Potential Predictability of North American Surface Temperature. Part I: Information-Based versus Signal-To-Noise-Based Metrics

Abstract

Abstract In this study, the potential predictability of the North American (NA) surface air temperature was explored using information-based predictability framework and Ensemble-Based Predictions of Climate Changes and their Impacts (ENSEMBLES) multiple model ensembles. Emphasis was put on the comparison of predictability measured by information-based metrics and by the conventional signal-to-noise ratio (SNR)-based metrics. Furthermore, the potential predictability was optimally decomposed into different modes by maximizing the predictable information (equivalent to the maximum of SNR), from which the most predictable structure was extracted and analyzed. It was found that the conventional SNR-based metrics underestimate the potential predictability, in particular in these areas where the predictable signals are relatively weak. The most predictable components of the NA surface air temperature can be characterized by the interannual variability mode and the long-term trend mode. The former is inherent to tropical Pacific sea surface temperature (SST) forcing such as El Niño–Southern Oscillation (ENSO), whereas the latter is closely associated with the global warming. The amplitude of the two modes has geographical variations in different seasons. On this basis, the possible physical mechanisms responsible for the predictable mode of interannual variability and its potential benefits to the improvement of seasonal climate prediction were discussed.