10B.2
Update on the NCEP GFS Forecast Skill Score Dropouts

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 20 January 2010: 4:15 PM
B306 (GWCC)
DaNa Carlis, NOAA/NWS/NCEP, Camp Springs, MD MD; and J. C. Alpert, B. Ballish, and K. V. Kumar

Intermittent low skill events, or “Dropouts”, in the NCEP Global Forecast System (GFS) have lead to investigations ranging from data quality control to using pseudo radiosonde profiles derived from ECMWF analysis as observations into GFS analysis. Using the ECMWF analysis as pseudo-RAOBS in the Gridpoint Statistical Interpolation (GSI)/GFS has been shown to eliminate more than 90% of NCEP GFS dropouts. With this perspective, our focus has been on investigating which observation data types ingested into the GFS has negative impact on forecast skill, but the large volume of observations for a model run limits the ability to draw definitive conclusions for a particular set of observation data. In some cases, we found that a particular data type was directly correlated with a poor 5-day 500 hPa anomaly correlation (AC) skill score for the GFS, but estimating observation impact by turning off certain types of data does not significantly enhance the skill of the GFS. Thus, we have turned our attention towards data quality control (QC) where, in this paper, we will discuss the impact of satellite cloud track winds on the GFS. Also, this study will show results from the high-resolution ECMWF pseudo-observation runs (or ECM runs) for dropout and nondropout cases and compare the results with the original low-resolution ECM runs.