Comparison of the efficiency of artificial neural networks method and regression model, sediment rating curve, for daily suspended sediment estimation

Estimation of soil erosion and sediment yield in a river is a difficult task and severalmethods have been suggested for its estimation. One the new methods in riverengineering and suspended sediment estimation is application of artificial neuralnetworks which uses the same algorithm of human brain to find out the internal relationbetween data based on the training process. The objective of current study is to explorethe capability of artificial neural networks method for estimation of daily suspendedsediment in Kharestan watershed located in the northwest of Fars province, Iran. Thestudy of efficiency is based on the comparison of neural network with regressionmodels. For this purpose, 22 years of water and sediment discharge data of ShoorKharestan River were considered and tested for outliers. Then the estimation was donebased on neural networks and linear regression method (sediment rating curve) andwere compared based on RMSE, MAE and R2. The results showed that estimation ofneural network is more accurate than that of linear regression (sediment rating curve).The estimations of RMSE, MAE and R2 for neural networks method was 19.27, 12.14and 0.98 respectively while these values for linear regression were 36.84, 20.75 and0.74 which showed the lower errors of neural networks method compared with linearregression.