Suspended sediment load prediction using artificial intelligence techniques: comparison between four state-of-the-art artificial neural network techniques

Abstract

Accurate prediction of suspended sediment (SS) concentration is a difficult task for water resource projects. In recent years, methodologies such as artificial intelligence (AI) algorithms have been applied for sediment load estimation and these models have provided efficient results. The present study investigates the abilities of four distinct AI approaches for estimating monthly SS load in Roodak station on Jajrood River, one of the longest waterways in the north of Iran, using the combinations of the present and antecedent monthly river flow data. This study aims to compare the predictive ability of artificial neural network (ANN), adaptive neuro-fuzzy inference systems (ANFIS), group method of data handling (GMDH), and least square support vector machines (LS-SVM) applied to predict the SS load. To develop the models, the monthly average river flow and the SS data for 50 years were obtained from Tehran regional water authority. Data were separated into three subsets (training, validation, and testing) and the SS concentration was predicted where the reliability of utilized approaches was assessed by statistical criterion including the correlation coefficient (R), mean absolute error (MAE), and root mean square error (RMSE). A comparison of the developed models revealed that the use of antecedent average river flow is able to enhance the prediction precision of suspended sediment concentration. The results indicate that the LS-SVM model generated superior results than the other models in terms of the mean error criteria, showing the ability of the model to reasonably predict the observed SS load values.