Systematic estimates of initial-value decadal predictability for six AOGCMs

Initial-value predictability measures the degree to which the initial state can influence predictions. In this paper, the initial-value predictability of six atmosphere–ocean general circulation models in the North Pacific and North Atlantic is quantified and contrasted by analyzing long control integrations with time invariant external conditions. Through the application of analog and multivariate linear regression methodologies, average predictability properties are estimated for forecasts initiated from every state on the control trajectories. For basinwide measures of predictability, the influence of the initial state tends to last for roughly a decade in both basins, but this limit varies widely among the models, especially in the North Atlantic. Within each basin, predictability varies regionally by as much as a factor of 10 for a given model, and the locations of highest predictability are different for each model. Model-to-model variations in predictability are also seen in the behavior of prominent intrinsic basin modes. Predictability is primarily determined by the mean of forecast distributions rather than the spread about the mean. Horizontal propagation plays a large role in the evolution of these signals and is therefore a key factor in differentiating the predictability of the various models.