Estimation methods to define reference evapotranspiration: a comparative perspective

Evapotranspiration is a key variable for hydrologic, climatic, agricultural, and environmental studies. Given the non-availability of economically and technically easy to implement direct measurement methods, evapotranspiration is estimated primarily through the application of empirical and regression models, and machine learning algorithms that incorporate conventional meteorological variables. While the FAO-56 Penman-Monteith equation worldwide has been recognized as the most accurate equation to estimate the reference evapotranspiration (ETo), the number of required climatic variables makes its application questionable for regions with limited ground-based climate data. This note provides a summary of empirical and semi-empirical equations linked to its data requirement and the problems associated with these models (transferability and data quality), an overview of regression models, the potential of machine learning algorithms in regression tasks, trends of reference evapotranspiration studies, and some recommendations of the topics future research should address that would lead to a further improvement of the performance and generalization of the available models. The terminology used in this note is consistent in both the theoretical and practical field of evapotranspiration, which is often dispersed in the academic literature. The goal of this note is to provide some perspective to stimulate discussion. HIGHLIGHTS An overview of trends in ETo studies is presented.; The main limitation of FAO-56 Penman-Monteith is the large number of meteorological variables required.; There is a wide variety of empirical equations for ETo estimation.; The application of machine learning algorithms is increasing due to their high performance for ETo estimation.; Some aspects of ETo estimation methods are discussed and recommended.;