GEOPHYSICS, VOL. 58, NO. 11 (NOVEMBER 1993); P. 1588--1595, 6 FIGS., 3 TABLES.

Downloaded 06/02/13 to 155.198.30.43. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Advantages of using the vertical gradient of

gravity for 3·D interpretation

I. Marson* and E. E. Klingele:J:

discontinuities and depth of the first interface are known.

ABSTRACT The use of derivatives of the gravity field, as described in
this paper, may be helpful in reducing the possible number of
Gravity gradiometric data or gravity data trans-
unknown geometrical parameters (Depth and horizontal
formed into vertical gradient can be efficiently pro- location of comers and edges for examples).
cessed in three dimensions for delineating density Cordell and Grauch (1985) presented a method to locate
discontinuities. Model studies, performed with the abrupt changes in density or magnetization of subsurface
combined use of maxima of analytic signal and of rocks by using the position of maxima of horizontal gradient
horizontal gradient and the Euler deconvolution tech- of magnetic or gravity anomalies. Their method was auto-
niques on the gravity field and its vertical gradient, mated by Blakely and Simpson (1986) and applied, among
demonstrate the superiority of the latter in locating others, to isostatic residual anomaly data of the United
density contrasts. Particularly in the case of interfering States. The joint use of analytic signal and horizontal gradi-
anomalies, where the use of gravity alone fails, the ent maxima for three-dimensional (3-D) gravity interpreta-
gravity gradient is able to provide useful information tion (mainly identification of density boundaries and the
with satisfactory accuracy. direction of dip of geological contacts) has been first dis-
cussed by Hansen et al. (1987). Finally, Reid et al. (1990), in
an interesting paper about the 3-D application of Euler
INTRODUCTION deconvolution to magnetic data, briefly discussed the appli-
cation of this technique to gravity data. The aim of this paper
Over the last few years increasing interest has been is to demonstrate the superior effectiveness of jointly using
expressed for measurements of gravity gradients by aircraft the above mentioned methods on the vertical gravity gradi-
or satellites [ESA, Aristoteles-Project; Gravity, Gradiome- ent, as compared to the gravity anomaly itself, to derive
ter Survey System (GGSS) Experiment]. To avoid the trans- information about vertical and horizontal position of density
formation of large amounts of gradiometric data into the changes. Synthetic examples given in this paper indicate that
vertical component of gravity and also to avoid the loss of the use of vertical gradient of gravity is more accurate than
accuracy, it seems reasonable to develop or to adapt algo- the use of gravity itself, mainly in the case of sharp discon-
rithms based on the direct use of the measured data. On the tinuities, shallow sources, and interfering anomalies. There-
other hand, conventional gravimetry is being applied more fore, when the vertical gradient is not measured directly, its
and more to delineate very shallow structures or cavities numerical computation from the gravity anomalies is recom-
where interfering sources are the major obstacle to a reliable mended.
interpretation. Because the vertical gravity gradient is more
sensitive than gravity itself to geologic structures and its METHOD
interpretation is less influenced by neighboring disturbing Analytic signal
bodies, it can be ideal data for interpretation. The availabil-
ity of efficient tools for the transformation of gravity anom- The basic concepts of the analytic signal in two dimen-
alies into their gradient expression opens the way for an sions for magnetic data were extensively discussed by
extensive use of the gravity gradient. During quantitative Nabighian (1972, 1974) and Green and Stanley (1975). Their
interpretation, the computation can be significantly reduced counterparts, in the case of gravity gradiometric data, have
if some parameters such as location of abrupt density been introduced by Klingele et al. (1991). According to

Manuscript received by the EditorDecember 31, 1991; revised manuscript received December 21, 1992.
*Department of Naval Architecture, Marine and Environmental Engineering, University of Trieste, Via Valerio 10, 1-34127 Trieste, Italy.
+Institut fur Geodasie und Photogrammetrie, ETH-Honggerberg, CH-8093 Zurich, Switzerland.
© 1993 Society of Exploration Geophysicists. All rights reserved.

1588
 Vertical Gradient for 3·0 Interpretation 1589
Nabighian (1984), one can define the analytic signal of the tion (3) the function! can be replaced by the vertical gradient
vertical gradient of gravity produced by a 3-D source as of gravity.
follows: Equation (3) has a more general meaning if one considers
that a given perturbing body can be represented by an
a2g a2g a2g appropriate distribution of point sources on the body's upper
A(x,y)=-+--j-. (I a)
axaz ayaz az 2 surface. In this case N is a "structural index" (Thompson,
1982), directly related to the particular distribution of point
For the field itself it can be written as:
Downloaded 06/02/13 to 155.198.30.43. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

sources. Since N can be considered as unknown, using

ag ag ag equation (3) does not require any assumption of a geologic
A (x,y)=-+--j-. (Ib) model. In practice, it is more convenient to find the solution
9 ax ay az for given N's and choose the set that is the most consistent,
The corresponding amplitudes for these expressions are both in terms of clustering and of uncertainties. Theoreti-
given by cally the value of N, for an infinite step in the value of g, is
equal to unity (Reid et al., 1990). From our experiences with

IA(x, y)1 = \j1(~)2

axaz + (~)2
2g)2 noisy synthetic data, a value around -0.5 gives better
ayaz + (aaz2 , (Ie) results. Similarly, the value of N for ag/az is around zero
(0.05). Therefore the only unknowns in equation (3) are xo,

I A,(x, Y)I " ~ (::r (::r (::r

+ + (Id)
Yo, and zo. The use and advantages of Euler's homogeneity
equation in two and three dimensions for automatic depth
estimates from magnetic data have been discussed by
Thompson (1982) and Reid et al. (1990), respectively, while
As in the two-dimensional (2-D) case (Nabighian 1972), the
its use in two dimensions for computer-assisted interpreta-
amplitude of the analytic signal can be used to locate the
tion of vertical gravity gradient data has been discussed by
horizontal position, as well as the depth of the corners and
Klingele et al. (1991). The application of equation (3) to
edges of the 3-D body. Since the horizontal coordinates of
gridded gravity and vertical gravity gradient data is straight-
the corners are related in a well defined way to the positions
forward, once the derivatives have been computed.
of the maxima of the signal amplitude (its maxima occur
It requires the choice of a suitable square (or rectangular,
directly over the edges of the source bodies), the use of the
in case of elongated anomalies) window (3 x 3 or greater) of
analytic signal for locating corners and edges of perturbing
gridded data on which a set of linear equations is formed. Of
bodies in the horizontal plane is quite efficient. Green (1976)
course the size of the square window depends on the grid
discussed the properties of the analytical signal for 2-D
spacing and the anomaly characteristics. Generally speak-
gravity data analysis. Hansen et al. (1987) demonstrated the
ing, good results can be achieved with a window of 9 x 9,
usefulness of this technique for the gravity field. Roest et al.
which produces 81 linear equations from which the three
(1992) used the same approach for solving the 3-D case in
unknowns and their uncertainties can be computed. The
magnetic interpretation. Since vertical gravity gradient and
window is moved to those areas that contain the maxima of
total field magnetic anomaly are related through Poisson's
the amplitude of the analytic signal or of the horizontal
equation, the same method can be applied to ag/az as well.
derivative. This process is repeated for different values of N.
The location of amplitude maxima of the analytic signal are
A rejection criterion, based on the uncertainty of the solu-
determined by the method of Blakely and Simpson (1986)
tion, is established, and the accepted solutions are plotted on
applied to the amplitude of the analytic signal of g and its
the (x, y) plane (see also Thompson, 1982). The solutions
vertical derivative. These are plotted on the (x, y) plane,
are rejected if the standard deviation of the three unknowns
and this information is used to assist the interpretation by
(xo, Yo, zo), evaluated from the covariance matrix of the
means of Euler's homogeneity equation.
linear system, is bigger than, say, 5 to 15 percent of the
Euler's homogeneity equation calculated depth (see also Reid et al., 1990). The optimal
choice of the structural index and the threshold of the
With reference to Thompson (1982), let f(x, y, z) be a rejection value is made according to the clustering of the
function in Cartesian coordinates with z positive downward solutions in the x, y plane. The combination of an objective
and the plane of observation horizontal at Z = o. f (x, y, z) rejection criterion (standard deviation) and a subjective one
is homogeneous of degree N if (clustering of solutions) greatly speeds up the interpretation
process still allowing the interpreter to use, to a certain
f itx , ty , tz) = tNf(x, y, z). (2) extent, his own judgement.
In this case, Euler's homogeneity equation is valid, giving The use of equations (1), (2), and (3) requires the compu-
tation of both horizontal and vertical derivatives. The hori-
af af af zontal derivatives were efficiently computed in the space
(x - xo) . - + (y - Yo) • - + (z - zo) . - = Nf, (3) domain by means of a standard seven-point formula, while
ax ay az
the vertical one were computed in the frequency domain by
where xo, Yo, and Zo are the coordinates of the source. means of the well-known Fourier transform procedure
It can easily be proven (Klingele et al., 1991) that the (Gunn, 1975). It is worth noting that, when the vertical
vertical gradient of the gravity effect of a point mass is a gradient has not been measured directly, it can easily be
homogeneous function of degree -3, while for g itself the computed from the gravity field itself with just one additional
function is homogeneous of degree -2. Therefore in equa- transformation.
 1590 Marson and Klingele
These computations and data manipulations introduce a the analytic signal, computed from the gravity signal and
certain degree of high-frequency noise, which must be from the vertical gradient, respectively (Figures 1c and 1d),
suppressed by filtering procedures. The most important demonstrates the higher resolving power of the latter. On the
requirement for successful application of these methods is other hand, the maxima of the horizontal gradient computed
high-quality and high-resolution gravity and gravity gradio- from the field and from the first vertical derivative give good
metric data. and comparable results (Figures le and It) while the Euler
solutions obtained from the gravity field are significantly
misplaced both in the X-, y-plane and in the z-direction
Downloaded 06/02/13 to 155.198.30.43. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

SYNTHETIC EXAMPLES (Figure 2a). Those obtained from the vertical gradient delin-

In the following, we will discuss some synthetic examples.

The gravity effects of simple geological structures were
computed using rectangular prisms by means of the expres- Table 1. Parameters of the test models: 1. Single prism; 2.1,
sion developed by Nagy (1966). To better simulate the most 2.2 Double prism; 3.1, 3.2 Transform fault. the co-ordinates
are in grid units and the rotation angle in degrees (positive
general case, when only the gravity field is available, the counterclockwise). All the structures have a constant density
vertical gradient was computed through the Fourier trans- contrasts 0.1 glcm-3.
form instead of using an analytical expression. Coordinates
are given in grid units to generalize the scale of the problem. Top Bottom Center Size a
Model grid units grid units grid units grid units deg
The test examples simulate geological situations in terms of
structures and scale of the problem, which are not easily 1 2 100 46/43 20 25
solved by means of inversion of the gravity field as it is. 2.1 2 100 60/60 20 0
Gravity and vertical gradient fields, produced by a single 2.2 5 100 40/40 20 0
rectangular prism (model 1) with parameters given in Table 1, 3.1 6 100 -96/88 200 25
are shown in Figures 1a and lb. Analysis of the maxima of 3.2 2 100 161-81 200 25

FIG. 1. Results obtained with model 1 of Table 1. The plotted area is 101 x 101 grid units. The figure shows: (a) the gravity field;
(b) the vertical gradient of g; (c) the maxima of the analytic signal of g and (d) of the vertical gradient; (e) the maxima of the
horizontal derivatives of g and (t) of the vertical gradient. Units are mGal for g and mGal/(grid units) for the vertical gradient
ofg.
 Vertical Gradient for 3·D Interpretation 1591
eate quite well the corners and edges of the top surface of the
prism (Figure 2b). In this example, the solutions obtained
with Euler deconvolution of the vertical gradient have been
subjected to quite a flexible rejection criterion (the solutions
a) b)
with standard deviations higher than 10 percent of the
calculated depth were rejected). It is possible to obtain
solutions extremely well localized at the four corners in
Downloaded 06/02/13 to 155.198.30.43. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

>',' .: :: ,'/. space (error not higher than 5 percent) using the vertical
" ... ,' ,', ." gradient of gravity and increasing the threshold of the
•• 1
.',
,.::
. , ", rejection criterion (the rejection criterion is restrictive when
:....
••••• I • ','

. it rejects solutions with standard deviation higher than 5

0' \ • • ,'. •
" " "
. '. ',:"':",
percent of the calculated depth). The results are summarized
. " in Tables 2 and 3.
The problem of solving for the source parameters in the
case of interfering anomalies has been tackled in the second
example. In this case, two prisms of the same size have one
corner at the same location but at different depths. Figures
3a and 3b show the gravity field and the vertical gradient
produced by this model (Table I, Model 2). In this case, it is
also difficultto locate, from the maxima of the analytic signal
of the gravity field and the positions of the prisms, while it is
easier from the maxima of the analytic signal of the vertical
FIG. 2. (a) Three-dimensional plot of the solutions of the gradient (Figures 3c and 3d). The upper prism is placed in the
Euler's deconvolution of g and (b) of the vertical gradient of right position, while the deeper one is not. The analysis can
model I of Table 1. be considerably improved by using the maxima of the
horizontal gradient computed from g and its vertical gradient
(Figures 3e and 3D. Just like the analytic signal case, no
realistic Euler solutions can be obtained using the gravity

Table 2. Results obtained from the analysis of the vertical gradient of g using a restrictive rejection criterion. The coordinates are
in grid units and the rotation angle in degrees (positive counterclockwise). All the structures have a constant density contrast of
0.1 g/cm -3.

Number Mean
Real depth Window of depth Quality of
Models (grid units) N (grid units) solutions (grid units) cr results

Simple prism 2 0.05 5 x 5 49 1.94 0.112 Excellent

Double prism
Shallower prism 2 0.05 5x 5 69 1.98 0.248 Excellent
Deeper prism 5 0.05 5x 5 24 4.42 0.219 Good
Common point 5 0.05 5x 5 20 3.26 0.565 Fair(?)
Transform fault
Shallower prism
Shallower prism 2 0.05 9 x 3 13 2.54 0.28 Good
Deeper prism 6 0.05 12 x 3 66 7.32 2.32 Fair
Common point 6 0.05 9x 9 28 5.14 1.04 Fair

Table 3. Results obtained from the analysis of g alone using a restrictive rejection criterion. The coordinates are in grid units and
the rotation angle in degrees (positive counterclockwise). All the structures have a constant density contrast of 0.1 g/cm-3.

Real depth Window Number of Mean depth Quality of

Models (grid units) N (grid units) solutions (grid units) c results

Simple prism 2 -0.45 5 x 5 129 7.8 3.3 Bad

Double prism
Shallower prism 2 -0.45 5x 5 75 5.2 0.9 Bad
Deeper prism 5 -0.45 5x 5 56 6.9 2.0 Poor
Common point 5 -0.45 5x 5 0
Transform fault
Shallower prism 2 -0.45 9x 3 33 11 1.3 Bad
Deeper prism 6 -0.45 12 x 3 } Solutions misplaced in x and y
Common point 6 -0.45 9 x 9 No acceptable depth
 1592 Marson and Kllngele

a)
Downloaded 06/02/13 to 155.198.30.43. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

c)
"''\.
d) r'
.i
'0: e) f)

,, ,...
. . •

" ......
,
....J. -.-
....
••••
i
I
.. -,
.
'
..f.·· ..••",
..s-; .... ,..
-\t •
,.1 ,»'
\....... ,
.,.. ". ,...
I···.....,.
~.,

•••• ~.
. J

FIG. 3. Results obtained with model 2 of Table 1. The plotted area is 101 x 101 grid units. The figure shows: (a) the gravity field;
(b) the vertical gradient of g; (c) the maxima of the analytic signal of g and (d) of the vertical gradient; (e) the maxima of the
horizontal derivatives of g and (1) of the vertical gradient. Units are mGal for g and mGal/(grid units) for the vertical gradient
of g.

a) b)

.:r-,
~ .
I' ••••

:.: .
::::............ .,

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

'. ..,
.: ... ',', .
• '0:'
..... .. :..
.:. ~ . .' ,',

..... ,', . .....

FIG. 4. (a) Three-dimensional plot of the solutions of the Euler's deconvolution of g and (b) of the vertical
gradient of model 2 of Table 1.
 Vertical Gradient for 3·D Interpretation 1593

a) b)
Downloaded 06/02/13 to 155.198.30.43. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

d) n' ~------------,
c) ............
......
--,', ..
..,.,. ..
:~:::::
......
• r, " ••

.. : .
," I.,',
.:1:::,',
:l>:
......
:::::
......
...... .
'I"
:11=
.......
..... ... II'
J :•
•• • '
••••• II
• • 41
•••••••••••• .)::. . ....
-t.....
::: .. ..I~';'::::: :: ~ :::. :! t: . . •
'I':~: :: :~::::: : : : ::::.
................ _..-
•
........••••...••.. "
II

. '"I •

.................... ",'",'
:.:.:.:.:.... . ," .
...•...•......•. _. ••. .:
I •••••••••••••••••••••

• • • • • • • • • • • I"
, .. ...:.:.:.:::::....
..... •••••••• ••• J;:.
",
:.~....
;;;;:.~~~:::=:==::.:.:.:.:.:
. . ::.;.:s~#...,.••..':
,

e) f)

-.
FIG. 5. Results obtained with model 3 of Table I. The plotted area is 101 x 101 grid units. The figure shows: (a) the gravity field;
(b) the vertical gradient of g; (c) the maxima of the analytic signal of g and (d) of the vertical gradient; (e) the maxima of the
horizontal derivatives of g and (0 of the vertical gradient. Units are mGal for g and mGaU(grid units) for the vertical gradient
of g.
 1594 Marson and Kllngele

field (Table 3, and Figure 4a). The plot of the solutions of one (edge). In practice, it would be preferable to use 2- or
Euler's equation for ag/az is shown in Figure 4b. The three 3-D deconvolution depending on the pattern of the measured
outermost comers of the shallower prism are resolved with field.
the same accuracy as in the first example (5 percent with
restrictive rejection criterion) whereas those of the deeper
CONCLUSIONS
one are resolved with an accuracy of about 10 percent (with
restrictive rejection criterion). In the case of the comers with
the same horizontal coordinates, it is interesting to note that The use of the vertical gradient, even when computed
Downloaded 06/02/13 to 155.198.30.43. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the depth obtained is located between the real depths of the from the gravity field, has definite advantages. It is possible
upper and the lower comer. to obtain realistic solutions for the geometry including depth
In the third example, two rectangular prisms are com- of the gravity source and to locate density changes in the
bined, as described in Table 1, to simulate a transform fault. horizontal plane. The use of the gravity itself can also
The field and its first vertical derivative produced by this accomplish the latter. However, in case of interfering anom-
model are shown in Figures 5a and 5b. The distribution of alies, the resolving power of the gravity field is smaller than
the maxima of the analytic signal of g, depicted in Figure 5c, that of the vertical gradient.
poorly delineates the edges of the perturbing body. A better The 3-D method described in this paper is best applied to
definition is obtained with the analytic signal of the vertical actual 3-D structures. In the case of a 2-D structure, it is
gradient (Figure 5d) and with the distribution of the maxima crucial to avoid parallelism between the grid and the struc-
of the horizontal gradient of g (Figure 5e) and of ag/az ture direction, because the computation of the derivative
(Figure 5t). parallel to the rows (or columns) of the grid will produce
Among these three, the best result is given by the hori- numerical instabilities.
zontal derivative of ag/az. The higher resolving power of the The choice of an efficient rejection criterion is of great
vertical gradient is evident from the comparison of the importance since erratic solutions are always present. In our
solutions obtained from applying the Euler's deconvolution approach a solution is accepted if its uncertainty is smaller
method to g with those obtained applying the same method than a given percentage of the solution itself. An overly
to ag/az (Figures 6a and 6b; Tables 2 and 3). In the case of severe rejection criterion applied to the depth coordinates
the maxima of both the analytic signal and the horizontal may reduce the number of available solutions and reduce the
gradient of ag/az, the mislocation of the contact in the upper delineating power of the method.
left part of the grid is not caused by the method itself but by A warning is necessary concerning the use of filtering
boundary effects in the computation of the vertical gradient. techniques. While it is true that a mild filter is necessary to
It is also worth remarking that this test model contains both avoid numerical instabilities (high-frequency noise) during
a pure 3-D structure (triple point and comer) and a more 2-D the transformations, it is also true that a drastic filter will

FIG. 6. Three-dimensional plot of the solutions of the Euler's deconvolution of g (a) and of the vertical
gradient (b) of model 3 of Table I.
 Vertical Gradient for 3·0 Interpretation 1595

significantly alter the signals. A careful choice of the filter source bodies from magnetic or gravity anomalies: Geophysics,
parameters is necessary. 51, 1494-1498.
Cordell, L., and Grauch, V. J. S., 1985, Mappingbasement magne-
The discontinuity analysis, using the distribution of the tizationzones from aeromagnetic data in the San Juan basin, New
maxima of the analytic signal and of the horizontal gradient, Mexico, in Hinze, W. J., Ed., The utility of regional gravity and
which has been proven to be quite effective, can serve to magnetic anomaly maps: Soc. Expl. Geophys., 181-197.
Green, R., and Stanley, J. M., 1975, Application of a Hilbert
delimit the area over which the Euler deconvolution should
transform method to the interpretation of surface-vehicle mag-
be applied. However, the window can also be moved over netic data: Geophys. Prosp., 23, 18-27.
Downloaded 06/02/13 to 155.198.30.43. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the entire grid to obtain all the solutions compatible with the Green, R., 1976, Accurate determination of the dip angle of a
rejection criterion. This approach seems to be quite efficient geological contact usingthe gravity method: Geophys. Prosp., 24,
265-272.
since the total time required to deconvolve a 100 x 100 grid Gunn, P. J., 1975, Linear transformations of gravity and magnetic
does not exceed a few seconds on a personal computer. fields: Geophys. Prosp., 23, 300-312.
Moreover, since the horizontal coordinates are better re- Hansen, R. 0., Pawlowski, R. S., and Wang, X., 1987, Joint use of
analytic signal and amplitude of horizontal gradient maxima for
solved than the depth, the distribution of the solutions in the three-dimensional gravity data interpretation: 57th Ann. Internat.
horizontal plane can be used for the discontinuity analysis. Mtg., Soc. Explo. Geophys., Expanded Abstract, 100-102.
Tools based on the derivatives of the gravity field for Klingele, E. E., Marson, 1., and Kahke, H. G., 1991, Automatic
interpretation of gravity gradiometric data in two dimensions:
automatic interpretation can be helpful in several circum- Vertical gradient: Geophys. Prosp., 39, 407-434.
stances, mainly when the scale of the problem does not Nabighian, M. N., 1972, The analytic signal of two-dimensional
permit a suitable use of the gravity field itself. For examples: bodies with polygonal cross-section: Its properties and use for
automated anomaly interpretation: Geophysics, 37, 507-517.
- - 1974, Additional comments on the analytic signal of two-
1) When the survey area doesn't cover the whole anom- dimensional magnetic bodies with polygonal cross-section: Geo-
aly. In this case only the available part of the anomaly physics, 39, 85-92.
can be used for the computation of the first vertical - - 1984, Toward a three-dimensional automatic interpretation
of potential field data via generalized Hilbert transforms: Funda-
derivative and consequently for the interpretation. mental relations: Geophysics, 49, 780-786.
2) In case of superposition of two or more anomalies or in Nagy, D., 1966, The gravitational attraction of a right rectangular
prism: Geophysics, 31, 362-371.
the presence of a linear regional field, the vertical Reid, A. B., Allsop, 1. M., Granser, H., Millet, A. J., and Somerton,
gradient has a superior power of resolution and permits 1. W., 1990, Magnetic interpretation in three dimensions using
a better separation of the solutions. Euler deconvolution: Geophysics, 55, 80-91.
Roest, W. R., Verhoef, J., and Pilkington, M., 1992, Magnetic
interpretation using the 3-D analytic signal: Geophysics, 57,
REFERENCES 116-125.
Thompson, D. T., 1982, EULDEPH: A new technique for making
ARISTOTELES, Phase: A study: Dornier Tech. Rep., Vol 2, 1988. computer-assisted depth estimates from magnetic data: Geophys-
Blakely, R. J., and Simpson, R. W., 1986, Approximating edges of ics, 47, 31-37.