Professional Documents
Culture Documents
Gsgpy 66 6 1805
Gsgpy 66 6 1805
Manuscript received by the Editor April 19, 2000; revised manuscript received March 16, 2001.
∗
Colorado School of Mines, Gravity and Magnetics Research Consortium, Golden, Colorado 80401. E-mail: mnabighi@mines.edu.
‡Pearson, deRidder, and Johnson, Inc., 12640 West Cedar Dr., Suite 100, Lakewood, Colorado 80228. E-mail: rohansen@prj.com.
°c 2001 Society of Exploration Geophysicists. All rights reserved.
1805
Mushayandebvu et al.’s (1999) equations for a nonzero Euler cept for n = 0. We carry it throughout both for generality and
index appear to be completely new: to handle this case along with that of n > 0. Let
∂M ∂M
(x − x0 ) + (z − z 0 ) + n M = 0, (3) r 2 = (ξ − x0 )2 + (η − y0 )2 (11)
∂x ∂z
and define the 3-D Hilbert transforms by [Nabighian (1984),
∂M ∂M
(x − x 0 ) − (z − z 0 ) + nH(M) = 0, (4) his equations (22) and (28)]
∂z ∂x Z ∞ Z ∞
1 x −ξ
where n is the Euler index and H denotes the Hilbert transform. Hx (M) = M(ξ, η) dξ dη,
Their derivation requires that the field M satisfy Laplace’s 2π −∞ −∞ r3
Z∞ Z ∞
equation because they use a complex analytic transformation 1 y−η
to obtain the second equation. We take some liberties in in- H y (M) = M(ξ, η) dξ dη, (12)
2π −∞ −∞ r3
terpreting their second equation, but our interpretation agrees
with theirs in the special case of a thin dike and furthermore H(M) = Hx (M)i + H y (M)j.
can be shown to be correct. Since we effectively derive the 3-D
In the Appendix we show that the 3-D Hilbert transforms of
equation later, we do not give the derivation here.
M in the sense of Nabighian (1984) also satisfy equation (10)
Equation (4) looks rather odd as it stands because two of
for all n but with different constants for the case n = 0. From
its terms involve M but the Hilbert transform of M appears in
equation (A-37) we obtain two equations which in compact
the third. Using the Cauchy-Riemann equations (Nabighian,
form can be written as
1972),
∂ ∂
∂M ∂ (x − x0 ) H(M) + (y − y0 ) H(M)
= − H(M), (5) ∂x ∂y
∂x ∂z ∂
∂M ∂ + (z − z 0 ) H(M) + nH(M) = β. (13)
= H(M), (6) ∂z
∂z ∂x
Thus, if a solution of Laplace’s equation satisfies an Euler
we can rewrite equation (4) as
equation with index n, so do its generalized Hilbert transforms.
∂ ∂
(x − x0 ) H(M) + (z − z 0 ) H(M) + nH(M) = 0. (7)
∂x ∂z EXAMPLE: THE VERTICAL MAGNETIC DIPOLE
This shows that equation (4) is equivalent to the assertion that To illustrate the 3-D case, we calculate the generalized
if a 2-D potential field is homogeneous with Euler index n, then Hilbert transforms of the potential of a vertical magnetic dipole
so is its Hilbert transform. We furthermore see that the same explicitly and show they satisfy Euler’s equation with index
transformation can be applied to the equations of index zero, two, as predicted.
so the general case can be written in the form The potential V for a unit dipole moment is given by
∂ ∂ z − z0
(x − x0 )
M + (z − z 0 ) M + n M = α, V = , (14)
∂x ∂z R3
∂ ∂ where R 2 = (x − x0 )2 + (y − y0 )2 + (z − z 0 )2 . The derivatives of
(x − x0 ) H(M) + (z − z 0 ) H(M) + nH(M) = β,
∂x ∂z V are
(8) ∂V 3(x − x0 )(z − z 0 )
=− ,
where it is understood that α and β will generally vanish unlesss ∂x R5
n = 0. ∂V 3(y − y0 )(z − z 0 )
=− , (15)
Equations (8) promise to be extremely useful in their own ∂y R5
right as a unification and generalization of Euler and Werner
deconvolution. However, they suggest more—that the algo- ∂V R 2 − 3(z − z 0 )2
= .
rithm might be extended to three dimensions, thereby provid- ∂z R5
ing both an extension of Werner deconvolution to the 3-D case By direct substitution from equations (15), we obtain
and a generalization of the Euler equations. In the next section, ∂V ∂V ∂V
we show this is the case. (x − x0 ) + (y − y0 ) + (z − z 0 )
∂x ∂y ∂z
3-D CASE £ ¤
−3 (x − x 0 ) + (y − y0 ) + (z − z 0 )2 (z − z 0 )
2 2
=
Suppose that M is a potential field, R5
z − z0
∇ 2 M = 0, (9) +
R3
and that M also satisfies the generalized Euler equation with 2(z − z 0 )
index n: =− . (16)
R3
∂M ∂M ∂M
(ξ −x0 ) + (η−y0 ) + (ζ −z 0 ) + n M = α, (10) This shows that V satisfies Euler’s equation with index two.
∂ξ ∂η ∂ζ
For brevity, we omit here the rather trivial but lengthy deriva-
where ξ , η, and ζ are the x, y, and z spatial positions, respec- tions required to calculate the Hilbert transforms of ∂ V /∂ x,
tively, and where α is a constant which normally vanishes ex- ∂ V /∂ y, and ∂ V /∂z for the vertical magnetic dipole. The results
are summarized in Table 1. The symmetry of the expressions area where various bodies with different structural indices n
obtained is remarkable. By applying Hilbert transforms, we might be encountered.
have achieved a circular rotation of the coordinate axes. This is On a more theoretical level, the new algorithm proves the
strongly reminiscent of the 2-D case (Nabighian, 1972) in which appropriateness of the generalized Hilbert transform opera-
the Hilbert transform of ∂ M/∂ x was shown to be ∂ M/∂z. tors (Nabighian, 1984). The rather delicate interaction between
the Euler equation, Laplace’s equation, and the generalized
DISCUSSION Hilbert transform could not plausibly be preserved for any
The results just derived provide a generalization of both other transformation. It also strengthens the notion of the 3-D
Euler deconvolution and Werner deconvolution in three di- analytic signal, although it may be that some modification may
mensions. At a practical level, the new algorithm helps stabilize yet be found which is invariant under changes in magnetization
the Euler algorithm by providing at each point three equations direction, as in the 2-D case.
for each measured field rather than one. The Hilbert trans- One outstanding issue is whether the constants which appear
forms of a field are locally independent of the field; thus, we on the right-hand side of the zero-index equations can in fact be
expect measurable reductions in the uncertainties of the depth calculated from invariants of the field. In that case, the number
estimates obtained using the new equations. Preliminary ex- of unknowns in the zero-index equations could be reduced to
periments confirm this expectation. match that in the n > 0 case. This issue is still under active
Furthermore, the additional equations should make it pos- investigation.
sible to compute the physical property contrast and the strike
and dip directions in the context of a particular model, such as CONCLUSIONS
a dipping contact. In addition, since the Hilbert transform of
a constant is zero, expression (13) does not include a constant We have shown that the extended Euler deconvolution of
regional value, usually inserted in the standard Euler equa- Mushayandebvu et al. (1999) generalizes both Euler decon-
tion (10). volution and Werner deconvolution. A reformulation of their
The new algorithm also generalizes immediately to the equations suggests a generalization to three dimensions, which
multiple-source case (Hansen and Suciu, 2000). This should we have carried out completely. This generalization unifies
have the immediate benefit of increasing the number of equa- Euler and Werner deconvolution in a general 3-D setting. Ex-
tions at each point, thereby stabilizing the depth estimates. plicit calculations for the potential of a vertical magnetic dipole
Unfortunately, physical property and strike and dip angle in- verify that the equations are valid in this case.
formation cannot be extracted in any obvious way except in
the single-source case. REFERENCES
The new algorithm suggests some possible variations of the Hansen, R. O., and Simmonds, M., 1993, Multiple-source Werner de-
Euler deconvolution method which use the fact that there are convolution: Geophysics, 58, 1792–1800.
three equations at each point rather than one. For example, for Hansen, R. O., and Suciu, L., 2002, Multiple-source Euler deconvolu-
tion: Geophysics, in press.
n > 0, it should be possible to solve for x0 , y0 , and z 0 pointwise Mushayandebvu, M. F., van Driel, P., Reid, A. B., and Fairhead,
using the three equations in three unknowns. Another possi- J. D., 1999, Magnetic imaging using extended Euler deconvolution:
bility for n > 0 is to eliminate the structural index n between Presented at the 69th Ann. Internat. Mtg., Soc. Expl. Geophys.
Nabighian, M. N., 1972, The analytic signal of two-dimensional mag-
pairs of equations, yielding a system of two equations at each netic bodies with polygonal cross-section: Its properties and use for
point which are still linear in x0 , y0 , and z 0 , do not contain n automated interpretation: Geophysics, 37, 507–517.
——— 1984, Toward a three-dimensional automatic interpretation of
explicitly, but are bilinear in the field variables. This approach potential field data via generalized Hilbert transforms: Fundamental
might prove useful when carrying out calculations over a large relations: Geophysics, 49, 780–786.
APPENDIX
DERIVATION OF 3-D HILBERT TRANSFORM RELATIONS
Suppose that M is a potential field satisfying the generalized The x-component of this equation is
Euler equation with index n: Z Z
∞ ∞
1 x −ξ ∂M
∂M ∂M ∂M (η − y0 ) dξ dη
(ξ − x0 ) + (η − y0 ) + (ζ − z 0 ) + n M = α, 2π −∞ −∞ r 3 ∂η
∂ξ ∂η ∂ζ Z ∞Z ∞
(A-1) 1 x −ξ ∂M
= [(η − y) + (y − y0 )] dξ dη
where α is a constant which normally vanishes except for n = 0. 2π −∞ −∞ r 3 ∂η
Let Z ∞Z ∞
1 η−y ∂M
r 2 = (ξ − x0 )2 + (η − y0 )2 , (A-2) = (x − ξ ) dξ dη
2π −∞ −∞ r 3 ∂η
∂ ∂ µ ¶
D= i+ j, (A-3) ∂M
∂ξ ∂η + (y − y0 )Hx
∂η
and define the 3-D Hilbert transforms by [Nabighian (1984), Z ∞Z ∞
equations (22) and (28)] 1 η−y ∂M
Z Z = [(x − x0 ) − (ξ − x0 )] dξ dη
1 ∞ ∞
x −ξ 2π −∞ −∞ r 3 ∂η
Hx (M) = M(ξ, η) dξ dη, µ ¶
2π −∞ −∞ r3 ∂M
Z Z + (y − y0 )Hx
1 ∞ ∞
y−η ∂η
H y (M) = M(ξ, η) dξ dη, (A-4) µ ¶ µ ¶
2π −∞ −∞ r3 ∂M ∂F
= −(x − x 0 )H y + Hy
H(M) = Hx (M)i + H y (M)j. ∂η ∂η
µ ¶
∂M
Multiply equation (A-1) by (1/2π)D(1/r ) and integrate over + (y − y0 )Hx , (A-10)
ξ and η from −∞ to ∞: ∂η
Z ∞ Z ∞ ·
1 ∂M ∂M where F = (ξ − x0 )M. The y-component is equal to
+ (η − y0 )
(ξ − x0 )
2π −∞ −∞ ∂ξ ∂η Z Z
¸ µ ¶ ∞ ∞
1 y − η ∂M
∂M 1 (η − y0 ) dξ dη
+ (ζ − z 0 ) + nM D dξ dη 2π −∞ −∞ r 3 ∂η
∂ζ r Z Z µ ¶
Z ∞Z ∞ µ ¶ 1 ∞ ∞
y − η ∂G
1 1 = − M dξ dη
= αD dξ dη. (A-5) 2π −∞ −∞ r3 ∂η
2π −∞ −∞ r µ ¶
∂G
We consider the terms in equation (A-5) one by one, working = Hy − H y (M), (A-11)
from the right. For the right-hand side we have ∂η
Z ∞ Z ∞ µ ¶
α 1 where G = (η − y0 )M.
D dξ dη = 0. (A-6)
2π −∞ −∞ r The final integral to be evaluated is
Finally, evaluating equation (A-27) on the observation surface For n = 0, the situation is not as clear. In that case, the dc term
ζ = z, we obtain may be undefined and an additional constant may appear in
∂ F̃ p the inverse Fourier transform. Thus, expression (A-32) should
( p, q, z) = p 2 + q 2 F̃( p, q, z) be modified to
∂z µ ¶ µ ¶ µ¶
ip ∂F ∂F ∂M
+p M̃( p, q, z). (A-28) Hx + Hy = (x − x0 )Hx
p2 + q 2 ∂x ∂y ∂x
µ ¶
∂M
Equation (A-28) can be rewritten in terms of generalized + (x − x0 )H y
Hilbert transforms using equation (17) of Nabighian (1984):
∂y
∂F + Hx (M) − βx , (A-33)
( p, q, z) = H̃x i p F̃( p, q, z) + H̃ y iq F̃( p, q, z)
∂z where βx is a constant which can be expected to vanish unless
− H̃x M̃( p, q, z). (A-29) n = 0 and the negative sign is simply for later convenience.
Substituting for the left-hand side of equation (A-33) in ex-
Taking the inverse Fourier transform yields pression (A-15), we are left with
µ ¶ µ ¶ µ ¶ µ ¶
∂F ∂F ∂F ∂M ∂M
(x, y, z) = Hx (x, y, z) + H y (x, y, z) (x − x0 )Hx + (y − y0 )Hx − βx . (A-34)
∂z ∂x ∂y ∂x ∂y
− Hx (M)(x, y, z). (A-30) An exactly similar argument for G yields the y-component:
However, µ ¶ µ ¶
∂M ∂M
∂F ∂M (x − x0 )H y + (y − y0 )H y − β y . (A-35)
= (x − x0 ) . (A-31) ∂x ∂y
∂z ∂z
Collecting all terms from the original Hilbert transforms and
Rearranging this last equation to put the remaining terms in F returning to vector notation, we have
on the left-hand side we obtain µ ¶ µ ¶
µ ¶ µ ¶ ∂M ∂M
∂F ∂F ∂M (x − x0 )H + (y − y0 )H
Hx + Hy = (x − x0 ) + Hx (M) ∂x ∂y
∂x ∂y ∂z µ ¶
µ ¶ ∂M
∂M + (z − z 0 )H + nH(M) = β. (A-36)
= (x − x0 )Hx ∂z
∂x
µ ¶
∂M A straightforward integration by parts argument, which we
+ (x − x0 )H y + Hx (M). omit, shows that the generalized Hilbert transform commutes
∂y
with differentiation for functions with reasonable behavior at
(A-32) infinity, which will be the case for M. We finally obtain
Finally, we return to the case where p 2 + q 2 = 0, i.e., the dc ∂ ∂
term. For Euler index n > 0, we expect that equation (A-32) (x − x0 ) H(M) + (y − y0 ) H(M)
∂x ∂y
will stand as written because in that case M falls off as O(1/r )
∂
at infinity and any dc coefficient will be bounded. This pre- + (z − z 0 ) H(M) + nH(M) = β. (A-37)
cludes the addition of any extra constant term to the equation. ∂z