9.3
Evaluation of a Mesoscale Short-Range Ensemble Forecasting System over the Pacific Northwest
Eric P. Grimit, University of Washington, Seattle, WA; and C. F. Mass
In January 2000, the University of Washington Department of Atmospheric Sciences began running a short-range (0-48 hours), five-member, mesoscale ensemble prediction system (EPS) using the Penn State/NCAR MM5 model. Each member forecast started with a different 0000 UTC initialization and continued out 48 hours with updated boundary conditions drawn from the corresponding, dynamically evolving, synoptic-scale forecasts. Sources for the five initializations/forecasts included: NCEP's AVN, NGM, and Eta, Canadian Meteorological Centre's GEM, and Fleet Numerical Meteorology and Oceanography Center's NOGAPS. Resolution of the numerical forecasts was 36-km in a coarse domain encompassing the West Coast of the U.S. and Canada as well as the eastern Pacific Ocean and 12-km in the nested domain over Washington and Oregon. In August 2000, the NGM-forced member was replaced by a forecast using NCEP's MRF, due to changes in the NGM data assimilation scheme earlier in the year that made it too similar to the Eta initialization.
Mesoscale, short-range ensemble forecasting (SREF) has been undertaken by a number of groups (Mullen and Baumhefner 1989,1994; Stensrud and Fritsch 1994a,b; Du et al. 1997; Stensrud et al. 1998; Hou et al. 2001) focusing mainly on the midwestern and eastern U.S. regions. In those regions, widespread, deep convection typically plays an important role in the mesoscale atmospheric behavior and it has been suggested that varying model physics may be just as beneficial as varying the atmospheric initial conditions to create useful ensemble forecasts. The value and performance of regional ensembles created by physics perturbation strategies may be very different in the Pacific Northwest. Convection is typically weak and less frequent than over the Midwest leaving the mesoscale structures to be determined largely by the interaction of the synoptic-scale flow with the resolved topography. Mesoscale variability is likely to be forced primarily by differences in the synoptic-scale flow rather than by differences in the model physics parameterizations.
Evaluation of the five-member ensemble using results from two six-month periods (January-June, 2000 and October 2000-March 2001) shows that running a multi-analysis, short-range ensemble is a worthwhile endeavor. The consensus forecast (the ensemble mean) at 12-km resolution exhibits comparable skill to 4-km deterministic MM5 forecasts during the same period. Limited ability for the ensemble to predict forecast skill is also shown by the high linear correlations between ensemble spread and ensemble mean errors. These high correlations increase when only the cases with extreme (very high or very low) spread are considered. Cases with a medium amount of spread display little spread/error correlation.
Probability characteristics of the ensemble forecasts are also studied. The missing rate (percent of time the verification falls outside the ensemble envelope) is found to be quite high, suggesting that the analyses used in this experiment do not adequately capture the initial state of the atmosphere. Precipitation forecasts are also included in the probability distribution analysis.
UW Mesoscale Ensemble Website: http://www.atmos.washington.edu/~epgrimit/ensemble.cgi
Session 9, Mesoscale predictability and ensembles
Wednesday, 1 August 2001, 9:30 AM-1:44 PM
Previous paper Next paper