Uncertainty Analysis of Artificial Intelligence Models in Forecasting River Flow (Case tudy: Karun River)

Accurate forecasting of hydrological process and taking into account their uncertainties are the major challenges in the water resources management. The goal of this paper is to forecast the monthly streamflow of Karun River at Armand hydrometric station using artificial intelligence (AI) methods. Applying the Monte-Carlo simulation (MCS) approach for considering the uncertainty of the AI forecasts and compareing their performance is another objective of this paper. In this regard, some AI-based models including Gene Expression Programming (GEP), Multivariate Adaptive Regression Spline (MARS) and Model Tree (MT) have been used. A 28-year historical data of Karun river (for period of 1981-2008) were also employed. To generate random numbers of stochastic variable, Thomas-Fairing (TF) parametric method was applied. The results of evaluating the performance of these models with indices like correlation coefficient (R), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), showed that the MT model had better performance than the others in both training and testing phases. The model accuracy indices for the MT model in training phase were R= 0.841 and RMSE= 36.789 m3/ s, while these indices for the test phase were R= 0.87 and RMSE = 44.253 m3/ s. The uncertainty results of the MARS, GEP and MT models showed that the MT model with R-factor= 1.67 and 95PPU=55.5% had the best performance for calculating uncertainty.