Suspended sediment estimation using regression and artificial neural network models: Kebir watershed, northeast of Algeria, North Africa

Abstract

The focus of this research was to identify potential equivalences between artificial neural networks and statistical regression and to verify these equivalences when applied to modelling sediment loads in the Kebir river. The use of feed-forward backpropagation neural networks such as MLP and LM were studied by applying relationship stream flow-sediment discharge data and geomorphology watershed parameters. Daily based water and sediment discharge were used as inputs for sediment rating curve and ANNs. In the present study the models were adopted by changing numbers of neuron in hidden layers and epoch. Results have shown that the ANN models were superior in reproducing sediment discharge compared to SRC. The findings further have suggested that LM could provide the most accurate estimates of sediment discharge, (R2 and EF of 0.94) compared to SRC had lower values of R2 and EF (0.89, 0.88), and resulted in underestimations of sediment discharge (−15%).