One-Day-Ahead Streamflow Forecasting Using Artificial Neural Networks and a Meteorological Mesoscale Model

Abstract

An approach to modeling daily flows using artificial neural networks (ANNs) is presented. In addition to previous streamflow values and mean areal rainfall sequences, a new runoff index was used and tested as ANN input. This runoff index was generated as a combination of two output variables of the weather research and forecasting (WRF) mesoscale model, which contains an integrated land surface model. Inclusion of the new index improved ANN model performance and increased simulation skill. A case study was conducted for the northeast Guadalquivir catchment in southeastern Spain. Accurate one-day-ahead streamflow forecasts were achieved in terms of overall fit and timing of peaks. Model performance was satisfactory, with a persistence index (PI) equal to 0.81 and a Nash–Sutcliffe efficiency R2 equal to 0.95 for an independent data set. These favorable results prove that WRF outputs contain useful information on the hydrologic state of a basin and can therefore be used as valuable ANN inputs.