The estimation and mapping of actual evapotranspiration (ETa) is an active area of applied research in the fields of agriculture and water resources. Thermal remote sensing-based methods, using coarse resolution satellites, have been successful at estimating ETa over the conterminous United States (CONUS) and other regions of the world. In this study, we present CONUS-wide ETa from Landsat thermal imagery-using the Operational Simplified Surface Energy Balance (SSEBop) model in the Google Earth Engine (GEE) cloud computing platform. Over 150,000 Landsat satellite images were used to produce 10 years of annual ETa (2010–2019) at unprecedented scale. The accuracy assessment of the SSEBop results included point-based evaluation using monthly Eddy Covariance (EC) data from 25 AmeriFlux stations as well as basin-scale comparison with annual Water Balance ETa (WBET) for more than 1000 sub-basins. Evaluations using EC data showed generally mixed performance with weaker (R2 < 0.6) correlation on sparsely vegetated surfaces such as grasslands or woody savanna and stronger correlation (R2 > 0.7) over well-vegetated surfaces such as croplands and forests, but location-specific conditions rather than cover type were attributed to the variability in accuracy. Croplands performed best with R2 of 0.82, root mean square error of 29 mm/month, and average bias of 12%. The WBET evaluation indicated that the SSEBop model is strong in explaining the spatial variability (up to R2 > 0.90) of ETa across large basins, but it also identified broad hydro-climatic regions where the SSEBop ETa showed directional biases, requiring region-specific model parameter improvement and/or bias correction with an overall 7% bias nationwide. Annual ETa anomalies over the 10-year period captured widely reported drought-affected regions, for the most part, in different parts of the CONUS, indicating their potential applications for mapping regional- and field-scale drought and fire effects. Due to the coverage of the Landsat Path/Row system, the availability of cloud-free image pixels ranged from less than 12 (mountainous cloud-prone regions and U.S. Northeast) to more than 60 (U.S. Southwest) per year. However, this study reinforces a promising application of Landsat satellite data with cloud-computing for quick and efficient mapping of ETa for agricultural and water resources assessments at the field scale.
Bibliographical noteFunding Information:
This work was performed under the U.S. Geological Survey (USGS) contract 140G0119C0001 in support of the USGS Land Change Science projects such as WaterSMART and Landsat Water Balance and USGS Landsat Science Team contract 140G0118C0007. We gratefully acknowledge the institutions and individuals who made various geospatial data freely available: Landsat (USGS Earth Resources Observation and Science (EROS) Center); Daymet air temperature (Oak Ridge National Laboratory through Dr. Peter Thornton); GridMET reference evapotranspiration (University of Idaho through Dr. John Abatzoglou); Climate Engine, Inc. We thank Lei Ji, Janet Carter, three anonymous journal reviewers, and associate editor for their constructive review feedback. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
- Cloud computing
- Eddy covariance
- Google Earth Engine
- Landsat thermal
- Water balance