 |
| Titel |
On the computation of a reliable signal covariance for the stochastic filtering of time-variable gravity field from GRACE |
| VerfasserIn |
C. Lorenz, B. Devaraju, N. Sneeuw |
| Konferenz |
EGU General Assembly 2009
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 11 (2009) |
| Datensatznummer |
250023133
|
|
|
|
|
|
| Zusammenfassung |
| Improving the mass estimates from GRACE time-variable monthly gravity field solutions by
stochastic filtering requires the knowledge of signal and error covariances of the monthly
solutions. Signal covariance is only poorly known for the time-variable field, and
it is mainly computed from geophysical models. In this contribution, the signal
covariance of time-variable field is computed from the GRACE monthly solutions,
which is not straightforward due to the presence of noise in the higher harmonic
degrees (l > 21–28). Hence, an isotropic signal covariance model, which can be
considered as a first approximation of the actual signal covariance, is computed by
fitting a Kaula-type power-law to the signal degree variances of the GRACE monthly
solutions. In order to ascertain a reliable signal covariance model power-law fits
to different subsets of the GRACE dataset were analyzed. The analysis indicates
that all the subsets chosen indicate at only one signal covariance model within
reasonable limits. This signal covariance model points to the less-noisy part of the
monthly degree variances, which in turn indicate the resolution of the monthly
time-variable gravity field solutions from GRACE. This analysis exemplifies the fact that a
reliable (isotropic) signal covariance model can be constructed from the GRACE
data itself. In the end, the computed signal covariance model and simulated error
covariances were used for constructing stochastic filters for three different months.
The results from the application of the stochastic filters indicate that the filtered
monthly solutions are of different resolutions. Therefore, care must be taken in
constructing and analyzing time-series from stochastically filtered GRACE datasets. |
|
|
|
|
|
|