Statistical Properties of Predictability from Atmospheric Analogs and the Existence of Multiple Flow Regimes

Abstract

The problem of predictability in relation to the existence of multiple flow regimes is examined for the atmosphere on the basis of observations. The existence of multiple-flow regimes and the dynamics of transitions between them has important implications for the predictability of a system. Initially close states that do not undergo the same transitions may diverge much more rapidly than those states that are not subject to any, or are subject to the same, transitions. The Lorenz system is given as an example of how this behavior can lead, at intermediate times in the error evolution, to a bimodal distribution of the error that is, however, not a property of the equilibrium distribution. The probability distribution of errors, defined as the distance between the best analogs found in a 35-yr record of 500-mb heights is followed during its evolution toward statistical equilibrium. Hierarchical clusters in the subspace of the leading empirical orthogonal functions (EOFs), identified in earlier studies with well-known anomalous circulation patterns, are found to have statistically significant differences in predictability and persistence. Following the trajectories that originate from a single cluster, the distribution of predictability and persistence is found to be bimodal for two of the clusters. This provides a dynamical criterion to classify the states, and evidence is found that bimodality is related to transitions from one regime to the other. These findings provide a strong dynamical evidence of the regime organization of planetary-scale flow and of preferred transition routes.