Eugenia Kalnay (PI)
University of Maryland College Park
ekalnay@atmos.umd.edu

Coupled Ocean-Atmosphere Breeding for Ensemble Forecasting and Data Assimilation

Since ENSO is essentially a coupled phenomenon, the skill of seasonal and interannual predictions strongly depends on how well coupled instabilities are captured by the coupled model initial conditions and ensemble perturbations. Therefore ensemble forecasts should include initial perturbations reflecting the uncertainty of the coupled ocean-atmosphere system. No current operational forecasting system attempts to do this. Breeding is probably the only practical way to attempt to find the coupled perturbations. Three papers (Cai et al, 2002, Peña and Kalnay, 2004 and Yang et al, 2004) have shown that coupled ENSO instabilities can indeed be determined with breeding. Beyond the research summarized in these three papers, which were carried out with simpler models or with “perfect model” scenarios, we have performed so far a single one year ensemble forecasting experiment with positive and negative bred vector perturbations upon the operational model control, starting in September 1996. The results are very encouraging, indicating that the initial conditions dynamically perturbed with bred vectors (BVs) lead to solutions that differ from the control forecasts much more than the operational randomly perturbed initial conditions for the same month. Furthermore, the average of the positive and negative bred vector integration is considerably better than the control beyond the first 3 months.   We propose to complete the project leading to the optimal use of bred vectors in ensemble forecasting and data assimilation within the operational NASA seasonal and interannual coupled forecasting system. This will include a) to explore the best way to compute the bred vectors (e.g., from differences between two months minus one month integrations, which will avoid the uncoupled jumps we are imposing in the present configuration); b) to determine the best amplitude for the ensemble bred vector perturbations; c) to develop an approach to use the bred vectors in the Optimal Interpolation by augmenting the background error covariance with the product of the bred vector and its transpose, a method that resulted in substantial improvements in a quasi-geostrophic model at essentially no computational cost (Corazza et al, 2002); and d) to develop a similar augmentation of the background error covariance of the currently uncoupled Ocean Ensemble Kalman Filtering (Keppenne and Rienecker, 2002, 2003, Keppenne et al, 2004), and e) to explore a new dynamical method for isolating ENSO-related coupled bred vectors within the “raw” bred vector fields.

+ Back to Participant Listing