The suspended sediment load modeling by artificial neural networks, neural-fuzzy and rating curve in Hlilrood watershed

With regard to financial and technical problems normally measured sediment data are limited in developing countries; therefore a model that uses water discharge data as input can be a reliable option for estimates of sediment. Due to widely application of the variety of models to predict the suspended sediment, this study aims to determine optimal prediction model based on the amount of discharge flow gauging stations of Halilrood River including, Soltani, Henjan, Cheshmeh Aroos, Meydan and Konaruiyeh. In this regard, efficiency of some rating curves models including one-linear, two-linear and the intermediate categories ones (by and without coefficients as CF1, CF2 and FAO) and black box models including artificial neural networks and neural-fuzzy in modeling sediment were evaluated. The results of the evaluation of the model using the parameters of MAE and RMSE showed that neuro-fuzzy models in major hydrometric stations studied, including Pole Baft, Henjan and Konaruiyeh with an equivalent amounts of 35.07, 11958.74 and 34235.27 ton/day for MAE and 42.07, 28672.78 and 52735.92 ton/day for RMSE, respectively are the best models to simulate the suspended sediment. The artificial neural network model of radial basis function in Meydan with 384.83 ton/day MAE and 669 ton/day RMSE amounts is the optimal model. Also two-linear sediment rating curve resulted the best simulation in Cheshmeh Aroos Station with MAE and RMSE as 1.7 and 4.1 ton/day and one-linear sediment rating curve with CF1 correction in Soltani Station with MAE and RMSE 9723.2 and 41235.6 ton/day, respectively are the best. According to changes of efficiency of models with varying location of gauging stations, it can be concluded that ecological conditions and statistical community determine the optimal model of the suspended sediment simulation.