Prediction of suspended sediment yield by artificial neural network and traditional mathematical model in Mahanadi river basin, India

Abstract

Estimation of sediment yield is essential towards understanding the mass balance between the ocean and land. Direct measurement of suspended sediment is difficult as it needs sufficient time and money. The suspended sediment yield depends on a number of variables, and their inter-relationships are highly non-linear and complex in nature. In this paper, soft computing-based sediment yield estimation algorithms are proposed for the Mahanadi river basin. A multilayer perceptron (MLP) artificial neural network (ANN) with an error back-propagation algorithm using historical monthly hydro-climatic data (temperature, water discharge and rainfall) was employed to predict the suspended sediment yield at the Tikarapara gauging station, which is the farthest downstream station in the Mahanadi river. The results demonstrated that water discharge and rainfall are significant controlling parameters of suspended sediment in the Mahanadi River. The comparative results show that the feed-forward back-propagation with Levenberg–Marquardt (FFBP–LM) is the best model for suspended sediment yield estimation, and provides more reasonable prediction for extremely high and low values. The performance of the sediment rating curve (SRC) model was below expectations as it produced the least accurate results for the peak sediment values, as well as overall model performance. It is also noticed that the multiple linear regressions (MLR) model predicted negative sediment yield at low values; which is completely unrealistic as suspended sediment yield cannot be negative in nature. It was also observed that suspended yield prediction by ANN was superior compared to that using MLR and SRC models. The proposed model will be beneficial for sediment prediction where estimates of suspended sediment values are unavailable.