Exploring the impacts of orthogonal updates in Hybrid Gain Data Assimilation algorithm
Hybrid Data Assimilation has been widely used in numerical weather prediction. Unlike the traditional covariance combination, Penny (2014) demonstrates a different hybrid concept named as the hybrid-gain data assimilation (HGDA). The implementation of HGDA takes two-step update processes, i.e. the EnKF controls the transient fastest-growing dynamic error modes first and the VAR controls the remaining climatological error modes. However, either hybrid concept requires a well-tuned combination weighting to optimal its performance.
Under an assumption that the correlation provided from each system should be independent, this study proposes a parameterless algorithm of HGDA to avoid the use of combination weight by extracting the correction from the variational analysis orthogonal to the ensemble perturbation subspace directly. This algorithm is implemented in a quasi-geostrophic model.
Results indicate that the orthogonal component is able to retain the useful information which is missed in the EnKF analysis but a detrimental effect also raises due to the inaccuracies in the VAR solution. The major finding based on the sensitivity experiments is the necessity of adopting a proper static background error covariance (B) in the two-step updated HGDA. When using the statistic B estimated from EnKF, the parameterless algorithm shows its advantage over the standard optimized HGDA.
Mr. Chih-Chien Chang National Central University, Taiwan