Department of Civil and Environmental Engineering
University of Washington
The University of Washington experimental west-wide seasonal hydrologic forecast system produces 6-12 month lead time hydrologic forecasts at approximately 100 forecast points in five major river basins within the western U.S. The system is an outgrowth of the North American LDAS (N-LDAS) project, and uses the N-LDAS 1/8 degree spatial grid, as well as N-LDAS vegetation, soils, and other data. At present the system is based on the University of Washington/Princeton University Variable Infiltration Capacity (VIC) macroscale hydrology model, driven by climate ensembles downscaled from the NCEP Seasonal Forecast Model (SFM), the NASA NSIPP1 global model, and an ensemble version (based on the ôSchaake Shuffleö) of the CPC official seasonal outlooks (12 month lead time). As a benchmark, we also produce parallel forecasts via the well-known Extended Streamflow Prediction (ESP) method, and a further conditioning of the ESP ensembles by ENSO and PDO state. The primary forecast products are: 1) monthly streamflow distributions and runoff volume statistics at the specified forecast points; and 2) west-wide spatial maps of monthly forecast ensemble averages for runoff, soil moisture, and snow water equivalent (SWE). We report results of initial real-time testing of the system with bi-monthly updates for the Pacific Northwest (for winter 2002-3), and for a larger expanded domain (most of the U.S. west of the Rocky Mountains) for winter 2003-4. To improve estimation of initial hydrologic conditions, we developed a simple method for assimilating observed snow water equivalent anomalies at the start of the forecast. We have also attempted to address the relative dearth of meteorological observations in the final months before the forecast start (which hampers the spin-up simulation of initial model state) using interpolated monthly precipitation percentiles and temperature anomalies from a set of real-time index stations. In this presentation, we survey the methods and results used during the first two years of the forecasting system, and discuss some challenges inherent in real-time forecast system implementation. We also describe ongoing work to extend the system to incorporate multiple hydrologic models so as to form multi-model ensembles, to increase the forecast update frequency, and to include shorter (two week lead) forecasts.