Summary
With the upcoming global warming forests are under threat. To forecast climate change impacts and adaptations, there is need for developing improved forest monitoring services, which are able to record, quantify and map bio-indicators of the forests’ health status across the globe. In this context, Earth observation (EO) can provide a substantial amount of up-to-date information about the biochemical and structural conditions of our forests at a local-to-global scale. Among the optical EO instruments in space, one of the most innovative instruments is the experimental Compact High Resolution Imaging Spectrometer (CHRIS) on board the PROBA-1 (Project for On Board Autonomy) satellite. CHRIS is capable of sampling reflected radiation at five viewing angles over the visible and near-infrared (VNIR) region of the solar spectrum with a relatively high spatial resolution (~17 m). The as such acquired spectrodirectional (combined multi-angular and spectroscopy) data may lead to new opportunities for space-based forest monitoring applications, yet the added value of canopy reflectance anisotropy measured over the whole VNIR spectral region is largely unknown. This is why the use of space-borne spectrodirectional data of a forested target has been investigated in this thesis.
An Alpine old-growth forest was chosen as study site because of its large heterogeneity in structure and composition. This heterogeneity in structure can be characterized by variation in canopy cover (CC) and leaf area index (LAI). The heterogeneity in composition can be characterized by variation in non-photosynthetic vegetation (NPV: e.g. dead standing trees and coarse woody debris) and photosynthetic vegetation (PV: e.g. foliage). Such a large heterogeneity exerts influence on reflectance anisotropy in the VNIR and therefore challenges the interpretation of spectrodirectional data. While reflectance anisotropy has traditionally been considered as a source of noise, in turn, when having an improved understanding of its functioning, it may actually become usable for mapping the canopy heterogeneity. The main objective of this thesis is therefore to analyze the link between canopy variables and spectrodirectional data with the purpose of (i) evaluating the use of spectrodirectional data, and (ii) developing a mapping application that enables to monitor forest heterogeneity at the subpixel scale.
Chapter 1 outlines the radiative transfer processes and models that are of importance for mapping quantitative forest properties, links the models in relation to forest growth development, sketches the potentials of multi-angular imaging spectrometry for forest monitoring, and after that, lists the research objectives of this PhD thesis.
Chapter 2 addresses the phenomena of canopy reflectance anisotropy in the VNIR by means of vegetation indices. For a set of broadband and narrowband vegetation indices the angular reflectance anisotropy of an old-growth forest and an Alpine meadow as measured by CHRIS was statistically evaluated. Not only the conventional broadband greenness indices but also narrowband greenness indices as well as light use efficiency and leaf pigment indices (e.g. Photochemical Reflectance Index: PRI) were subject to canopy anisotropy effects. The forest produced more pronounced reflectance anisotropy than the meadow due to its heterogeneous canopy structure. A model-based quantification of the underlying forest canopy variables showed that angularity of PRI can be partly explained by the angular-dependent variation in observed PV and NPV proportions.
Chapter 3 builds further on the findings of chapter 2 by assessing the influence of structural and compositional variables on the performance of vegetation indices. Therefore, a sensitivity study was employed in which the detectability of chlorophyll (Cab) content at the canopy level was analyzed on the basis of modeled reflectance data. Statistical analysis revealed that most of the chlorophyll indices outperform single wavelengths in assessing Cab content, with best results obtained by the Maccioni index ([R780 – R710] / [R780 – R680]). The Maccioni index was highly sensitive to variations in Cab content but relatively insensitive to variations in LAI, CC and NPV. The modeling results provided a theoretical framework for evaluating how reliable Cab content can be assessed under various canopy conditions. This evaluation was applied for three distinct coniferous forest types (young, early mature and old-growth stands). It is concluded that the presence of woody elements considerably perturb the relationships between a vegetation index and foliar biochemistry.
Having identified some mechanisms that govern reflectance anisotropy in the VNIR, chapter 4 investigated the mappable information content that can be exploited from space-borne measured reflectance anisotropy. More precisely, the anisotropic reflectance of an Alpine old-growth forest was quantified and analyzed across the VNIR using the so-called Minnaert-k parameter. This parameter describes the curvature of reflectance anisotropy and is obtained through inversion of the parametric Rahman–Pinty–Verstraete (RPV) model. Results indicated that for a forested ecosystem with a bright underlying snow cover a switch from bell-shaped (k > 1) to bowl-shaped (k < 1) reflectance anisotropy patterns takes place in the red edge and early NIR part of the spectrum. It was found that CC plays an important role in determining at which wavelength this switch takes place. The strong spectral dependency of reflectance anisotropy dynamics, with in particular the usable information content in the red edge, encourages the use of multi-angular spectrometers for forestry applications.
Chapter 5 implemented the newly acquired knowledge of reflectance anisotropy dynamics into a mapping application. The uniqueness of CHRIS was exploited to the fullest: information on canopy properties was independently derived from both the angular and spectral domains. In the angular domain a map of the Minnaert-k parameter was generated while in the spectral domain a CC map was generated. It was evaluated that both maps are complementary in the red edge (722 nm). Both maps provided information on canopy structure but also have there single-source limitations. Merging Minnaert-k with CC produced a unique new data layer that provides information on the horizontal and vertical heterogeneity of the forest canopy at the sensor subpixel scale.
Chapter 6 contains the final conclusions and gives recommendations for further research. The overall conclusion is that space-borne spectrodirectional data are able to simultaneously derive information on forest foliar biochemistry (from the spectral domain) and on forest cover heterogeneity (from the angular domain). This creates new opportunities to monitor heterogeneous ecosystems such as forests, woodlands and shrublands at a local-to-regional scale. The results presented in this thesis should therefore encourage further research in this field as a means to develop future spectrodirectional EO instruments and to apply derived mapping products into forest monitoring schemes across the globe.