Journal of Geophysics and Engineering

J. Geophys. Eng. 14 (2017) 303–315 (13pp) https://doi.org/10.1088/1742-2140/aa5832

Basement depth estimation from gravity

anomalies: two 2.5D approaches coupled
with the exponential density contrast model
V Chakravarthi1, K Mallesh and B Ramamma
Centre for Earth and Space Sciences, University of Hyderabad, Gachibowli, Hyderabad—500046,

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

Telangana, India

E-mail: vcvarthi@rediffmail.com, malleshkamal08@gmail.com and ramageophd@gmail.com

Received 20 July 2016, revised 23 November 2016

Accepted for publication 10 January 2017
Published 13 February 2017

Abstract
We develop two automatic techniques in the spatial domain using the exponential density
contrast model(EDCM) to trace the bottom surface of a 2.5D sedimentary basin from the
observed gravity anomalies. The interface between the sediments and basement is described with
a finite strike polygonal source, whose depth ordinates become the unknown parameters to be
estimated. The proposed automatic modeling technique makes use of the forward difference
approximation and the inversion solves a system of normal equations using the ridge regression
to estimate the unknown parameters. Furthermore, the proposed inversion technique
simultaneously estimates the regional gravity background that is associated with the residual
gravity anomaly. In either case, forward modeling is realized in the spatial domain through a
method that combines both analytical and numerical approaches. The utility of each algorithm
was successfully tested on a theoretically produced noisy residual gravity dataset. The validity of
the inversion technique is also exemplified with the noisy gravity anomalies attributable to a
synthetic structure in the presence of regional gravity background. We demonstrate that the
magnitude of gravity anomaly is offset dependent and that it would influence the modeling
result. Additionally, some applications with real gravity datasets from the Gediz and Büyük
Menderes grabens in western Turkey using the derived EDCMs have produced geologically
reasonable results which are in close agreement with those reported previously.

Keywords: gravity anomalies, basement interfaces, 2.5D, modeling, inversion, exponential

density contrast model
(Some figures may appear in colour only in the online journal)

Introduction anomalies can be analyzed using appropriate modeling and

inversion schemes to decipher the basement configurations.
One of the major applications of the gravity method is to deci- Undoubtedly, a thorough knowledge of the density of
pher the concealed basement geometries under the sedimentary rocks both at the surface and subsurface is always essential
load. In thick sedimentary basins electric and electromagnetic for applying a few gravity corrections to the gravity mea-
methods cannot provide information at sufficient depth, whereas, surements and also to obtain reliable geologic interpretations
gravity data yields valuable information on the basement (Mar- of the gravity anomalies. However, in case of sedimentary
telet et al 2013). Usually, negative gravity anomalies are rocks the density is rarely uniform (Kinsman 1975,
observed over the sedimentary basins because the density of Rusakov 1990, Ramillien and Wright 2002, Braitenberg
sedimentary rocks is less than the surrounding basement. These et al 2006, Klinger et al 2011, Azab and El-Khadragy 2013).
Athy (1930) was the first researcher to express the rela-
1
Author to whom any correspondence should be addressed. tions between the depth of burial and the density, porosity,

1742-2132/17/020303+13$33.00 303 © 2017 Sinopec Geophysical Research Institute Printed in the UK

J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

and compaction of different types of sediment by exponential calculate the gravity anomalies of sedimentary basins, Bhas-
equations. Manger (1963) had documented large amount of kara Rao et al (1993) developed a few graphical methods to
measured data pertaining to the porosity and density mea- analyze the Fourier transforms of the gravity anomalies of
surements of sedimentary rocks and concluded that the por- simple 2D geometric models. Bhaskara Rao and Mohan Rao
osity of sandstones generally decreases whereas the density (1999) proposed a method based on the Bott’s (1960)
increases with depth of burial and age. From an extensive approach to analyze the gravity anomalies of 2D sedimentary
study and analysis of the density logs measured in 435 deep basins, where forward modeling was performed in the spectral
wells in Western Canada, Maxant (1980) had shown that the domain followed by their transformation to the space domain
density-depth relationship seldom follows a linear trend; and by Filon’s (1928) method. Chai and Hinze (1988) also cal-
in the case of shale, the correlation between density and depth culated anomalies of prismatic bodies in the wave number
could be explained efficiently by exponential density-depth domain and converted the anomalies back to the spatial
relationship. Cowie and Karner (1990) have demonstrated domain by applying a shift-sampling technique. Chappel and
from the measured density-depth data of different strati- Kusznir (2008) derived wave number domain expressions to
graphic units in sedimentary basins that the sediment densities calculate the gravity anomalies of sedimentary basins with
exhibit a wide range but the mean density clearly increases irregular bounding surfaces. All the above-mentioned tech-

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

with depth with the highest rate in the top few hundred niques, except the ones reported by Granser (1987) and Chai
meters. Castagna et al (1993) have demonstrated that the and Hinze (1988), are 2D. The methods proposed by Granser
measured densities for shale as a function of depth showed (1987) and Chai and Hinze (1988) are applicable to analyze
more or less similar behavior, although the samples were either profile data or two dimensional data. Nevertheless,
collected from a wide variety of locations with different truncation errors would crop up in all the enlisted methods
geologic settings and histories. Based on the analysis of the when the gravity anomalies transform back to the spatial
density samples taken from 716 drill sites of the Deep Sea domain from frequency domain (Chakravarthi and Sundar-
Drilling Project, Tenzer and Gladkikh (2014) showed that the arajan 2007, Chakravarthi et al 2016).
density increases nonlinearly with the increasing sediment In some cases, sedimentary basins that are formed par-
depth due to compaction. ticularly due to strike slip motions (for e.g., pull-apart basins)
Cai and Zhdanov (2015) argue in line with Cordell more often posses finite strike lengths; hence, the anomalous
(1973) that the density-depth relationship of sedimentary mass needs to be approximated by limited/finite strike
rocks, in general, does not strictly follows any mathematical models to analyze the gravity anomalies produced by them.
formulation because of the influence of several geologic Nevertheless, 3D models are more expensive in terms of both
factors such as compaction, stratigraphic layering, cementa- data requirement and computational time than 2D; therefore,
tion, facies change, diagenesis etc. From the actual mea- use of 2.5 dimensionality (2.5D) in the quantitative inter-
surement of sedimentary rock density from deep bore holes, pretation of gravity anomalies is justified.
Cordell (1973) had established that the density contrast In this context, Rama Rao and Radhakrishna Murthy
decreases drastically at shallow depths and less progressively (1989) have developed two forward modeling schemes cou-
at deeper depths following an exponential law. Although pled with relevant codes to compute the gravity anomalies of
several mathematical functions are in use to describe the 2D and 2.5D polygonal sources using uniform or linear or
density contrast variation of sedimentary rocks with depth exponential density variations. To accommodate linear or
viz., linear (Pedersen 1985, Reamer and Ferguson 1989, exponential density variation they proposed subdivision of
Hansen 1999, Hamayun et al 2009, D’Urso 2014), quadratic each side of the polygon into segments for e.g., 1 for the
(Bhaskara Rao 1986, 1990, Gallardo-Delgado et al 2003), linear density and 10 for the exponential density. This tech-
cubic polynomial (Garcia-Abdeslem 2005); each one has its nique, though efficient, invariably consumes significant
own limitations in its application as demonstrated by Chak- amount of time (Visweswara Rao et al 1994, Chakravarthi
ravarthi and Sundararajan (2006) and Chakravarthi (2009). et al 2016). Mickus and Peeples (1992) have devised a
Hence, the choice of exponential density contrast model technique based on the inverse theory of Backus and Gilbert
(EDCM) in the analysis of gravity anomalies of sedimentary (1967, 1968, 1970) to trace the bottom surface of a 2.5D
basins is more appropriate to obtain reliable interpretations. sedimentary basin from the observed gravity and magnetic
However, the major intricacy associated with the EDCM fields. However, this technique finds limited practical appli-
is that no closed form analytical expressions for the gravity cation particularly in cases where the geological settings
anomalies could be derivable in the spatial domain for for- warrant the use of variable density. In recent past, a 2.5D
ward modeling (Radhakrishna Murthy 1998, Chakravarthi inversion technique coupled with EDCM to estimate the
and Sundararajan 2004). Owing to this difficulty, methods basement depths from the observed gravity anomalies was
have been proposed in the spectral domain to realize forward developed by Chakravarthi et al (2013); wherein the sedi-
modeling. For e.g., Cordell (1973) had developed a recursive mentary pile above the basement was described with a collage
method combining both the gravity field and its vertical of vertical prisms having finite but variable strike lengths.
derivative (determined by convolution in discrete Fourier However, this technique is difficult to implement if the
series) to solve the structure of a sedimentary basin from the observed gravity anomalies are available at random intervals
observed gravity anomalies, Granser (1987) proposed a for- along a profile. On the other hand, the interpretation algo-
ward modeling algorithm based on Taylor series expansion to rithms of Chakravarthi et al (2016) are free from the above

304
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

throughout the volume of the basin given by

Dr (z) z dx dydz
Dg2.5D (0, r , 0) = G òv (x 2 + y - r 2 + z 2 ) 3 /2
, (1 )

where G is universal gravitational constant, (x, y, z ) are

source coordinates, dx dy dz is the volume of an element
within the source, and r is the offset of the profile, AA¢ , from
the origin O (0, 0, 0). Dr (z ) represents EDCM at any depth z
within the structure given by Cordell (1973)

Dr (z) = Dr 0 e-lz . (2 )

Here Dr0 is the density contrast observed at the ground

surface and l is a decay factor expressed in reciprocal length
units. Upon substitution equation (2) and performing partial

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

Figure 1. Schematic representation of the depth structure of a finite integration with respect to y, equation (1) becomes
strike sedimentary basin (solid line in black) and its approximation
by a 2.5D finite strike polygonal source (solid line in blue). The z e-lz
finite strike, 2Y, of the basin prevents one to represent it by a 2D
Dg2.5D (0, r , 0) = G Dr 0 òs (x 2 + z 2)
source. Here, r, is the offset distance of a selected profile, AA′.
⎡
Y-r
´⎢
⎢⎣ (x 2 + Y - r 2 + z 2)
constraint, but again these are 2D. Above all, the enlisted
methods are strictly valid for residual gravity anomalies ⎤
Y+r ⎥ dx dz ,
alone. + (3 )
In the present paper, two automatic techniques in the (x 2 + Y + r 2 + z 2) ⎥⎦
spatial domain are developed; one based on the principles of
automatic modeling and the other on inversion (Chakravarthi where s stands for surface integration. Applying Stokes’
et al 2016) to estimate the basement depths from a set of theorem, equation (3) takes the form
randomly distributed gravity anomaly data at any profile
offset considering (i) finite strike length for a sedimentary Dg2.5D (0, r , 0) = G Dr 0 ∮z e-lz
basin, and (ii) exponential density contrast variation within ⎡
xY-r
the sedimentary pile. The proposed automatic modeling ´ ⎢arctan
technique uses the forward difference approximation and the ⎢⎣ z (x + Y - r 2 + z 2 )
2

inversion technique solves the system of normal equations to ⎤

estimate the unknown parameters. In case of automatic xY+r ⎥ dz .
+ arctan
modeling, the anomalies attributable to a synthetic model in z (x 2 + Y + r 2 + z 2) ⎥⎦
(4 )
the presence of pseudorandom noise are analyzed and in case
of inversion the noisy anomalies are analyzed both with and Approximating the outline of the basin by a polygon CDEK
without regional gravity background. In either case the esti- (shown as a solid line in blue in figure 1), the x term in
mated parameters are compared with the assumed ones. To equation (4) can be expressed for the Jth side, such as CD as
demonstrate the practical applicability of the proposed tech-
niques three real field gravity anomaly profiles from western x = a + z coti , (5 )
Turkey are interpreted and judged against the interpretations
previously reported (Sari and Şalk 2002). where a = xj - zj coti and i is the angle made by the side CD
with the x-axis. Here, (xj , zj ) are the coordinates of the vertex
C. Upon substitution of equation (5) in equation (4) the
gravity effect of the Jth side of the polygon can be obtained as
Forward modeling—theoretical considerations z j + 1 -l z
DgCD (0, r , 0) = G Dr 0 ∮ e
zj
Figure 1 shows the depth structure of a typical finite strike ⎡
(a + z coti) Y - r
sedimentary basin in the xz-plane of the Cartesian co-ordinate ´ ⎢arctan
system. The basin possesses uniform cross-section throughout ⎢⎣ z (a + z coti2 + Y - r 2 + z 2)
its strike length (2Y) along the y-axis perpendicular to the xz- ⎤
(a + z coti) Y + r ⎥ dz .
plane. Let the z-axis be positive vertically downwards along + arctan (6 )
which the depth dimension of the basin is scaled and x-axis z (a + z coti2 + Y + r 2 + z 2) ⎥⎦
runs transverse to the strike of the basin. The gravity anomaly
of the basin at any observation, O′(0, r, 0), on the xy-plane can Here zj + 1 is the depth ordinate of the vertex D, whose coor-
be obtained by integrating the gravity effect of an element dinate is represented by xj + 1. The total gravity effect of the

305
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

model, the geometry of which is shown in figure 2(b). In this

case, we presume the length of the model as 40 km along the
strike (half-strike length is 20 km) with 32 irregularly spaced
vertices describe the geometry of the undulating density
interface (figure 2(b)).
Presuming that the sedimentary basin is homogeneous
throughout its volume with a density contrast of −0.32 gm cm−3,
the residual gravity anomalies calculated from 32 unequally
spaced observations in the interval xj Є [0 km, 64 km] using
equation (7) (shown as solid dots in figure 2(a)) are compared
with the anomalies obtained from an analytical solution (shown
as a solid line in figure 2(a)) of Radhakrishna Murthy (1998). It is
to note that the x-coordinates of the vertices of the polygon in this
case do not coincide with the observer locations on the profile.
The fact that the error between the two anomalies barely exceeds

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

±1 × 10−3 mGal confirms the reliability and accuracy of the
proposed numerical method.
Further, figure 2(c) shows the gravity anomalies pro-
duced by a structure (figure 2(d)) along two selected profiles
at two different offsets 0 and 18 km; one each in the interval
xj Є [0 km, 62 km]. It is to note that the depth structure shown
in figure 2(d) is exactly the same as the one shown in
figure 2(b) but in this case the density contrast within the
structure obeys exponential decrease with depth (figure 2(e)).
The surface density contrast and the decay constant of EDCM
(equation (2)) are presumed as −0.32 gm cm−3 and 0.3 km−1
respectively. In this case, distances to the observer locations
Figure 2. (a) Gravity anomalies calculated at zero offset by analytical
on the profile form the x ‐coordinates of the vertices of the
(Radhakrishna Murthy 1998) and proposed methods using uniform polygon. The magnitudes of the gravity anomalies calculated
density contrast, (b) geometry of a finite strike sedimentary basin, (c) at two different offsets are different over the lengths of the
gravity anomalies at 0 and 18 km offsets with exponential decrease profiles (figure 2(c)), though the structure remains the same
in density contrast, (d) geometry of a basin in which the color (figure 2(d)). For e.g., the model produces a maximum gravity
gradation from yellow to red indicates the increase in density of
sediments with depth, (e) prescribed exponential density contrast
anomaly (absolute) of 22 mGal at the 22nd km on the profile
model used in forward modeling. at zero offset; whereas it is 19 mGal at 18 km offset
(figure 2(c)). It is to realize that even a 3 mGal difference in
the anomaly would seriously affect the interpretation if
basin at the observation, O¢ (0, r , 0), can be obtained as appropriate offset is not chosen in the analysis.
N
Dg2.5D (0, r , 0) = å Dgj (0, r , 0) , (7 )
j=1 Analysis of gravity anomalies
where N stands for the number of sides of the polygon. It is to
note that closed form analytic solution does not exist for In general, interpretation of gravity anomalies is carried out
equation (6) in the spatial domain; hence, necessity arises to by specifying a set of approximate depths to the density
solve the equation by a numerical approach. The vertices of interface supplemented from known geologic information/
the anomalous source are covered sequentially in the clock- drilling and/or other geophysical inputs. The gravity response
wise direction. of the structure is then calculated with a suitable forward
In the presence of regional background, Ψ, the gravity modeling algorithm and compared with the observed anom-
anomaly of the basin can be represented as aly. The difference between the two anomalies is minimized
Dgtotal (0, r , 0) = Dg2.5D (0, r , 0) + Y , (8 ) by adjusting the depths to the interface within the permissible
limits based on some convergence criteria. For optimum
where depth estimates the model gravity response should mimic the
N1 observed response and the corresponding estimated structure
Y= åfi x i . (9 ) is geologically sensible.
i=0 If the profile along which the interpretation is intended
Here N1 stands for the degree of polynomial and x is the covers the lateral dimensions of a sedimentary basin com-
observer location on the profile. pletely then the relief of the basin at the first and last obser-
We demonstrate the validity of the proposed numerical vations become zero. If the distances to observer locations on
method of anomaly calculation on a finite strike synthetic a profile and the x ‐coordinates of the vertices of the polygon

306
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

are the same then one needs to estimate (N - 2) depth parameters to be estimated from a given set of observed data
parameters from N observed anomalies. The present auto- to a number less than or equal to the number of observations.
matic modeling and inversion algorithms initiate the structure For this reason, the regional background in this case needs to
of a sedimentary basin presuming that the observed gravity be presumed either constant throughout the length of the
anomaly at each observation on the profile is being produced profile or would vary linearly as a function of observer
by an infinitely extending horizontal slab in which the density location. In the present inversion technique we have assumed
contrast decreases with depth following equation (2). that the regional background varies linearly along the profile
According to Cordell (1973), the thickness of such a slab can for which case the number of unknown parameters to be
be estimated as estimated becomes N, being equal to the number of obser-
⎛ vations. In case higher order polynomials are required to
-1 lg B i ⎞
z Bi = log ⎜1 - ⎟, (10) simulate the regional background, the number of depth
l ⎝ 2pG Dr 0 ⎠ parameters to be solved shall be cut down appropriately by
where, gBi is the gravity anomaly observed at any station, means of using known basement depths as constraints in the
i (xi ). Because the width of a sedimentary basin is always inversion.
finite across its strike, the depth (thickness) estimates realized To start with the coefficients f0 and f1 of the polynomial

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

from equation (10) are only approximate; thereby, the model (equation (9)) are set to zero and subsequently updated
gravity anomalies calculated from equations (6) to (7) iteratively along with the depth parameters of the basement.
obviously differ in magnitude from the observed anomalies, The difference between the observed (Dgobs (xi )) and model
Dgobs (xi ). The difference between the observed, Dgobs (xi ), gravity anomaly (Dgtotal (xi )) at any observation, xi , at the end
of kth iteration can be expressed as
and model gravity response, Dg2.5D (xi ), at all observations
can be quantified by a root mean square error (Chakravarthi N-1
¶Dg (xi )
et al 2016) defined by Dgobs (xi ) - Dgtotal (xi ) = å ¶zi
dzi + df0 xi + df1 .
j=2
(14)
å i= 1 [Dgobs (xi) - Dg2.5D (xi )]2 .
N

Jrms = (11) The N normal equations that are constructed and solved to
Nobs
estimate the improvements in N unknown parameters by
Here N = Nobs. In case of automatic modeling, the basin minimizing the rms error (equation (11)) using the ridge
depths are improved based on the forward difference regression algorithm (Marquardt 1970) can be expressed in a
approximation matrix form as

z Bi k+ 1 = z Bi k + Dz Bi k , (12) (A + dI ) X = B, (15)
where A is a N ´ N matrix, whose elements Anj ¢ are
where
given by
-1 ⎛ l [Dgobs (xi ) - Dg2.5D (xi )] ⎞
Dz Bi k = log ⎜⎜1 - ⎟⎟ . (13)
N N
¶Dg (xm ) ¶Dg (xm )
l ⎝ 2pG Dr 0 e –lz Bi
k
⎠
Anj ¢ = åå ¶a j ¢ ¶a n
, j ¢ = 1, 2, ¼, N , (16)
n = 1m = 1

Here k represents the number of iterations. It is to note that in X = da n , (17)

case of automatic modeling the gravity anomalies attributable
solely to a sedimentary basin are used to estimate the base- N
¶Dg (xm )
ment depths, whereas in case of inversion the residual gravity
B= å [Dgobs (xm) - Dgtotal (xm )] ¶a j ¢
,
m=1
anomalies may be associated with regional background
j ¢ = 1, 2, ¼, N . (18)
(Radhakrishna Murthy 1998, Chakravarthi et al 2013). The
proposed inversion algorithm estimates the regional back- Here, a n = zn, n = 1, 2, ...,N - 2,
ground in addition to the depth parameters of the interface.
The modified depths obtained from equation (13) are used to aN - 1 = f0
update the gravity response of the structure and a new rms and
error, Jrms1 is calculated. If the magnitude of Jrms1 is less than
aN = f1 .
Jrms then the value of Jrms1 is assigned to Jrms and z Bi k + 1 to
z Bi k and the process continues till (i) the specified number of Also, d is the damping factor and I is a diagonal matrix
iterations completed, or (ii) the resulting rms error becomes containing the diagonal elements of the matrix A. The partial
less than the predefined allowable error, or (iii) the new rms derivatives required in equations (16) and (18) are evaluated
error exceeds its preceding value. numerically following the procedure described by Chakra-
In case of inversion, the number of unknown parameters varthi et al (2013). The application of the ridge regression
to be estimated from N (=Nobs) observed anomalies is algorithm is detailed by Chakravarthi (2003) and Chakra-
(N - 2) + N 1 + 1. Although one may choose any degree varthi et al (2016). The estimated improvements in the
of polynomial (equation (9)) to describe the regional back- unknown parameters, da n , solved from equation (15) are
ground and then to realize forward modeling using added to/subtracted from the existing parameters, a n ,
equation (8), it is always necessary to restrict the unknown n = 1, 2, ... , N and the inversion process continues for the

307
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

Table 1. Assumed and estimated coefficients of linear regional

background, synthetic example.
Assumed Estimated
Offset
(km) f0(mGal) f1(mGal km−1) f0(mGal) f1(mGal km−1)
18 −0.33 −0.023 −0.429 −0.016
15 −0.365 −0.015
0 −0.302 −0.0143

setting different values to the offset parameter independently

(18, 15 and 0 km) to study its effect on the interpretation, if
any. In a real case, the strike length of a basin needs to be

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

established either from existing geological and/or Bouguer
anomaly maps. The offset of the profile along which inter-
pretation is intended can be easily found. The values of Dr0
and l of EDCM (equation (2)) remain unchanged during the
process of analysis in both cases.
When the offset was set to 18 km, the modeling algo-
rithm has performed 15 iterations before it got terminated. For
the same anomaly and chosen offset the inversion took 12
iterations. The rms error (equation (11)) for the starting model
in either case was 2.09 mGal. In case of modeling, the error
Figure 3. (a) Observed and theoretical anomalies by automatic
was drastically reduced to 0.13 mGal at the end of the 3rd
modeling and inversion at different offsets in the presence of iteration, beyond which it showed a progressive decay with
pseudorandom noise, (b) assumed and estimated structures by the iteration number (figure 3(c)). On the other hand, in case
modeling and inversion for different offsets, (c) changes in rms of inversion the decay of rms error was found to be rather
errors with iteration for different offsets.
gradual when compared to modeling. The optimum estimated
structure corresponding to an rms error of 0.05 mGal
specified number of iterations or until one of the following (figure 3(c)) between the observed and modeled gravity
criteria is fulfilled anomalies was obtained at the end of the 15th iteration in case
of modeling and 12th iteration in case of inversion
(i) the resulting misfit becomes less than the predefined (figure 3(b)). The theoretical gravity anomalies obtained from
allowable error, or both modeling and inversion are shown in figure 3(a) with the
(ii) the resulting damping factor attains an unusually large corresponding estimated structures in figure 3(b). One can
value (Chakravarthi 2003). notice from figure 3(b) that the estimated structure from
modeling exactly mimics the structure obtained from
inversion.
On the other hand, when the algorithms are applied on
Applications
the same anomaly with 15 km offset; modeling took 10
The applicability of each proposed interpretation methodol- iterations and inversion 12 iterations for a proper conv-
ogy is established by analyzing the gravity anomalies pro- ergence. On the other hand, for 0 km offset modeling per-
duced by a synthetic structure before being applied to analyze formed 7 iterations and inversion 11 iterations. Beyond the
the real world gravity anomalies. concluding iterations the modeling technique was terminated
because the resulting rms error in each case exceeds its pre-
ceding value, whereas in case of inversion the damping factor
Synthetic example
attained a large value. The changes in rms errors associated
The gravity anomalies shown in figure 2(c), which are pro- with the improvements in model space at both offsets from
duced by a structure at 18 km offset, are treated as the modeling are shown in figure 3(c). The theoretical gravity
observed anomalies for modeling purpose. These anomalies anomalies and the estimated depth structures from modeling
are further corrupted with pseudorandom noise (Gaussian) for s = 0 km and s = 18 km offsets are shown in figures 3(a)
having zero mean and a standard deviation of 0.1 mGal before and (b) for comparison. The model gravity responses and
the analysis attempted. The noisy anomalies of the structure inferred depth structures realized from the inversion technique
are shown in figure 3(a) as a solid line in black. Automatic for both the offsets (0 and 15 km) are exactly the same as the
modeling and inversion techniques, described in the text, are ones obtained from modeling; hence are not shown in
applied on these noisy anomalies to recover the structure by figures 3(a) and (b) for brevity.

308
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

Figure 4. (a) Observed and theoretical anomalies by inversion at different offsets in the presence of both regional background and
pseudorandom noise, (b) assumed and estimated structures by inversion for different offsets, (c) changes in rms errors with iteration for
different offsets, (d) error (%) between the estimated and assumed depths for different assumed offsets.

Figure 5. Faulting along Gediz graben, western Anatolia. Main faults are shown in thick lines with solid ticks on downthrown side; minor
faults are in thinner lines with open ticks on downthrown side (reproduced from Sari and Şalk 2002, Chakravarthi and Sundararajan 2007).
Gravity anomalies along the Profiles 2 and 3 are interpreted.

It can be seen from figure 3(a) that the theoretical gravity becomes more and more underestimated (figure 3(b)). A few
responses realized from both modeling and inversion at dif- insignificant deviations in the estimated structures are inevi-
ferent offsets equally explain the observed anomaly but the table (around 30th km) even considering an appropriate offset
inferred depth structures are dissimilar (figure 3(b)). The of 18 km in the interpretation, however, such deviations can
structures deciphered by setting the offset to 18 km closely be ignored because the anomalies used in the analysis are
mimic the assumed one, where as the other two structures noisy.
obtained with zero and 15 km offsets show undue deviations. To study the combined effect of both regional back-
It is observed that as the magnitude of the offset decreases the ground and pseudorandom noise on the interpretation,
resulted structure obtained from either modeling or inversion regional background described with a set of predefined

309
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

Figure 6. Faulting along Büyük Menderes graben, western Anatolia. Main faults are shown in thick lines with solid ticks on downthrown
side; minor faults are in dashed lines (reproduced from Sari and Şalk 2002). Gravity anomalies along the Profile XY are interpreted.

Figure 7. Derived hyperbolic (reproduced from Sari and Şalk 2002) and exponential density contrast models, Gediz and Büyük Menderes
grabens, western Turkey.

310
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

Figure 9. (a) Observed and theoretical gravity anomalies by
Figure 8. (a) Observed and theoretical gravity anomalies by
automatic modeling and inversion along Profile 3, Gediz graben,
automatic modeling and inversion along Profile 2, Gediz graben,
western Turkey. Estimated regional is also shown, (b) inferred depth
western Turkey. Estimated regional is also shown, (b) inferred depth
structures of the graben by automatic modeling and inversion.
structures of the graben by automatic modeling and inversion.
Estimated structures by Sari and Şalk (2002), and Chakravarthi and
Estimated structure by Sari and Şalk (2002) based on 2D modeling is
Sundararajan (2007) are also shown for comparison, (c) changes in
also shown for comparison, (c) changes in rms errors with iteration.
rms errors with iteration.

coefficients (table 1), by setting N1 = 1 in equation (9), is Field example

added to the noisy anomalies shown in figure 3(a) and sub-
The Gediz and Büyük Menderes are two prominent finite
sequently inverted to recover the basement structure. The
strike (150 km) east–west trending onshore grabens in the
assumed regional field and the regional associated noisy
western Turkey (Eyidogan and Jackson 1985, Paton 1992,
anomalies are shown in figure 4(a).
Sari and Şalk 2002). These grabens, bounded by normal fault
The analysis was performed on the anomalies indepen-
systems, are filled with thick-sectioned sediments above the
dently, as in the previous case, by setting the offsets to 18 km,
metamorphic basement complex. The master fault bounding
15 km and 0 km respectively. In each case, the inversion
the Büyük Menderes graben is on the northern side, whereas
algorithm has performed four iterations and then terminated
the master fault associated with the Gediz graben is towards
because the resulting rms error attained a value less than the the south (figures 5 and 6).
predefined allowable error (figure 4(c)). The theoretical Paton (1992) had analyzed the gravity anomalies of the
gravity anomalies at the end of the concluding iteration clo- two grabens along four selected transects; three across the
sely mimic the observed anomalies in all cases (figure 4(a)); Gediz graben and one over the Büyük Menderes graben
whereas the corresponding estimated structures show devia- respectively. These interpretations were carried out with an
tions from each other (figure 4(b)). assumption that the sedimentary fill within the grabens is
The deciphered structure with 18 km offset was slightly uniform, which however is not so in reality (Sari and
overestimated in the range xj Є [30 km, 55 km] with a max- Şalk 2002). Based on the borehole data, Sari and Şalk (2002)
imum error of 3.3% found at the 54th km (figure 4(d)). On the described the density contrast variation of the sedimentary
other hand, when the offset was set to 15 and 0 km the rocks within the grabens by hyperbolic functions (figure 7)
algorithm in each case has yielded a structure that was and used them in the analyses of two gravity profiles across
underestimated considerably over the length of each profile. the Gediz graben (figures 8 and 9) and one over the Büyük
With 15 km offset a maximum error of 14.2% in the estimated Menderes graben (figure 10). However, the interpretations
depth was found and with zero offset 18.2% respectively, carried out by both Paton (1992) and Sari and Şalk (2002) are
both observed at 19.5 km (figure 4(d)). Furthermore, the based on 2D approaches. While justifying the need to
predicted regional background by setting the offset to 18 km approximate the grabens to 2.5D sources, Chakravarthi and
(coefficients of the polynomial given in table 1) closely Sundararajan (2007) have analyzed the gravity anomalies of
matches with the assumed regional, whereas the other the Gediz graben for its basement structure along a profile
regional fields estimated with 15 and 0 km offsets do not presuming the sediment fill above the basement as collage of
(figure 4(a)). vertical prisms (figure 9(b)).

311
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

Table 2. Derived hyperbolic (Sari and Şalk 2002) and exponential density contrast models, Gediz and Büyük Menderes grabens, western
Turkey.
Hyperbolic density contrast model Exponential density contrast model (EDCM)
−3
Name of the graben/Profile No. Δρ0 (gm cm ) β (km) Δρ0 (gm cm−3) λ (km−1)
Gediz/2 −1.407 0.859 −1.182 1.019
Gediz/3 −1.407 0.620 −1.320 1.6708
Büyük Menderes/XY −0.98 2.597 −0.798 0.3808

Table 3. Rms errors for initial and estimated models by automatic

modeling and inversion, Gediz and Büyük Menderes grabens,
western Turkey.
Rms error (mGal)

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

Automatic modeling Inversion
Name of the graben/
Profile No. Initial Final Initial Final
Gediz/2 1.59 0.95 1.59 0.07
Gediz/3 0.67 0.02 0.67 0.04
Büyük Menderes/XY 1.54 0.23 1.54 0.14

Table 4. Estimated coefficients of linear regional background, Gediz

and Büyük Menderes grabens, western Turkey.
Name of the graben/Profile No. f0 f1
Gediz/2 −4.02 0.110
Gediz/3 0.06 −0.002
Büyük Menderes/XY 0.23 0.004

Figure 10. (a) Observed and theoretical gravity anomalies by

Büyük Menderes graben, respectively. The modeling and
automatic modeling and inversion along Profile XY, Büyük inversion techniques got terminated at the end of the said
Menderes graben, western Turkey. Estimated regional is also shown, concluding iterations because one of the termination criteria
(b) inferred depth structures of the graben by automatic modeling in each case was fulfilled.
and inversion. Estimated structure by Sari and Şalk (2002) based on The magnitudes of the initial and final rms errors
2D modeling is also shown for comparison, (c) changes in rms errors
with iteration. (equation (11)) between the observed and modeled gravity
anomalies in each case are given in table 3 and the changes in
In the present case, the gravity profiles considered by Sari rms errors with iteration are shown graphically in figures 8(c),
and Şalk (2002) are re-interpreted by the proposed algorithms 9(c) and 10(c).
using the derived EDCMs of respective grabens. The con- The gravity anomalies subsequent to the analysis are
stants of EDCMs (Dr0 and l ) obtained by fitting equation (2) shown in figures 8(a)–10(a) along with the observed ones. In
to the hyperbolic density contrast models (Sari and Şalk 2002) all cases, the nature of fit between the observed and model
are given in table 2 and shown in figure 7. gravity anomalies is satisfactory (figures 8(a)–10(a)). The
The interpreted results of the three gravity profiles by the coefficients of regional gravity background estimated from
proposed techniques are shown in figures 8–10. It is to note the inversion for each one of the profile are given in table 4
that these profiles do not bisect the strike lengths of respective and shown in respective figures from 8(a) to 10(a).
grabens but run at different offsets, hence the offsets of the The recovered basement structures along the three pro-
profiles measured from the geologic maps (Sari and files from automatic modeling comply well with the structures
Şalk 2002) are used in the interpretation for reliable results. deciphered from inversion (figures 8(b)–10(b)). The max-
The modeling technique has performed 23 iterations for a imum depths to the basement estimated from the present
proper convergence of the anomalies along the Profile 2 and analyses and the ones reported by Sari and Şalk (2002) along
26 iterations in each case for the Profiles 3 and XY respec- the three selected profiles under consideration are given in
tively. For the same set of gravity anomalies when inversion table 5.
has applied it took 6 and 4 iterations for the Profiles 2 and 3 of By and large, the deciphered structural models of the two
the Gediz graben, and 79 iterations for the Profile XY of the grabens from the present analyses (figures 8(b), 9(b) and

312
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

Table 5. Maximum depths to the basement estimated from gravity modeling and inversion, Gediz and Büyük Menderes grabens, western
Turkey.
Name of the graben/ Estimated depth from Estimated depth from Estimated depth by Sari and
Profile No. Distance (km) modeling (km) inversion (km) Şalk (2002) (km)
Gediz/2 16.0 1.91 1.75 2.07
Gediz/3 46.0 1.65 1.60 1.75
Büyük Menderes/XY 24.0 2.26 2.32 2.60

10(b)) compare reasonably well with those reported by Sari further the offset deviates from the optimum offset the more
and Şalk (2002), however, with a few exceptions. For e.g., all the final solution deviates.
the structural models derived from the present analyses (from The derived exponential density contrast models for
both modeling and inversion) divulge (i) relatively deeper Gediz and Büyük Menderes grabens in western Turkey are
basement over the shoulders (except for the Profile XY on the used to analyze the observed gravity anomalies of respective

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

northwestern part) and (ii) relatively shallower basement in grabens. The estimated structures of the two grabens from
the depocentres compared to the ones reported by Sari and present modeling and inversion show moderate deviations
Şalk (2002) (table 5 and figures 8(b)–10(b)). The interpreted from the structures previously reported by Sari and Şalk
depth structure of the Gediz graben along the Profile 3 by the (2002) which are based on the interpretation of gravity
stacked prism model (Chakravarthi and Sundararajan 2007) is anomalies using a 2D approach with hyperbolic density
also shown in figure 9(b) for comparison. A maximum depth functions.
of 1.65 km to the basement estimated at the 46th km on the
Profile 3 from the present interpretation remarkably coincides
with a figure of 1.64 km inferred by Chakravarthi and Sun-
dararajan (2007). Conclusion

Two interpretation techniques, based on the principles of

automatic modeling and inversion, using the exponential
Discussion density contrast model are developed in the spatial domain to
trace the basement structures of finite strike sedimentary
The applicability of the proposed automatic modeling and basins from the observed gravity anomalies. The sediment-
inversion is exemplified with a synthetic example and also on basement interface is described by a finite-strike polygonal
a few real field gravity anomalies. In case of the synthetic source with several vertices, whose depth ordinates become
example, pseudorandom noise added to the gravity anomalies the unknown parameters to be estimated. Unlike the case with
produced by a residual source (sedimentary basin) at 18 km modeling, the inversion technique simultaneously estimates
offset (treated as observed anomalies) are analyzed by both regional trend described by a linear equation. The proposed
automatic modeling and inversion to recover the structure; interpretation techniques are automatic as they generate the
setting the offset parameter to 0, 15 and 18 km independently. initial parameters from the observed gravity anomalies, esti-
At the end of the concluding iteration the theoretical gravity mates the improvements in corresponding parameters and
anomalies in each case equally fit the observed noisy modifies them in an iterative approach following the criteria
anomalies but the estimated depth structures differ from each of minimization of rms error between the observed and
other. The structures deciphered from both automatic mod- modeled gravity anomalies. Forward modeling is realized
eling and inversion with 18 km offset are exactly the same through an approach that combines both analytical and
and explained well the assumed structure even in the presence numerical methods, because no closed form analytical gravity
of pseudorandom noise. On the other hand, even though the expression could be derivable in the spatial domain using an
estimated structures from automatic modeling correlate well exponential density contrast model. The reliability of the
with the corresponding structures obtained from inversion for proposed forward modeling is established by comparing the
different offsets (0 km and 15 km); these structures do not anomalies realized from the present method with those
comply well with the assumed one. anomalies obtained from an analytical method. It is demon-
In the presence of both regional background and pseu- strated in detail with a synthetic example that the magnitude
dorandom noise the inversion technique by setting 18 km of gravity anomalies over a finite strike geologic structure is
offset has yielded a structure that is more or less consistent offset dependent and that this parameter plays a vital role in
with the assumed structure with a few marginal errors. the interpretation.
However, the inversions performed on the gravity anomalies The advantage of the proposed techniques is that they are
with 15 and 0 km offsets fail to recover the structure. In a fairly applicable to analyze the gravity anomalies even when
nutshell, it is demonstrated that in both automatic modeling the profile fails to bisect the strike length of a sedimentary
and inversion the final solution was offset dependent as the basin.

313
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

In the proposed interpretation methodologies it is pre- Chakravarthi V, Pramod Kumar M, Ramamma B and
sumed that the density of underlying basement below the Rajeswara Sastry S 2016 Automatic gravity modeling of
sedimentary column is uniform, and that the regional back- sedimentary basins by means of polygonal source geometry
and exponential density contrast variation: two space domain
ground along a profile varies linearly as a function of observer based algorithms J. Appl. Geophys. 124 54–61
location across the strike in case of inversion. However, these Chakravarthi V, Rajeswara Sastry S and Ramamma B 2013
assumptions may or may not be valid elsewhere. Therefore, MODTOHAFSD—a GUI based JAVA code for gravity
the strategies presented here yield reliable interpretations analysis of strike limited sedimentary basins by means of
growing bodies with exponential density contrast–depth
where the enlisted assumptions are more or less valid. variation: a space domain approach Comput. Geosci. 56
131–41
Chakravarthi V and Sundararajan N 2004 Ridge regression
algorithm for gravity inversion of fault structures with variable
References density Geophysics 69 1394–404
Chakravarthi V and Sundararajan N 2006 Discussion on the
gravitational attraction of a right rectangular prism with density
Athy L F 1930 Density, porosity and compaction of sedimentary varying with depth following a cubic polynomial Geophysics
rocks Bull. Am. Astron. Pet. Geol. 14 1–24 71 X17–9

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

Azab A and El-Khadragy A A 2013 2.5D gravity/magnetic model Chakravarthi V and Sundararajan N 2007 3D gravity inversion of
studies in Sahl El Qaa area, Southwestern Sinai, Egypt Pure basement relief—a depth dependent density approach
Appl. Geophys. 170 2207–29 Geophysics 72 I23–32
Backus G E and Gilbert F J 1967 Numerical applications of a Chappell A and Kusznir N 2008 An algorithm to calculate the
formalism for geophysical inverse problems Geophys. J. R. gravity anomaly of sedimentary basins with exponential
Astron. Soc. 13 247–76 density-depth relationships Geophys. Prospect. 56 249–58
Backus G E and Gilbert F J 1968 The resolving power of gross earth Cordell L 1973 Gravity anomalies using an exponential density-
data Geophys. J. R. Astron. Soc. 16 169–205 depth function-San Jacinto Graben, California Geophysics 38
Backus G E and Gilbert F J 1970 Uniqueness in the inversion of 684–90
inaccurate gross earth data Phil. Trans. R. Soc. A 226 123–92 Cowie P A and Karner G D 1990 Gravity effect of sediment
Bhaskara Rao D 1986 Modeling of sedimentary basins from gravity compaction: examples from the North Sea and Rhine Graben
anomalies with variable density contrast Geophys. J. R. Astron. Earth Planet. Sci. Lett. 99 141–53
Soc. 84 207–12 D’Urso M G 2014 Gravity effects of polyhedral bodies with linearly
Bhaskara Rao D 1990 Analysis of gravity anomalies of sedimentary varying density Celest. Mech. Dyn. Astron. 120 349–72
basins by an asymmetrical trapezoidal model with quadratic Eyidogan H and Jackson J A 1985 A seismological study of normal
density function Geophysics 55 226–31 faulting in the Demirci, Alasehir and Gediz earthquakes of
Bhaskara Rao D and Mohan Rao C P V N J 1999 Two-dimensional 1969–1970 in W Turkey: implications for the nature and
interpretation of gravity anomalies over sedimentary basins geometry of deformation in the continental crust Geophys. J.
with an exponential decrease of density contrast with depth R. Astron. Soc. 81 569–607
Proc. Indian Academy of Sciences-Earth and Planetary Filon L N G 1928 On a quadrature formula for trigonometric
Sciences vol 108, pp 99–106 integrals Proc. R. Soc. Edin. 49 38–47
Bhaskara Rao D, Prakash M J and Ramesh Babu N 1993 Gravity Gallardo-Delgado L A, Perez-Flores M A and Gomez-Trevino E
interpretation using Fourier transforms and simple geometrical 2003 A versatile algorithm for joint inversion of gravity and
models with exponential density contrast Geophysics 58 magnetic data Geophysics 68 949–59
1074–83 Garcia-Abdeslem J 2005 The gravitational attraction of a right
Bott M H P 1960 The use of rapid digital computing methods for rectangular prism with density varying with depth following a
direct gravity interpretation of sedimentary basins Geophys. J. cubic polynomial Geophysics 70 j39–42
R. Astron. Soc. 3 63–7 Granser H 1987 Three-dimensional interpretation of gravity data
Braitenberg C, Wienecke S and Wang Y 2006 Basement structures from sedimentary basins using an exponential density-depth
from satellite-derived gravity field: South China Sea ridge function Geophys. Prospect. 35 1030–41
J. Geophys. Res.—Solid Earth 111 B05407 Hamayun K, Prutkin I and Tenzer R 2009 The optimum
Cai H and Zhdanov M 2015 Application of Cauchy-type integrals in expression for the gravitational potential of polyhedral bodies
developing effective methods for depth-to-basement inversion having a linearly varying density distribution J. Geod. 83
of gravity and gravity gradiometry data Geophysics 80 G81–94 1163–70
Castagna J P, Batzle M L and Kan T K 1993 Rock physics—the link Hansen R O 1999 An analytical expression for the gravity field of a
between rock properties and AVO response Offset-Dependent polyhedral body with linearly varying density Geophysics 64
Reflectivity—Theory and Practice of AVO Analysis 75–7
(Investigations in Geophysics Series 8) ed P Castagna and Kinsmann D J J 1975 Rift basins and peculiarities of sediment
M M Backus (Tulsa, OK: Society of Exploration accumulation in the conditions of sagged margins of continents
Geophysicists) pp 124–57 Petroleum and Global Tectonics ed G Fischer and S Judson
Chai Y and Hinze W J 1988 Gravity inversion of an interface above (Princeton, NJ: Princeton University Press) pp 61–91
which the density contrast varies exponentially with depth Klinger F L, Nacif S, Martinez M P, Gimenez M E, Ruiz F and
Geophysics 53 837–45 Alvarez O 2011 Gravimetric model of the Gastre trough,
Chakravarthi V 2003 Digitally implemented method for automatic province of Chubut Argentina: Boletín Geológico y Minero
optimization of gravity fields obtained from three-dimensional 122 299–310
density interfaces using depth dependent density US Patent Manger G E 1963 Porosity and bulk density of sedimentary rocks,
6,615,139 Contributions to Geochemistry, U.S. Atomic Energy
Chakravarthi V 2009 Automatic gravity inversion for simultaneous Commission US Geological Survey Bulletin 1144-E E1–56
estimation of model parameters and regional gravity Marquardt D W 1970 Generalized inverses, ridge regression, biased
background: an application to 2D pull-apart basins Curr. Sci. linear estimation, and nonlinear estimation Technometrics 12
96 1349–60 591–612

314
J. Geophys. Eng. 14 (2017) 303 V Chakravarthi et al

Martelet G, Perrin J, Truffert C and Deparis J 2013 Fast mapping of of finite and infinite strike length with the density contrast
magnetic basement depth, structure and nature using differing with depth Comput. Geosci. 15 1265–77
aeromagnetic and gravity data: combined methods and their Ramillien G and Wright I C 2002 Sea mount gravity anomaly
application in the Paris Basin Geophys. Prospect. 61 857–73 modeling with variably thick sediment cover Mar. Geophys.
Maxant J 1980 Variation of density with rock type, depth, and Res. 23 13–23
formation in the Western Canada basin from density logs Reamer S K and Ferguson J F 1989 Regularized two-dimensional
Geophysics 45 1061–76 Fourier gravity inversion method with application to the Silent
Mickus K L and Peeples W J 1992 Inversion of gravity and magnetic Canyon caldera, Nevada Geophysics 54 486–96
data for the lower surface of a two and one-half dimensional Rusakov O M 1990 The thickness and density of sedimentary cover
sedimentary basin Geophys. Prospect. 40 171–93 in the indian ocean Geophys. J. 3 390–9
Paton S 1992 Active normal faulting, drainage patterns and Sari C and Şalk M 2002 Analysis of gravity anomalies with
sedimentation in southwestern Turkey J. Geol. Soc. 149 hyperbolic density contrast: an application to the gravity
1031–44 data of Western Anatolia J. Balkan Geophys. Soc. 5
Pedersen L B 1985 The gravity and magnetic fields from ellipsoidal 87–96
bodies in the wave number domain Geophys. Prospect. 33 Tenzer R and Gladkikh V 2014 Assessment of density variations of
263–81 marine sediments with ocean and sediment depths Sci. World
Radhakrishna Murthy I V 1998 Gravity and Magnetic Interpretation J. 2014 823296
in Exploration Geophysics (Memoir 40) (Bengaluru: Visweswara Rao C, Chakravarthi V and Raju M L 1994 Forward

Downloaded from https://academic.oup.com/jge/article/14/2/303/5106843 by guest on 25 March 2022

Geological Society of India) modelling: gravity anom-alies of two-dimensional bodies of
Rama Rao P and Radhakrishna Murthy I V 1989 Two fortran77 arbitrary shape with hyperbolic and parabolic density functions
function subprograms to calculate gravity anomalies of bodies Comput. Geosci. 20 873–80

315