H:\CCTWP\sphcov.wpd
Abstract: In this note the covariances between spherical harmonic coefficients and linear
functionals applied on the anomalous gravity potential, T, are derived. The functionals are the
evaluation functionals, and those associated with first and second derivatives of T. These
equations forms the basis for the prediction of spherical harmonic coefficients, a feature now
implemented in the GRAVSOFT program GEOCOL.
1. Introduction.
When using least-squares collocation (LSC) available information about the spherical
harmonic coefficients of the anomalous gravity potential, T, can be used directly as
observations, or used in a remove-restore procedure, see Forsberg & Tscherning (1981).
However, if coefficients are to be predicted, we need the explicit covariances (or values of
functionals applied on a reproducing kernel, see Tscherning (1993)). As shown in section 2
the covariances are simply the observation functional applied on the solid spherical harmonic
function of a specific degree and order, multiplied by a constant depending on the degree.
The equations have been implemented in the form of new versions of the subroutines
COVAX, COVBX and COVCX (Tscherning, 1976) in the program COVFIT (Knudsen,
1987). This program has been used for the calculation of covariances, and used to illustrate
the order of magnitude of typical covariances.
The question of which method is the most efficient to be used for the calculation of the
predicted coefficients and their error-estimates is discussed in section 3.
2. Covariances between spherical harmonic coefficients and point related quantities.
Let P, Q be two points with coordinates (latitude, longitude, r),
,
respectively, and having the spherical distance
. R is the mean radius of the Earth, Yij are
the surface spherical harmonics, Pi the Legendre polynomials and
the degree-variances.
Then the covariance between the values of the anomalous potential T in P, Q is
The covariance between the coefficient GM Cij /R and the anomalous potential is obtained by
applying the functional Lij on the covariance function, where
Then we have
When applying the functional twice we get its norm, squared,
or
The covariance with an arbitrary functional Lk is then
Example:
The correlation can also be calculated
3. Prediction of spherical harmonic coefficients.
An approximation to T determined using LSC will have the form
where bn are the solutions to the normal equations, Ln are the observation functionals and N is
the number of observations. The value of a predicted quantity is obtained by applying the
associated functional on this expression. For the spherical harmonic coefficients we then have
The calculation of the observation functional applied on the solid spherical harmonic function
is generally done by recursion, starting with the (0,0) term. This means that a better
computational strategy would be to calculate all coefficients simultaneously up to and
inclusive degree i. However, we get a problem when we want to calculate the error estimates
because we then must store all the quantities cov(Ln,Lij) in the expression eq. (10)
simultaneously, and subsequently evaluate the expressions for all degrees up to and inclusive
degree i. This will obviously be possible when the degree i and the number of observations are
small.
The strategy adopted is for the moment to predict all quantities up to degree i, but only
compute the error estimate for the ij'th coefficient.
4. Conclusion.
The procedures described above have been implemented in a new version of COVFIT,
denoted covfit12.f and in a new version of GEOCOL, geoco115.f .
Both programs are now in the testing phase. They can be found at ftp.gfy.ku.dk, directory cct.
References.
Forsberg, R. and C.C.Tscherning: The use of Height Data in Gravity Field Approximation by Collocation. J.Geophys.Res., Vol. 86, No. B9, pp. 7843-7854, 1981.
Knudsen, P.: Estimation and Modelling of the Local Empirical covariance Function using gravity and satellite altimeter data. Bulletin Geodesique, Vol. 61, pp. 145-160, 1987a.
Tscherning, C.C.: A FORTRAN IV Program for the Determination of the Anomalous Potential Using Stepwise Least Squares Collocation. Reports of the Department of Geodetic Science No. 212, The Ohio State University, Columbus, Ohio, 1974.
Tscherning, C.C.: Covariance Expressions for Second and Lower Order Derivatives of the Anomalous Potential. Reports of the Department of Geodetic Science No. 225, The Ohio State University, Columbus, Ohio, 1976b.
Tscherning, C.C.: Computation of covariances of derivatives of the anomalous gravity potential in a rotated reference frame. Manuscripta Geodaetica, Vol. 18, no. 3, pp. 115-123, 1993.
This is a contribution to the ESA sponsored project E2M.
Last update 1999.05.20,
comments to: cct@gfy.ku.dk