Recent Northern Hemisphere mid-latitude medium-range deterministic forecast skill

• Published in: Tellus A: Dynamic Meteorology and Oceanography
• Main Authors: Rolf H. Langland, Ryan N. Maue
• Format: Article in Journal/Newspaper
• Language: English
• Published: Stockholm University Press 2012
• Subjects: numerical weather prediction; anomaly correlation coefficient; forecast skill; Arctic Oscillation; Oceanography; GC1-1581; Meteorology. Climatology; QC851-999; Arctic
• Online Access: https://doi.org/10.3402/tellusa.v64i0.17531; https://doaj.org/article/d75c4d72971048e5a8ef2911cc7b5b53

A multi-model archive of global deterministic forecasts and analyses from three operational systems is constructed to analyse recent Northern Hemisphere mid-latitude forecast skill from 2007 to 2012 and its relation to large-scale atmospheric flow anomalies defined by the Arctic Oscillation (AO) index. We find that the anomaly correlation coefficient (ACC) in 120-hr forecasts of 500 hPa geopotential height has similar variability on synoptic, monthly, and seasonal time scales in each of the three forecast systems examined here: the European Centre for Medium-Range Weather Forecasts, the National Centers for Environmental Prediction Global Forecast System, and the U.S. Navy Operational Global Atmospheric Prediction System. The results indicate that forecast skill as measured by the ACC is significantly correlated with the AO index and its transitions between negative and positive phase. Intervals of exceptionally high ACC skill during the 2009–2010 and 2010–2011 winters are associated with periods in which the AO remained in a persistent negative phase pattern. Episodes of low ACC, including so-called ‘forecast skill dropouts’ most frequently occur during transitions between negative and positive AO index and with positive AO index. The root mean square error (RMSE) of 120-hr forecast 500 hPa height is also modulated by the AO index, but to a lesser extent than the ACC. In two recent winters, the RMSE indicates lower 120-hr forecast accuracy during periods with negative AO index, which is opposite to ‘skill’ patterns provided by the ACC. These results suggest that the ACC is not in all situations an optimal metric with which to quantify model forecast skill, since the ACC can be higher when the large-scale atmospheric flow contains strong anomalies even if there is no actual improvement in model forecasts of that atmospheric state.