Nonparametric framework for long-range streamflow forecasting

The extended streamflow prediction (ESP) procedure of the National Weather Service River Forecast System (NWSRFS) produces long-range forecasts of streamflow through the use of hydrologic models and historical hydrologic data. An important element of the ESP procedure is converting hydrologic-model output to estimates of a forecast random variable. In this paper, nonparametric statistical procedures are developed for combining hydrologic models and historical hydrologic data into long-range streamflow forecasts. Although these procedures are developed for use within the ESP system, they should be broadly applicable to problems of long-range streamflow forecasting. Two notable features of the procedures developed in this paper are: (1) Climate information is easily incorporated; and (2) hydrologic-model errors can be accommodated. Results are presented for a test implementation of ESP for a basin in the southeastern United States during the severe drought period of 1988. The relative importance of climate information and soil moisture information for long-range streamflow forecasting is compared and contrasted.