Thursday 25 August 2005

G2

PTH0021
Estimation of data noise in global gravity field modeling
Liu, Xianglin¹, Ditmar, Pavel¹, Klees, Roland¹
1 Delft Institute Of Earth Observations And Space System, Delft University Of Technology, The Netherlands
Author email: p.ditmar@lr.tudelft.nl
New satellite missions that have been launched or will be launched soon - CHAMP (2000), GRACE (2002), and GOCE (2006) - make it possible to study the Earth's gravity field model with an unprecedented accuracy and resolution. In order to fully benefit from the acquired information, the data processing strategy should result in the statistically optimal model. This means that the covariance matrix of observational noise has to be estimated. We propose the following procedure for this purpose. It is assumed that the noise covariance matrix C can be represented as C = PDP, where D is a diagonal positive-definite matrix (with a mean over diagonal elements being equal to one), and P is a Toeplitz matrix. The matrix D reflects a dependence of noise on time and is assumed to be given a priori (up to a scaling factor). The matrix P reflects a dependence on frequency. It is estimated from the data themselves. First, a 'trial' gravity field model is obtained. Then, data noise is estimated as the difference between the observed and adjusted quantities. The covariance matrix of noise is obtained and used to compute the noise PSD (Power Spectral Density). Finally, the square root of the PSD is transformed back into the time domain, which results in an estimation of the matrix P. Importantly, the proposed procedure can easily be generalized to data with gaps. Numerical simulations with noise models introduced above show that the proposed procedure results in a reasonable estimation of noise parameters. Synthetic gravity field models produced with the true and with the estimated noise model show a negligible difference. The proposed procedure has been used also in processing real accelerations of the CHAMP satellite.

Return to Poster Presentations