Atmospheric multiple equilibria and non‐Gaussian behaviour in model simulations

Multiple equilibria from general circulation models (GCMs) are investigated via Gaussian mixture models. The cross‐validation technique based on bootstrap, used to find the possible number of clusters, is shown to have a tendency to display a bias in the number of clusters when the sample is not statistically independent. The method is shown, however, to give satisfactory results when applied with a strategy based on statistical independence. The mixture model is successfully applied to a few dynamical systems and data from the UK Universities Global Atmospheric Modelling Programme (UGAMP) GCM. The same analysis is then carried out and applied to the wintertime 500 mb geopotential height field taken from a 50‐year run of the Hadley Centre HadAM3 GCM. Three regions, the North Pacific‐American, the North Atlantic, and the northern hemisphere are investigated for nonlinear circulation regimes. It is shown that the model displays robust regime behaviour over the North Pacific‐American sector, while only marginal regime behaviour is observed over the North Atlantic, and no trace of hemispheric regimes.