Artificial Neural Networks Application to Sensitivity Analysis of Effective Parameters on Suspended Sediment Load; C ase Study: Ligvanchay River

In this study, Artificial Neural Network (ANN) as a black box model was used in order to evaluate the temperature and water discharge effects on suspended sediment load of Ligvanchay River. For this purpose, the hydrological data such as mean river flow discharge, mean daily temperature and mean daily suspended sediment load of Ligvanchay at the outlet hydromerey station were collected and divided to seven categories. These data were then entered to the ANN and the model with Levenberg-Marquardt (LM) of back propagation training algorithm was executed about 2000 times to determine the optimum structure of ANN model includes numbers of hidden layers and neurons in each layer. After determine the optimum structure, other back propagation training algorithms were tested for doing more comparison among the algorithms; in this manner the results of training and verifying as graphs and tables were presented. Furthermore, the results of the ANN model were compared with the results of some other classical black box models such as multi and single linear regression, auto regressive (AR) time series models. The obtained results of the study showed better performanceof the ANN model in comparison of the others when only one neuron was considered as model input. However when more than one neuron was used as input data, MLR and AR (2) showed a hit better results. This may be because of noise propagation in ANN non-linear model in comparison with other linear models. According the results, it can be obviously seen that the suspended sediment load of Ligvanchay has only a little sensitivity to the environment temperature.