Investigating the applicability of Neural Network method for estimating daily suspended sediment yield (Case study: Zard Drainage Basin, Khozestan Province)

In this study, to drawn the model of daily suspended sediment yield, simultaneous water and sediment discharge data of Machine Hydrometric Station, which is located on Zard River, Ramhormoz, Iran, were used. For this purpose, after elimination of statistical deficiency and exclusion of deviated data, the data were divided into two parts: 80% of the data were allocated for training and the other 20% of data were used for the examination of neural network. After standardization of the data, by using training data series, neural network with back propagation error algorithm was developed. Furthermore, by using the training data series, regression equation was developed between water and sediment discharges. For evaluating these two methods, the examination data series and the statistical parameters of R2, MAE and RMS were used. The amounts of R2, MAE and RMS for the neural network method are as follows: R2=0.62, MAE=1854 and RMS=3184. The amounts of the mentioned parameters for estimation using regression equation are: R2=0.54, MAE=1934 and RMS=3251. The results have shown that the estimation of suspended sediment yield using neural network model is more accurate in comparison to the regression equation estimates. But for reaching an optimum model, the processes of the data preparation, network architecturing and network training should be performed carefully and accurately. It is concluded that for the estimation of river suspended sediment yield, this model should be considered and used.