Skip to Main content Skip to Navigation
Theses

Etude de l'optimisation d'un système d'observation adaptatif pour l'amélioration de la prévision des dépressions météorologiques

Abstract : Forcasting severe cyclogenesis within numerical prediction systems remains a problematic issue even for short range forceasts (1 or 2 days). Observation targeting is designed to improve such cases and to cope with hard to predict situations. The predictability issue was among the scientific objectives of FASTEX which consisted in the first real-time test of such an observing strategy. Some additional observations were deployed over pre-computed sensitive areas (targets) that varied from day to day, according to the meteorological situation.

The sensitive areas of FASTEX were computed using adjoint techniques. So those unstable structures account for the dynamical properties of the atmosphere. It has been shown however, that targeting efficiency strongly depends on the data assimilation system that will process the targeted observations. In order to cope with this efficiency problem, we developped an approach called the {\textquotedblleft}sensitivity to observations{\textquotedblright}. Its computation enables us to draw up sensitivity maps that take into account all of the observations to be used within a given data assimilation system.

The formulation of the sensitivity to the observations is based on the adjoint of the assimilation operator. This linear computation combines forecast sensitivities using the adjoint forecast model and the influence of the adjoint of a variational data assimilation operator: it has been developped within a quasi-operational framework using the 3D-Var of ARPEGE (at Météo France).

In a diagnostic context, the sensitivity to the observations can be used as a critical tool to analyze the deployments of the adaptive observations that were tested during FASTEX. This technique provides us with an insight into the complex interactions of the different types of observations (conventional and targeted, at least) that intervene in the data assimilation process. The sensitivity to the observations appears to be a powerful tool to diagnose how the targeted observations individually impact the subsequent forecast. We show a case study for FASTEX IOP17.

This adjoint technique, when applied to the variational data assimilation process, is helpful to detect some unsuitable usage of the targeted data handled within such a process. We illustrate this kind of implementation on the IOP18 and isolate a few inconsitencies in the assimilation system that trigger spurious effects on the subsequent forecasts.

However, real-time is the natural context of targeting techniques; so we adapted the adjoint approach to prognosticate some suitable adaptive deployments of observations. This prognostic usage addresses the complex issue of the optimization of the targeting techniques. But its complexity makes optimization an untracktable problem for the moment. We consequently focused on some sub-optimal sampling strategies as a first step towards this prime objective and to renew the adjoint targeting techniques of FASTEX. In a simulated prognostic context, we test a real-time feasible startegy by discriminating between proposed realistic deployments. To do so, we use a statistical measure of the quality of the subsequent forecasts.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00266577
Contributor : Alexis Doerenbecher <>
Submitted on : Tuesday, March 25, 2008 - 9:38:09 AM
Last modification on : Friday, April 5, 2019 - 8:14:08 PM
Long-term archiving on: : Friday, May 21, 2010 - 12:51:08 AM

Identifiers

  • HAL Id : tel-00266577, version 1

Collections

Citation

Alexis Doerenbecher. Etude de l'optimisation d'un système d'observation adaptatif pour l'amélioration de la prévision des dépressions météorologiques. Océan, Atmosphère. Université Paul Sabatier - Toulouse III, 2002. Français. ⟨tel-00266577⟩

Share

Metrics

Record views

345

Files downloads

427