Comparison of Artificial Neural Network Models for Sediment Yield Prediction at Single Gauging Station of Watershed in Eastern India

This paper describes the application of two different neural network models, the standard-back propagation (SBP) model and the radial basis neural network (RBNN) model, to predict monthly sediment yield as a function of monthly rainfall and runoff during the rainy season for a watershed area in India. Four scenarios were considered to determine the type and number of inputs for the artificial neural network (ANN) model. It was observed that in the small and forested watershed of Nagwa, the inclusion of monthly precipitation and average discharge values improved the performance of the ANN model in the estimation of monthly sediment yield. The momentum rate, number of nodes at the hidden layer, number of nodes at the prototype layer, linear coefficient, learning rule, and transfer functions were optimized based on lowest root-mean-square error and highest correlation coefficient values. The optimized parameters were used for the SBP and RBNN models. During validation periods, the RBNN model was closer to the observed values than SBP. The mean annual observed sediment yield was 3.7 t/ha. The mean annual simulated sediment yields were found to be 3.1 and 3.5 t/ha in SBP during training and validation periods. RBNN simulated mean annual sediment yields of 3.6 and 3.5 t/ha during training and validation periods. The results are indicative that the RBNN model is more appropriate for forecasting/simulating the sediment yield at a single point of interest in agricultural watersheds. DOI: 10.1061/(ASCE)HE.1943-5584.0000601. (C) 2013 American Society of Civil Engineers.