We describe an implementation of the Variable Infiltration Capacity (VIC) macroscale hydrology model over the western U.S. at 1/8 degree spatial resolution for experimental ensemble hydrologic prediction at lead times of six months to a year. Seasonal climate forecast ensembles are statistically downscaled directly from the NCEP Global Spectral Model (GSM) and the NASA NSIPP model. In addition, official probabilistic CPC outlooks are downscaled via a weather generation approach and used to drive hydrologic forecasts. As a benchmark, we also use the VIC model to produce parallel forecasts via the well-known Extended Streamflow Prediction (ESP) method. The ESP forecasts are further composited to provide ENSO and PDO conditioned ensembles, which past work has shown can considerably reduce seasonal forecast error variance. We report results from initial testing in real-time, which began with bi-monthly updates in winter 2002-03. Several upgrades to the hydrologic forecasting system that were implemented during the test period are described, including the development of a method for assimilating snow water equivalent observations at the start of the forecast, and the implementation of real-time daily surface forcing estimation using a set of index stations. We also describe the development of a set of reservoir system models for the western U.S., and their implementation within the system to produce ensemble forecasts of reservoir system storages, operations and releases. Other forecast products from this system include monthly spatial 1/8 degree maps of soil moisture, snowpack and runoff (as percentiles), streamflow hydrographs and streamflow volume forecasts.