JWSS - Journal of Water and Soil Science

---

Many sediment transport equations have been developed for estimation of the river sediment materials during the past four decades. There are significant differences in the results from these equations when applied to compute sediment transport for a specific river. Therefore, application of an equation for estimation of a river sediment load is not an easy task. In this study, 12 important sediment transport equations including Meyer-Peter and Muller, Einstein, Bagnold, Engelund and Hansen. Toffaleti, Ackers and White, Yang, Van Rijn, Wiuff, Samaga et al, Beg and Fazel were tested against the measured field data of four major Khuzestan rivers, namely, the Karoon, the Dez, the Karkheh, and the Maroon. For accurate results and rapid computation, a computer program was developed for this purpose. Over 490 measured data from the gauging stations of these rivers are selected. Using these data, the hydraulic parameters and the bed material of the gauging stations are determined.

The results of the computer program are analyzed and compared with the measured data. The results from this study show that those equations which are based on the energy exchange of the flow, are generally in good agreement with the measured data for Khuzestan Rivers. From these equations, the Engelund and Hansen’s equation generally predicts satisfactory results for the all gauging stations except for the Maroon River gauging station. And finally if the sediment load computed by the Beg’s method is multiplied by a factor of 0.1, the results approximately match those obtained from the Engelund and Hansen’s method.

---

  Prediction of sediment load transported by rivers is a crucial step in the management of rivers, reservoirs and hydraulic projects. In the present study, in order to predict the suspended sediment of Taleghan river by using artificial neural

network, and recognize the best ANN with the highest accuracy, 500 daily data series of flow discharge on the present day, flow discharge on the past day, flow depth and hydrograph condition (respectively with the average of 13.83 (m³/s), 15.42 (m³/s), 89.83 (cm) and -0.036) as input variables, and 500 daily data series of suspended sediment, as the output of the model were used. The data was related to the period of 1984-2005. 80 different neural networks were developed using different combinations of variables and also changing the number of hidden-layer neurons and threshold functions. The accuracy of the models was then compared by R² and RMSE. Results showed that the neural network with 3-9-1 structure and input parameters of flow discharge on the present day, flow discharge on the past day and flow depth was superior (R²= 0.97 and RMSE= 0.068) compared to the other structures. The average of the observed data of sediment and that predicted by the optimal model (related to test step) were 1122.802 and 1184.924 (tons per day), respectively.