To forecast streamflow in headwater basins, the NWS utilizes what is known as the National Weather Service River Forecast System (NWSRFS). The system consists of three conceptual models: (1) a spatially lumped rainfall-runoff model, (2) a spatially lumped soil moisture accounting model, which is a modified version of the Sacramento model, and (3) a channel routing model in the form of a cascade of nonlinear reservoirs.
In routine operation, streamflow forecast is computed once a day for up to 10 days into the future. For instance, the Ohio River Forecast Center disseminates forecasts for about 110 river gauges. The forecast for a gauge specifies estimates of stages at 7 AM in 24, 48, and 72 hours, and estimates of flood crest and time to crest, if a flood event is probable. During floods, updated forecasts are produced more frequently.
The NWSRFS has received two enhancements. Georgakakos and Smith [1990] implemented an extended Kalman filter that updates the system states based upon discharges at the basin outlet observed up to the forecast time. The implementation assumes that the dominant sources of forecast uncertainty are errors in estimates of model parameters and errors in observations of model inputs. Under these constrictive assumptions, the authors have overcome the deficiency of earlier hydrologic Kalman filters which lacked a procedure for estimating the time-varying covariance matrix of errors. Tests confirmed improvements in forecasts as well as the operational feasibility of the method.
The second advance has been reported by Sorooshian et al. [1993] who had developed a global optimization procedure well suited to estimating parameters of conceptual hydrologic models, which are known to have response surfaces (relations between output errors and parameter values) dotted with numerous local optima. The strategy, dubbed the shuffled complex evolution method, was subjected to a test in which values of 13 parameters of the Sacramento soil moisture accounting model were optimized. Although the tests were limited, the new method consistently outperformed the well-known multistart simplex method, raising hopes that the day may be approaching when a hydrologist will be able to entrust the tedious task of parameter estimation to an automatic algorithm.
Applied activities have been the subject of two papers. Rango and Martinec [1994] have reported on a comparative test of seven hydrologic models conducted by the World Meteorological Organization. The test, which included flood events, provided an assessment of potential accuracy of daily snowmelt runoff forecasts for lead times of up to 20 days. Barrett [1993] has described an application of NOAA's knowledge and technology to the development of a hydrometeorological forecast system for the Nile River in Egypt. The system will provide daily forecasts of inflow to the High Aswan Dam and hydrographs at selected gauges upstream.