Comparison of reference evapotranspiration estimates by several models in the region of Western São Paulo Plateau (Brazil)

Abstract

Evapotranspiration is the way in which water from the Earth’s surface passes into the atmosphere in the vapor state and plays an important role in the global hydrological cycle. Reliable and direct measurement of evapotranspiration is a high-cost activity in the implementation of techniques and equipment maintenance. This study sought to compare the estimates of reference evapotranspiration made by means of multiple regression and machine learning techniques for the region of the Western São Paulo Plateau. The results showed good performances for estimating the reference evapotranspiration through multiple regression and machine learning techniques. The two methods that presented the best performance were the multilayer perceptron method (ETo-MLP, rRMSE = 0.62%) and the adaptive neuro-fuzzy inference system (ETo-ANFIS; rRMSE = 0.75%), both machine learning techniques. Machine learning models are more convenient and comparatively faster to implement than other models, especially when climate data are limited. The results can be applied to the area of water resource management, especially to help estimate evapotranspiration for irrigation and water balancing. In addition, the results of this study can also be applied to predict crop productivity.