We’ve Hit a Roadblock: The role of atmospheric blocks in degrading UFS subseasonal forecast prediction skill

Accurate weather predictions on subseasonal to seasonal (S2S) timescales (weeks to months in advance) have long been a challenge. While S2S forecast skill has improved in the last decade, many unknowns still exist regarding the reasons for larger forecast errors on these extended timescales. This project focuses on post-processed data from the National Oceanic and Atmospheric Administration’s (NOAA) Unified Forecast System (UFS) to identify periods of large forecast temperature and precipitation error. We identify time periods of large error that are categorized as “more predictable” by the Weighted Permutation Entropy (WPE) metric. WPE measures the complexity of a time series, like its counterpart permutation entropy, but also incorporates amplitude information. WPE was applied to 20 years of 2-meter surface temperature forecasts that isolated specific periods when the UFS had large errors. A common theme for cases of large UFS forecast errors on S2S timescales was the presence of atmospheric blocking and related large-scale regimes. Next, UFS errors were compared to several deep learning weather model errors during time periods when atmospheric blocks had significant impacts on forecast temperature error. This comparison was performed to investigate whether atmospheric blocking regimes create large errors in both UFS and deep learning models, or whether they are specific to the UFS alone. Results demonstrate the utility of WPE for large forecast skill and error exploration, as well as the role of atmospheric blocking in degrading S2S forecast prediction skill across different types of weather prediction systems.