A. Kazemi - Department of Irrigation and Reclamation Engineering, University College of Agriculture and Natural Resources, University of Tehran, ۳۱۵۸۷-۷۷۸۷۱ Karaj, Islamic Republic of Iran.
N. Ghahreman - Department of Irrigation and Reclamation Engineering, University College of Agriculture and Natural Resources, University of Tehran, ۳۱۵۸۷-۷۷۸۷۱ Karaj, Islamic Republic of Iran.
M. Ghamghami - Department of Irrigation and Reclamation Engineering, University College of Agriculture and Natural Resources, University of Tehran, ۳۱۵۸۷-۷۷۸۷۱ Karaj, Islamic Republic of Iran.
A Ghameshloo - Department of Irrigation and Reclamation Engineering, University College of Agriculture and Natural Resources, University of Tehran, ۳۱۵۸۷-۷۷۸۷۱ Karaj, Islamic Republic of Iran.

Accurate estimation of Evapotranspiration (ET), as a key component in the hydrological cycle, is essential in agricultural water management. In the current study, an approach based on the Bayesian Model Averaging (BMA) was used to combine eight ET empirical models, namely, Blaney-Criddle, Makkink, Penman, FAO-Penman-Monteith, Priestly-Taylor, Thornthwaite, Turc and Wang to improve the accuracy of ET estimations compared to individual models. The results of eight models and ۲۴۷ combinations of them (without replacement) were compared to the results of the Water Balance (WB) model as the reference of comparison. This study was performed using warm season (April-September) data of ۲۰۰۵-۲۰۱۴ from Gorganrood-Gharesoo Basin, north of Iran. The performance of the eight models and all possible combinations were evaluated based on four statistical metrics i.e. Root Mean Square Error (RMSE), Kling-Gupta (KGE), Coefficient of Determination (R۲), and Bias. Then, the best-performing combination, (BMA-Best), was determined. Based on the WB method, the BMA-Best combination had better performance than the single models according to most of the metrics. In a few cases in which individual models showed slightly better performance than BMA-Best combination, the differences were not statistically significant. On average, the BMA-Best combination increased the R۲ by more than ۵۰% and decreased RMSE by more than ۷۰%. According to results of the current study, BMA provides a more reliable estimation of ET and it is recommended for use rather than the individual models. Moreover, the BMA-best combination mostly consisted of energy-based ET models, suggesting that these models have a better performance in climatic conditions of the study area.

Bias, EM algorithm, Statistical analysis, Water balance.