(more details claude.fischer++at++meteo.fr )
The second workshop on variational configurations in ALADIN took place in Toulouse along June and July, with the participation of Radmila Bubnova, Claude Fischer, Andras Horanyi, Wafaa Sadiki, Maria Siroka and Cornel Soci.
Wafaa and Cornel successfully implemented the screening, i.e. the new procedure for observation selection, in ALADIN. This step is required for the further use of TOVS observations for example. The screening is ready both for shared and distributed memory applications. Wafaa and Cornel could also focus their efforts on first scientific tests mainly for the choice of a suitable length in the horizontal data thinning process.
Maria and Andras managed to make 3D-Var work, using the old ALADIN-France domain. First encouraging, yet sometimes amazing, results were obtained using the NMC-statistics for the background error covariances. The single-obs experiments carried out by Maria showed that the basic multivariate analysis is running properly, but the analyzed increments remain rather large-scale and "feel" strongly the biperiodization. Andras performed preliminary full-obs analysis and subsequent model forecasts to show that a complete 3D-Var in feasible and that the analyzed state is robust for use in configuration 1. There are only a few technical problems left, related to distributed memory. The main scientific issues are now the treatment of the extension zone and the selection of scales to be analyzed in ALADIN.
Claude focused on the adaptation to non-hydrostatic dynamics and to distributed memory of the linear tangent (TL) and adjoint (AD) codes, while Radmila introduced the dependance on orography in the computation of forecast error statistics.
Single-obs experiment : a +1K (Obs-Guess) increment is analyzed at 500 hPa in the 3D-Var multivariate system. The result is a +0.4267 analysis increment at model level 13 and an anticyclonic wind correction.
 |
Home |