The application of energy balance models for estimation of evapotranspiration ($ET$) still has challenges to be addressed for large scale applications. The algorithm for automated calibration using inverse modeling at extreme conditions (CIMEC) is based on the definition of endmembers that represent the extreme conditions of the $ET$ spectrum, between hot (dry and sparse vegetation) and cold (wet and dense vegetation) surfaces, with pre-defined quantiles for the endmember selection. The main goal was to assess geeSEBAL algorithm uncertainties related to the (i) automated calibration, including the use of additional filters (land cover, homogeneity, and domain area) and (ii) the use of a global climate grid as input data. Based on a sensitivity analysis, we defined new set of quantiles to increase the accuracy of $ET$ estimates in subtropical humid climates, since the default quantiles were adjusted to semiarid climates with dry summers. To validate our $ET$ estimates we used eddy covariance measurements from five flux towers located in the South of Brazil. Processing 132 Landsat cloud free images and using adjusted quantiles, we found a root mean square error (RMSE) of 0.91 mm d ^− 1 and a coefficient of determination (R²) of 0.82 with geeSEBAL driven by meteorological measurements. Using the pre-defined quantiles, we found an RMSE of 1.16 mm d ^− 1 (27% higher) and R² of 0.75. The upscaling instantaneous $ET$ to daily $ET$ resulted in an underestimation of the daily $ET$ using the pre-defined quantiles, while the optimized quantiles corrected the daily estimates. Furthermore, our results suggested a low sensitivity of geeSEBAL to meteorological inputs, since RMSE slightly increased to 1.04 mm d ^− 1 (14.3% higher) and R² decreased to 0.76 (8.5% smaller) when driven by global climate data. For data scarce areas, geeSEBAL is a feasible alternative for cropland $ET$ estimation and water resources management in subtropical humid climates.

能量平衡模型在估算蒸散量中的应用（$乙吨$) 对于大规模应用仍然有挑战需要解决。在极端条件下使用逆向建模 (CIMEC) 的自动校准算法基于代表极端条件的端元定义$乙吨$光谱，在热（干燥和稀疏植被）和冷（潮湿和密集植被）表面之间，具有用于端元选择的预定义分位数。主要目标是评估与 ( i ) 自动校准相关的 geeSEBAL 算法不确定性，包括使用附加过滤器（土地覆盖、同质性和域面积）和 ( ii ) 使用全球气候网格作为输入数据。基于敏感性分析，我们定义了一组新的分位数以提高$乙吨$亚热带湿润气候中的估计值，因为默认分位数已调整为夏季干燥的半干旱气候。验证我们的$乙吨$估计我们使用了来自巴西南部的五个通量塔的涡流协方差测量值。处理 132 张 Landsat 无云图像并使用调整后的分位数，我们发现均方根误差 (RMSE) 为 0.91 mm d  ^{−  1}，使用由气象测量驱动的 geeSEBAL 的决定系数 (R²) 为 0.82。使用预定义的分位数，我们发现 RMSE 为 1.16 mm d  ^{−  1}（高 27%），R² 为 0.75。瞬间升级$乙吨$ 每天 $乙吨$ 导致低估了每日 $乙吨$使用预定义的分位数，而优化的分位数更正了每日估计值。此外，我们的结果表明，geeSEBAL 对气象输入的敏感性较低，因为在全球气候数据的驱动下，RMSE 略微增加至 1.04 mm d  ^{−  1}（高 14.3%），而 R² 降低至 0.76（小 8.5%）。对于数据稀缺的地区，geeSEBAL 是一种可行的农田替代方案$乙吨$ 亚热带湿润气候下的估算和水资源管理。