| Ensemble forecasts are used in numerical weather prediction to quantify forecast uncertainty in an objective and flow-dependent manner. An overview will be given of the Ensemble Prediction System (EPS) which is operational at the European Centre for Medium-Range Weather Forecasts (ECMWF). Firstly, the design of the EPS will be described. Initial uncertainties in the EPS are represented by perturbations based on singular vectors of the propagator of the prediction model. The propagator in the singular vector computation is defined over a period of 48 hours. This is sufficiently short so that a linear approach is deemed to be still useful and it is sufficiently long to be comparable with typical error doubling time-scales of the synoptic-scale flow in the extra-tropics. Model uncertainties of parameterized processes such as clouds, convection, radiation, gravity wave drag are represented by perturbing the total tendencies due to these processes with multiplicative noise. Secondly, diagnostics of the EPS forecasts will be presented including a comparison with other global ensembles. Issues that have been identified in this diagnostic work will be discussed. Thirdly, recent developments will be described. This includes the construction of initial perturbations from an ensemble of variational assimilations in which the observations, model tendencies and sea surface temperatures have been perturbed. |
|