A six simulation ensemble AMIP (Atmospheric Model Intercomparison Project) experiment was conducted with the Canadian Climate Centre second generation General Circulation Model (CCC GCM2). Each simulation is forced with observed monthly mean SSTs and sea-ice extent from January 1979 through December 1988. One simulation was initiated from January 1, 1979 FGGE analyses while the remaining 5 simulations were initiated from January 1 states which were randomly selected from a long control simulation. Variability in the seasonal means of 1000 mb height, 500 mb height, 850 mb temperature and other climate variables is analysed using a "2 way mixed effects" model. With this methodology it is possible to diagnose the contributions to simulated inter-annual variability which originate from the external forcing (the prescribed SSTs and sea-ice extent), sources internal to the simulated climate system (such as the atmosphere's internal dynamics and its interaction with the land surface), and high frequency "weather noise".
These variance components are described and compared with similar quantities
computed from NMC analyses and are interpreted as measures of "potential
predictability". It is demonstrated that the potential predictability of
the simulated climate is similar to that of the observed climate, and that
most of the potential predictability in the simulated climate is due to
the external forcing. Verification of the simulations against the analyses
and against themselves suggests that it may be possible to realize most
of the potential predictability with ensemble forecasts of mean seasonal
conditions obtained by forcing climate models.