Spectral diversity predicts functional, phylogenetic, and species diversity across a hyper-diverse biogeographic region

Abstract:

As plant biodiversity loss progresses, there is an urgent need to monitor the spatial distribution of multiple axes of biodiversity. Remote sensing approaches have been a primary tool in this endeavor, using a top-down approach to collect imagery and subsequently link the variation of spectral signatures to biodiversity through validation with ground observations. This approach rests upon the principle that spectral signatures can reveal much about the identity and properties of plants and that the variation of plant reflectance spectra aggregate to indicate community-level diversity at different spatial scales. In this study we explicitly explore this aggregation by taking a novel “bottom-up” approach to calculate leaf-level spectral diversity for 1,267 sites across two adjacent biodiversity hotspots and seven biomes, using community-weighted leaf reflectance spectra of 986 species. After comparing predictive performance of seven spectral diversity metrics, our dendrogram-based metric accurately predicted three out of four biodiversity dimensions examined, with low cross-validation bias for species richness (mean RMSE = 2.44 ± 0.21; cf. mean site richness, 14.70 ± 8.43), functional diversity (mean RMSE = 1382.60 ± 115.73; cf. mean site diversity, 4108.09 ± 2644.557), and phylogenetic diversity (mean RMSE = 259.59 ± 14.49 ; cf. mean site diversity, 1267.53 ± 526.30). We found that the relationship between spectral diversity and biodiversity significantly differed between sampling regions and biomes, indicating what we view as a new ecological concept: “spectral redundancy.”