A critical requirement to produce skillful and reliable Ensemble Streamflow Predictions (ESP) is skillful and reliable ensemble atmospheric forcing. Current atmospheric forecasts must be both re-scaled and downscaled to match the input requirements of hydrologic forecast models. Procedures also are needed to compensate for the strong tendency for atmospheric ensemble forecasts to under estimate the spread corresponding to the actual uncertainty. This paper presents the technique being used in the NWS Ensemble Streamflow prediction (ESP) system to produce ensemble forecasts for the Advanced Hydrologic Prediction Services (AHPS) program. To produce ensemble members the technique re-scales space-time patterns of historical observations so that the marginal distributions of the re-scaled data are consistent with current forecast marginal distributions. This assures that the space-time Spearman rank correlation structure of the ensembles is the same as the historical data at all space and time scales and between all atmospheric variables. It also assures that ensemble members from different tributaries are mutually consistent so that they can be combined to produce a skillful and reliable space-time structure of hydrologic forecasts at all forecast points.