Performance mapping and weighting for the evapotranspiration models of the OpenET ensemble

Authors: REITZ, M - Us Geological Survey (USGS); VOLK, J - Desert Research Institute; OTT, T - Desert Research Institute; Anderson, Martha; SENAY, G - Us Geological Survey (USGS); MELTON, F - National Aeronautics And Space Administration (NASA); KILIC, A - University Of Nebraska; ALLEN, R - University Of Idaho; FISHER, J - Jet Propulsion Laboratory; RUHOFF, A - University Of Rio Grande Do Sul; PURDY, A - Jet Propulsion Laboratory; HUNTINGTON, J - Desert Research Institute

Submitted to: Water Resources Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/26/2025
Publication Date: 8/9/2025
Citation: Reitz, M., Volk, J., Ott, T., Anderson, M.C., Senay, G.B., Melton, F., Kilic, A., Allen, R.G., Fisher, J., Ruhoff, A., Purdy, A., Huntington, J. 2025. Performance mapping and weighting for the evapotranspiration models of the OpenET ensemble. Water Resources Research. 61(8). Article e2024WR038899. https://doi.org/10.1029/2024WR038899.
DOI: https://doi.org/10.1029/2024WR038899

Interpretive Summary: Evapotranspiration (ET) represents the total amount of liquid water in the water cycle that is converted to water vapor, through either uptake and transpiration by plants or evaporation from soils, wet leaves, lakes, or other surfaces. Remotely sensed ET data available from OpenET provide daily gridded ET information at the field scale, which are useful for a variety of applications, such as quantifying how much water is consumed by crops. OpenET is an ensemble of six individual models. In this study, we evaluate the performance of these models against ground-based data from a national-scale set of flux towers. We characterize each flux tower location in terms of a range of variables, such as climate and land cover descriptors. We model and map the performance of the OpenET models as a function of these variables, and find that the different OpenET models have strengths in different types of settings. The results have the potential to lead to improvements in the high-impact data set produced by OpenET, through either motivating improvements to the individual models or improving the ensemble ET value by weighting models more highly in the settings where they perform better.

Technical Abstract: Because evapotranspiration (ET) accounts for the majority of precipitation in the water cycle, improvements to the accuracy, resolution, and coverage of ET data are highly useful for informing hydrologic models and assessments. The OpenET collaboration of six remotely sensed ET modeling teams has demonstrated that an ensemble approach generally provides improved accuracy relative to individual ensemble members when averaged over time or space. The performance of individual models has been shown to vary by land cover type and climate zone, but a thorough study of the dependence of performance on setting has not yet been conducted. In this paper, we quantify the performance of the OpenET models relative to flux tower data as a function of several explanatory variables, such as land cover type and reference ET. We find that performance metrics, such as R-squared, can be modeled as a function of these explanatory variables. We apply these models to produce mapped estimates of model performance metrics across the conterminous U.S. We find that weighting models according to these performance metrics improves the ensemble accuracy relative to the current OpenET method, improving monthly MAE by 2% in agricultural settings and by greater amounts in settings where model predictions vary significantly, for example by 10% in shrublands. We produce weight maps that can be used to generate weighted ensemble estimates for OpenET data. The results will be useful for informing model selection and for providing insight on the variable controls on model performance that could lead to future model refinement.