Prediction of daily sediment discharge using a back propagation neural network training algorithm: A case study of the Narmada River, India

Abstract

Abstract Most of the studies on Artificial Neural Network (ANN) models remain restricted to smaller rivers and catchments. In this paper, an attempt has been made to correlate variability of sediment loads with rainfall and runoff through the application of the Back Propagation Neural Network (BPNN) algorithm for a large tropical river. The algorithm and simulation are done through MATLAB environment. The methodology comprised of a collection of data on rainfall, water discharge, and sediment discharge for the Narmada River at various locations (along with time variables) and application to develop a three-layer BPNN model for the prediction of sediment discharges. For training and validation purposes a set of 549 data points for the monsoon (16 June-15 November) period of three consecutive years (1996–1998) was used. For testing purposes, the BPNN model was further trained using a set of 732 data points of monsoon season of four years (2006–07 to 2009–10) at nine stations. The model was tested by predicting daily sediment load for the monsoon season of the year 2010–11. To evaluate the performance of the BPNN model, errors were calculated by comparing the actual and predicted loads. The validation and testing results obtained at all these locations are tabulated and discussed. Results obtained from the model application are robust and encouraging not only for the sub-basins but also for the entire basin. These results suggest that the proposed model is capable of predicting the daily sediment load even at downstream locations, which show nonlinearity in the transportation process. Overall, the proposed model with further training might be useful in the prediction of sediment discharges for large river basins.