Fine-scale leaf chlorophyll distribution across a deciduous forest through two-step model inversion from Sentinel-2 data
Fine-scale leaf chlorophyll distribution across a deciduous forest through two-step model inversion from Sentinel-2 data
复制标题
DOI：
10.1016/j.rse.2021.112618
复制
发表时间：
2021-08-05
影响因子：
13.5
通讯作者：
Liu, Jane
中科院分区：
文献类型：
--
作者：
Li, Yingjie;Ma, Qingmiao;Liu, Jane
叶子叶绿素含量（LCC）是一个关键的生理性状，对于监测植物健康和准确模拟陆地碳循环至关重要。然而，关于 LCC 变异性在精细时间步长和跨区域空间范围的精细空间分辨率下的空间连续信息很少。在这项研究中，我们改进了基于物理的两步反演方法，使用先进的冠层到叶片反射率转换模型，根据 Sentinel-2 多光谱仪器 (MSI) 数据估计精细空间分辨率 (20 m) 的 LCC。第一步是使用由几何光学模型（4 尺度）构建的查找表将 MSI 冠层反射率转换为叶片反射率。第二步是使用 PROSPECT-5 叶片光学模型根据建模的叶片反射率估算 LCC。 MSI 得出的叶子反射率和 LCC 都根据加拿大温带混合森林地点的现场测量进行了验证，以检查叶面积指数 (LAI) 和 LCC 反演的准确性。结果表明，冠层水平反演稳健，测量的叶片反射率与 MSI 衍生的叶片反射率之间存在密切关系（R-2 = 0.995，p < 0.001，RMSE = 0.0143）。与测量的 LCC 样品相比，建模的 LCC 结果也很强劲：分别为 R-2 = 0.849、p < 0.001 和 RMSE = 0.304 mu g/cm(2)。用于 LAI 和 LCC 推导的最重要的 Sentinel-2 MSI 波段集中在 705 nm（波段 5）。重要的是，这种两步辐射传输反演方法大大改进了Sentinel-2应用平台当前采用的LAI和LCC算法，后者低估了LAI 52.93%，高估了LCC 44.45%。这项工作强调了基于物理的两步反演方法在非常精细的空间和时间分辨率下从 Sentinel-2 导出叶子和冠层特征的潜力，可用于广泛的陆地生态应用。
Leaf chlorophyll content (LCC) is a key physiological trait and is crucial for monitoring plant health and accurately modeling the terrestrial carbon cycle. However, spatially-continuous information on LCC variability at fine time-steps, and at fine spatial resolutions across regional spatial extents, is sparse. In this study, we improved a physically-based, two-step inversion approach by using an advanced canopy-to-leaf reflectance conversion model to estimate LCC at fine spatial resolution (20 m) from Sentinel-2 Multi-Spectral Instrument (MSI) data. The first step is to convert MSI canopy reflectance to leaf reflectance using look-up tables constructed from a geometric optical model (4-Scale). The second step is to estimate LCC from the modeled leaf reflectance using the PROSPECT-5 leaf optical model. Both leaf reflectance and LCC derived from MSI were validated against field measurements at a mixed temperate forest site in Canada to examine the accuracy of leaf area index (LAI) and LCC retrievals. The results demonstrate robust canopy-level inversions with strong relationships between measured and MSI-derived leaf reflectance (R-2 = 0.995, p < 0.001, RMSE = 0.0143). The modeled LCC results were also strong when compared to measured LCC samples: R-2 = 0.849, p < 0.001, and RMSE = 0.304 mu g/cm(2), respectively. The most important Sentinel-2 MSI band for LAI and LCC derivation was centered at 705 nm (Band 5). Importantly, this two-step radiative transfer inversion approach substantially improved upon the current LAI and LCC algorithms adopted by the Sentinel-2 Application Platform, which underestimated LAI by 52.93% and overestimated LCC by 44.45%. This work highlights the potential of the physically-based two-step inversion method for deriving leaf and canopy traits from Sentinel-2 at very fine spatial and temporal resolutions, for a wide range of terrestrial ecological applications.