Using ensemble information to improve the representation of covariances in our variational analysis scheme.
In a data assimilation system, forecast errors are specified in terms of covariances, which determine how the information from observations is spread to the model variables. Most current systems used climatologically-averaged covariance statistics, but these give poor results in regions where instabilities have recently occurred. For example, in frontal regions, covariances are seen to "stretch" along the direction of the front, as shown in the illustration. In order to capture these effects, we can use covariances derived from an ensemble data assimilation, which reflect the location of recent instabilities.
Using 'raw' ensemble covariances leads to problems with sampling error and the ability of the analysis to fit the observations, particularly if the ensemble size is small. To get around these problems, one can blend the ensemble covariances with the standard climatological covariances so that the weaknesses in one covariance source can be compensated by the other. This is the idea behind the so-called 'hybrid' data assimilation technique, which leads to a coupling between the data assimilation and ensemble forecasting systems.
Development of a hybrid data assimilation system, coupling our 4D-Var system with the Met Office Global and Regional Ensemble Prediction System (MOGREPS).