Challenges and algorithmic solutions for data assimilation in coupled systems

Coupled ocean-atmosphere-ice models are central to extended-range prediction of the earth system models. Data assimilation in these coupled models presents a unique set of challenges. The spatial and temporal scales vary by an order of magnitude between fluids, the covariances between components of the coupled system are poorly known, the assimilation methods tend to differ, and coupled model often do not have tangent linear and adjoint models. In this presentation, we show several algorithmic solutions that allow us to resolve these challenges. Specifically, we introduce the interface solver method that augments existing stand-alone systems for ocean and atmosphere by allowing them to be influenced by relevant measurements from the coupled fluid. To propagate information between fluids, we use coupled ensembles, for which we develop coupled localization and verification methods. Finally, to enable use of variational 4DVAR methods for coupled systems, we develop localized ensemble-based tangent linear models. The developed methods are tested using a combination of realistic simulations for Mediterranean as well as a set of idealized models. The talk concludes with overview of the U.S. Navy’s plans for global coupled Data Assimilation.