Proposed Daily Streamflow-Forecasting Model for Reservoir Operation

Abstract

Daily reservoir operations rely increasingly on accurate streamflow forecasts, which today are derived from computer modeling. This paper documents development of a hydrologic-process-based daily streamflow-forecasting model with a built-in parameter-estimation routine. The model [called the National Taiwan University (NTU) Daily Model] closely captures the physical reality while maintaining the robustness of the statistical model, both of which must be considered in the day-to-day operations of a reservoir. Several equations are used to describe various simplified hydrologic processes, such as direct runoff, baseflow, infiltration, evapotranspiration, and change in ground-water storage. It is assumed that these equations can generate total daily streamflow from precipitation forecasts and initial conditions. The model parameters are automatically determined by a built-in Gauss-Newton algorithm. Model performance was analyzed by five statistical error measurements, including correlation coefficient, coefficient of efficiency, relative error, standard error, and volume error. The proposed model was calibrated and verified using historical streamflow data from three gauging stations in Taiwan. The results obtained indicate that the NTU Daily Model for streamflow forecasting is satisfactory for use in daily operation of the reservoir.