Prediction of Daily Streamflow Using Jordan-Elman Networks

The prediction of daily streamflow is required for future planning in water resource activities. This study presents the application of the Jordan-Elman network with the Levenberg-Marquardt algorithm. Prediction was made by using flow data of gauging station no. 2122 on Birs River, Switzerland between 2000 and 2010. The data, 4018 days in total, were used as calibration and validation sets for the chosen Jordan-Elman Neural Network architecture. Of the data obtained, 2922 days (1 super(st) January 2000 - 31 super(st) December 2007) were reserved for calibration, and remaining data were used for validation. In total, six different models were developed, based on the prediction of current flow from up to six-days-ahead flows. Mean square error (MSE), Nash-Sutcliffe Sufficiency Score (NSSS) and coefficient of correlation (R-value) were used as performance criteria. Model Mg (six-days- ahead flows) gave the best results, with respect to all prediction performance criteria.