Abstract
Spectral induced polarization (SIP) data are commonly analysed using phenomenological models. Among these models the Cole–Cole (CC) model is the most popular choice to describe the strength and frequency dependence of distinct polarization peaks in the data. More flexibility regarding the shape of the spectrum is provided by decomposition schemes. Here the spectral response is decomposed into individual responses of a chosen elementary relaxation model, mathematically acting as kernel in the involved integral, based on a broad range of relaxation times. A frequently used kernel function is the Debye model, but also the CC model with some other a priorly specified frequency dispersion (e.g. Warburg model) has been proposed as kernel in the decomposition. The different decomposition approaches in use, also including conductivity and resistivity formulations, pose the question to which degree the integral spectral parameters typically derived from the obtained relaxation time distribution are biased by the approach itself. Based on synthetic SIP data sampled from an ideal CC response, we here investigate how the two most important integral output parameters deviate from the corresponding CC input parameters. We find that the total chargeability may be underestimated by up to 80percent and the mean relaxation time may be off by up to three orders of magnitude relative to the original values, depending on the frequency dispersion of the analysed spectrum and the proximity of its peak to the frequency range limits considered in the decomposition. We conclude that a quantitative comparison of SIP parameters across different studies, or the adoption of parameter relationships from other studies, for example when transferring laboratory results to the field, is only possible on the basis of a consistent spectral analysis procedure. This is particularly important when comparing effective CC parameters with spectral parameters derived from decomposition results.
Electrical properties, Electromagnetic theory, Hydrogeophysics
1 INTRODUCTION
Over the last decade the spectral induced polarization (SIP) method has seen a rapid increase of its use in hydrogeological and environmental studies (e.g. Kemna etal.2012). SIP data are given as frequency dependent electrical impedance or admittance measurements that can be converted to resistivities or conductivities by accounting for the measurement geometry.
Two basic types of phenomenological models are in use for the quantitative description of SIP data. The Cole–Cole (CC) model and its variants (e.g. Cole & Cole 1941; Pelton etal.1978; Dias 2000) are typically used if the spectrum exhibits a distinct peak. Here three parameters, the CC relaxation time, the CC chargeability and the CC frequency exponent, account for the position of the peak along the frequency axis, the magnitude and the frequency dispersion of the response, respectively.
Alternatively, SIP data can be decomposed into individual responses of a chosen elementary relaxation model, mathematically acting as kernel in the involved integral, based on a broad range of relaxation times. A frequently used kernel function is the Debye model (e.g. Uhlmann & Hakim 1971; Lesmes & Morgan 2001; Tarasov etal.2003; Nordsiek & Weller 2008), the special case of the CC model with strongest possible frequency dispersion. However, also the CC model with some other a priorly specified frequency dispersion (e.g. the Warburg model with intermediate frequency dispersion) has been proposed as kernel in the decomposition (Tarasov etal.2003; Florsch etal.2014; Revil etal.2014). The choice of the latter follows pore-scale modelling results, which indicate that the elementary polarization response of a single pore or grain exhibits a broader frequency dispersion than the Debye model response (e.g. Wong 1979; Titov etal.2002; Bücker & Hördt 2013). Therefore a CC decomposition (CCD) (with fixed frequency dispersion) seems to be more adequate from a petrophysical point of view. The decomposition approach yields a relaxation time distribution (RTD) from which integral spectral parameters similar to those of the CC model can be computed, especially a mean relaxation time and a total chargeability.
It is important to note that the decomposition of SIP spectra based on CC-type kernel functions is highly ambiguous and the resultant RTD depends on the frequency dispersion of the kernel. For instance the equivalent representation of a CC model response as a superposition of Debye model responses based on an appropriate RTD was already recognized by Cole & Cole (1941). However, even if the frequency dispersion of the kernel is fixed in the decomposition, which is commonly done due to this inherent equivalence, the resulting inverse problem typically still is ill-posed, requiring regularization (Florsch etal.2012, 2014; Weigand & Kemna 2016). This leads to considerable uncertainties in the decomposition results on top of the contribution on the account of data errors. Several studies have assessed uncertainties in the estimation of spectral parameters for the CC model (Ghorbani etal.2007; Chen etal.2008) and the Debye decomposition (DD; Florsch etal.2012; Keery etal.2012) in detail.
With the growing number and sophistication of SIP studies relating phenomenological model parameters to petrophysical properties, for example to permeability (e.g. Binley etal.2005; Revil & Florsch 2010; Zisser etal.2010; Weller etal.2015), consistency of data analysis procedures becomes important to ensure comparability of results. However, the variety of different models in use today is quite large, and formulations based on either resistivity or conductivity further complicate the interpretation of results across studies. For example, Revil etal. (2015) used four different approaches to determine characteristic relaxation times from SIP data. As pointed out by Weller etal. (2015), the total chargeability derived from the DD is biased by the shape of the spectrum and the analysed frequency range, while the CC chargeability accounts for all frequencies. Gurin etal. (2015) also mention that the total chargeability from the DD is not comparable to the CC chargeability, and is dependent on the position of the spectral peak and the frequency range covered by the data.
Despite these insights, to our knowledge it has not yet been systematically studied how the total chargeability and the mean relaxation time derived from the CC decomposition (including the Debye and Warburg cases) of CC-type data deviate from the original CC chargeability and CC relaxation time. We here investigate these relationships based on synthetic SIP data sampled from an ideal CC response for a broad range of CC parameter values. The results of our study yield quantitative information on the possible over- or underestimation of chargeability and relaxation time when different models are used to analyse SIP data.
2 METHODS
2.1 The Cole–Cole model
Based on the original formulation of Cole & Cole (1941), the complex conductivity (|$\hat{\sigma }$|) formulation of the CC model is given as (e.g. Tarasov & Titov 2013)
\begin{equation}\hat{\sigma }(\omega ) = \sigma _\infty \left(1 - \frac{m}{1 + (j \omega \tau _{(\sigma )})^{c}}\right),\end{equation}
(1)
with ω denoting the angular frequency, σ0 and σ∞ the conductivity in the low- and high-frequency limit, respectively, m = (σ∞ − σ0)/σ∞ the chargeability, j the imaginary unit, τ(σ) the CC relaxation time, and c the CC frequency exponent.
An alternative formulation in terms of complex resistivity (|$\hat{\rho }$|) was proposed by Pelton etal. (1978) as
\begin{equation}\hat{\rho }(\omega ) = \rho _0 \left(1 - m \left[1 - \frac{1}{1 + (j \omega \tau _{(\rho )})^c}\right] \right),\end{equation}
(2)
\begin{equation}m = \frac{\sigma _\infty - \sigma _0}{\sigma _\infty } = \frac{\rho _0 - \rho _\infty }{\rho _0},\end{equation}
(3)
with ρ0 = 1/σ0 being the DC resistivity, ρ∞ = 1/σ∞, and τ(ρ) the CC relaxation time in the resistivity formulation. Importantly, from |$\hat{\rho }(\omega ) = 1 / \hat{\sigma }(\omega )$| it follows that τ(σ) and τ(ρ) are not identical but related by (Florsch etal.2012)
\begin{equation}\tau _{(\sigma )} = \tau _{(\rho )} (1 - m)^{1/c}.\end{equation}
(4)
It is easy to show that the peak frequencies of the imaginary components of |$\hat{\sigma }(\omega )$| and |$-\hat{\rho }(\omega )$| are related to the relaxation times τ(σ) and τ(ρ) according to
\begin{equation}f_{\text{peak}}^{(\sigma ^{\prime \prime })} = \frac{1}{2 \pi \tau _{(\sigma )}} \quad \text{and} \quad f_{\text{peak}}^{(-\rho ^{\prime \prime })} = \frac{1}{2 \pi \tau _{(\rho )}}.\end{equation}
(5)
In combination with eq. (4) this implies that σ″(ω) and −ρ″(ω) peak at different frequencies (with ″ denoting the imaginary component).
2.2 The Cole–Cole decomposition
Based on a broad range of relaxation times spanned by N discretized values τk, the discrete form of the CCD in conductivity and resistivity formulation, respectively, can be written as
\begin{equation}\hat{\sigma }(\omega ) = \sigma _\infty \left(1 - \sum _{k=1}^{N} \frac{m_{k}^{(\sigma )}}{1 + (j \omega \tau _k)^{\tilde{c}}}\right),\end{equation}
(6)
\begin{equation}\hat{\rho }(\omega ) = \rho _0 \left(1 - \sum _{k=1}^{N} m_{k}^{(\rho )} \left[1 - \frac{1}{1 + (j \omega \tau _k)^{\tilde{c}}}\right] \right),\end{equation}
(7)
where the kernel (assumed elementary relaxation model) is adopted from the CC model in eqs (1) and (2), respectively. In eqs (6) and (7), |$m_{k}^{(\sigma )}$| and |$m_{k}^{(\rho )}$| denote the chargeability weights at the sampled relaxation times τk in conductivity and resistivity formulation, respectively. Usually the relaxation time range is chosen to cover at least the measurement frequency (f) range of the data according to the inverse relationship τ = 1/(2πf) (e.g. Weigand & Kemna 2016). The frequency dispersion of the kernel functions in the decompositions is controlled by the chosen fixed value for |$\tilde{c}$|, with the DD resulting for |$\tilde{c}=1$| and a Warburg decomposition (Revil etal.2014) resulting for |$\tilde{c} = 0.5$|.
The RTD is given by the distribution |$m_k^{(\sigma )}(\tau _{k})$| or |$m_k^{(\rho )}(\tau _k)$|, respectively, from which the following integral parameters can be derived:
The total chargeability|$m_{\text{tot}} = \sum _{k=1}^{N} m_k$| (e.g. Nordsiek & Weller 2008) is the analogue to the chargeability m in eqs (1) and (2) when a single CC model is used to describe SIP data. However, in the typical case that the fixed value of |$\tilde{c}$| in the decomposition is considerably larger than the c value of a single CC model describing the same SIP data, that is, if the elementary relaxation model has a much stronger frequency dispersion than the given data, mtot primarily accounts for polarization within the considered frequency range of the decomposition (spanned by the chosen τk values), whereas m also contains significant contributions from outside this range.
The median relaxation time τ50 is the relaxation time at which 50percent of the total chargeability is reached (Nordsiek & Weller 2008; Zisser etal.2010).
The mean logarithmic relaxation time τmean is the chargeability-weighted logarithmic mean value of the RTD (Nordsiek & Weller 2008): |$\tau _{\text{mean}} = \exp (\frac{\sum _{k=1}^{N} m_k \log (\tau _k)}{\sum _{k=1}^{N} m_k})$|.
The arithmetic mean of the RTD is given by |$\tau _a = \frac{\sum _{k=1}^{N} m_k \tau _k}{\sum _{k=1}^{N} m_k}$| (Tong etal.2004).
Although originally proposed for the DD, the above parameters can be computed analogously from the RTD of the CCD.
3 NUMERICAL EXPERIMENTS
Synthetic SIP data were generated assuming a CC model response in resistivity formulation (eq. 2) by systematically varying τ(ρ) in the range from 1.59 × 10−5 s to 159.15s and c from 0.05 to 1. Assumed measurement frequencies were selected using eq. (5) such that the polarization peaks lie within the frequency range covered by the data, with 30 frequencies in total. CCDs of the resulting spectra were performed for the |$\tilde{c}$| values 0.3, 0.5 and 1.0, according to eq. (7) (resistivity formulation), using the open-source code from Weigand & Kemna (2016), with 20 relaxation times per decade, and two decades of relaxation times outside the low- and high-frequency limits of the data (222 relaxation times in total, 140 representing the data frequency range).
The input CC chargeability m was compared to the mtot parameters computed from the CCD results. For the relaxation times, the three parameters τmean, τ50, and τa based on the decomposition with |$\tilde{c} = 1$| (DD) were related to the input CC parameter τ(ρ). In addition, τmean resulting from the CCD for the |$\tilde{c}$| values 0.3 and 0.5, respectively, was compared to τ(ρ). Finally, the analogue of eq. (4) for the mean relaxation times |$\tau _{\text{mean}}^{(\sigma )}$| and |$\tau _{\text{mean}}^{(\rho )}$| obtained from the DD (|$\tilde{c} = 1$|) in conductivity (eq. 6) and resistivity (eq. 7) formulation, respectively, was evaluated for different input CC chargeability values in terms of |$\tau _{\text{mean}}^{(\sigma )} / \tau _{\text{mean}}^{(\rho )}$|.
4 RESULTS
4.1 Chargeability recovery
Original CC chargeabilities m are increasingly underestimated by |$m_{\text{tot}}^{(\rho )}$| by up to 80percent for decreasing c values, that is, decreasing frequency dispersion of the data, and for τ(ρ) approaching either the low or high relaxation time limit in the decomposition (Fig.1). This behaviour is found for different |$\tilde{c}$| values in the range |$c < \tilde{c}$|.
Figure 1.
Open in new tabDownload slide
Relationship between original CC chargeability m and CCD-derived total chargeability |$m_{\text{tot}}^{(\rho )}$| using the indicated values of |$\tilde{c}$| for varying values of input CC time constant τ(ρ) and CC frequency exponent c (|$< \tilde{c})$|. A fixed value m = 0.5 was used in the CC model to generate the SIP data. The resistivity formulations of both the CC model and the DD scheme were used.
4.2 Relaxation time recovery
The relaxation times |$\tau _{\text{mean}}^{(\rho )}$| and |$\tau _{50}^{(\rho )}$| recovered from the CCD (eq. 7) over- or underestimate the original CC relaxation time τ(ρ) by up to three orders of magnitude for τ(ρ) approaching the high or low relaxation time limit, respectively, and for decreasing c values (Fig.2). The absolute deviation patterns are similar to those obtained for |$m_{\text{tot}}^{(\rho )}$| (Fig.1). The arithmetic mean relaxation time (|$\tau _{a}^{(\rho )}$|) shows only a narrow band of CC parameters where the agreement with the original τ(ρ) value is good; outside this range, deviations up to three orders of magnitude exist.
Figure 2.
Open in new tabDownload slide
Relationship between original CC relaxation time τ(ρ) and the indicated DD-derived relaxation times for varying values of input CC parameters (τ(ρ), c). A fixed value m = 0.5 was used for the input CC chargeability. Displayed is the log10-difference between τ(ρ) and the DD-derived relaxation times. The resistivity formulations of both the CC model and the DD scheme were used.
The deviation of the CCD-derived (eq. 7) relaxation time |$\tau _{\text{mean}}^{(\rho )}$| for different values of |$\tilde{c}$| exhibits a similar pattern as the corresponding |$m_{\text{tot}}^{(\rho )}$| results (Fig.1). The absolute deviation from the original τ(ρ) value increases with decreasing c (|$< \tilde{c}$|) and with τ(ρ) approaching one of the relaxation time limits (Fig.3).
Figure 3.
Open in new tabDownload slide
Relationship between original CC relaxation time τ(ρ) and CCD-derived relaxation time |$\tau _{\text{mean}}^{(\rho )}$| for different values of |$\tilde{c}$|. The resistivity formulations of both the CC model and the DD scheme were used.
The ratio between the DD-derived relaxation times |$\tau _{\text{mean}}^{(\sigma )}$| and |$\tau _{\text{mean}}^{(\rho )}$| (using eqs 6 and 7, respectively) diverts from the behaviour of the CC counterparts τ(σ) and τ(ρ) (described by eq. 4) and shows much less dependence on m (Fig.4). However, the ratio deviates from 1 (the value which indicates perfect agreement between τ(σ) and τ(ρ)) for larger m values and also depends on τ(ρ), that is, the position of the peak in the spectrum relative to the analysed frequency range.
Figure 4.
Open in new tabDownload slide
Relationship between |$\tau _{\text{mean}}^{(\sigma )}$| and |$\tau _{\text{mean}}^{(\rho )}$| obtained from DD results in conductivity and resistivity formulation, respectively, for different values of input CC frequency exponent c (0.05, 0.31, 1) and CC relaxation time τ(ρ) (0.038s (a), 1.59 × 10−5 s (b)) plotted as ratio |$\tau _{\text{mean}}^{(\sigma )} / \tau _{\text{mean}}^{(\rho )}$| (solid curves). For comparison the relationship between the corresponding CC relaxation times τ(σ) and τ(ρ) according to eq. (4) is shown (dashed curves).
5 DISCUSSION
The CCD-derived parameters mtot, τmean and τ50 increasingly diverge from the original CC model parameters with decreasing values of c and for τ(ρ) approaching the relaxation time limits in the decomposition. This result can be mostly attributed to the limited frequency range covered by the data, and thus the limited relaxation time range considered in the CCD. Outside the data frequency range the CCD response decreases faster than the original CC response (Figs 5a and c) (if the used kernel has a stronger frequency dispersion than the data), which causes a bias in the inferred mean and median relaxation times towards the centre of the RTD. However, a non-zero response outside the data range always remains due to the spectral width of the response of even a single kernel term, which can extend over more than one decade beyond the data frequency limits (increasing with decreasing |$\tilde{c}$| value, cf. Figs 5a and c). One decade can also be considered as the approximate theoretical resolution limit below which two peaks in a spectrum can no longer be distinguished (Florsch etal.2014). Since the deviations of the CCD-derived relaxation times can easily exceed one decade, they must be considered significant and therefore should to be taken into account when analysing SIP data.
Figure 5.
Open in new tabDownload slide
Debye versus Warburg decomposition of two CC model responses with different frequency dispersion. Decompositions are based on the shaded frequency range only (a,c). Original CC chargeability is m = 0.5 and CC relaxation time τ(ρ) = 0.159s. (a) Responses of input CC model for c = 0.3 and CCD results for |$\tilde{c} = 1$| (|$m_{\text{tot}}^{(\rho )} = 0.30$|) and |$\tilde{c} = 0.5$| (|$m_{\text{tot}}^{(\rho )} = 0.35$|). (b) RTDs of the CCDs from panel (a). (c) Responses of input CC model for c = 0.5 and CCD results for |$\tilde{c} = 1$| (|$m_{\text{tot}}^{(\rho )} = 0.41$|) and |$\tilde{c} = 0.5$| (|$m_{\text{tot}}^{(\rho )} = 0.50$|). (d) RTDs of the CCDs from panel (c). The resistivity formulations of both the CC model and the DD scheme were used.
As notable in the RTD results in Fig.5(b), in this example the CCDs for |$\tilde{c} = 1$| (DD) and |$\tilde{c} = 0.5$| (Warburg decomposition) yield similar chargeability weights at the data frequency edges, which explains the larger spread of the response outside the data frequency range for the |$\tilde{c} = 0.5$| case (Fig.5a). However, with increasing original c value the contribution outside the data frequency range decreases, and thus the underestimation of m is also reduced (Figs 5c and d). In fact for the trivial, but practically irrelevant, case of a decomposition kernel with identical frequency dispersion like the data (i.e. |$\tilde{c} = c$|), the derived parameters mtot, τmean and τ50 perfectly resemble the original CC parameters if the spectral peak lies well within the analysed data frequency range (cf. Figs 1–3).
The observed variation in the reconstruction quality of chargeabilities and relaxation times also influences the relationship between the mean relaxation times derived from the CCD in conductivity and resistivity formulations (Fig.4). We found that |$\tau _{\text{mean}}^{(\sigma )}$| and |$\tau _{\text{mean}}^{(\rho )}$| agree only within one order of magnitude over the whole range of c values. Therefore, should |$\tau _{\text{mean}}^{(\sigma )}$| and |$\tau _{\text{mean}}^{(\rho )}$| be compared or jointly used in a study, the discrepancy between both quantities should be taken into account. However, we again stress that errors can be much larger when CC and CCD-derived parameters are jointly interpreted and their systematic deviations ignored—no matter if conductivity or resistivity formulations are used.
We finally note that the CCD scheme depends on various settings which can have significant influence on the results, with the selected range of relaxation times having the strongest effect. As shown in previous work (Weigand & Kemna 2016), the relaxation time range should extend by at least one order of magnitude (better two) to both sides of the data frequency limits. The amount of regularization, here a smoothness constraint imposed on the RTD, has from our experience only marginal influence on the derived integral parameters mtot and τmean. However, regularization at all is crucial to overcome the inherent ill-posedness of the problem (Florsch etal.2014).
6 CONCLUSIONS
In this study we investigated the relationships of chargeability and relaxation time parameters recovered from a spectral decomposition of synthetic CC-type SIP data with the corresponding CC input parameters. As kernel function in the decomposition we considered CC models with different frequency dispersion, including the Debye model and the Warburg model. We found that total chargeability and mean relaxation time derived from the decomposition results can deviate significantly from their CC model counterparts, as well as among each other for different decomposition approaches. Total chargeability was underestimated by up to 80percent and mean relaxation time varied by up to three orders of magnitude compared to the CC input parameters. The deviations depend on the frequency dispersion of the analysed spectrum and the proximity of its peak to the frequency range limits considered in the decomposition. This behaviour is found for all considered kernels (Debye, Warburg, CC), except for the trivial, and practically irrelevant, case of a CC kernel with a frequency dispersion identical to the one of the data. Importantly, our results do not promote the use of a particular kernel function, which should be chosen based on the physics of the underlying polarization mechanism in the given application.
We conclude that a quantitative comparison of SIP parameters across different studies, or the adoption of parameter relationships from other studies, is only possible on the basis of a consistent, well documented spectral analysis procedure. This is especially important when comparing effective CC parameters with spectral parameters derived from decomposition results. Non-adherence to a consistent analysis procedure can result in severe misinterpretations in the growing field of applications, for example when linking SIP parameters to textural, hydraulic, or (bio)geochemical properties, or when transferring laboratory results to the field.
Part of this work was conducted in the framework of the SFB TR32 ‘Patterns in soil-vegetation-atmosphere systems: monitoring, modelling and data assimilation’ funded by the Deutsche Forschungsgemeinschaft (DFG). We thank the editor Jörg Renner and two anonymous referees for their very constructive reviews.
REFERENCES
Binley A. Slater L. f*ckes M. Cassiani G.
2005
Relationship between spectral induced polarization and hydraulic properties of saturated and unsaturated sandstone
Water Resour. Res.
41
12
1
13
Bücker M. Hördt A.
2013
Long and short narrow pore models for membrane polarization
Geophysics
78
6
E299
E314
Chen J. Kemna A. Hubbard S.
2008
A comparison between Gauss–Newton and Markov-chain Monte Carlo–based methods for inverting spectral induced-polarization data for Cole–Cole parameters
Geophysics
73
6
F247
F259
Cole K. Cole R.
1941
Dispersion and absorption in dielectrics I. Alternating current characteristics
J. Chem. Phys.
9
341
351
Dias C.
2000
Developments in a model to describe low-frequency electrical polarization of rocks
Geophysics
65
2
437
451
Florsch N. Camerlynck C. Revil A.
2012
Direct estimation of the distribution of relaxation times from induced-polarization spectra using a Fourier transform analysis
Near Surf. Geophys.
10
6
517
531
OpenURL Placeholder Text
Florsch N. Revil A. Camerlynck C.
2014
Inversion of generalized relaxation time distributions with optimized damping parameter
J. appl. Geophys.
109
119
132
Ghorbani A. Camerlynck C. Florsch N. Cosenza P. Revil A.
2007
Bayesian inference of the Cole–Cole parameters from time-and frequency-domain induced polarization
Geophys. Prospect.
55
4
589
605
Gurin G. Titov K. Ilyin Y. Tarasov A.
2015
Induced polarization of disseminated electronically conductive minerals: a semi-empirical model
Geophys. J. Int.
200
3
1555
1565
Keery J. Binley A. Elshenawy A. Clifford J.
2012
Markov-chain Monte Carlo estimation of distributed Debye relaxations in spectral induced polarization
Geophysics
77
2
E159
E170
Kemna A.
2012
An overview of the spectral induced polarization method for near-surface applications
Near Surf. Geophys.
10
6
453
468
OpenURL Placeholder Text
Lesmes D. Morgan F.
2001
Dielectric spectroscopy of sedimentary rocks
J. geophys. Res.
106
B7
13329
13346
Nordsiek S. Weller A.
2008
A new approach to fitting induced-polarization spectra
Geophysics
73
6
F235
F245
Pelton W. Ward S. Hallof P. Sill W. Nelson P.
1978
Mineral discrimination and removal of inductive coupling with multifrequency IP
Geophysics
43
3
588
609
Revil A. Florsch N.
2010
Determination of permeability from spectral induced polarization in granular media
Geophys. J. Int.
181
3
1480
1498
OpenURL Placeholder Text
Revil A. Florsch N. Camerlynck C.
2014
Spectral induced polarization porosimetry
Geophys. J. Int.
198
2
1016
1033
Revil A. Binley A. Mejus L. Kessouri P.
2015
Predicting permeability from the characteristic relaxation time and intrinsic formation factor of complex conductivity spectra
Water Resour. Res.
51
8
6672
6700
Tarasov A. Titov K.
2013
On the use of the Cole–Cole equations in spectral induced polarization
Geophys. J. Int.
195
1
352
356
Tarasov A. Titov K. Münch M. Kemna A.
2003
Induced polarization spectra of sands and clays measured in the time domain
Proc. Int. Conf. Geophysics of the 21st Century – a Leap into the Future
SEG/EAGE, September 1–4, 4 p
OpenURL Placeholder Text
Titov K. Komarov V. Tarasov V. Levitski A.
2002
Theoretical and experimental study of time domain-induced polarization in water-saturated sands
J. Appl. Geophys.
50
4
417
433
Tong M. Weinan W. Li L. Yizhong J. Deqin S.
2004
Estimation of permeability of shaly sand reservoir from induced polarization relaxation time spectra
J. Pet. Sci. Eng.
45
1
31
40
OpenURL Placeholder Text
Uhlmann D. Hakim R.
1971
Derivation of distribution functions from relaxation data
J. Phys. Chem. Solids
32
11
2652
2655
Weigand M. Kemna A.
2016
Debye decomposition of time-lapse spectral induced polarisation data
Comput. Geosci.
86
34
45
Weller A. Slater L. Binley A. Nordsiek S. Xu S.
2015
Permeability prediction based on induced polarization: insights from measurements on sandstone and unconsolidated samples spanning a wide permeability range
Geophysics
80
2
D161
D173
Wong J.
1979
An electrochemical model of the induced-polarization phenomenon in disseminated sulfide ores
Geophysics
44
7
1245
1265
Zisser N. Kemna A. Nover G.
2010
Relationship between low-frequency electrical properties and hydraulic permeability of low-permeability sandstones
Geophysics
75
3
E131
E141
© The Authors 2016. Published by Oxford University Press on behalf of The Royal Astronomical Society.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
