CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Methods for improving background error covariance matrix rebuild in data assimilation Authors:  Sibo Cheng - Electricite de France (France) [presenting]
Jean-Philippe Argaud - Electricite de France (France)
Didier Lucor - Limsi Universite Paris-Sud (France)
Angelique Poncot - Electricite de France (France)
Bertrand Iooss - Electricite de France (France)
Abstract: A recurrent obstacle in data assimilation is the lack of information for background error covariance modeling. In the case of meteorology, the background covariance is often estimated from an observations ensemble or forecast differences. However, for many industrial fields, the modeling remains highly empirical relying on some form of expertise and physical constraints enforcement in the absence of historical observations/predictions. The Desroziers \textit{a posteriori} tuning algorithm of 2001 is well known in variational methods for adjusting the ratio between background and observation error covariance matrices. We have developed two novel sequential adaptive methods: \textbf{CUTE}(Covariance Updating iTerative mEthod) and \textbf{PUB}(Partially Updating BLUE method) for building background error covariance matrices in order to improve the assimilation result under the assumption of a good knowledge of the observation error covariances. We have compared these two methods with the Desroziers approach in a twin fluid mechanics experiment framework together with a linear observation operator. The optimality in terms of assimilation errors is similar for all three methods. However, experiments show that the two new methods own a non-negligible advantage concerning correlation rebuild under the hypothesis that the background error is dominant over the one of observations.