GEOPHYSICS, VOL. 68, NO. 6 (NOVEMBER-DECEMBER 2003); P. 1962–1968, 3 FIGS., 1 TABLE.

10.1190/1.1635050

3D Euler deconvolution: Theoretical basis for automatically

selecting good solutions

João B. C. Silva∗ and Valéria C. F. Barbosa‡

by volume and strive to determine both. The peculiarity of

ABSTRACT magnetic interpretation is that magnetization is not a scalar
but a vector property which is related to surface rather than to
We derive the analytical estimators for the horizontal
volume distributions of magnetic poles (a magnetic pole is an
and vertical source positions in 3D Euler deconvolution
abstraction and is invoked here for its simplicity in explaining
as a function of the x-, y-, and z-derivatives of the mag-
magnetic phenomena). As a result, the fundamental ambigu-
netic anomaly within a data window. From these expres-
ity in magnetic interpretation is more complex than in gravity
sions we show that, in the case of noise-corrupted data,
and MT interpretations because not only the magnetization in-
the x-, y-, and z-coordinate estimates computed at the
tensity but also magnetization inclination and declination may
anomaly borders are biased toward the respective hori-
couple with parameters defining the source volume. For exam-
zontal coordinate of the data window center regardless
ple, for the same magnetization intensity, a wide, thin tabular
of the true or presumed structural indices and regard-
source may produce large or small magnetic anomalies above
less of the magnetization inclination and declination. On
its center, depending on whether the magnetization inclination
the other hand, in the central part of the anomaly, the
is, respectively, high or low.
x- and y-coordinate estimates are very close to the re-
This scenario is further complicated when the magnetic
spective source horizontal coordinates regardless of the
anomalies are not isolated, as is the case in aeromagnetic in-
true or presumed structural indices and regardless of the
terpretations. In this situation, the anomalies rarely fall off to
magnetization inclination and declination. This contrast-
zero, so their base level is unknown. Automatic determination
ing behavior of the horizontal coordinate estimates may
of the base level couples it with the source position and with the
be used to automatically delineate the region associated
magnetization and source geometry parameters, worsening the
with the best solutions. Applying the Euler deconvolu-
interpretation ambiguity. For this reason, magnetic interpreta-
tion operator inside this region would decrease the dis-
tion methods which require no magnetization vector estima-
persion of all position estimates, improving source loca-
tion are highly promising in automatic aeromagnetic interpre-
tion precision.
tations and have been widely investigated and applied. They
include analytic signal methods and their extensions (e.g., Hsu
et al., 1996; Debeglia and Corpel, 1997; Thurston and Smith,
INTRODUCTION
1997; Hsu et al., 1998; Bastani and Pedersen, 2001; Zhang, 2001;
Thurston et al., 2002) and the Euler deconvolution method and
Any geophysical interpretation is subject to a fundamental its extensions (e.g., Roest and Pilkington, 1993; Peirce et al.,
ambiguity involving the product of a physical property and a 1998; Mushayandebvu et al., 2001).
volume. Small sources with high-intensity physical properties A drawback of Euler deconvolution is the scattering of the
and large sources with low-intensity physical properties tend to solutions estimated at different data window positions. Fair-
produce approximately the same geophysical responses. Clas- head et al. (1994) suggest using only the windows located
sic examples are the product of conductivity multiplied by around the maximum of the horizontal gradient of the reduced-
thickness in 1D magnetotelluric inversion and density multi- to-the-pole anomaly. Silva et al. (2001) show that the solution
plied by volume in 3D gravity interpretation. In the case of scattering associated with an isolated anomaly has two causes:
magnetic data, the same ambiguity is present. Magnetic data the choice of a wrong structural index and the presence of
are sensitive to the dipole moment, but geophysicists express noise in data. While the solution dispersion caused by choos-
the dipole moment as the product of magnetization multiplied ing the wrong structural index can be eliminated by estimating

Manuscript received by the Editor June 18, 2002; revised manuscript received May 7, 2003.
∗
Federal University of Pará, Department of Geophysics, CG, Caixa Postal 1611, CEP 66017-900 Belém, Pará, Brazil. E-mail: joaobcs@
supridad.com.br.
‡LNCC, Av. Getúlio Vargas, 333, Quitandinha, Petrópolis, Rio de Janeiro 25651-075, Brazil. E-mail: [email protected].
°c 2003 Society of Exploration Geophysicists. All rights reserved.
1962
 Selecting the Best 3D Euler Solutions 1963

the structural index via Barbosa et al.’s (1999) method, the re- where p is the vector of unknown parameters:
duction of solution scattering caused by noise in data can only  
be achieved by applying the Euler deconvolution operator in
x0
 
data windows where the amplitudes of the total-field anomaly p =  y0  , (4)
and its derivatives are large. z0
Our paper presents the theoretical basis for constructing an
¿ À ¿ À ¿ À
automated algorithm that detects the region of the anomaly
∂h ∂h ∂h ∂h ∂h ∂h
 
 ∂x , ∂x , ,
producing consistent estimates of the source position. First, the
¿ ∂x ∂y ∂ x ∂z 
À
estimators of the horizontal and vertical source positions, x̂ o ,
 À ¿ À ¿ 
ŷo , and ẑ o , are derived analytically as a function of the x-, y-, and  ∂h ∂h ∂h ∂h ∂h ∂h 
z-derivatives of the magnetic anomaly within a data window. S S=
T
, , , , (5)
 ∂x ∂y ∂y ∂y ∂ y ∂z 
From the expressions for x̂o and ŷo , and in the case of noise- ¿ À ¿ À ¿ À
 
corrupted data, we show that these estimators at the anomaly  ∂h ∂h ∂h ∂h ∂h ∂h 
borders are biased toward the respective horizontal coordinate , , ,
∂ x ∂z ∂ y ∂z ∂z ∂z
of the data window. On the other hand, in the central part of
the anomaly, x̂o and ŷo produce consistent estimates very close and
to the correct source coordinates, defining, in this way, a flat
area. ST y =
¿ À ¿ À ¿ À ¿ À
These properties of estimators x̂ o and ŷo as a function of the
∂h ∂h ∂h ∂h ∂h ∂h ∂h
horizontal coordinates of the data window center may be used  ,x + ,y + ,z +η ,h 
 ∂x ∂x ∂x ∂y ∂ x ∂z ∂x 
to delineate the flat area where good and consistent estimates ¿ À ¿ À ¿ À ¿ À
 
for all source positions are produced. This can be done, for  ∂h ∂h ∂h ∂h ∂h ∂h ∂h 
 ,x + ,y + ,z +η ,h  ,
example, by fitting a first-degree polynomial to x̂o (or ŷo ) within  ∂y ∂x ∂y ∂y ∂ y ∂z ∂y 
a moving data window. The window positions where an abrupt ¿ À ¿ À ¿ À ¿ 
À

change in the x- (or y-) coefficient of the fitted polynomial  ∂h ∂h ∂h ∂h ∂h ∂h ∂h 
,x + ,y + ,z +η ,h
occur delineate the x- (or y-) coordinate of the border of the ∂z ∂ x ∂z ∂ y ∂z ∂z ∂z
flat area.
In contrast with Fairhead et al.’s (1994) method, our ap- (6)
proach requires no reduction-to-the-pole operation and ac-
cepts a larger number of reliable solutions. As a result, the and where hu(x, y), v(x, y)i is the discrete inner product of
method is simpler to implement and may produce more pre- u(x, y) and v(x, y), defined by
cise estimates of source location because these are computed X
NX X
NY
as the average of the estimates obtained with all reliable data hu(x, y), v(x, y)i = u[x ∗ + (n x − 1)1x, y ∗
windows. n x =1 n y =1

+ (n y − 1)1y] · v[x ∗ + (n x − 1)1x, y ∗ + (n y − 1)1y],

ANALYTICAL ESTIMATORS FOR HORIZONTAL
AND VERTICAL SOURCE POSITIONS (7)

We start from Euler’s 3D equation: where N X and N Y are the number of points of a data window
along the x- and y-directions, respectively; x ∗ and y ∗ are the
∂ ∂ coordinates of the observation on the lower-left corner in the
(x − xo ) h(x, y, z) + (y − yo ) h(x, y, z) data window; and 1x and 1y are the data spacing along the x-
∂x ∂y
∂ and y-directions, respectively.
+ (z − z o ) h(x, y, z) = −ηh(x, y, z), (1) Estimators x̂o , ŷo , and ẑ o are then given by
∂z
1
x̂o = [(BC − F 2 )(G + H + P + I η)
where h(x, y, z) is the total-field anomaly produced by a point T
or line source at (xo , yo , z o ) and η is the structural index related − (DC − E F)(J + K + Q + Lη)
to the nature or geometry of the source.
By presuming that a value for η is tentatively assumed by
+ (D F − E B)(M + N + R + Oη)], (8)
the interpreter and that the base level is known to be null, 1
ŷo = [−(DC − E F)(G + H + P + I η)
equation (1) can written as T
+ (AC − E 2 )(J + K + Q + Lη)
∂h ∂h ∂h ∂h ∂h ∂h
xo + yo + zo =x +y +z + ηh. (2) − (AF − D E)(M + N + R + Oη)], (9)
∂x ∂y ∂z ∂x ∂y ∂z
and
We obtain estimators for parameters xo , yo , and z o by solving
1
the following least-squares normal equation associated with ẑ o = [(D F − E B)(G + H + P + I η)
equation (2): T
− (AF − D E)(J + K + Q + Lη)
ST Sp = ST y, (3) + (AB − D 2 )(M + N + R + Oη)], (10)
1964 Silva and Barbosa

where terms A–R are defined in Table 1 and where

Table 1. Definitions of terms A–R used in equations (8), (9),
and (10). T = ABC − AF 2 − D 2 C + 2D E F − E 2 B. (11)
Term Definition BIAS AND SYMMETRY OF ESTIMATES x̂o , ŷo , AND ẑ o WITH
* + NOISE-CORRUPTED DATA
∂h ∂h
A , Theory
∂x ∂x
* + Assume that an isolated magnetic anomaly decays to a con-
∂h ∂h stant base level γ at its borders. As a result, at this place, terms
B ,
∂y ∂y A–H, J–K, M–N, and P–R will decay to zero. In the case of an
* + anomaly corrupted by zero-mean random noise realizations,
∂h ∂h we have the following additional properties:
C ,
∂z ∂z
1) terms I, L, and O will also decay to zero because by replac-
* +
∂h ∂h ing h by h + γ these terms assume the form h∂h/∂α, hi +
D , h∂h/∂α, γ i, with α = x, y, or z. The first term decays to
∂x ∂y
zero because h and ∂h/∂α decay to zero and the second
* +
∂h ∂h term also decays to zero because it is proportional to the
E , sum of values oscillating about zero;
∂ x ∂z
2) terms A, B, C, G, K, and R will decay less rapidly to
* + zero than all other terms because they are either sums
∂h ∂h
F , of squared values (A–C) or positively weighted sums of
∂ y ∂z x-, y-, and z-coordinates (G, K, and R), whereas all other
* + terms are either the product of values oscillating about
∂h ∂h
G ,x zero (D, E, F, H, J, M, N, P, and Q) or a quantity consisting
∂x ∂x of a product of values oscillating about zero plus a sum
* + of values oscillating about zero (I, L, and O).
∂h ∂h
H ,y Thus, by retaining only terms A, B, C, G, K, and R in
∂x ∂y
equations (8)–(10), we obtain approximate expressions for the
* +
∂h position parameters at the anomaly borders:
I ,h ¿ À
∂x ∂h ∂h
,x
* + G ∂x ∂x
∂h ∂h x̂o ≈ = ¿ À , (12)
J ,x A ∂h ∂h
∂y ∂x ,
∂x ∂x
* + ¿ À
∂h ∂h ∂h ∂h
K ,y ,y
∂y ∂y K ∂y ∂y
* + ŷo ≈ = ¿ À , (13)
B ∂h ∂h
∂h ,
L ,h ∂y ∂y
∂y
* + and ¿ À
∂h ∂h ∂h ∂h
M ,x ,z
∂z ∂ x R ∂z ∂z
* + ẑ o ≈ = ¿ À. (14)
C ∂h ∂h
∂h ∂h ,
N ,y ∂z ∂z
∂z ∂ y
* + That is, estimates x̂o , ŷo , and ẑ o at the anomaly borders
∂h are, respectively, the weighted averages of the x-, y-, and
O ,h
∂z z-coordinates of the points defining the data window, in-
* + dependent of the assumed structural index evidenced by
∂h ∂h equations (12)–(14).
P ,z
∂ x ∂z In summary, the following approximate bias and symmetry
* + properties hold for estimators x̂o , ŷo , and ẑ o at the anomaly
∂h ∂h borders, regardless of the presumed structural index:
Q ,z
∂ y ∂z 1) estimator x̂o , as a function of the data window position,
* + is biased toward the x-coordinates of the data window
∂h ∂h center, asymmetric along the x-direction with respect to
R ,z
∂z ∂z xo , and symmetric (constant) along the y-direction with
respect to yo ;
 Selecting the Best 3D Euler Solutions 1965

2) estimator ŷo , as a function of the data window position, In an aeromagnetic survey, z o and the average observation al-
is biased toward the y-coordinates of the data window titude Hflight are always substantially different from each other.
center, asymmetric along the y-direction with respect to Accordingly, the flat areas corresponding to z o and Hflight in the
yo and symmetric (constant) along the x-direction with plot of ẑ o against the window center will also be very different.
respect to xo ; As a result both flat areas may be always clearly identified.
3) estimator ẑ o , as a function of the data window position, is
biased toward the z-coordinates of the data window cen-
ter. If the data were collected (or gridded) at an approxi-
mately constant level, it will be symmetric along both the
x- and y-directions with respect to xo and yo , respectively.

These properties are said to be approximate because the

presence of noise on the one hand creates favorable conditions
for their very existence and on the other hand prevents them
from being exact. We stress that the above bias and symmetry
properties at the borders depend neither on the value assigned
to the structural index nor on the magnetization inclination
and declination.

Example
The bias and symmetry in x̂o , ŷo , and ẑ o is illustrated us-
ing a noise-corrupted anomaly produced by a spherical source
with radius of 1 km located at xo = 12 km, yo = 10 km, and
z o = 2 km and magnetized by induction only. The magnetiza-
tion vector has an inclination of 60◦ , declination of 0◦ , and mod-
ulus of 1 A/m. The total-field anomaly was computed on plane
z = 0 km at the nodes of a 120 × 100 grid with a grid spacing
of 0.2 km in both x- and y-directions (north–south and east–
west, respectively) assuming a null base level. The theoreti-
cal anomaly was contaminated with pseudorandom Gaussian
noise with zero mean and standard deviation of 1 nT. In all tests,
the gradients were obtained using the moving window method
based on the equivalent layer principle (Leão and Silva, 1989)
that solves an underdetermined damped least-squares prob-
lem. In applying the Euler deconvolution method, we used
a 15 × 15 moving data window spanning the area limited by
x ∈ [4 km, 20 km] and y ∈ [2 km, 18 km]. Note that the area
where the magnetic anomaly is computed is larger than the
area where the Euler operator is applied. The purpose is to
increase the precision in computing the anomaly gradients by
the equivalent layer method. An estimated parameter vector
is obtained by solving equation (3), and the values of x̂o , ŷo ,
and ẑ o are assigned to the central point of the moving data
window.
Figures 1a–c show the estimates x̂o , ŷo , and ẑ o , respectively,
against the central position of the data window, assigning the
correct value of 3 to η. Figures 1a and 1b show that, at the
anomaly borders, x̂o and ŷo are biased toward the arithmetic
averages of the x- and y-coordinates of the data points defining
the data window. Graphically, we see that the behavior of x̂o
and ŷo can be approximated by a first-order polynomial sur-
face. On the other hand, Figure 1c shows that ẑ o is uniformly
biased toward zero at the borders because we assumed the data
were collected at z = 0 (theoretically, it decays to zero only at
infinity). FIG. 1. Estimates (a) x̂o , (b) ŷo , and (c) ẑ o against the x- and
In contrast, close to the central part of the contour map y-coordinates of a 15 × 15 moving data window center for the
(x ∈ [8 km, 14 km] and y ∈ [7 km, 12 km]) where the highest case of a noise-corrupted anomaly at 60◦ inclination produced
by a sphere with the center at xo = 12 km, yo = 10 km, and
absolute values of the vertical and horizontal gradients take z o = 2 km and magnetized by induction only using the correct
place, x̂o , ŷo , and ẑ o assume nearly constant values approxi- structural index (η = 3). Contour intervals: 1 km for Figures 1a
mately equal to xo , yo , and z o , as expected. and 1b, 0.3 km for Figure 1c.
1966 Silva and Barbosa

In view of the above-described properties, estimator x̂ o will POTENTIAL APPLICATIONS

be approximately asymmetric in the x-direction with respect
We have established some theoretical properties of the
to the coordinate xo = 12 km and symmetric in the y-direction
source position estimates in 3D Euler deconvolution method.
relative to the coordinate yo = 10 km, as seen in Figure 1a. Like-
Although this is only a theoretical analysis, understanding these
wise, ŷo will be approximately asymmetric in the y-direction
with respect to coordinate yo = 10 km and symmetric in the
x-direction relative to the coordinate xo = 12 km, as illustrated
in Figure 1b. Finally, Figure 1c displays the symmetric behavior
of ẑ o along the x- and y-directions with respect to x o = 12 km
and yo = 10 km, respectively.
The behavior of estimates x̂o and ŷo , either at the anomaly
borders or in the central part of the anomaly, does not de-
pend on the tentative value assigned to η. This can be verified
in Figures 2a and 2b, respectively, which show, for the same
source and data layout, the estimates x̂o and ŷo as a function
of the data window center using a wrong structural index of
1. On the other hand, ẑ o is underestimated in the central part
of the anomaly if a tentative structural index smaller than the
correct one is used (Figure 2c) and vice-versa. However, at the
borders, ẑ o falls off to zero even when using a wrong structural
index.
We have shown numerically that by using noise-corrupted
data, the estimates x̂o and ŷo are close to the respective x- and
y-coordinates of the true source in the neighborhood of the
central part of the anomaly and biased toward the arithmetic
averages of the respective x- and y-coordinates of the moving
data window center at the anomaly borders. Our theoretical
analysis indicates that the bias in x̂o and ŷo at the anomaly bor-
ders occurs, only if the data are contaminated by random noise
because in this case all terms in equations (8) and (9) present
very small values, except A, B, C, G, K, and R. In contrast,
for the case of noise-free data, equations (12)–(14) are invalid
because there is no guarantee that terms D, E, F, H, I, J, L, M,
N, P, O, and Q decay more rapidly than terms A, B, C, G, K,
and R. In this case any analysis about the bias of estimates
x̂o , ŷo , and ẑ o at the anomaly borders must be performed via
equations (8), (9), and (10), respectively. To validate this as-
sertion, we present below evidences that the theoretical biases
on estimates of horizontal source coordinates at the anomaly
borders for the cases of noisy and noise-free data are very
different.
Figure 3 shows the estimate x̂o as a function of the data win-
dow center for noise-free data produced by the same spherical
source used previously. In Figures 3a and 3b, the magnetization
inclination is 90◦ and the assumed structural indices are 3 and 1,
respectively; in Figure 3c, the inclination is 60◦ and the assumed
structural index is 3. In Figures 3a and 3b, x̂o is still biased at the
borders but not to the theoretical bias occurring in noisy data,
that is, to the x-coordinate of the moving data window. More-
over, Figure 3c demonstrates that, for 60◦ inclination, the bias
does not show up within the processed area, indicating that the
behavior of x̂o at the anomaly borders is highly dependent on
the magnetization inclination for noise-free data. In addition,
note from Figure 3b that, different from the noise-corrupted
case, the bias in the case of noise-free data is dependent on
the structural index because I , L, and O in equation (8) are FIG. 2. Estimates (a) x̂o , (b) ŷo , and (c) ẑ o against the x- and
nonnegligible. y-coordinates of a 15 × 15 data window center for the case of
Finally, we stress that our tests with synthetic data produced a noise-corrupted anomaly at 60◦ inclination produced by a
the same results qualitatively when applied to other interpre- sphere with the center at xo = 12 km, yo = 10 km, and z o = 2 km
and magnetized by induction only using a wrong structural
tation models (horizontal cylinder and thin dike) and under index (η = 1). Contour intervals: 1 km for Figures 2a and 2b,
different unknown base levels. 0.2 km for Figure 2c.
 Selecting the Best 3D Euler Solutions 1967

properties is the first step in implementing a new method to We trust that such a method might produce more reliable
select only the best source position estimates via a simple cri- results than a method that simply selects as the best solutions
terion. This criterion would consist in delimiting the area con- those related to data windows positioned near the peaks of the
taining the best estimates by mapping the abrupt changes in anomalies for the following reasons. First, it does not require
the gradients of x̂o and ŷo plotted as a function of the moving that the anomaly be reduced to the pole to guarantee a sin-
window center. gle, well-defined peak per anomaly. Further, near and anomaly
are subjective terms. As a result, defining an anomaly algorith-
mically and establishing how far from the anomaly peak the
data window should be centered to produce a good solution
are certainly not trivial tasks. In addition, our tests indicate
that the flat area containing good solutions comprises even
data points where the anomaly amplitude is small. Therefore,
it defines a larger neighborhood around the anomaly peaks
relative to a criterion based just on the anomaly amplitude.
This is an important feature because the final location esti-
mate is the average of all good location estimates. Thus, the
larger the number of good solution; the more precise the final
estimate.

CONCLUSIONS

We have derived analytical estimators for the horizontal and

vertical source positions in 3D Euler deconvolution as a func-
tion of the x-, y-, and z- coordinates of the observations and
have analyzed the behavior of these estimators as a function
of the x- and y-coordinates of the moving data window center
using an assumed value for the structural index. For noise-free
data, our numerical tests confirm the theoretical prediction
that, in the central part of the anomaly, the surfaces repre-
senting the source coordinates’ estimates (x̂ o , ŷo , and ẑ o ) as a
function of the x- and y-coordinates of the moving data win-
dow center present approximately constant values defining a
flat area. These values are very close to the true source coor-
dinates if the assumed structural index is correct. If a wrong
structural index is used, our numerical tests show that constant
values close to the true coordinates still occur for x̂o and ŷo ,
but ẑ o will underestimate the true coordinate if the assumed
structural index is smaller than the correct one and vice-versa.
With noise-corrupted data, these flat areas in the central part
of the anomaly are approximately maintained. On the other
hand, we have shown theoretically that at the anomaly bor-
ders, x̂o becomes biased toward the x-coordinate of the data
window center. As a result, it also becomes symmetric along
the y-direction relative to yo , the true source y-coordinate, and
asymmetric along the x-direction relative to xo , the true source
x-coordinate. Mutatis mutandis, the same is true for ŷo . Esti-
mator ẑ o , on the other hand, is biased toward the z-coordinate
of the observation associated with the horizontal coordinates
of the moving data window center.
The above theoretical properties of 3D Euler estimators are
not just mathematical peculiarities presented just for academic
interest. They may be used to delineate the flat area of estimates
x̂o and ŷo where good estimates for all source positions are pro-
duced. This can be done by fitting a first-degree polynomial to
FIG. 3. Estimate x̂o against the x- and y-coordinates of a 15 × 15 x̂o (or ŷo ) within a moving data window. The window positions
data window center for the case of a noise-free anomaly pro- where an abrupt change in the x- (or y-) coefficient of the fit-
duced by a sphere with the center at xo = 12 km, yo = 10 km, ted polynomial occur delineate the x- (or y-) coordinate of the
and z o = 2 km and magnetized by induction only. (a) Magne- border of the flat area. Because the flat areas are associated
tization with 90◦ inclination and true structural index (η = 3).
(b) Magnetization with 90◦ inclination and a wrong structural with consistent estimates, computing the estimates of x̂ o and
index (η = 1). (c) Magnetization with 60◦ inclination and true ŷo , and ẑ o only within them, will decrease the dispersion of
structural index (η = 3). Contour interval: 1 km. these estimates, improving their precision.
1968 Silva and Barbosa

ACKNOWLEDGMENTS Hsu, S., Sibuet, J., and Shyu, C., 1996, High-resolution detection of
geologic boundaries from potential-field anomalies: An enhanced
We are grateful to Afif Saad, A. B. Reid, P. B. Keating, and analytic signal technique: Geophysics, 61, 373–386.
Leão, J. W. D., and Silva, J. B. C., 1989, Discrete linear transformations
R. O. Hansen for their comments and thoughtful suggestions. of potential field data: Geophysics, 54, 497–507.
The authors were supported in this research by fellowships Mushayandebvu, M. F., van Driel, P., Reid, A. B., and Fairhead, J. D.,
from Conselho Nacional de Desenvolvimento Cientifico e Tec- 2001, Magnetic source parameters of two-dimensional structures us-
ing extended Euler deconvolution: Geophysics, 66, 814–823.
nológico (CNPq), Brazil. Peirce, J. W., Goussev, S. A., Charters, R. A., Abercrombie, H. J., and
DePaoli, G. R., 1998, Intrasedimentary magnetization by vertical
REFERENCES fluid flow and exotic geochemistry: The Leading Edge, 17, 89–92.
Roest, W. R., and Pilkington, M., 1993, Identifying remanent magne-
Barbosa, V. C. F., Silva, J. B. C., and Medeiros, W. E., 1999, Stability tization effect in magnetic data: Geophysics, 58, 653–659.
analysis and improvement of structural index estimation in Euler Silva, J. B. C., Barbosa, V. C. F., and Medeiros, W. E., 2001, Scat-
deconvolution: Geophysics, 64, 48–60. tering, symmetry, and bias analysis of source position estimates in
Bastani, M., and Pedersen, L. B., 2001, Automatic interpretation of Euler deconvolution and its practical implications: Geophysics, 66,
magnetic dike parameters using the analytical signal technique: Geo- 1149–1156.
physics, 66, 551–561. Thurston, J. B., and Smith, R. S., 1997, Automatic conversion of mag-
Debeglia, N., and Corpel, J., 1997, Automatic 3-D interpretation of netic data to depth, dip, and susceptibility contrast using the SPITM
potential field data using analytic signal derivatives: Geophysics, 62, method: Geophysics, 62, 807–813.
87–96. Thurston, J. B., Smith, R. S., and Guillon, J., 2002, A multimodel
Fairhead, J. D., Bennett, K. J., Gordon, D. R. H., and Huang, D., 1994, method for depth estimation from magnetic data: Geophysics, 67,
Euler: Beyond the “black box”: 64th Ann. Internat. Mtg., Soc. Expl. 555–561.
Geophys., Expanded Abstracts, 422–424. Zhang, J., 2001, An analysis of the accuracy of magnetic source-body
Hsu, S., Coppens, D., and Shyu, C., 1998, Depth to magnetic source geometry determined from the 3-D analytic signal: Geophysics, 66,
using the generalized analytic signal: Geophysics, 63, 1947–1957. 579–581.