Long-term monitoring of evapotranspiration using the SEBAL algorithm and Google Earth Engine cloud computing

• Published in: ISPRS journal of photogrammetry and remote sensing 2021-08, Vol.178, p.81-96
• Main Authors: Laipelt, Leonardo, Henrique Bloedow Kayser, Rafael, Santos Fleischmann, Ayan, Ruhoff, Anderson, Bastiaanssen, Wim, Erickson, Tyler A., Melton, Forrest
• Format: Article
• Language: English
• Subjects: Cloud computation; ERA5 land; geeSEBAL; Google earth engine; Landsat; Meteorological reanalysis
• ISSN: 0924-2716; 1872-8235
• DOI: 10.1016/j.isprsjprs.2021.05.018

The geeSEBAL application estimates and displays evapotranspiration maps and times series based on Landsat images and global meteorological data from ERA5 Land reanalysis. Codes and applications are available at https://github.com/et-brasil/geesebal and https://etbrasil.org/geesebal, respectively. [Display omitted] Accurate estimation of evapotranspiration (ET) is essential for several applications in water resources management. ET models using remote sensing data have flourished in recent years allowing spatial and temporal assessments at unprecedented resolutions. This study presents geeSEBAL, a new tool for automated estimation of ET, based on the Surface Energy Balance Algorithm for Land (SEBAL) and a simplified version of the CIMEC (Calibration using Inverse Modeling at Extreme Conditions) process for the endmembers selection, developed within the Google Earth Engine (GEE) environment. The tool framework is introduced, and case studies across multiple biomes in Brazil are presented by comparing daily ET estimates with eddy covariance (EC) data from 10 flux towers. Based on 224 Landsat images using ERA5 Land as meteorological inputs, daily ET estimates of geeSEBAL yielded an average root mean squared difference (RMSD) of 0.67 mm day−1 when compared to EC data corrected for the energy balance closure. Additional analyses indicate a low geeSEBAL sensitivity to meteorological inputs, yielding an average RMSD of 0.71 mm day−1 when driven by in situ meteorological measurements. On the other hand, we found a higher sensitivity of the automated CIMEC algorithm to the selection of endmembers for internal calibration. For instance, by adjusting the endmembers percentiles to tropical biomes we found an error that was 36% lower compared to the standard CIMEC percentiles. Finally, we assessed the long-term effects (1984–2020) of land cover changes on surface energy fluxes and water use in agriculture for key areas in Brazil, from deforested areas in the Amazon to irrigated crops in the Pampas and Cerrado biomes. A comparison with a land surface temperature-based (SSEBop) and a vegetation-based (MOD16) model was also performed to assess relative advantages and disadvantages. This analysis showed that geeSEBAL has a significant potential for long-term assessment of ET in data-scarce areas, due to its lower sensitivity to meteorological inputs. geeSEBAL codes are written in Python and JavaScript and are freely available on GitHub (https://github.com/et-brasil/geesebal