The 4-km ensemble continued to use three dynamic cores (WRF-ARW, WRF-NMM and ARPS) for a total of 26 ensemble members. Six of the ARW members were designed to facilitate the evaluation of several new microphysics and PBL schemes available in WRF-ARW. Three ARW members contained initial condition perturbations of different scales to allow for the study of multi-scale error growth and predictability. Forecasts from consistent NCEP SREF members were used to provide lateral boundary conditions at hourly intervals. The ensemble-transform-initialized SREF provided initial condition perturbations for most SSEF members. Level-2 radial velocity and reflectivity data from over 120 operational WSR-88D radars were analyzed into the initial conditions of all except for three members using the ARPS 3DVAR together with its cloud analysis package. Thirty-hour forecasts were produced once per day, initialized at 00 UTC. In addition, a 1-km deterministic forecast was produced daily, covering the same CONUS grid. To complete the daily 30-hour forecasts, a dedicated Cray XT-4 with 18,000+ CPU cores was used for about 5 hours a day.
This paper will discuss the scientific problems and issues to be addressed, report on the design of the prediction experiments, the associated scientific and technical challenges as well as logistic issues. It will present ensemble and deterministic forecasting examples, and preliminary verification results aginst observations. Particular emphasis will be given to the value of radar data assimilation and the effect of resolution on short- and longer-range forecasts. The relative performance of different physics packages will also be examined. A companion paper (Kong et al. 2010) will focus on the realtime production and performance evaluation of the ensemble/probabilistic forecasting products.