ESRL GFS Reforecast Project

ESRL/PSD has undertaken a "reforecasting" effort, a companion effort to the NCEP/NCAR reanalysis. 15-member ensemble  forecasts have been run over  the last 25+ years using initial conditions from the reanalysis and a recent, unchanging version of NCEP's global forecasting system (GFS, formerly known as MRF). Details of the model  configuration can be found here.  Go  here for a web form to download the data. You can also access the forecast files directly via ftp. Some sample NCL sripts for reading and plotting the refcst netCDF files are available. Real-time experimental probabilistic forecasts, calibrated using the forecast archive are being developed.

Why undertake a "reforecasting" effort if NCEP runs a continually improved version of the model? Reasons include:

  • Improved medium-range weather forecast products.   In order to diagnose and make effective corrections to forecast errors, a long time series of forecasts from an unchanged model is necessary.  It is likely that effective statistical corrections to forecasts from an older version of the MRF will allow it to outperform newer versions of the MRF without statistical post-processing at long leads.
  • Improved probabilistic forecasts of extreme events.   These events occur rarely by definition. Accurate probabilistic forecasts require a database of many extreme events.  The events that are predictable must be distinguished from events that are not predictable, and calibrated probabilities must be developed for the predictable events using the long time series of ensembles.
  • Diagnosing model errors.  Especially for phenomena with low-frequency variability, a long time series of forecasts are necessary in order to diagnose model errors.   This information about model error can be used by NCEP to develop future versions of the medium-range forecast model.


For more information, see the article published in the Bulletin of the American Meteorological Society.


[an error occurred while processing this directive]