Artificial Neural Network Approach for Hydrologic River Flow Time Series Forecasting

Stream flows are highly non-linear and complex in nature in any hydrological system. Lack of data availability for modeling requires use of specific model with high accuracy and efficiency which are capable in generating accurate hydrological response. In this paper, capability of artificial neural network model was investigated by utilizing eight years' (2009-2016) of daily river flow and rainfall data for the Sot river catchment in Uttar Pradesh state of India. For investigating the effect of variables on river flow prediction, four ANN models were calibrated and validated for the period of 2009 to 2014 and 2015 to 2016, respectively, using different combinations of input parameters. Accuracy of developed ANN models was assessed by four performance indicators, namely correlation coefficient, root-mean-square error, modified Nash-Sutcliffe efficiency and Modified Index of Agreement. The results of the study indicate that performance of ANN model improves drastically after including the lag 1 river flow as an input parameter. It is concluded that the ANN model has the ability to solve highly complex non-linear river flow prediction and can be further used for the generation of synthetic river flow data of the Sot river catchment.