Methods of ensemble prediction for seasonal forecasts with a coupled ocean-atmosphere model

Computational methods for efficient seasonal ensemble prediction with a coupled ocean-atmosphere model, consisting of a global atmosphere and a Pacific basin ocean, are described. Nonlinearly modified Lyapunov vectors, termed bred modes, and finite time normal modes, termed cyclic modes, that grow fastest over a month are found to be suitable ensemble perturbations. The skill of seasonal ensemble prediction is examined in hindcast simulations for the period 1980 to 2000. In general, ensemble mean forecasts are significantly more skilful than the control forecasts. We find that cyclic mode perturbations are generally more effective than bred vectors in improving ensemble forecasts.