Remotely sensed carbon content: The role of tree composition and tree diversity

Optical remote sensing permits modeling of variables related to forest biomass, which is a critical determinant of carbon (C) stocks and fluxes. Plant functional characteristics can be captured by (hyper)spectral data, but it remains unclear whether the links between spectral information and C content are driven largely by tree composition, tree diversity, or by forest attributes not commonly measured in field inventories (e.g., physical canopy structure). Here, we examine the relationship between hyperspectral reflectance and aboveground C content in forests, testing the relative importance of tree composition and diversity in mediating this relationship. We use hyperspectral imaging data from an airborne survey with precisely geo-located field data on canopy trees in two forests in southern Que ' bec, Canada. Spectral data covering visible (VIS), near infrared (NIR) and short-wave infrared (SWIR) wavelengths were extracted for 2626 tree crowns within 64 field plots distributed along an elevational gradient. We applied a continuum removal to the spectra and subsequently performed a principal component analysis to reduce the spectral dimensionality. From the spectral principal components, for each plot we quantified (i) spectral composition (based on average reflectance per plot), and (ii) spectral diversity using the convex hull volume. From field data, we calculated variables characterizing the taxonomic, functional and phylogenetic composition and diversity of canopy trees. Carbon content was calculated using allometric equa-tions based on tree size. We applied a structural equation model based on partial-least squares to test both in-direct effects of spectral composition and diversity on C storage of trees (via on-the-ground tree composition or diversity), and also direct effects (reflecting forest characteristics unmeasured in field surveys). We found that spectral composition, particularly from the VIS, is related to C content largely indirectly, via changes in tree composition along the elevational gradient (a transition from deciduous to coniferous species with increasing elevation). Though spectral diversity was significantly related to tree species diversity, no direct or indirect effects on C content were detected. Overall, our findings support (i) the importance of tree composition (but not diversity) in mediating the link between hyperspectral data and forest C content, and (ii) the use of hyperspectral remote sensing as an effective surrogate of taxonomic, functional, and phylogenetic composition of tree com-munities with strong links to C storage.