COS 138-5 - Linking spectral data to biodiversity: Can we predict tree composition and diversity from hyperspectral imagery along an elevation gradient?

Background/Question/Methods

Biodiversity is a multidimensional concept - encompassing taxonomic, functional, and other elements of variation of life on Earth. Our capacity to assess changes in biodiversity through space and time is dependent on our ability to quantify biodiversity dimensions. Remote sensing can be used to overcome the inherent spareness of field-based biodiversity inventories, with hyperspectral imaging emerging as a leading remote sensing method for plant biodiversity assessment. One approach is to assess plant biodiversity solely from plant spectra (i.e., electromagnetic radiation reflected from plants). This method is built on the assumption that plant spectra are signals of foliar and canopy traits, as well as taxonomic identity, and therefore, correspond to multiple dimensions of biodiversity. We used hyperspectral imaging data from an airborne survey in combination with precisely geo-located field data on canopy trees to examine the degree of correspondence between spectral, taxonomic, and functional composition and diversity along an elevation gradient spanning the temperate to boreal forest biomes in southern Québec, Canada. We then compare how spectral, taxonomic, and functional composition and diversity respond to environmental gradients.

Results/Conclusions

Taxonomic and functional composition and diversity were more strongly associated with spectral composition and diversity derived from the visible to near infrared wavelengths (VNIR; 401-996 nm) than from the short-wave infrared wavelengths (SWIR; 883-2523 nm). Yet, VNIR spectral composition and diversity were relatively weakly associated with taxonomic and functional composition (avg. m₁₂² = ~ 0.4) and diversity (avg. r = ~0.3). These weak associations may be driven by non-linear relationships between spectral and taxonomic/functional diversity - beyond a point, increases in spectral diversity were no longer associated with increases in the other diversity dimensions. Our results suggest that spectral composition and diversity capture information beyond the taxonomic and functional characteristics of on-the-ground vegetation, thereby calling into question the generality of the assumption that remotely sensed plant spectra are proxies of taxonomic identity and functional characteristics. However, despite these weak relationships, we find that spectral, taxonomic, and functional composition and diversity are related to environmental gradients in similar ways. Overall, we conclude that spectral composition and diversity are limited indicators of taxonomic/functional composition and diversity but that there is promise in applying spectral data to further our understanding of how tree composition and diversity vary across environmental gradients.