Comparison of artificial intelligence methods in estimation of suspended sediment load (Case Study: Sistan River)

Background and objectives: Correct estimation of suspended sediments volume in rivers is one of important issues in river engineering, water resources and environment projects. Sistan river is one of split main branch of Helmand river, which task of irrigate 70% agricultural plain and is responsible for providing part of Hamoon water in Helmand. Given the many problems caused by sediment in rivers, sediment science researchers have done many effort ‌to achieve sediment transport relations according laboratory and field studies. Because of multiplicity parameters involved in sediment transport and complexity process of erosion and transport particles, most of the sediment relationships need to solution complex mathematical equations, however, it aren’t accurate results. Also regression relations between water discharge and sediment discharge aren’t good correlation. Cobaner et al. (2008) is compared the potential of neuro-fuzzy technique with those of the three different artificial neural network technique in suspended sediment concentration estimation. The comparison results shown the neuro-fuzzy models perform better than the other models for the particular data sets (8). Aytek and Kişi (2008) develop an explicit model based on genetic programming. Their research’s results indicated that the proposed GP formulation performs quite well compared to sediment rating curves and multi linear regression models and is quite practical for use (3).
Materials and methods: The recent years using of ‌ smart systems in order to increase accuracy of estimating of river sediments are common. In this study were used the smart systems‌ including‌ Artificial Neural Networks (ANNs), Adaptive Neuro-Fuzzy Inference System (ANFIS) and ‌ Genetic Expression Programming (GEP) in order to estimation of suspended sediment load in Sistan River. Root mean square error (RMSE), mean bias error (MBE) and determination coefficient (R2) statistics are used for evaluating the accuracy of the models.
Results: All smart ways estimate suspended sediment load better than empirical relations. The third scenario of ANFIS from artificial intelligence (AI) methods with RMSE=20983.43 and R2=0.97 is the best result in estimation suspended sediment load. Also AI methods obtained at 95% absent aren’t significant difference between results and according to error rates all AI methods are highly accurate.
Conclusion: According to the obtained results in this study used three methods to estimate the suspended sediment load are suitable but Genetic Expression Programming is preferable to the other two models because of develop a mathematical model. The dramatic impact of the classification of discharge is clear in the precision of the suspended sediment load estimation. According to this research results, suggest estimation of suspended sediment load is suggested using AI methods in Sistan River.