Modelling of error covariances by the ensemble Kalman filter
LORENC, A.C.
The ensemble Kalman (EnKF) filter method is summarised and reviewed for its applicability to data assimilation for numerical weather prediction (NWP). Algorithms with sequential processing of observations, using either perturbed observations, or an ensemble adjustment Kalman filter, should be relatively easy to implement, and affordable in computer time, compared to the currently more popular four-dimensional variational assimilation (4D-Var). The use of an existing forecast model, the analysis algorithm, and parallel computing, are all simpler than 4D-Var, but limited-area versions are more complicated. The EnKF represents the probability distribution of forecast errors using a covariance from a relatively small sample. The noise this introduces is potentially serious, but can be ameliorated by localising the covariances using a Schur product. The ensemble size also limits the number of observed pieces of information which can be fitted in a region. The Schur product removes the ability to impose balance through the covariances, so a separate initialisation may be needed. Good...
Accès à la notice sur le site du portail documentaire de Météo-France