The prediction of river sediment load is an essential issue in water resource engineering problems. In this study, artificial neural network employed in order to estimate the daily sediment load on rivers. Two different algorithms, multi-layer perceptron (MLP) and hybrid MLP-FFA (MLP integrated with the FFA) were used for this purpose in the Lake Mahabad, Iran. For this purpose, nine different scenarios are considered as inputs of the models. Performance of selected models was evaluated on basis of performance criterion namely root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R 2 ) for choosing best fit model. The results indicated that the new hybrid model MLP-FFA is successful in estimating sediment load with high accuracy as compared with its alternatives with RMSE = 2018 ton/day, MAE = 1698 and R 2 = 0.95, which were much lower than those of MLP-based model with RMSE = 3044 ton/day, MAE = 2481 and R 2 = 0.90. The results of the present study confirmed the suitability of proposed methodology for precise modeling of suspended sediment load.

Selecting appropriate inputs for intelligent models is important due to reduce costs and save time and increase accuracy and efficiency of models. The purpose of this study is using Shannon entropy to select the optimum combination of input variables in time series modeling. Monthly time series of precipitation, temperature and radiation in the period of -was used from Tabriz synoptic station. Precipitation, temperature and radiation parameters with different delays are considered as input to the Shannon entropy. The results showed that time series with three delays provide the better results for the modeling. Applying Bayesian network and multivariate linear regression analysis were performed. Models performance was evaluated using three criteria: coefficient of determination (R ), root mean square error (RMSE), and the dispersion. Index (SI). The results indicated that Bayesian neural network model shows the best performance to simulate time series of precipitation, temperature and radiation in compare to multivariate linear regression analysis. The results showed that Shannon entropy has better performance in selection of the appropriate entry into intelligent models.