Seasonal Predictions of the Winter North Atlantic Oscillation: Variability, Forecast Skill and the Signal‐to‐Noise Paradox in a Large Ensemble

ABSTRACT We investigate the variability of the North Atlantic Oscillation (NAO) in the Met Office Global Seasonal Forecasting System (GloSea) using a 132‐member ensemble of coupled model forecasts, which is larger than has previously been available. Consistent with previous studies, we find that the signal‐to‐noise ratio is too small to match the correlation skill, and we additionally find that this result is statistically significant for years when the El Niño Southern Oscillation is active, and therefore skill is higher. We also show that correcting the signal‐to‐noise ratio by only increasing the signal would produce total variability that is still within real world estimates, removing the necessity for nonlinear mechanisms to increase the signal at the expense of the noise. Finally, we find an inverse relationship between yearly ensemble spread and ensemble mean, suggesting that the negative phase of the NAO may be less predictable than the positive, although the relationship is partly due to a longitudinal shift in the NAO pattern, which in positive NAO years moves the centre of variability away from the traditional Azores location.