Journal of Applied Geophysics 75 (2011) 472–478

Contents lists available at SciVerse ScienceDirect

Journal of Applied Geophysics

journal homepage: www.elsevier.com/locate/jappgeo

3D inversion of gravity data by separation of sources and the method of local

corrections: Kolarovo gravity high case study
Ilya Prutkin a,⁎, Peter Vajda b, Robert Tenzer c, Miroslav Bielik b, d
a
Institute of Earth Sciences, Jena University, Burgweg 5 11, 07749 Jena, Germany
b
Geophysical Institute, Slovak Academy of Sciences, Dubravska cesta 9, 845 28 Bratislava, Slovakia
c
National School of Surveying, Division of Sciences, University of Otago, 310 Castle Street, Dunedin, New Zealand
d
Department of Applied and Environmental Geophysics, Faculty of Natural Sciences, Comenius University, Mlynska dolina, 842 15 Bratislava, Slovakia

a r t i c l e i n f o a b s t r a c t

Article history: We present a novel methodology for 3D gravity/magnetic data inversion. It combines two algorithms for pre-
Received 30 March 2011 liminary separation of sources and an original approach to 3D inverse problem solution. The first algorithm is
Accepted 12 August 2011 designed to separate sources in depth and to remove the shallow ones. It is based on subsequent upward and
Available online 14 September 2011
downward data continuation. For separation in the lateral sense, we approximate the given observed data by
the field of several 3D line segments. For potential field data inversion we apply a new method of local cor-
Keywords:
Potential field methods
rections. The method is efficient and does not require trial-and-error forward modeling. It allows retrieving
3D inversion unknown 3D geometry of anomalous objects in terms of restricted bodies of arbitrary shape and contact sur-
Gravity interpretation faces. For restricted objects, we apply new integral equations of gravity and magnetic inverse problems. All
Anomalous body steps of our methodology are demonstrated on the Kolarovo gravity anomaly in the Danube Basin of Slovakia.
© 2011 Elsevier B.V. All rights reserved.

1. Introduction
Our main goal was to develop an approach for interpretation of poten-
tial field data, which has to face the following challenges: i) to be really
three-dimensional (to avoid an assumption that the object sought is an
infinite cylinder); ii) to abandon the simplest bodies like prisms, ellip-
soids, cylinders of finite length for objects of more complex geometry;
iii) to take into account the non-uniqueness of the inverse problem. Our
approach provides different admissible solutions to the geological section
based on potential field data. Later it can be discriminated among these
solutions based on additional geophysical, geological or tectonic con-
straints (if such information exists). We clarify all details of the approach
while interpreting the Kolarovo gravity high data in the southern Slovak
part of the Pannonian Basin.
This gravity anomaly is located in the south-eastern part of the
Danube Basin (a part of the Pannonian Basin in Central Europe). The
anomaly belongs to one of the largest and most famous gravity
highs in the Western Carpathian area. Therefore it has been of signif-
icant interest to geophysicists and geologists since early sixties of the
twentieth century. In terms of the Bouguer anomaly, the Kolarovo
high reaches a magnitude of 28 mGal. It is supposedly associated
with deep seated sources within the pre-Neogene basement (Kubeš
et al., 2010). According to Kubeš et al. (2010), the interpretations of
⁎ Corresponding author. Tel.: + 49 3641 948751; fax: + 49 3641 948662.
E-mail address: Ilya.Prutkin@uni-jena.de (I. Prutkin).
these sources are insufficient and problematic and thus remain open
for further investigation.
The quantitative interpretation of this gravity high (Bielik et al., 1986;
Sitárová et al., 1984) has revealed that the density of the anomalous body
varies from 2900 to 3050 kg/m3, compared to the density of the upper
crust of 2700 kg/m3. Sitárová et al. (1984) determined a source body as
a right-rectangular homogeneous prism with a top plane at 4.5 km
with a center of mass at 9.5 to 12.5 km. Bielik et al. (1986) interpreted
the Kolarovo anomaly by means of 3D gravimetric forward modeling
using polyhedrons. The TFM (Truncation Filtering Methodology) inter-
pretation of the Kolarovo high yielded an estimate of the depth of the
center of mass of the body at 8.7 km (Vajda et al., 2002).
We start with separation of sources. First, we remove shallow
sources down to the depth corresponding to the assumed basement.
Our algorithm is based on the subsequent upward and downward
continuation of the observed gravity data. During this procedure, we
integrate data along the area of investigation only. This has been
made possible by prior subtraction of a model of the regional field,
which we regard as a 2D harmonic function, from the observed
data, thus forcing the data outside the investigation area to be zero.
To illustrate the regional field subtraction and the separation of
sources in depth, we refer to (Prutkin and Casten, 2009; Prutkin and
Saleh, 2009), where this algorithm was applied to isolate the contri-
bution of the Moho interface to the total field. Then, the 3D Moho to-
pography was found by means of the method of local corrections
applied to the thus separated residual field (Prutkin and Saleh, 2009).
After obtaining the separated signal of interest, it is interpreted by
means of direct gravity inversion, using the method of local corrections.

0926-9851/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.jappgeo.2011.08.012
 I. Prutkin et al. / Journal of Applied Geophysics 75 (2011) 472–478
We do not use the time-consuming forward modeling approach. In the
method of local corrections, a solution is calculated from gravity data
automatically by successive approximations. The non-uniqueness and
instability of the inverse problem are taken into account.
Now we expand our methodology to be able to retrieve unknown
geometry both for a contact surface and for a 3D restricted body of
arbitrary shape. We approximate the given data by the field of sever-
al 3D line segments. This procedure allows to separate sources both
in depth and also in the lateral direction, and to estimate their
masses. In the next step it is possible to find a shape of a restricted
object by means of the method of local corrections using new inte-
gral equations of gravity (and magnetic) inverse problems that
were described in Prutkin (2008) and applied to the core–mantle
boundary recovery.
The approximation of the Kolarovo gravity anomaly with 3D line
segment sources permits to study different models of the geological
section. We are able to investigate various options to join segments
and to relate the obtained gravitational effect to the same object (a
restricted body or a contact surface). We have obtained a solution
to the geological section which is in good agreement with results of
previous interpretations mentioned above. Moreover, our new inter-
pretation methodology provides three new admissible models: a
body with more complicated geometry, an uplift of a relatively denser
magmatic material as a contact surface, and a combination of an
anomalous body and an elevation of the contact surface.
2. Separation of sources in depth
We begin with the algorithm based on the upward and subse-
quent downward continuation (Prutkin and Casten, 2009), see also
(Martyshko and Koksharov, 2005; Martyshko and Prutkin, 2003).
The main purpose of this algorithm is to find a part of the observed
field which is harmonic above a given depth h. We can treat this func-
tion as an effect of the half-space below the depth h. To find such
function means to eliminate all sources located in the layer from the
Earth’s surface, which we approximate by the plane z = 0, down to
the prescribed depth. The algorithm allows separating the effects of
shallow and deeper source objects.
Firstly, we continue the data upward to the level z = h to diminish
the influence of the sources in the near-surface layer. This operation
causes the main errors in the vicinity of the boundary of the area. To re-
duce these errors we need a model of the regional field to be subtracted
from the observed field prior to the upward continuation. Secondly, we
continue the obtained function downward to the plane z = −h, i.e. over
the distance 2h downwards. Since the downward continuation proce-
dure is a linear ill-posed inverse problem, the regularization is applied.
The function, which we treat as a regional field, is assumed to sat-
isfy two-dimensional Laplace equation in the area of investigation
S and to have the same values at the boundary of the area, as the ob-
served field:
8 2 2 field is very similar to the original one. It is a confirmation that the Kolar-
<∂ f ∂ f
þ 2 ¼0 within area S ð1Þ
:∂ x ∂ y
2
f ¼ Δg on its boundary ∂S
If we subtract the values of this function, the residual field will be
equal to zero at the boundary of the area. In this way no errors are in-
troduced, when we integrate the residual field while upward contin-
uation within the restricted area. According to the properties of
harmonic functions, this function has no extreme values within the
area, so we do not create artificial anomalies. Besides, as it is known
from calculus of variations, a solution to the problem (1) minimizes
the norm of the gradient, therefore, we obtain the smoothest possible
function with given values on the boundary. All these properties
allow us to regard this function as a model of the regional field.
473
After subtracting the suggested model of the regional field we per-
form upward continuation of the residual field by means of the Poisson
integral:
1 h  
∫∫     3=2 U ðx; y; 0Þdxdy ¼ U x′ ; y′ ; h ð2Þ
2π 2 2
x−x′ þ y−y′ þ h2
Formula (2) gives a solution of the Dirichlet problem: to find a
harmonic function in the upper half-space with given values on the
boundary (on the plane z = 0 ).
At the second step we continue the obtained function downward
to the depth h , i.e., over the distance 2h downwards. To do this, we
apply a formula similar to Eq. (2):
1 h1  
∫∫     3=2 U ðx; y; −hÞdxdy ¼ U x′ ; y′ ; h ; ð3Þ
2π ′ 2 ′ 2
x−x þ y−y þ h 2
1
where h1 = 2h. Eq. (3) provides a function, which is harmonic in a
half-space above the plane z = − h . This time we have to, and we
do, treat the formula as an integral equation: the right hand side is
given, while the unknown function U(x, y,−h) on the left hand side
is sought.
Although the signal of the shallow sources is attenuated after apply-
ing the upward continuation, it is not completely removed. As a conse-
quence, the downward continuation is performed through the sources.
It is worth emphasizing that the problem of downward continuation
is a linear ill-posed inverse problem, therefore some regularization
must be applied. Since the integral operator A in Eq. (3) is self-adjoint
and positive, we apply the Lavrent'ev's approach (Lavrent'ev et al.,
1986). If we write Eq. (3) in the following form: Au= uh, u is an un-
known field at the level z = −h, uh is the obtained field at height h,
then the Lavrent’ev’s regularization gives: (A + αI)u = uh, where I is
the identity operator, and α is a regularization parameter.
The choice of the regularization parameter decides how well the
solution fits the data. To find a suitable regularization parameter,
we use the method of residuals (Lavrent'ev et al., 1986), which ex-
ploits information about data noise: we look for the smoothest possi-
ble solution, which approximates the given data with the same
accuracy as the data noise level.
At last, we calculate (by an upward continuation) the field on the
Earth’s surface z = 0 using a formula, similar to Eq. (3). We obtain a
part of the field, which is harmonic down to the depth h, so that we
now can treat it as an effect of the deeper sources, deeper than z =−h.
We apply the described methodology to the Kolarovo gravity anoma-
ly. Our goal is to remove shallow sources, thus we take for height h a value
equal to 2 km, which corresponds to the minimum thickness of the Neo-
gene sediments. The initial data and the results of our processing are
shown in Fig. 1. The amplitude has practically not changed. The obtained
ovo gravity high is caused mainly by deeper sources.
3. Approximation with 3D line segments
To separate sources not in depth only, but also in the lateral direc-
tion, as well as to estimate their masses, we propose to approximate
data by simple sources prior to 3D inversion. As a source for gravity
data approximation, we use 3D line segments. The effect of a line seg-
ment can be evaluated by a quite simple formula.
Let r0 = (x0, y0, z0) be the radius-vector of an observation point,
r1 = (x1, y1, z1) and r2 = (x2, y2, z2)—radius-vectors of endpoints of a
3D line segment, w—its line density. We introduce following notations:
α is the angle between vectors r0−r1 and r2−r1, c = |r2−r1| is the
length of the line segment, c0 = |r0−r1|, c2 = |r0−r2|, a0 = z0−z1, a1 =
474 I. Prutkin et al. / Journal of Applied Geophysics 75 (2011) 472–478

20
segment generates a field similar to a point source, two segments are
0
usually sufficient to approximate an effect of a restricted object with
5
complex geometry. Besides, each 3D line segment is described by 7 pa-
10 rameters only: 3 coordinates of both two ends and the line density.
This is an important point, because we obtain these parameters while ap-
10 proximating the data by means of nonlinear minimization.
0 10 To demonstrate the advantages of the 3D line segment approxima-
km

0
5 15
tion, we begin with a synthetic field example. The gravitational effect of
−5
two rectangular prisms is calculated, one of them being cubic, the other
−10 being stretched in the vertical direction. In Fig. 2 we present the synthetic
field (Fig. 2a) and projections of the prisms, as well as the obtained 3D
line segments to the horizontal (Fig. 2b) and vertical (Fig. 2c) coordinate
−20 planes. Differences between the given data and the field of line segments
(Fig. 2d) have RMS of 0.0188 mGal. This approximation yields not only
−30 −20 −10 0 10 20 the separation of sources and reliable mass estimates, but also reasonable
km information about the shape of the sought source body.
Then we approximate the Kolarovo gravity high with a field of sev-
−15 −11 −7 −3 1 5 9 13 17 21 25 eral 3D line segments. The gravity data set, given at an equidistant reg-
mGal ular grid, was obtained from Pohánka (2001). The origin of local planar
20 coordinates was chosen at the point of latitude 47∘57′ N, longitude
18∘00′ E, and height above sea level of 110 m, the x-coordinate being
0
5
easting and y-coordinate northing. The gravity data are represented
10
by the vertical component of the gravitation acceleration vector. Since
the given area lies in lowlands and is relatively flat (variations in height
are of only a few meters), no topographic correction was applied, while
10 10 the data were vertically reduced to the reference altitude of 110 m
0 0
using the free air gradient. The regular equally spaced grid of data was
km

5
15
−5
produced from the original irregularly placed data points by an in-
house interpolation method described in (ibid). The interpolation
−10
method smooths the data according to a given parameter, the so called
smoothing distance. Gravity data were interpolated onto a 200 by 200-
meter grid. A smoothing distance of 400 m was chosen, while the mean
−20 minimum distance between the original data points was 425 m, which
resulted in minimum smoothing effect. A mean value over the regular
−30 −20 −10 0 10 20 grid was subtracted from the gridded gravity.
km We take the residual field after subtracting the model of the re-
gional field and after eliminating the signal of shallow sources by
−15 −11 −7 −3 1 5 9 13 17 21 25 means of the upward and downward continuation. Three line seg-
mGal ments are sufficient for fairly accurate approximation of the given
data. The RMS of differences between the observed data and the
Fig. 1. Initial data and their preliminary processing. Top-gridded data of the Kolarovo
field of the line segments is 0.57 mGal. The gravity field (without re-
gravity high; bottom—contribution of shallow sources down to the basement is removed
by means of the algorithm based on upward and downward continuation. gional component) and the field of 3D line segments are presented in
Fig. 3. All ends of line segments are located at depths between 6 and
9 km. It should be noted that we were seeking 21 parameters of the
(z2−z1)/c. For the vertical derivative of the gravitational potential we line segments respective to nearly 60,000 observations. Hence, this
have procedure is quite stable.
"    #
1 1 a1 c0 cos α−a0 c−c0 cos α 4. Method of local corrections for 3D data inversion
Vz ðr0 Þ ¼ w⋅ a1 − þ   þ cos α ð4Þ
c0 c2 c20 1−cos2 α c2
In this section our approach to the 3D potential field data inversion is
reviewed. We begin with the case of a restricted source body. We seek
for the general case cos2 α ≠ 1 (the observation point does not belong to
geometry of an unknown homogeneous body; the only requirement is
the continuation of the line segment). In the special case cosα = 1 in-
that the sought body is assumed to be star convex relative to some
stead of Eq. (4) we obtain
point of the body. This means that there exists a point of the body
" # such that the line segment from this point to any other point of the
w a0 þ a1 c0 a0 þ a1 ðc0 −2cÞ
Vz ðr0 Þ ¼ − ð5Þ body is contained in the body. In this case we can introduce spherical
2⋅ c20 c22 coordinates r, θ, φ relative to the point. The body boundary can be deter-
mined by the equation r = r(θ, φ), where r(θ, φ) is a single-valued func-
For cos α =−1 we have the same as in Eq. (5), but the endpoints tion. We introduce a Cartesian coordinate system, the origin of which
of the segment change their places: c0 = |r0− r2|, c2 = |r0− r1|, a0 = coincides with the center of the spherical coordinate system. Let r =
z0−z2, a1 = (z1− z2)/c. (x, y, z) be the radius-vector of an observation point. The gravitational
The computation time to calculate the effect of a 3D line segment in inverse problem can be reduced to the following nonlinear equation re-
one observation point exceeds the corresponding time for a point source spective to an unknown function r(θ, φ):
only about 2 times. At the same time, it is a considerably flexible tool for
the approximation of measurements: one 3D line segment approximates
perfectly an effect of an object stretched in arbitrary direction; a short GΔσ∫∫ K ðx; y; z; θ; φ; r ðθ; φÞÞ dθdφ ¼ U ðx; y; zÞ; ð6Þ
 I. Prutkin et al. / Journal of Applied Geophysics 75 (2011) 472–478 475

a 20
b
10

15

10
5

5
km

km
0

−5

−10
−5

−15

−20
−20 −15 −10 −5 0 5 10 15 20 −10
km −10 −5 0 5 10
km
0 2 4 6 8 10 12 14 16 18 20
mGal
c d 20

10 mGal gravity data

model field
15

10
5 mGal
5
km

−5

3 km
−10

6 km
−15

9 km −20
−20 −15 −10 −5 0 5 10 15 20
-20 -15 -10 -5 0 5 10 15 20
km
km

−0.06 −0.04 −0.02 0.00 0.02 0.04 0.06

mGal

Fig. 2. Approximation of synthetic data with 3D line segments. a) Synthetic gravity field of two prisms; b) and c) projections of the prisms and the obtained 3D line segments to the
horizontal and vertical coordinate planes; d) residuals of approximations, RMS = 0.0188 mGal.

where G is the gravitational constant, Δσ is the density contrast, U(x, y, reduction of time expenditures to solve it, approximately by an
z) is the given field (gravitational potential, some of its derivative, or order of magnitude.
their combination). The density contrast Δσ is included in Eq. (6) as a After discretization of Eq. (6) by numerical integration we obtain
numerical factor. It allows obtaining an equivalent set of solutions gen-  
erating—with different values of the density contrast—the same gravita- cGΔσ ∑∑ K i 0 j 0 rij ¼ Ui0 j0 ; ð7Þ
i j
tional field. For a fixed value of Δσ, the solution of the inverse problem is
unique—for this preselected class of solutions—according to Novikov's
where c is the weight of the cubature formula, rij = r(θi, φj), and
theorem (Novikov, 1938).
Ui0j0 = U(θi0, φj0, ρ(θi0, φj0)), Ki0j0(rij) = K(θi0, φj0, ρ(θi0, φj0), θi, φj, rij).
We propose the method of local corrections for solving nonlinear
Our goal is to develop an iterative procedure for solving the sys-
potential inverse problems similar to that given in Eq. (4). The idea
tem of nonlinear Eq. (7). Suppose that rijn are the values of the un-
is that, in computing the integrals in Eq. (6) and also in evaluating
known function obtained at the n-th step. The idea of a local
the field values on some closed surface r = ρ(θ, φ), containing the ob-
correction described above similar to the case of a contact surface,
ject sought within it, the very same nodes {θi, φj} are used. In each it-
see e.g., Prutkin and Saleh (2009) leads to the fundamental equation
eration an attempt is made to reduce the difference between the
to find the next approximation
given and approximate field values at a fixed node solely by modify-
ing the value of the unknown function at the same node. These con-       
nþ1 n n
GΔσ Kij rij −Kij rij ¼ α Uij −Uij ; ð8Þ
siderations lead to decomposition of the inverse problem and
476 I. Prutkin et al. / Journal of Applied Geophysics 75 (2011) 472–478

20 where V(r) denotes the gravitational potential of the body. For the
function Ψ(r) introduced in Eq. (9) the following relation holds

10
GΔσ 2π π r 3 ðθ; φÞsin θ dθd φ
4 ΨðrÞ ¼ ∫0 ∫0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð10Þ
2
8 4
r 2 ðθ; φÞ−2brðθ; φÞ þ a2
12
0 12
km

16 where b = x sin θ cos φ + y sin θ sin φ + z cos θ, and a 2 = x 2 + y 2 + z 2.

8
The integrand in Eq. (10) is algebraic relative to the unknown func-
−10 tion and does not contain its derivatives.
4
In the case of the core–mantle boundary (Prutkin, 2008), harmonic
coefficients of the gravitational potential are given on the Earth’s surface.
−20 Based on Eq. (9), in this case we can calculate from them the harmonic
coefficients of Ψ(r) and then generate its values on the Earth’s surface
which we use as a closed surface containing an object with unknown ge-
−30 −20 −10 0 10 20
km ometry (the Earth's core). In the case of a local object, we benefit from the
preliminary approximation with the 3D line segments. After this proce-
−15 −11 −7 −3 1 5 9 13 17 21 25 dure, the parameters of segments are known. Therefore, while searching
mGal for the geometry of an object causing the same gravity effect as a given set
20
of line segments, we can easily calculate values of Ψ(r) on any closed sur-
face allegedly containing the unknown body.
To improve the solution, one can proceed from the local correc-
tions method to solving the integral equation for the inverse problem
10
4 by Newton’s method. This method is applied rather rarely, since it en-
sures convergence (yet extremely rapid) only in the presence of an
8 4
12 initial approximation close to the solution sought. In our case, such
0 12
an approximation is provided by the method of local corrections.
km

20
8 16 New integral equation for the magnetic inverse problem, as well as
integral equations for the linearized gravitational and magnetic in-
−10 verse problems can be found in Prutkin (2008).
4
It should be noted that in the case of a contact surface we can
apply the method of local corrections directly to gravity data on the
−20 physical surface. The approximation with 3D line segments is used
only if we need to separate sources. The formula to find the next ap-
−30 −20 −10 0 10 20 proximation is easier than for a restricted object. For more details we
km refer readers to Prutkin and Saleh (2009).

−15 −11 −7 −3 1 5 9 13 17 21 25 5. Model solutions for the geological section

mGal
Fig. 3. Approximation of the Kolarovo gravity anomaly with line segments. Top—
observed data after subtracting the model regional field (a 2D harmonic function);
bottom—gravitational effect of three line segments. The RMS of residuals is
0.57 mGal.
where Uijn) is the solution of the direct problem corresponding to the
model rijn obtained at the n-th step. The coefficient α on the right hand
side of Eq. (8) is introduced to slow down the change of the model.
The integral equation of the 1st kind in Eq. (6) is ill-posed and re-
quires regularization. Diminishing the coefficient α in Eq. (8), we
can prevent highly oscillating solutions. Therefore, this factor is really
similar to a regularization parameter.
From Eq. (8) it is clear that time expenditures to solve the inverse
problem depend dramatically on complexity of the kernel K in Eq. (6). If
we use to derive Eq. (6) a representation of the field in the form of a vol-
ume integral, then the kernel function in the corresponding equation is
rather complicated, and, in addition, it is a transcendental function of r
(θ,φ). If we apply a representation in the form of a surface integral, then
the integrand includes derivatives of the unknown function, but the dif-
ferentiation of the approximately computed function r(θ,φ) is an ill-
posed problem. We have derived new integral equations for gravitational
and magnetic inverse problems which are free from both disadvantages.
We introduce the following function (cf. Prutkin, 2008):
1 1 ∂V
ΨðrÞ ¼ V ðrÞ− r⋅∇V ðrÞ ¼ V− r ; ð9Þ
2 2 ∂r
After the separation of gravitational signals we transform the for-
mal sources used as anomalous sources approximation (3D line seg-
ments) into objects of more geological meaning. For a given set of
segments we obtain a restricted body or a contact surface generating
the same gravity.
For the Kolarovo gravity anomaly, the central line segment has sub-
stantially bigger line density than the two other. One can treat this as
separation of sources and relate the field of the two lighter segments
to the basement topography. We attribute the effect of the main seg-
ment to the anomalous body (a basaltic intrusion). According to results
from available interpretations, we assume for the ambient medium a
density of 2700 kg/m3. The density of the anomalous body equals to
3000 kg/m3. The main segment is substituted by an equivalent (relative
to the external field) homogeneous body with the corresponding densi-
ty contrast. The body is located below the upper boundary of the base-
ment. Depths to the basement vary from 2 to 3 km. In Fig. 4 (top) a
projection to the horizontal coordinate plane is shown, basement to-
pography is presented by means of depth isolines. The body (with the
line segment inside) is also displayed. Two other figures show projec-
tions to the vertical coordinate planes. In the upper half the solid line
corresponds to given gravity, the dashed line to the model field. Depths
to the center of mass, as well as to the upper and lower boundary of the
body are in good agreement with the previous interpretations.
By means of our methodology we can obtain alternative solutions
of the geological section that are also admissible from the viewpoint
of gravity data. For instance, we can attribute the whole gravity effect
to a contact surface separating the underlying layer with density
 I. Prutkin et al. / Journal of Applied Geophysics 75 (2011) 472–478 477

are nearly flat, the observed field is dominantly caused by the shape
of the source body. In this case, we obtain an intrusion with more
complex geometry (see Fig. 6, top). The body is equivalent in the
gravitational effect to all three line segments obtained by the line seg-
ment approximation.
The last admissible solution of the geological section presented is
based on the following considerations. We have applied to the given
gravity the algorithm of the upward and downward continuation and
observed that even if we remove sources up to the depth of 10 km, a re-
sidual signal with the amplitude of 10 mGal is still present (a half of the
initial amplitude). It looks like a low-frequency isometric anomaly. We
relate this component of gravity to an elevation of the dense layer below
the body. If we subtract this low-frequency signal, the residual field can
be approximated quite accurately (on the level of 0.75 mGal) with using
only two 3D line segments. This approximation is more preferable, be-
cause two segments have lower number of parameters than three seg-
ments, so this procedure is more stable. In Fig. 6 (bottom) the last
solution of the geological section is presented. It includes the contact
surface (the elevation of the upper boundary of the dense layer) and
the intrusion of the dense material above (it corresponds to both the
3D line segments).

6. Summary and conclusions

The objective of this paper was to introduce the new methodology

for 3D inversion of potential field data (with focus on gravity anoma-
lies) in terms of signal separation both in depth and in lateral sense
and the method of local corrections. To introduce the methodology
we have demonstrated its capabilities on a case of the Kolarovo grav-
ity high in the Danube Basin, southern Slovakia. The methodology be-
longs to the realm of direct inversion methods, making use of no
forward modeling. It works with a preselected class of solutions:

20
16 18

18 20 16

10 14 16

14
12
0 20 18
10
km

12
16
8
14 18
−10 20
16
18
16

−20 20
18

−30 −20 −10 0 10 20

Fig. 4. Anomalous body below basement with topography. Top—projection of the body
km
equivalent to the central 3D line segment onto the horizontal coordinate plane (the line
segment is shown also), basement topography is presented by depth isolines. Middle 5 7 9 11 13 15 17 19 21 23 25
and bottom—projections to the vertical coordinate planes. Solid line—given data, dashed km
line—model field.

3000 kg/m 3 from the upper layer with density 2700 kg/m 3 and esti-
mating its depth. In this case, we do not need the approximation
with 3D line segments. We invert directly the gravity data on the
physical surface (without regional component). The depths of the
contact surface obtained by the method of local corrections vary
from 22.5 km to 7.5 km. The solution is presented in Fig. 5 in two
ways: as a map of the contact surface topography (depth isolines
are shown) and as a 3D surface. For visualization of 3D objects we
apply Poisson Surface Reconstruction software (Kazhdan et al., 2006). Fig. 5. Undulations of the density contrast contact surface. Top—the obtained 3D topography
The next admissible solution indicates that both the basement to- of the density interface as a map of depth isolines, bottom—shaded 3D surface of the density
pography and the upper boundary of the dense layer below the body interface.
478 I. Prutkin et al. / Journal of Applied Geophysics 75 (2011) 472–478
Fig. 6. Two additional admissible solutions. Top—intrusion with complex geometry
causing the same gravity as the three line segments. Bottom—anomalous body above
a dense layer with topography. The body is equivalent (by its gravitational effect) to
two line segments used for the approximation.
source bodies of compact star convex geometry, and 3D contact sur-
faces representing subsurface density interfaces. The ultimate goal
was not to find the best interpretation for the Kolarovo gravity data.
The goal was to demonstrate that the methodology yields a suite of
admissible solutions that can be further discriminated based on addi-
tional geophysical, geologic or tectonic constraints. The quest for the
best Kolarovo interpretation will be the subject of a separate paper
currently under preparation.
The following conclusions are drawn from our study:
1. A new algorithm is proposed to eliminate the contribution of poten-
tial field sources from the Earth’s surface to a prescribed depth h,
based on applying the upward and downward continuation proce-
dures. The solution to the 2D Dirichlet problem is used to produce a
model of the regional field. Subtracting the regional field from the ob-
served data prior to the upward continuation allows the integration of
gravity data in the restricted study area. The downward continuation
procedure produces the part of the field which is harmonic above the
preselected depth h. The algorithm is successfully applied to separate
the effects of shallow and deeper sources for the Kolarovo gravity
anomaly interpretation.
2. Preliminary approximation using the 3D line segments allows sepa-
rating the sources not only in depth, but also in lateral direction.
Moreover, it provides reasonable mass estimates and valuable infor-
mation about the shape of the source body. We divide the interpre-
tation process into two steps: we begin with data approximation
by the field of formal sources (3D line segments) with no geological
meaning; in the second step we join segments and attribute their ef-
fects to a restricted body or a contact surface with the same gravity.
In this latter step different solutions for the geological section can be
investigated.
3. The method of local corrections is developed to recover 3D geometry
of a contact surface or a restricted body. The method offers a simple
and effective procedure for solving the nonlinear inverse problem
without any linearization. This method does not make use of non-
linear minimization, which reduces the computer calculation time
by an order of magnitude.
4. To find geometry of a 3D restricted object, we propose new inte-
gral equations for gravitational and magnetic inverse problems.
Their integrands are algebraic relative to the unknown function
and do not contain its derivatives. The use of modified potentials
instead of the conventional gravitational and magnetic potentials
does not introduce any additional difficulties.
5. Our methodology is successfully applied in the interpretation of
the Kolarovo gravity high in Slovakia. We separated sources and
inverted the gravity data by means of the herein described algo-
rithms. This methodology provides various model solutions for
the geological section: a body inside a layer with topography, an
intrusion with complex geometry, an elevation of a contact sur-
face, a body above a layer with topography. Among these admissi-
ble solutions, one can make a choice based on additional geological
and geophysical constraining information.
Acknowledgements
The Kolarovo gravity data interpolated to a dense regular grid at a
level surface were produced by Vladimir Pohanka. Peter Vajda was
supported by the Slovak Research and Development Agency under
contract no. APVV-0194-10 and by Vega grant agency under project
no. 2/0107/09. Miroslav Bielik was supported by Vega grant agency
under project no. 1/0461/09.
References
Bielik, M., Fusán, O., Plančár, J., Biela, A., Túnyi, I., 1986. Some knowledge on deep-
seated structure of the Danube Basin (in Slovak) Geologické Práce, Správy 84,
119–134.
Kazhdan, M., Bolitho, M., Hoppe, H., 2006. Poisson surface reconstruction. In: Polthier, K.,
Sheffer, A. (Eds.), Proceedings of the fourth Eurographics symposium on Geometry
processing (SGP ’06). Eurographics Association, Aire-la-Ville, Switzerland, pp. 61–70.
Kubeš, P., Bezák, V., Kucharič, L., Filo, M., Vozár, J., Konečný, V., Kohút, M., Gluch, A.,
2010. Magnetic field of the Western Carpathians (Slovakia): reflections on the
structure of the crust. Geologica Carpathica 61 (5), 437–447.
Lavrent'ev, M.M., Romanov, V.G., Shishatskii, S.P., 1986. Ill-posed problems of mathematical
physics and analysis. Translations of Math. Monographs v. 64. Amer. Math. Soc,
Providence.
Martyshko, P.S., Koksharov, D.E., 2005. On density determination in the layered medium
by the gravity data (in Russian) Geophysical Journal 27 (4), 678–684.
Martyshko, P.S., Prutkin, I.L., 2003. Technology of depth distribution of gravity field
sources (in Russian) Geophysical Journal 25 (3), 159–168.
Novikov, P.S., 1938. Sur le problème inverse du potentiel. Doklady Akademii Nauk SSSR
18, 165–168.
Pohánka, V., 2001. Application of the harmonic inversion method to the Kolárovo gravity
anomaly. Contributions to Geophysics and Geodesy 31 (4), 603–620.
Prutkin, I., 2008. Gravitational and magnetic models of core–mantle boundary and
their correlation. Journal of Geodynamics 45 (2–3), 146–153.
Prutkin, I., Casten, U., 2009. Efficient gravity data inversion for 3D topography of a contact
surface with application to the Hellenic subduction zone. Computers & Geosciences
35 (2), 225–233.
Prutkin, I., Saleh, A., 2009. Gravity and magnetic data inversion for 3D topography of the
Moho discontinuity in the northern Red Sea area, Egypt. Journal of Geodynamics 47
(5), 237–245.
Sitárová, A., Bielik, M., Burda, M., 1984. Interpretation of the Kolárovo gravity anomaly
(in Slovak) Geologické Práce, Správy 81, 171–182.
Vajda, P., Bielik, M., Pohánka, V., 2002. Testing the application of the Truncation Filtering
Methodology in interpreting real gravity data: the Kolárovo gravity anomaly. Contri-
butions to Geophysics and Geodesy 32 (1), 57–66.