The signal-to-noise error in decadal climate modes

Large ensembles of climate models were designed with the explicit goal of disentangling externally forced signals from internally generated noise in regional climate. The first, pioneering large ensembles depicted a climate system with rich internal variability leading to a diverse array of regional climates across simulations, even on decadal timescales. Models simulated the familiar spatial patterns of the major modes of climate variability (e.g. NAO, AMV, PDO) and their impacts with the random timeseries expected of internal variability. However, with the advent of even larger ensembles from multiple modeling groups, a growing number of studies noticed that when you average over the random timeseries of these modes, isolating the externally forced signal common to all simulations, the resultant ensemble mean timeseries has an impressive correspondence with observations. That is, the externally forced signal has a strong influence on these decadal modes of climate variability and their impacts. A large forced signal with the correct amplitude should have been identifiable with fewer ensemble members; however, the amplitude of the forced signal in models is far too weak relative to that in observations. Succinctly, the newfound, large role for external forcing was previously obscured by an erroneously low signal-to-noise ratio in climate models. This new interpretation of large ensemble output implies that regional climate variability may be both more accurately and precisely predictable than previously assumed.