Clustering of Leaf and Canopy Spectra: How Do They Correspond to Functional Traits and Types?

Abstract:

Recent ecological research has strongly linked plant function to leaf-level biochemical and plant structural traits over a wide range of species and environmental conditions. Broad classifications of plant functional types (PFTs) have conventionally been used in mapping and modeling efforts, as they capture major patterns in the variation of these traits. While strong relationships between canopy imaging spectrometer data and leaf biochemical composition have been established in tropical ecosystems, these relationships have not been well-documented over a broader range of ecosystems. In this study, we sought to characterize relationships between leaf-level functional traits and leaf and canopy-scale reflectance. We examined how variation within each data set corresponds to conventional PFTs and how additional sources of variation at canopy level can impact these relationships. We collected leaf trait data (pigments, water, dry matter, thickness, carbon & nitrogen) and leaf reflectance from a diverse set of species and PFTs across multiple sites and seasons in California as part of the Hyperspectral Infrared Imager (HyspIRI) mission study. Image data were acquired by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Using ordination analysis, we assessed major axes of variation in leaf traits, leaf spectra and canopy spectra by species and functional types. Hierarchical agglomerative clustering analysis was used to evaluate existing groups within each dataset and to determine how well these match conventional PFTs. We also investigated the role of pixel composition and structure on spectral clustering within the AVIRIS data. Our results show that both leaf spectral clusters and biochemical trait clusters correspond to differences among species and sites, as well as conventional PFTs. Greater variation is present in the leaf spectra than in the biochemical data, indicating the spectra may be sensitive to additional leaf traits. Canopy-level spectral clusters were different from leaf-level clusters, though species and site differences were still observed. We demonstrate these differences can be attributed to pixel composition and canopy structure that must be accounted for if sensors such as HyspIRI will be used to accurately retrieve canopy biochemistry and leaf functional traits.