Prediction of reference evapotranspiration using NASA-POWER: Testing with weather stations in southeastern Córdoba

Abstract

Computation of reference evapotranspiration (ETo) is useful for different applications. However, its estimation is limited by the lack of terrestrial observation instruments and networks. Currently, gridded satellite reanalysis products and machine learning models are available as an alternative to estimate spatially explicit meteorological variables at local and regional scales. The aim of this work was to evaluate and predict ETo using the NASA-POWER product in southeastern Córdoba. Through the Penman-Monteith model, ETo was estimated from temperature, humidity, wind speed and solar radiation recorded at stations, those estimated by NASA-POWER and those predicted by the Extreme Gradient Boosting machine learning model (XGBoost). NASA-POWER estimated ETo with errors less than 1 mm.day-1 when compared to that observed at the stations. XGBoost obtained a better accuracy and lower estimation errors. From the NASA-POWER product
and XGBoost it is possible to reconstruct the lack of meteorological records in southeastern Córdoba to estimate ETo accurately.