Verification of a Downscaling Sequence Applied to Medium Range Meteorological Predictions for Global Flood Prediction

We describe a prototype system for medium range (up to two week lead) flood prediction in large rivers, which is intended for global implementation - particularly in river basins having limited in situ meteorological observations. The procedure draws from the experimental North American Land Data Assimilation System (NLDAS) and the University of Washington West-wide Seasonal Hydrologic Forecast System for streamflow prediction. Meteorological forecasts based on a numerical weather prediction model serve both as the forcing for hydrologic model initialization and forecasts for lead times up to fifteen days. The hydrologic component of the system is the Variable Infiltration Capacity (VIC) macroscale hydrology model. In the prototype, VIC is spun up for forecast initialization using daily ERA-40 precipitation, wind, and surface air temperature. In hindcast mode, VIC is driven by global NCEP ensemble 15-day re-forecasts (NOAA/ESRL) that are bias corrected with respect to ERA- 40 and spatially disaggregated using two higher spatial resolution satellite products: Global Precipitation Climatology Project (GPCP) 1DD daily precipitation and Tropical Rainfall Measuring System (TRMM) 3B42 precipitation are used to spatially disaggregate NCEP re-forecasts precipitation during the 15-day forecast period. The use of forecast models and satellite remote sensing data in this procedure reduces the need for in situ precipitation and other observations in parts of the world where surface networks are critically deficient, but where a global hydrologic forecast capability arguably would have the greatest value. The prototype system was implemented at one-half degree spatial resolution and tested during the 1979-August 2002 period. For the Mississippi R. Basin (where ample data for model evaluation exist) we evaluate the spatial disaggregation step in which observed precipitation products (NARR) are first aggregated to a coarser resolution (for the sole purpose of the evaluation) and then used in the spatial disaggregation step. The output of this procedure is compared to the original high resolution data. We also compare our disaggregation scheme with the analog technique of Hamill and Whitaker. Finally, we verify forecast error statistic¬s resulting from the application of the entire downscaling sequence.