Efficient estimation of Reference Evapotranspiration (ET0) becomes necessary for water resources management and irrigation practices. Despite research advancement in the recent decades, results inconsistencies have been reported related to chaotic, stochastic and black box approaches for multi-step ahead prediction of ET0. This study aimed at applying ensemble approaches to improve single and multi-step ahead predictions of ET0. To do so, several Artificial Intelligence (AI) based techniques including Support Vector Regression (SVR), Adaptive Neuro Fuzzy Inference System (ANFIS), Artificial Neural Network (ANN) and Multiple Linear Regression (MLR) models, were employed for one, two and three-steps ahead predictions of ET0 for numerous climatic stations in Iraq, North Cyprus and Turkey. Monthly meteorological parameters were used as inputs for the models development. Finally, two linear ensemble methods (simple averaging and weighted averaging) and a nonlinear ensemble method (neural ensemble) were applied for the single models performance and reliability improvements. The results showed superior performance of AI based models with regard to the MLR model. In addition, promising improvements were achieved in single and multi-step ahead ET0 modeling by the application of the ensemble models. The overall results revealed that the ensemble models proposed in this study could lead to increase in single models performance in the validation phase up to 60%, 33% and 24% for Turkey, North Cyprus and Iraq stations, respectively.