pore size
size distribution Sample #493 Clayey sandstone). (b) Thin section. (c) The distribution of Sample #436 (clayey sandstone). (b) Thin section. (c) The
quadrature conductivity data (filled circles) was only fitted with a Cole–Cole quadrature conductivity data (filled circles) was only fitted with a Cole–Cole
model (plain line) for the frequencies around the peak frequency (R2 = 0.98). model (plain line) for the frequencies around the peak frequency (R2 = 0.93).

Figure 12. Pore size distribution and quadrature conductivity. (a) Pore size Figure 14. Pore size distribution and quadrature conductivity. (a) Pore size
distribution of Sample #498 (clayey sandstone). (b) Thin section. (c) The distribution of Sample #439 (mudstone). (b) Thin section. (c) The quadrature
quadrature conductivity data (filled circles) was only fitted with a Cole–Cole conductivity data (filled circles) was only fitted with a Cole–Cole model
model (plain line) for the frequencies around the peak frequency (R2 = 0.98). (plain line) for the frequencies around the peak frequency (R2 = 0.99).
1026 A. Revil, N. Florsch and C. Camerlynck

information on small pores seems lost due to the high-frequency

dielectric behaviour.
Note that hat the Debye function is sharper in frequency domain
and thus is expected to yield broader distributions in relaxation
times and pore size and such broad pore size distributions in the
pore sizes are not supported by the data. In Fig. 20, we compare the
variance of the distributions determined from induced polarization
porosimetry and mercury porosimetry. We see that SIP porosimetry
is doing a fair job in reproducing the broadness of the pore size
distributions, at least the main peak since the small pores cannot
been observed through SIP porosimetry.

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

5 C O N C LU S I O N S
The following conclusions have been reached.

(1) The distribution of relaxation times seems to be controlled by

the distribution of the pore sizes. This observation is still not fully
understood by the current mechanistic models. Pore scale numerical
modelling may be needed to evaluate the ability of the local pore
scale physics to reproduce these observations. The pore scale model
developed recently by Bücker & Hördt (2013) may be a step in this
Figure 15. Test of the relationship between the capillary entry pressure as direction.
a function of the porosity and permeability. The experimental data are from (2) For a collection of five sandstones and one mudstone core
Huet et al. (2005). They correspond to 89 sets of mercury-injection (Hg-air) samples, we found that the Cole–Cole relaxation time is propor-
capillary pressure data. Core samples include both carbonate and sandstone tional to the characteristic pore size
squared and to the pore size
lithologies. The permeability is here expressed in mD. squared determined from the capillary entry pressure. The Cole–
Cole relaxation time seems also correlated to the peak of the pore
size distribution determined from Hg porosimetry.
porosimetry. The Warburg decomposition (c = 0.5 in the equa- (3) The diffusion coefficients of the counterions moving along
tions of Section 3.3) yields pore size distributions that are nar- the Stern layer of pure silica sands and clays are consistent with the
rower and compatible with the observed pore size distribution from mobility of the Stern layer polarization (POLARIS) model devel-
mercury porosimetry. The pore size distributions determined from oped by Revil (2012, 2013a,b). This shows the consistency in the
Hg-porosimetry are shown in terms of the peak (with the vertical model between the components describing the relaxation time and
bar) and the width of the distribution at half the value of the peak those needed to compute the amplitude of the quadrature conduc-
(horizontal bar). The Warburg decomposition is generally repro- tivity at high and low frequencies. To the best of our knowledge,
ducing the maximum and the width of the distribution from Hg there is no other model able to reproduce such a consistency.
porosimetry. That said, the details of the pore size distribution for (4) The variance as well as the peak of the pore throat size dis-
small pore sizes are lost showing the limitations of the method. The tribution can be recovered from the SIP data. Finer structures of

Figure 16. Relationship between the Cole–Cole relaxation time and two different pore sizes reported in Table 1. (a) Correlation between the Cole–Cole
relaxation time and the pore size
determined using
2 = 8Fk (R2 = 0.90) and k determined from the capillary entry pressure. (b) Correlation between the
Cole–Cole relaxation time and the peak of the pore size distribution determined from the mercury intrusion experiments (R2 = 0.94). The best correlation
coefficient is obtained with the pore size determined from the mercury porosimetry using the pore size corresponding to the peak of the distribution.
 Influence of pore size on induced polarization 1027

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

Figure 17. Fit of the SIP data of sample 498. (a) Fit of the amplitude (Debye decomposition). (b) Fit of the phase (Debye decomposition). (c) Distribution of
pore throat sizes (Debye decomposition). (d) Fit of the amplitude (Warburg decomposition). (e) Fit of the phase with a Warburg decomposition. (f) Distribution
of pore throat sizes (Warburg decomposition). (g, h, i, j, k, l) Same for Sample 493. The pore size distributions determined from Hg-porosimetry are shown in
terms of the peak (with the vertical bar) and the width of the distribution at half the value of the peak (horizontal bar).
1028 A. Revil, N. Florsch and C. Camerlynck

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

Figure 18. Fit of the SIP data of sample 490. (a) Fit of the amplitude (Debye decomposition). (b) Fit of the phase (Debye decomposition). (c) Distribution of
pore throat sizes (Debye decomposition). (d) Fit of the amplitude (Warburg decomposition). (e) Fit of the phase (Warburg decomposition). (f) Distribution of
pore throat sizes (Warburg decomposition). (g, h, i, j, k, l) Same for Sample 439 (mudstone).
 Influence of pore size on induced polarization 1029

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

Figure 19. Fit of the SIP data of sample 499. (a) Fit of the amplitude with a Debye decomposition. (b) Fit of the phase (Debye decomposition). (c) Distribution
of pore throat sizes with the Debye decomposition. (d) Fit of the amplitude with the Warburg decomposition. (e) Fit of the phase (Warburg decomposition).
(f) Distribution of pore throat sizes with the Warburg decomposition. (g, h, i, j, k, l) Same for Sample 436.
1030 A. Revil, N. Florsch and C. Camerlynck

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

Figure 20. Comparison between the inverted pore size distribution using the SIP data (black filled circles) and the pore size distributions from the mercury
porosimetry (peak and variance materialized by the vertical and horizontal bars, respectively). In these figures, we have moved the position of the peaks to
match each other.

the pore size distribution, especially for small pore sizes, seem AC K N OW L E D G E M E N T S
more difficult to obtain from this procedure. The Debye decompo-
We thank the National Science Foundation for the project ‘SIP to
sition seems unable to provide the correct variance regarding the
invert permeability’ (in collaboration with Lee Slater at Rutgers
pore size distribution for the six samples presented in this study.
University and Andrew Binley at Lancaster University, NSF award
At the opposite, the Warburg decomposition provides narrower dis-
NSF EAR-0711053) and the Office of Science (BER), US. Depart-
tributions of relaxation times, which translates to distributions of
ment of Energy (award DE-FG02–08ER646559). We thank Egon
pore throat sizes that are compatible with those observed through
Zimmermann for the construction of his impedance meter, Markus
Hg-porosimetry.
Hilpert and two anonymous referees for their constructive com-
ments regarding a previous version of our paper, Andreas Hördt,
The present findings open the door to induced polarization and an anonymous referee for their constructive comments regard-
porosimetry as currently done with NMR. NMR and SIP seem ing this version of the paper.
to be a natural tools to invert pore size distributions not only in
the laboratory but also in the field through joint inversion based on
a petrophysical model. Indeed both techniques are sensitive to the
pore size and we have developed an approach to remove surface
conductivity from the total measured conductivity of a porous rock REFERENCES
by taking advantage of the relationship between surface conductiv- Archie, G.E., 1942. The electrical resistivity log as an aid in determining
ity and quadrature conductivity (Revil 2013b; Weller et al. 2013). some reservoir characteristics, Trans. AIME, 146, 54–62.
In other words, this means we can isolate the formation factor and Attwa, M. & Günther, T., 2012. Application of spectral induced polarization
use the quadrature conductivity, corrected for the formation factor, (SIP) imaging for characterizing the near-surface geology: an environ-
to determine the pore size distribution. Therefore it is possible to mental case study at Schillerslage, Germany, Aust. J. Basic Appl. Sci.,
formulate a joint inversion problem for the SIP and the NMR and 6(9), 693–701.
take advantage of their very different sensitivity maps to get an Attwa, M. & Günther, T., 2013. Spectral induced polarization measurements
for predicting the hydraulic conductivity in sandy aquifers, Hydrol. Earth
image (tomogram) of the pore size distribution. If confirmed with
Syst. Sci., 17, 4079–4094.
additional datasets, this may open new doors in hydrogeophysics to
Avellaneda, M. & Torquato, S., 1991. Rigorous link between fluid perme-
image non-intrusively some characteristics of the pore size distri- ability, electrical conductivity, and relaxation times for transport in porous
bution with a combination of SIP/NMR data. Finally, drainage or media, Phys. Fluids A, 3, 2529–2540.
imbition of a porous material should have a specific signature on the Avena, M.J. & De Pauli, C.P., 1998. Proton adsorption and electrokinetics
distribution of relaxation times and this aspect will be investigated of an Argentinean Montmorillonite, J. Colloid Interface Sci., 202, 195–
in a future contribution. 204.
 Influence of pore size on induced polarization 1031

Bernabé, Y. & Revil, A., 1995. Pore-scale heterogeneity, energy dissipation mercury-injection capillary pressure data, in Proceedings if the Interna-
and the transport properties of rocks, Geophys. Res. Lett., 22(12), 1529– tional Petroleum Technology Conference, Doha, Qatar, 21–23 November
1552. 2005, Paper IPTC 10994, 12 pp.
Binley, A., Slater, L.D., Fukes, M. & Cassiani, G., 2005. Relationship Johnson, D.L., Plona, T.J. & Kojima, H., 1986. Probing porous media with
between spectral induced polarization and hydraulic properties of sat- 1st sound, 2nd sound, 4th sound and 3rd sound, in Physics and Chemistry
urated and unsaturated sandstones, Water Resour. Res., 41, W12417, of Porous Media, Vol. 2, pp. 243–277, eds Jayanth, R., Banavar, J. &
doi:10.1029/2005WR004202. Winkler, K.W., Am. Inst. Phys..
Börner, F.D., 1992. Complex conductivity measurements of reservoir prop- Katz, A.J. & Thompson, A.H., 1987. Prediction of rock electrical conduc-
erties, in Proceedings of the Third European Core Analysis Symposium, tivity from mercury injection measurements, J. geophys. Res., 92(B1),
Paris, pp. 359–386. 599–607.
Bücker, M. & Hördt, A., 2013. Analytical modelling of membrane polariza- Kemna, A., 2000. Tomographic inversion of complex resistivity, theory and
tion with explicit parametrization of pore radii and the electrical double application, PhD thesis, Bochum University, 196 pp.
layer, Geophys. J. Int., 194, 804–813. Kemna, A. et al., 2012. An overview of the spectral induced polarization
Chen, J., Kemna, A. & Hubbard, S.S., 2008. A comparison between Gauss- method for near-surface applications. Near Surface Geophys., 10, 453–
Newton and Markov-chain Monte Carlo-based methods for inverting 468.

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

spectral induced-polarization data for Cole Cole parameters, Geophysics, Koch, K., Kemna, A., Irving, J. & Holliger, K., 2011. Impact of changes
73(6), F247–F259. in grain size and pore space on the hydraulic conductivity and spectral
Chen, J., Hubbard, S.S., Williams, K.H., Flores Orozco, A. & Kemna, A., induced polarization response of sand, Hydrol. Earth Syst. Sci., 15, 1785–
2012. Estimating the spatiotemporal distribution of geochemical param- 1794.
eters associated with biostimulation using spectral induced polarization Leroy, P., Revil, A., Kemna, A., Cosenza, P. & Ghorbani, A., 2008. Spectral
data and hierarchical Bayesian models, Water Resour. Res., 48, W05555, induced polarization of water-saturated packs of glass beads, J. Colloid
doi:10.1029/2011WR010992. Interface Sci., 321, 103–117.
Cole, K.S. & Cole, R.H., 1941. Dispersion and absorption in dielectrics. I. Lesmes, D.P. & Frye, K.M., 2001. Influence of pore fluid chemistry on
Alternating current characteristics, J. Chem. Phys., 9, 341–351. the complex conductivity and induced polarization responses of Berea
Comparon, L., 2005. Etude expérimentale des propriétés électriques et sandstone, J. geophys. Res., 106(B3), 4079–4090.
diélectriques des matérieux argileux consolidés, PhD thesis, Institut de Marshall, D.J. & Madden, T.R., 1959. Induced polarization, a study of its
Physique du Globe de Paris, 400 pp. causes, Geophysics, 24, 790–816.
Davis, C., Atekwana, A., Atekwana, E., Slater, L.D., Rossbach, S. & Nordsiek, S. & Weller, A., 2008. A new approach to fitting induced polar-
Mormile, M.R., 2006. Microbial growth and biofilm formation in ge- ization spectra, Geophysics, 73(6), F235–F245.
ologic media is detected with complex conductivity measurements, Pelton, W.H., Ward, S.H., Hallof, P.G., Sill, W.R. & Nelson, P.H., 1978.
Geophys. Res. Lett., 33, L18403, doi:10.1029/2006GL027312. Mineral discrimination and removal of inductive coupling with multifre-
Deceuster, J. & Kaufmann, O., 2012. Improving the delineation of quency IP, Geophysics, 43, 588–609.
hydrocarbon-impacted soils and water through induced polarization (IP) Revil, A., 2012. Spectral induced polarization of shaly sands: Influ-
tomographies: A field study at an industrial waste land, J. Contam. ence of the electrical double layer, Water Resour. Res., 48, W02517,
Hydrol., 136–137, 25–42. doi:10.1029/2011WR011260.
Deceuster, J., Chaballe, M. & Kaufmann, O., 2005. 3D resistivity and IP Revil, A., 2013a. Effective conductivity and permittivity of unsaturated
tomographies as efficient tools to monitor remedial actions over a gas porous materials in the frequency range 1 mHz–1GHz, Water Resour.
plume, in Proceedings of the Symposium on the Application of Geophysics Res., 49, 306–327.
to Engineering and Environmental Problems (SAGEEP), San Diego, CA, Revil, A., 2013b. On charge accumulations in heterogeneous porous mate-
pp. 155–166. rials under the influence of an electrical field, Geophysics, 78(4), D271–
de Lima, O.A.L. & Sharma, M.M., 1992, A generalized Maxwell–Wagner D291.
theory for membrane polarization in shaly sands, Geophysics, 57, 431– Revil, A. & Florsch, N., 2010. Determination of permeability from spectral
440. induced polarization data in granular media, Geophys. J. Int., 181, 1480–
Dukhin, S.S. & Shilov, V.N., 2002. Nonequilibrium electric surface phenom- 1498.
ena and extended electrokinetic characterization of particles, in Electroki- Revil, A. & Skold, M., 2011. Salinity dependence of spectral induced po-
netics and Electrophoresis, Surfactant Sci. Ser., Vol. 106, pp. 55–85, ed. larization in sands and sandstones, Geophys. J. Int., 187, 813–824.
Delgado, A.V., Marcel Dekker. Revil, A., Karaoulis, M., Johnson, T. & Kemna, A., 2012a. Review: some
Fiandaca, G., Auken, E., Vest Christiansen, A. & Gazoty, A., 2012. Time- low-frequency electrical methods for subsurface characterization and
domain-induced polarization: full-decay forward modeling and 1D later- monitoring in hydrogeology, Hydrogeol. J., 20(4), 617–658.
ally constrained inversion of Cole–Cole parameters, Geophysics, 77(3), Revil, A., Koch, K. & Holliger, K., 2012b. Is it the grain size or the
E213–E225. characteristic pore size that controls the induced polarization relaxation
Florsch, N., Camerlynck, C. & Revil, A., 2012. Direct estimation of the time of clean sands and sandstones?, Water Resour. Res., 48, W05602,
distribution of relaxation times from induced-polarization spectra us- doi:10.1029/2011WR011561.
ing a Fourier transform analysis, Near Surface Geophys., 10, 517– Revil, A., Atekwana, E., Zhang, C., Jardani, A. & Smith, S., 2012c.
531. A new model for the spectral induced polarization signature of bac-
Gazoty, A., Fiandaca, G., Pedersen, J., Auken, E., Christiansen, A.V. & terial growth in porous media. Water Resour. Res., 48, W09545,
Pedersen, J. K., 2012. Application of time domain induced polarization doi:10.1029/2012WR011965.
to the mapping of lithotypes in a landfill site, Hydrol. Earth Syst. Sci. Revil, A., Skold, M., Hubbard, S.S., Wu, Y., Watson, D. & Karaoulis, M.,
Discuss., 9, 983–1011. 2013a. Petrophysical properties of saprolites from the Oak Ridge In-
Ghorbani, A., Camerlynck, C., Florsch, N., Cosenza, P., Tabbagh, A. & tegrated Field Research Challenge site, Tennessee, Geophysics, 78(1),
Revil, A., 2007. Bayesian inference of the Cole–Cole parameters from D21–D40.
time and frequency domain induced polarization, Geophys. Prospect., Revil, A., Eppehimer, J.D., Skold, M., Karaoulis, M., Godinez, L. & Prasad,
55(4), 589–605. M., 2013b. Low-frequency complex conductivity of sandy and clayey
Hördt, A., Blaschek, R., Kemna, A. & Zisser, N., 2007. Hydraulic con- materials, J. Coll. Interface Sci., 398, 193–209.
ductivity estimation from induced polarization data at the field scale-the Revil, A., Barnier, G., Karaoulis, M. & Sava, P., 2014. Seismoelectric cou-
Krauthausen case history, J. appl. Geophys., 62, 33–46. pling in unsaturated porous media: Theory, petrophysics, and saturation
Huet, C.C., Rushing, J.A., Newsham, K.E. & Blasingame, T.A., 2005. Mod- front localization using an electroacoustic approach, Geophys. J. Int., 196,
ified Purcell/Burdine model for estimating absolute permeability from 867–884.
1032 A. Revil, N. Florsch and C. Camerlynck

Schmutz, M., Revil, A., Vaudelet, P., Batzle, M., Femenı́a Viñao, P. & APPENDIX A: THE COLE–COLE MODEL
Werkema, D.D., 2010. Influence of oil saturation upon spectral induced
polarization of oil bearing sands, Geophys. J. Int., 183, 211–224. The Cole–Cole model is usually used to fit complex conductiv-
Schwarz, G., 1962. A theory of the low-frequency dielectric dispersion of ity spectra. For the complex conductivity, the real (in phase) and
colloidal particles in electrolyte solution, J. Phys. Chem., 66, 2636–2642. imaginary (quadrature) components of the complex conductivity
Schwartz, N. & Furman, A., 2012. Spectral induced polarization signature are given by (Cole & Cole 1941),
of soil contaminated by organic pollutant: Experiment and modeling,  
J. geophys. Res., 117, B10203, doi:10.1029/2012JB009543.  1 sinh [c ln (ωτ0 )]
σ = σ∞ − Mn 1 −   , (A1)
Scott, J. & Barker, R., 2003. Determining pore-throat size in Permo-Triassic 2 cosh [c ln (ωτ0 )] + sin π2 (1 − c)
sandstones from low-frequency electrical spectroscopy, Geophys. Res.  
Lett., 30(9), 1450, doi:10.1029/2003GL016951. 1 Mn cos π2 (1 − c)
Swanson, B.F., 1981. A simple correlation between permeabilities and mer- σ  = − π . (A2)
2 cosh [c ln (ωτ0 )] + sin 2 (1 − c)
cury capillary pressures, J. Petrol. Tech., 33(12), 2498–2504.
Tarantola, A. & Valette, B., 1982. Generalized nonlinear inverse problems where Mn = σ∞ − σ0 ≥ 0. Including the high frequency effect
solved using the least square criterion, Rev. Geophys. Space Phys., 20(2), in the complex apparent conductivity (associated with the high-

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

219–232. frequency displacement current) yields (Revil 2013a),
Tarasov, A. & Titov, K., 2007. Relaxation time distribution from time do-
∗  
main induced polarization measurements, Geophys. J. Int., 170, 31–43. σeff = σeff − iσeff = σ ∗ − i ωε∞ . (A3)
Tarasov, A. & Titov, K., 2013. On the use of the Cole–Cole equations in
spectral induced polarization, Geophys. J. Int., 195, 352–356. The components of the apparent complex conductivity are,
Tikhonov, A.N., 1977. Solutions of Ill Posed Problems, Scripta Series in  
 1 sinh [c ln (ωτ0 )]
Mathematics, V.H. Winston. σeff = σ∞ − Mn 1 − π  ,
Timur, A., 1969. Pulsed Nuclear Magnetic Resonance studies of porosity, 2 cosh [c ln (ωτ0 )] + sin 2 (1 − c)
movable fluid, and permeability of sandstones, J. Petrol. Tech., 21, 775–
(A4)
786.
Tong, M., Li, L., Wang, W. & Jiang, Y., 2006a. Determining capillary-  π  
pressure curve, pore size distribution and permeability from induced po-  1 Mn cos (1 − c)
σeff = 2
π  + ωε∞ . (A5)
larization of shaley sand, Geophysics, 71, N33–N40. 2 cosh [c ln (ωτ0 )] + sin 2 (1 − c)
Tong, M., Li, L., Wang, W. & Jiang, Y., 2006b. A time-domain induced-
polarization method for estimating permeability in a shaly sand reservoir,
Geophys. Prospect., 54, 623–631.
Tong, M., Wang, W., Li, L., Jiang, Y. & Shi, D., 2004. Estimation of perme- APPENDIX B: CONNECTING THE
ability of shaly sand reservoir from induced polarization relaxation time D E B Y E A N D WA R B U R G
spectra, J. Petrol. Sci. Eng., 45, 31–40. DECOMPOSITIONS
Vanhala, H., 1997. Mapping oil-contaminated sand and till with the spec-
tral induced polarization (SIP) method, Geophys. Prospect., 45, 303– The complex conductivity is written as:
326.  ∞
h(τ )
Vaudelet, P., Revil, A., Schmutz, M., Franceschi, M. & Bégassat, P., 2011a. σ ∗ (ω) = b + a dτ, (B1)
Induced polarization signature of the presence of copper in saturated 0 1 + (iωτ )c
sands, Water Resour. Res., 47, W02526, doi:10.1029/2010WR009310. where h(τ ) is the function to be retrieved and c = 0.5 for a
Vaudelet, P., Revil, A., Schmutz, M., Franceschi, M. & Bégassat, P., 2011b.
Warburg decomposition of the spectra. Making the classical change
Changes in induced polarization associated with the sorption of sodium,
lead, and zinc on silica sands, J. Coll. Interface Sci., 360, 739–752.
of variables, we obtain,
Vinegar, H.J. & Waxman, M.H., 1984. Induced polarization of shaly sands, Hs (s) = τ h(τ ), (B2)
Geophysics, 49, 1267–1287.
Wang, M. & Revil, A., 2010. Electrochemical charge of silica surface at high
ionic strength in narrow channels, J. Coll. Interface Sci., 343, 381–386. z = − ln(ωτ0 ) ⇔ ωτ0 = e−z , (B3)
Warburg, E., 1899. Über das Verhalten sogenannter unpolarisiebarer Elek-  
troden gegen Wechelstrom (About the behaviour of so-called “impolariz- τ
s = ln ⇔ τ = τ0 es (B4)
able electrodes” in the present of alternating current), Ann. Phys. Chem., τ0
67, 493–499.
Weller, A., Nordsiek, S. & Debschütz, W., 2010. Estimating permeability of In eq. (B1), the function h(τ ) is normalized. However, when per-
sandstone samples by nuclear magnetic resonance and spectral-induced forming the inversion, it is easier to retrieve a function encapsulating
polarization, Geophysics, 75(6), E215–E226. the constant a and to normalize at the end. In the following, we can
Weller, A., Slater, L. & Nordsiek, S., 2013. On the relationship between drop the constant a. A normalization of the retrieved h(τ ) is re-
induced polarization and surface conductivity: Implications for petro- quired as soon as we want to retrieve this constant and a normalized
physical interpretation of electrical measurements, Geophysics, 78(5), distribution. In addition, we set, without loss of generality, τ0 = 1,
D315–D325. and we get in the general case:
Wong, J., 1979. An electrochemical model of the induced-polarization phe-  ∞
nomenon in disseminated sulfide ores, Geophysics, 44(7), 1245–1265.
Zimmermann, E., Kemna, A., Berwix, J., Glaas, W., Munch, H.M &
σz (z) = b + Hs (s)(z − s)ds, (B5)
−∞
Huisman, J.A., 2008a. A high-accuracy impedance spectrometer for mea-
suring sediments with low polarizability, Measur. Sci. Technol., 19(1–9), with,
105603. 1
Zimmermann, E., Kemna, A., Berwix, J., Glaas, W. & Vereeken, H., 2008b. (z − s) ≡ . (B6)
EIT measurement system with high phase accuracy for the imaging of
1 + i c e−c(z−s)
spectral induced polarization properties of soils and sediments, Measur. The convolution integral can be easily obtained through Fourier
Sci. Technol., 19(1–9), 094010. transform. Coming back to eq. (B1), which is for interest (here in
 Influence of pore size on induced polarization 1033
 
the case c = 0.5), we get ωc = e−c z and τ c = ec s , in such a way we σ̃η (η)˜ ∗η,c (η)
−1 −1
can re-write: HWiener
s (s) = FT [H̃η (η)] = −FT , (B12)
 ∞ |
˜ η,c (η)|2 + λ2
Hs (s)
σz (z) = b + ds = σz (z) + iσz (z). (B7) where λ is a damping factor, and (∗) denotes for the complex con-
−∞ 1 + i e
c −c(z−s)
jugate of the function. The Warburg case corresponds to c = 0.5.
Following here Florsch et al. (2012), we only consider the imag- Because the complex conductivity spectrum can be written as a con-
inary part in this Fourier approach. Calculating the imaginary part volution product, the Fourier transform appears to be an efficient
of σz (z) is facilitated when using the notation i c = exp(iπ c/2). We tool to determine the kernel of the convolution integral. This also
get: permits to swap between Warburg and Debye decompositions. We
 ∞  

sin π2 c e−c(z−s) come back on eq. (B11), just adding the index c to H to specify
σz (z) = − Hs (s)   ds. (B8) explicitly that we try to recover H for a chosen c. Then, the general
−∞ 1 + 2 cos π2 c e−c(z−s) + e−2c(z−s)
equation is
It follows, by using the Fourier-convolution theorem and defining
σ̃η (η) = −H̃η,c (η) · 
˜ η,c (η). (B13)
in this step z (z):

Downloaded from https://academic.oup.com/gji/article-abstract/198/2/1016/2874197 by guest on 24 March 2020

    The Debye and Warburg cases are given by:
   sin π2 c e−cz
FT σz = −FT [Hs ] · FT  
1 + 2 cos π2 c e−cz + e−2cz σ̃η (η) = −H̃η,1 (η) · 
˜ η,1 (η), (B14)

= −FT[Hs ] · FT[z,c ]. (B9) σ̃η (η) = −H̃η,1 (η) · 

˜ η, 1 (η), (B15)
2 2
The Fourier transform in eq. (B8) can be determined numerically. respectively. Therefore, the Debye and Warburg cases are related to
We write eq. (B9) using the ‘∼’ notation to denote the corresponding each other by:
functions in the Fourier domain, η is the variable in the Fourier dual
space. We obtain σ̃η (η) = −H̃η (η) · ˜ η,c (η), and therefore: H̃η,1 (η) · 
˜ η,1 (η) = H̃η, 1 (η) · 
˜ η, 1 (η). (B16)
2 2
 
−1 −1 σ̃η (η) Then if we have determined the RTD using a Debye decom-
Hs (s) = FT [H̃η (η)] = −FT . (B10) position, we can derive the corresponding RTD using a Warburg
˜ η,c (η)
decomposition by computing:
As explained in Florsch et al. (2012), this form is not expected  
to work well, because  ˜ η,c (η) can be really close to zero and then H̃η,1 (η) ·  ˜ η,1 (η)
Hz,1 (z) = FT−1 . (B17)
having this function in the dominator will cause some instabilities 2 ˜ η,1 (η)
in getting Hs (s). Actually, one could assume that where 
2
˜ η,c (η)
is small, σ̃η (η) should be also small, making the ratio in eq. (B9) Conversely, from a Warburg to a Debye decomposition, we have:
limited. It is not the case in practice, because the data are always  
H̃η,1 (η) ·  ˜ η,1 (η)
noisy, and then σ̃η (η) = −H̃η (η) · ˜ η,c (η) must be written as: −1
Hz,1 (z) = FT 2 2
. (B18)

˜ η,1 (η)
σ̃η (η) = −H̃η (η) · 
˜ η,c (η) + N (η), (B11)
It is natural that these two transforms do not involve the data set,
where N (η) denotes the non-modelled noise in the dual domain. since it is just a transform between relaxation time distributions. Of
Following Florsch et al. (2012), the Fourier inversion scheme can course, this relationship can be generalized easily to the transform
be stabilized by involving the Wiener regularization and a L-curve from any Cole–Cole RTD to another one if they only differ by their
procedure to obtain the optimum damping. We write: constant c.