Over the past decade, a diverse set of variations on the Ensemble Kalman Filter (EnKF) has been proposed. Differences include whether observations should be perturbed or not, whether and how the ensemble covariance should be regularised to inhibit undesirable behaviour, whether a single, double or multiple filter should be employed, and whether the observation-error covariance should be used directly, or via an ensemble representation. The Australian Bureau of Meteorology is exploring the use of the EnKF as a next-generation operational assimilation system. Understanding the properties of, and reasons for the differences between, the different flavours of EnKF is a necessary precursor to developing such a system. Here, results from some initial studies, pertaining to a subset of the above list, will be presented.

The first aspect of the EnKF that has been considered is whether the observation-error covariance should be used directly, or via an ensemble representation, as recently promoted by Evensen (2003, Ocean Dynamics). It is shown that this will lead to the loss of rank in the ensemble whenever the ensemble size is less than the number of observations, and collapse of the ensemble to a single member when the ensemble size does not exceed half the number of observations. This collapse can be prevented by some numerical approximations in the analysis equations, which however lead to a mis-estimation of the analysis-error covariance.

The second aspect that is being explored is background-error covariance regularisation.
 

Convenor: Les Muir.