Performance Evaluation of Artificial Intelligence and Heuristic Regression Methods for Rainfall-Runoff Modelling: An Application in Aksu Stream

Abstract

In this study, taking into account the Aksu Stream data, daily total precipitation (P) and daily mean flow (Q) values were using time lagged, 8 different Rainfall-Runoff models were created and runoff value estimated for the future. The Rainfall-Runoff models have been tried with different methods and this methods performances compared for Rainfall-Runoff process. Artificial Intelligence (AI) methods, Artificial Neural Networks (ANN), Adaptive Neuro Fuzzy Inference System (ANFIS) and Heuristic Regression (HR) methods, Multivariate Adaptive Regression Splines (MARS) and Support Vector Machine (SVM) were used for describing the Rainfall-Runoff modelling. The performance of the methods is determined and compared with the Root Mean Square Error (RMSE), Correlation Coefficient (R) and Mean Absolute Error (MAE) coefficients. Although AI methods performance was very close, the lowest error value was obtained in the Rainfall-Runoff model created with the ANFIS method (RMSE=3.682, R=0.934, MAE=1.103). In the HR methods, the highest performance was observed on the Rainfall-Runoff model created with MARS (RMSE=3,101, R=0,952, MAE=1,302). In the performance evaluation, it was seen that HR methods have higher performance than AI methods for modelling Rainfall-Runoff process.