Investigation of the Performance of Classical and Artificial Intelligence Approaches in Prediction of Roughness Coefficient in Meanders

Accurate prediction of the river's roughness coefficient is always one of the most important and substantial issues in the hydraulic modeling of open channels. In the current research, the Support Vectore Machine (SVM) and Adaptive Neuro-Fuzzy Inference System (ANFIS) intelligent approaches were used to estimate the hydraulic roughness of meandering rivers, and the impacts of different variables including: channel slope, channel sinuosity and also, hydraulic parameters such as Reynolds number on prediction of the roughness coefficient in these types of channels were investigated. On the other hand, the obtained results were compared with classical methods. In order to model the roughness coefficient, two experimental data series related to the sinusoidal shaped channels were used. The obtained results showed that the SVM and ANFIS intelligent methods are more accurate and reliable in estimating the Manning roughness coefficient in natural rivers than semi-experimental formulas. It was observed that in estimation of the Manning roughness coefficient, the model with input variables of α (shape factor), Sr (sinusoidal coefficient), S0 (channel slope) and Re (Reynolds number) leads to the more accurate results. The results showed that in estimating the roughness coefficient in the meandering rivers, the effect of the shape factor on increasing the accuracy of the models is more than the sinusoidal coefficient. Also, the results of sensitivity analysis indicated that the channel slope is the most effective parameter in estimating the roughness coefficient in the meandering rivers.