Development of the signal‐to‐noise paradox in subseasonal forecasting models: When? Where? Why?

Abstract Subseasonal forecast models are shown to suffer from the same inconsistency in the signal‐to‐noise ratio evident in climate models. Namely, predictable signals in these models are too weak, yet there is a relatively high level of agreement with observed variability of the atmospheric circulation. The net effect is subseasonal forecast models show higher correlation with observed variability than with their own simulations; that is, the signal‐to‐noise paradox. Also, similar to climate models, this paradox is particularly evident in the North Atlantic sector. The paradox is not evident in week 1 or week 2 forecasts, and hence is limited to subseasonal time‐scales. The paradox appears to be related to an overly fast decay of northern annular mode regimes. Three possible causes of this overly fast decay and for the paradox in the Northern Hemisphere are identified: a too‐fast decay of polar stratospheric signals, overly weak downward coupling from the stratosphere to the surface in some models, and overly weak transient synoptic eddy feedbacks. Though the paradox is clearly evident in the North Atlantic, it is relatively muted in the Southern Hemisphere: southern annular mode regimes persist realistically, the stratospheric signal is well maintained, and eddy feedback is, if anything, too strong and zonal.