Evaluation of Experimental Models and Artificial Intelligence in Estimation of Reference Evapotranspiration (Case Study: Boroujerd Station)

The FAO Penman-Monteith is a baseline method to estimate reference evapotranspiration. In many cases, it is difficult to access all data, so replacing simpler models with ‎lower input data and appropriate accuracy is necessary. ‎ The purpose of this study is to investigate the capability of the experimental ‎models, gene expression programming, stepwise regression, and Bayesian network in estimating ‎reference evapotranspiration.‎ In this research, daily information of the Boroujerd synoptic station in the period of 1996 -2017 was used as model inputs. ‎Based on the correlation between input and output parameters, six input patterns were ‎determined for modeling. The results showed that the Kimberly-Penman model has the ‎best performance among the experimental models.‎ Gene expression programming with fourth pattern ‎‎and Default Model Operators (R2 = 0.98 and RMSE = 0.9), Bayesian Network with sixth pattern (R2=0.91 and RMSE = 1.01), and stepwise regression with sixth pattern have the most accurate patterns at R2 = 0.91 and RMSE = 0.9 in the ‎training stage.‎ Comparison of the performance of the three models showed that the gene expression ‎programming model was superior to the other two models with the Average Absolute Relative Error (AARE) of 0.12 and the Mean Ratio (MR) of 0.94.‎ The results showed that gene expression programming had an acceptable ability to estimate ‎reference evapotranspiration under the weather conditions of Boroujerd and could be introduced as a ‎suitable model.‎