Post-doc in model error covariance

Look here for job postings within the Modeling community.

Moderators: arango, robertson

Post Reply
Message
Author
User avatar
arango
Site Admin
Posts: 1347
Joined: Wed Feb 26, 2003 4:41 pm
Location: DMCS, Rutgers University
Contact:

Post-doc in model error covariance

#1 Unread post by arango »

Estimation and representation of model error covariances in data assimilation

Research Institute : CNRM/GAME (Météo-France/CNRS) URA 1357, 42 av. Coriolis, 31057 Toulouse Cedex 1, France
Supervisors : Gérald Desroziers and Loïk Berre
Email : Gerald.Desroziers@meteo.fr, Loik.Berre@meteo.fr


We are opening an 18 month post-doctoral position for a scientist or researcher holding a PhD degree on either of the following subjects: data assimilation, atmospheric modelling, meteorology, statistics, and applied mathematics. This position is opened in the context of the ADTAO project. This project, lead by Serge Gratton (Université Paul Sabatier) and Michel Dayde (IRIT) in Toulouse, manages research actions conducted by OMP, CNES, Météo-France, CERFACS, and IRIT. The project is funded by RTRA, and its main objective is to design the next generation of operational data assimilation systems by improving the representation of model errors in large, multi-scale and highly nonlinear dynamical systems.

The specification of realistic background and observation error covariances plays a central role in data assimilation, because it determines spatial filtering and propagation of observations to a large extent. Moreover, background errors can be considered as the sum of predictability errors (corresponding to the forecast evolution of analysis errors) and of model errors (corresponding to model imperfections). While using ensemble assimilation techniques allows the predictability error to be simulated, estimating and representing model error covariances is in fact one of the main scientific and technical challenges in data assimilation. In addition, knowledge about model error is crucial not only for calculating realistic background error covariances, but also for specifying appropriate model error covariances in the weak-constraint formulation of 4D-Var. Finally, the representation of model errors is also a major issue in ensemble prediction systems.

The objective of the proposed position is to get more knowledge about model error covariances, by combining ensemble assimilation techniques and innovation-based covariance estimates. In particular, the Météo-France operational ensemble 4D-Var system will be used to estimate predictability error covariances. In order to estimate model error covariances, these predictability error covariances will be compared to innovation-based covariance estimates, which give information about “total” background error covariances. Different innovation-based techniques will be considered, such as covariances of analysis residuals, diagnostics based on the minimum of the cost function, and differences between forecasts and a reference analysis. Global estimates will be calculated first, and regional estimates will be considered in a second step. The focus will be on model error variances at the beginning, and then spatial correlation estimates will be examined.

Once model error covariances have been estimated, their impact on data assimilation will be investigated. This corresponds to the representation of model error contributions in the ensemble 4D-Var system, with direct effects on associated background error covariance estimates. Representing these model error covariances can be done by using inflation techniques, and it is planned to compare an additive inflation technique with the current near-operational multiplicative inflation technique which is available in the ensemble 4D-Var system. These model error covariance estimates will also be tested in a weak-constraint formulation of 4D-Var. Contacts will be established with the ensemble prediction community as well, to consider experimentation of such model error covariances in ensemble prediction systems (such as the Météo-France operational PEARP system).

The position will be held in Toulouse. We invite candidates to send their application to Gérald Desroziers and Loïk Berre as soon as possible, and latest by 31 December 2010. It should include a resume, motivation letter, the names of 2 or 3 persons who can be contacted as references, and a copy of the most recent international publications.


4D-Var : Four-Dimensional Variational data assimilation
ADTAO : Algorithmes de nouvelle génération pour l’Assimilation de Données dans le système Terre Atmosphère Océan
ARPEGE : Action de Recherche à Petite Echelle et Grande Echelle
CERFACS : Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique
CNES : Centre national d’études spatiales
ECMWF : European Centre for Medium-range Weather Forecasts
IRIT : Institut de Recherche en Informatique de Toulouse
OMP : Observatoire Midi-Pyrénées
RTRA : Réseau Thématique de Recherches Avancées

Post Reply