Journal of Geodesy (1996) 70:214-223
Geodesy
Spherical harmonic analysis, aliasing, and filtering
Christopher Jekeli
Department of Geodetic Science and Surveying, The Ohio State University, 1958 Neil Ave., Columbus, OH 43210, USA
Received 5 October 1994; Accepted 14 June 1995
Abstract. The currently practiced methods of harmonic
analysis on the sphere are studied with respect to aliasing
and filtering. It is assumed that a function is sampled on
a regular grid of latitudes and longitudes. Then,
transformations to and from the Cartesian plane yield
formulations of the aliasing error in terms of spherical
harmonic coefficients. The following results are obtained:
1) The simple quadratures method and related methods are
biased even with band-limited functions. 2) A new
method that eliminates this bias is superior to Colombo's
method of least squares in terms of reducing aliasing. 3)
But, a simple modification of the least-squares model
makes it identical to the new method as one is the dual of
the other. 4) The essential elimination of aliasing can
only be effected with spherical cap averages, not with the
often used constant angular block averages.
Introduction
The harmonic (or spectral) representation of functions
on the sphere has proved useful in modeling the gravity
field, the magnetic field, the topography, the ocean tides,
etc, not just for the Earth, but for any roughly spherical
planet (Rapp 1977; Schmitz and Cain 1983; Kaula 1993).
Such harmonic representations enable a more precise
interpretation by wavelength of their magnitudes, their
coherence, and their measurability. Usually, these
functions are richly endowed with harmonics to a high
degree, decaying relatively slowly with increasing detail.
For example, the Earth's gravity field, after the first few
harmonics (which certainly dominate), decays
approximately linearly on the logarithmic scale, and
Journal of
© Springer-Verlag 1996
today's representations include over 130 thousand terms
(Rapp and Pavlis 1990).
It is a significant consequence of this fine structure in
the functions that the harmonic components determined
from a finite set of function values are biased estimates of
the true components. This is a well known phenomenon
in signal analysis, where even though one applies
orthogonal operators to the data, they cannot filter out
harmonics finer in detail than that dictated by the
sampling interval. This is known as "aliasing." Another
interpretation, though less precise, is that of attempting to
determine all (possibly an infinite number of) harmonics
from a given number of function values. In the general
case, this is an underdetermined problem and the solution
invariably lumps together harmonics in a prescribed way.
Spectral aliasing in functions defined on the line or on the
Cartesian plane is well understood; it is less obvious on a
non-Euclidean surface such as the sphere.
The spherical harmonic analysis is examined here
with the object of clearly understanding the effect of
aliasing in conventional techniques, as well as in the
modem techniques developed by Colombo (1981) that are
used to compute today's models. In so doing, a new
formula for the analysis is rendered that reduces the
aliasing effect better than the least-squares approach. It is,
however, also shown that the new technique is equivalent
to a modification of the least-squares model, requiring
some extra computations. The results also indicate clearly
which types of averaging (filtering) do and do not
eliminate aliasing. It is assumed throughout that the data
are provided discretely at every node of a uniform grid.
(Aliasing in the case of a nonuniform data distribution
was studied by Sansb 1990.) Though the errors in
measuring the function on the sphere are not considered
initially, their effect on the derived procedures is discussed
in conclusion.
 215

Preparatory Concepts
The infinite sum on the far fight in (4) is an alias of G k
A periodic function integrable over its period may be for any frequency, k, such that Ikl < k N = I(/2, where kr~ is
represented either in terms of its independent (space or the so-called Nyquist frequency. The determination of the
time) variable or, where it is continuous, in terms of its spectnma {Gk} of the function g from the discrete sequence
Fourier spectrum, being the set of coefficients in its {gs}' using (2b), is subject to an aliasing error as
corresponding Fourier series (Priestley 1981): formulated in (4). Note that if the function g does not
contain harmonics with frequency above the Nyquist
g(x) = ~ Gk e i2=kxrr , xE R (la) frequency, then there is no aliasing error. Such functions
k~-~ will be called band-limited. I t is further noted that a
function with infinite bandwidth can be filtered to make it
I"T
Gk = 1J0 g(x) e-i2=kx/rdx , ke Z (lb) band-limited. This procedure is discussed later with
respect to the particular application of spherical harmonic
analysis.
where i2 = -1, T is the period (g(x+T) = g(x)), R i s the set Obviously, similar formulas hold for periodic
of real numbers, Z is the set of integers, and the integer k functions defined on the Cartesian plane. They are
may be termed "frequency," or "wavenumber." In practice, specialized here to the case where g is a real function of
for example from measurements, one knows only discrete two variables, 0 and X, and is periodic in 0 with period
values of g and the corresponding spectrum is the Discrete and in k with period 2re. Then
Fourier Transform (DFT):
K-I o g(O,~.) = ~ ~ Gk,me i(2ke+ r~) , e,~.e R (5a)
gs = ~ Gk e i2=ks/K , s = 0 ..... K-1 (2a) k=-~ m---~
k=0
2~
o K-1 t f
G k = A x Z gs e -i2~ks/K , k = 0 ..... K - 1 (2b) Gk m = _-:-~-| I g(0,k) e -i(2k°+ m~)d0dX , k,me Z
T s=0 • "-~2nJ0 J0
where it is assumed that the function values are known on (5b)
a regular grid, i.e., one with constant spacing, Ax = T/K.
where, with * denoting the complex conjugate,
The unsymmetric transform (2a,b), where k is
nonnegative, is preferred here as being more conventional Gk,-m= G*-lqm (5c)
and avoiding the separate treatment of even and odd K.
Since the DFT is periodic with frequency period K: For a regular grid with constant spacings given by A0 =
o o ~/K and AX = 2~/M, where 0 s = sA0, ~'t = tAX, the
Gk+K = Gk, for any k (2c) Discrete Fourier Transform pair is

it is noted that equations (2a,b) are equivalent to the K-1 M - 1 o

symmetric forms, where k (and s) ranges from -(K-I)12 to gs,t = g(0s,~,t) = k~--0=m=0ZGk,m e i21t(sk/K + tin/M) ,
(K-1)12 (if K is odd). In the notation of (2a, b), the (6a)
frequencies from (K+I)/2 to K-I are used to represent the S = 0 ..... K - l , t = 0 .....M-1
negative frequencies from -(K-1)/2 to -I.
o K-1 M-1
The relationship between the spectra (Ib) and (2b) can Gk,m = ~ Z Z gs,t e "i2~(sk/K+ tin/M)
be found by substituting x = sAx into (la): s=0 t=0
(6b)
k = 0 ..... K - l , m = 0 .....~ 2
g~ = g(sAx) = .~, Gj e i2~j~m
Because the values gs,t are real,
(3)
K-1 O O
= ,~, Z Gjx+k e i2g~/I~ Gk.M-m= ~ - ~ m (6:)
k=0 j=-~
which explains the limit of interest on m. M is assumed
from which, by comparing with (2a), one has for any k, to be even, without
0
loss in generality.
O, 0 O,
By the periodicity
0
in k and m, G0,m = G0~-m and Gk,0 --- GK-k,0, while ~0,0 is
O real. Hence, the complex, discrete spectrum, I G k ~ ,
G~ = .y_, GjK+~ = G~ + .y_, Gj~,~ (4) contains only KM independent coefficients. Analogous to
j~ j=-~
j#o (4), the relationship between the total spectrum (5b) and
216

the sample spectrum (6b) is, for any k and m, to the coefficients of Heiskanen and Moritz (1967, p.31,
(1-76)) by
o 0o oo

Gk,m = Gk,m + E E GuK+k,vM+m (7)

I.l-~e0
The double sum in (7) represents the aliasing error for
frequencies Ikl < K/2 and Iml < M/2.
Instead of the plane, the real function g(0,~) can also
be defined on a unit sphere and then expressed in terms of
a series of spherical harmonics rather than sinusoids.
There are many variations in such a representation. For
simplicity, the complex exponential form is used here and
its relationship to the more customary form is also given
for the reader's convenience. The spherical harmonic
series defines the complex Legendre spectrum {?n,m} of
the function as follows:
n
g(O,~,)= ~ ~ '~n,m Pn,lm¢cos0)e im~ ;
n=O m=-n
(Sa)
0_<0<n, 0 < ~,_< 2n
/~an,m - i bn,m) , m>0
~n,m = an'0 , m=0 (10)
/~an,m + i bn,m) , m < 0
Because of (8d), one need only solve for coefficients with
non-negative orders although, formally, the series must
include m < 0.
The Fourier Spectra of Pn,m and sin0 Pn,m
It will be necessary to know the Fourier spectra of
Pn,m arid of sin0. Pn,m' both defined for 0 _< 0 _ n, as
usual, and for 0 < 0 and 0 > n by periodic extension.
This means that the number of frequencies of the spectrum
in each case is infinite, even though these functions are
polynomials of sinusoids (Hobson 1965, p.96). For 0 <
0 _<re, and any integer, k, let
Pn,m(0) = Pn,m(COS0) and Pn,m(0+ krc) = Pn,m(0)

2re (11a)
- - - 1 f0
~/n,m -- f0 g(0,~,) Pn,lmgCOS0) e-im~ sin0 d0 dX ;
4~rn Sn,m(0) = sin0 Pn,m(COS0) and Sn,m(0 + KS) = Sn,m(0)

n>0, -n _.<m < n

(lib)
(8b)
In accord with (la) and (lb), the corresponding Fourier
transform pairs are
where

m:0 Pn,m(0) = E p~,m ei2k0 , 0e R (12a)

(Sc)
, m~0

and, since g is real, p~,m= Pn,m(0)e"i2k°dO , k~ Z (12b)

~n,-m = ~n,m (8(1)

~d
The integers n and m are, respectively, the degree and order
of the harmonic coefficient, Yn,m" The Pn,min (8a) an (8b)
Sn,m(0) = ~ o~,naei2k0 , 0e R (13a)
are called associated Legendre functions of the first kind, km.e~
defined, for the present purpose, for non-negative integers
n,m with m < n and for 0 _<0 _<n. Furthermore, they are
assumed to be normalized such that o~'m = sin0 Pn,m(0) e -i2k0 dO , ke Z (13b)

~- (Pn,m(COSO))2 sin0 dO =Em (9) It is noted that because Pn,mand Sn,m are real functions,

p~'m=(p.nl~m)* and o n,m

k = (on,m)* (14)
(again, for simplicity, the conventional over-bar notation
is omitted). The complex harmonic coefficients are related The orthogonality of the associated Legendre functions
 217

implies that the infinite sequences {p~,m} and {OJ.~} are Substituting (12a) and (8a) into (5b) and taking note of
orthogonal for n ~ j: (18a, b), one obtains

Z #~m (5.kn,m= 2Em 8(j,n) (15) Gk,m = Z yn,m p~,tm, (19)

k=-~ 'K n=lml
where it is further noted that Yn,m = 0 for Iml > n.
where 8(j,n) is the delta function, equal to zero unless j =
Similarly, substituting (13a) and (5a) into (8b) yields the
n, when it is one. inverse relationship
Once these coefficients are computed for n,m = 0,1
and for all integers, k, of interest, the remaining spectra
(n,m > 1) for these k can be determined using well Ok ~n,lml (20)
known recursion formulas for the associated Legendre
function (Abramowitz and Stegun 1972).
These relationships are possible because of the definitions
The discrete inverse Fourier transforms of these
(lla, b). The alternative suggested by Ricardi and Burrows
functions sampled on a uniform grid
(1972) implicitly assumes periodicity of the associated
{0s = s A0 : s = 0 ..... K-I; A0 = x/K} are
Legendre functions over (0,2x) and, though their spectral
K-1 domains are then finite, in this form they cannot be used
Pn,m;s = Z ~ , m ei2r:sk/K (16a)
k=0 with (5a) and (5b) that embody periodicity in latitude over
(0,n). On the other hand, Goldstein (1978) devised an
K-1
Sn,m;s = Z cro n,m
k ei2mk/K (16b) extension of the function g onto the larger domain, 0 < 0
k=0 < 2re, 0 < ~, _<2n, thus taking advantage of the associated
Legendre functions as finite polynomials of sinusoids. In
with (see (4)):
the present case the infinite nature of the their spectral
eo domains will not cause computational problems because
~,m
=
Z PuK+k
n,m ,
"
°n,m = Z
(~k n,m
(~wK+k
(17)
in the end only the discrete spectra (17) are needed.
u=-~ w~
Albertella and Sacerdote (1995) also developed procedures
The discrete spectra in (17) are also amenable to similar to the present ones.
computation by recursion formulas (see Appendix) since
Quadrature Formulas
the left-hand sides are simply finite sums of associated
Legendre functions. Other methods to compute these Because the function (e.g., the gravity anomaly) is
include simply determining the Discrete Fourier supposed to be given on a regular grid, all estimates of its
Transform (A.1) of the Legendre functions determined Legendre spectrum have the form of so-called quadrature
from recursion formulas, or substituting their equivalent formulas. A subset of these is the discretization of the
finite polynomial of sinusoids into (A.1). integral (8b). Numerous variations have been proposed;
for example, the simplest is
The Relationships Between the Fourier and
Legendre Spectra K-1 M-1
~n,m -- ~ - - Z Z gs,tPn,m(COS0s) sin0s e -i2rctm/M (21)
The following orthogonality relationships hold: 2Em~ s=0 t=0

1 2~
i m m' (18a)
More common is an attempt to incorporate the fact that
usually the data are averages over angular cells. In this
-"J0 , if m = m' case, the Legendre spectrum of the m e a n anomaly is
sought and the spherical harmonic function values in (21)
and are replaced by their averages over the cells. Either way,
(20) or its alternative with averaged spherical harmonics is
only an approximation and leads to estimates that are
lf0 ei2(k ' - k)0 dO =
I0,, iifl kk =¢ k'k' (18b) biased due to aliasing not just from frequency components
higher than the Nyquist limit, but from components of all
degrees greater than or equal to the order. Indeed, it is well
and for later use:
known (Pavlis 1988; Albertella et al. 1993) that a band-
limited function analyzed according to (21) is not
M-1
Z eifJ-
= JO, j*m recovered exactly by the estimated spectrum. This feature
m);Zt (18c)
t=o {M, j = m is demonstrated analytically below, but only for the case
218

exemplified by (21). The analogy for the case of cell- The first sum contains exactly K harmonic coefficients for
averages will be evident because the integrals of the any given m, and these are to be solved, given Gk;m. One
Legendre functions have spectra like (17), but multiplied could have separated the first two sums in (26) at some
by (sinkA0)/(kA0). integer larger than Iml + K, but then the problem is
With the aid of (13a) and (17), equation (21) becomes underdetern~oined because there are only K Fourier
coefficienotsGk,m for any given m.
o K-1 o n ImI .2o
yn,m- ~ Z G.~ (-Jk,m (22) Let Gm and Ym be ofinite K-vectors containing,
2~m k--O respectively, coefficients Gt,m and ~,m as follows:
The error in this estimate is derived by substituting (7)
oT [6 o °o]
into (22), adding and subtracting Tn,m' and making use of Gm = O,m, Gl,m ..... GK-1, (27)
(17), (19) and (20). Some manipulations yield
y y = [YI rrl ,m, ~ rrl +l,m..... "~rnl+K-l,m] (28)
~n'm=yn,m+ ~ --Z Z [~kdmlG-nl~lmlyj,m+
28m k=-~j~ ml Also, define the KxK matrix
(23)
0mHml ... 0ml+K'14ml
~-k ]tj ,vM+m
k=0 v=-.o j~ ml •
Rm = (29)
NOW shift the v=0 term to the first sum to get, with (15), [~.~ ml .~lml+K-l,lml
"'" VK-1
O
Tn,m = Yn,m + Then, the collection of equations (26) for a f'Lxedm and for
k = 0 ..... K - l i s

28m =lml~k=0 O

Gm = Rm Ym + 8Rm 8Tm + ARm ATm (30)

" ~ K-1 oi,lvM+mIOn, lml ,.]
=-~ J=l ml k=O
...j where 8R m, 8Ym,ARm,and AYm are semi-infinite matrices
v~0 and vectors whose elements can be inferred from (26). In
particular, 8ym contains harmonic coefficients of order m
Clearly, for band-limited data, where Yn,m = 0, n > M/2,
and degree greater than Iml + K - 1; while AYm contains
the last set of sums, where j > M/2, vanishes. But, the
all coefficients with order greater than or equal to M/2.
estimate is still biased Caliased") by the second set that
The relative domains for these coefficients are shown in
contains all harmonic coefficients with degrees j _> Iml,
Figure 1, where it is noted that K is completely
including Tn,m ! The analysis procedure derived next
independent of M.
shifts the total error of the estimate to frequencies beyond
the Nyquist frequency.

A New Method m,order //~7

Combining (7) and (19), and using (17) leads to

Gk,m = Z Z °LlvM+ml-
Irk yj,vM+m (25)
v=-~ j~ vM+m]

This is the true model linking the two-dimensional

discrete Fourier transform of the data (6b) to the actual
Legendre spectrum. It is equivalent to the model (8a) for
the nontransformed data. Again, extracting the v = 0
contribution, this can be broken into three parts:
Iml +K-1
Gk,m= Z
j~ ml
~kdml ~tJ,m+ Z
j~ ml+K
~k~ml ~tj'm
K n, degree
(26)
Fig.l: Domain of spherical harmonic coefficients estimated by (31).
+ Z j~ ml Aliasing is due to coeffieients AT and 8y in their respective domains,
v~0 see eq. (32).
 219
Solving (30) for Tm yields the following estimate of true model, only an approximation, due to the truncation.
the harmonic coefficients For a more compact notation, define the following KM-
vector
A O
Tm = R~ Gm (31)
g = {gs,t}, (35)
with an aliasing error given by
the (N - Iml)-vector (which differs from the K-vector, Tin,
by the number of elements)
Ta = -R~ (SRm 8Tin + ARm ATm) (32)
""F
'~m = [~ ml,m, '~ ml+l,m . . . . . ~N-l,m], (36)
R m is nonsingular as shown later (equation (49)). For a
band-limited function the aliasing error (32) should
and the KMx(N-Iml) matrix
vanish. Band-limited now refers to two Nyquist
frequencies, just like in the planar case: N(m) = M/2 and
- Ip. (cos a ~ e tmzt~j~r~'''''N-I (37)
N(n) = m + K. Thus, the spherical analysis of functions in-- I Jlml~ vsj Is=0,...,K-1;t=0,...,M-1
having harmonics (n,m) that are nonzero only for n <
N(n) and m < N(m) will be perfect, without aliasing, The tildes (~) are included now to differentiate between
according to (31). The use of the quadratures formula (21) this and the following modification to the approach. With
would introduce errors as given by the first double sum in (35), (36), and (37), the model (34) becomes
(24).
N-1
It is convenient, especially in efforts to eliminate the g -- X Ym ~'m (38)
aliasing effect, to refer the Nyquist limit to a single m=-N+l
frequency in terms of the degree, n. This is N =
min(K,M/2), where usually, of course, K=M/2. This, Note that there are KM data (observations), but only
however, understates the number of recoverable spectral approximately N 2 coefficients (if K=M/2, then N 2 =
componentos. That is, from the KM data, gs,t, that KM/2). The true model (33) can be written like (38) as
determine Gr~ (31) yields exactly KM real coefficients follows:
contained in the complex numbers Y'm, 0 < m < M/2; see N-1
Figure 1. Note that for each such m, the estimate (31)
requires the inversion of a KxK matrix.
g =m=-N+l
E (YmT'm + 8~m~Y~m)+ I mlX>-y AYmAym (39)
N
The Least-Squares Solution where ~Yrn is a semi-infinite matrix with KM rows and its
It is now shown that the procedure above is similar to elements are spherical harmonic functions with degrees n
the least-squares approach developed by Colombo (1981) >_N. 5~m contains coefficients with degree n > N, ATm is
and implemented by Hwang (1993). However, there is a the same as in (30) and AY m can be infen'ed from (33).
subtle difference that is easily eliminated. Equation (8a), By the discrete orthogo~nality of the complex
with summations reversed, is rewritten as follows to exponential (18c), the matrices Ym are also orthogonal:
separate symbolically the coefficients to be estimated:
(~*m)XYj = 0 , j;~m (40)
g(0s,Xt) =
Multiplying (38) by (Y'*m)T and denoting the normal matrix
_l~m + E E + E (Tn,mPn, lm[cOs0s)eim~'t)
~m=-N+l n- I m=-N+l n=N IInl>_N = I by Ain = (Y"*m;~m yields the "least-squares" solution:
(33)
A ~ol 1""* V
~'m = Am ~Ym] g (41)
The usual model for the least-squares approach is the
truncated spherical harmonic series, being the first part in
(33): The aliasing error is obtained by substituting the true
model (39) and duly noting the orthogonalities:
N-1 N-1
gs,t ~ E I;
m=-N+l n~ml
Tn,m Pn,lm~cOS0s)eimXt ;
(42)
~in = ~'m + Am (8~m ~)~m + AYm ATm)
(34)
0_<s<K-1,0 <t<_M-1
Figure 2 shows the parts of the spectrum, 8y'm and ATm,
The symbol = is used to emphasize that this is not the that alias the estimated part (K=M/2 is assumed). Note
220

the similarities of the equation pairs (41), (42) and (31), W m = {e "i2rc(sk/K+ tin/M)Ik=0,...,K-1 (46)
Js=0,...,K-1; t=0,...,M-I
(32); however, the solution (41)is aliased more than the
solution (31). Because the Am are (N-Iml)x(N-Iml)
Substituting (45) into (31) and equating with (44) yields
matrices, there are fewer computations associated with
(41) (matrix inverses of diminishing dimension as m --+ A~ (Ym)T g 1 D "1 II/"T
=
~,'~m ,,rag (47)
N-I).

~
which holds for arbitrary g. Therefore, since
m, order *T
Wm (Win) = KM I, where I is the identity matrix,

*T *T

Y h m = ( W m ) (Wm) Rm (48)

Finally, it can easily be shown that

*T *T *
(Ym) (Wm) = KIVI r m , so that (48) b e c o m e s

Am --- KM Rm Rm (49)

which is Parseval's theorem in view of (29) and because

the elements of A m are

I
N= K=M/2 n, degree
Fig,2: Domain of spherical harmonic coefficients estimated by (41),
the conventional least-squares approach. Aliasing is due to
coefficients Ay and ~ in their respective domains, see eq. (42).
A Modification To The Least-Squares Model
By now it is evident that the difference between (41)
and (31) lies in the model. Augmenting (34) to include
degrees up to Iml + K - 1, the model becomes
M/2
g ~ Z Ym )tin
m;-M/2+l
where Ym is similar to Ym but now each has K columns
instead of (N ;..Iml), and Ym is given by (28).
Correspondingly, Am becomes the KxK matrix Am; and
the least-squares solution (41) becomes
Since as many coefficients are computed as data given,
this is no longer a "least-squares" solution. If g is band-
limited, then both (31) and (44) yield the true spectrum,
which means that (31) and (44) are identical for any g.
Hence, in general, the aliasing error in this modified
approach is also given by (32). o
The 2-D Fourier transform Gm (K-vector) is a linear
operator on g:
o --
Gm - ~1 Wm
T g
where Wm is a KMxK matrix:
K-I
aj,k = M ~ Pj~n~(cOS0s) Pk~ml(COS0s) (50)
s=0
One can start with (49), being Parseval's Theorem,
and show, first, that since Am is a full rank matrix, Rm is
non-singular; and, second, that solutions (31) and (44) are
equivalent. Thus, there is a perfect duality between these
solutions: (44) operates directly o n the signal, g; (31)
operates on its Fourier transform G; both yield identical
results.
(43)
Harmonic Analysis of Averages
In order to make use of all gravity data around the
world and to keep the computation of spherical harmonic
models tractable, usual practice is to transform the data
into mean quantities, i.e., averages over certain defined
blocks or areas on the Earth's surface, often delineated by
constant lines of latitude and longitude, e.g., l°xl ° or
30'x30' blocks. This yields a new function, g, in
principle, defined for every point on the sphere, but
sampled, like g, on a regular grid. g has its own unique
Legendre spectrum, {yn,m}, and all the previous results
hold with respect to this case. The question, however, is
how to relate ~[n,mand Yn,rn, the latter being ultimately
sought.
This relationship depends on the spherical area over
which the averaging is performed. It is derived for the
uniformly weighted angular block averages (e.g., l°xl °
mean anomalies) by Gaposchkin (1980). The relationship
for the generally weighted angular block average is deduced
(45)
as follows. Let the average be defined in terms of a
convolution with the general weighting function, b:

~(o,k) = 4-~ b(0,k,0',~.') g(0',k') sin0' d0' Ok'
And, suppose Bk,ra is the frequency response of the
weighting function (the filter) in terms of the Cartesian
variables 0,)~. Then by the convolution theorem for
periodic functions on the plane
Gk.m = Bk, m Gk, m
m
Using (19) for both Gk,m and Glum yields
Z ~'n,mp~,lml= Bk,m--~Im~'n,mp~,lml
n=lml n= I
Now multiply both sides by o3_~, sum over k, and use the
orthogonality (15) to get
~,m = ~ ~ Bk,m p~,lml O!,lkml ~/n,m
28rn I
For purposes of discussion, suppose the filter, b,
removes all power beyond the Nyquist frequencies on the
plane, so that Bk,m = 0 for Ikl > K/2 or Iml > M/2. (The
response for the uniformly weighted block average is
rather inefficient in this respect.) Then, clearly, 7i,m = 0
for Iml > M/2; but ~,m ~ 0 for any degree, j, when Iml <
M/2. In terms of the previous notation (see Figures 1 and
2), AT--m= 0, but 87"rn ~ 0. This means that there remains
considerable aliasing of the spectrum estimated by any
method of harmonic analysis if the averaging is performed
over constant angular blocks. In fact, with a perfect
angular block average (in addition to being zero at
frequencies higher than the Nyquist, Bk, m = 1 for Ikl <
K/2 and Iml < M/2), equation (54) with (15) becomes
7j,m = 7j,m for any j if Iml < M/2. That is, the spherical
spectrum denoted by the region 8"y in Figure 2 is passed
fully by this filter and therefore represents an aliasing error
(see equation 42).
A better filter is the so-called spherical cap average
(Jekeli 1981). In this case (for constant size caps, only!),
the frequency response of the filter is best formulated in
terms of its Legendre spectrum, ~n; and
~"n,m = ~n ~n,m
For the uniformly weighted average over a spherical cap of
radius, ~c'
10- coS~tc
2 ; ~ < ~c
b(~) = (56)
, otherwise
221
where c o s v = cos0 cos0' + sin0 sin0' cos(X - X'). The
filter response, [~n' is the familiar Pellinen smoothing
factor (Sj6berg 1980). Figure 3 shows ~n for ~c = 0"5°"
(5D
1
0.8
0.6
(52)
¥o = 0.5 degrees
(53) o
-0.2 , , , ~ . . . . . . . , , , , ) , , ,
harmonic degree, n
Fig.3: Pellinen smoothing factor (thin line) versus Gaussian smoothing
(54)
factor (thick line).
If the filter were designed such that [3n = 0 for n _>K
then from (55), clearly 7n,m = 0 for n > K (i.e., both/XTm
= 0 and 8ym = 0); and the Legendre spectrum of the
averaged function estimated by (31), (41), or (44) is not
aliased. Such a filter is approximately given by the
Gaussian weighting function (Jekeli 1981):
b ( ~ ) = e "a(l-c°s~); a>0; 0<~<n (57)
The corresponding response is also shown in Figure 3 for
a = 40263. Both responses were designed to attenuate all
harmonics with n > 360 to less than 20%. (The filter
(51) with (57), being a global average, is approximately
a cap average since b(~)~0, ~>~Ja).)
It is seen from (54) and (55) that the frequency
responses of the constant angular block average, Bk,m, and
of the cap average, [3n, are completely different, the first
capable of removing Ay, only, and the second able to
remove both A7 and 87m. Therefore, also the practice of
approximating the response of the constant angular block
average by I~n is quite incorrect.
Data Errors and Least-Squares Collocation
So far, the data measurement noise and an a priori
(55)
spectrum with weights have not been considered. These
are important aspects of modem harmonic analyses.
However, the computational tractability of the analysis
for high-degree models (such as N = 360) restricts the
allowable variances and covariances among the errors of
the given data, as well as the weights of the a priori
harmonic coefficients (Colombo 1981). In essence, the
orthogonality of the matrices Ym (see 41) must be availed
222
to reduce the dimensionality of the normal matrices to be
inverted.
If (38) is written as
N
g = Y y" (58)
*'.-T ~ T
where Y =[Y_N+I ..... ~N.1] and 7 =IT-N+ 1..... ~/N-1], then
the least-squares solution with weights based on the noise
variance/covariance matrix, p-l, is given by
y ~Y P Pg (59)
In order to take advantage of the orthogonality (40), it is
required that
YmP =0; ifmel (60)
This happens only if the errors in the d a t a {gs,t} are
uncorrelated in longitude, and if the correlation in latitude
is the same for all longitudes.
The absence of redundancy of data in the new
approach obviates weighting the data, and no restrictions
need be imposed on the correlations in the data noise for
either formulation, (44) or (31). The variance/covariance
matrix of the estimated coefficients is obtained by simple
error propagation.
Colombo's (1981) least-squares collocation (1.s.c)
method of harmonic analysis, in theory, is supposed to
yield the best estimator of the Legendre spectrum for a
given set of data. In case the data (again, globally and
regularly distributed on a grid) are perfect measurements
(no measurement noise), 1.s.c. minimizes the aliasing
error. The 1.s.c. estimator is also of the quadratures type,
as defined by Colombo (1981, p.43, eq.2.60):
K-1 M-1
y_, y_., TIn,m gs,t e (61)
s=0 t--0
where ~i~,mare "weights" to be determined by inverting a
KxK matrix of discrete spectral components of the
Toeplitz/circulant autocovariance matrix of the data.
Although originally formulated to estimate only
coefficients up to degree and order N, it is easily extended
to estimate any number of coefficients, such as those in
(28). Again, the formulation (61) is preserved in the
presence of data noise only if the noise covariance matrix
has a structure identical to the physical autocovariance
matrix of the data.
Considering (37) and (45), it is readily seen that all
the other methods discussed here, namely (22), (31), (41),
and (44) are of the quadratures type, but with different
weights; and those in (61) are the best in terms of
minimizing the aliasing error for a particular set of
estimated coefficients. But the optimality of (61) is
based on the assumed isotropy and homogeneity of the
(physical) covariance function of the data. These
assumptions are quite strong and do not hold uniformly
for all frequencies. Furthermore, the covariance function
usually is based on a model with questionable fidelity at
the high frequencies. Therefore, in practical applications
1.s.c. (for example, Colombo 1981; Gleason 1987) may
not be significantly superior to the estimation based on
the harmonic analysis (31) or (44) with appropriate
filtering, e.g., with (57), because aliasing thus can be well
controlled and/or eliminated.
Summary
The methods of harmonic analysis on the sphere were
investigated from the point of view of aliasing. The
aliasing error was formulated in terms of spherical
harmonic coefficients in the case when the function is
sampled on a regular grid of latitudes and longitudes. This
yielded several results, some of which are known, but
perhaps more analytically illustrated, here: 1) The simple
quadratures method and related methods are biased even
with band-limited functions. 2) A new method that
eliminates this bias is also superior to Colombo's method
of least squares in terms of reducing aliasing, because it
estimates more coefficients at little added computational
cost. 3) But, a simple modification of the model makes
the least squares method identical to the new method as
one is the dual of the other. The new method solves for
as many coefficients as data and, in the presence of data
noise, it need not use fabricated noise covariances to make
the solution numerically tractable. 4) The essential
elimination of aliasing can only be effected with spherical
cap averages, not with the often used constant angular
block averages.
Appendix - The C o m p u t a t i o n of t3~'m
Analogous to (2b)
K-t
°nm
Pk' = -1- ~ Pn.m(COS0s)e "i2~ksm , k = 0 ..... K-1
K s=0
(A.1)
Also define the Discrete Fourier Transforms of cos0 and
sin0 on the interval (0,re):
o K-I o
~k = I_._~ COS0s e"i2~ks/I¢, ~k+K = ~k (A.2a)
K s=0
K-I o o
~k° = K1s=0
~ sin0s e "i2~ks/K , ~k+K = ~k (A.2b)
The recursion formulas for the normalized associated

Legendre functions are
Substituting (A.3a-d) into (A.1) and using the
convolution theorem for the product cos0s Pn,m(COS0s)
On,m
then yields recursion formulas for the Ok , k --- 0 ..... K-l:
Acknowledgments. This work was supported by contract F19628-93-
K-0033 with the Air Force Phillips Laboratory, Hanscom AFB, MA.
The author is grateful to the reviewers whose valuable comments led to
many improvements.
223
References
Abramowitz M, Stegun IA (eds.) (1972) Handbook of Mathematical
Functions. Dover Publications.
Albertella A, Sacerdote F (1995) Spectral analysis of block averaged
data in geopotential global model determination. To appear in
Manuscripta Geodaetica.
Albertella A, Sacerdote F, Sansb F (1993) Geodetic calculus with
block-averages observations on the sphere. Surveys in
Geophysics 14: 395-402.
Colombo OL (1981) Numerical methods for harmonic analysis on the
sphere. Report no.310, Department of Geodetic Science, The
Ohio State University.
Gaposchkin EM (1980) Averaging on the surface of a sphere. J
Geophys Res 85(B6):3187-3193.
Gleason DM (1987) Developing an optimally estimated earth gravity
model to degree and order 360 from a global set of 30' by 30'
mean surface gravity anomalies. Manuscripta Geodaetica
12:253-267.
Goldstein JD (1978) Application of Fourier techniques to the
computation of spherical harmonic coefficients. The Analytic
Sciences Corp., Tech. Info. Mere. TIM-868-3.
Heiskanen WA, Moritz H (1967) Physical Geodesy. W.H. Freeman,
San Francisco.
Hobson EW (1965) The Theory of Spherical and Ellipsoidal
Harmonics. Chelsea Publishing.
Hwang C 1(1993) Fast algorithm for the formation of normal equations
in a least-squares spherical harmonic analysis by FFT.
Manuscripta Geodaetica 18(1):46-52.
Jekeli C (1981) Alternative methods to smooth the earth's gravity field.
Report No.327, Department of Geodetic Science and Surveying,
Ohio State University.
Kaula WM (1993) Higher Order Statistics of Planetary Gravities and
Topographies. Geophys Res Lett 20.
Pavlis NK (1988) Modeling and estimation of a low degree
geopotential model from terrestrial gravity data. Report No.386,
Department of Geodetic Science and Surveying, The Ohio State
University.
Priesfley MB (1981) Spectral Analysis and Time Series. Academic
Press.
Rapp RH (1977) Determination of Potential Coefficients to Degree 52
from 5° mean gravity anomalies. Bulletin G6od6sique 51(4):301-
323.
Rapp RH, Pavlis NK (1990) The development and analysis of
geopotential coefficient models to spherical harmonic degree
360. J Geophys Res 950313):21885-21911.
Ricardi LI, Burrows ML (1972) A recurrence technique for expanding
a function in spherical harmonics. IEEE Trans Comp June:583-
585.
Schmitz DR, Cain JC (1983) Geomagnetic Spherical Harmonic
Analysis, 1. Techniques. J Geophys Res 88(B2):1222-1228.
Sans?) F (1990) On the aliasing problem in the spherical harmonic
analysis. Bulletin Gtodtsique 64:313-330.
Sjtberg L (1980) A recurrence relation for the I~n-function. Bulletin
Gtodtsique 54(1):69-72.