Estimation of Manning Roughness Coefficient in Alluvial Rivers with Bed Forms Using Soft Computing Models

Flow conditions (flow discharge, flow depth, and flow velocity) in natural streams are mainly determined via the flow resistance formula such as Manning’s equation. Evaluating the accurate Manning’s roughness coefficient (n), especially in rivers with bed form during floods, to obtain more reliable results has always been of interest to scholars. The interaction between the flow and bed form is very complex since the flow conditions control bed forms, and vice versa. The main goal of the present study is to predict n in rivers with bed forms, using soft computing models, including multilayer perceptron artificial neural network (MLPNN), group method of data handling (GMDH), support vector machine (SVM) model, and genetic programming model (GP). To this end, the energy grade line ( $${S}_{f}$$ S f ), flow Froude number (Fr), the relative submergence ( $$y/{d}_{50}$$ y / d 50 ; y = flow depth and d50 = bed sediment size), and the bed form dimensionless parameters ( $$\Delta /{d}_{50}$$ Δ / d 50 , $$\Delta /\lambda$$ Δ / λ , and $$\Delta /y$$ Δ / y ; ∆ = bed form height and λ = bed form length) were used as the input variables, and n was used as the output variable. The results showed that all the test models have acceptable accuracy, while the SVM model showed the highest level of accuracy with the coefficient of determination $${R}^{2}=0.99$$ R 2 = 0.99 in the verification stage. The sensitivity analysis of SVM and MLPNN models and the structural analysis of GMDH and GP models indicated that the most important parameters affecting n are Fr, $${S}_{f}$$ S f , and $$\Delta /\lambda$$ Δ / λ .