Recent Upgrades to the NCEP Short Range Ensemble Forecasting System (SREF) and Future Plans

The SREF system was developed to provide a regional, short-range (0-3 days) ensemble prediction system that produces operationally relevant and useful guidance on the probability distribution of weather elements or events. The probabilistic information provided by SREF will help meet the NWS strategic goal of “providing probabilistic gridded products to the NWS/WFOs, service centers and other users”.

In September 2003, NCEP added 5 Eta-KF (Kain-Fritsch convective parameterization) members to the original ten member system. The system consisted of 5 ETA-KF, 5 Eta-BMJ (Betts-Miller-Janjic convective parameterization) and 5 Regional Spectral Model (RSM) members, for 15 total members. SREF is run twice per day (09z and 21z) out to 63hrs; with output available every 3hrs using regional bred initial state perturbations.

Previous studies have shown the importance of multi-physics ensemble forecasts in improving ensemble diversity. Also, clustering of ensemble members by parent model in the operational SREF is a main concern of NWS field forecasters. Therefore, a new physics based ensemble system at 32 km horizontal resolution (SREF-32) was developed and evaluated. This implementation incorporates additional physics diversity by running various convective and cloud microphysics parameterizations along with single paired breeding (one positive and one negative perturbation to the initial state) to improve system diversity and forecast spread information. This system was implemented into NCEP operations on August 17, 2004.

In addition to improved physics diversity, a change was implemented in the SREF initial condition breeding technique (1) to control unrealistically high initial condition perturbations that sometimes occur in cold season when the atmosphere becomes extremely unstable; and (2) more importantly, to prevent too small initial condition perturbations during the warm season. The initial condition perturbations are now scaled by a factor inversely proportional to the difference between the previous run's 12-hour perturbed 850 mb temperature forecasts. Finally, the Eta and RSM model members were upgraded to the versions used in NCEP operations.

This paper will overview recent upgrades to the SREF system and summarize system performance for several key weather parameters (temperature, moisture, winds and heights) at several levels. The upgraded system will be benchmarked using both deterministic (Root Mean Square Error, Biases, Correlations) and probabilistic (Ranked histograms, statistical consistencies, outlier plots ) verification statistics. Finally planned upgrades in 2005 will be discussed.