ENSEMBLE PREDICTION USING DYNAMICALLY CONDITIONED PERTURBATIONS

We apply the technique developed in the companion paper by Molteni and Palmer (1993) as a means of providing dynamically conditioned perturbations for ensemble forecasting with a primitive-equation model. Four wintertime initial states are chosen-three at random and one because of substantial development in the large-scale flow within four days, which the control forecast completely missed. A set of singular vectors are created using a quasi-geostrophic model linearized about basic states taken from data close to the chosen initial dates. These are interpolated onto the primitive-equation-model grid, and used as perturbations to the initial state. An ensemble forecast is made from the perturbed initial states. The dispersion of this ensemble is compared, for each date, with that from a second ensemble with initial perturbations constructed from 6-hour-forecast errors. Throughout the forecast period, it is found that the amplitude of the perturbations is noticeably larger using the singular vectors. The dispersion of the ensembles using the forecast-error perturbations did not indicate that the control forecast from the case with substantial development was likely to be poor. By contrast, the (envelope) dispersion of the ensemble using the singular vectors was notably larger for this case than the other three. A number of members of this ensemble were particularly skilful in predicting weather-related elements of the flow, such as low-level temperature change. It is found that the evolution of perturbations which are initially localized over the western Pacific, or western Atlantic, can develop blocking-like structures several days later over the eastern oceans. With the initial amplitudes used in this paper, the development of these structures is in part nonlinear. The growth of the singular-vector perturbations was not as large in the primitive-equation model as in the quasi-geostrophic model, probably due to interpolation problems exacerbated by inconsistent orographic representation. Further work to overcome these, and other, problems is indicated.