Abstract Accurately quantifying large-scale terrestrial evapotranspiration (ET) remains hampered by poor parameterization of the physical processes that relate to ET. Previous studies suggested that the calibration-free complementary relationship (CR) method that requires only routine meteorological data performed better than main-stream atmospheric reanalyses, land surface or remote sensing models in estimating large-scale ET. Here we simultaneously evaluate the latest machine learning-based upscaling of eddy-covariance measurements (FLUXCOM) and the CR estimates against the water-balance derived ET rates of 18 large Hydrologic Unit Code-2 (HUC2) and 327 medium HUC6 basins across the conterminous United States. Overall, CR and FLUXCOM perform comparably in representing the multiyear mean and temporal variations in annual ET at both, HUC2 and HUC6, scales for the 1979–2013 period. Such equally good skills also hold true for the 2003–2015 period, during which FLUXCOM was driven solely by remote sensing data. However, the CR generally captures the long-term linear tendencies in annual ET rates somewhat better than FLUXCOM. Because of its minimal data requirement, the calibration-free version of the CR may continue to serve as a benchmarking tool for large-scale ET simulations.

中文翻译：

---

对大规模蒸散估计进行基准测试：从无校准互补关系 (CR) 方法和 FLUXCOM 的角度来看

摘要 准确量化大规模陆地蒸散 (ET) 仍然受到与 ET 相关的物理过程参数化不佳的阻碍。以往的研究表明，仅需要常规气象数据的免校准互补关系 (CR) 方法在估计大尺度 ET 方面比主流大气再分析、地表或遥感模型表现更好。在这里，我们同时评估了最新的基于机器学习的涡流协方差测量 (FLUXCOM) 的升级和 CR 估计值与 18 个大型水文单元代码 2 (HUC2) 和 327 个中型 HUC6 流域的水平衡得出的 ET 率美国。总体而言，CR 和 FLUXCOM 在表示 HUC2 和 HUC6 的年度 ET 的多年平均和时间变化方面表现相当，1979-2013 年期间的量表。这种同样出色的技能在 2003-2015 年期间也适用，在此期间 FLUXCOM 完全由遥感数据驱动。然而，CR 通常比 FLUXCOM 更好地捕捉年 ET 率的长期线性趋势。由于其最低数据要求，CR 的免校准版本可能会继续作为大规模 ET 模拟的基准工具。