J Geod

DOI 10.1007/s00190-011-0533-4

ORIGINAL ARTICLE

Spherical harmonic modelling to ultra-high degree

of Bouguer and isostatic anomalies
G. Balmino · N. Vales · S. Bonvalot · A. Briais

Received: 1 July 2011 / Accepted: 14 November 2011

© Springer-Verlag 2011

Abstract The availability of high-resolution global digital the reference ellipsoid’s surface to the Earth’s surface was
elevation data sets has raised a growing interest in the fea- performed by high-order Taylor expansion with all grids of
sibility of obtaining their spherical harmonic representation required partial derivatives being computed in parallel. The
at matching resolution, and from there in the modelling of main application was the production of a 1 × 1 equian-
induced gravity perturbations. We have therefore estimated gular global Bouguer anomaly grid which was computed
spherical Bouguer and Airy isostatic anomalies whose spher- by spherical harmonic analysis of the Earth’s topography–
ical harmonic models are derived from the Earth’s topogra- bathymetry ETOPO1 data set up to degree and order 10,800,
phy harmonic expansion. These spherical anomalies differ taking into account the precise boundaries and densities of
from the classical planar ones and may be used in the con- major lakes and inner seas, with their own altitude, polar caps
text of new applications. We succeeded in meeting a num- with bedrock information, and land areas below sea level.
ber of challenges to build spherical harmonic models with The harmonic coefficients for each entity were derived by
no theoretical limitation on the resolution. A specific algo- analyzing the corresponding ETOPO1 part, and free surface
rithm was developed to enable the computation of associ- data when required, at one arc minute resolution. The fol-
ated Legendre functions to any degree and order. It was lowing approximations were made: the land, ocean and ice
successfully tested up to degree 32,400. All analyses and cap gravity spherical harmonic coefficients were computed
syntheses were performed, in 64 bits arithmetic and with up to the third degree of the altitude, and the harmonics of
semi-empirical control of the significant terms to prevent the other, smaller parts up to the second degree. Their sum
from calculus underflows and overflows, according to IEEE constitutes what we call ETOPG1, the Earth’s TOPography
limitations, also in preserving the speed of a specific regu- derived Gravity model at 1 resolution (half-wavelength). The
lar grid processing scheme. Finally, the continuation from EGM2008 gravity field model and ETOPG1 were then used
to rigorously compute 1 × 1 point values of surface grav-
G. Balmino (B) · N. Vales ity anomalies and disturbances, respectively, worldwide, at
CNES, Groupe de Recherches de Geodesie Spatiale, the real Earth’s surface, i.e. at the lower limit of the atmo-
Geosciences Environnement Toulouse, Observatoire sphere. The disturbance grid is the most interesting product
Midi-Pyrenees, 14, Avenue Edouard Belin, 31400,
of this study and can be used in various contexts. The surface
Toulouse, France
e-mail: [email protected] gravity anomaly grid is an accurate product associated with
EGM2008 and ETOPO1, but its gravity information contents
S. Bonvalot are those of EGM2008. Our method was validated by com-
IRD, Bureau Gravimétrique International,
parison with a direct numerical integration approach applied
Geosciences Environnement Toulouse, Observatoire
Midi-Pyrenees, 14, Avenue Edouard Belin, 31400, to a test area in Morocco–South of Spain (Kuhn, private
Toulouse, France communication 2011) and the agreement was satisfactory.
Finally isostatic corrections according to the Airy model, but
A. Briais
in spherical geometry, with harmonic coefficients derived
CNRS, Bureau Gravimétrique International,
Geosciences Environnement Toulouse, Observatoire Midi-Pyrenees, from the sets of the ETOPO1 different parts, were computed
14, Avenue Edouard Belin, 31400, Toulouse, France with a uniform depth of compensation of 30 km. The new

123
 G. Balmino et al.

world Bouguer and isostatic gravity maps and grids here them here since our approach is very different: it introduces
produced will be made available through the Commission a new type of anomaly and uses radically different methods.
for the Geological Map of the World. Since gravity values The present study was triggered by the goal of computing
are those of the EGM2008 model, geophysical interpretation a global Bouguer map, and by the work of Kuhn et al. (2009),
from these products should not be done for spatial scales who define and compute what they call complete spherical
below 5 arc minutes (half-wavelength). Bouguer gravity anomalies (ΔgCSB ) using spherical terrain
corrections over the whole Earth with respect to a local but
Keywords Bouguer gravity anomalies · Isostatic gravity full spherical Bouguer shell. Such an anomaly, at any gravity
anomalies · Earth’s topography · Spherical harmonics · observation point P of altitude H , involves: (i) the gravita-
Surface gravity anomalies · Surface gravity perturbations tional effect of the Bouguer shell of constant thickness H and
density ρ, that is 4πGρ H (twice the value of the usual plateau
term—G being the gravitational constant); and (ii) the spher-
ical terrain correction with respect to the shell, computed
1 Introduction over the whole planet by numerical integration over spheri-
cal volume elements having a size which increases with the
The availability of high-resolution global digital elevation distance to point P. Then the free air correction, atmospheric
data sets has raised possibility to address in best conditions correction and normal gravity (at the reference ellipsoid sur-
the basic question we want to answer: what is the Earth’s face) are used to achieve the computation of ΔgCSB . Kuhn
gravitational potential, or gravity field when: (i) the planet is et al. (2009) have computed high-resolution Bouguer gravity
stripped of all masses above a reference surface (geoid), and anomalies over Australia only, but it was acknowledged that
(ii) mass deficiencies below the reference surface are restored the same principles can be applied globally.
to a given density. Our goal has been to do a similar computation but in one
This residual field then contains information on sub-sur- step and, with some approximation, by replacing the integra-
face density variations. tion stage by the use of spherical harmonic (SH) models: on
This problem has for a very long time received sev- the one hand models of the Earth’s topography parts (lands,
eral solutions, especially by regional or local approxima- oceans, inner seas, lakes, polar caps) taking into account the
tion. Considering the attraction of neighbouring topographic different geometrical situations and different densities, on
masses, mostly in planar geometry (local flat Earth) with the other hand model of the gravity perturbations induced
some spherical correction, the computations were based on by the different parts or their normalized counterparts when
numerical integration over finite mass elements with more or appropriate (i.e. the replacement by material of conventional
less sophistication. The basics are described in several text- density). To do so, we used a method which we developed
books—see for instance Heiskanen and Moritz (1967), Torge in a very different context, which aimed at precisely finding
(2001). This conventional approach, by which one defines the gravitational spherical harmonic coefficients of a homo-
classical Bouguer anomalies, is a simplified realization of geneous body (an asteroid, a comet nucleus or a natural
the mass normalization process described above, where mass satellite of odd shape) from the spherical harmonic coeffi-
layers are usually approximated in two steps by: (i) flat cients describing its shape (Balmino 1994). The theory also
plates of finite thickness, infinite extent and uniform density included the case of a body composed of several layers with
(Bouguer plate), (ii) volume elements, e.g. pseudo-rectangu- different densities and therefore was readily applicable to
lar/spherical prisms (tesseroids), to take into account the real the computation of the gravity perturbations due to homo-
topography of the Earth (deviation from the plates, so-called geneous matter between two surfaces, which is exactly what
terrain correction applied at regional scale). These two steps we need here.
are sometimes merged into a single one in which the total We define the spherical Bouguer gravity at the Earth’s
attraction of the whole mass column from the reference sur- surface by:
face to the Earth’s surface is directly computed. Moreover, the
gB (P) = g(P) − A(topo) (1)
modern point of view, corresponding to Molodensky theory
(see Heiskanen and Moritz 1967; and Hofmann-Wellenhof where g(P) is the measured gravity at point P and A(topo)
and Moritz 2005), is to stay with the attraction of the sub- is the total attraction (at P) of the topographic masses them-
tracted or normalized masses, at the Earth’s surface; that is to selves (between the surface and the geoid) or of their sub-
compute a gravity perturbation over this surface without free- stitutes according to normalization conventions. A(topo) is
air correction. In practice, a great number of techniques and computed by taking into account the whole Earth with its
mathematical tricks have been used to improve and speed up real shape and surface density: topographic masses include
the computational process which becomes heavy as the res- matter above the geoid (and lack of it below the geoid in
olution and size of the domain increase. We will not discuss some continental areas), bathymetry, ice and lakes.

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

The spherical Bouguer gravity anomaly is then defined in surface [SE ], defined by the lower limit of the atmosphere
the context of Molodensky theory, as: and based on ETOPO1 and additional data, on a world-
wide 1 × 1 grid;
ΔgB (P) = gB (P) − γ (Q) (2)
• to perform 1 × 1 SHS of the Molodensky gravity anom-
where γ (Q) is the normal gravity at point Q on the tellu- alies on [SE ] associated with EGM2008, so as to finally
roid, corresponding to P. Since the Molodensky (surface) produce a global spherical Bouguer gravity anomaly grid
anomaly is Δg(Molod) = g(P) − γ (Q), we have: (and map).

ΔgB (P) = Δg(Molod) − A(topo) (3) In a similar manner, we derived the SH gravity coefficients
In our case, Δg(Molod) is derived from a global Earth grav- of the compensation of all topographic components for an
ity model, here EGM2008 (Pavlis et al. 2008) and, following Airy isostatic model with fixed compensation depth, and we
Pavlis (1998) and Rapp and Pavlis (1990) it is computed as: produced a global 1 × 1 grid of the gravity corrections on
  [SE ], and a final grid of the isostatic anomalies.
∂T 2 All SHAs and SHSs were done up to degree and order
g(Molod) = − − T + εproj + εh + εγ (4)
∂r r Q 10,800 corresponding to the 1 resolution (half-wavelength).
However, for the EGM2008 surface gravity anomalies the
where T = W − U is the disturbing gravitational potential,
model limitations (maximum degree and order 2,160, plus
with W the Earth’s gravity potential and U the normal poten-
some terms up to degree and order 2,190) were obviously
tial (of the dynamic reference ellipsoid). εproj , εh , εγ (respec-
applied.
tively, named isozenithal projection correction, and first and
In a first part we will recall the theory; subsequently we
second ellipsoidal corrections) are small, long-wavelength
will give its application to our case, emphasizing the han-
correcting terms (not exceeding 3, 3 and 100 μGal, respec-
dling of the different Earth’s surface components. Then we
tively) which we here compute from the EGM2008 spherical
will explain our strategy with respect to several challenging
harmonic model truncated at degree 60. The precise evalua-
problems and describe the adopted solutions, especially con-
tion on the telluroid of the term depending on T is delicate
cerning the SHA and synthesis of models of very high degree
if one wants to minimize the computational effort (for very
and order (examples will be given with verification tests and
large data sets or grids)—as will be the case for the compu-
error analysis); in this part we will also address the problem
tation of A(topo) from spherical harmonics, and will be the
of precisely and efficiently computing a geodetic function at
subject of Sect. 4.3. Equation (3) must be slightly modified
the Earth’s surface, i.e. at (a large number of) points with dif-
to agree with the actual definition of the reference ellipsoid
ferent altitudes, and we will give a method based on Taylor
which includes the mass of the atmosphere; an atmospheric
expansions of high order. These critical problems all being
correction term δgAC must be added to the gravity anom-
solved, the next part will show an example over the Morocco–
alies. δgAC is approximately a function of the elevation H
South of Spain area, and comparisons with results of Kuhn
with respect to sea level. With H in meters, δgAC is given in
(private communication 2011) using the approach of Kuhn
milligal, by (NGA 1999):
et al. (2009) will be given. For sake of completeness, the
δgAC = 0.87 exp[−0.116(H/1,000)1.047 ] if H > 0 derived Airy isostatic anomalies will also be presented over
δgAC = 0.87 if H ≤ 0 (5) the same area (Morocco, in short). Finally we will present
the worldwide 1 × 1 grids and maps of spherical gravity
To compute A(topo) the core of the work has been: perturbations computed from our ETOPG1 model, of spher-
ical Bouguer anomalies derived from the EGM2008 surface
• to perform spherical harmonic analysis (SHA) of the gravity anomalies and from the ETOPG1 perturbations, and
heights of the Earth’s topography–bathymetry compo- of Airy isostatic gravity anomalies.
nents from the ETOPO1 database (Amante and Eakins
2009) and other databases for inner seas and lakes, such
as ILEC (http://wldb.ilec.or.jp/), at 1 equiangular reso- 2 Theory
lution, plus SHAs of the second and third powers of these
height values (as will be shown); It is based on the expression of Newton’s integral in spheri-
• then to transform the obtained SH coefficients into gravi- cal harmonics, which has been addressed by many authors,
tational SH coefficients, thus producing what we call the e.g. Rummel et al. (1988); Balmino (1994); Wieczorek and
Earth’s TOPography derived Gravity model at 1 resolu- Phillips (1998); Tsoulis (2001); Ramillien (2002); Kuhn and
tion, or ETOPG1; Featherstone (2003).
• to perform spherical harmonic synthesis (SHS) of the We briefly recall the basics of the method, from our ear-
gravity perturbations produced by ETOPG1 at the Earth’s lier work (Balmino 1994). We want to find the gravitational

123
 G. Balmino et al.

SH coefficients of a homogeneous body of density ρ from a infinite—even if S is finite, but is truncated at degree L in
SH model of its shape. In a reference coordinate system [R] practice. Here the K lm coefficients are given by:
fixed in the body, the shape is described by a truncated series ⎡ ⎤
 (ϕ,λ)
r
S giving the radius vector r as a function of the latitude ϕ ⎢ ⎥ ∗
and longitude λ: K lm = ραlm ⎣ r l+2 dr ⎦ Y lm (ϕ, λ) dσ1 (12)
σ1 0
r (ϕ, λ) = R0 (1 + S) (6) 
where σ1 is the unit sphere and αlm = 1/ (2 − δ0m )(2l + 1)
where R0 is a reference length and M Rl .
In short, we will note T jq = (A jq , B jq )shape , and K lm =

J 
j
(C lm , S lm )gravit . Our goal is to find the K lm ’s from the
S= (A jq cos qλ + B jq sin qλ) P jq (sin ϕ) T jq coefficients. The rigorous analytical solution (Balmino
j=0 q=0
1994), is quite heavy; it requires the introduction of n − j

J 
j coefficients (a generalization of the Clebsch–Gordan, or 3− j
= T jq Y jq (ϕ, λ) (7) coefficients, Wigner 1959) for the integrals of products of
j=0 q=− j any number of surface spherical harmonic functions. Practi-
In the case of a planetary topography, H , measured with cal computations were performed up to degree and order 180
respect to a sphere of radius R0 , we simply have H = R0 S. and for integrals of products of up to 72 functions (using a “in
In Eq. (7) the P jq are the Legendre polynomials (q = 0) house” extended precision arithmetic library). We encoun-
and associated functions (q > 0) of the first kind, with the tered numerical problems beyond those limits—which are
usual geodetic normalization (Heiskanen and Moritz 1967), insufficient (by far) for our concern. It is much easier to adopt
the Y jq are defined by: the numerical solution which is recalled below.
Using Eqs. (6), (7) and (12), we find:
Y jq (ϕ, λ) = P jq (sin ϕ) eiqλ (8)  l  l+3
∗ 4πR03 R0 (k)
Y j,−q = (−1) q
Y jq (9) K lm = ρ γlk T lm (13)
(2l + 1)M R
k=0
√
with q ≥ 0, i = −1, and where the superscript * indicates with
the complex conjugate; coefficients A, B and T are normal- 
(k) ∗
 
ized according to the P s and Y s and are related by: T lm = [4π(2 − δ0m )]−1 S k (ϕ, λ) Y lm (ϕ, λ) dσ1 (14)
σ1
T jq = (A jq − i B jq )/(2 − δ0q )  
∗ l +3 (1)
T j,−q = (−1)q T jq (10) and γlk = 1
l+3 ; T lm is obviously equal to T lm . We
k
note that the summation on k stops at l + 3 for each degree
Let us note that we may have a T 00 term if r  = 0 (zero (k)
l. The T lm ’s for k > 1 correspond to coupling (products)
mean over the sphere).
of the coefficients T jq (i.e. all possible combinations of k
On the other hand the gravitational potential of the body
products).
is also represented by a spherical harmonic series, written at (k)
each point P as: The T lm are derived by successive SHAs of the S k func-
tion mean values computed on regular (equiangular) grids by
l=L   +l
GM  R l  SHS. Integrals of SH functions are computed analytically or
U = (C lm cos mλ + Slm sin mλ) numerically according to the degree and order (see Sect. 5.1).
r r
l=0 m=0 From this general expression we can derive formulas for dif-
×P lm (sin ϕ) ferent cases.
l=L   +l
GM  R l 
= K lm Y lm (ϕ, λ) (11) (a) Body composed of several homogeneous layers
r r
l=0 m=−l
There are N + 1 layers. The radius vector of the outer sur-
with G: the gravitational constant, M: mass of the body, R:
face (ν ) of the νth layer (ν = 0, 1, . . . N ) is modelled by:
reference length (usually close to the body mean radius),
C lm , Slm and K lm : dimensionless harmonic coefficients of rν = Rν [1 + Sν (ϕ, λ)] (15)
degree l and order m, and r, ϕ, λ are the spherical coordinates
with rν+1 < rν (∀ϕ, λ), and:
of the point P in the reference system [R]. The C lm , S lm , K lm
 ν
and K l,−m coefficients (m ≥ 0) are related to each other by Sν (ϕ, λ) = T jq Y jq (ϕ, λ) (16)
equations similar to (10). The series for U is in principle j,q

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

As before we define: An elegant way of dealing with lateral variable density is to

 introduce a surface density function (Kuhn and Featherstone
ν(k) ∗
T lm = [4π(2 − δ0m )]−1 Sνk (ϕ, λ) Y lm (ϕ, λ) dσ (17) 2003). Instead we adopted the following classical approach.
σ1 We perform the above transformation (Eq. 20) for a topo-
and also ρν = ρν − ρν−1, (ρ−1 = 0) with ρν the density of graphic layer made of several parts of different densities cor-
layer ν. Introducing ρ, the mean density of the body, we have: responding to the various Earth’s surface elements and their
normalization. If ρω∗ is the true, or normalized density, of a
 3
3 
N
gravi 1 Rν given component extending over a domain Dω of the Earth,
K lm (N + 1 layers) = ρν
ρ 2l + 1 R0 the spherical Bouguer coefficients for all components are (in
ν=0
shorter notation):
 l    
 ρ ∗ 3  R0 l
l+3
ω(1)  k ω(k)
Rν ν(k) l+3
× γlk T lm (18) K lm (topo) = ω
T lm + γl T lm
R ρ 2l +1 R
k=1 ω k=2
This equation is fundamental for deriving the Airy isostatic (21)
potential coefficients (with N = 1 and R1 = R0 − D, D ω(k)
being the depth of compensation). The T lm harmonics are computed by SHA of topographic
heights over the convex envelope of the Dω domain solely,
(b) Case of one layer, with density ρ0 , limited by two complex defined by meridians and parallels which bound the domain,
surfaces (0 ), (1 ) and by setting the analyzed function to zero inside the enve-
lope and outside the domain (a tapering function has been
The radius vectors of these surfaces are defined by applied in some cases). Equation (21) will be truncated (as
Eq. (15). In addition, we here assume that the layer is thin to be shown later) at kmax = min(l + 3, K ) and will be called
with respect to the size of the body and we take R1 = R0 , transformation up to the Kth power of the topography H,
ν which indeed means that we take into account H/R0 up to
which usually implies (a) nonzero T 00 term(s). Then:
the power K . We pay a special attention to the degree and
 l 
l+3
gravi 3 ρ0 R0 order zero term, given exactly by:
K lm (one layer) = γlk
2l + 1 ρ R  
3  ∗
N
k=1 gravi ω(1) ω(2) 1 ω(3)
0(k) 1(k) K 00 (N parts) = ρω T 00 + T 00 + T 00
× [T lm − T lm ] (19) ρ 3
ω=1

From here we define different types of spherical harmonic (22)

coefficients representing the potential of a layer of matter, It generalises the concept of spherical Bouguer shell, and
now called topography: (i) rigorous Bouguer potential coef- requires to carry the transformation up to K = 3.
ficients when (1 ) is the Earth’s geoid, for instance defined Finally, and as shown in the following section, for some
from a global gravity field model; (ii) ellipsoidal Bouguer components of the Earth’s surface this procedure must be
potential coefficients when (1 ) is an ellipsoid (of revolu- applied twice to account for the correct removal of some
tion); and (iii) spherical Bouguer potential coefficients, when parts and normalization of others.
(1 ) is a sphere of radius R0 , given by:
 l
gravi ρ0 3 R0
K lm (topo/sphere) = 3 Application
ρ 2l + 1 R
 

l+3 We apply the above approach to the following components of
k (k)
× T lm + γl T lm (20) the Earth’s surface (Table 1) where all elevations, taken from
k=2 ETOPO1 or ILEC (or other data source in some cases) are
The latter is the type of coefficients which will be computed given with respect to sea level and assumed to be orthometric
and used in this paper—with the possibility of applying long heights:
wavelength ellipsoidal corrections. In the last equation, the
first term (k = 1) has been isolated and the superscript (a) all lands above or below sea-level (except the ice caps),
dropped; this term is the usual single layer approximation (b) all oceans,
in the transformation T jq → K lm . (c) closed seas and lakes,
(d) ice caps. Ice shelves were considered but not treated
(c) Case of one layer made of several parts with different separately: assuming they are purely floating ice (ignor-
densities ing the mechanical constraints at the places they are

123
 G. Balmino et al.

Table 1 The different topographic components analyzed: all from • ρ I the ice density, equal to 917 kg/m3 ,
ETOPO1 except closed seas and lakes (ILEC database) and ice shelves • ρsl the density of water for inner seas and lakes; it
(US National Snow and Ice Data Center)
has been assigned a constant value for a given entity
Components Contents although some are known to have parts with different
Lands with H > 0 All—except polar caps (according to
densities,
ETOPO1 data set) • ρ ∗ = ρsl in case (c), ρ ∗ = ρ I in case (d):
Lands with H < 0 Chott Algeria-Tunisia, Qattara depression,
Death Valley, Jordan Valley,
Netherlands (part), Turfan depression Figure 1 illustrates cases (c) and (d)—other cases are well
Oceans All known.
Closed seas Aral sea, Caspian sea, Dead sea
Lakes Baikal, Balkach, Bear lake, Constance,
Erie, Eyre, Huron, Ladoga, Leman, 3.1 Digital terrain model (DTM)-induced gravity
Malawi/Nyassa, Maracaibo, Michigan, disturbance (DIG)
Onega, Salton, Slaves (Can.), Superior,
Tanganyka, Tiberiade, Titicaca,
Victoria, Winnipeg, Yssyk-Koul
This is the gravity field induced by the topographic-water-
Ice caps • Over Greenland ice masses and their normalization, computed from a given
• Over Antarctica + ice shelves (Ross, DTM, here ETOPO1.
Ronne-Filchner, Larsen, Amery) The computation of such a field to produce Bouguer anom-
alies implies in each case the following “removal” of surface
anchored on the bed-rock), their gravitational effect can densities (in the algebraic sense—it is indeed an addition in
be ignored. some configurations):

Data errors were detected in some parts (especially the Arctic

and Antarctic areas); they could be corrected under simple (a) t ρc
assumptions. (b) t (ρc − ρw )
Specific polygons and grids were built to precisely delin- (c), (d) (t0 − t)ρ ∗ and tρc .
eate all different areas (e.g. lakes and seas—with their free
surface altitude).
Equation (21) is applied in each case. Note that this is
We define:
the correct sign for these quantities when computing the
K lm (topo), i.e. the coefficients of the DTM-induced grav-
• t the elevation of the solid rock topography [called bed-
ity disturbance field itself.
rock in case (d)]; we have t < 0 in case (b), t > 0 or
t < 0 in all other cases,
• t0 the elevation of the closed sea or lake surface in case 3.2 Airy isostatic compensation
(c) (it may be >0 or < 0), or of the surface ice (t0 > 0)
in case (d), In this well-known model the topographic constructs float on
• ρc the crust density, taken here equal to 2, 670 kg/m3 , the upper mantle, a denser under-layer of constant density ρm ;
• ρm the upper mantle density, equal to 3, 270 kg/m3 , therefore we have roots or anti-roots under the topographic
• ρw the ocean density, equal to 1, 027 kg/m3 , components.

Fig. 1 Geometry of considered layers for closed seas and lakes and for ice caps, and of Airy isostatic model for these cases. t1 is the elevation of
the root (or anti-root) at compensation level (of depth D)

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

Let us further define: In all cases, with the signs adopted in the last two
equations, the K lm coefficients must be added to the
• ρ0 = ρc in case (a), ρ0 = ρc − ρw in case (b) Bouguer anomaly field coefficients to obtain the isostatic
• ρ1 = ρm − ρc anomalies.
• ρ ∗ = ρc − ρ ∗ where, as before, ρ ∗ = ρsl in case (c),
ρ ∗ = ρ I in case (d)—cf. Fig. 1.
4 Challenging problems and strategies
Then, adopting a depth of compensation D, that is a level
at which the pressure induced by all surface loads is con- As we have seen it, the theory is fairly simple and adopting
stant, we can compute the elevation t1 of the roots or anti- the numerical approach makes the problem tractable, pro-
roots, measured with respect to the sphere of radius R1 = vided that we meet some challenges related to the desired
R0 − D: resolution of the representation, which we want to match the
  information contents in the ETOPO1 database.
ρ0 R0 2
• cases (a) and (b): t1 = −t (23)
ρ1 R1 4.1 The meaning of spherical DIG potential coefficients

   2
ρ ∗ ρ∗ R0 We made several approximations from which geometrical
• cases (c) and (d): t1 = − t + t0 origin is shown in Fig. 2.
ρ1 ρ1 R1
(24) Rigorously one should analyze OS/R0 and OG/R0
(Fig. 2a) and their successive powers to get the SH coef-
The factor (R0 /R1 )2 accounts for the ratio of the sur- 0(k) 1(k)
ficients T lm and T lm according to formula (19). Firstly
face elements sustained by the same solid angle, at lev- we ignore the geoid height, because: (i) its maximum value
els R0 and R1 , which enter into the equilibrium equation; (∼±100 m) is much smaller than H in most places on Earth;
it is equal to one in planar geometry. Equation (23) was (ii) the computation of the terrain effect is then made inde-
already derived by Rummel et al. (1988). Using Eq. (18) with pendent of the geoid model; (iii) it makes the computational
solely the ν = 1 term (to isolate the compensation part), we effort smaller; (iv) comparisons with other approaches in
find the corrections to the gravitational SH coefficients due spherical geometry are easier; (v) the impact of any geoid
to t1 : model may be evaluated subsequently. Secondly, H being
given as a function of the geodetic latitude  (and longitude
(k)
• for cases (a) and (b), a formula in which the T lm enter λ), we map H onto the sphere with the geocentric latitude ϕ
directly, thus not requiring any additional SHA: (properly converted from ) instead of ϕ  (Fig. 2b). This is
justified by the fact that (Fig. 2c): ϕ  − ϕ ≈ f sin 2H/R0
  l

oceans
3 ρ0 R0 is always <2.10−4 degree (0.8  ) in absolute value, and also:
K lm (Airy) = |H  − H | ≈ |H |( − ϕ)2 /2 < 0.06m. The analyses of all
2l + 1 ρ R
lands functions (H/R0 )k being done in such spherical approxima-

l+3  k−1 tion, the synthesis (computation of grids) will be performed in
ρ0
× γlk (−1)k accordance. However, it may be interesting to quantify it with
ρ1 respect to the ellipsoidal approximation. Assuming H  = H
k=1
 l+3−3k  and ϕ  = ϕ (Fig. 2d), the correction (to be added to the
R0 − D (k) spherical DIG potential coefficients) is, following Eq. (12):
× T lm (25)
R0 ⎡ ⎤
 R E(ϕ)+H R0 +H
⎢ ⎥
• for case (c) or (d): δ K lm = ραlm ⎣ r l+2 dr − r l+2 dr ⎦
σ1 R E (ϕ) R0
 3  l
3 ρ1 R0 − D R0 − D ∗
K lm (Airy) = − ×Y lm (ϕ, λ) dσ1 (27)
2l + 1 ρ R0 R
The difference of the two integrals is:

l+3
1(k)
× γlk T lm (26) 
l+3
k=1 l = γlk [R l+3−k
E − R0l+3−k ]H k (28)
k=1
1(k)
with T lm being the SH coefficients of t1 as given by (24) where one may take R E (ϕ) = R0 (1 − ε2 sin2 ϕ + ε4 sin4 ϕ −
and obtained by SHA. · · · ) with sufficient accuracy and ε2 , ε4 being given in terms

123
 G. Balmino et al.

Fig. 2 Approximations made (a) (b)

in the analysis of the topography S
and the derivation of spherical
Mapping
DIG potential coefficients. a G H (Σ0)
Given H (ϕ, λ), GS should be • S
f ⇐ ellipsoid (geoid height ignored)
sphere
H
analyzed (for rigorous DIG (flattening)

ge
oi
coefficients); b H is mapped on

d
the sphere; c the difference (Σ1)
H − H  can be neglected (h is ϕ ϕ' Φ ϕ
O
introduced in the context of R0
Sect. 4.3); d origin of the
ellipsoidal correction (c) (d) H
S
H' H
H(or h)
RE RE
S0
r0
ϕ ϕ' Φ ϕ
O
R0 R0

of the reference ellipsoid second eccentricity e , or its flat- followed our increase of knowledge of the Earth’s topogra-
tening f (ε2 = 21 e2 ≈ f ; ε4 = 38 e4 ≈ 23 f 2 ). phy or other geophysical or geographical functions. SH func-

H is written H = R0 j,q T jq Y jq (ϕ, λ) and cor- tions need to be computed up to degree and order (d/o) 10,800
rections are then obtained by transforming the products in the present work, which corresponds to the 1 resolution
sin2 ϕ Y jq (ϕ, λ) and sin4 ϕ Y jq (ϕ, λ), and using the orthog- (half-wavelength) of ETOPO1, and soon to d/o 21,600 if one
onality property of the Y jq (ϕ, λ) functions (Balmino 2003). wants to represent the data sets which become available at
We only give here corrections of first order in f and first and the 30 resolution (∼1 km). The problem which plagues most
second order in H/R0 ; we write δ K lm = δ 1 K lm + δ 2 K lm algorithms is of numerical nature and is due to the Institute
with: of Electrical and Electronic Engineers’ (IEEE) standard for
(k) (k) binary floating-point arithmetic; this limits the range of all
δ k K lm = −(l + 3 − k)ε2 [al−2,m T l−2,m + blm T l,m real numbers which can be represented on current computers
(k) to ∼10−305 to ∼ 10+305 . In particular, the computation of
+cl+2,m T l+2,m ] (29) all SH functions for high d/o cannot be done without several
The a, b, c coefficients are given by: tricks, which may be costly.
There is no problem with the Legendre polynomials, only
1
ank = with the associated Legendre functions (ALF): P lm (x), x =
2n + 3 sinϕ, m > 0. Writing these as y m H lm (x), where y = cosϕ
 
(n − k + 1)(n − k +2) (n + k + 1)(n + k + 2) 1/2 and H lm (x) = polynomial of degree l − m, the problem is
×
2n+1 2n + 5 known to come from both terms and is exemplified in Fig. 3a
  for l = 21,600 and on Fig. 3b for m = 3,600 (computations
1 (n−k)(n−k −1) (n + k)(n + k − 1) 1/2
cnk = have been done with a special in-house library developed by
2n−1 2n−3 2n+1
the first author for demanding celestial mechanics applica-
2n(n + 1) − 2k 2 − 1
bnk = (30) tions).
(2n − 1)(2n + 3) Figure 3a shows that P lm (x) becomes nonsignificant
(they are =0 whenever n < 0 or k > n). above some maximum order, a phenomenon used by Jekeli
Of greater interest are the long wavelength corrections et al. (2007)—see below. On Fig. 3b it is clear that the H lm ’s
which we have computed in parallel to the correcting terms increase regularly with l before becoming significant, then
considered for the surface gravity anomalies (see Sect. 1). oscillate around a stable value; and the P lm ’s behave simi-
They will be shown later on together with other results. larly (starting at ∼cos mϕ), which we will use in our algo-
rithm.
4.2 Spherical harmonic functions of high degree and order When using recursive relations on the H lm (x), as it has
become customary in geodesy, usual tricks, e.g. Wenzel sin-
There has been an increasing need in resolution and precision gle normalization (1998), or Horner’s schemes (in cos ϕ) on
of SH representations in the last years, which has generally partial sums such as in Holmes and Featherstone (2002),

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

(a) Degree fixed : l = 21600 ; ϕ = 75°

extremum at Hlm
log⏐Hlm⏐ m = 10443

Plm
log⏐Plm⏐ cos m ϕ

log (cos m ϕ)
values of Plm
are significant only
for m < mmax

mmax order m

log⏐Hlm⏐
log⏐Plm⏐ ⏐ lm⏐
log⏐P
truncated at -300)
log⏐Hlm⏐

⏐Hlm⏐

⏐Plm⏐
⏐Plm⏐

degree ⏐Hlm⏐
degree
(b) Order fixed : m = 3600 ; ϕ = 75°

Fig. 3 Behaviour of the Legendre associated functions P lm (x), and their components cosm ϕ and H lm (x), at 75◦ North latitude: a for degree
21,600 and all orders; b for order 3,600 (cosm ϕ = 10−2,113 ) and degrees up to 21,600

do not work at such level. Use of specialized library is an therefore by converting their limiting criterion: for a given
alternative but is too time consuming for operational situa- latitude ϕ and fixed order m, we retain the terms of degree
tions. A breakthrough was made recently (Jekeli et al. 2007) l > l0 (m) − D(m, ϕ), where l0 (m) is derived from Jekeli
which started from the recognised fact that the meaningful et al’s rule (except near the poles where we adopted a dif-
ALFs at a given latitude and for a given degree follow a ferent formulation) and D is an empirical function (Fig. 4).
semi-empirical rule which allows to safely neglect more and By computing the H lm (x) by recursive formulas with fixed
more terms—indeed to limit the order of the expansion, as order (stable close to poles) we take advantage of the Partial
|ϕ| increases (cf. Fig. 3a), to m max (ϕ, l) = l cosϕ + C (C Sums-Longitude Recursion (PSLR) technique for the SHS
is an empirical constant depending on the spectrum of the of equiangular grids (Bosch 1983) and its dual formulation
studied function); this circumvents all numerical problems (LRPS) in SHA (Balmino 2003)—all this in view of the exist-
provided that one makes use of recursive formulas for the ing software developed along many years. Besides we can
P lm (x) themselves with fixed degree (which are in principle retain 64 bits arithmetic, though making verifications in 128
unstable near the poles). Based on Fig. 3b, our approach has bits, for optimization purposes (on small computers), with-
been to modify Jekeli et al.’s strategy by working by order— out loosing precision near the poles. The drawback is that

123
 G. Balmino et al.

(a) log Plm (sin ϕ ) abs.val. ( Plm) (b)

0
10
ϕ=45°
-5
10

m = 80 (fixed)
-10
10 D
m
l0(m)
-15
10
80 100 120 140 160 180 200
degree l

degree lmin
l0(m) truncation inf. in l

Fig. 4 Example of Jekeli et al. ’s (2007) criterion transposed to the case oscillates around a stable value. The significant degrees are retained
of associated Legendre functions of fixed order and variable degree. ϕ above lmin (m) = l0 (m) − D(m, ϕ) where D is an empirical function
being fixed, P lm (sin ϕ) increases regularly with l, up to l ∼ l0 (m), then

we have to compute some intermediate values of the H lm (x) the computation is done on the telluroid and we assimilate
which are out of the IEEE range but necessary for starting the ETOPO1 elevations to normal altitudes, equal to zero
the scheme. The core of the algorithm is described in appen- over the oceans.
dix A and is based on the idea of Wenzel (1998); it consists The rigorous calculation of δg(S) or g(S) is too heavy,
in an iterated re-normalization of the H lm (x) as l increases. since it requires to recompute the ALFs at each point S (see
Finally, depending on the maximum degree and latitude, we Fig. 2c), but it has been done over limited mountainous areas
apply Horner’s schemes in cos ϕ and (or) with respect to the for verification. We may adopt a semi-rigorous method: we
normalization factor. already noted that the latitude difference ϕ − ϕ  remains
quite small even for significant altitudes (for instance it is
4.3 Precise and efficient computation of a geodetic function 0.0001208◦ at 4,000 m and latitude ϕ = 45◦ ). Therefore the
at the Earth’s surface ALFs need not be recomputed at each grid point but solely
once for each parallel (but the exact radius vector value at
The Bouguer correction, or the isostatic one, is a gravity per- each point S is used). We have inter-compared these methods
turbation δg = −∂ T /∂r and must be computed at the Earth’s for computing g, also with the upward continuation from
surface [SE ], that is at variable altitude for each point. This the ellipsoid using Taylor expansions of order 1 and order 2,
hampers the application of fast techniques (such as PSLR) in over a 2◦ × 2◦ part of the Atlas mountains in Morocco where
SHS. The problem is general and we must question ourselves H ranges from 134 to 3,785 m, and using EGM2008. Results
on the accuracy of usual methods for gridding values of any are given in Table 2.
geodetic function on [SE ] from a high d/o SH model, such Obviously the rigorous method is very costly. The semi-
as the surface gravity anomalies—for instance derived from rigorous method works very well in terms of precision and
a global model like EGM2008. can be taken as reference. The Taylor expansions are not
Most of the time, the function is gridded on the ellipsoid, precise enough at this level. Looking at the first (radial)
then upward (or downward) continued (using 1st, 2nd,. . . and second derivatives of g(g1 and g2 ), it appears
derivatives—also evaluated on the same grid). This is hardly that these terms, evaluated with EGM2008, are unrealisti-
sufficient in high mountains, even for expansions up to 2,160 cally large over the highest mountain summits (< 5% of
(2190) like in EGM2008 as we are going to see it. First, the area): g1 reaches −0.492 and 0.376 mGal/m (this
all altitudes (H ) of ETOPO1 and altitudes of the lakes and is greater than the free-air gradient !) and g2 ranges
closed seas are converted to ellipsoidal altitudes (h) using from −5.0610−4 to +5.4910−4 mGal/m2 ; elsewhere g1
EGM2008. The surface of the oceans is assumed to coincide is ∼0.01 mGal/m and g2 ∼ O(10−5 ) mGal/m2 . Con-
with the geoid. Therefore, [SE ] is referred to the EGM2008 sequently we experimented more with the semi-rigorous
ellipsoid. For surface (Molodensky) gravity anomalies g method but found that it was costly at high resolution (1 × 1

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

Table 2 Comparison of different SHS methods: 5 × 5 grids of surface gravity anomalies (in mGal) over part of the Atlas mountains, from
EGM2008 and elevation data set DTM2006
Method Min. value Max. value r.m.s. differences (mGal) CPU timea

(1) Rigorous −41.39 249.12 – 761

(2) Semi-rigorous −41.39 249.05 (2)–(1): 0.004 73
(3) Taylor 1 −41.30 237.02 (3)–(2): 4.783 10
(4) Taylor 2 −41.39 255.30 (4)–(2): 2.070 16
a IBM Power 5 (1.9 GHz) processor—time in seconds

Table 3 Computation of 1 × 1 grids of δg at degree and order 10,800 Indeed we have been able to implement the simultaneous
by the semi-rigorous method; comparison of CPU time over areas of computation of a geodetic function and its radial derivatives
different sizes (IBM Power 5 (1.9 GHz) processor)
of high order, and the final summation at grid nodes, in a
Area Latitude extension Longitude extension CPU time (sec) quite efficient way. Details are given in Appendix B. Then
we have been able to compare the two approaches. Antic-
Everest 1◦ 2◦ 7,363
ipating once more on the obtained ETOPG1 model (see next
France 9.5◦ 13.5◦ 247,301
section) we have computed grids of gravity perturbations δg
Morocco 11◦ 12◦ 316,902
on [SE ] over several areas and for varying orders N of the
Taylor expansion (Table 4).
The conclusions are:
grids with d/o 10,800) over large areas (see Table 3—
with some anticipation on the Bouguer gravity field model, • CPU time t is clearly linear in N : for example in the case
ETOPG1, which we obtained from ETOPO1 and other data, of Everest t ≈ 1,750 + 11.5N ; but it does not increase
and which will be described in Sect. 5). Although computa- much with N (cf. France area);
tions over the whole Earth could be done with it, we decided • the computer time saving with respect to the semi-rigor-
to re-visit the Taylor expansion approach to save computer ous method is more than an order of magnitude over large
time. areas;
Any degree l component of a geodetic functional has • N = 40 will ensure a precision always better than
the factor (a/r )l where a is a reference length (close to ∼ 0.1 mGal (worst case over Everest) and certainly much
the Earth’s equatorial radius). We want to study the conver- better (few microgals) almost everywhere on Earth.
gence of the following Taylor expansion of order N around
r0 and for various values of r (see Fig. 2c, with r = OS and Therefore Taylor expansion with N = 40 has been adopted
r0 = OS0 ): for all subsequent SHS of grids.
 a l  l 
N  
a 1 ∂ k (a/r )l
= + (r − r0 )k (31)
r r0 k! ∂r k r =r0
k=1 5 Main results
We test the convergence in spherical approximation; we
expand (a/r )l with r0 = a, r = a + h, around h = 0, that 5.1 Analysis of the Earth’s topography
is: (k)
  The SH coefficients T lm are derived by a standard quadra-
 a l  N
(−1)k k h k
=1+ l (32) ture method applied to 1 × 1 equiangular mean values, and
r k! a accelerated by the Longitude Recursion-Partial Sums algo-
k=1
rithm (LRPS, Balmino 1994). Integrals of ALFs are com-
k−1
where lk = j=0 (l + j). puted in different ways according to the order m and the
Results, as shown on Fig. 5, are striking: one needs to go maximum degree L: the method of Gerstl (1980) is applied
to very high orders N as the elevation increases, all the more up to d/o 1,800 or above for low orders, then a modifica-
as the degree increases too. tion of it (Balmino 2003) is used up to d/0 2,700 or above
Those findings are for individual terms, and may be tem- for low m (with one initial normalization); beyond those
pered by considering the (decreasing) values of the function limits, numerical Simpson method is applied (with subdi-
SH coefficients. Also one has to weigh the performances vision of the integration interval when necessary) on the
of the method with respect to the semi-rigorous approach. ALFs computed according to Appendix A. De-smoothing

123
 G. Balmino et al.

Fig. 5 Investigation on the

Taylor expansion order required
(a) degree l = 2160
to compute (a/r )l at the Earth’s
surface by upward continuation
from the ellipsoid. a Degree
l = 2,160; b degree l = 10,800

relative error
relative errors are
10 to 15 orders
of magnitude
below in
quadruple precision

N : order of Taylor expansion

(b) degree l = 10800
relative error

2. 10-5

8. 10 -13

order of Taylor expansion

relative errors are
10 to 15 orders
of magnitude below
in quadruple precision

Table 4 Computation of gravity perturbations δg with full ETOPG1 model (lmax = 10,800) at ETOPO1 points (1 × 1 grids): (i) by Taylor
expansion; (ii) by semi-rigorous method

Area lat. min/max N min/max diff. CPU CPU time

long.min/max (Taylor) with semi-rig. method time semi-rig.method
⇓ (mGal) (sec) (sec)
30 – 4.27 / + 39.51 2094
27.5°/ 28.5° 35 – 1.20 / + 23.74 2145
Everest 7363
86.0°/ 88.0° 40 – 0.13 / + 0.004 2205
60 – 4.10 –7/ + 1.10 – 7 2438
42.0°/ 51.5° 20 – 1.16 / + 0.06 14525
France 247301
– 5.0°/ 8.5° 30 – 3.10–5/ + 3.10– 5 15696
28.0°/ 39.0°
Morocco 30 – 5.10 –7/ + 5.10– 7 20391 316902
– 12.0°/ 0.0°
Comparison of performances for various orders N

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

factors based on Meissl–Pellinen coefficients (Jekeli 1981)

σl : sqrt of degree variances
are finally introduced.
fit of σl by power law A / lα
Figure 6 shows the behaviour of the SH coefficients (here
sqrt(var.) in meters

(α = 1.882)
multiplied by R0 ) resulting from the analysis of ETOPO1
data set. An empirical law: A/l 1.882 fits well with these spec-
tral components.
We then look at the magnitude of the SH harmonics of the
kth power of the topography. Figure 7 shows their dimension-
less spectral components; at first glance one would decide
that going beyond k = 3 is superfluous (Fig. 7a), but the
(k)
gravitational contribution of the T lm must be weighted by
Degree γlk (Fig. 7b): higher-order terms (k > 1) become much more
important at resolution higher than ∼7 km(l > 2,800 =
Fig. 6 Results of analysis of the whole Earth topography (land topo.
l0 ); this contradicts a statement of Wieczorek and Phillips
and oceans/lakes bathymetry) from ETOPO1: square root of SH coef-
ficients degree variances (in meter) (1998) on the decrease of these terms. This might be also

Fig. 7 Spectral behaviour of

(k) (a) sqrt (degree
( variances) : Lands
the SH coefficients T lm for the
lands only (excluding lakes,
closed seas and ice covered
parts) up to k = 3. a S k stands
for the kth power of the land
topography; b The coefficients SHA of S1 ......... well fitted by C / l 3/2
dimensionless sqrt(var.)

(k)
T lm are weighted by γlk , and
shown with respect to the
contribution to topography,
r.m.s.(δ Hl ), of harmonics of SHA of S2
degree > l

SHA of S3

(b) weighted sqrt (degree

( variances) : Lands
Resolution = 20000 km / l r.m.s.(δHl) :

10 km 3 km 2 km 40. m
5 km
dimensionless weighted sqrt(var.)

30. m
r.m.s.(δHl) 20. m
γl1 σl (SS1) 10. m

0. m

γl3 σl (S3) γl2 σl (S2)

123
 G. Balmino et al.

Fig. 8 Morocco–South of Spain-ocean area, selected for many tests and comparisons in this study. Topography is from ETOPO1; values are
between −5, 247 and 3,771 m

true for k > 3 (which we have not investigated). However, intuitively expected; this is confirmed by a close look at a
Fig. 7b also shows that the contribution to topography, r.m.s. West–East 200-km long section which crosses rough topog-
(1)
(δ Hl ) = R0 ∗ r.m.s.[T lm ], of those terms of degree > l0 is raphy and then a more gentle area (right of Fig. 9b) where
rather small. This is why we decided to truncate expansion smaller differences are visible. We call this error of repre-
(13) at k = 3 at most; this is in agreement with Wieczorek sentation, which may be attributed to both the SHA and the
(2007). Nevertheless this remains, at this stage, an empiri- SHS which are approximations of exact operators—SHS is
cal approximation which demands external comparison (see not exactly the inverse of SHA.
Sect. 6). We can further perform a SHA of this error, then a new SH
Next we investigated the precision of our SHA of the of the differences and see if the process converges. Results
topography, by recomputing (by SHS to d/o 10,800) 1 × 1 over the test area are given in Table 5. There is indeed a sign
mean values and comparing them with the original ETO- of convergence, but is it worth the effort? We will answer it
PO1 mean values, to quantify the error on the DTM-induced in the next section.
gravity field—which we do in Sect. 5.2. We did it for many
areas on Earth; the largest differences appear over the Hima- 5.2 The ETOPG1-induced gravity disturbance model
layas (25◦ N to 40◦ N, 70◦ W to 95◦ W) with: min = −196 m,
max = +167 m, r.m.s. = 12.8 m. We then selected an area This model results from the conversion into gravitational SH
which we found quite representative of the Earth’s relief in coefficients, of the sets of SH coefficients representing the
terms of topographic features and values: Morocco–South of Earth’s topography (lands, oceans, ice caps and underlying
Spain (Morocco area in short). It has mountainous and flat bed-rock) provided by the ETOPO1 database, plus elevation
parts, significant bathymetry and its topography ranges from data of main closed seas and large lakes, as explained in the
−5, 247 to +3,771 m (Fig. 8). It will be used subsequently previous sections. It may be symbolically noted {K lm (T jq )}.
for testing several steps of our approach. It is complete to d/o 10,800. The behaviour of its spectral
The evaluation of the precision of our SHA–SHS proce- components is shown on Fig. 10. Compared to the EGM2008
dures over this area (Fig. 9) shows acceptable differences, characteristics (to d/o 2,160), it is obvious that they are signif-
although geographically correlated: differences are larger icantly larger, the well known sign of a compensation mech-
over highest and rough topography (Fig. 9a) as could be anism such as isostasy.

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

(a) mean 1' x 1' differences Table 5 Reduction of the error of representation of the topography (in
meter) by iterated SHA–SHS
Iter.→ 0 1 2

Min −66.8 −44.6 −40.4

Max +67.3 +44.0 +41.6
r.m.s. 4.2 2.7 2.5
latitude

Morocco test area

longitude
(b) Section AB in Atlas mountain : ϕ = 31°

Fig. 10 Spectral characteristics of the ETOPG1 field model to degree

10,800, compared to EGM2008 to degree 2,160

from the error of representation of the topography by SH (see

Sect. 5.1). The impact of the former is difficult to clarify. To
try to quantify the latter we may evaluate the global r.m.s.
error on the induced gravity perturbation δg. We start from:
GM   a l+2 
δg(r, ϕ, λ) = 2 (l + 1) K lm Y lm (ϕ, λ)
a r m
l
(33)
We propagate the variances and we average over the unit
sphere σ1 :
 
GM 2  
σ 2 (δg)σ1 ≈ (l + 1) 2
σ 2 (K lm ) (34)
a2 m l

We make a single layer approximation with the surface den-

sity ρ equal everywhere to ρc (equivalent rock topography),
that is:
ρ 3
K lm ≈ T lm (35)
Fig. 9 Error of representation of the topography in the test area: dif- ρ 2l + 1
ferences between the recomputed topography (from SH) and ETOPO1.
a Over the whole area: min/max/r.m.s. = −67, +67, 4.2 m; b compari- Using:
son along section AB: the fit is much better when altitude gradients are
 σ 2 (δ H )
small (here East of longitude −7.7◦ ) σ 2 (T lm ) ≈ (36)
l m R02
The model has uncertainties coming from the errors in the
where σ 2 (δ H )1/2 is approximated by r.m.s.(δ H ), the mean
original data sets. Some, of tens or even hundreds of meters,
error of representation, we find:
were discovered in the ETOPO1 data, e.g. over the Dead
Sea and Jordan Valley, over lake Titicaca,. . . (we could cor- 3 G M ρ r.m.s.(δ H )
r.m.s.(δg) = {σ 2 (δg)σ1 }1/2 ≈ (37)
rect for some of them). In addition, there is an error coming 2 a2 ρ R0

123
 G. Balmino et al.

Fig. 11 Error on DTM-induced gravity disturbance δg due to estimated error of representation of the topography by SH, over the Morocco test
area. min = −2.2 mGal; max = +2.2 mGal; r.m.s. = 0.12 mGal

µ gal
→ 603

→ 602

→ 601

→ 600

→ 599

→ 598

Fig. 12 Ellipsoidal correction on the gravity perturbations, limited to degree and order 120. It ranges between 598 and 603 μGal

If we take r.m.s.(δ H ) ≈4 m (cf. Morocco area—representa- on gravitational harmonics. Therefore, we decided not to per-
tive of Earth),we have r.m.s.(δg) ≈ 0.5 mGal for the whole form several SHA–SHS iterations worldwide.
Earth, which is rather small compared to other error sources Finally we computed the ellipsoidal corrections on the
(errors in ETOPO1 as said above, in the reference global gravity perturbations (Fig. 12), to d/o 60, then to d/o 120—
gravity field model used to compute surface anomalies…). which did not make significant difference. They exhibit long
This is confirmed by a direct computation over the Morocco wavelength patterns with prominent zonal components and
test area, Fig. 11) where the full error of representation (at have a magnitude around 0.5 mGal. They may be applied or
iteration 0—see Table 5) is analyzed and converted into errors not depending on the usage of ETOPG1.

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

5.3 The derived Airy isostatic model Obviously the mean magnitude of the K lm coefficients
decreases very quickly due to the factor [(R0 − D)/R0 ]l ,
We computed the SH coefficient corrections according to which is confirmed on Fig. 13: around degree 400, Airy cor-
Sect. 3.2 with a constant compensation depth D = 30 km. rection is smaller than the topographic effect by a factor of
∼5, and it is more than an order of magnitude below it at
degree ∼1,000; then it becomes negligible for larger degrees.
We call δg I S O the gravity perturbations derived from K lm ,
to be added to the Bouguer anomalies so as to obtain the
usual isostatic anomalies (Sect. 3.2).
At regional scale, we compare the DTM-induced gravity
disturbances and the isostatic corrections over our test area.
The former have been limited to degree 400 for sake of mag-
nitude comparison and for enhancing geographical correla-
tion. As can be seen from Fig. 14, the Airy model removes
a significant part to the topographic masses’ gravitational
effect.

6 External comparison

As discussed in Sect. 5.1 a critical point in our approach is

the truncation (at k = 3) of the expansion giving the gravita-
tional SH coefficients in terms of the harmonics of the relative
elevations H/R0 at the kth power. Outside validation only
can bring confidence (or not) on this choice. The most reli-
able way of doing it is the direct numerical integration of the
attraction of the topographic masses effect (and/or of their
normalization), which has been the approach of Kuhn et al.
(2009). As recalled in Sect. 1, Khun uses a numerical scheme
where the gravity contribution at point P of the whole topog-
raphy is computed over the full sphere by quadrature over the
Fig. 13 Comparison of the spectral components of the ETOPO1- topographic masses elements having a size which increases
induced gravity disturbance and Airy isostatic field (fixed compensation with the distance to point P (to reduce the computational
depth D = 30 km) effort).

Fig. 14 Airy isostatic corrections over the Morocco test area, with a compensation depth of 30 km (a); and DTM-induced gravity disturbances
(here labelled spherical Bouguer corrections) to degree 400 (b)

123
 G. Balmino et al.

Fig. 15 Difference between the combined DTM (ETOPO1)-induced paper) over the Morocco–South of Spain test area; min = −37.344,
gravitational effect from Newtonian integration based on point values max = 48.040, mean = −0.067, SD = 2.340 (Unit: mGal). (Kuhn, pri-
(grid/node registered data) and the spherical harmonic approach (this vate communication 2011)

Kuhn (private communication 2011) nicely provided a test 7 Worldwide maps of spherical Bouguer anomalies
computation over the Morocco area. He made two calcula- and isostatic corrections
tions: one based on ETOPO1 mean (cell/pixel) values and
another one based on ETOPO1 point (grid/node) values. In The different parts of the global, spherical approach which we
the latter case, the values are taken as mean values of a block explained in the preceding sections, have been applied to the
shifted by half a grid element. In both instances, the computa- production of several 1 ×1 grids (and maps) in the context of
tional point (i.e. at which the gravity disturbance is evaluated) the “World Gravity Map” project of the Commission for the
is the center of each considered block. Therefore Kuhn’s sec- Geological Map of the World (CGMW 2010). The project,
ond case was retained for comparison, which is summarized sponsored by UNESCO, consists in the production of a new
by the difference map in Fig. 15. worldwide Bouguer anomaly map. We therefore computed
There is almost no bias (−0.067 mGal) in the differences four main grids, all of them on the physical Earth’s surface
which is not surprising since our truncation limit (k = 3) (lower limit of the atmosphere); they are point values at the
ensures that the K 00 term is computed with no approxi- nodes of the ETOPO1 equiangular mesh.
mation. The agreement is fairly good on average with a
standard deviation of 2.34 mGal. However, the minimum (a) Surface gravity anomalies (g) from EGM2008: this
and maximum values and the geographical distribution of computation was of course done up to the maximum
the differences (which exhibit a clear correlation with the degree (2,159, plus terms up to 2,190) of the reference
topography) probably indicate that the truncation at the third model (Fig. 16).
power of the elevation is not sufficient in areas of high/rough (b) Gravity perturbations (δg) from the ETOPG1 model, to
topography at such high resolution. Finally part of the dif- d/o 10,800 (Fig. 17).
ference may also be attributed to the effective spatial resolu- (c) Spherical Bouguer anomalies (gB ): by subtracting the
tion of the spherical harmonic models at d/o 10,800 and to grid (δg) from the (g) one (Fig. 18).
Gibbs effect of which the SHAs suffer when modelling step (d) Isostatic corrections (δgISO ) according to the Airy model
functions. with a 30 km compensation depth: the computation was

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

Fig. 16 Surface (Molodensky) gravity anomalies from EGM2008 and ETOPO1; min = −364 mGal, max = +670 mGal

Fig. 17 Gravity perturbations (ETOPG1) from ETOPO1, at Earth’s surface; min = −1,123 mGal, max = +626 mGal

done only to d/o 2,160—which is by far sufficient con- Other grids have been derived such as the one of isostatic
sidering the spectral decay of the Airy isostatic correc- anomalies (by adding gB and δgISO ). They will be analyzed
tion SH coefficients (Fig. 19). and documented by the CGMW project committee.

123
 G. Balmino et al.

Fig. 18 World spherical Bouguer anomalies from EGM2008 and ETOPO1; min = −529 mGal, max = +1,005 mGal

Fig. 19 Airy isostatic corrections derived from ETOPO1; depth of compensation is 30 km; min = −761 mGal, max = +460 mGal

8 Conclusions the available data set. Thanks to the approach, especially

the successive re-normalization strategy in the computa-
We have designed a way to handle global topography and tion of spherical harmonic functions, we do not foresee
derived gravity at a very high resolution through SHA and limitations in the mid-future when higher resolution data
synthesis, though with some approximation. The present sets are used. The method has been validated in one case
work has established and used expansions up to degree but this test is incomplete; further verification will be
and order 10,800, compatible with the 1 resolution of necessary.

123
Spherical harmonic modelling to ultra-high degree of Bouguer and isostatic anomalies

The way Bouguer, and also isostatic anomalies, are under- • we multiply H̃lm (x) and H̃l−1,m (x) by f
stood and computed differs from the common usage, but • we continue the recursion in l.
meets modern concerns which tend to take the real Earth
The process is iterated (still with m fixed), each time we
into account.
have | H̃lm (x)| ≥ G. At the end H̃ν(k),m and H̃ν(k)−1,m are
A 1 × 1 grid of worldwide surface gravity perturbations
divided by f , for k = 1, . . . N f, and we define ν(N f + 1) =
has been computed from the ETOPO1 data set and other
L.
data for lakes and closed seas, taking into account usual
We note that N f = 0 if the condition | H̃lm (x)| ≥ G was
normalization conventions for the densities of various topo-
never fulfilled. For a given value H̃lm (x) (m fixed) there is a
graphic components. Higher-resolution terrain corrections
k for which ν(k) < l ≤ ν(k + 1); therefore its true value is:
may benefit from it by integrating residual topography up to
shorter distances. 1 × 1 grids of spherical Bouguer anoma- H lm (x) = H̃lm (x)/ f k+1
lies and of spherical Airy isostatic corrections have also been
It cannot be written as such in most cases if it exceeds the
produced.
IEEE range [10−s , 10+S ] but instead we get its logarithmic
The philosophy behind this novel computation of a so-
value:
called “complete spherical Bouguer anomaly” will be clar-
ified with respect to different communities of users in the log |H lm (x)| = −(k + 1) log f + log | H̃lm (x)|
context of the World Gravity Anomaly Maps to be published
by CGMW. with sign[H lm (x)] = sign[ H̃lm (x)].
Finally the value of the ALF is obtained as follows:
Acknowledgments This research was supported by the CNES Pro- Let Z = log |P lm (x)| = mlog(cosϕ) − (k + 1)log f +
gramme Directorate and the Toulouse Space Center, the Centre National log | H̃lm (x)|
de la Recherche Scientifique and the Institut National des Sciences de
l’Univers, the Institut de Recherche pour le Developpement. We espe- • if Z < −s : P lm (x) = 0. (it is actually an underflow)
cially thank Michael Kuhn who kindly applied his method over the • if not: P lm (x) = 10 Z × sign [ H̃lm (x)]
Morocco area for comparison with our method. We thank the reviewers
who made very detailed and constructive comments. In practice, we never encountered an overflow. As an
example, in the case of Fig. 3b the algorithm finds six nor-
Appendix A: computation of associated Legendre malizations: ν(1) = 4,418, ν(2) = 5,098, ν(3) = 5,988,
functions (ALF) of very high degree ν(4) = 7,156, ν(5) = 8,735 and ν(6) = 11,110.
Actually we have seen that we can take into account the
We compute the H lm (x) by standard stable recursion with ALFs for l > lmin = l0 (m) − D(m, ϕ) only, that is we do
fixed order m, variable degree l from m to L. The recursion not have to recover the ALFs, as shown above, for l ≤ lmin
is modified by a normalization factor which is applied sev- which is some time saving. The determination of lmin is semi-
eral times, as becomes necessary with respect to the IEEE empirical. It has been established by numerous tests up to d/o
limitations on real numbers. 32,400.
We make choice of a factor f , very small but larger (by L being the maximum degree we define K = 20 +
10 to 20 orders of magnitude) than the smallest real number 8.10−3 L and d L = 23 π/L(this is 1.5× the maximum res-
which the computer can represent. For instance f = 10−280 . olution). For a given latitude ϕ, that is at polar distance
We then define G=1/ f . We initialize a counter N f = 0 and d = 90 − |ϕ|:
a table ν(k) by ν(0) = m − 1; then we compute:
√  • the overall computation is carried up to:
H̃0,0 = f ; H̃1,1 = 3 f ; H̃l,l = 1 + 1/(2l)
H̃l−1,l−1 for l ≥ 2 m max (L) = min[Lcosϕ + K , L]
√
H̃l,l−1 = x 2l + 1 H̃l−1,l−1 ; l ≥ 1 or max [35 − 5 log2 (d L /d), 0] if d < d L and L > 720
(x)
H̃lm = αlm [x H̃l−1,m (x) − H̃l−2,m (x)/αl−1
m
]; {d-condition}
l ≥ 2, 0 ≤ m ≤ l − 2 . . . (R)
• for each order m ≤ l ≤ L , from m max (L) to 1 because
where αlm = [(2l − 1)(2l + 1)/[(l − m)(l + m)]]1/2 computations are done in reverse order (in m and in l) for
We apply (R), with m fixed and l variable (from m to L), improved accuracy:
as long as | H̃lm (x)| < G.
As soon as this condition is not verified, and if m < L: m ≤ lcosϕ + K ⇒ lmin = max[m, (m − K )/cosϕ]
3π 1
or m ≤ 35 − 5log2 (dl /d) ⇒ lmin = if
• we increment N f by 1 2 d2 7−m/5

• we define : ν(N f ) = l {d-condition }is true.

123
 G. Balmino et al.

Appendix B: implementation of the Taylor expansion Balmino G (1994) Gravitational potential harmonics from the shape of
method an homogeneous body. Cel Mech Dyn Astr 60(3):331–364
Balmino G (2003) Ellipsoidal corrections to spherical harmonics of
surface phenomena gravitational effets. Festschrift zum 70. Ge-
Let us consider a geodetic functional written as: F(r, ϕ, λ) = burtstag von H. Moritz, Publication of Graz Techn. University, pp
w(r, ϕ) f (r, ϕ, λ) with: 21–30
 L Bosch W (1983) Effiziente Algorithmen zur Berechnung von Raster-
G M    a l+q
L Punkwerten von Kugelfunktionsentwicklungen. Memorandum,
f (r, ϕ, λ) = q δl (C lm cos mλ D.G.F.I., Munich
a r Commission for the Geological Map of the World-CGMW (2010) Res-
m=0 l=m
 olutions of the CGMW General Assembly, UNESCO, Paris
Gerstl M (1980) On the recursive computation of the integrals of the
+Slm sin mλ) P lm (sin ϕ) associated Legendre functions. Manuscr Geod 5:181–199
Heiskanen WA, Moritz H (1967) Physical Geodesy. Freeman and co,
San Francisco
and where w(r, ϕ), q and δl are given by the table below for
Hofmann-Wellenhof B, Moritz H (2005) Physical Geodesy. Springer,
the quasi-geoid height (ζ ), the gravity anomaly g and the Berlin
gravity perturbation (or disturbance) δg. Holmes SA, Featherstone WE (2002) A unified approach to the Clen-
shaw summation and the recursive computation of very high
Param. w(r, ϕ) q δl degree and order normalised associated Legendre functions. J
F Geod 76:279–299
Jekeli C (1981) Alternative methods to smooth the Earth’s gravity field,
ζ 1 /γ (r ∗, ϕ) 1 1. Rep N0.310, Department of Geodetic Science and Surveying, Ohio
g 1 2 l −1 State University, Columbus
δg 1 2 l +1 Jekeli C, Lee KJ, Kwon JH (2007) On the computation and approx-
imation of ultra-high-degree spherical harmonic series. J Geod
In the case of ζ, r ∗ is at the telluroid and is (in principle) 81:603–615
Kuhn M, Featherstone WE (2003) On the optimal spatial resolution
obtained by iteration. The Taylor expansion of order N is: of crustal mass distributions for forward gravity field modelling.

N  k In: Gravity and geoid 2002, Proceedings, pp 195–200
r − r0 Kuhn M, Featherstone WE, Kirby JF (2009) Complete spherical Bou-
f (r, ϕ, λ) = k guer gravity anomalies over Australia. Aust J Earth Sci 56:213–223
a
k=0 NGA (1999) Gravity station data format and anomaly computations.
(−1)k  L
Technical Report, Geospatial Sciences Division, St Louis (Mo)
with k = GaM q k! m=0 m
k
Pavlis NK (1988) Modeling and estimation of a low degree geopotential
We now use the PSLR technique (Bosch 1983); each km model from terrestrial gravity data. Rep. N0. 386, Dpt. of Geodetic
is written as: Sc. and Surveying, Ohio State Univ., Columbus
Pavlis NK, Holmes SA, Kenyon SC, Factor JK (2008) An Earth Gravita-
km = Akm cos mλ + Bmk sin mλ tional Model to Degree 2160: EGM2008. EGU General Assembly,
Vienna, Austria, 13–18 April 2008
and (PS part of the algorithm): Ramillien G (2002) Gravity/magnetic potential of uneven shell topog-
raphy. J Geod 76(3):139–149
  L  l+q+k
   Rapp RH, Pavlis NK (1990) The development and analysis of geopo-
Akm a C lm
= δl l+q
k
P lm (sin ϕ) tential coefficient Models to spherical harmonic degree 360. J Geo-
Bmk r0 Slm phys Res 95(B13):21885–21911
l=m
Rummel R, Rapp RH, Sünkel H, Tscherning CC (1988) Comparisons of
where the l+qk have been defined in the text. global topographic-isostatic models to the Earth’s observed grav-
ity Field. Rep. N0. 388, Dpt. of Geodetic Sc. and Surveying, Ohio
(ϕ, λ) being now a grid node (ϕi , λ j ) with λ j = λ0 + State Univ., Columbus
jλ, (λ0 : origin of longitudes, λ : grid stepsize in longi- Torge W (2001) Geodesy, 3rd edn. de Gruyter, The Netherlands
tude), all km (λ j ) = km, j are computed simultaneously by Tsoulis D (2001) Terrain correction computations for a densely sampled
(LR part of the scheme): DTM in the Bavarian Alps. J geod 75(5/6):291–307
Wenzel G (1998) Ultra-high degree geopotential models GPM98A, B
km, j = 2 cos mλ km, j−1 − km, j−2 and C to degree 1800. Joint meeting of the Intern. Gravity Com-
mission and Intern. Geoid Commission, 7–12 September, Trieste
Wieczorek MA (2007) Gravity and topography of the terrestrial planets.
This recursion is initialized by km,0 = km (λ0 ) and km,1 = In: Treatise on geophysics, vol 10. Elsevier, Amsterdam, pp 165–
km (λ1 ). 206
Wieczorek MA, Phillips RJ (1998) Potential anomalies on a sphere:
applications to the thickness of the lunar crust. J Geophys Res
103(E1):1715–1724
References Wigner EP (1959) Group theory and its application to quantum mechan-
ics of atomic spectra. Academic Press, New-York
Amante C, Eakins BW (2009) ETOPO1 1 arc-minute global relief
model: procedures, data sources and analysis. NOAA Technical
Memorandum NESDIS NGDC-24, Boulder (Co)

123