Session 11A.4 Evaluation of a mesoscale short-range ensemble forecast system over the Northeast United States

Thursday, 4 August 2005: 8:45 AM
Ambassador Ballroom (Omni Shoreham Hotel Washington D.C.)
Brian A. Colle, Stony Brook University/SUNY, Stony Brook, NY; and M. Jones and J. S. Tongue

Presentation PDF (284.5 kB)

This talk will highlight the long-term verification results from the Stony Brook University Short-Range Ensemble Forecast (SBU SREF) system using the MM5, which has been running operationally since May 2003 down to12-km grid spacing over the Northeast United States for the 0000 UTC cycle using 7 initial conditions (ICs) and 12 physics (PHYS) members (http://fractus.msrc.sunysb.edu/mm5rte). Surface verification for the 48-h forecasts has been completed for both the warm (April-September, 2003) and cool (October 2003-March, 2004) seasons. All ensemble members have an appreciable diurnal temperature and wind speed bias that tends to vary by PBL type. The largest temperature variations during the warm season are for the Eta PBL during the day, since cloud cover and precipitation are strongly dependent on what convective parameterization is used. The Eta-PBL cool biases during the day partially cancel the warm biases from other PBL members, resulting in the PHYS ensemble mean having the most skill on average. In contrast, because of greater clustering of PBL and convective parameterizations solutions during the cool season, the IC members are more useful than the physics members on average, but none of the members outperform the GFS-MM5 for sea-level pressure. The MM5 ensemble outperforms the NCEP Eta model and it is nearly equal in skill on average with deterministic MM5 initialized 12-hours later. Using either a 14-day or Model Output Statistics (MOS) bias calibration improves the ensemble under-dispersion and bias of temperature and winds.

It will be shown that the SBU SREF system has some ability to predict skill of the ensemble-mean and represent forecast uncertainty by a correlation between ensemble spread and errors of the ensemble-mean. The usability of ensemble spread varies from little representation of forecast uncertainty (e.g. 2-meter temperature) to good representation of forecast uncertainty (e.g. 10-meter wind direction). Although the reliability of ensemble probability of precipitation is only moderate for most accumulation thresholds, probabilistic precipitation is more skillful than sample climatology. For 24-hour precipitation forecasts over the Northeast, the physics ensemble has the greatest skill and reliability during the warm season, while the initial condition ensemble provides the largest benefit during the cool season.

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