Extended Data Fig. 5: NAO not solely driven by AMV.
From: North Atlantic climate far more predictable than models imply
a, Time series of observed (black curve) and variance-adjusted lagged ensemble forecasts (years 2–9; red curve, ensemble-mean; red shading, 5%–95% confidence interval diagnosed from the error variance) 8-yr running-mean December-to-March NAO. b, As in a, but for the AMV-matched forecasts. AMV matching uses the same procedure as NAO matching (see Methods), except that the 20 ensemble members are selected on the basis of AMV instead of NAO. If the NAO signal were solely driven by AMV, then selecting the most skilful AMV ensemble members via AMV matching should increase the NAO skill. However, AMV matching clearly reduces the NAO skill (ACC reduces from 0.79, P < 0.01, to 0.37, P = 0.1). By contrast, NAO matching clearly improves the forecasts of AMV (Fig. 2c, d). We therefore conclude that the NAO signal is not solely driven by AMV.
