87th AMS Annual Meeting

Wednesday, 17 January 2007: 10:45 AM
Evaluation of Streamflow Forecasts Based on Coupled GFS-Noah Ensemble Forecasting System
213A (Henry B. Gonzalez Convention Center)
Dingchen Hou, NOAA/NWS/NCEP/EMC and SAIC, Camp Springs, MD; and Z. Toth, K. Mitchell, D. Lohmann, and H. Wei
A major thrust of coupled hydrological modeling is directed toward forecasting streamflow for river basins of various size, based on the operational Numerical Weather Prediction (NWP) systems. Hydrological forecasts are sensitive to atmospheric forcing, especially precipitation forecasts. Since precipitation forecasts exhibit large uncertainties, hydrologic forecasts must be framed in a probabilistic form. It is generally accepted that to capture case dependent variations in forecast uncertainty, one must follow an ensemble approach.

The recent implementation of the National Centers for Environmental Prediction (NCEP) North America Land Surface Data Assimilation System (NLDAS), the coupling of the Noah Land Surface Model (Noah LSM) with the Global Ensemble Forecast System (GEFS), and the development of a river routing model provide an opportunity for exploring the feasibility of distributed ensemble stream-flow predictions. To generate river flow initial conditions (also used for forecast evaluation), observed precipitation values are used to force the NLDAS system that in turn produces the required runoff values for forcing the river flow model. Experimental river flow forecasts are then generated by forcing the same river flow model with runoff values from coupled atmosphere - land surface model forecasts (GFS - Noah LSM), executed in an ensemble mode, using the operational NCEP GEFS system. Both the land surface and river flow models are represented on a 1/8-degree latitude/longitude grid that spans the CONUS domain.

A subjective evaluation and statistic verification of preliminary experiments indicates the following. (1) The variability in the ensemble river flow forecasts is on the same order of magnitude as the error in the mean of the ensemble. (2) For large basins, the ensemble river flow forecasts appear to well capture analyzed variations. (3) For medium- and small basins, a serious under-dispersion, which is due to the lack of variability of precipitation forcing on the scale of the river flow model, is present at short to medium lead times.

Supplementary URL: