Is spectral pixel-to-pixel variation a reliable indicator of grassland biodiversity? A systematic assessment of the spectral variation hypothesis using spatial simulation experiments

Covering 30%–40% of the terrestrial surface, grasslands are important hosts of biodiversity, crucial for nutrient cycles and carbon sequestration. However, these ecosystems face a pressing threat in the form of biodiversity loss, which can disrupt their functioning and resilience. Addressing this challenge requires effective monitoring of biodiversity changes on large scales. Remote sensing emerges as a valuable tool in this endeavour, enabling the assessment of grassland biodiversity through the analysis of vegetation patterns, species composition, and ecosystem health over extensive areas. According to the spectral variation hypothesis (SVH), the link between pixel-to-pixel spectral variation and species diversity in remote sensing images can be used to retrieve plant diversity based on spectral data. Nevertheless, the transferability of the proposed relation across ecosystem types, seasons and spatial resolutions remains unclear. The absence of comprehensive data has hindered systematic assessments of the SVH so far, which would ideally incorporate coherent sets of diversity estimates from remote sensing data and in-situ plant diversity measurements. With this study, we present a combined approach that brings together trait data from field measurements, simulations of spatial species distributions and radiative transfer models for a systematic and in-depth analysis of the SVH in temperate grasslands. Based on simulated grassland communities with different diversity levels, we assessed the spectral-to-species diversity relationship across (1) three temperate grassland types, (2) three seasons and, (3) five spatial resolutions (from 10 m to 0.2 m pixel size). We used the mean Euclidean distance and Rao's Q as measures for spectral diversity and different biodiversity indices to describe the species and trait diversity of the simulated grassland communities. Based on 45000 simulated grassland communities in five different spatial resolutions, we found that the spectral-to-species diversity relationship is not stable across grassland types and seasons, despite the spectral diversity metric used. Correlations with spectral diversity were inconsistent for the different applied diversity indices and no single index outperformed the others. Spectral diversity was mainly driven by the spatial resolution (i.e. pixel size) of the image and not by species richness (SR) or functional trait diversity (FD) per se. Our results further underline that the link between SR and FD is not always prominent in plant communities and the basic assumption of the SVH is fulfilled only under certain conditions. Consequently, we argue that FD, which is an important driver of the spectral signature of a plant community, is not inevitably linked to the number of present species in an image. We conclude that the interplay of SR and FD is crucial for the expression of the spectral-to-species diversity relationship. This study clearly underlines the context-dependency of the SVH and we point out that, although of promising value for distinct ecosystems, it is not universally applicable in grasslands.