Investigating the Uncertainty of Data-Based Models in Forecasting Monthly Flow of the Hablehroud River

Accurate and reliable forecasts of river flow are required for proper management of watershed systems. In recent years, data-driven models and especially artificial intelligent based models have been successfully used in various areas related to water resources. However, uncertainty analysis of these models has been less appreciated in prior studies. In the present study, the output uncertainty of five data-driven models including modular, PCA (Principle Component Analysis), TLRN (Time-Lagged Recurrent Network), ANFIS (Adaptive-Network-based Fuzzy Inference System) and SVM (Support Vector Machine) type models in forecasting river flow has been investigated using 95PPU, p-factor and d-factor quantities. Using the observed meteorological and flow data during 1998-2012 in Hablehroud Basin, different structures of the proposed models were trained and tested. The final values of p-factor and d-factor for each model type were obtained. The results showed that SVM with a p-factor of 82% produces the most reliable forecasts in the present study.