Modeling shear stress distribution in natural small streams by soft computing methods

Abstract

In this study, artificial neural networks (ANNs) and adaptive neuro-fuzzy inference system (ANFIS) were used to estimate shear stress distribution in streams. The methods were applied to the 145 field data gauged from four different sites on the Sarimsakli and Sosun streams in Turkey. The accuracy of the applied models was compared with the multiple-linear regression (MLR). The results showed that the ANNs and ANFIS models performed better than the MLR model in modeling shear stress distribution. The root mean square errors (RMSE) and mean absolute errors (MAE) of the MLR model were reduced by 47% and 50% using ANFIS model in estimating shear stress distribution in the test period, respectively. It is found that the best ANFIS model with RMSE of 3.85, MAE of 2.85 and determination coefficient (R2) of 0.921 in test period is superior to the MLR model with RMSE of 7.30, MAE of 5.75 and R2 of 0.794 in estimation of shear stress distribution, respectively.