We have performed short-range (24 hour) forecasts using the NCEP Eta model in which closure constants in the cumulus parameterization scheme are varied across a range of plausible values. Although some previous studies have constructed ensembles by using models with different cumulus parameterization schemes or different values of closure constants, our study differs in that a multiple-dimension parameter space is constructed. This is done by varying two parameters within the Kain-Fritsch and Betts-Miller-Janjic schemes rather than only one parameter. Test cases will be shown in which two parameters are varied using four different values of each parameter, producing 4-by-4 matrices of simulations (i.e., 16-member ensembles). Results indicate that there are noticeable interactions between the two parameter specifications which would not be captured if only one parameter were varied at a time.

Ongoing research aims toward assessing value added by the ensembles in comparison to a reference forecast. Ensemble evaluations include relative operating characteristic as well as decomposition of mean square error into domain-mean and spatially-dependent components. Of special interest is the minimal ensemble size required to capture most of the value of the ensemble approach. Updated results will be presented in the conference.