Comparative assessment of modelled and empirical reference evapotranspiration methods for a brazilian savanna

Abstract

The Brazilian savanna, locally known as Cerrado, native vegetation has been replaced by an agro-pastoral land cover over the last few decades, which has affected its hydrological cycle and the availability of its water resources. Reference evapotranspiration (ETo) estimates are an important tool for water resources managements because they estimate the potential for atmospheric water loss. The Penman-Monteith method (FAO-PM) is recommended by FAO due to its consistent results in different regions and climates, but it requires a high data demand. Several empirical ETo methods with less inputs have been tested around the world, and machine learning approaches have been studied in the recent years. Therefore, in the present paper, 21 empirical methods, classified as mass transfer-based, radiation-based, temperature-based, or combination methods, and 4 equations generated by genetic algorithm (GA) were evaluated against the FAO-PM method. Radiation-based methods were more accurate than the others, especially De Bruin-Keijman and Priestley-Taylor methods, which were the two models with best performance. Mass transfer-based and temperature-based models were deemed unsuitable due to their high errors and low correlation when compared to the other ETo methods. Among the combination methods, the Copais method showed the fifth best result from all the empirical equations tested. The GA equations were generated based on our results, considering the mass transfer-based and temperature-based models, however only one equation, using air temperature and solar radiation as inputs, presented a performance as good as the best methods found in our study. Therefore, alternatives empirical equations and GA approach could meet ETo estimates similar to the FAO-PM method using less inputs.