Adaptive ensemble size reduction and preconditioning

B. Uzunoglu, I.M. Navon, M. Zupanski

The ensemble size in sequential atmospheric ensemble based data assimilation using the Heikes and Randall (1995a,b) global shallow-water model is reduced by projecting the ensemble on a limited number of its leading EOFs. The ensemble size is determined by retaining the modes containing the main directions of variability of the system (most energetic modes of the flow). The efficiency of this approach for adaptively updating the ensemble size in the Maximum Likehood Ensemble Filter (MLEF) by Zupanski (2005), Zupanski et al. (2005) used for ensemble data assimilation is assessed for different fractions of variability conserved and compared in terms of RMS and similarity index error indicators with the full ensemble run. An illustration of the feasibility and effectiveness of the method is presented in framework of twin experiments for the above shallow water model. A reduction of up to a factor of four in the number of members of the ensemble was obtained, yielding comparable ensemble data assimilation (ENSDA) results with the full ensemble run. This novel approach results in sizable computational resource economy for general ensemble data assimilation methods.