本文为瑞士苏黎世大学的毕业论文，共136页。

草原，无论是人造还是天然的，几乎覆盖了地球陆地表面的20%，代表着独特的颜色、结构和生物多样性。这些自然和半自然栖息地对生物多样性保护具有重要的生态意义，因为它们为动植物提供了广泛的栖息环境，其中许多被列为濒危或受威胁的物种。此外，草原在全球碳循环中发挥着重要作用，至少占全球碳资源库的10%。据估计，它们占陆地初级净生产力（NPP）的20%。面对全球气候变化和对农业生产力日益增长的需求，草原生态系统的压力预计将增加。因此，制定管理战略确保保护生物多样性，提高对草原碳循环及其空间代表性的认识，特别是在当前和未来的气候条件下至关重要。

本论文评估了高光谱遥感和生态系统过程模拟在草地生态环境生物多样性保护中的协同应用，以及在局部到区域尺度上提高对生物地球化学循环的理解。本文工作的主要目标是探索高光谱遥感在绘制物种丰富的草原生境图、获得与生态系统生产力相关植被特性方面的潜力。

本文主要由三个部分组成。第一部分讨论了利用野外光谱辐射计收集的不同草地生境的年内和年际光谱测量来探索其季节光谱可分性。在第二部分中，本论文重点研究了根据野外光谱辐射计记录的光谱数据预测草地地面上生物量的统计模型，并评估了将这些模型扩展到星载高光谱数据（Hyperion）的可能性。最后，第三部分探讨了利用新的统计学方法从航空高光谱遥感（HyMap）中提取草地叶片生物化学浓度（nitrogen）的定量信息。随后，利用HyMap提取的叶片生物化学信息，以空间分布的方式初始化和驱动生态系统过程模型（Biome-BGC），从而估算草地生态系统的NPP。

本文第一部分的研究结果表明，利用高光谱遥感可以成功地识别出物种丰富的草原生境。确定了在生长季节对草原类型可分离性贡献最大的特定光谱部分。更重要的是，生长季的开始是从高光谱遥感测量中识别草原生境的最佳时期。论文第二部分的研究结果表明，在生长季采集多个样本的情况下，建立高光谱遥感草地生物量估算的稳健统计模型是可行的。由于物候期、空间格局和管理的变化，季节性采样（特别是在生长季早期）对于覆盖正常发生的变异性非常重要。此外，还得出结论，当使用窄带NDVI类型指数时，可以最好地实现田间生物量估算统计模型的放大。最后，本文第三部分的结果表明，在校准统计模型之前，只要对HyMap反射光谱进行连续去除变换，就可以实现对叶片生化信息的准确预测。研究还表明，以空间分布模式运行Biome-BGC模型是可能的，从而得出研究区域的详细NPP估计。然而，更重要的发现是，利用C:N空间预测的研究区域NPP显著高于利用文献中广泛应用的C:N值或利用区域测量的平均C:N值估算的NPP。这些发现表明，如果使用标准方法，不将空间分布的C:N值输入模型运行，则某些生态系统的碳固存动态（在我们管理的草地生境的情况下）在区域上可能会被低估。

总的来说，本论文的工作显示了高光谱遥感在物种丰富的草原生态系统的光谱可分性和生物量及叶片氮浓度反演方面的巨大潜力。此外，它强调了使用高光谱遥感的高精度产品作为生态系统过程模型输入的重要性。将这两种协同技术结合起来，有望提高我们对区域尺度上陆地碳循环动力学的理解。

Grasslands, both man made and natural, cover nearly 20% of the Earth’s land surface representing a unique variety of colours, structure and biodiversity. These natural and semi natural habitats are of great ecological importance for biodiversity conservation, since they provide a wide range of habitats for plants and animals, many of which are classified as endangered or threatened. Furthermore, grasslands play a significant role in the global carbon cycle, accounting for at least 10% of the global carbon pools. Additionally, it is estimated that they are responsible for as much as 20% of the total terrestrial net primary production (NPP). In the face of global climate change and the growing demand for agricultural productivity, pressure on grassland ecosystems is expected to increase. Therefore, the development of management strategies that ensure the protection of biodiversity and an improved knowledge of the grasslands carbon cycle and its spatial representation are of critical importance, specifically also under current and future climatic conditions. This dissertation evaluates the synergistic use of hyperspectral remote sensing and ecosystem process modelling for biodiversity conservation of grassland habitats and for improved understanding of their biogeochemical cycles at local to regional scales. The major goals of the presented work are to explore the potential of hyperspectral remote sensing for mapping species-rich grassland habitats and for deriving vegetation properties relevant to ecosystem productivity. The dissertation consists of three main research parts. The first part addresses the use of intra and inter-annual spectral measurements of different grassland habitats collected with a field spectroradiometer for exploring their seasonal spectral separability. In the second part, the dissertation focuses on the development of statistical models for predicting above-ground biomass of grasslands from spectral measurements recorded with a field spectroradiometer and evaluates the potential of up-scaling these models to spaceborne hyperspectral data (Hyperion). Finally, the third part explores the extraction of quantitative information on grassland foliar biochemical concentrations (nitrogen) from airborne hyperspectral remote sensing (HyMap) using novel statistics. Subsequently, the HyMap extracted foliar biochemical information is used to initialise and drive an ecosystem process model (Biome-BGC) in a spatially-distributed manner to estimate NPP of the grassland ecosystems. Results from the first part of this dissertation demonstrated that species-rich grassland habitats could be discriminated successfully by using hyperspectral remote sensing. Specific parts of the spectrum that contributed best to the separability of the grasslands types during the growing seasons were identified. More importantly, it was shown that the beginning of the growing season was the best period for discriminating grassland habitats from hyperspectral remote sensing measurements. Results from the second part of the dissertation showed that construction of robust statistical models for grassland biomass estimation from hyperspectral remote sensing is feasible, provided that multiple samples during the growing season were collected. Seasonal sampling, especially early in the growing season, was shown to be very important in order to cover normally occurring variability, due to variations in phenology stage, spatial patterns and management. Furthermore, it was concluded that up-scaling of field developed statistical models for biomass estimation could best be achieved when narrow band NDVI type indices were used. Finally, results from the third part of this dissertation revealed that accurate predictions of foliar biochemical information could be achieved, provided that continuum-removal transformation was applied to the HyMap reflectance spectra prior to calibrating the statistical models. It was also shown that it was possible to run the Biome BGC model in a spatially distributed mode, thus deriving detailed NPP estimates of the study area. More important findings, however, were that NPP of the study area using spatial predictions of C:N was significantly higher than NPP estimated using C:N values widely applied in literature or even from using regionally measured mean C:N values. These findings suggested that carbon sequestration dynamics of certain ecosystems − in our case of managed grassland habitats − might be underestimated regionally if standard methods are used that do not feed spatially distributed C:N values into model runs. Overall, the work presented in this dissertation has demonstrated the high potential of hyperspectral remote sensing for the spectral separability and the retrieval of biomass and foliar nitrogen concentration of species-rich grassland ecosystems. Furthermore, it emphasised the importance of using high accuracy products derived from hyperspectral remote sensing as input to ecosystem process models. Coupling these two synergistic technologies is expected to enhance our understanding on terrestrial carbon cycle dynamics at local to regional scales.

1.       引言

2. 相关材料

3. 研究方法

4. 利用多时相光谱辐射计数据研究草原沿干旱梯度的光谱可分性

5. 星载高光谱遥感在草地生境地面上生物量景观尺度上的应用

6. 耦合成像光谱与生态系统过程建模——空间分布叶片生物化学浓度估计对草地生境NPP建模的重要性

7. 结论与展望

更多精彩文章请关注公众号：