The pronounced impact of topography on meteorological conditions has largely limited evapotranspiration (ET) remote sensing techniques to relatively flat terrains. This study addresses this limitation by adapting and assessing the performance of two common ET models based on thermal infrared data in rugged mountainous regions: a physically-based energy balance model (TSEB-PT), and a contextual model (LST-VI). The latter derives the evaporative fraction (EF), defined as ratio of the latent flux (LE) to available energy, from spatial relationships between land surface temperature (LST) and Vegetation Index (VI), by assuming uniform meteorological conditions. The LST-VI model hence requires the normalization of LST data for meteorological variability effects induced by topography prior to EF estimation, while TSEB-PT requires the spatialization of meteorological data at the thermal sensor's resolution. This study provides for the first time a quantitative assessment of methods for correcting topographical effects at thermal data resolution within a steep-sided valley, and compares them when applied to EB- and EF-based models. Both ET models are applied to 30 m resolution Landsat data across a 20 km by 44 km area in the High Atlas mountain of Morocco from 2020 to 2022. The models' results are evaluated at two eddy covariance sites with or without considering topographic effects: an agricultural foothill site, and an elevated rocky site, located at 900 and 3850 m.a.s.l., respectively. By taking into account topography, the RMSE (and % error) on simulated LE at the foothill site was reduced by 29 W/m2 (29 %) and 10 W/m2 (16 %) for TSEB_PT and LST-VI respectively. At the elevated site however, the RMSE (and % error) reduction was 50 W/m2 (50 %) and 64 W/m2 (59 %) for TSEB_PT and LST-VI respectively. Analysis of the spatial variability over the study area indicates that the EF distributions (corrected for topographical effects) between east-facing and west-facing slopes are similar for LST-VI (mean difference of 0.01) and significantly different for TSEB_PT (mean difference of 0.19). Normalizing LST for topographic effects at the thermal sensor resolution is hence an effective way of estimating ET in mountains despite the inherent uncertainties in the available meteorological data.

中文翻译：

---

从热红外数据估计山区水分受限地区的蒸散量：基于能量平衡和蒸发分数的两种方法的比较


地形对气象条件的显著影响在很大程度上将蒸散 （ET） 遥感技术限制在相对平坦的地形上。本研究通过在崎岖山区调整和评估两种基于热红外数据的常见 ET 模型的性能来解决这一限制：基于物理的能量平衡模型 （TSEB-PT） 和上下文模型 （LST-VI）。后者通过假设均匀的气象条件，从地表温度 （LST） 和植被指数 （VI） 之间的空间关系中得出蒸发分数 （EF），定义为潜通量 （LE） 与可用能量的比率。因此，LST-VI 模型需要在 EF 估计之前对地形引起的气象变率效应的 LST 数据进行归一化，而 TSEB-PT 要求以热传感器的分辨率对气象数据进行空间化。本研究首次对陡峭山谷内热数据分辨率下校正地形影响的方法进行了定量评估，并在应用于基于 EB 和 EF 的模型时对其进行了比较。从 2020 年到 2022 年，这两种 ET 模型都应用于摩洛哥高阿特拉斯山 20 公里 x 44 公里区域内的 30 米分辨率 Landsat 数据。模型的结果在两个涡流相关站点进行评估，考虑或不考虑地形影响：一个农业山麓站点和一个高耸的岩石站点，分别位于 900 和 3850 m.a.s.l.。通过考虑地形，TSEB_PT 和 LST-VI 在山麓现场模拟 LE 的 RMSE（和百分比误差）分别降低了 29 W/m2 （29 %） 和 10 W/m2 （16 %）。 然而，在升高部位，TSEB_PT 和 LST-VI 的 RMSE（和 % 误差）降低分别为 50 W/m2 （50%） 和 64 W/m2 （59%）。对研究区域空间变异性的分析表明，LST-VI 的东向和西向斜坡之间的 EF 分布（针对地形效应进行了校正）相似（平均差为 0.01），TSEB_PT 的显著差异（平均差为 0.19）。因此，尽管可用气象数据存在固有的不确定性，但在热传感器分辨率下对地形效应的 LST 进行归一化是估计山区 ET 的有效方法。