NCEP Global Forecast System (GFS) model forecasts sometimes suffer from unforeseen forecast skill busts or “dropouts” supposedly from the imprecise treatment of a wide variety of conventional and non-conventional observations in the assimilation system or possible model errors. Since, the European Centre for Medium-Range Weather Forecasts (ECMWF), often do not experience these forecast skill dropouts, the Committee on Operational Processing Centers (COPC) requested that real-time diagnostics be developed comparing GFS and ECMWF analyses and forecasts. NCEP has recently completed the implementation of the first phase of this project entitled “GFS Forecast Divergence versus ECMWF” for 00Z and 12Z cycles. This web based system alerts the 24x7 operational staff whenever NCEP 5-day forecasts differ excessively from those of the ECMWF, which could potentially be an early precursor of a poor or dropout forecast.

5-day forecast correlations computed between GFS versus ECMWF forecasts (similar to the anomaly correlation (AC) computations with the exception that the ECMWF 5-day forecasts are treated as the truth) predict with remarkable success a majority of the Northern and Southern Hemispheric dropouts, when the GFS 5-day forecasts differ significantly with the ECMWF forecasts. These correlations are produced for 850, 500, 250 and 100 hPa heights at days 1-5 of the forecasts from 00Z and 12Z cycles in both the Northern and Southern Hemispheres. Prediction of forecast dropouts in advance also has significant practical importance to possibly correct problems leading to bad forecasts sooner than in the past.

In addition, this website alerts us to extreme analysis differences in heights, temperatures and winds based on 3 dimensional volumetric integrals of analysis differences between the GFS and ECMWF. Both the forecast-forecast correlations and extreme analysis differences lead to many research opportunities towards improving operations. In the next phase of this project, NCEP has plans to implement similar GFS forecast correlations versus the GFS forecast from 6 hours earlier, but will also include the 06Z and 18Z cycles. These added correlations are likely to be even more useful for predicting poor 5-day forecasts in advance. We have been finding frequent large areas of extreme analysis vector wind differences on the order of 50 knots in the tropics around 300 to 150 hPa. Evidence is shown that these differences appear to be due to differences in the use of satellite wind observation time windows and 4-DVAR versus 3-DVAR analyses. The merits and applications of this integrated system in predicting the GFS dropouts and possibly improving the analysis will be highlighted.