An examination of forecasts from three winter storms is presented using data from the operational National Centers for environmental predictions (NCEP) stepped terrain (Eta), global spectral model (AVN), and the Short-Range Ensembles Forecast system (SREF). The current SREF system is composed of 10 forecast members. These cases demonstrate both the strengths and limits of the current NCEP short-range ensemble suite. Additionally, they show the large variation in the forecasts produced by the Eta and AVN and why ensembling may offer an improved forecast methodology over any single deterministic model.

Comparisons of forecasts of frontogenesis and quantitative precipitation (QPF) between the SREF, AVN, and Eta forecasts are presented. Preliminary results show the large variation in the placement of strong forcing and the QPF between the two models and the SREF mean. These data also show the impact of grid resolution on the forecasts. During the case from 6 January 2002, the Eta was run on a 12-km grid and examined on a 40-km grid. The 12-km Eta data showed detailed areas of forcing and QPF relative to the coarser AVN and SREF forecasts. However, these detailed forecasts did not necessarily translate into more accurate forecasts.

Data were displayed using traditional techniques and formats. These included the forecast position and intensity of the surface cyclone, the 850-hPa isotherms, QPF, and 850-hPa frontogenesis. Additional displays of ensemble data included traditional spaghetti plots, consensus forecasts, and probabilistic forecasts.

In at least one case, it appeared that updating the SREF forecasts with the next cycle's forecasts from the Eta and AVN, creating a 12-member ensemble, offered an improved forecast over the original 10-member SREF ensemble. Since the latter two forecasts were from more recent data, they were considered to be more skillful than any single member of the 10-member SREF ensemble. Therefore, weights were assigned to all members, with larger weights assigned to the more recent Eta and AVN forecasts. This led to improved forecast over the non-weighted 12-member ensemble and the original 10-member SREF ensemble forecast.