Assimilating Nonlocal Observations using a Local Ensemble Kalman Filter

Many ensemble Kalman filter data assimilation schemes benefit from spatial localization, often in both the horizontal and vertical coordinates. On the other hand, satellite observations are often sensitive to the dynamics over a broad layer of the atmosphere; that is, the observation operator that maps the model state to the observed satellite radiances is a nonlocal function of the state. Similarly, errors in satellite retrieval observations can be correlated over significant distances. This nonlocality can present problems for assimilating satellite observations with local ensemble Kalman filter schemes. In this paper, we propose a technique in which the observation operator is applied to the global model state and then appropriate observations are selected to estimate the atmospheric state for each model grid point. The issue of how to choose appropriate observations is investigated with numerical experiments on a seven layer primitive equation model, the SPEEDY model. We assimilate both simulated point observations and either nonlocal radiance-like or retrieval-like observations with a particular ensemble Kalman filter, LETKF. The best analysis results are obtained from a scheme that updates the state at a given location by assimilating all those observations that are strongly correlated to the model state near that location. ? Corresponding author. email:[email protected]