Professional Documents
Culture Documents
Featherstone1998 Article AMeissl-modifiedVaníčekAndKleu
Featherstone1998 Article AMeissl-modifiedVaníčekAndKleu
Abstract. A deterministic modi®cation of Stokes's inte- include the area in which the geoid is desired. In other
gration kernel is presented which reduces the truncation words, when the residual Stokes integration is perform-
error when regional gravity data are used in conjunction ed over a limited area, a truncation error
dNst results in
with a global geopotential model to compute a gravi- the geoid height.
metric geoid. The modi®cation makes use of a combi- Following Rapp and Rummel (1975), the gravimetric
nation of two existing modi®cations from VanõÂ cÏek and estimate of the geoid height
N in the remove-restore
Kleusberg and Meissl. The former modi®cation applies scheme can be divided into the following three contri-
a root mean square minimisation to the upper bound of butions:
the truncation error, whilst the latter causes the Fourier
series expansion of the truncation error to coverage to N NLÿ1 Nst dNst
1
zero more rapidly by setting the kernel to zero at the where NLÿ1 is the long-wavelength contribution of the
truncation radius. Green's second identity is used to global geopotential model, complete to spherical har-
demonstrate that the truncation error converges to zero monic degree and order
L ÿ 1, and Nst is the residual
faster when a Meissl-type modi®cation is made to the geoid contribution from Stokes's formula
VanõÂ cÏek and Kleusberg kernel. A special case of this
Z 2pZ p
modi®cation is proposed by choosing the degree of R
modi®cation and integration cap-size such that the Nst S
wDgL sin w dw da
2
4pc 0 0
VanõÂ cÏek and Kleusberg kernel passes through zero at the
truncation radius. where R is the Earth's radius, c is normal gravity on the
reference ellipsoid, DgL are the residual terrestrial
gravity anomalies which have been reduced by the
Key words. Gravimetric geoid determination corresponding spherical harmonic degree and order
Modi®ed kernels Truncation error reduction
L ÿ 1 of the global geopotential model, w is the
spherical distance of each gravity anomaly from the
computation point, and S
w is the spherical Stokes
integration kernel, which is expressed as a Fourier series
of Legendre's polynomials Pn
cos w as
1 Introduction
X
1
2n 1
The routine remove-restore approach to gravimetric S
w Pn
cos w
3
n2
nÿ1
geoid computation is to combine a global geopotential
model with terrestrial gravity observations and a digital In practice, the Stokes integration in Eq. (2) is usually
terrain model about the area of interest. Formally, performed over a limited or truncated spherical cap,
Stokes's (1849) formula requires that gravity data are which is bound by the surface spherical distance w w0 .
used over the whole Earth, even to compute the residual As Stokes's integration kernel is non-zero in the region
geoid undulations relative to the geopotential model. w0 < w p, the neglect of the residual gravity anoma-
However, for many reasons the global coverage of lies in these remote zones can adversely aect the
gravity data is incomplete. As such, the computation is gravimetrically computed geoid. This comprises the
performed using only those data that surround and truncation error, and is given by
Z 2pZ p
R
dNst S
wDgL sin w dw da
4
Correspondence to: W.E. Featherstone 4pc 0 w0
155
It is this truncation error that should be reduced, reduces to Eq. (3) when w0 0 because the orthogo-
ideally to zero, in the region w0 < w p in order for the nality relations of Legendre's polynomials over the
2
correct residual geoid height to be evaluated from only a sphere show that Qn
0 nÿ1 for 2 n < 1.
truncated integration of Eq. (2). A kernel modi®cation is Equation (5) is substituted into Eq. (4) for L 2 to
used eectively to alter the shape of Stokes's kernel in give
such a way that the truncation error in Eq. (4) is reduced,
preferably to the point at which it can be safely ne- R X 1
dNst
2n 1Qn
w0
glected. Only when this condition is satis®ed can the tern 8pc n2
dNst be safely omitted from Eq. (1). Therefore, the Z 2pZ p
objective of a kernel modi®cation is to reduce the trun- Dg Pn
cos w sin w dw da
7
cation error to a level which is acceptable for modern 0 w0
geodetic applications, such as the determination of or-
thometric heights from the Global Positioning System. and Eq. 1.71 of Heiskanen and Moritz (1967) is used to
Many authors have investigated the theoretical basis replace the integral term, which yields a series expansion
for the reduction of the truncation error using modi®ed of the truncation error in terms of the surface spherical
integration kernels. These include deterministic ap- harmonics of the gravity anomalies
proaches (Molodensky et al. 1962; de Witte 1967; Wong RX 1
and Gore 1969; Meissl 1971a; Heck and GruÈninger dNst Qn
w0 Dgn
8
1987; VanõÂ cÏek and Kleusberg 1987; VanõÂ cÏek and SjoÈberg 2c n2
1991) and stochastic approaches (Wenzel 1982; SjoÈberg
1984; SjoÈberg 1991; VanõÂ cÏek and SjoÈberg 1991). The Equation (8) shows that the truncation error is a
practical application of modi®ed kernels in gravimetric function of the truncation coecients. Similarly, Eq. (6)
geoid determination and the estimation of the magni- shows that the truncation coecients are a function of
tude of the truncation error have also been made by, for the integration kernel. Therefore, a modi®cation to the
example, Rapp and Rummel (1975), Jekeli (1980, 1981), integration kernel provides the basis for reducing the
Wichiencharoen (1984), SjoÈberg (1986), Petrovskaya impact of the truncation error on any gravimetric de-
and Pishchukhina (1990), VanõÂ cÏek et al. (1990) and termination of the geoid.
Smeets (1994). This, of course, neglects any errors in the global
This paper presents a deterministic Meissl-type geopotential model, terrestrial gravity and terrain data,
modi®cation to the Molodensky-type kernel formulated and discretisation or omission errors associated with the
by VanõÂ cÏek and Kleusberg (1987), and is analogous with numerical evaluation of Stokes's formula. None of these
the combination made by Heck and GruÈninger (1987). auxiliary error sources will be considered here.
The Fourier series of the truncation error is de®ned,
followed by a summary of the relevant deterministic
modi®cations of Wong and Gore (1969), VanõÂ cÏek and 3 The relevant deterministic kernels
Kleusberg (1987) and Meissl (1971a). An approach
similar to that taken by Meissl is subsequently applied 3.1 Wong and Gore's (1969) modi®cation
to the already-modi®ed kernel of VanõÂ cÏek and Kleus-
berg. Green's second identity is then used to demon- Equation (3) shows that the unmodi®ed Stokes kernel
strate that the Fourier series of the truncation error can be expressed as a Fourier series of Legendre's
associated with this hybrid modi®ed kernel converges to polynomials from 2 to 1. When a global geopotential
zero faster when the kernel is set to zero at the trunca- model of spherical harmonic degree and order
L ÿ 1 is
tion radius w0 . included in Eq. (2), the low-degree terms from 2 to
L ÿ 1, inclusive, are no longer required in Stokes's
integral due to the orthogonality of spherical harmonics
2 The basis of kernel modi®cation over the sphere. However, if the integration is only
performed over a spherical cap of radius w0 , this
Using the approach of Molodensky et al. (1962), Eq. (3) condition breaks down. Therefore, these low-degree
is rede®ned in the region w0 < w p as harmonics must be removed in some way, otherwise the
unmodi®ed Stokes formula will distort the long-wave-
X
1
2n 1 length contribution to the geoid.
S
w Qn
w0 Pn
cos w
5 This can be achieved by removing the corresponding
n2
2
degree and order of gravity anomalies implied by a
where global geopotential model, as is the usual practice in the
Z remove-restore technique. Alternatively, the low-degree
p
Legendre polynomials
2 n < L can be removed from
Qn
w0 S
wPn
cos w sin w dw
6
w0 the Stokes kernel (de Witte 1967; Wong and Gore 1969).
A combination of these can also be used, where the
which de®nes the Molodensky truncation coecients, degree of kernel modi®cation
M ÿ 1 is not necessarily
and whose numerical values can be computed using the equal to the degree and order of the global geopotential
recursive algorithms given by Paul (1973). Also, Eq. (5) model
L ÿ 1. However, either L or M must at least
156
correspond to the maximum degree and order of the kS M
wkkDgL k kS
wkkDgk for w > w0 :
global geopotential model utilised.
The removal of the Legendre polynomials from Eq. Therefore, the truncation error for a combination of
(3) introduces a kernel that is higher than second-degree, a low-degree modi®ed integration kernel and a high-
and has been called the spheroidal Stokes kernel by degree
L > M global geopotential model is
VanõÂ cÏek and Kleusberg (1987) and VanõÂ cÏek and SjoÈberg Z 2pZ p
R
(1991); also see de Witte (1967) and Wong and Gore dNLM S M
wDgL sin w dw da
(1969). This is 4pc 0 w0
X RX 1
Mÿ1
2n 1 QM
w Dgn
S M
w S
w ÿ Pn
cos w
9 2c nL n 0
n2
nÿ1 " #
RX 1 X
M ÿ1
2k 1
When using this spheroidal Stokes kernel in Eq. (2), Qn
w0 ÿ enk
w0 Dgn
13
the truncation coecients (Eq. 6) become the spheroidal 2c nL k2
kÿ1
truncation coecients, and are given by
Z p where
Z
QMn
w 0 S M
wPn
cos w sin w dw
10 p
w0 enk
w0 Pk
cos wPn
cos w sin w dw
14
w0
The corresponding truncation error, in spatial and
spectral form, respectively, is and whose numerical values can be determined using the
recurrence relation given in Paul (1973).
Z 2pZ p
R In the following discussions only a high-degree global
dN M S M
wDg sin w dw da geopotential model
L ÿ 1 in conjunction with a low-
4pc 0 w0
degree kernel modi®cation
M ÿ 1 will be considered.
RX 1
QM
w Dgn
11
2c n2 n 0
3.2 VanõÂcÏek and Kleusberg's (1987) modi®cation
Simply by observing Fig. 3 in VanõÂ cÏek and Kleusberg
(1987), one can see that the spheroidal kernel (Eq. 9) In a similar manner to Molodensky et al. (1962), who
tapers o to zero for smaller truncation radii
w0 . make a modi®cation to the spherical Stokes kernel (cf.
Moreover, kS M
wk kS
wk for w > w0 . Therefore, Eqs. 5 and 8), VanõÂ cÏek and Kleusberg (1987) make a
the impact of the truncation error is reduced, as is the modi®cation to the spheroidal Stokes kernel (Eq. 9).
requirement for integration over the whole Earth. This approach minimises the upper bound of the
However, if higher degrees of Legendre polynomials are truncation error in a root mean squares sense (VanõÂ cÏek
removed from the kernel, it can act as a high-pass ®lter and SjoÈberg 1991).
and thus prevent the inclusion of intermediate wave- The VanõÂ cÏek and Kleusberg kernel is de®ned by
length geoid information, which is strongly suspected to
be in error in the geopotential model. X
M ÿ1
2k 1
M
Svk
w S M
w ÿ tk
w0 Pk
cos w
15
Alternatively, the truncation error can be reduced by 2
k2
subtracting the global geopotential model-derived
gravity anomalies from the terrestrial gravity anomalies and the VanõÂ cÏek and Kleusberg (or spheroidal Mo-
used in Eq. (2). This approach has already been referred lodensky) truncation coecients, QM n
w0 , and the
to as the remove-restore technique, and the corre- modi®cation coecients, tk
w0 , are determined via
sponding truncation error is
X
Mÿ1
2k 1
Z 2pZ p QM
R n
w0 tk
w0 enk
w0
dNL S
wDgL sin w dw da k2
2
4pc 0 w0
X
Mÿ1
2k 1
RX 1
Qn
w0 ÿ enk
w0
Qn
w0 Dgn
12 k2
kÿ1
2c nL Z p
M
Svk
wPn
cos w sin w dw
16
and is thus reduced because kDgL k kDgk for w > w0 , w0
which indicates the use of the remove-restore technique as
a means to reduce the truncation error to a certain extent. Equations (16), (14) and (6) produce
M ÿ 2 equations
To avoid the possibility of high-pass ®ltering by a that are linear in tk
w0 . Once the integration cap-size
high-degree modi®ed kernel, a preferable approach is to
w0 is selected, the tk
w0 coecients can be evaluated
use only a high-degree global geopotential model in for 2 k < M using matrix inversion. These coecients
conjunction with a low-degree spheroidal Stokes kernel are then used in Eq. (15) to compute the VanõÂ cÏek and
(cf. Jekeli 1981). This approach still reduces the trun- Kleusberg modi®ed kernel, and thence the geoid.
cation error on the gravimetric geoid because The corresponding truncation error for M L is
157
Z 2pZ p
L R of the Meissl truncation error (cf. Jekeli 1981; Heck and
dNvk S M
wDgL sin w dw da GruÈninger 1987). Consequently, Meissl's modi®cation
4pc 0 w0 vk
to the Stokes kernel causes the truncation error to
RX 1
converge to zero at a greater rate, especially when used
QM
w0 Dgn
17
2c nL n in conjunction with the remove-restore approach.
The unit normal
^ n to the boundary b is in the di- It is argued that if the integration kernel is set to
rection of decreasing w (i.e. towards the geoid compu- zero at the truncation radius, the ®rst term on the right-
tation point). Therefore, hand side of Eq. (26) vanishes. Most importantly, this
M
guarantees that the truncation error converges to zero
M ÿoSvk
w0 according to n
n1 1
as opposed to n1 1
as n ! 1.
rSvk
w0 n^
23
ow M
Therefore, under the condition Svk
w0 0, Eq. (26)
and becomes
M
ÿoPn
cos w0 ÿ1 oSvk
w0
rPn
cos w0 n^
24 Q0M
n
w 0 Pn
cos w0 sin w0
ow n
n 1 ow
Z p
where here and in what follows the notation r2 Svk
M
w Pn
cos w sin w dw
28
w0
og
w0 og
w
The truncation error associated with this hybrid mod-
ow @w
i®ed integration kernel (Eq. 20) is
is used for an arbitrary function g
w which is usually Z 2pZ w0
M
Svk
w or Pn
cos w. L R M
H R 2p dNmvk Svk
w0 DgL sin w dw da
The change of variable b db 0 sin w0 da, where 4pc 0 0
Z 2pZ p
sin w0 is the linear radius of the integration cap, is also R
applied to the right-hand side of Eq. (22). As the inte- S M
wDgL sin w dw da
29
4pc 0 w0 vk
gration kernel and Legendre polynomials are both iso-
tropic functions of w only, the integration of Eq. (22) with which is expressed in spectral form as
respect to azimuth
a can be performed, which yields
Z p L RX 1
Pn
cos w r2 SvkM
w dNmvk Q0M
w0 Dgn
30
M
Svk
wPn
cos w sin w dw 2c nL n
w0 n
n 1
Therefore, the series expansion of the truncation er-
sin w0 M oPn
cos w0
Svk
w0 ror converges to zero at a greater rate when the Meissl-
n
n 1 ow type modi®cation is made to the VanõÂ cÏek and Kleusberg
M
oSvk
w0 kernel. Moreover, the use of a higher than second-de-
ÿ Pn
cos w0
25
ow gree reference ®eld, as is required by VanõÂ cÏek and SjoÈ-
berg (1991), ensures that the low-degree harmonics in
Substituting Eq. (16) in Eq. (25) de®nes the VanõÂ cÏek and the gravity ®eld are absent, thus allowing this rapidly
Kleusberg truncation coecients as converging series to start from L
M.
" The Meissl-modi®ed VanõÂ cÏek and Kleusberg trunca-
1 oPn
cos w0 M tion coecients, Q0M
QM n
w0 , are derived as follows: ®rstly,
n
w0 Svk
w0 sin w0
n
n 1 ow the error kernel associated with this modi®cation
M scheme is de®ned as
oSvk
w0
ÿ Pn
cos w0 sin w0 8 M
ow < Svk
w0 for 0 w w0
Z p #
M
2 M Kmvk
w; w0
31
ÿ r Svk
w Pn
cos w sin w dw
26 : M
w0
Svk
w for w0 < w p
The second and third terms on the right-hand side of such that
1
Eq. (26) converge to zero according to n
n1 as n ! 1. Z p
However, the behaviour of the ®rst term is less Q0M
n
w0 M
Kmvk
w; w0 Pn
cos w sin w dw
0
straightforward because the ®rst derivatives of Legend- Z w0
re's polynomials do not necessarily converge to zero as M
Svk
w0 Pn
cos wsin w dw QM
n
w0
32
rapidly. This is illustrated through the following recur- 0
sive relationship between Legendre's polynomials
Using Eqs. (21) and (23) in Eq. (32), and noting that the
oPn
cos w ÿn M
term Svk
w0 is constant, yields
Pnÿ1
cos w ÿ cos w Pn
cos w
ow sin w
M
Svk
w0 sin w0 oPn
cos w0
which when evaluated at w w0 gives Q0M M
n
w0 Qn
w0 ÿ
33
n
n 1 ow
oPn
cos w0 ÿn
Pnÿ1
cos w0 ÿ cos wPn
cos w0 Then, using the recursive relationship for Legendre's
@w sin w0
polynomials (Eq. 27) gives the spectral form of the
27 truncation error (Eq. 30) as
159
L RX 1 M
Svk
w0 zero faster by forcing the kernel to zero at the truncation
dNmvk Pnÿ1
cos w0 radius.
2c nL
n 1
The practical implementation of this new modi®ca-
ÿ cos w0 Pn
cos w0 QM tion is relatively straightforward by using Eq. (20) to
n
w0 Dgn
34
replace the Stokes kernel in Eq. (2). The VanõÂ cÏek and
Kleusberg kernel is determined using Eqs. (9), (15), (14)
which is analogous with Eq. (19). and (16). Alternatively, the degree of spheroidal modi-
®cation or the integration cap-size, or both, can be se-
lected in such a way that the VanõÂ cÏek and Kleusberg
4.2 The selection of w0 and M kernel passes through zero at the truncation radius.
The validity of this Meissl-modi®cation to the VanõÂ -
From Eq. (31), the residual geoid solution inside the cÏek and Kleusberg kernel is veri®ed through the appli-
integration cap may become biased as a consequence of cation of Green's second identity, which clearly shows
M
subtracting the constant value Svk
w0 . However, this that the truncation error is diminished at a greater rate
should not be the case, as the zero-degree harmonics are when using this approach.
absent from both the gravity anomalies and geoid by
virtue of the formulation of Stokes's boundary-value Acknowledgements. This research was supported by Australian
problem and Bruns's equation. As such, this bias in the Research Council grant AA49331318 and an Academic Links and
geoid is dealt with dierently by the use of geometrical Interchange Scheme grant from the British Council to Will
constraints (Heiskanen and Moritz 1967). Featherstone; both are gratefully acknowledged. We would also
Nevertheless, this question of a biased residual geoid like to thank the three reviewers for their time taken to consider
solution can be eliminated by choosing a condition such this manuscript.
that the Meissl-modi®ed VanõÂ cÏek and Kleusberg kernel
passes through zero at the truncation radius. This avoids
M
the need to subtract the value of Svk
w0 according to References
Eq. (20), whilst still eliminating the second term on the
right-hand side of Eq. (34), and thus maintains the rapid de Witte L (1967) Truncation errors in the Stokes and Vening
convergence of the truncation error series de®ned by Eq. Meinesz formulae for dierent order spherical harmonic gravity
(28). terms. Geophys J R Astron Soc 12: 449±464
Heck B, GruÈninger W (1987) Modi®cation of Stokes's integral
This can be achieved in practice by choosing values formula by combining two classical approaches. IUGG General
of M and w0 to de®ne the coecients tk
w0 in such a Assembly, Vancouver
way that when used in Eq. (15) the kernel is zero at the Heiskanen WH, Moritz H (1967) Physical geodesy. Freeman, San
truncation radius. However, this requires that some it- Francisco
eration be performed as the tk
w0 coecients are Jekeli C (1980) Reducing the error in geoid undulation computa-
themselves functions of w0 . This approach is similar to tions by modifying Stokes's function. OSU Rep 301, Dept Geod
Sci Surv, Ohio State University, Columbus
that presented by Heck and GruÈninger (1987) where the Jekeli C (1981) Modifying Stokes's function to reduce the error of
Wong and Gore (1967) modi®cation is chosen for some geoid undulation computations. J Geophys Res 86(B8): 6985±
particular degree of modi®cation M such that the ®rst 6990
zero point of the spheroidal kernel (Eq. 9) coincides with Meissl P (1971a) Preparations for the numerical evaluation of
the truncation radius, or vice versa. second-order Molodensky-type formulas. OSU Rep 163, Dept
Geod Sci Surv, Ohio State University, Columbus
Meissl P (1971b) On the linearisation of the geodetic boundary
value problem. OSU Rep 152, Dept Geod Sci Surv, Ohio State
5 Summary and conclusion University, Columbus
Molodensky MS, Eremeev VF, Yurkina MI (1962) Methods for
study of the external gravitational ®eld and ®gure of the Earth.
Modi®cations to Stokes's formula are necessary in order
Israeli Programme for the Translation of Scienti®c Publications,
to reduce the truncation error which exists when gravity Jerusalem
data are used over a limited spherical cap to compute Paul MK (1973) A Method of evaluating the truncation error co-
the geoid. A proportion of this truncation error can be ecients for geoidal height. Bull Geod 47: 413±425
reduced by using a global geopotential model via the Petrovskaya MS, Pishchukhina KV (1990) Methods for compact
remove-restore technique. However, an error remains approximation of geoid height. Manuscr Geod 15: 253±260
Rapp RH, Rummel R (1975) Methods for the computation of
due to the neglect of residual gravity anomalies in the
detailed geoids and their accuracy. OSU Rep 233, Dept Geod
remote zones outside the integration cap. Sci Surv, Ohio State University, Columbus
In this paper, three deterministic modi®cations to the SjoÈberg LE (1984) Least-squares modi®cation of Stokes's and
original Stokes integration kernel have been combined Vening Meinesz's formula by accounting for truncation and
in order to reduce the truncation error still further. The potential coecient errors. Manuscr Geod 9: 209±229
Wong and Gore modi®cation suppresses the low-degree SjoÈberg LE (1986) Comparison of some methods of modifying
Stokes's formula. Boll Geod Sci A 45: 229±248
Legendre polynomials from Stokes's kernel, the VanõÂ cÏek
SjoÈberg LE (1991) Re®ned least-squares modi®cation of Stokes's
and Kleusberg modi®cation applies a root mean square formula. Manusr Geod 16: 367±375
minimisation to the upper bound of the truncation er- Smeets I (1994) An error analysis of the height anomaly deter-
ror, whilst the Meissl modi®cation causes the truncation mined by a combination of mean terrestrial gravity anomalies
error, when expressed as a Fourier series, to converge to and a geopotential model. Boll Geod Sci A 53: 57±96
160
Stokes GG (1849) On the variation of gravity on the surface of the Wenzel H-G (1982) Geoid computation by least-squares spectral
Earth. Trans Camb Phil Soc 8: 672±695 combination using integration kernels. In: (ed) Proc IAG
VanõÂ cÏek P, Kleusberg A (1987) The Canadian geoid ± Stokesian General Meeting, Tokyo. pp 438±453
approach. Manuscr Geod 12: 86±98 Wichiencharoen C (1984) A comparison of gravimetric geoid un-
VanõÂ cÏek P, SjoÈberg LE (1991) Reformulation of Stokes's theory for dulations computed by the modi®ed Molodensky truncation
higher than second-degree reference ®eld and modi®cation of method and the method of least-squares spectral combination
integration kernels. J Geophys Res 96(B4): 6529±6539 by optimal integration kernel. Bull Geod 58: 494±509
VanõÂ cÏek P, Zhang C, Ong P (1990) Computation of a ®le of geoidal Wong L, Gore R (1969) Accuracy of geoid heights from modi®ed
heights using Molodensky's truncation method. UNB Rep 147, Stokes kernels. Geophys J R Astron Soc 18: 81±91
Dept Surv Eng, University of New Brunswick