Estimation of actual evapotranspiration using NASA-POWER data and Support Vector Machine

Abstract

An important issue for agricultural planning is to estimate evapotranspiration accurately due to its fundamental role in the sustainable use of water resources. In this sense, it is essential to have reliable and precise evapotranspiration measurements to improve models or products, mainly related to predicting droughts. The main objective of the present study is to evaluate the Support Vector Machine Regression’s (SVR) potential to estimate the actual evapotranspiration (ETa) through a NASA-Power dataset in the Pampean Region of Argentina. The results obtained were compared with ETa values (water balance), based on information from 12 agro-meteorological stations (1983 – 2012). After training and validating the SVR algorithm, we observed statistical mean errors of 0.39 ± 0.07 mm/d, 0.54 ± 0.09 mm/d, and 0.67 ± 0.07 for the MAE, RMSE, and R2, respectively. The results show the feasibility of applying machine learning algorithms for obtaining ETa values in agricultural plains without agro-meteorological data.