Assimilation rétrospective de données par lissage de rang réduit : application et évaluation dans l'Atlantique Tropical

Abstract : The Kalman filter is widely used in data assimilation for operational oceanography, in particular for forecasting problems. Yet, now that data assimilation applications tend to diversify, with reanalysis problems for instance, the three-dimensional (3D) formulation of the filter doesn't allow an optimal use of the observations. The four-dimensional extention of the 3D methods, called smoothers, allows a better use of the observations, assimilating them on a retrospective way. We study in this work the implementation and the effects of a reduced-rank smoother on reanalysis, with a realistic tropical Atlantic ocean circulation model. First we expose some sensitive steps required for the smoother implementation, most notably the covariances evolution parametrisation of the filter. The smoother's benefits for reanalysis are then exposed, compare to a 3D reanalysis. It shows that the global error can be reduced by 15% on assimilated variables (like temperature). The smoother also leads to an analyzed solution dynamically closer to the reference (compare to the filter), as we can observe with phasing of Brazil rings for instance. Finally, we studied a case of smoothing based on optimal interpolation (instead of the filter). This case is inconsistent with the theory but often used in operational centers. Results shows that the smoother can improve the reanalysis solution in an OI case (reducing the global error from 10 to 15%), but still the dynamical evolution of error covariances (filter) are needed to get a correction according with the real error structures.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00683971
Contributor : Abes Star <>
Submitted on : Friday, March 30, 2012 - 12:07:43 PM
Last modification on : Thursday, January 3, 2019 - 4:30:09 PM
Long-term archiving on: Wednesday, December 14, 2016 - 7:18:27 PM

File

24363_FREYCHET_2012_archivage....
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-00683971, version 1

Collections

STAR | CNRS | LEGI | UGA

Citation

Nicolas Freychet. Assimilation rétrospective de données par lissage de rang réduit : application et évaluation dans l'Atlantique Tropical. Sciences de la Terre. Université de Grenoble, 2012. Français. ⟨NNT : 2012GRENU003⟩. ⟨tel-00683971⟩

Share

Metrics

Record views

776

Files downloads

358