Artificial Intelligence Techniques for river flow forecasting in the Seyhan River Catchment, Turkey

Abstract

The use of Artificial Intelligence methods is becoming increasingly common in the modeling and forecasting of hydrological and water resource processes. In this study, applicability of Adaptive Neuro Fuzzy Inference System (ANFIS) and Artificial Neural Network (ANN) methods, Generalized Regression Neural Networks (GRNN) and Feed 5 Forward Neural Networks (FFNN), for forecasting of daily river flow is investigated and the Seyhan catchment, located in the south of Turkey, is chosen as a case study. Totally , 5114 daily river flow data are obtained from river flow gauges station of¨Uçtepeof¨ of¨Uçtepe (1818) on Seyhan River between the years 1986 and 2000. The data set are divided into three subgroups, training, testing and verification. The training and testing data 10 set include totally 5114 daily river flow data and the number of verification data points is 731. The river flow forecasting models having various input structures are trained and tested to investigate the applicability of ANFIS and ANN methods. The results of ANFIS, GRNN and FFNN models for both training and testing are evaluated and the best fit forecasting model structure and method is determined according to criteria of 15 performance evaluation. The best fit model is also trained and tested by traditional statistical methods and the performances of all models are compared in order to get more effective evaluation. Moreover ANFIS, GRNN and FFNN models are also verified by verification data set including 731 daily river flow data at the time period 1998-2000 and the results of models are compared. The results demonstrate that ANFIS model is 20 superior to the GRNN and FFNN forecasting models, and ANFIS can be successfully applied and provide high accuracy and reliability for daily River flow forecasting.