Short-range ensembles were run using the non-hydrostatic Pennsylvania State University -National Center for Atmospheric Research (NCAR) Mesoscale Model MM5. The model was configured with a 36-km outer domain covering the eastern Pacific and western United States, and a 12-km inner domain covering the Intermountain West. Simulations were run for the period of the Intermountain Precipitation Experiment (IPEX), a field and research program examining precipitation processes over northern Utah that was held from 31 January - 25 February 2000. Ensemble members were generated using two methods. The first method was to perturb the initial and boundary conditions provided by the NCEP Eta Model using the technique developed by Errico and Baumhefner (1987). The parameters that were perturbed were: 1) u and v wind; 2) sea level pressure; 3) height; 4) dew point depression; 5) temperature; and 6) precipitable water. The second method consisted of varying MM5 physics packages, including a combination of Reisner and "simple ice" microphysics parameterizations, and Betts-Miller-Janic, Grell, and Kain-Fritsch convective parameterizations. Six members from each method were included in the ensemble.
The performance and utility of the ensemble was evaluated using high-density surface data from the MesoWest networks, a collection of 40 collaborating mesonets over the western United States, and conventional and specialized radiosonde observations collected during IPEX. Recent studies have suggested that including different model physics configurations in an ensemble prediction system (EPS) would result in larger spread compared to an EPS based entirely on initial condition perturbations. We will present results, however that show much of the spread within our EPS is due mainly to the uncertainty of the initial state. This may be due to the strong forcing for mesoscale wind and precipitation systems that is provided by the topography of the Intermountain West. Although the spread was larger for the initial condition component of the EPS, upper-level error statistics were nearly identical to those of the model physics component of the ensemble. Implications of these findings for EPS design and operational weather prediction over the Intermountain West will also be discussed.