Perturbations by the Ensemble Transform

In the ensemble data assimilation, the background error covariance is estimated from perturbations of the ensemble forecast, while characteristics of the ensemble forecast strongly depend on how the initial ensemble is generated. The ensemble transform is a popular perturbation method that widely used as an ensemble perturbation generator, however, linear combinations of different perturbations in the ensemble transform (off-diagonal components of the transform matrix) may harm the global balance of the meteorological field. In this paper, we discuss this issue and show the structure of initial perturbations. Results of forecast experiments using the local ensemble transform Kalman filter (LETKF) for a simplified global model and a regional NWP model are shown. The spin-up issue in a cloud resolving model is shown with the comparison to an alternative method (diagonal LETKF).