Evaluation of evapotranspiration data and gridded products using robust linear estimators in Colombia

Abstract Evapotranspiration (ET) is an essential agroclimatic variable, fundamental for estimating crop water requirements and managing irrigation. However, quantifying ET has been challenging due to limited monitoring stations, leading to the development of alternative methods, including the use of datasets and gridded products derived from satellite remote sensing or reanalysis of global models. Despite its complexity, the FAO-Penman Monteith reference evapotranspiration (ET 0 ) model is widely used, but challenges arise in upscaling it over extensive areas, especially where data are insufficient or heterogeneous as in the Colombian Thornthwaite regions. Datasets from platforms such as TERRACLIMATE, NASA POWER, and DMETREF-EUMETSAT provide valuable information for estimating ET 0 at different spatial and temporal resolutions; however, there are limitations in terms of their coverage and accuracy. In this study, the ET 0 derived from these three satellite-based datasets were compared with the monthly ET 0 estimated from national legacy data from the available IDEAM weather stations (1981–2020) in Colombia, using robust linear estimators for the entire country. Results show that TERRACLIMATE ( R 2 = 0.75, RMSE = 311 mm) performs better than NASA POWER ( R 2 = 0.5, RMSE = 531 mm) and DMETREF-EUMETSAT ( R 2 = 0.20, RMSE = 1666 mm) using Ordinary Least Squares regression, particularly in certain Thornthwaite regions. However, challenges remain, especially in arid areas, where the analyzed datasets show lower accuracy. The results of this study suggest that using certain ET data and gridded products, particularly TERRACLIMATE, is an effective approach for characterizing and monitoring this important agroclimatic variable.