Вы находитесь на странице: 1из 9

GEOPHYSICS, VOL. 67, NO. 4 (JULY-AUGUST 2002); P. 1095–1103, 6 FIGS., 1 TABLE.

10.1190/1.1500370

The moments of the impulse response: A new paradigm


for the interpretation of transient electromagnetic data

Richard S. Smith∗ and Terry J. Lee‡

ABSTRACT strongly dependent on any distortion of the waveform


We define the nth moment of the transient electro- from an ideal impulse. Hence, it is not critical to decon-
magnetic impulse response as the definite integral with volve the data prior to estimating the moments.
respect to time of the “quadrature” magnetic-field im- If a conductor has a single exponential decay, the nth
pulse response weighted by time to the nth power. In moment of the decay is proportional to the nth power of
this context, the quadrature response is defined as the the time constant of the exponential. Thus, it is relatively
full impulse response with the in-phase component (i.e., easy to estimate the time constant from the moments.
the delta function component at zero time) removed. The For a conductive sphere model, the expressions for the
low-order moments are equivalent to familiar quantities: moments are more complicated, but are still simpler
the zeroth moment (n = 0) is numerically equal to the than the full transient solution or the frequency-domain
frequency-domain inductive limit, and the first moment solution.
is the resistive-limit response. The higher order moments In a field example, the high-order moments empha-
can be of particular benefit: successively they put greater size local highly conductive features, but also show the
emphasis on the late-time data, and hence can bring out noise present in the late-time data. A discrete feature on
features in the data that are more conductive or deeper. the profile evident in moments 3 through 10 has been
An advantage of calculating moments (and hence the modeled as a spherical conductor with its center at 90 m
inductive and resistive limit) is that these data are not depth, a radius of 45 m, and a conductivity of 9.4 S/m.

INTRODUCTION gives the impulse response. A more detailed discussion of these


topics can be found in Grant and West (1965).
The magnetic-field transient-electromagnetic (TEM) re- In this paper, we use the term “impulse response” to mean
sponse consists of the secondary magnetic field response as a the secondary magnetic-field response resulting from a short
function of time after the excitation field has suddenly changed. pulse in the magnetic field. Functionally, this is the same as
If the change is a negligibly short cessation of the excitation the voltage resulting from a step in magnetic field, except of
field, then the measured response is the step response. If the course the units are different. Also, our definition of the im-
sudden change is a short pulse of unit time-amplitude (area), pulse response has the delta function at zero time removed.
then the measured response is the impulse response. In theory, This is equivalent to removing the in-phase component of the
the impulse response can be derived from the step response us- secondary response (Smith, 2001), so we call this modified im-
ing the time-derivative operator. Transient data are normally pulse response the “quadrature” impulse response.
interpreted by assuming a model structure, calculating the tran- The inverse transform of the frequency-domain response
sient response (step or impulse) and trying to match the calcu- requires calculation of the response at multiple frequencies,
lated data with the measured data. For a few cases, the transient which can be computationally expensive. Hence, there has been
response can be calculated directly. Otherwise the transient a need to derive simpler approaches for the interpretation of
response can be calculated by inverse Fourier transformation transient data. One solution is to use simple time-domain mod-
of the calculated frequency-domain response (e.g., Morrison els such as wire loop circuits (Grant and West, 1965; Barnett,
et al., 1969). If the frequency-domain excitation has constant 1984; Smith and West, 1988) or thin sheets (Grant and West,
power at all frequencies, then the inverse Fourier transform 1965; Macnae and Lamontagne, 1987; Smith 2000).

Published on Geophysics Online January 7, 2002. Manuscript received by the Editor August 9, 2000; revised manuscript received November 9, 2001.

Fugro Airborne Surveys, 2060 Walkley Road, Ottawa, Ontario K1G 3P5, Canada. E-mail: rsmith@fugroairborne.com.
‡P.O. Box 1984, Canberra, ACT 2601, Australia. E-mail: TerryJLee@bigpond.com.
°c 2002 Society of Exploration Geophysicists. All rights reserved.
1095
1096 Smith and Lee

Even in the frequency domain, the calculation of a model re- sponse moments for the wire-loop and sphere models. In cases
sponse at many frequencies can be time-consuming, requiring when the electromagnetic systems only measure an approxima-
the evaluation of special functions such as Bessel functions. Us- tion to the impulse response, we argue that a good estimate of
ing the example of a sphere in a uniform field, Grant and West the moments can be obtained simply from the approximate im-
(1965) show that the limiting cases of high-frequency response pulse response. Finally, a field data set is used to show how the
(the inductive limit) and low-frequency response (the resis- higher order moments enhances the detection of a deep con-
tive limit) can reveal critical information about the conductive ductive feature. By fitting the fourth-order moment, we show
body. For example, the ratio of the resistive to inductive limits that a sphere of radius 45 m, depth 90 m, and conductivity
can be used to estimate the transition frequency or effective 9.4 S/m can be used to model the deep feature.
time constant of a conductor. This suggests a second approach
[advocated by Macnae et al. (1998, 1999)] to evaluate these two MOMENTS OF THE IMPULSE RESPONSE
limits from time-domain data and use these quantities for mod-
eling. Evaluation of the inductive and resistive limits requires The nth moment of the impulse response I (t) is defined as
deconvolving the data, a procedure that assumes that the ideal Z ∞
step response can be approximated by a sum of weighted ex- M =
n
t n I (t) dt (1)
ponentials with time constants that are empirically selected. 0
By differentiating the exponential functions, convolving these (all symbols are defined in Table 1). Integrals of this form are of
with a known system waveform, and fitting this to a measured interest to mathematicians, being related to the Mellin trans-
response, it is possible to determine the weights. The inductive form (Sneddon, 1972) and occurring in the moment problem
limit is the sum of the weights of all the exponentials and the (Widder, 1941). These authors show that in order to produce
resistive limit is the sum of the weights multiplied by the re- valid mathematical results, restrictions need to be placed on
spective time constants (Macnae et al., 1998, 1999; Stolz and the integrand (e.g., existence and integrability). In the specific
Macnae, 1998). This exponential decomposition is related to case of equation (1), as t → ∞, I (t) must vanish faster than
procedures for inverting Laplace transforms (Bellman et al., t n grows. For example, the response of a half-space decays at
1966; Pressens, 1971a, b; Longman, 1968; Longman and Sharir, t −5/2 (Lee and Lewis, 1974; O’Brien and Smith, 1985), so the
1971). The decomposition procedure requires that the range moments do not formally exist for n ≥ 2. However, in practice,
and the spacing of the exponential time constants be carefully the upper limit of integration is actually the end of the off-time,
selected and that the inversion be carefully regularized. By so the integrals can be evaluated.
comparison, using moments is more straightforward. We begin by showing how these moments relate to other
The resistive-limit concept used by Macnae and coworkers quantities used in frequency- and time-domain interpretation.
has been extended by Smith (2000) to the realizable resistive For the purpose of this discussion, we assume that the sec-
limit to map the apparent conductance of the ground. In the ondary step-on response of a conductor can be written in the
resistive limit, the response of the ground as a whole is the sum form
of the responses of the individual parts weighted by the con-
ductivity of the part [Wait (1982) demonstrates this for a three- R(t) = −Bu(t)H (t), (2)
layer nonpermeable earth]. Hence, the contributions from all
where B is the response at t = 0, u(t) is the unit-step-on func-
the material close to the transmitter and receiver can dominate
tion, and H (t) is a dimensionless function characterizing the
over the response of relatively small features that are deeper,
secondary decay associated with currents in the ground. The
even if they are more conductive. This is consistent with our ob-
negative sign is included for consistency with other workers.
servation that apparent conductance maps (Smith, 2000) tend
The function H (t) is continuous and differentiable with the
to reflect the material nearer to the surface. The profiles of
constraints that it has a value of 1 at t = 0+, vanishes at large
the resistive limit also show a significant background response,
values of t, and has no delta-function–like features or un-
which is problematic to remove and then interpret with the
damped oscillations. The impulse response I (t) is the time
automatic anomaly picking programs described by Stolz and
derivative of R(t):
Macnae (1997) and Macnae et al., (1998).
½ ¾
Electromagnetic responses can often be written as sums of ∂ H (t)
terms, where each term is related to a pole or a pole pair in the I (t) = −B δ(t)H (t) + u(t) . (3)
∂t
complex frequency plane. In what follows, we show that the
higher order moments are dominated by the terms associated In this paper, we remove the first term containing the delta
with the poles closest to the origin. The ratios of successive function at zero time. This is equivalent to removing the in-
moments give characteristic times, which have interpretative phase response (Smith, 2001), so we call what results the
value. quadrature impulse response:
There might be other ways of enhancing particular features ∂ H (t)
in the data; for example, if the moment method were extended I (t) = −Bu(t) . (4)
or generalized, it might be possible to derive analytic rela- ∂t
tions that allow other information to be determined. The use The quadrature response is the response which is normally
of higher order moments is not the only way of focusing on the measured by “off-time” impulse response systems, because
lower order terms, a suitable modification of the transmitter they restrict their measurements to times when t > 0. In prac-
waveform can achieve the same goal (e.g., Lee, 1979). tice, even the systems which measure in the “on-time” also mea-
In this paper, we define the moments of the quadrature im- sure the quadrature response: the airborne systems remove the
pulse response and then derive the quadrature impulse re- in-phase component using the procedure described by Smith
Moments of the Impulse Response 1097

(2001), and step-response systems like Utem (West et al, 1984) These two quantities can be related to frequency-domain quan-
and Spectrem (Leggatt et al., 2000) remove the late-time re- tities. If the Fourier transform of I (t) = ∂ R/∂t is denoted Î (ω),
sponse, which is an approximation to removing the in-phase then the inductive (high-frequency) limit of Î (ω) by the initial-
part of the impulse response. Using the constraints H (0) = 1 value theorem of the Laplace transform can be shown to be
and H (∞) = 0, it is possible to show that related to the time-domain step response:

M 0 = B, (5) lim Î (ω) = R(0). (7)


ω→∞
and using integration by parts (Papoulis, 1962; Lamontagne, 1975). The right-hand side is sim-
Z ∞ ply B, so the high-frequency limit is related to the zeroth-order
M1 = B H (t) dt. (6) moment:
0
lim Î (ω) = M 0 . (8)
ω→∞
Table 1. Symbols. Macnae et al. (1999) also show that the high-frequency limit
B inductive limit and the early-time step response are identical for responses
t time which can be decomposed into decaying exponentials.
u(t) unit-step-on function The resistive limit in the frequency domain is the low-
H (t) dimensionless function of time; H (0) = 1, frequency limit of the derivative with respect to ω of the imag-
H (∞) = 0 inary component. Bracewell (1965) shows that this is equal to
R(t) step response the first-order moment
I (t) impulse response ¯ Z ∞
Î (ω) Fourier transform of I (t) 1 ∂ Î (ω) ¯¯
= t I (t) dt. (9)
Mn
ω
moment of the impulse response of order n
angular frequency
i ∂ω ¯ω=0 0
K (t) a realizable impulse response Bracewell (1965) generalized this to the higher-order moments
w(t) a weighting function which smears and delays µ ¶n n ¯ Z ∞
the ideal impulse response 1 ∂ Î (ω) ¯¯
τ = t n I (t) dt, (10)
∂ωn ¯ω=0
convolution integration variable
t0 position that the approximately symmetric w(t) i 0
is a maximum which implicitly assumes that the higher order derivatives of
T half-width of the weighting function w(t)
Ai weights of the exponential decomposition the spectrum exist.
τi time constants of the exponential decomposition The zeroth-order moment is thus equivalent to the inductive
M̃ n moment of the impulse response of order n limit, and the first-order moment is equivalent to the resistive
calculated from the smeared impulse K (t) limit in the frequency domain and the integral of the step re-
t0 integration variable sponse in the time domain.
τw time
√ constant (in wire-loop model section)
i −1 WIRE-LOOP MODEL
Hi magnetic field in the radial (i = r ), latitudinal
(i = ϑ), or longitudinal (i = ϕ) directions The response of a simple wire-loop circuit is one of the simple
mj dipole moment in the j = r , ϑ, or models used in interpretation of EM data (Grant and West,
direction perpendicular to ϑ 1965; Barnett, 1984; Duncan, 1987, Smith and West, 1988). For
a sphere radius
r0 radial distance to the transmitter this model, the time-dependent part of the step response, H (t),
r radial distance to the receiver can be written as
ϑ latitudinal angle to the receiver
ϕ longitudinal angle to the receiver H (t) = u(t)e−t/τw , (11)
j
Fi function given in equations (17-45), (17-48),
where τw is the time constant defined by the ratio of the loop
and (17-49) of Grant and West (1965)
P` Legendre coefficients degree ` inductance to the loop resistance (Grant and West, 1965). Us-
P`1 Legendre coefficients degree `, order 1 ing integration by parts, it is possible to prove by induction that
` summation variable for the wire loop case
X + iY Frequency dependent part of the sphere
response M n = Bn!τwn , (12)
I`+1/2 (ka) modified
√ Bessel function of order ` + 12
k = iωµσ where B in this case involves only the coupling coefficients.
µ magnetic permeability (assumed equal to Thus, it is a simple matter to determine the characteristic time
free space permeability) constant of a conductor
σ electrical conductivity
M n+1
γ = ` + 12 τw = , (13)
4(t) Fourier transform of X + iY (n + 1)M n
s Laplace transform variable
c constant on real axis of complex plane s where n can take on any value.
c0 constant on real axis of complex plane λ Some insight into the different moments can be obtained
λ integration variable by considering the integrand t n e−t/τ . This function has a max-
Jγ −1 Bessel function degree γ − 1 imum contribution at the turning point, where n − t/τ = 0. As
λk kth zeros of Jγ −1 (λ) the order increases, then the time at which the function is a
1098 Smith and Lee

X · ¸
maximum (t = nτ ) increases, which effectively means that 2 2 Jγ +1 (λk ) 2λ
higher moments place greater emphasis on the late-time data. 4(t) = e−λk t/(µσ a ) . (20)
k
Jγ0 −1 (λk ) µσ a 2
We now turn our attention to the more general situation
where there is more than one time constant. Two applications of the recurrence relations in equa-
tion (9.6.26) of Abramowitz and Stegun (1965) allows us to re-
SPHERE MODEL
write the denominator as a term very similar to the numerator
The electromagnetic response of a sphere is taken directly  
from the expressions given in Grant and West (1965). The X 2  Jγ +1 (λk ) 
magnetic field expressions are of the form 4(t) =
2
e−λk t/(µσ a )   2 , (21)
 1  µσ a 2
k − Jγ +1 (λk )
mj X∞
a 2`+1 j© ª 2γ
Hi = (X + iY ) `+2
Fi `, P`1 (cos ϑ), P` (cos ϑ) ,
4π `=1 (rr0 ) which cancels, giving the simple exponential expression
(14) X 2 2) 4γ
4(t) = − e−λk t/(µσ a , (22)
where Hi denotes the component in the radial (i = r ), latitu- k
µσ a 2
dinal (i = ϑ), or longitudinal (i = ϕ) direction; m j is the dipole
moment in the radial, latitudinal, or direction perpendicular where γ is defined below equation (16) and is a function of
to latitudinal; a is the sphere radius; r0 is the radial distance the index ` of the sum in equation (14). This is the only time-
from the center of the sphere to the transmitter; r is the radial dependent term in the expression. The zeroth-order moment
distance to the receiver; ϑ is the latitudinal angle; and Fi is
j of 4(t) is
a function dependent on `, P` , and P` (the Legendre coeffi-
1 Z ∞ X 1
cients), the exact form of which are given in Grant and West 4(t) dt = 4γ . (23)
k λk
2
(1965) equations (17-45), (17-48), and (17-49). The only fre- 0

quency dependent term is X + iY , which is given in equation The sum of the inverse squares of the zeros of the Bessel func-
(17-44) of Grant and West (1965) as tion are given on page 502 of Watson (1944) as
 µ ¶ 
` + 12 0
X 1 1
 − ka I` + 1/2 (ka) + I` + 1/2 (ka)  = , (24)
  k λk
2 4γ
X + iY =  µ ¶  , (15)
 ` + 12 0

I` + 1/2 (ka) + I` + 1/2 (ka) so the zeroth-order moment of 4(t) is unity. The higher-order
ka moments of 4(t) can be calculated by integration by parts
where I`+1/2 giving
√ (ka) is the
√ modified Bessel function of order Z
` + 12 , k = iωµσ , i = −1, the prime denotes the derivative ∞
4γ X 1
with respect to the argument, and we have made the simpli- t n 4(t) dt = n!(µσ a 2 n+1
) ¡ 2 ¢n+1 . (25)
0 µσ a 2
λ
fying assumption that µ = µ0 . This expression can be further k k
simplified using the second and fourth recurrence relations in Once again, Watson (1944) gives an equation for the sum of
equation (9.6.26) of Abramowitz and Stegun (1965) to the inverse squares of the zeros of the Bessel functions for n up
· ¸
Iγ +1 (ka) to 4. The higher order sums require fewer terms, so in principle
X + iY = , (16) we can estimate the dominant time constants.
Iγ −1 (ka)
where γ = ` + 12 . The inverse Laplace transform, 4(t), of Validation of numerical solution
X + iY is given by
Z c−i∞ · ¸ Because the temporal decay has been integrated out, the
1 Iγ +1 (ka)
4(t) = ets ds, (17) moments of the sphere response can be calculated in a straight-
2πi c−i∞ Iγ −1 (ka) forward manner. For the zeroth-order moment, it is possible
to sum the expansion containing the Legendre functions ana-
where s = iω. Changing the integration variable to λ using
lytically; however, our implementation sums the expansion nu-
s = −λ2 /µσ a 2 , we can write
merically using the recurrence relation for the higher degree
Z ic0 +∞ · ¸
1 2 t/(µσ a 2 ) Jγ +1 (λ) 2λ polynomials. As a check of the program, we have calculated the
4(t) = e−λ dλ. zeroth-order moment for a sphere with a radius of 91.4 m, when
2πi ic0 −∞ Jγ −1 (λ) µσ a 2
the transmitter-receiver configuration is a coplanar horizontal
(18) dipole system mounted on the wingtips of an aircraft, sepa-
This integral can be evaluated using the residue theorem, rated by 18.3 m and flying 91.4 m above the top of the sphere.
X · ¸ This configuration was chosen, because the inductive-limit re-
−λ2 t/(µσ a 2 ) Jγ +1 (λ) 2λ
4(t) = lim (λ − λk )e , sponse for this configuration is presented in Figure 18-13 of
residues
λ→λk Jγ −1 (λ) µσ a 2 Grant and West (1965). The solution of Grant and West (1965)
(19) was also used for validating the numerical inductive-limit solu-
tion of King and Macnae (2001). The profile and contour map
where λk are the zeros of Jγ −1 (λ). Expanding the Jγ −1 (λ) as a of the response presented in Grant and West (1965) has been
Taylor series, and then setting λ = λk , we get digitized by hand and plotted as circles on Figure 1, along with
Moments of the Impulse Response 1099

the numerical results for the zeroth-order moment obtained as a continuous line. The agreement between the two quantities
by our method (solid line). The results agree to the digitization is quite good.
accuracy: the value at the peak position, 0.182, compares with
a peak value of 0.183 in the caption of Grant and West (1965). NONIDEAL WAVEFORMS
Equation (9) allows us to compare the first-order moment
against the frequency-domain resistive limit. In this case, the In reality, electromagnetic systems cannot measure the ideal
derivative with respect to frequency is estimated using a fi- quadrature impulse response, so equation (1) cannot be used to
nite difference of the sphere response at 0.156 and 0.318 Hz. calculate the moments. The response K (t) actually measured
The frequency-domain sphere response is calculated using the with a real system is a smeared version of the ideal impulse I (t).
EMSphere program developed by PetRos Eikon Inc, which is This smearing process can be written as a convolution integral
based partially on the formulation of Debye (1909). The sphere Z ∞
has a conductivity of 0.1 S/m, a radius of 50 m, and a depth be- K (t) = w(τ )I (t − τ ) dτ , (26)
low the transmitter of 220 m. The receiver is offset by 50 m −∞
vertically and 130 m horizontally from the transmitter and the where the integration variable is the lag τ , and we define w(τ ) as
plotting point is at the receiver. The background rock is in- a pulse-weighting function that smears and delays the impulse.
sulating. The z-component frequency-domain resistive limit is Note that I (t) is a causal function. For convenience, we can
plotted as circles on Figure 2; the first-order moment is plotted define the function w(τ ) to have unit time-amplitude area:
Z ∞
w(τ ) dτ = 1. (27)
−∞

In many ground electromagnetic systems, the transmitter cur-


rent is essentially a step-off and the observed response is the
voltage from the receiver coil. To within a proportionality
factor, this is practically the magnetic field impulse response,
where the duration of the impulse is the current turn-off time.
Thus, the weighting function is a boxcar of short duration, or a
band-limited version of a boxcar. For the airborne electromag-
netic system which provides the example data of this paper, the
magnetic field response is observed, and the transmitter cur-
rent waveform (weighting function) is a half sine pulse of rela-
tively short duration. For the purposes of this discussion, we will
assume that the weighting function is symmetric about some
midpoint t0 . Hence equation (27) can be split into two parts:
Z t0
FIG. 1. The inductive-limit response digitized from 1
w(τ ) dτ = , (28)
Figure 18-13 of Grant and West (1965), plotted with circles, −∞ 2
compared with the zeroth-order moment (solid line). The
model is a sphere radius 91.4 m with a transmitter receiver and
system flying 91.4 m above the top of the sphere. The trans- Z ∞
mitter and receiver are horizontal coplanar coils separated by 1
w(τ ) dτ = . (29)
18.3 m. t0 2
Typically, w(τ ) is nonzero over a small range of times, the
duration of which is equal to the ramp switch-off time, which is
typically a fraction of a millisecond [e.g., 0.42 ms cited in Smith
and Balch (2000)]. This pulse width is generally engineered to
be shorter than the range of time constants that characterize
the responses the system is intended to measure. If the finite
width of the pulse is denoted 2T , then practically the positive
and negative infinite limits of the convolution integral can be
replaced with t0 + T and t0 − T , respectively.
For the purposes of the rest of the discussion in this section,
we will assume that the quadrature impulse response can be
represented as a sum of decaying exponentials. Geophysicists
frequently describe the response of structures in such terms.
These structures include not only spheres and wire loops (dis-
cussed above), but more general ones as well. For example,
Kaufman (1978) provided some analyses that extend the mod-
eling to less regular shapes. Also, we know that an exponential
FIG. 2. The z-component resistive-limit response from EM- decay can be expected from each “wavelength in the lambda
Sphere (circles) compared with the first-order moment (dashed domain” for a ground that can be modeled by a slab (Lee, 1982)
line). or a ground whose conductivity decreases exponentially with
1100 Smith and Lee

depth (Lee and Ignetik, 1994). The pure quadrature impulse 1.26 ms after the start of the waveform. The first two measure-
response is written ment windows being before the center of the pulse, the third
X window is near the center, and all subsequent windows are af-
I (t) = Ai e−t/τi , (30) ter the center of the pulse. The centers of each measurement
i windows are listed in the caption of Figure 3. The profile of Fig-
where the Ai are the amplitudes of each exponential and the ure 3 is dominated by data acquired in the pulse, and there are
τi are the characteristic decay constants. Substituting this ex- three distinct features evident in these data: one is to the left
pression into equation (1) and integrating by parts n + 1 times, of 6500, another at about 6750, and the third at 6900. There is a
we get the expression bedrock feature at 6670, which is barely visible at this scale, but
X is more apparent when the vertical scale is expanded (bottom
M n = n! Ai τin+1 . (31) panel of plot).
i The moments of the impulse response M 0 toM 9 have been
This is the quantity we would like to determine from our mea- calculated using equation (33) and shown in Figure 4. Be-
surement of K (t), the smeared version of the ideal impulse. cause windows 1 and 2 occur before the midpoint, these are
Substituting equation (30) into equation (26) gives
Z ∞ X
K (t) = w(τ ) Ai u(t − τ )e−(t−τ )/τi dτ . (32)
−∞ i

The formula we propose using to obtain an approximate


estimate M̃ n of the nth moment is
Z ∞
M̃ n = 2 (t − t0 )n K (t) dt. (33)
t0

If equation (32) is substituted into this expression, the integra-


tion variable is changed to t 0 = t − t0 . If then we undertake the
t 0 integration using integration by parts n + 1 times, we get the
result
X Z t0
M̃ n = 2n! Ai τin+1 w(τ )e−(t0 −τ )/τi dτ. (34)
i t0 −T

The argument of the exponential is in the range [−T /τi , 0].


Because electromagnetic systems are designed so that T is
small compared with the τi , the exponential function will be
close to unity. This will be particularly true for the higher order
moments as the τ n+1 factor results in the terms associated with
large τ dominating. For these terms, the argument of the ex-
ponential is particularly small, and the exponent will be even
closer to unity. The convolution integral will therefore yield a
value close to 12 [equation (28)], and we can write
X
M̃ n ≈ n! Ai τin+1 . (35)
i

This is the same result as the moment of the pure quadrature


impulse (equation 31). Hence, equation (33), can be used to es-
timate the moments even when the smeared impulse response
has been measured.

FIELD EXAMPLE

As an example of how the approximate moments of the


impulse response can be used in the interpretation of field
data, we take an example of airborne electromagnetic data FIG. 3. The Megatem horizontal in-line (x) component, ac-
collected over the Reid-Mahaffy test site (Ontario Geologi- quired on traverse line 15 of the Ontario Geologic Survey/
Operation Treasure Hunt Reid-Mahaffy test site. The wave-
cal Survey, 2000). The Ontario Geological Survey, Ministry of form repetition rate is 90 Hz, and each curve is a different
Northern Development and Mines, defined this test site as part measurement window of the transient response. The window
of their “Operation Treasure Hunt” program. The horizontal times for windows 1 to 20 are 0.28, 0.72, 1.28, 1.85, 2.21, 2.34,
in-line (x) component of the magnetic-field response acquired 2.45, 2.56, 2.71, 2.87, 3.04, 3.21, 3.41, 3.62, 3.84, 4.08, 4.34, 4.62,
4.95, 5.34 ms after the start of the transmitter half-sine wave-
with the Megatem system has been plotted in Figure 3. For form. The center of the pulse, t0 , is at 1.26 ms. The top panel
this survey, the ideal impulse response has been delayed and shows all windows, and the bottom panel the late time windows
smeared by a half-sine with T ∼ 1 ms, the center of which (t0 ) is at an expanded scale.
Moments of the Impulse Response 1101

not included in the integral. For display purposes, the data in the higher order moments. This known conductor is about
have been converted to the same units by normalizing by the 100 m deep, so the higher order moments are better at empha-
moment of an exponential with a time constant of 0.48 ms sizing this deeper, more conductive feature. This is because
[calculated using equation (12) with B = 1]. This value was se- the t n in the integral of equation (1) puts greater emphasis
lected as it corresponds to a typical time constant for the pro- on the late-time data. The higher order moments will show
file line (Figure 5). The first-order moment has been checked an increase in the noise evident in the data; for example, the
against the resistive limit calculated using the method of Smith three features to the left of 6600 are noise related. However,
(2000) and found to agree to within a few percent. Note that the signal-to-noise ratio is comparable to the late-time tran-
the first-order moment shows no obvious feature at 6670. This sient data. Note that the presentation of the data on Figure 4
is consistent with our experience from using the method of is quite different from the conventional display of transient
Smith (2000) that features in the near surface are mapped with electromagnetic data. In conventional presentation of tran-
the first-order moment, but deeper features are generally not sient electromagnetic data, the early-time features have large
apparent. amplitudes and the late-time features smaller amplitudes. In
When the data are normalized by a specific time constant, this presentation, the largest amplitude features are associated
as n increases, each subsequent moment will increase if the with the late-time, low-amplitude, slowly decaying features.
time constant of the decay at that location is greater than the As an example of how the time constant can be estimated
normalizing time constant. Thus, the formational conductor for this profile line, Figure 5 shows the estimated time constant
to the left of 6500 appears to have a large time constant, and when the fourth and fifth moments are used in equation (13).
the known bedrock conductor at 6670 becomes more apparent The feature with the greatest time constant is the bedrock con-
ductor at 6670.
The simplicity of the formulation for calculating the mo-
ments of the sphere response means that it is straightforward
to interactively model the moments derived from measured
data. In this case, the third moment of the bedrock conductor
has been extracted from the profile of Figure 4, and the back-
ground response removed. The resulting data, shown with a
dashed line on Figure 6, has a sharp peak at an offset of −110 m
on the profile, and a slight shoulder indicating another smaller
body at about 0 offset. This second peak indicates that the ideal
model for the bedrock conductor is not a sphere model. In fact,
the z-component data indicates that a thin plate model may ac-
tually be more appropriate. However, we have modeled these
x-component data with a sphere model as there is moderately
good depth information (in the form of drillhole data) and we
wish to show that relatively useful information can be extracted
even if the model is not ideal.
The anomaly has been interactively forward modeled and
a relatively good fit (in terms of width and amplitude) is ob-
FIG. 4. The moment of the impulse response M 0 to M 9 for the
profile line plotted in Figure 3. The data have been normalized tained using a sphere at 90 m depth with a radius of 45 m
by the moments of a conductor with a time constant of 0.48 ms.
The bedrock conductor at 6670 becomes apparent as a clear
feature on this data.

FIG. 6. The third moment extracted from the profile of Figure 4,


with the background removed (dashed line). This data has been
matched with the third moment of a sphere with a radius of
FIG. 5. The time constant profile for the line shown in Figure 3 45 m, a depth below surface of 90 m, and a conductivity of
estimated from the fourth and fifth moments. 9.4 S/m (solid line).
1102 Smith and Lee

and a conductivity of 9.4 S/m (solid line on Figure 6). This The wire-loop model has moments that exist in a simple an-
is consistent with the drill information for this line, which alytic form. These formulas provide us with a simple means of
has a narrow intersection of sulfides at about 120-m depth. estimating a characteristic time constant from the data. Analy-
An exact agreement in depth is not necessarily expected be- sis of the wire-loop model also gives us an understanding why
cause of the inadequacy of the sphere model and the fact than the higher order moments will emphasize the late-time data.
the drill hole only intersects a small volume of rock, whereas The full transient response of a sphere model is traditionally
the electromagnetic system has a footprint of several hundred calculated by evaluating the modified Bessel function in equa-
meters. We expect that other combinations of the sphere radius tion (14) at a range of frequencies and then transforming these
and conductivity would likely give an equally good fit to the data to the time domain by inverse Fourier transform. How-
data. ever, the moments of the impulse response for a sphere model
have a simpler mathematical form, making modeling and inter-
DISCUSSION
pretation simpler. This simplification, for this and other models,
could make the process of modeling and inverting TEM data
The method of moments of the impulse response uses significantly less complex.
a t n term in the moment integral. It might be possible to
extend this concept to include more general functions f n (t) ACKNOWLEDGMENTS
in the integral. This could have application in emphasizing cer-
We thank Ross Groom and Danielle Parker of PetRos
tain features in the data (or its spectrum), or in allowing a
Eikon Inc. for their assistance in generating the EMSphere
simpler interpretation of data. For example, a different type of
data shown in Figure 2. We are grateful to Sharon Taylor of
moment might be used to interpret data over a layered half-
Falconbridge for providing ground truth on the bedrock con-
space or for interpreting conductors embedded in a layered
ductor. Richard Smith is grateful to Fugro Airborne Surveys
half-space.
for permission to publish this paper. Terry Lee thanks David
Throughout the analysis of this paper, the moment integral
Denham of AGSO for encouragement in undertaking this
has been written with an upper limit of infinity. If the moment
work. Megatem is a registered trademark of Fugro Airborne
kernel t n I (t) is nonzero at large values, such as will occur for
Surveys.
half-spaces when n ≥ 2, then the moment integrals do not exist.
However, electromagnetic systems use a repetitive waveform
REFERENCES
with pulses of alternating sign and, as a consequence, have a
finite off time. Hence, when these moments are calculated from Abramowitz, M., and Stegun, I. A., 1965, Handbook of mathematical
functions: Dover Publications.
measured data, the upper limit of the moment integral is set to Barnett, C. T., 1984, Simple inversion of time-domain electromagnetic
the end of the off-time and the moment integrals must all exist. data: Geophysics, 49, 925–933.
The theoretical and measured moments should be reasonably Bellman, R., Kalaba, R. E., and Lockett, J. A., 1966, Numerical inver-
sion of the Laplace transform: American Elsevier Publishing Co.
close if the measured response has decayed to the noise level Bracewell, R., 1965, The Fourier transform and its application:
before the end of the off-time. For the case of the inductive- McGraw-Hill.
limit (zeroth-order moment), Smith and Balch (2000) argue Debye, P., 1909, Der lichtdruck auf kugeln von beliebigern material:
Ann. d. Phys. 30, 57.
that small nonzero responses at the end of the off-time will Duncan, A. C., 1987, Interpretation of down-hole transient EM data
have a minimal effect. However, for higher order moments, using current filaments: 13th Geophys. Conf. Austral. Soc. Expl.
Geophy., 18, 36–39.
the effect would become greater, possibly severe, so a more so- Grant, F. S., and West, G. F., 1965, Interpretation theory in applied
phisticated analysis that takes into account the repetitive wave- geophysics: McGraw-Hill.
form may be required. One possible approach to this analysis Kaufman, A., 1978, Frequency and transient responses of electromag-
netic fields created by currents in confined conductors: Geophysics,
would be to adapt the exponential decomposition of Stolz and 43, 1002–1010.
Macnae (1998), as this methodology is able to account for wave- King, A., and Macnae, J. C., 2001, Modelling of the EM inductive-limit
form periodicity. surface currents: Geophysics, 66, 476–481.
Lamontagne, Y., 1975, Applications of wideband, time-domain EM
measurements in mineral exploration: Ph.D. thesis, Univ. of Toronto.
Lee, T., 1979, Transient electromagnetic waves applied to prospecting.
CONCLUSIONS Proc. IEEE, 67, 1016–1021.
——— 1982, Asymptotic expansions for transient electromagnetic
The moments of the quadrature impulse response are a set of fields: Geophysics, 46, 38–46.
quantities that can be used for the interpretation of TEM data. Lee, T. J., and Ignetik, R., 1994, Transient electromagnetic response of
a halfspace with exponential conductivity profile and its application
The zeroth- and first-order terms are related to the inductive- to salinity mapping: Expl. Geophys., 25, 39–51.
limit and resistive-limit responses, which have historically been Lee, T., and Lewis, R., 1974, The transient EM response of a large loop
used for interpretation of frequency-domain data and more on a layered ground: Geophy. Prosp. 22, 430–444.
Leggatt, P. B., Klinkert, P. S., and Hage, T. B., 2000, The Spectrem air-
recently time-domain data. The higher order moments put borne electromagnetic system—Further developments: Geophysics,
greater emphasis on the late-time data. This can bring out con- 65, 1976–1982.
ductive features that are deeper than other less conductive Longman, I. M., 1968, On the numerical inversion of the Laplace
transform of a discontinuous original: J. Inst. Math. Its Appl., 4,
features, and normally only seen in the late-time data. 320–328.
The higher order moments are only weakly affected by any Longman, I. M., and Sharir, M., 1971, Laplace transform inversion of
rational functions: Geophys. J. Roy. Astr. Soc., 25, 99–305.
departure in the impulse response from the ideal impulse. Macnae, J., and Lamontagne, Y., 1987, Imaging quasi-layered conduc-
Hence, it is not necessary to deconvolve the data to get the in- tive structures by simple processing of transient electromagnetic
ductive limit, the resistive limit, and the higher order moments. data: Geophysics, 52, 545–554.
Macnae, J., King, A., Stolz, N., and Klinkert, P., 1999, 3-D EM inversion
Good approximations to all these quantities can be calculated to the limit: in Oristaglio, M. and Spies, B., Eds., Three-dimensional
directly using equation (33). electromagnetics: Soc. Expl. Geophys., 7, 489–501.
Moments of the Impulse Response 1103

Macnae, J., King, A., Stolz, N., Osmakoff, A., and Blaha, A., 1998, Fast quantitative modelling, estimating bird position and detecting
AEM data processing and inversion: Expl. Geophys., 29, no. 1/2, perfect conductors: Geophys. Prosp., 49, 405–416.
163–169. Smith, R. S., and Balch, S. J., 2000, Robust estimation of the band
Morrison, H. F., Phillips, R. J., and O’Brien, D. P., 1969, Quantitative limited inductive-limit response from impulse-response TEM mea-
interpretation of transient electromagnetic fields over a layered surements taken during the transmitter switch-off and the transmit-
half-space: Geophys. Prosp., 17, 82–101. ter off-time: Theory and an example from Voisey’s Bay, Labrador,
O’Brien, D. M., and Smith, R. S., 1985, Transient electromagnetic re- Canada: Geophysics, 65, 476–481.
sponse of a layered conducting medium at asymptotically late times: Smith, R. S., and West, G. F., 1988, Inductive interaction between po-
J. Austral. Math. Soc. Ser. B, 27, 1–30. larizable conductors—An explanation of a negative coincident-loop
Ontario Geological Survey, 2000, Airborne magnetic and electromag- transient electromagnetic response: Geophysics, 53, 677–690.
netic surveys, Reid-Mahaffy airborne geophysical test site survey: Sneddon, I. H., 1972, The use of integral transforms: McGraw-Hill.
Ont. Geol. Surv., Misc. Release-Data (MRD)-55. Stolz, E., and Macnae, J., 1997, Fast approximate inversion of TEM
Papoulis, A., 1962, The Fourier integral and its applications: McGraw- data: Expl. Geophys., 28, 317–322.
Hill. ——— 1998, Evaluating EM waveforms by singular-value decomposi-
Pressens, R., 1971a, On a numerical method for the calculation of tion of exponential basis functions : Geophysics, 63, 64–74 (errata in
transient responses: J. Franklin Inst., 292, 57–64. Geophysics, 64, 310).
——— 1971b, Gaussian quadrature formulas for the numerical inte- Wait, J. R., 1982, Geoelectromagnetism: Academic Press.
gration of Bromwich’s integral and the inversion of the Laplace Watson, G. N., 1944, A treatise on the theory of Bessel function:
transform: J. Eng. Math. 5, 2–9. Cambridge University Press.
Smith, R. S., 2000, The realizable resistive limit: A new concept for map- West, G. F., Macnae, J. C., and Lamontagne, Y., 1984, A time-domain
ping geological features spanning a broad range of conductances: EM system measuring the step response of the ground: Geophysics,
Geophysics, 65, 1124–1127. 49, 1010–1026.
——— 2001, On removing the primary field from fixed-wing time- Widder, D. V., 1941, The Laplace transform: Princeton University
domain airborne electromagnetic data: Some consequences for Press.

Вам также может понравиться