Advances in ecological remote sensing over recent decades have contributed substantial new insights into patterns and processes of biophysical and biochemical variation in species-rich tropical forests. Imaging spectroscopy, also known as hyperspectral imaging (HSI), has served as the technological backbone for many of these advances [1
], as the fine spectral resolution (2–10 nm) and extended spectral range (380–2500 nm) provided by HSI facilitate the quantitative retrieval of a wide range of biochemical attributes [2
]. Variation in remotely-sensed biochemical traits has been used to assess forest canopy functioning, including water stress, pest pressure, and canopy fluxes in nutrients and carbon [5
]. Further, HSI data have shown significant potential for mapping community composition, including species and plant functional groups, based largely on their biochemical traits [8
]. However, a challenge remains to understand whether trait variation (measured directly or remotely sensed) arising from taxonomic differences is confounded by environmental gradients [11
Phylogenetic relatedness, i.e.
, the degree to which species share evolutionary histories, can explain a substantial proportion of morphological and genetic variation among species within ecological communities [13
]. However, only recently have sufficient data on the relationships of rare tropical tree species become available to allow researchers to assess phylogenetic patterns of trait variation, in addition to their impact on emergent, community-level properties [15
]. Recent phylogenetic studies have contributed insights into the assembly of tropical forest tree communities [18
], but little work has been done to relate phylogenies with remote sensing approaches to tropical canopy community ecology. In the context of high-diversity tropical forests, extensive DNA sequence data, e.g., from DNA barcoding efforts, may prove particularly valuable for understanding the spectral properties of closely-related taxa. For the most part, phylogenetic methods have not been incorporated into remote-sensing based analyses of foliar trait variation. An exception is the work of Asner and Martin (2009) [12
], which quantified the taxonomic and environmental components of trait variation across a gradient of soil fertility in a lowland Amazonian forest, using a hierarchical ANOVA procedure to partition taxonomic variance of biochemical traits into family, genus-within-family, and species-within-genus components. However, while this approach quantifies the entire pattern in phylogenetic relatedness in a community, it does not allow for variance in the distance between taxa at the same taxonomic rank. Explicit phylogenetic analyses using phylogenies derived from DNA sequence data may resolve the role of relatedness in determining patterns of trait variation, particularly among species and genera [22
Many traits that are capable of being detected or quantified from HSI, such as foliar nitrogen, chlorophyll, phosphorus, and specific leaf area vary among tropical tree species in proportion to phylogenetic relatedness or exhibit “phylogenetic signal” [12
]. Moreover, patterns of multivariate trait co-variation, such as stoichiometric ratios between foliar nutrients, are tightly linked with patterns of species diversity [24
], so much so that species-specific chemical portfolios are now utilized to identify species using an approach called Spectranomics
, developed by Asner and Martin (2009) [25
]. Other spectroscopic techniques, such as near-infrared spectroscopy of dried plant materials have been utilized to discriminate species on the basis of their chemical properties [26
]. It might therefore be anticipated that, along with the biochemical traits themselves, variations among the reflectance spectra of co-existing species within a tropical tree community may directly reflect the phylogenetic structure of that community. Methods to quantify the degree of phylogenetic signal are well established for assessing patterns of interspecific trait variation [13
], and adapting such methods to spectral analysis may be achieved by calculating phylogenetic signal for each wavelength across the reflectance spectrum.
Here we ask whether tropical canopy tree foliar reflectance spectra exhibit phylogenetic signal, as measured by the Pagel’s λ [28
], for three distinct Neotropical rainforest canopy communities that vary in soil fertility and community composition. We then investigate the relationship between foliar reflectance spectra and phylogenetic structure by (1) examining the relationships between spectra and a suite of nine biochemical traits that relate to a wide range of leaf functions and which are known to be expressed in leaf reflectance spectra; and (2) directly quantifying the phylogenetic signal of these biochemical traits.
Within three Neotropical forest communities, we found evidence for phylogenetic structure in foliar reflectance spectra primarily within the visible (400–750 nm) and shortwave infrared (1500–2500 nm) spectral regions, both of which are known to be sensitive to foliar biochemistry (reviewed in Ollinger, 2011 [62
]). Moreover, spectral regions exhibiting significant phylogenetic signal broadly overlapped with the spectral regions selected by the multi-model ensemble as those most important to the quantitative prediction of the nine foliar traits measured in this study (Figure 2
). The general co-incidence of phylogenetic signal and biochemical correlations within the visible and short spectral regions affirms the importance of these broad regions to quantifying foliar traits and detecting taxonomic differences, and further suggests a biochemical basis for species identification via Spectranomics
]. However, both the specific spectral regions that displayed phylogenetic structure varied considerably among sites, which may indicate differences in the evolutionary and ecological processes acting within the tropical tree communities at each site.
Within the visible portion of the spectra, we observed contrasting patterns of phylogenetic organization between high and low fertility sites. The phylogenetic signal for species reflectance within discrete regions of the visible spectrum at both high-fertility sites, Tam-I and BCI, may indicate a phylogenetic basis for the absorption of light by chlorophyll and other pigments, which are the predominant biochemical constituents active within this region [3
]. While we found evidence for a strong chlorophyll band selection in this region, we did not observe phylogenetic signal for chlorophyll itself. However, non-pigment biochemical constituents are known to have an indirect influence on the efficacy of the photosynthetic apparatus, as elements such as N and P are directly and indirectly involved in the biochemistry of photosynthesis [30
Soil fertility may play a role in the among-site differences in phylogenetic signal of this spectral region. Nutrient limitation is known to facilitate habitat associations in low-resource tropical tree communities, favoring species that allocate limited resources towards longer-lived, better-defended foliage [11
]. Across a diverse phylogeny, this process may be expected to yield the convergence of distantly related species on similar trait values, which would reduce phylogenetic signal compared to a Brownian expectation [23
]. However, within a relatively high-fertility site, local-scale differences in resource availability (including light availability) may contribute to the maintenance of a wide range of life-history strategies, and therefore permit greater conservation of differences among species [20
]. Although other explanations are possible, it is not implausible to speculate that resource-based constraints on photosynthesis may explain differences in patterns of phylogenetic signal within the visible range between the high and low-fertility sites in this study.
Phylogenetic structure was observed across broad regions within the SWIR when comparing all species included in the study and was observed to some degree among species present at each site, albeit λ was not significant at all sites or within all community phylogenies (Figure 3
). The majority of bands selected by the ensemble for importance to foliar biochemical traits were also located within the SWIR. However, a comparison of patterns of phylogenetic organization between the two Tambopata sites reveals that while more phylogenetic signal is found for foliar reflectance spectra at Tam-U, a larger number of foliar traits
show evidence of phylogenetic signal at Tam-I. The relationship between foliar biochemistry and reflectance spectra is complex, particularly within the SWIR, where numerous biochemical constituents may absorb at overlapping wavelengths [67
]. Direct links between leaf biochemistry and spectral reflectance within the SWIR are driven by changes in the vibrational or rotational properties of molecules [62
]. The relationship between these absorption features and elemental leaf traits, such as N, P, or more broadly-defined functional traits such as LMA, is indirect—contingent upon the allocation of elemental resources to specific biochemical compounds (such as N to Rubisco), or biochemical compounds to foliar functional traits (such as lignin to LMA) [31
]. Patterns of phylogenetic structure within the SWIR may relate more to these specific biochemical differences among species than to laboratory estimates of elemental concentrations or broader functional traits.
A low coefficient of variation (CV) within the near-infrared (750–1500 nm) spectral region across all sites likely contributes to the apparent lack of site-specific phylogenetic signal or importance to prediction of foliar biochemistry within this spectral range (Figure 2
b and Figure 3
). Given the high diversity of tropical tree species found at each site, low dispersion of reflectance values is likely to result in distantly-related species with spectral values more similar than would be predicted by Brownian model of evolution. However, a high spectral CV does not necessarily predict a high value for lambda, as this variation may or may not correspond to patterns of phylogenetic relatedness. Similarly, low spectral dispersion among species reduces the utility of this region for the prediction of foliar biochemistry (Supplementary Figures S10
The pattern of phylogenetic organization of traits and spectra at BCI was more contingent upon the methods of phylogenetic construction than for either site at Tambopata. While λ values generated using the Phylocom
phylogeny revealed significant phylogenetic structure across a substantial portion of the reflectance spectrum and for a few foliar traits (Ca, lignin), values of λ generated using the DNA-based phylogeny revealed no significant phylogenetic signal for spectra or traits. This contrast is likely driven by differences in phylogenetic resolution between the two phylogenies (Supplementary Figures 1–7
). The megatree from which the Phylomatic
phylogeny is derived is resolved only to the genus level, and assumes a “comb-shaped” and unresolved phylogeny below this level (i.e.
, all species equally related) [38
]. The Phylomatic
phylogeny emphasizes evolutionary signal at the basal branches of the phylogeny, whereas the DNA-based phylogeny also includes resolution at the tips of the phylogeny. Thus, the degree of trait divergence among congeneric species in our DNA-based phylogeny outweighs the estimate for phylogenetic signal restricted to the basal nodes in the Phylomatic
phylogeny. While this interpretation holds true across all sites, the effects are most pronounced within the BCI community phylogenies, where a relatively higher proportional representation of recently evolved clades such as Inga
increases the influence of these evolutionary relationships at the tips of the phylogenetic branches on the overall pattern of phylogenetic signal.
Previous work by Asner and Martin (2011) [12
] established the substantial contribution of taxonomic organization to between-species variation of several foliar biochemical traits, many of which contribute to the spectral reflectance signatures of canopy tree species. In the current study, we demonstrate that phylogenetic organization is directly expressed in patterns of spectral variation among species. The patterns of phylogenetic signal observed in the reflectance spectra of three distinct tropical canopy tree communities provide a direct confirmation of the phylogenetic basis of foliar reflectance inferred from the relationships of foliar spectra and traits, and foliar traits and taxonomy [68
]. Quantifying this component, using λ or other phylogenetic metrics [14
], may have many applications in remote sensing, from improving species identification algorithms to providing insights into the processes driving community assembly. Further, incorporating phylogenetic analyses into spectral regression models, such as PLS, RF, and SVM, may facilitate more robust models, by accounting for the statistical non-independence of species [60
]. The increasing availability of well-resolved phylogenies and methods for phylogenetic analysis create an opportunity for all those interested in the evolutionary underpinnings of ecological remote sensing.