Geophysical Research Letters | 2019
A Priori Identification of Skillful Extratropical Subseasonal Forecasts
Abstract
The current generation of subseasonal operational model forecasts has, on average, low skill for leads beyond 3 weeks. This is likely a fundamental property of the climate system, due to the relative high amplitude of unpredictable weather variability compared to potentially predictable, but generally weaker, climate signals. Thus, for subseasonal forecasts to be useful, their high versus low skill events should be identified at time of forecast. We show that a linear inverse model (LIM), an empirical‐dynamical model constructed from covariability statistics of wintertime (December–March) weekly averaged observational analyses, can be used to identify, a priori, the expected extratropical subseasonal surface and midtropospheric forecast skill. The LIM s predicted signal‐to‐noise ratio identifies the subset (10%–30%) of Weeks 3–6 forecasts—of the LIM and two operational models from the National Centers for Environmental Prediction and the European Centre for Medium‐Range Weather Forecasts—with relatively higher skill versus the much larger remainder of forecasts whose skill cannot be distinguished from random chance. Plain Language Summary Our current understanding of weather prediction is that usable daily forecasts cannot be made more than 15 days in advance. This is a consequence of chaos: Any small initial uncertainty in our picture of the atmosphere (e.g., wind, temperature, and pressure) when the forecast is made will lead to errors growing to become as large as the weather we are trying to predict. Recently, however, focus has turned to “subseasonal” forecasts, predictions of weekly averaged weather made 3 to 6 weeks ahead, because climate phenomena (e.g., El Niño) sometimes produce persistent weather patterns that might be predicted even though individual storms within them cannot be. To identify when these “forecasts of opportunity” will occur, we developed a statistical subseasonal forecast model capable of predicting when its own forecasts—and those of sophisticated physical models from U.S. and European operational centers—will be usable. Our model successfully identifies the 20%–30% of forecasts at Weeks 3 and 4 and 10% of forecasts at Weeks 5 and 6 that are usable. Our results show a path forward to develop techniques for identifying usable subseasonal forecasts beforehand, so that practical forecast guidance may be given to end users in a variety of societal contexts.