Ensemble initial conditions through 4D-Var assimilation

The methodology to generate members of an ensemble is an important issue in operational forecasting; currently a number of procedures are followed at different operational centres. In the present work we propose and evaluate a methodology where a dynamically consistent set of initial conditions is generated through 4D-Var assimilation procedure. The perturbations are generated through 4D-Var assimilation with different information (observation) content, either through a variation in the frequency of observations or a variation in the length of assimilation window. We adopt a representative non-linear system, viz. the three-variable Lorenz system and show that ensemble forecasts generated through 4D-Var assimilation show less error than those generated from initial conditions (with similar amplitude of perturbation) adopted arbitrarily. Another advantage of the procedure is that it provides a (variable-specific) estimate of the maximum allowed amplitude and spread of perturbations.