NWC REU 2012
May 21 - July 31



Photo of author

Accuracy of GFS Forecasts: An Examination of High-Impact Cold Season Weather Events

Daniel Diaz, Steven Cavallo, and Brian Fielder


What is already known:

  • Wave packets can be problematic for global NWP models due to the fact that model error can propagate in a pattern similar to that of a wave packet when these features are present.
  • Waves on the jet stream impact weather at the surface. Impacts can include the onset of atmospheric blocking, and cyclogenesis.

What this study adds:

  • Forecast model error was diagnosed for the 2011-2012 Northern Hemisphere cold season using the GFS model while identifying particular high-impact weather events.
  • Model error was sometimes associated with wave packets in the upper atmosphere, with corresponding model error reflected at the surface.
  • Our expectation was that model error is generally enhanced with wave packets . Although circumstances exist in which a degradation of skill followed our expectations, there are instances identified where model error did not follow expectations, and hence our expectations were not proven in all cases.


Numerical weather prediction models are thought to handle certain situations with a diminished level of skill, such as with baroclinic Rossby wave packets in the midlatitudes and with large-scale regime changes associated with the onset of atmospheric blocking. Using analysis and forecast data from the NCEP GFS model, this study examines the forecast skill of the 2011–2012 cold season in the Northern Hemisphere with the hypothesis that relatively large model error is primarily associated with baroclinic Rossby wave packets and the onset of atmospheric blocking events. Forecast skill is diagnosed by examining forecast model error through the use of Hovmöller diagrams and the 500 hPa geopotential height anomaly correlation skill score. These diagnostics identify cases which exhibit relatively large error in the forecast model, including tropopause-level wave packets and associated forecast error at the surface. One such event, a 955 hPa surface low, is examined in detail during November 2011, which is found to be associated with relatively large error downstream at later times. Although there are some instances of increased model error in wave packets, coherent patterns of error are not present with every wave packet identified in this study. Additionally, the onset of a long-lived blocking ridge of high pressure around 20° E longitude, which persisted intermittently from the middle of October 2011 to March 2012, is examined. This ridge is found to become stationary after a series of breaking baroclinic Rossby waves impact it, and low skill is seen during its transition from a non-stationary to a stationary ridge. However, no definitive conclusions can be made regarding the source or characteristics of model error for this particular event, and further examination is necessary to gain a better understanding of this event.

Full Paper [PDF]