See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/258787075

Inductive Source Induced Polarization

Article  in  Geophysical Journal International · February 2013

DOI: 10.1093/gji/ggs040

CITATIONS READS

7 255

3 authors:

David Marchant Eldad Haber

University of British Columbia - Vancouver University of British Columbia - Vancouver
9 PUBLICATIONS   44 CITATIONS    143 PUBLICATIONS   3,840 CITATIONS   

SEE PROFILE SEE PROFILE

Douglas W. Oldenburg
University of British Columbia - Vancouver
302 PUBLICATIONS   8,450 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Multiscale Finite Volume Methods to forward model Maxwell's Equations View project

TKC project (UBC-GIF) View project

All content following this page was uploaded by Douglas W. Oldenburg on 05 January 2015.

The user has requested enhancement of the downloaded file.

Inductive Source Induced Polarization
David Marchant
UBC-Geophysical Inversion Facility
Dept Earth and Ocean Science,
6339 Stores Road,
Vancouver, V6T1Z4 dmarchant@eos.ubc.ca
Eldad Haber
UBC-Geophysical Inversion Facility
Dept Earth and Ocean Science,
6339 Stores Road,
Vancouver, V6T1Z4 haber@math.ubc.ca
Douglas W Oldenburg
UBC-Geophysical Inversion Facility
Dept Earth and Ocean Science,
6339 Stores Road,
Vancouver, V6T1Z4 doug@eos.ubc.ca

propose a new way to process the data that is based on the

character of the fields at low frequency accompanied with a
SUMMARY new inversion methodology for an effective IP parameter.
We present a new survey methodology to map the
METHOD
distribution of chargeable material in the subsurface
using inductive electromagnetic sources and observations
of the magnetic fields in the frequency domain. An Maxwell’s equations, in the frequency domain, assuming a
accompanying inversion algorithm is developed, and the time dependence of e − iωt , are
technique is tested on synthetic data.
∇ × E − iωµH = 0 (1a)
Key words: Induced polarization, Electromagnetics, ∇ × H − σE = s (1b)
Geophysical Inversion
Here, E and H are the electric and magnetic fields, σ is the
conductivity and µ is magnetic susceptibility. If we define H
INTRODUCTION such that H = H 0 + H S and ∇ × H 0 = s we can rewrite the
equations in terms of E and Hs as
The presence of chargeable material in the ground can often
provide an excellent proxy for the distribution of sulphides in ∇ × E − iωµH s = iωµH 0 (2a)
the subsurface. Traditionally, chargeability is mapped using
s
induced polarization (IP) surveys (Seigel, 1959) in which ∇ × H − σE = 0 (2b)
current is injected into the ground using two transmitter
electrodes and the potential difference is measured across Eliminating E from the system yields an equation for Hs
other pairs of electrodes placed away from the transmitter.
These data are now commonly inverted to recover 2D or 3D ∇ × ρ∇ × H s − iωµH s = iωµH 0 (3)
chargeability structures (e.g. Oldenburg and Li (1994), Li and
Oldenburg, (2000)). 1
where ρ is the resistivity, ρ = . We first assume a real,
σ
Although IP methods have had great success, surveying large non-dispersive (frequency independent) resistivity distribution
areas can be prohibitively time consuming and expensive. IP ρ . Differentiating Equation 3 twice with respect to ω gives
methods can also fail in some geological situations. If the
surface material is too resistive it can be impossible to push ∂H s ∂H s
enough current into the ground and to excite an observable ∇ × ρ∇ ×
∂ω
− iωµ
∂ω
= iµ H 0 + H s ( ) (4)
polarization effect. In areas with very resistive overburden,
the voltage and power limitations of the transmitter hardware ∂ 2H s ∂2H s ∂H s
∇ × ρ∇ × − i ωµ = 2i µ (5)
adversely affect the resulting signal. ∂ω 2 ∂ω 2 ∂ω

The magnetic induced polarization (MIP) method (Seigel, Expanding Hs about ω = 0 , and dropping the higher terms
1974) was developed in order to address these issues. In this leaves
method, the transmitter is still a grounded source but the
∂H s 1 ∂2H s
secondary magnetic field is measured. A 3D inversion H S (ω ) ≈ ω+ ω2 (6)
technique for MIP data was developed by Chen and ∂ω ω =0
2 ∂ω 2 ω =0
Oldenburg (2003). MIP eliminates the time consuming task
of placing receiver electrodes, but current still needs to be ∂H s
We can see from Equations 4 and 5 that is purely
injected into the ground. The use of an inductive source was ∂ω ω =0
examined by Hohmann et al. (1970) and the use of natural
sources was investigated by Gasperikova and Morrison (2001) ∂2H s
imaginary, and that is purely real, giving
but the method has not been developed for practical ∂ω 2 ω =0
application.
∂H s 1 ∂2H s
In this work we propose a new inductive source IP (ISIP)
( )
ℑ H s (ω ) ≈
∂ω
ω (
and ℜ H s (ω ) ≈ ) 2 ∂ω 2
ω2 .
ω =0 ω =0
technique. The technique is based on the ability to measure the
differences in the magnetic fields at two low frequencies. We

nd
22 International Geophysical Conference and Exhibition, 26-29 February 2012 - Brisbane, Australia 1
Inductive Source Induced Polarization Marchant, Haber and Oldenburg

Figure 2. Secondary magnetic fields generated by a 100m

Figure 1. Secondary magnetic fields generated by a 100m
loop centred at x=600m, y=600m above a conductive (non-
chargeable) block in a half space with ω1 = 1hz and
ω2 = 2hz . ISIP data (1e and 1f) show very little response
to this model.
We define the quantities d ℜ and d ℑ
ω 22
( )
d ℜ = ℜ H s (ω 2 ) −
ω12
(
ℜ H s (ω1 ) )
( )ω
(
d ℑ = ℑ H s (ω 2 ) − 2 ℑ H s (ω1 ) ) (7b)
ω1
where ω1 and ω 2 are two frequencies which are sufficiently low
so that their inductive responses to a real conductivity earth
are only minimally different, that is, we are working in
resistive limit regime. For any real, non-dispersive resistivity
distribution, we see from Equation 6 that both d ℜ and d ℑ
should approximately equal zero. Signal in either of these
quantities indicates the presence of chargeable material and
we will use these data in our inversion.
Discretization
For numerical evaluation, we discretize Maxwell’s equations
on a orthogonal, staggered grid and use a finite volume
approach (Haber and Ascher, 2001). If we assume that
µ = µ 0 , then ∇ ⋅ µH s = 0 . This is added as stabilization term
to Equation 3 to yield
s s s
∇ × ρ∇ × H − ∇ρ∇ ⋅ µH − iωµH = iωµH 0
We choose to put Hs on cell edges, and ρ at cell centres. Our
matrix equations are
loop centred at x=600m, y=600m above a conductive block
with a chargeability of 0.1.
where curl and grad are discrete forms of the curl and
graident operators, and Av is an averaging matrix.
To show the merits of our choice of data we perform a forward
modelling. We use the Cole-Cole model (Pelton et al., 1978)
to describe the complex resistivity. In the frequency domain
(7a)
  1 
ρ (ω ) = ρ 0 1 − η 1 − 
 (10)
 1 + (iωτ )
c
  
where ρ 0 is the resistivity at zero frequency, η is the
chargeability, and τ is a time constant. The parameter c is a
constant which controls the frequency dependence of the
material.
Figure 1 shows the vertical components of the secondary
magnetic field computed using Equation 8. A square loop
transmitter, 100m on a side, is centered at x=600m, y=600m,
and is placed above a conductive (2 Ωm ), non-chageable
block in a resistive halfspace (100 Ohm-m). Both the real and
the imaginary components of the response are shown at 1 hz
( ω1 ) and 2 hz ( ω 2 ). Figures 1e and 1f show the computed
ISIP data. For this non-dispersive resistivity distribution, the
values are approximatly equal to zero.
Figure 2 shows results of an identical survey, but this time
carried out above a chargeable block, with resistivities
(8) following a Cole-Cole model (Equation 10, ρ 0 =2 Ωm ,
η =0.1, τ =0.1, c=1) In this case, there is strong dipolar signal
observed in both the real and imaginary data as in Figures 2e
and 2f. These are the data we will invert.

(curl T
)
diag (A v ρ )curl + grad diag (A v ρ )gradT µ − iωµI H s
(9)
= iωµH 0

nd
22 International Geophysical Conference and Exhibition, 26-29 February 2012 - Brisbane, Australia 2
Inductive Source Induced Polarization Marchant, Haber and Oldenburg

Formulating the ISIP data equations s

∂H sI
(TI + ωµI ) ∂H R + TR ( )
= curl T diag curl H sR A v (16b)
∂ρ I ∂ρ I
The effect of chargeability is to cause a pertubation in the
resistivity with frequency. Let ρ1 equal the real, frequency
If we consider the sensitivity to changes in the imaginary part
independent, back ground conductivity. Let ρ 2 be the of the conductivity about a purely real conductivity, and we
resistivity that would be observed at ω2 . The frequencies consider only low frequencies such that ωµ ≈ 0 then this
ω1 and ω 2 are closely spaced, so the reisistivity that would be simplifies to
observed at ω 2 is equal to the resistivity at ω1 , plus a small ∂H sR
TR
∂ρ I
(
= curlT diag curl H sI A v ) (17a)
pertubation, or ρ 2 = ρ1 + ∆ρ . We can write
∂H sI
∂H s
TR
∂ρ I
(
= curlT diag curl H sR A v ) (17b)
H s (ω 2 , ρ 2 ) ≈ H s (ω 2 , ρ1 + ∆ρ ) ≈ H s (ω 2 , ρ1 ) + ∆ρ (11)
∂ρ ω2 , ρ1
We can now define the quantities GI, GR and m to be
Taking the Cole-Cole model, setting c = 1 , and expanding
about ω = 0 gives
( )
G I = −ω2TR−1curl T diag curl H SI A v diag (ρ1 ) (18a)
GR = −ω2TR−1curlT diag (curl H )A diag (ρ )
S
R v 1 (18b)
ρ (ω ) ≈ ρ 0 − iρ 0ητω (12) m = ητ (19)
this gives
The first order perturbation is purely imaginary, so at low dℜ = GIm (20a)
frequency ∆ρ is purely imaginary, which we can denote ℑ
d = G Rm (20b)
as ∆ρ I . Substituting this into Equation 11 and separating the
real and the imaginary parts yields
The recovery of the “chargeability,” m, from the ISIP data
∂H sR
( s
) (
ℜ H (ω2 , ρ 2 ) ≈ ℜ H (ω2 , ρ1 ) s
) +
∂ρ I
∆ρ I therefore reduces to solving a standard linear inverse
problem.
ω 2 , ρ1
(13a)
ω2 ∂H sR
ω1
(
≈ 22 ℜ H s (ω1 , ρ1 ) +
∂ρ I
) ∆ρ I
ω 2 , ρ1

∂H sI
( ) (
ℑ H s (ω2 , ρ 2 ) ≈ ℑ H s (ω2 , ρ1 ) + ) ∂ρ I
∆ρ I
ω 2 , ρ1
(13b)
ω ∂H sI
ω1
(
≈ 2 ℑ H s (ω1 , ρ1 ) +
∂ρ I
) ∆ρ I
ω 2 , ρ1
Figure 3. True resistivity model used in the synthetic test in
Combining this with the equations for the ISIP data (Equation log10 Ωm . The overburden is 40m thick and has a
7) leaves resistivity of 1000 Ωm and the background is 100 Ωm . The
block has a resistivity of 2 Ωm , a chargeability of 0.1, and
∂H sR ∂H sR time-constant of 0.1.
dℜ ≈ ∆ρ ≈ − ρ1ητω2 (14a)
∂ρ I ∂ρ I
ω2 , ρ1 ω2 , ρ1 Inverse modelling
∂H sI ∂H s
dℑ ≈ ∆ρ I ≈ − I ρ1ητω 2 (14b) The goal of the inversion is to recover a chargability
∂ρ I ∂ρ I distribution which can predict the observations while
ω2 , ρ1 ω2 , ρ1
minimizing a predefined objective function. Prior to inverting
To evaluate this we need to compute the sensitivities of Hs to for chargeability, we must have a real background resistivity,
changes in the imaginary part of the resistivity. We define the ρ1 , from which to compute the sensitivity elements GI and
operators TR and TI to be GR. This background can be obtained by inverting the low
frequency ( ω1 ) data from all transmitters and receivers if the
TR = curl T diag (A v ρ R )curl + graddiag (A v ρ R )gradT µ (15a) data are sufficiently numerous. The desired sensitivities can be
computed via Equation 18. The inverse problem is solved by
TI = curl diag (A v ρ I )curl + graddiag (A v ρ I )grad µ
T T
(15b)
minimizing the usual objective function φ = φd + βφm
Differentiating Equation 9 with respect to ρ I , and separating where φd is the data misfit, φm is a measure of the amount of
the real and the imaginary parts gives structure present in the model and β is the regularizatoin
parameter. A Gauss-Newton procedure is used and the matrix
∂H sR ∂H s system solved using Conjugate Gradient techniques. The
TR
∂ρ I
− (TI + ωµI ) I = curlT diag curl H sI A v
∂ρ I
( ) (16a)

nd
22 International Geophysical Conference and Exhibition, 26-29 February 2012 - Brisbane, Australia 3
 Inductive Source Induced Polarization Marchant, Haber and Oldenburg

Figure 4. True chargeability distribution.

Figure 5. Chargeability distribution recovered from
sensitivity is kept in factored form (Haber et al, 2004) with the inversion of synthetic data.
forward modelling operator factored as a Cholesky
decomposition. Positivity on the chargeability is achieved by REFERENCES
implementing a projected gradient method. (Kelly, 1999).
Chen, J. and Oldenburg, D. W., 2003, 3-D inverion of
Synthetic Example magnetic induced polarization data, in J. Macnae and G. Liu,
eds., ASEG, Three-dimensional electromagnetics III: 11p.
A synthetic resistivity model was created containing a
conductive, chargeable block ( ρ 0 =2 Ωm , η =0.1, τ =0.1, Gasperikova, E. and Morrison, H. F., 2001, Mapping of
c=1), buried beneath a resistive overburden (1000 Ωm ). The induced polarization using natural fields: Geophysics, 66,137-
background had a resistivity of 100 Ωm . The resistive 147.
overburden was 40m thick. The block had dimensions of 80m
Haber, E. and Ascher, U. M., 2001, Fast fintie volume
on a side, with the top of the block being 60m below the
simulation of 3d electromagnetic problems with highly
surface. The synthetic resistivity model is shown in Figure 3.
discontinuous coefficients: SIAM Journal on Scientific
Computing, 22, 1943-1961.
The frequency domain response of this complex resistivity
distribution was simulated for 9 square loop transmittors. The
Haber, E., Ascher, U. M. and Oldenburg, D. W., 2004,
transmitters were on the surface, were 50m on a side, and were
Inversion of 3D electromagnetic data in frequency and time
distributed as a 3 by 3 grid. A grid of 64 recievers (placed
using an inexact all-at-once approach: Geophysics, 69,1216-
every 40m, 1m above the surface) recorded the three
1228.
components of the magnetic field. Data were modelled at 1hz
and 2hz. The real, synthethic ISIP data were then calculated
Hohmann, G. W., Kintzinger, P. R., Van Voorhis, G. D. and
using Equation 7a.
Ward, S. H., 1970, Evaluation of the measurment of induced
electrical polarization with an inductive system: Geophysics,
The ISIP data were contaminated with normaly distributed,
35, 901-915.
random noise prior to inversion. The noise had a standard
deviation of 5% of the data value plus a small floor. The data
Kelley, C. T., 1999, Iterative methods for optimization: SIAM,
were then inverted using the methodology presented
Philadelphia.
previously. The true and resulting chargeability models are
shown in Figures 4 and 5.
Li, Y. and Oldenburg, D. W., 2000, 3-D inversion of induced
polarization data: Geophysics, 65, 1931-1945.
Conclusion
Oldenburg, D. W. and Li, Y., 1994, Inversion of induced
In this work we have explored a new imaging technique to polarization data: Geophysics, 59, 1327-1341.
recover the distribution of chargeable material in the earth
given a magnetic sources and magnetic receivers. Our method Pelton, S. H., Ward, S. H., Hallof, P. G., Sill, W. R., and
is based on low frequency measurements so that the inductive Nelson, P. H., 1978, Mineral discrimination and removal of
responses to a real conductivity distribution are minimally inductive coupling with multifrequency IP: Geophysics, 43,
different at the two frequencies. That is, we are working in the 588-609.
resistive limit regime. We have developed a mathematical
framework and applied it to synthetic data set that Seigel, H. O., 1959, Mathematical formulation and type
demonstrates the potential of the method. curves for induced polarization: Geophysics, 24, 547-565.

Seigel, H. O., 1974, The magnetic induced polarization (MIP)

method: Geophysics, 39, 321-339.

nd
22 International Geophysical Conference and Exhibition, 26-29 February 2012 - Brisbane, Australia 4

View publication stats