The Infrared Atmospheric Sounding Interferometer (IASI) provides atmospheric observations with high spectral resolution and its data have been shown to have a significant positive impact on global Numerical Weather Prediction (NWP) and trace gas retrievals. A fundamental component of the retrieval of atmospheric composition is the radiative transfer model used to simulate the observations. An accurate representation of the expected emission spectrum measured by the satellite is essential given that differences in the reproduced atmospheric spectra propagate through a retrieval procedure and produce an altered estimate of the atmospheric state. The importance of the assumptions within the forward model are discussed and it is established that in the simulation of spectra from satellite-borne instruments the choice of the model parameters can have a large impact upon the resulting output. These assumptions are explored in the context of the Reference Forward Model (RFM), which is further configured to optimise its output for simulating the IASI spectrum in the troposphere. In order to ascertain the consistency of different radiative transfer models, comparisons are carried out between the RFM and the Radiative Transfer model for TOVS (RTTOV) in order to quantify any discrepancies in the reproduction of IASI measurements. Good agreement is shown across the majority of the spectrum, with exceptions caused by CO₂ line mixing effects and the H₂O continuum. Alongside model comparisons, the RFM is validated against real IASI measurements. Being a Fourier Transform Spectrometer, there are a large number of channels available from the IASI instrument, which leads to a very large quantity of data. However, this can lead to problems within retrievals and data assimilation. Choosing an optimal subset of the channels is an established method to reduce the amount of data; maintaining the information contained within it whilst eliminating spectral regions with large uncertainties. The method currently used at the UK Met Office to select their spectral channels is re-assessed and a modified method is presented that improves upon the modelling of spectrally correlated errors.

Savannahs regulate an agro-ecosystem crucial for the production of domestic livestock, one of the main sources of income worldwide as well as in South African rural communities. Nevertheless, globally these ecosystem functions are threatened by intense human exploitation, inappropriate land use and environmental changes. Leaf area index (LAI) defined as one half the total green leaf area per unit ground surface area, is an inventory of the plant green leaves that defines the actual size of the interface between the vegetation and the atmosphere. Thus, LAI spatial data could serve as an indicator of rangeland productivity. Consequently, the accurate and rapid estimation of LAI is a key requirement for farmers and policy makers to devise sustainable management strategies for rangeland resources. In this study, the main focus was to assess the utility and the accuracy of the PROSAILH radiative transfer model (RTM) to estimate LAI in the South African rangeland on the recently launched Landsat 8 sensor data. The Landsat 8 sensor has been a promising sensor for estimating grassland LAI as compared to its predecessors Landsat 5 to 7 sensors because of its increased radiometric resolution. For this purpose, two PROSAIL inversion methods and semi- empirical methods such as Normalized difference vegetation index (NDVI) were utilized to estimate LAI. The results showed that physically based approaches surpassed empirical approach with highest accuracy yielded by artificial neural network (ANN) inversion approach (RMSE=0.138), in contrast to the Look-Up Table (LUT) approach (RMSE=0.265). In conclusion, the results of this study proved that PROSAIL RTM approach on Landsat 8 data could be utilized to accurately estimate LAI at regional scale which could aid in rapid assessment and monitoring of the rangeland resources. / Environmental Sciences / M. Sc. (Environmental Science)

Savannahs regulate an agro-ecosystem crucial for the production of domestic livestock, one of the main sources of income worldwide as well as in South African rural communities. Nevertheless, globally these ecosystem functions are threatened by intense human exploitation, inappropriate land use and environmental changes. Leaf area index (LAI) defined as one half the total green leaf area per unit ground surface area, is an inventory of the plant green leaves that defines the actual size of the interface between the vegetation and the atmosphere. Thus, LAI spatial data could serve as an indicator of rangeland productivity. Consequently, the accurate and rapid estimation of LAI is a key requirement for farmers and policy makers to devise sustainable management strategies for rangeland resources. In this study, the main focus was to assess the utility and the accuracy of the PROSAILH radiative transfer model (RTM) to estimate LAI in the South African rangeland on the recently launched Landsat 8 sensor data. The Landsat 8 sensor has been a promising sensor for estimating grassland LAI as compared to its predecessors Landsat 5 to 7 sensors because of its increased radiometric resolution. For this purpose, two PROSAIL inversion methods and semi- empirical methods such as Normalized difference vegetation index (NDVI) were utilized to estimate LAI. The results showed that physically based approaches surpassed empirical approach with highest accuracy yielded by artificial neural network (ANN) inversion approach (RMSE=0.138), in contrast to the Look-Up Table (LUT) approach (RMSE=0.265). In conclusion, the results of this study proved that PROSAIL RTM approach on Landsat 8 data could be utilized to accurately estimate LAI at regional scale which could aid in rapid assessment and monitoring of the rangeland resources. / Environmental Sciences / M. Sc. (Environmental Science)

Grasslands are one of the most important ecosystems on Earth, covering approximately one-third of the Earth’s surface. Grassland biodiversity is important as many services provided by such ecosystems are crucial for the human economy and well-being. Given the importance of grasslands ecosystems, in recent years research has been carried out on the potential to monitor them with novel remote sensing techniques. Improved detectors technology and novel sensors providing fine-scale hyperspectral imagery have been enabling new methods to monitor plant traits (PTs) and biodiversity. The aims of the work were to study different approaches to monitor key grassland PTs such as Leaf Area Index (LAI) and biodiversity-related traits. The thesis consists of 3 parts: 1) Evaluating the performance of remote sensing methods to estimate LAI in grassland ecosystems, 2) Estimating plant biodiversity by using the optical diversity approach in grassland ecosystems, and 3) Investigating the relationship between PTs variability with alpha and beta diversity for the applicability of the optical diversity approach in a subalpine grassland of the Italian Alps To evaluate the performance of remote sensing methods to estimate LAI, temporal and spatial observations of hyperspectral reflectance and LAI were analyzed at a grassland site in Monte Bondone, Italy (IT-MBo). In 2018, ground temporal observations of hyperspectral reflectance and LAI were carried out at a grassland site in Neustift, Austria (AT-NEU). To estimate biodiversity, in 2018 and 2019 a floristics survey was conducted to determine species composition and hyperspectral data were acquired at two grassland sites: IT-MBo and University of Padova’s Experimental Farm, Legnaro, Padua, Italy (IT-PD) respectively. Furthermore, in 2018, biochemistry analysis of the biomass samples collected from the grassland site IT-MBo was carried out to determine the foliar biochemical PTs variability. The results of the thesis demonstrated that the grassland spectral response across different spectral regions (Visible: VIS, red-edge: RE, Near-infrared: NIR) showed to be both site-specific and scale-dependent. In the first part of the thesis, the performance of spectral vegetation indices (SVIs) based on visible, red-edge (RE), and NIR bands alongside SVIs solely based or NIR-shoulder bands (wavelengths 750 - 900 nm) was evaluated. A strong correlation (R2 > 0.8) was observed between grassland LAI and both RE and NIR-shoulder SVIs on a temporal basis, but not on a spatial basis. Using the PROSAIL Radiative Transfer Model (RTM), it was demonstrated that grassland structural heterogeneity strongly affects the ability to retrieve LAI, with high uncertainties due to structural and biochemical PTs co-variation. In the second part, the applicability of the spectral variability hypothesis (SVH) was questioned and highlighted the challenges to use high-resolution hyperspectral images to estimate biodiversity in complex grassland ecosystems. It was reported that the relationship between biodiversity (Shannon, Richness, Simpson, and Evenness) and optical diversity metrics (Coefficient of variation (CV) and Standard deviation (SD)) is not consistent across plant communities. The results of the second part suggested that biodiversity in terms of species richness could be estimated by optical diversity metrics with an R2 = 0.4 at the IT-PD site where the grassland plots were artificially established and are showing a lower structure and complexity from the natural grassland plant communities. On the other hand, in the natural ecosystems at IT-MBo, it was more difficult to estimate biodiversity indices, probably due to structural and biochemical PTs co-variation. The effects of canopy non-vegetative elements (flowers and dead material), shadow pixels, and overexposed pixels on the relationship between optical diversity metrics and biodiversity indices were highlighted. In the third part, we examined the relationship between PTs variability (at both local and community scales, measured by standard deviation and by the Euclidean distances of the biochemical and biophysical PTs respectively) and taxonomic diversity (both α-diversity and β-diversity, measured by Shannon’s index and by Jaccard dissimilarity index of the species, families, and functional groups percent cover respectively) in Monte Bondone, Trentino province, Italy. The results of the study showed that the PTs variability metrics at alpha scale were not correlated with α-diversity. However, the results at the community scale (β-diversity) showed that some of the investigated biochemical and biophysical PTs variations metrics were associated with β-diversity. The SVH approach was also tested to estimate β-diversity and we found that spectral diversity calculated by spectral angular mapper (SAM) showed to be a better proxy of biodiversity in the same ecosystem where the spectral diversity failed to estimate alpha diversity, this leading to the conclusion that the link between functional and species diversity may be an indicator of the applicability of optical sampling methods to estimate biodiversity. The findings of the thesis highlighted that grassland structural heterogeneity strongly affects the ability to retrieve both LAI and biodiversity, with high uncertainties due to structural and biochemical PTs co-variation at complex grassland ecosystems. In this context, the uncertainties of satellite-based products (e.g., LAI) in monitoring grassland canopies characterized by either spatially or temporally varying structure need to be carefully taken into account. The results of the study highlighted that the poor performance of optical diversity proxies in estimating biodiversity in structurally heterogeneous grasslands might be due to the complex relationships between functional diversity and biodiversity, rather than the impossibility to detect functional diversity with spectral proxies.