3B.2 Evaluation of the May 2016 GFS Upgrade

Monday, 23 January 2017: 4:15 PM
Conference Center: Tahoma 4 (Washington State Convention Center )
Glenn H. White, EMC, College Park, MD; and V. Tallapragada, F. Yang, R. Treadon, T. A. Dorian, P. Bhattacharjee, S. Saha, and J. S. Woollen

The National Centers for Environmental Prediction (NCEP) upgraded the GFS Analysis and Forecast System in May, 2016.  The upgrade includes changes to both the Global Data Assimilation System (GDAS) and to the GFS Global Spectral Model and post-processing software.

 The GDAS changes included an upgrade of the 3D Hybrid Ensemble Variational assimilation system to a 4D Hybrid Ensemble Variational assimilation system, where 4D increments are constructed from the best combination of 4D ensemble perturbations. The ensemble provides an updated estimate of the situation dependent background error every hour as it evolves through the ensemble window.  Other changes to the GDAS include new satellite data, better use of satellite data including an extension of the microwave radiance assimilation to all sky conditions, an update of the Community Radiative Transfer Model (CRTM), and improved bias correction for aircraft temperature observations.

 The changes to the GFS Global Spectral Model and post-processing software included corrected land surface characteristics for grassland and cropland categories to reduce summertime warm dry biases over the Great Plains, upgraded convective gravity wave drag, and an upgraded tracer adjustment in the semi-Lagrangian dynamic core.

 There were also changes to model products, including a new Global Forecast Icing Severity (GFIS) icing severity parameter, improved Global Forecast Icing Potential (GFIP) products, the addition of an hourly forecast product out to 120 hours, and the addition of five more vertical levels (7, 5, 3, 2, and 1 hPa).

 This upgraded system was tested using more than three years of forecasts plus the Hurricane Sandy case.  New evaluation methodologies were developed in coordination with other NCEP centers and National Weather Service regional headquarters and forecast offices. Maps of three months of real time operational and upgraded forecasts were available to operational forecasters for evaluation, and selected case studies recommended by forecasters were conducted on more than three years of forecasts.

 Objective verification against observations and the model’s own analyses showed that week 1 forecasts were significantly improved except in the upper stratosphere. The upgraded system produced much smaller analysis increments. Anomaly correlations for 5 day forecasts of 500 hPa height improved significantly by .004 in the Northern hemisphere and .007 in the Southern Hemisphere, with larger improvements for zonal wavenumbers 10-20.  RMS errors of selected atmospheric fields decreased by 10% for day 1 forecasts, 4% for day 3 and 2% for day 5 when verified against the model’s own analyses. Forecasts of 2 m temperatures and dew points and 10 m winds verified against surface observations over the continental United States and Alaska improved significantly. RMSE errors for 96 hour forecasts of 2m temperature and dew point improved by 2% in the western United States and 3% in the east while 96 hour forecasts of 10 m winds improved by 4% in the west and 8% in the east when verified against surface stations. The new system increased a GFS tendency towards too much drizzle, but showed significant improvement in forecasting 2-25 mm/day precipitation thresholds over the continental United States.  Forecasts of CAPE over the United States and jet streams showed significant improvement. Forecasts of tropical storm genesis, track and intensity also improved. Synoptic evaluation of the new GFS yielded no major new concerns; the operational and experimental systems often produced very similar forecasts with the GFSX performing better in some cases. An extensive evaluation site is available at:


This implementation followed a new implementation procedure with a considerably longer official evaluation period and more active engagement with and participation of the other NCEP centers and NWS regional headquarters and forecast offices and with forecast users outside the NWS.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner