0% found this document useful (0 votes)
201 views9 pages

Sutarno 1991 - Phase Smoothing

phase-smoothed robust M-estimation of magnetotelluric impedance function, D sutarno and K.Vozoff

Uploaded by

RandiRusdiana
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
201 views9 pages

Sutarno 1991 - Phase Smoothing

phase-smoothed robust M-estimation of magnetotelluric impedance function, D sutarno and K.Vozoff

Uploaded by

RandiRusdiana
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

GEOPHYSICS, VOL. 56, NO. 12 (DECEMBER 1991); P. 1999--2007,4FIGS.

Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Phase-smoothed robust M-estimation of magnetotelluric


impedance functions

D. Sutarno* and K. Vozoff*

zontal magnetic field components H x and H y (input) are


ABSTRACT related deterministically to the horizontal electric field com-
ponents E, (output).
Accurate estimation of impedance functions is es- A variety of methods have been proposed for the numer-
sential for the correct interpretation of magnetotelluric ical computation of impedance functions and their associ-
(MT) measurements. Noise is inevitably encountered ated errors in the presence of noise. Most ofthese are based
when MT observations are conducted and, conse- on classical least squares (LS) regression. It is well known
quently, impedance estimates are usually based on that the concept, as well as the computation, of LS estimates
least-squares (LS) regression. Least squares ulti- is quite simple. Furthermore, with the Gaussian error as-
mately assumes simple Gaussian statistics. However, sumption, LS procedures are statistically optimal in a pre-
estimation procedures based on LS would not be cise sense. Unfortunately, the presence of outliers (abnor-
statistically optimal, as outliers (abnormal data) are mal data) superimposed on a common Gaussian noise
frequently superimposed on a normal ambient MT background which constitutes the normal ambient noise
noise field which is approximately Gaussian. In this field, often makes such assumptions about the error unten-
situation, the estimation can be seriously misleading. able and results in a distortion of the estimates. One appeal-
An alternative method for making unbiased robust ing approach to dealing with outliers is to make the estima-
estimates of MT impedance functions is based on tion procedure robust, i.e. to modify it in such a manner that
regression M-estimation and the Hilbert Transform, it is resistant to the effects of the outliers.
operating on minimum-phase MT impedance func- In the geophysical context, robust procedures were intro-
tions. In the resulting regression estimates, outlier duced by Claerbout and Muir (1973). Egbert and Booker
contamination is removed and other departures from (1986) and Chave et al. (1987) proposed the application of a
Gauss-Markov optimality are not critical. Using MT robust procedure which can be classified as an M-estimator
data from the Columbia River Plateau and the to geomagnetic data. More recently, Jones et al. (1989),
EMSLAB Lincoln line, it is shown that the method Chave and Thomson (1989), and Sutarno and Vozoff (1989)
can produce usable MT impedance functions even also implemented this procedure with MT data. This paper
under conditions of severe noise contamination and in aims at improving the performance of the robust procedure.
the absence of remote reference data. Our object is to obtain unbiased robust impedance estimates
in the absence of remote reference (RR) data.

INTRODUCTION ROBUST IMPEDANCE ESTIMATION

In the frequency domain, the equations governing the In the real situation, equation (1) does not hold exactly due
magnetotelluric (MT) relations between the signal compo- to the presence of noise. Therefore, it is necessary to
nents of the electric and magnetic fields at the earth's surface estimate the impedance functions from many noisy data, and
are given by: the problem becomes statistical. Suppose n independent sets
of evaluations at a given harmonic are available to estimate
i = x or y. (1)
Zij. Then the following equation, which represents a linear
From the viewpoint of linear system theory, the tensor regression problem, can be formed
impedances Zix and Ziy are transfer functions of a dual
input, single output linear system through which the hori- x = V0il + r. (2)

Manuscript received by the Editor December 18, 1990; r~vise~ ma~uscript received May 28, 1991. .
*Centre for Geophysical Exploration Research, Macquane University, New South Wales, 2109, Austraha.
© 1991 Society of Exploration Geophysicists. All rights reserved.

1999
2000 Sutarno and Vozoff
The vector x consists of n observations of the ith horizontal
electric field component, the n x 2 matrix V gives n values kt 2 ItI S to;
pet) = I 2 (6)
of the dual horizontal magnetic field components, the ele- { toltl-"2to Itl~to,
ments of !JA are impedance tensor elements (the unknown
parameters to be estimated), and the elements of rare with tuning constant to = 1.5. This hybrid form, which is
unknown errors. The above equation assumes that noise is based on a density function with a Gaussian center and
restricted to an output, while the input is assumed to be Laplacian tails, results in robustness of the M-estimator for
Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

noise free. Therefore this assumption may be approximately data with a "least informative" distribution (Huber, 1981).
valid if the magnetic field (input) may be measured much Note that the tuning constant to together with the scale
more accurately than the electric field (output). Except when parameter s determine the transition point, above which the
the magnetic channel is contaminated by artificial noise, the residuals are considered to be large. For p(t) given by (6) the
assumption is normally used in practice. All of the quantities corresponding Huber weight function is
in (2) are complex. However, it will be simpler to consider
the real case first, from which the complex analog can be I ItI S to;
deduced. Wet) = (7)
{
The M-estimator (or Maximum likelihood type estimator tolltl ItI ~ to·
of Huber, 1981) for !JA in (2) is the value of !JA which The scale parameter s must also be estimated robustly, for
minimizes which the median absolute deviation (MAD),

med {Iri - med (rdl}


s=-------- (8)
U'MAD

is one suitable choice. In the above equation, U'MAD is the


where pet) is a general function, called the loss function, and
expected value of the MAD for the appropriate probability
S is a scale parameter. Equivalently, the estimates may be
density function (PDF).
defined as the solutions of equations For the MT problem, in which the data are complex rather
than real [the vectors and matrices in equation (2) are
n
complex], the determination of the robust scale as well as
L l\I(rils)uij = 0, j = 1, 2, (4) iterative reweighting may be performed either in a real or a
i= I
complex framework. However, the use of a complex frame-
where 1\I(t) = dp(t)ldt, is called the influence function. work, within which the scale and weights are determined
Alternatively, by defining the weight function to be Wet) = based on the magnitude of the residuals, is preferable
1\I(t)lt, the above equations become because it is rotationally (i.e., phase) invariant (Zeger, 1985).
In the complex framework, the normal equation (5) becomes
(5)
U*WU!JA
- -- = U*Wx
- -' (9)

where W is a diagonal matrix. This alternative form of


where * denotes the Hermitian conjugate.
M-estimator might more properly be called a W-e stimator,
The convexity of the Huber p function makes the iterative
but M-estimators are more familiar and the difference is
M-estimators less sensitive to the starting value, for which
minor. The W comes from "weighted," since equation (5)
the LS estimator is still a possible choice. Furthermore, this
can be viewed as the normal equation for a weighted
characteristic also ensures that convergence to a unique
least-squares (WLS) regression problem. However, unlike
estimate is guaranteed in the iterative WLS solution. How-
the ordinary WLS regression, the weights Wi are neither
ever, the Huber weights fall off slowly for large residuals and
equal nor U-determined: they depend on the normalized
provide inadequate protection against severe residuals.
residual (r ils).
Therefore, more severe types of weight function such as the
In general, equation (5) is not linear unless Wet) = 1,
one proposed by Thomson (1977), need to be applied for a
which corresponds to the LS estimator. Thus iterative
few iterations after convergence with the Huber weights has
methods are necessary. For this, an initial solution is chosen
been achieved. The Thomson weights have the form
as the starting value, and both the residuals and a scale
estimate are computed. The weights are calculated from Wet) = exp {_e"(lt!- Ill}, (10)
these, and the solution to equation (5) is found. This proce-
dure is repeated using the residuals and the scale estimate where ex is a parameter that determines the scale at which
from the previous iteration at each stage until convergence is downweighting begins, and is analogous to to in equation (7).
achieved. Because the weights are data-adaptive, the influ- Empirically, it was found that the nth quantile of the
ence of data corresponding to large residuals will be mini- appropriate probability distribution is a suitable choice for ex
mized. (Chave et al., 1987). It should be noticed that, if outliers have
There are a great number of possible forms for the loss been eliminated, the magnitude of residuals are the square
function p(t) and thus of the W(t) to be used in the root of the sum of the squares of the two almost normally
regression M-estimation. The most widely used is a "stan- distributed variates, and hence are Rayleigh-distributed with
dard hybrid" (Huber, 1981), pdf (Johnson and Kotz, 1970)
M-Estimation of MT Impedance Functions 2001

f(t) == t(exp (-t 2/2)) (t ~ 0). (11) utilized by Boehl et al. (1977), stems from the fact that only
the phase function can be estimated consistently using the
The MAD for the above PDF is 0.44845, whereas its quan- standard (4-channel) analysis techniques (Vozoff, 1972),
tiles are given by (Chave et al., 1987) whereas estimates for the amplitude are biased.
The new robust procedure proposed here, incorporates
a. =-V. J2 log ( n - I
.n
+ 0.5
) i == 1, ... , n. (12) phase smoothing in the regression M -estimation, The proce-
dure is essentially an iteratively reweighted least-squares
Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

approach, which is similar to the one proposed by Sutarno


PHASE SMOOTHED ROBUST IMPEDANCE ESTIMATION and Vozoff (1989). Likewise, the procedure uses a complex
The superiority of the robust impedance estimation pro- framework. The main steps of the procedure are as follows:
cedures over the conventional procedures (which are based (a) Use the LS estimator for the preliminary estimator
oo(m)(m = 0).
on LS) was demonstrated by Jones et al. (1989), Chave and
Thomson (1989), and Sutarno and Vozoff (1989). Using (b) Compute the predicted outputs
synthetic and real data, Sutarno and Vozoff (1989) showed i~m) == u ..p'~m) i==I, ••• ,n
I lJ t-'J '
that the robust procedures always yield impedance estimates
which are better than the conventional estimates. However,
and residuals
since the underlying procedures are essentially based on the
sort of model assumption that is intrinsic to the transfer r~m) = x. - i~m) i==l, ••• ,n.
I I I'
function approach (that noise is restricted to the output), the
procedures still need to employ the remote reference (RR) (c) Use the magnitude of residuals to obtain the scale
technique of Gamble et al. (1979) in order to minimize the estimator (8).
bias effect caused by noise in the input. Unfortunately, RR is (d) The Huber weights (7) are computed using the scaled
more costly than single site observations, requiring more magnitude of residuals and applied to matrix W of the
field equipment and personnel. Biasing can still occur if there weighted normal equation (9).
is insufficient separation between the measuring and refer- (e) The weighted normal equation (9) is solved and the
ence sites or if there are effects of possible nonuniform fields entire process is repeated until convergence is
that are coherent between the measuring and reference sites. achieved. The iteration formula is given by
Therefore, it remains a very desirable goal to obtain unbi-
ased, robust impedance estimates using only one set of
equipment.
It is well known that the real and imaginary parts of a (f) Use the Thomson weights [equation (10)] and a fixed
causal transfer function such as the MT impedance function scale estimate derived from the above final iteration,
are not independent bur are related by the Hilbert transform again solving the weighted normal equation (9) and
(HT). Furthermore, with a minimum-phase assumption, the terminating when convergence is achieved.
relationship at radian frequency Wo may be written (Boehl et (g) Apply the Hilbert transform operation to compute the
al., 1977): amplitude of the impedance from the phase and obtain

In IZ(wo)1 == -
1 foo cI>(w)
- - dw (13a)
the phase-smoothed estimate of the impedance at each
frequency.
11' -00 Wo - w (h) Replace the original estimate by the phase smoothed
estimate and perform one-step iteration toward the
and
Thomson estimator.

<I>(wo) == - -
1 foo In IZ(w)1 dw. (13b)
U) Repeat step (g) in order to find the final phase-
smoothed estimate of the impedance.
11' -00 Wo - w
Note that this robust regression procedure can be used with
Boehl et al. (1977) and Fischer and Schnegg (1980) presented the standard (4-channel) or the RR (6-channel) technique. To
methods for calculating the amplitude of the impedance and employ the RR technique, elements of matrix U* in equation
thus the apparent resistivity from the phase or vice versa, for (14) should be replaced by the remote referen& field values.
,..-1

a one dimensional response or a two dimensional magneto-


FIELD EXAMPLES
telluric response rotated to its principal axes. Recently, Yee
and Paulson (1988) derived such a relation from the Cauer MT data from the Columbia River Plateau
integral representation for the impedance tensor, which is
constructed based on the imposition of passivity and sym- This MT data set was taken in an extremely noisy area in
metry requirements on the impedance tensor. In this way, the Columbia River Plateau, by Phoenix Geophysics Inc.,
they suggested that the dispersion relations are shown to be using a remote reference MT system. The data are very
valid for the general three dimensional (3-D) earth system. highly contaminated with common noise at the two loca-
As pointed out, this is because the impedance function can tions. It was indicated that a hydroelectric dam and alumi-
be associated with a minimum-phase system. It is then num refinery which exist in this area were responsible for the
possible to find a consistent estimate of the MT impedance noise (Sutarno and Vozoff, 1989). The impedance function
function from the phase. The basic idea behind such a example was computed for site 101 using site 102 horizontal
procedure, namely "phase smoothing," which was first magnetic field as a reference. The distance between the two
2002 Sutarno and Vozoff

sites is about 3 miles. (No site map was available). Prelimi- ods. In addition, a diagnostic procedure, namely, the "nor-
nary processing of the data was conducted using standard mal scores test" (Hettmansperger, 1984), was done in order
MT analysis. The mean and linear trends were subtracted 10 examine the error distribution resulting from the prelimi-
from each segment of 32 points. The segments were multi- nary fit. For this purpose a real framework was used, so that
plied by a Hanning window and the 6th and 8th discrete the residuals may be regarded as having independent Gaus-
Fourier transform (DFT) coefficients were calculated for sian real and imaginary parts. From the diagnostic test, it is
each segment. To calculate the Fourier coefficients for the found that for frequencies higher than 0.035 Hz, most of
Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

output frequencies of the low bands in the low range, the residuals deviate from normal behavior. Figure 1 shows
cascade decimation procedure of Wight and Bostick (1986) normal scores plots of Ex and E; residuals for two frequency
was used. The impedance and thus the apparent resistivity values, 0.281 and 0.375 Hz. All the plots show a typical
and phase values were then computed using the new robust long-tailed behavior caused by outliers. The apparent resis-
method as well as the conventional standard and RR meth- tivity computed by the new method, along with the results

F-O.28130 Hz N-16Z0 F"O.Z8'30 Hz N"1SZO


~

.~
•••
• .".
<P
*
)

f !
e 0
::)
-e
O.
,J
t ~
"0
I)
,.J
. r r
fIJ lfl
CI
"
~

-
w
-0

•• •I
~'"
-0

v
)
0'
~o.
• ·0

-2.0 -0.1 01 2.0


Normal Scorn Normal Scores

r·o,37500 Hz f a O. 37500 Hz N-1620

f

o •
~
"0
J
W
,. -0

~O
r -0

-0
I
r

if
-t.o 07 2.0 3.4 ·3 .... -'l.O -0 r 01 ~."
Normal Scores Normal Scores

FIG. 1. Normal scores plots of Ex (left) and E y (right) residuals for MT data from the Columbia River Plateau site
WI, for frequencies 0.281 Hz (top) and lU75 Hz (bottom). Number of samples, N == 1620.
M-Estimation of MT Impedance Functions 2003
Ilf . . , , - - - - - - - - - - - - - - - - - - - - - - - - - - - - ,

--~-- RMT. REF. (a)


- - -)( - - STO. M::THOD
--..- STO. ROBUST
---e-- RR ROBUST
lOS
Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

I~

~
E
.: \
~

...
~
,
.,.. ........ -- ... .(
\
)('
>C

.,.. .....
~ 101 '"

Ie!'

10-1 -t---;r--1-rTTTTTr--r-T"'T.........TTT1-...,-....-rMrTT1.----.-r"TTTTT. . .--,~""T'1rTT1rr1

10-5 10- 1 Ie!' 101 I~


FREQUENCY (Hertz)

Ilf ~--------------------------.

- . + - - RMT. REF. (b)


--* -- STO. M::THOO
----...- STO. ROBUST
--e-- RR ROBUST
lOS

i 101

Ie!'

10- 1 ;---;r--1-rTTTTTT--r-T"'T.........TTT1-...,-....-rMrTT1.----.-,-"TTTTT. . .--.~""T'1l"'TTrrI

ur5 10- 1 Ie!'


FREQUENCY (Herlz)

FIG. 2. (a) A comparison of results for Pxy on MT data from the Columbia River Plateau site 101, by four different
analyses: standard, RR, robust on four-channel data (Std. robust), and robust on six-channel data (RR robust). The
two robust analyses provide very smooth and consistent results. (b) As Figure 2a, but for Pvx' In addition to
smoothness, the two robust results are much more consistent than the other pair. -
2004 Sutarno and Vozoff

from the conventional standard and RR methods, are shown for two frequencies, 0.0012 and 0.0016 Hz. Here also, the
in Figures 2a and 2b, and the corresponding phases can be plots show a typical long-tailed behavior. The apparent
seen in Figures 2c and 2d. Based on smoothness, slope, and resistivity computed using the new method, along with the
compatibility of Pa and <I>, it can be seen from the figures that results from the standard and RR methods are depicted in
the conventional analyses yield impedance estimates that are Figures 4a and 4b, whereas their corresponding phases are
very poor in these qualities. The new method markedly shown in Figures 4c and 4d. As can be seen, the figures again
improves the impedance estimates. Furthermore, the proce- show the superiority of the new method over conventional
Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

dure also provides more consistent impedance estimates. As methods.


shown, application of the procedure to the 4-channel data
gives almost the same results as applied to the 6-channel
(RR) data. This is a very powerful test of its effectiveness. CONCLUSION

MT data from the EMSLAB Lincoln line Regression M -estimation which is robust to violations of
the usual Gaussian error assumption has been applied in the
This field example was recorded at one of the EMSLAB
processing of MT data to estimate the impedance tensors.
Lincoln lines, by the Geological Survey of Canada. The
impedance function example was computed for the "quiet The outlier rejection facility provided by the procedures
period" of the MT record of the Lincoln line site 1, using site improves the impedance estimates. When the impedance
13 (135 km distant) horizontal magnetic field as a reference. function is assumed to be a minimum-phase system, it is then
As reported (Jones et al., 1989), the MT data in this period possible to compute the impedance amplitude from the
are badly contaminated. A telluric noise spike of about 200 phase by applying the phase smoothing process, which is
mVIkm and a polar substorm event dominate the data. As based on the Hilbert transform. Since the phase function can
before, except for interpolating across a few short data gaps, be estimated consistently from the noisy MT data, incorpo-
no preliminary editing of data was deemed necessary on the ration of the process in the regression M-estimation solves
basis of visual inspection, and the same process is conducted the bias problem and results in robust, consistent impedance
on the data. Figure 3 shows normal scores plots of residuals estimates.

135..,.....---------------------------.
---+-- RMT. REF. ... Cc)
- - -)( - - STD. M::THOD I
,
- 4 - STD. ROBUST
- - G - RR ROBUST
I \
I \
•••
80
~
... \
.•.. ~
.....
l 45 ,I
'\ I
.
\ ./
O+---,r-r-rrTTTTr-r-r-rTT1rTTr-"-"T""'1r-T'1"TTT...--,--rTTTTT..---.r-r-nrTTm
Icr l 10'1 uP Idl
FREQUENCY (Hzl

-45..,.....-------------------...----------:----:---.
....
--0+-- RMT. REF.
I \
Cd)
--* -- STD. M::THOD
-----...- STD. ROBUST ,I \
- 4 - RR ROBUST
I 'J \
••• -80 \
;;. \
.•.. \
....
i ·1

I
, /

-/
"-,I
-llIO-l--r"-rT1rrnr--'--'-"T'TTTT,.,--,r-r-rrTTnr-r-T""TTTrTTT--r-.,-",rrn1
10. 1 IcP
FREQUENCY (Hz I

FIG. 2. (c) (d) Phases for Figures 2a and 2b. The two robust results are very smooth and consistent.
M-Estimation of MT Impedance Functions 2005

F-O.OOI17 Hz N-168 F-O.OOI17 Hz N-I68


Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

15.


• •
•• •

a .,..• a
~ • ~
'0
,/ '0

.
••
-I.
.,# ..•• -2.
M
LIJ
•• >.
LIJ

••
-5. • -7.

-5.

-Z.' -1.7 -0.' 0.' 1.7 2.' -Z.' -1.7 -0.' 0.' 1.7 2.'
Normal Scores Normal Scores

F-O.OOI56 Hz N-168 F-O.OOI56 Hz N-168

••

a
~
'0
,.... • a
~
'0
7#'0·
0 •
.
••
M
-0.
..••
>.
-I.

......

LIJ LIJ •
-I. -5.

-Z. -t. ' I I


'i
-Z.' -1.7 -0.' 0.1 1.7 Z.• -Z.' -1.7 -0.' 0.' 1.7 2.'
Normal Scores NormaI Scores

FIG. 3. Normal scores plots of Ex (left) and E)' (right) residuals for the quiet period of the MT data from the
EMSLAB Lincoln line site 1, for frequencies 0.0012 Hz (top) and 0.0016 Hz (bottom). Number of samples, N =
168.
2006 Sutarno and Vozoff

Iff
--+-- RMT. REF. (0 )

,
~ ----
- - .. - - STD. t.£THOD
STD. ROBUST
- - 4 - RR ROBUST

\
Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

10' \
\\

r ~
i:
I:
e
~
..... , ,
,\

\ J \
~ .... .........
Irl
~. 4 > ~t
'\ ..-
I :".'~
,,:'\
~ II \

,
.:' \
\\ II <,
II '\
\\ J \
.rl \
\ <,
101 "
10"" 10"' 10'2 10"1
FREQUENCY (He,' z I

Iff
"-\i --+-- RMT. REF. (b)

\ STD. t.£THOD
- - -)t - -
______ STD. ROBUST
~ RR ROBUST
\\
\
10'
\
\

I' ,
r
i
e
.....
, I
, '\

i
Irl "- <,
--"

10' +-----.--r--r-!'"TTTrr--.......,.--r--r-!'"TTTrr--.......,.--r--r-r-rTTr!
10"" 10"' 10"1
FREQUENCY (He,' z I

FIG. 4. (a) A comparison of results for Px on the quiet period of the MT data from the EMSLAB Lincoln line site
1, by four different analyses: standard, RR, robust on four-channel data (Std. robust), and robust on six-channel
data (RR robust). Here also, the two robust results are very smooth and consistent. (b) As Figure 4a, but for Pyx'
In addition to smoothness, again the two robust results are very consistent.
~Estimation of MT Impedance Functions 2007
13II...,----""""""T------------------------,
--..,-_____ . REF. (c)
--->t"- - sro. t.£THOO
--JI.- slq. ROBUST
, RR\~OBUST

...•••..
• 10
\\
\
.....
Downloaded 10/13/14 to 129.2.29.132. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

~~~

0
10-4 10"5 10.2 10"1
FREOlEHCY (Hz)

·45
--~ •• RMT. REF. (d)
- - -)( - - STD. t.£THOO
- . A - - STD. ROBUST

---lD-- RR ROBUST

·..••
~
-10

~

.....
if -I
,.--- <,
00--

-I

10-4 10. 5 10.2 10'1


FREQUENCY (Hz)

FIG. 4. (c) (d) Phases for Figures 4a and 4b. Again, the two robust results are very smooth and consistent.

ACKNOWLEDGMENTS Egbert, G. D., and Booker, J. R., 1986,Robust estimation of geomag-


netic transfer functions: Geophys. J. Roy. Astr. Soc., 87, 175-194.
Fischer, G., and Schnegg, P.·A., 1980, The dispersion relations of
We thank Bob Anderson, Alan Jones, Phoenix Geophys- the magnetotelluric response and their incidence on the inversion
ics Inc., and the Geological Survey of Canada for provision problem: Geophys. J. Roy. Astr. Soc., 62, 661-673.
of the data and permission to release it. Don McNeil and Gamble, T. D., Goubau, W. M., and Clarke, J., 1979, Magnetotel-
luries with a remote reference: Geophysics, 44, 53-68.
Dave Kleinbaum of Macquarie University and Ken Paulsen Hettmansperger, T. P., 1984, Statistical inference based on ranks:
of the University of Saskatchewan helped to improve the John Wiley and Sons, Inc.
first version of the paper. EMSLAB was supported by the Huber, P. J., 1981, Robust statistics: John Wiley and Sons, Inc.
Johnson, N. L., and Kotz, S., 1970, Continuous univariate distri-
US National Science Foundation and the Geological Survey bution -I: Houghton Mifflin Co.
of Canada. Jones, A. G., Chave, A. D., Egbert, G. D., Auld, D., and Bahr, K.,
1989, A comparison of techniques for magnetotelluric response
estimationrJ. Geophys. Res., 94, 14201-14213.
Sutarno, D., and Vozoff, K., 1989, Robust M-estimation of Magne-
totelluric impedance tensors: Expl. Geophys., 20, 383-398.
REFERENCES Thomson, D. J., 1977, Spectrum estimation techniques for charac-
terization and development of WT4 waveguide 1: Bell Syst. Tech.
J., 56,1769-1815.
Boehl, J. E., Bostick, F. X., and Smith, H. W., 1977,An application Vozoff, K., 1972, The magnetotelluric method in the exploration of
of the Hilbert transform to the magnetotelluric method: Tech. sedimentary basins: Geophysics, 37, 98-141.
Rep. Electr. Geophys. Res. Lab., University of Texas, Austin. Wight, D. E., and Bostick, F. X., 1986, Cascade decimation-A
Chave, A. D., Thomson, D. J., and Ander, M. E., 1987, On robust technique for real time estimation of power spectra, in Vozoff, K.,
estimation of power spectra, coherences, and transfer functions: ed., Magnetotelluric methods: Geophysics Reprint Series no. 5:
J. Geophys. Res., 92, 633-648. Soc. Expl. Geophys., 215-218.
Chave, A. D., and Thomson, D. J., 1989, Some comments on Yee, E., and Paulson, K. V., 1988, Concerning dispersion relations
magnetotelluric response function estimation: J. Geophys. Res., for the magnetotelluric impedance tensor: Geophys. J., 95, 549-
94, 14215-14225. 559.
Claerbout, J. F., and Muir, F., 1973, Robust modeling with erratic Zeger, S. L., Exploring an ozone spatial time series in the frequency
data: Geophysics, 38, 826-844. domain: 1. Amer. Stat. Assoc., 80, 323-331.

You might also like