Monitoring surface albedo at a fine spatial resolution in forests can enrich process understanding and benefit ecosystem modeling and climate-oriented forest management. Direct estimation of surface albedo using 10 m reflectance imagery from Sentinel-2 is a promising research avenue to this extent, although questions remain regarding the representativeness of the underlying model of surface reflectance anisotropy originating from coarser-resolution imagery (e.g., MODIS). Here, using Fennoscandia (Norway, Sweden, Finland) as a case region, we test the hypothesis that systematic stratification of the forested landscape into similar species compositions and physical structures prior to the step of carrying out angular bin regressions can lead to improved albedo estimation accuracy of direct estimation algorithms. We find that such stratification does not lead to statistically meaningful improvement over stratification based on conventional land cover classification, suggesting that factors other than forest structure (e.g., soils, understory vegetation) may be equally important in explaining within-forest variations in surface reflectance anisotropy. Nevertheless, for Sentinel-2-based direct estimation based on conventional forest classification, we document total-sky surface albedo errors (RMSE) during snow-free and snow-covered conditions of 0.015 (15 %) and 0.037 (21 %), respectively, which align with those of the coarser spatial resolution products in current operation.

中文翻译：

---

成分和结构分层并不能改善 Fennoscandian 森林中 Sentinel-2 衍生的表面反照率的直接估计


以精细的空间分辨率监测森林中的地表反照率可以丰富过程理解，并有利于生态系统建模和以气候为导向的森林管理。在这种程度上，使用 Sentinel-2 的 10 m 反射率图像直接估计表面反照率是一个很有前途的研究途径，尽管关于源自较粗分辨率图像（例如 MODIS）的表面反射各向异性基础模型的代表性仍然存在疑问。在这里，以 Fennoscandia（挪威、瑞典、芬兰）作为案例区域，我们检验了以下假设：在执行角度 bin 回归的步骤之前，将森林景观系统地分层为相似的物种组成和物理结构可以提高直接估计算法的反照率估计准确性。我们发现，与基于传统土地覆盖分类的分层相比，这种分层并不会导致具有统计学意义的改进，这表明森林结构以外的因素（例如，土壤、林下植被）在解释表面反射各向异性中的森林内变化可能同样重要。尽管如此，对于基于常规森林分类的基于 Sentinel-2 的直接估计，我们记录了无雪和有雪条件下的总天空表面反照率误差 （RMSE） 分别为 0.015 （15 %） 和 0.037 （21 %），这与当前操作中较粗糙的空间分辨率产品一致。