GEP and MLR approaches for the prediction of reference evapotranspiration

In this study, reference evapotranspiration (ET_o) is modeled as one of the major items of hydrological applications from different combinations of climatic variables using two different techniques: gene expression programming (GEP) and multiple linear regression (MLR). The data used in modeling were collected from weather stations in Egypt through the CLIMWAT database. The Penman–Monteith FAO-56 equation was considered as a reference target for ET_o values depending on the entire climatic variables. The developed ET_o models’ performances were compared and evaluated with regard to their predictive abilities using statistical criteria to identify the superiority of one modeling approach over the others and determine climatic variables which have a significant effect on ET_o. The results indicated that GEP and MLR models’ contribution toward mean relative humidity and wind speed at 2 m height is greater compared to that of other variables. Meanwhile, when adding temperature data to models, solar radiation has a slight effect on increasing the accuracy of ET_o estimate. Moreover, the lower statistical error criteria values of GEP models confirmed their better performance than MLR models and other empirical equations.