Quantitative remote sensing of vegetation functional traits and disturbance dynamics in Australian eucalypt forest. A multiscale analysis from leaf to landscape

Quantitative remote sensing estimation of vegetation functional traits is essential to assess and monitor forest ecological functions and the dynamic response of vegetation to changing environmental conditions, such as more frequent drought events. Vegetation biochemical traits, such as chlorophyll a+b content, carotenoid content, and anthocyanin contents, provide insights in plant physiological processes. Modern optical sensors facilitate retrieval of plant traits, contributing to better understanding of ecosystem function and biodiversity. This thesis focuses on developing quantitative approaches, based on optical imaging spectroscopy, for retrieval of Australian Eucalypt leaves’ pigment contents. Additionally, an approach was developed to assess the sensitivity of forest to water stress across topographic gradients using the Harmonized Landsat Sentinel-2 (HLS) multispectral time series analysis. The research of this thesis encompasses three ecological levels (leaf, tree canopy, and ecosystem) using field spectroscopy, drone imagery, and satellite data of complementary spatial scales.
Chapter 2 investigated the adaptation and validation of the Fluspect-Cx leaf radiative transfer model to simulate leaf optical properties and estimate leaf biochemical traits through inversion of two native Australian eucalypt species (Eucalyptus dalrympleana and E. delegatensis). Comparison of measured and simulated optical properties revealed the need to recalibrate the refractive index and in vivo specific absorption coefficients (SACs) of biochemical constituents. The recalibration enabled Fluspect-Cx to simulate leaf optical properties of the eucalypt leaves within the wavelength range of 450-1600 nm with a mean RMSE < 1% and to retrieve, through a multivariate model inversion, contents of total chlorophylls (C_ab), carotenoids (C_car), anthocyanins content (C_ant), dry matter (C_w), and leaf water (C_m) with root mean square error (RMSE) of 8.46 µg.cm⁻² , 3.83 µg.cm⁻² , 1.69 µg.cm⁻² , 0.0036 g.cm⁻² and 0.0013 cm respectively. Additionally, the study explored the potential of optical spectral indices sensitive to the retrieved biochemical traits as their alternative estimators, achieving accurate results for estimation of C_car.
In Chapter 3, the recalibrated Fluspect-Cx was coupled with the 3-dimensional Discrete Anisotropic Radiative Transfer (DART) model, to estimate C_ab, C_car, and C_ant of sunlit eucalypt crowns from hyperspectral Unoccupied Airborne Systems (UAS) (~6 cm) and Sentinel-2A (20 m) images acquired over the Terrestrial Ecosystem Research Network (TERN) supersite Tumbarumba. Using machine learning regression, the retrieved quantitative estimates gained plausible ranges, revealing rather low C_ab and C_car contents (maximum around 9 and 5 µg.cm⁻² , respectively) and relatively high Cant (maximum close to 9 µg.cm⁻² ), typical for an early growth spring phenophase. Obtained UAS maps indicated that the top parts of tree crowns contained less photosynthetic pigments than crown edges. Notable differences were found between retrievals from UAS and Sentinel-2A images , which could be attributed to the scale effects, i.e., presence of mixed over- and under-story reflectance in Sentinel-2A pixels.
Chapter 4 assessed the water stress sensitivity across topographic gradients derived from digital elevation model attributes across sites in south-east Australian forests. A newly proposed approach for a potential water stress exposure classification was coupled with temporal trends of selected vegetation spectral indices characterising plant biochemical and biophysical/morphological traits at three eucalypt dominated forest sites in south-east Australia. When analysing trends of spectral indices time series derived from HLS multispectral images, the results revealed different sensitivity of forest ecosystems to drought seasons between 2016 and 2020 across the established potential water stress classes. The findings identified a correspondence between observed forest browning and declining remotely sensed spectral indices due to drought conditions. The study demonstrated that the developed method for forest potential water stress stratification offers valuable insights for forest management decisions aimed at mitigating the impact of future drought episodes.
The research findings of this thesis contribute to the development of forestry operational tools and applications using hyperspectral and multispectral imaging. The methods and tools in this thesis enable extraction of plant functional traits characterising key Australian eucalypt species from optical remote sensing data across different spatial and functional scales. With the onset of new imaging spectroscopy spaceborne missions like the European Space Agency’s Copernicus Hyperspectral Imaging Mission for the Environment (CHIME) and National Aeronautics and Space Administration’s Surface Biology and Geology (SBG), the integration of derived remote sensing products into forest monitoring programs will allow for pre-visual assessment of forest health and stress driven by the escalating impacts of global climate change.