ABSTRACT Questions Earth observation is an important tool in biological monitoring. Previous studies have shown that variation in grassland canopy reflectance can explain gradients in plant community composition. However, there is a need for methods that allow the interpretation of the overall spectral signal of complex grassland vegetation in terms of existing knowledge about individual species' habitat preferences. Can community‐level grassland canopy reflectance be interpreted in terms of the distribution patterns and habitat preferences of individual plant species? Location The island of Öland, Sweden. The sites represent a succession from grazed grassland on previously arable fields to old semi‐natural pastures. Methods We collected data on plant species occurrences and mean hyperspectral reflectance (414–2351 nm) during peak vegetation season in 104 (4 m × 4 m) field plots. We simplified the spectral data into orthogonal components using Principal Component Analysis. Joint species distribution modelling was used to analyse 100 species' relationships with the components. Information on species' habitat preferences was included to explain species' model‐responses. Results Grassland canopy reflectance was associated with variation in the occurrences of individual plant species (41 species with Tjur's D > 0.15, 15 species with D > 0.30), particularly species with distinct habitat preferences. The main gradient in the reflectance data—representing features in the red, blue, SWIR and NIR spectra—was associated with species' preferences for mineral nitrogen availability. In contrast, species' phosphate preferences showed stronger associations with reflectance in the green and red‐edge spectra. Conclusions Data on individual species and their habitat preferences can be used to interpret patterns of variation in community‐level canopy reflectance in grazed grasslands. Vegetation in phosphorus‐poor grasslands—which are valuable for biodiversity conservation—showed a consistent pattern (characterised by features in the green and red‐edge) in the spectral data.
Löfgren et al. (Thu,) studied this question.