P. Moazami Goodarzi - MSc Student, Civil Engineering Department, Iran University of Science & Technology
E. Jabbari - Associate Professor, Civil Engineering Department, Iran University of Science & Technology

In order to control the sedimentation destructive consequents and to evaluate the erosion in the basin, it is of great importance to estimate the river sediment transport rates precisely. When there are several mathematical and experimental models with unacceptable accuracy or when there is deficiency in records of input variables to the classical equations, machine learning techniques are perfect completes. Multiple Additive Regression Trees or TreeNet is one of machine learning techniques, which has been applied limitedly in water engineering environment so far, as reports say; therefore, its efficiency in comparison with other common models in this field is not evident. In this context, this model was compared with ANN and ANFIS models and proved competitive results in two different conditions regarding the target variable coefficient of variation, without giving negative predictions like the other two models did. Also the other advantages which distinguish the TreeNet model from the other two are its running speed and not being parametric. In present research, the two Kareh Sang and Sira stations (located on Haraz and Karaj Rivers respectively) data have been utilized. In Kareh Sang station, a parameter named day of the year which simulates the periodic property of sediment transport was introduced to models as a predictor variable. Applying this parameter leaded to the considerable increase in models accuracy, particularly in that of TreeNet model, which indicates that there are different relations between sediment discharge and flow discharge in different time periods of the year

TreeNet, Suspended sediment, Day of the year, ANN, ANFIS