 |
| Titel |
Model error estimation in ensemble data assimilation |
| VerfasserIn |
S. Gillijns, B. Moor |
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| ISSN |
1023-5809
|
| Digitales Dokument |
URL |
| Erschienen |
In: Nonlinear Processes in Geophysics ; 14, no. 1 ; Nr. 14, no. 1 (2007-01-31), S.59-71 |
| Datensatznummer |
250012127
|
| Publikation (Nr.) |
 copernicus.org/npg-14-59-2007.pdf |
|
|
|
|
|
| Zusammenfassung |
A new methodology is proposed to estimate and account for systematic model
error in linear filtering as well as in nonlinear ensemble based filtering.
Our results extend the work of Dee and Todling (2000) on constant bias errors to
time-varying model errors. In contrast to existing methodologies, the new
filter can also deal with the case where no dynamical model for the
systematic error is available. In the latter case, the applicability is
limited by a matrix rank condition which has to be satisfied in order for the
filter to exist.
The performance of the filter developed in this paper is limited by the
availability and the accuracy of observations and by the variance of the
stochastic model error component. The effect of these aspects on the
estimation accuracy is investigated in several numerical experiments using
the Lorenz (1996) model. Experimental results indicate that the
availability of a dynamical model for the systematic error significantly
reduces the variance of the model error estimates, but has only minor effect
on the estimates of the system state. The filter is able to estimate additive
model error of any type, provided that the rank condition is satisfied and
that the stochastic errors and measurement errors are significantly smaller
than the systematic errors. The results of this study are encouraging.
However, it remains to be seen how the filter performs in more realistic
applications. |
| |
|
| Teil von |
|
|
|
|
|
|
|
|