Météo-France has implemented a short-range ensemble prediction system known as Prévision d'Ensemble ARPEGE (PEARP). This system is a global ensemble performing forecasts up to 4.5 days. It uses the operational global numerical weather prediction model Action de Recherche Petite Echelle Grande Echelle (ARPEGE) and benefits from variable horizontal resolution, so that it is comparable to some limited-area mesoscale systems over France. Perturbations to the initial conditions are computed by combining an ensemble data assimilation system with singular vectors. Model uncertainties are represented through a 'multiphysics' approach with ten different physical parametrization sets. The article describes the set-up of the system and provides an assessment of the approaches used to represent initial conditions and model uncertainties. The positive impact of the variable horizontal resolution of PEARP is also illustrated. As a global ensemble forecast system (EFS), PEARP is also used to forecast cyclone tracks. It is shown that it has correctly predicted the landfall of hurricane Sandy. The performance of PEARP as run operationally with these features in 2014 is assessed objectively and compared with that of four operational global EFSs using classical probabilistic scores. This comparison is based on The Observing System Research and Predictability Experiment (THORPEX) Interactive Grand Global Ensemble (TIGGE) data. This is one of the first evaluations of EFSs for short-range forecasts. The reliability and global skill of the five EFSs are evaluated over a three-month period with scores computed against observations. PEARP shows skill comparable to or better than the other EFSs.

Abstract.A hybrid scheme obtained by combining 3DVar with the Assimilation in the Unstable Subspace (3DVar-AUS) is tested in a QG model, under perfect model conditions, with a fixed observational network, with and without observational noise. The AUS scheme, originally formulated to assimilate adaptive observations, is used here to assimilate the fixed observations that are found in the region of local maxima of BDAS vectors (Bred vectors subject to assimilation), while the remaining observations are assimilated by 3DVar. The performance of the hybrid scheme is compared with that of 3DVar and of an EnKF. The improvement gained by 3DVar-AUS and the EnKF with respect to 3DVar alone is similar in the present model and observational configuration, while 3DVar-AUS outperforms the EnKF during the forecast stage. The 3DVar-AUS algorithm is easy to implement and the results obtained in the idealized conditions of this study encourage further investigation toward an implementation in more realistic contexts.

Four methods for initialization of ensemble forecasts are systematically compared, namely the methods of singular vectors (SV) and bred modes (BM), as well as the ensemble Kalman filter (EnKF) and the ensemble transform Kalman filter (ETKF). The comparison is done on synthetic data with two models of the flow, namely, a low-order model introduced by Lorenz and a three-level quasigeostrophic atmospheric model. For the latter, both cases of a perfect and an imperfect model are considered. The performance of the various initialization methods is assessed in terms of the statistical reliability and resolution of the ensuing predictions. The relative performance of the four methods, which is statistically significant to a range of about 6 days, is in the order EnKF > ETKF > BM > SV. The difference between the former two methods and the latter two is on the whole more significant than the differences between EnKF and ETKF, or between BM and SV separately. The general conclusion is that, if the quality of ensemble predictions is assessed by the degree to which the predicted ensembles statistically sample the uncertainty on the future state of the flow, the best initial ensembles are those that best statistically sample the uncertainty on the present state of the flow.