Multi-station artificial intelligence based ensemble modeling of reference evapotranspiration using pan evaporation measurements

In this study, different Artificial Intelligence (AI) techniques including Feed Forward Neural Network (FFNN), Adaptive Neuro Fuzzy Inference System (ANFIS), Support Vector Regression (SVR), empirical models including Hargreaves and Samani (HS), Modified Hargreaves and Samani (MHS), Makkink (MK), Ritchie (RT) and conventional Multilinear Regression (MLR), were employed to model Reference Evapotranspiration (ET0) in fourteen stations from several climatic regions in Turkey, Cyprus, Iraq, Iran and Libya. For this purpose, 12 parameters of monthly climate data were collected and used as input parameters to the models. The parameters were subjected to quality assurance tests to ensure their validity and acceptability. The study was conducted in three sections: (i) Sensitivity analysis was conducted to determine the dominant inputs. (ii) Single models were trained and their performances were accessed on the basis of ET0 derived from pan evaporation method. (iii) Finally, three ensemble methods of simple averaging, weighted averaging and neural ensemble were applied in strategy 1 (for AI models) and strategy 2 (for empirical models) to improve the predicting performance. The results revealed that depending on the climate of the regions, pan evaporation, solar radiation, temperatures and surface pressure are the most dominant parameters, empirical and MLR models could be employed to achieve the valuable results, AI based models are superior in performance to the other models, also promising improvement in ET0 modeling could be achieved by ensemble modeling. The results of this study affirmed that the provided ensemble approaches can increase the performance of single models in the verification phase up to 22%, for strategy 1 and 55%, for strategy 2. Also, the results demonstrated that AI based ensemble modeling is preferable to empirical ensemble modeling.