MULTISCALE MODELING FROM EIGEN-1S, EIGEN-2, EIGEN-GRACE01S,

GGM01, UCPH2002 0.5, EGM96

Martin J. Fengler
(1)
, Willi Freeden
(1)
and Martin Gutting
(1)
(1)
Geomathematics Group, University of Kaiserslautern,
P.O.Box 3049, 67653 Kaiserslautern, Germany
M. Gutting will present these results at the workshop.
Abstract
Spherical wavelets have been developed by the Geomathematics Group Kaiserslautern for several years and
have been successfully applied to georelevant problems. Wavelets can be considered as consecutive band-
pass lters and allow local approximations. The wavelet transform can also be applied to spherical harmonic
models of the Earths gravitational eld like the most up-to-date EIGEN-1S, EIGEN-2, EIGEN-GRACE01S,
GGM01, UCPH2002 0.5, and the well-known EGM96. Thereby, wavelet coecients arise. In this paper it
is the aim of the Geomathematics Group to make these data available to other interested groups. These
wavelet coecients allow not only the reconstruction of the wavelet approximations of the gravitational
potential but also of the geoid, of the gravity anomalies and other important functionals of the gravitational
eld. Dierent types of wavelets are considered: bandlimited wavelets (here: Shannon and Cubic Polynomial
(CuP)) as well as non-bandlimited ones (in our case: Abel-Poisson). For these types wavelet coecients are
computed and wavelet variances are given. The data format of the wavelet coecients is also included.
Keywords: Multiscale Modeling, Wavelets, Wavelet Variances, Wavelet Coecients, Gravitational Field Model
Conversion
1 INTRODUCTION
During the last years spherical wavelets have been brought into existence (cf. e.g. [3], [4], [5] and the references
therein). It is time to apply them to well-known models in order to oer easy access to the multiscale methods.
Therefore, the spherical harmonics models EIGEN-1S, EIGEN-2, GGM01, UCPH2002 0.5, EGM96 and also
EIGEN-GRACE01S are transformed into bilinear wavelet models (see [3], [4] or [6]) and the coecients of these
models are made available by the Geomathematics Group via the worldwide web. Moreover, the very same
coecients enable a multiscale reconstruction of the geoid, the gravity disturbance, the gravity anomalies and
the vertical gravity gradients by special types of reconstructing wavelets.
Our wavelets are constructed from so-called scaling functions, i.e. kernels that depend on a scale j, of the form

j
(x, y) =

n=0
n=1

j
(n)
2n + 1
4R
2
_
R
2
|x||y|
_
n+1
P
n
_
x
|x|

y
|y|
_
, (x, y)
ext
R

ext
R
, (1)
where
ext
R
denotes the outer space of the sphere
R
of radius R, P
n
is the Legendre polynomial of degree n
and
j
(n) is called the symbol of the scaling function. This symbol is a sequence of numbers that determines
the shape of the scaling function and has special properties (for details see [2], [3] or [4]). These properties make
these kernels an approximation of the Dirac distribution that converges to it for j tending to innity. Wavelets
are dened by taking the dierence of two consecutive scales which is performed by the renement equation for
the symbols:

j
(n)

j
(n) =
2
j+1
(n)
2
j
(n) , n = 0, 2, 3, . . . . (2)
The primal and dual wavelets
j
,

j
are obtained by either taking the square root or applying the third
binomial formula to equation 2. Generally, they look the following way:

j
(x, y) =

n=0
n=1

j
(n)
2n + 1
4R
2
_
R
2
|x||y|
_
n+1
P
n
_
x
|x|

y
|y|
_
, (3)

j
(x, y) =

n=0
n=1

j
(n)
2n + 1
4R
2
_
R
2
|x||y|
_
n+1
P
n
_
x
|x|

y
|y|
_
. (4)
____________________________________________
Proc. Second International GOCE User Workshop GOCE, The Geoid and Oceanography,
ESA-ESRIN, Frascati, Italy, 8-10 March 2004 (ESA SP-569, June 2004)
This construction allows an approximation of level J of a potential F:
F
J
=
(2)
J
F =
(2)
J
0
F +
J1

j=J
0

j
(
j
F) =
J
0
(
J
0
F) +
J1

j=J
0

j
(
j
F), (5)
where denotes the convolution in L
2
(
R
). These convolution integrals can be discretized for numerical
evaluation by methods presented in [1] or [4]. For the wavelet coecients that we oer to other groups we chose
the equiangular grid discussed in e.g. [1].
The important examples that we used for the computation of our multiscale representations are the Shannon
type, the CuP type, and the Abel-Poisson type wavelets:
1.1 Shannon Wavelets
In the case of Shannon scaling functions the symbol
j
(n) reads as follows

SH
j
(n) =
_
1 for n [0, 2
j
)
0 for n [2
j
, ),
(6)
and for the corresponding wavelets we choose the P-scale version to resolve the renement equation (2), i.e.

SH
j
(n) =
SH
j
(n) =
_
_

SH
j+1
(n)
_
2

SH
j
(n)
_
2
. (7)
1.2 Cubic Polynomial (CuP) Wavelets
In the CuP case the symbol takes the following form:

CP
j
(n) =
_
(1 2
j
n)
2
(1 + 2
j+1
n) for n [0, 2
j
)
0 for n [2
j
, )
(8)
and for the corresponding wavelets we apply again the P-scale version.
1.3 Abel-Poisson Wavelets
For the Abel-Poisson scaling function the symbol takes the following form
AP
j
(n) = e
2
j
n
, n [0, ), with
some constant > 0. We choose = 1. Since
AP
j
(n) = 0 for all n N this symbol leads to a non-bandlimited
kernel. It should be noted that the Abel-Poisson scaling function has a closed form representation which allows
the omission of a series evaluation and truncation, and when constructing bilinear Abel-Poisson wavelets we
want to keep such a representation as an elementary function. Thus, we decide to use M-scale wavelets whose
symbols are deduced from the renement equation (2) by the third binomial formula:

AP
j
(n) =
_

AP
j+1
(n)
AP
j
(n)
_
,

AP
j
(n) =
_

AP
j+1
(n) +
AP
j
(n)
_
. (9)
Since the Abel-Poisson scaling function and its corresponding wavelets are non-bandlimited we obtain just a
good approximation by the numerical integration method based on an equiangular grid (we choose the parameter
of polynomial exactness suciently large enough).
2 MULTISCALE REPRESENTATION OF THE GRAVITATIONAL POTEN-
TIAL
The Earths gravitational potential V in a point x of the outer space of
R
, i.e. the gravity potential W
without the part caused by centrifugal force, possesses the following representation by convolutions with
scaling functions and wavelets:
V (x) =
_

R
(
j
0
V )(y)
j
0
(x, y)d(y) +
J
max
1

j=j
0
_

R
WT
j
(V ; y)

j
(x, y)d(y), (10)
where J
max
is some suitably chosen maximal level of approximation and WT
j
(V ; y) = (
j
V )(y) denotes the
wavelet transform of V at scale j in the point y. In discrete form we get:
V (x) =
GM
R
N
j
0

i=1
w
j
0
i
v

j
0
i

j
0
(x, y
j
0
i
) +
GM
R
J
max
1

j=j
0
N
j

i=1
w
j
i
v

j
i

j
(x, y
j
i
) . (11)
The weights of the integration are named w
j
0
i
, w
j
i
and the corresponding knots are y
j
0
i
, y
j
i
. The scaling function
coecients v

j
0
i
and the wavelet coecients v

j
i
result from the convolutions (12):
GM
R
v

j
0
i
= (
j
0
V )(y
j
0
i
) ,
GM
R
v

j
i
= WT
j
(V ; y
j
i
) = (
j
V )(y
j
i
). (12)
Fig. 1 and 2 show examplarily a part of the multiscale resolution of (10) where the details (Fig. 2) are added
to the approximation of scale 7 (Fig. 1). By subtracting the non-centrifugal part of the ellipsoidal normal
Figure 1: Approximation at Scale 7 of V from
EIGEN2 using the CuP scaling function, [m
2
/s
2
].
Figure 2: Wavelet detail at scale 7 of V from EIGEN2
using the CuP wavelet, [m
2
/s
2
].
potential V
ell
= U from V the disturbing potential T = V V
ell
can be obtained (see [7], [8] or [9]) and
this subtraction can be performed for the coecients v

j
0
i
, v

j
i
in order to obtain a multiscale representation
of the disturbing potential similar to (11), but with coeents t

j
0
i
and t

j
i
that are related to v

j
0
i
, v

j
i
by the
equations (13):
GM
R
t

j
0
i
=
GM
R
v

j
0
i
(
j
0
V
ell
)(y
j
0
i
) ,
GM
R
t

j
i
=
GM
R
v

j
i
(
j
V
ell
)(y
j
i
) (13)
A more detailed derivation can be found in [2].
3 FUNCTIONALS OF THE DISTURBING POTENTIAL
By virtue of the Bruns formula N = T/ (cf. [7], [8] or [9]) a multiscale representation of the geoid undulations
can be computed from the multiscale decomposition of the disturbing potential, i.e. from the coecients t

j
0
i
and t

j
i
. Thereby, the normal gravity is taken spherically as =
GM
R
2
. Thus, the geoid heights N are described
by
N(x) = R
N
j
0

i=1
w
j
0
i
t

j
0
i

j
0
(x, y
j
0
i
) +R
J
max
1

j=j
0
N
j

i=1
w
j
i
t

j
i

j
(x, y
j
i
) . (14)
In Fig. 3 and 4 parts of a multiresolution of N are presented (see [2] for a full multiresolution).
Figure 3: Multiscale geoid heights N at Scale 7 from
EGM96 using the CuP scaling function, [m].
Figure 4: Wavelet detail at scale 7 of geoid heights N
from EGM96 using the CuP wavelet, [m].
By denition the gravity disturbances correspond to the negative rst radial derivative of T which leads to the
multiscale representation (15):
g(x) =
GM
R
_
_
N
j
0

i=1
w
j
0
i
t

j
0
i

r
x

j
0
(x, y
j
0
i
) +
J
max
1

j=j
0
N
j

i=1
w
j
i
t

j
i

r
x

j
(x, y
j
i
)
_
_
=
GM
R|x|
N
j
0

i=1
w
j
0
i
t

j
0
i

g
j
0
(x, y
j
0
i
) +
GM
R|x|
J
max
1

j=j
0
N
j

i=1
w
j
i
t

j
i

g
j
(x, y
j
i
) (15)
with

g
j
0
(x, y
j
0
i
) = |x|

r
x

j
0
(x, y
j
0
i
) =

n=0
n=1

g
j
(n)
2n + 1
4R
2
_
R
2
|x||y
j
0
i
|
_
n+1
P
n
_
x
|x|

y
j
0
i
|y
j
0
i
|
_
(16)
where
g
j
(n) = (n + 1)
j
(n) and the corresponding reconstructing wavelets

g
j
are constructed by applying
the symbol

g
j
(n) = (n + 1)

j
(n).
Analogously, the multiscale descriptions of the gravity anomalies g = g
2
|x|
T and the vertical gravity
gradients g
r
=

2
T
r
2
assume the following shape:
g(x) =
GM
R|x|
N
j
0

i=1
w
j
0
i
t

j
0
i

g
j
0
(x, y
j
0
i
) +
GM
R|x|
J
max
1

j=j
0
N
j

i=1
w
j
i
t

j
i

g
j
(x, y
j
i
) , (17)
g
r
(x) =
GM
R|x|
2
N
j
0

i=1
w
j
0
i
t

j
0
i

g
r
j
0
(x, y
j
0
i
) +
GM
R|x|
2
J
max
1

j=j
0
N
j

i=1
w
j
i
t

j
i

g
r
j
(x, y
j
i
) , (18)
where

g
j
0
(x, y
j
0
i
) =

n=0
n=1

g
j
(n)
2n + 1
4R
2
_
R
2
|x||y
j
0
i
|
_
n+1
P
n
_
x
|x|

y
j
0
i
|y
j
0
i
|
_
, (19)

g
r
j
0
(x, y
j
0
i
) = |x|
2

2
r
2
x

j
0
(x, y
j
0
i
) =

n=0
n=1

g
r
j
(n)
2n + 1
4R
2
_
R
2
|x||y
j
0
i
|
_
n+1
P
n
_
x
|x|

y
j
0
i
|y
j
0
i
|
_
(20)
with the symbols

g
j
(n) =
g
j
(n) 2
j
(n) = (n 1)
j
(n) ,
g
r
j
(n) = (n + 1)(n + 2)
j
(n) , (21)
and corresponding symbols

g
j
(n) and

g
r
j
(n) for the respective wavelets. As an example we show in Fig. 5
and 6 for the gravity anomalies g again scale and detail 7 with CuP scaling functions or wavelets, respectively.
Figure 5: g at Scale 7 from EGM96 using CuP,
[mgal].
Figure 6: Wavelet detail at scale 7 of g from EGM96
using CuP, [mgal].
4 WAVELET VARIANCES
The distribution of space-dependent signal energy of the disturbing potential is described by the scale and space
variances of T at scale j and point x (see [2], [5]):
Var
j ; x
(T) =
2
j ; x
(T) =
_

R
_

R
T(y)T(z)
j
(y, x)
j
(z, x)d(y)d(z) . (22)
Using the wavelet coecients t

j
i
one can compute this quantity with the help of a Shannon kernel SH( , ) as

j ; x
(T) =
GM
R

j ; x
,
2
j ; x
=
_
_
N
j

i=1
w
j
i
t

j
i
SH(y
j
i
, x)
_
_
2
. (23)
The wavelet variances of the geoidal heights N are then given by
j ; x
(N) = R
j ; x
; and for the gravity
disturbances, the gravity anomalies, and the vertical gravity gradients the variances can be obtained by a
convolution with special Shannon kernels similar to (16), (19) and (20):

2
j ; x
(g) =
_
_
GM
R|x|
N
j

i=1
w
j
i
t

j
i
SH
g
(y
j
i
, x)
_
_
2
, (24)

2
j ; x
(g) =
_
_
GM
R|x|
N
j

i=1
w
j
i
t

j
i
SH
g
(y
j
i
, x)
_
_
2
, (25)

2
j ; x
(g
r
) =
_
_
GM
R|x|
2
N
j

i=1
w
j
i
t

j
i
SH
g
r
(y
j
i
, x)
_
_
2
. (26)
Fig. 7 to 10 demonstrate the development of the energy distribution of the disturbing potential for j = 4 to 7.
5 WAVELET COEFFICIENTS
We supply to the end-user the scaling function or wavelet coecients, v

j
0
i
or v

j
i
corresponding to the locations
of the equiangular grid on
R
as well as the integration weights w
j
0
i
, w
j
i
. The wavelet coecients are ordered as
Figure 7:
4; x
(T) (rescaled to maximal 400), [m
2
/s
2
]. Figure 8:
5; x
(T) (rescaled to maximal 250), [m
2
/s
2
].
Figure 9:
6; x
(T) (rescaled to maximal 150), [m
2
/s
2
]. Figure 10:
7; x
(T) (rescaled to max. 100), [m
2
/s
2
].
a square matrix, in addition the weights form a rst column. Comments are located in the rst 20 lines of the le
and are indicated by a % sign. The coecients, a detailed model description and further gures can be found and
downloaded at the following web page: http://www.mathematik.uni-kl.de/~wwwgeo/waveletmodels.html
References
[1] Driscoll J. R., Healy D. M., Computing Fourier Transforms and Convolutions on the 2-Sphere, Adv. Appl.
Math., Vol. 15, 202250, 1994.
[2] Fengler M. J., Freeden W., Gutting M., Darstellung des Gravitationsfelds und seiner Funktionale mit
Multiskalentechniken, Zeitschrift f ur Vermessungswesen (ZfV), 2004. (submitted)
[3] Freeden W., Multiscale Modelling of Spaceborne Geodata, BG Teubner, Stuttgart, Leipzig, 1999.
[4] Freeden W., Gervens T., Schreiner M., Constructive Approximation on the Sphere (With Applications to
Geomathematics), Oxford Science Publications, Clarendon, Oxford, 1998.
[5] Freeden W., Maier T., On Multiscale Denoising of Spherical Functions: Basic Theory and Numerical
Aspects, Electron. Trans. Numer. Anal. (ETNA), Vol. 14, 4062, 2002.
[6] Freeden W., Michel V., Multiscale Potential Theory (with Application to Earths Sciences), Birkhauser,
2004. (in print)
[7] Groten E., Geodesy and the Earths Gravity Field I, II, D ummler, 1979.
[8] Heiskanen W. A., Moritz H., Physical Geodesy, W. H. Freeman and Company, 1967.
[9] Torge W., Geodasie, Sammlung Goschen, de Gruyter, 1975.