Earth & Environmental Sciences, Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA; Department of Environmental Science, Policy and Management, UC Berkeley, Berkeley, CA 94720, USA; Environmental Change Institute, School of Geography and the Environment, University of Oxford, Oxford OX1 3QY, UK.
Environmental Change Institute, School of Geography and the Environment, University of Oxford, Oxford OX1 3QY, UK.
Sci Total Environ. 2019 May 20;666:1301-1315. doi: 10.1016/j.scitotenv.2019.01.379. Epub 2019 Feb 16.
Recent work has shown that leaf traits and spectral properties change through time and/or seasonally as leaves age. Current field and hyperspectral methods used to estimate canopy leaf traits could, therefore, be significantly biased by variation in leaf age. To explore the magnitude of this effect, we used a phenological dataset comprised of leaves of different leaf age groups -developmental, mature, senescent and mixed-age- from canopy and emergent tropical trees in southern Peru. We tested the performance of partial least squares regression models developed from these different age groups when predicting traits for leaves of different ages on both a mass and area basis. Overall, area-based models outperformed mass-based models with a striking improvement in prediction observed for area-based leaf carbon (C) estimates. We observed trait-specific age effects in all mass-based models while area-based models displayed age effects in mixed-age leaf groups for P and N. Spectral coefficients and variable importance in projection (VIPs) also reflected age effects. Both mass- and area-based models for all five leaf traits displayed age/temporal sensitivity when we tested their ability to predict the traits of leaves of other age groups. Importantly, mass-based mature models displayed the worst overall performance when predicting the traits of leaves from other age groups. These results indicate that the widely adopted approach of using fully expanded mature leaves to calibrate models that estimate remotely-sensed tree canopy traits introduces error that can bias results depending on the phenological stage of canopy leaves. To achieve temporally stable models, spectroscopic studies should consider producing area-based estimates as well as calibrating models with leaves of different age groups as they present themselves through the growing season. We discuss the implications of this for surveys of canopies with synchronised and unsynchronised leaf phenology.
最近的研究表明，随着叶片的衰老，叶片的性状和光谱特性会随时间和/或季节发生变化。因此，目前用于估算冠层叶片特性的田间和高光谱方法可能会因叶片年龄的变化而产生显著偏差。为了探究这种影响的程度，我们使用了一个包含不同叶片年龄组（发育、成熟、衰老和混合年龄）叶片的物候数据集，这些叶片来自秘鲁南部的冠层和新兴热带树木。我们测试了从这些不同年龄组开发的偏最小二乘回归模型在预测不同年龄叶片特性时的性能，这些特性是基于质量和面积的。总体而言，基于面积的模型优于基于质量的模型，基于面积的叶片碳（C）估计值的预测效果有显著提高。我们在所有基于质量的模型中都观察到了特定性状的年龄效应，而基于面积的模型在混合年龄叶片组中则显示了 P 和 N 的年龄效应。光谱系数和变量重要性投影（VIPs）也反映了年龄效应。当我们测试这些模型预测其他年龄组叶片特性的能力时，所有五个叶片特性的基于质量和基于面积的模型都表现出了年龄/时间敏感性。重要的是，当预测其他年龄组叶片的特性时，基于质量的成熟模型的整体表现最差。这些结果表明，广泛采用的使用完全展开的成熟叶片来校准模型以估算遥感树冠层特性的方法会引入误差，这可能会根据树冠叶片的物候阶段产生偏差。为了实现时间稳定的模型，光谱研究应该考虑制作基于面积的估计值，并使用不同年龄组的叶片来校准模型，因为它们会在整个生长季节中呈现出来。我们讨论了这对具有同步和不同步叶片物候的树冠调查的影响。
