To examine seasonal climate predictability using regional models, in this study we developed and tested a high resolution Regional Climate Model (RCM). The model was based on the NCEP operational Eta model (as of 24 July, 2001, namely, the Eta model version in the NCEP 25-year Regional Reanalysis), with changes made to make the model run over a longer time period and to update the sea surface temperature (SST), sea ice, greenness fraction, and albedo fields on a daily basis. The model was run on the same large domain as does the operational Eta model and Regional Reanalysis, with a resolution of 32 km and 45 levels, as used in the Regional Reanalysis. Presently, the model can be executed off of analyzed lateral boundary conditions of the NCEP Global Reanalysis I and II or predicted lateral boundary conditions from the NCEP global Seasonal Forecast Model (SFM).

To test the skill of the Eta RCM in predicting warm season precipitation, two summertime cases (2001 and 2002) were chosen. Most previous studies of RCM seasonal simulation were driven by analysis lateral boundary conditions and observed SST, and were initialized from one single date. In contrast, we used both predicted boundary condition fields and predicted SST, and executed 5 members whose starting dates vary by one day. The lateral boundary condition fields were provided by the NCEP SFM model. The SST fields were provided by the NCEP coupled ocean-atmosphere general circulation model consisting of sixteen forecasts made from a combination of four weekly oceanic initial conditions and four atmospheric initial conditions within one calendar month. The study period is from June to September and the executions were started from late May and continued to early October. Results obtained from the two years are compared and our focus is the interannual variability in precipitation.

We examine the resulting ensemble mean to demonstrate whether a) the Eta RCM successfully captures both wet and dry interannual anomalies in total precipitation over some U.S. regions between the two years and b) there are substantial member-to-member variabilities in both total monthly precipitation and surface energy and water balance. These results suggest that the Eta RCM is sensitive to the choice of initial starting date and that previous RCM studies that employed only "one member" and initialized from one single date may be misleading by failing to represent the inherent internal variability, indicating that a choice of ensemble strategy embracing different initial starting dates, tuning parameters, and physical packages is important to warm season precipitation prediction.