Introduction

Genetic diversity of forests provides a foundation for resilience to climate change, biological invasions, and other anthropogenic threats (Crutsinger et al. 2008; Schaberg et al. 2008). High genetic diversity of overstory forest species has been linked to increased productivity and fitness (Aravanopoulos and Zsuffa 1998; Arcade et al. 1996; Jelinski 1993; Knowles and Grant 1981; Mitton et al. 1981), a higher tolerance to pollutants (Bergmann and Hosius 1996; Müller-Starck 1985; Oleksyn et al. 1994), and cascading trophic effects on arthropod (Johnson et al. 2006) and fungal (Tang et al. 2022) biodiversity. As genetic diversity is a basis for adaptation and enhanced resilience, it is vital to preserving forest ecosystems, yet anthropogenic disturbances have resulted in significant declines in forest genetic diversity, reducing the future resistance of affected species (Schaberg et al. 2008). While the 15th Sustainable Development Goal of the United Nation includes aims to stop biodiversity loss, including loss of genetic diversity (Le Blanc 2015), the resources with which to quantify and map genetic diversity are constrained because genetic analyses of forest species require extensive field and lab work (Walters and Scholes 2017).

Remote sensing, in particular imaging spectroscopy, has emerged as a powerful tool for quantifying biodiversity at large spatial scales to understand drivers of biodiversity and inform protection priorities (Asner et al. 2017; Féret and Asner 2011). Imaging spectroscopy generates high-spectral-resolution data spanning the visible to shortwave-infrared (SWIR; 400–2500 nm) electromagnetic spectrum. Applied to vegetation, spectroscopy captures the molecular constituents of leaves, mediated by leaf structure. Leaf traits such as leaf mass per area (LMA), chlorophyll content, and secondary compounds, among others are an expression of adaptation of a species to its environment (Ordoñez et al. 2009; Wright et al. 2004). While such quantitative traits often have a substantial allelic basis (Hallgren et al. 2003; Marron and Ceulemans 2006) that is predominantly polygenic (Bourgaud et al. 2001; Orians et al. 2000), heritability of leaf traits derived from spectroscopy has not been widely tested. Due to the capability of spectroscopy to capture these traits, spectral variation tracks genetic variation among and within forest stands (Blonder et al. 2020; Cavender-Bares et al. 2016; Deacon et al. 2017; Madritch et al. 2014; Martin et al. 2007). According to the spectral variability hypothesis, the variability of canopy reflectance spectra within an area is positively related to plant diversity (Palmer et al. 2000, 2002). Consistent with this hypothesis, leaf-level spectroscopy has revealed heritable spectral differences within and among species of Quercus (oak; Cavender-Bares et al. 2016) and Dryas (an Arctic shrub; Stasinski et al. 2021) as well as within populations of Populus tremuloides (aspen; Deacon et al. 2017) and Metrosideros polymorpha (ohia, Martin et al. 2007). Ploidy levels and genetic varieties of P. tremuloides have been successfully classified using canopy-level imaging spectroscopy (Blonder et al. 2020; Madritch et al. 2014). Further, some studies have revealed patterns of leaf spectra consistent with phylogeographic variation within species (e.g., Quercus oleoides, Fagus sylvatica—European beach—and P, tremuloides; Cavender-Bares et al. 2016; Blonder et al. 2020; Czyż et al. 2020; Madritch et al. 2014) or among species (e.g., Neotropical trees; McManus et al. 2016). To develop imaging spectroscopy as a tool for characterizing genetic variation at the landscape level, we must first understand how spectra vary within continuous forest stands, including variation among conspecific varieties and their hybrids, especially at fine spatial and taxonomic scales. This gap in our understanding of how spectroscopy captures functional variation challenges conservation agendas that seek to include genetic diversity.

Metrosideros polymorpha Gaudich. (Myrtaceae) is an ideal model species for testing the capacity of spectroscopy to characterize functional genetic variation of forest canopies at fine spatial and taxonomic scales. This dominant tree species comprises a large number of vegetatively distinct varieties and races distributed nonrandomly within continuous forests that span environmental gradients and ecotones within the climatically variable Hawaiian Islands (Dawson and Stemmermann 1990; Stacy et al. 2020; Stacy and Sakishima 2019; Treseder and Vitousek 2001). The many forms of M. polymorpha, along with the four other species of Hawaiian Metrosideros, appear to derive from a single colonization of Hawaii by the genus ~ 2.6–3.9 million years ago (Choi et al. 2021; Dupuis et al. 2019; Percy et al. 2008; Wright et al. 2000). Diversification within this group is largely the result of adaptive radiation associated with Hawaii’s diverse abiotic conditions (Ekar et al. 2019; Izuno et al. 2022; Morrison and Stacy 2014; Stacy et al. 2014, 2020; Stacy and Sakishima 2019). On Hawaii Island, the youngest and largest island in the chain, Metrosideros occurs continuously (barring deforestation) from sea level to 2470 m above sea level wherever mean precipitation exceeds 50 cm annually (Stemmermann and Ihsle 1993). The Metrosideros community on Hawaii Island comprises just four varieties of M. polymorpha associated with different environments: M. polymorpha var. incana (new lava flows at low-to-middle elevations and dry areas), M. polymorpha var. glaberrima (mature substrates at all but lowest and highest elevations), M. polymorpha var. polymorpha (all substrates at high elevations), and M. polymorpha var. newellii (riparian zones; Dawson and Stemmermann 1990). All taxon pairs can be crossed to make F1 hybrids (Corn 1979; Rhoades 2012; Stacy et al. 2017), and hybridization between varieties occurs to varying degrees where ranges overlap (Corn and Hiesey 1973; Stacy et al. 2016). Thus, M. polymorpha on Hawaii Island presents the opportunity to examine the utility of spectroscopy to discern very closely related, co-occurring tree taxa and their hybrids and to examine the expression and differentiability of leaf traits in the hybrids.

Here, we use a common-garden population of the four varieties of M. polymorpha on Hawaii Island and their F1 hybrids derived from controlled crosses to address the following questions: Do leaf-level reflectance spectra differentiate the four varieties of M. polymorpha on Hawaii Island? Are patterns of spectral inheritance in F1 hybrids distinct and intermediate to those of their parental varieties, as expected for highly polygenic traits? Finally, for a single variety occurring on multiple islands, we ask: do the reflectance spectra differ between common-garden trees from Hawaii Island and Oahu? We include a discussion of the spectral data in light of evidence of differential adaptation of the four varieties to contrasting environments and to islands of different ages.

Methods

Common garden population

The 54 reproductively mature trees used in this study were raised from seed at Panaewa Farm, College of Agriculture, Forestry, and Natural Resources Management, University of Hawaii Hilo, located 75 m above sea level on east Hawaii Island. Seeds were derived from controlled crosses in natural populations on Hawaii Island and Oahu (Rhoades 2012; Stacy et al. 2017, unpub. data), supplemented by open-pollinated seeds, and all trees were maintained at the farm for use in studies of life history traits and hybrid fertility (Stacy et al., unpub. data). The 8-to-14-year-old trees represented the four varieties of M. polymorpha on Hawaii Island (hereafter designated glaberrima, incana, newellii, and polymorpha; Fig. SI 1), four inter-varietal F1 hybrid genotypes from Hawaii Island, and a single variety (incana) from Oahu (Table 1). With two exceptions, all genotypes comprised trees derived from > 1 site, or trees for which parents were derived from > 1 site in the case of F1 hybrids; the exceptions were individuals of incana from Hawaii Island and Oahu that were derived from controlled field crosses at a single site on each island. All trees were maintained within a 72’ × 35’ coldframe until 2020 when some Hawaii Island-derived trees were outplanted in a common garden adjacent to the coldframe. We assessed the effect of this outplanting on leaf spectra by comparing greenhouse and common-garden trees of incana-polymorpha F1 hybrids following the methods below and found no significant differences. Thus, we determined that outplanting had negligible effects on the spectra, and all samples were combined for analysis of spectra among genotypes.

Table 1 M. polymorpha varieties and F1 hybrids used in this study
Full size table

Leaf measurements

We measured leaf reflectance spectra on six trees from each of the nine genotypes (treating incana from Hawaii Island and Oahu as separate genotypes). A minimum of 11 leaves were collected from each plant, placed in zip lock bags, and stored on ice for transport to the laboratory for analysis within four hours. We selected leaves from sunlit portions of the plant with minimal discoloration (e.g. chlorosis) and sooty mold. Five representative leaves per tree were selected and wiped clean with water and patted dry prior to spectral measurements. Spectral measurements were collected using a leaf clip and field spectrometer at 1-nm intervals from 350 to 2500 nm (Analytical Spectra Devices Inc., Boulder, CO, USA). Spectra were calibrated using a white reference and corrected using parabolic correction to optimize spectrometer measurements (Hueni and Bialek 2017). Parabolic correction was performed to correct for differences in temperature sensitivity of sensors within the field spectrometer. A jump in the spectra often occurs around 1000 nm due to the silicon-based sensors for the visible to near infrared and can be corrected post hoc according to Hueni and Bialek (2017). Finally, the brightness normalization was applied to all spectral measurements, as it minimizes noise (Kruse et al. 1993; Myneni et al. 1989). Reflectance values below 400 nm were removed, as wavelengths between 350 and 400 nm have a low signal-to-noise ratio. Leaf spectra were averaged by plant. Following spectral measurements of all leaves, leaf area was calculated using ImageJ from a leaf scan collected with an EPSON scanner at 600 dots per square inch. Once dried for 72 h at 65 degrees Celsius, leaves were weighed, and leaf mass per area (LMA) was quantified for each plant.

Analysis

To assess whether leaf spectra can differentiate the varieties of M. polymorpha and their hybrids, we used principal component analysis (PCA) and analysis of variance (ANOVA). Using the pca function from the scikit learn python package (version 0.24.1; Virtanen et al. 2020), which uses a covariance matrix for the eigen decomposition, we reduced the 2100 dimensions of the reflectance data to the first 10 principal components (PC). PCA was applied spearately to different genotype groupings detailed below. For each of the first 10 PCs, we evaluated its ability to separate the genotypes using an ANOVA according to the methods in Cavender-Bares et al. (2016), followed by Tukey’s pairwise HSD tests. These methods were performed in python using the statsmodels package (version 0.12.2; Seabold and Perktold 2010) to compare genotypes separately for each of the following six groups: (1) Hawaii Island incana, glaberrima, newellii, and polymorpha; (2) glaberrima-incana, glaberrima, and Hawaii Island incana; (3) incana-polymorpha, Hawaii Island incana, and polymorpha; (4) newellii-polymorpha, newellii, and polymorpha; (5) glaberrima-polymorpha, glaberrima, and polymorpha; and (6) incana from Oahu and Hawaii Island (Table 1).

Although the PCA allowed us to determine if the spectra were differentiable, we used the spectral similarity index (SSI; Eq. 1; Somers et al. 2009, 2012, 2015) to quantify spectral overlap between varieties. The SSI calculates the spectral distance between populations i and j for each wavelength:

$$\text{SSI}= \frac{\left|{\overline{R} }_{b,i}-{\overline{R} }_{b,j}\right|}{sd\left({R}_{b,i}\right)+sd\left({R}_{b,j}\right)}$$
(1)

where R is the brightness-normalized reflectance for each group over n spectral bands. Rather than performing pairwise comparisons, population j was represented by pooled reflectance data from all varieties (including Oahu and Hawaii Island incana). In doing so, we estimated the degree to which each variety diverged spectrally from all varieties. SSI has been used to estimate species turnover (Somers et al. 2015) and as a means of determining which wavelengths distinguish classes (Asner et al. 2018). Here, we plotted SSI across the entire spectrum to quantify the degree of separation, with higher SSI values indicating a higher degree of spectral overlap, between spectra of the M. polymorpha varieties. Further, we calculated the mean SSI by taking an average of 1/SSI across all bands (Eq. 2).

$$\mathrm{mean SSI}= \frac{1}{n}\sum_{b=0}^{n}\frac{sd\left({R}_{b,i}\right)+sd\left({R}_{b,j}\right)}{\left|{\overline{R} }_{b,i}-{\overline{R} }_{b,j}\right|}$$
(2)

To understand within-variety variation, we calculated the mean spectra and coefficient of variation (CV) for each variety across the spectra. The CV is a standardized measure of variation that allows for visual comparison among samples across the full spectrum. Here, we use the CV to visually assess regions of the spectrum that show the greatest variation within each genotype of M. polymorpha. Although this investigation is useful for visualizing diversity in terms of reflectance between varieties, band-by-band assessments of CV are limited because spectra are derived from broader features related to chemical interactions with light.

Lastly, we examined variation among genotypes in leaf traits derived from the reflectance data. Leaf chemical traits were estimated from reflectance spectra using chemometric equations specific to M. polymorpha developed by Asner et al. (2018). These spectral−chemical relationships were determined using the partial least squares regression (PLSR) – prediction residual error sum of squares (PRESS) method that has been used to develop universal chemometric equations for broadleaf species (Asner et al. 2009, 2015; Asner and Martin 2008). As these methods approximate leaf traits, we use them as a means of comparing leaf traits between groups rather than interpreting their absolute value. We estimated eight chemical traits (Table SI 1) using the equations specific to M. polymorpha, including the photosynthetic pigments chlorophylls a and b, the structural molecules lignin and cellulose, and the secondary traits phenols and tannins. Chlorophylls a and b were summed and represented as chlorophyll a + b. Further, nonstructural carbohydrates (NSC) like sugars and starch were estimated along with total nitrogen (N) and total carbon (C). Leaf mass per unit area (LMA) was calculated using leaf area and dry weights quantified from the collected leaves, described above. When discussing leaf trait data, we refer to the chemical leaf traits estimated from the reflectance data as well as the LMA calculated from leaves. Significance of differences in leaf traits between genotypes in the groupings described above was quantified using ANOVA and Tukey HSD tests. All analyses were done using python version 3.6.9.

In summary, we first used principal component analysis (PCA) to reduce this highly dimensional dataset into fewer components that captured a larger proportion of the variance. We then determined whether any of the components could separate the varieties as well as F1 hybrids from their parent varieties using ANOVA and Tukey HSD. To understand differences in reflectance between and within the varieties, we used the spectral similarity index (SSI) and compared their coefficient of variation (CV) and leaf traits.

Results

Spectral divergence among varieties

PCAs of leaf spectra (Table 2) separated all varieties in pairwise comparison except glaberrima and incana. Two PCs (PC1 and PC5) derived from the reflectance spectra significantly differentiated the varieties (p = 0.003 for each; Table 2). In the pairwise comparison, incana and newellii were separable in both PC1 and PC5. PC1 additionally separated glaberrima and polymorpha as well as newellii and polymorpha. Glaberrima and newellii as well as incana and polymorpha were differentiable in PC5. The only taxon pair that was not separable in the first 10 PCs of the reflectance data was glaberrima and incana.

Table 2 Results showing the statistical separability of M. polymorpha varieties using spectra
Full size table

When visually comparing spectra of the four Hawaii Island varieties, the mean brightness-normalized spectra vary most in the visible (400–700 nm) and shortwave-infrared (SWIR; 1500–2500 nm) wavelength regions (Fig. 1a). According to the spectral separability index (SSI), separation between the varieties occurred across the spectra (Fig. 1b), with polymorpha having the greatest mean SSI (29; Table 3). Both polymorpha and Hawaii Island incana were most distinct in the visible and parts of the infrared while glaberrima had the greatest separability in the SWIR and infrared (Fig. 1b). Glaberrima and incana were similar in their degree of spectral overlap with SSI values of 9 and 11, respectively (Table 3). Newellii, which had the highest separability after ~ 1800 nm, had the lowest mean SSI (7; Fig. 1b; Table 3).

Fig. 1
figure 1

a Mean brightness-normalized reflectance (represented as a percentage) and b coefficient of variation (CV) of reflectance values for the four M. polymorpha Hawaii Island varieties and Oahu incana. See Fig. SI 2 for reflectance prior to brightness-normalization. c Spectral separability of all Hawaii Island and Oahu genotypes. Spectral separability was calculated for each wavelength (Eq. 1). Higher values indicate less spectral overlap. See Table 3 for average SSI values

Full size image
Table 3 Within-variety spectral similarity of samples on Hawaii Island and Oahu
Full size table

Among Hawaii Island varieties, incana had the least within-variety variation among the spectra, while newellii and polymorpha displayed the most variation according to the CV (Fig. 1c).Within-variety variation was greatest in the visible and SWIR regions of the spectrum (Fig. 1c). The CV of newellii peaked in the visible region, where newellii had not only the greatest within-variety variation but also the highest reflectance values. This result is also expressed in the estimated chemical data (Fig. 2), where newellii had a greater variability relative to the other varieties and lower values of chlorophyll a + b than polymorpha. While polymorpha likewise had a high CV in the visible, this variety had the lowest reflectance in this region compared to the other varieties, and this corresponded to high chlorophyll a + b. In the SWIR region, which is influenced by many leaf traits, within-variety variation was greatest for polymorpha and newellii, followed by glaberrima. Newellii had higher total N than all other varieties but lower LMA, total phenols, and lignin than some other varieties. Polymorpha had lower cellulose than newellii and higher LMA than glaberrima and newellii. Both polymorpha and glaberrima had a wide variation in NSC, and polymorpha had high variation in LMA. Incana had low variation in all the leaf traits except for tannins. Leaf traits were less useful than PCA for discriminating the varieties (Fig. 2). Cellulose, chlorophyll a + b, lignin, phenols, total N, and LMA separated newellii from all other varieties (Fig. 2). Beyond this, only polymorpha and glaberrima differed significantly in leaf traits (chlorophyll a + b and LMA; Fig. 2).

Fig. 2
figure 2

Boxplots of leaf traits for the Hawaii Island varieties glaberrima (G), polymorpha (P), newellii (N), and incana (I) as well as Oahu incana (OI). Hawaii Island varieties with traits that differed at a significance of p < 0.05 as determined by ANOVA and Tukey HSD are noted with an asterisk. Incana from Hawaii Island and Oahu were also compared using ANOVA, and significant differences are likewise noted with an asterisk. Boxplots denote quartile ranges, with the lower and upper bounds of the box indicating the 25th and 75th percentile. Middle lines in the box represent the median of the data, and the whiskers end at the group minimum and maximum. Outliers are shown as points

Full size image

Spectral patterns in hybrids

The four F1 hybrid genotypes demonstrated different patterns of leaf reflectance relative to their parental taxa. Spectral PC1 scores separated glaberrima and incana as well as glaberrima and glaberrima-incana hybrids (Table 4). Mean spectra of glaberrima-incana F1s fell between the mean spectra of their parent varieties but were closer to glaberrima in the visible and closer to incana between approximately 2000 and 2500 nm (Fig. 3a). Overall, the shape of the CV across the spectrum within glaberrima-incana F1s mirrored that of incana (Fig. SI 4a). None of the leaf traits differed between the glaberrima-incana F1s and either of their parent varieties (Fig. 4a). Variation in F1 leaf traits was often intermediate to or less than that of the parent varieties, except for total C and LMA (Fig. 4a).

Table 4 Results showing the statistical separation of M. polymorpha F1 hybrids and their parent varieties
Full size table
Fig. 3
figure 3

Mean brightness-normalized reflectance (represented as a percentage) of a the F1 hybrids glaberrima-incana (GI) and its parents, glaberrima (G) and incana (I); b the hybrid incana-polymorpha (IP) and its parents, incana (I) and polymorpha (P); the hybrid newellii-polymorpha (NP) and its parents, newellii (N) and polymorpha (P); d the hybrid glaberrima-polymorpha (GP) and its parents, glaberrima (G) and polymorpha (P). See supplementary information Figure SI 4 for CV of F1 hybrids

Full size image
Fig. 4
figure 4

Boxplots of nine leaf traits for all F1 hybrids measured and their parents. The left figure a represents glaberrima (G), the F1 hybrid glaberrima-incana (GI), incana (I), the hybrid incana-polymorpha (IP), and polymorpha (P). The right figure b displays newellii (N), the hybrid newellii-polymorpha (NP), polymorpha (P), the hybrid glaberrima-polymorpha (GP), and glaberrima (G). Genotypes with traits that differed at a significance of p < 0.05 as determined by ANOVA and Tukey HSD are noted with an asterisk. Only groups of the F1 hybrid and their parents (Table 1) were compared using ANOVA and Tukey HSD. None of the traits differed significantly among GI, I, and G according to ANOVA

Full size image

The incana-polymorpha F1 trees showed intermediate values for many of the leaf traits and within-genotype spectral variation, though their mean spectra most often mirrored those of polymorpha (Fig. 3b). Consistent with this trend, PC3 scores (but not PC1 or PC2 scores) separated incana-polymorpha F1s from incana but not polymorpha (Fig. 3b; Table 4). Similar to the glaberrima-incana F1s, spectral variation (CV) of incana-polymorpha most resembled that of incana in shape, but was often intermediate or closer to the other parent (here, polymorpha) in magnitude (Fig. SI 4b). Both incana-polymorpha F1s and polymorpha had higher chlorophyll a + b than incana (Fig. 4a). Many of the other leaf traits of the F1s displayed values intermediate to those of the parent values, though the variation of the hybrid data was often greater than that of either parent.

The mean spectral values of newellii-polymorpha F1s were intermediate in the visible, closely followed polymorpha in the infrared and beyond ~ 1700 nm, and were lower than either parent between 1500 and 1700 nm (Fig. 3c). Newellii and polymorpha was the only pair of genotypes in the newellii-polymorpha group that was separable by any PC scores (Table 4). Leaf trait data indicated that many of the F1 traits were intermediate to those of the parent varieties, but within-F1 variation was lower than variation within either parent for lignin, NSC, and total C (Fig. 4b). In contrast, tannin levels varied more among F1 trees than among trees of either parent. Four of the leaf traits separated newellii and polymorpha, while total N, lignin, and LMA separated newellii and newellii-polymorpha (Fig. 4b). Polymorpha and newellii-polymorpha F1s did not differ in any of the leaf traits.

Mean spectra of glaberrima-polymorpha F1s largely fell between those of the parent varieties, but more closely followed glaberrima in the visible and polymorpha in the SWIR (Fig. 3d). Only the parent varieties were differentiable using reflectance spectra (Table 4). Glaberrima-polymorpha F1s had lower total C relative to polymorpha, although within-group variation of total C was greater in the hybrid than either parent (Fig. 4b). LMA was the only leaf trait for which glaberrima-polymorpha F1s were intermediate to the parents in both median value and within-group variation. The glaberrima-polymorpha outlier values for tannins, lignin, and cellulose were taken from the same plant (Fig. 4b).

Comparing populations across islands

Trees of incana from Oahu and Hawaii Island were compared with assess inter-island divergence of leaf spectra (Fig. 1). Across the full spectrum, except in the infrared (~ 750–1700 nm), mean spectral reflectance of incana was greater for trees from Oahu than those from Hawaii Island (Fig. 1a). The CV was similarly greater for Oahu trees across the spectra, and the shapes of the CV were similar only in the visible (Fig. 1c). PC1 scores significantly differentiated leaf spectra of incana from the different islands (p value < 0.05), and Oahu incana, with an SSI of 4, had the lowest SSI of all the varieties by nearly a factor of four (Fig. 1b; Table 3). Six of the leaf chemical traits differed between islands (Fig. 2). Oahu incana had higher cellulose and total N concentrations, but lower lignin, phenols, LMA, and tannins. Qualitative comparisons suggested that within-group trait variation was greater for Oahu incana in cellulose, lignin, NSC, and tannins.

Discussion

We measured the leaf spectra of several genotypes of a landscape-dominant tree species and demonstrated separation of ecologically diverged varieties across the geographic scale of east Hawaii Island. Leaf reflectance data successfully distinguished all but one pair of varieties of M. polymorpha on Hawaii Island as well as populations of the same variety from different islands. Spectral reflectance measures from four classes of F1 hybrids led to less successful discrimination of intraspecific hybrids from their parental varieties, as expected. However, the results suggest that reflectance spectra should be useful for the detection of M. polymorpha hybrid zones using airborne imaging spectroscopy and that with increased sample size, discrimination of individual F1 hybrids from parental taxa may be possible.

Spectral divergence among varieties

Biodiversity estimates based on imaging spectroscopy, in accordance with the spectral variability hypothesis, have been made across many landscapes (Féret and Asner 2011; Schäfer et al. 2016), but few studies have investigated how spectral variability captures intraspecific variation at finer scales (Cavender-Bares et al. 2016; Czyż et al. 2023; McManus et al. 2016). The current study demonstrates the potential of reflectance spectra to capture the genetic variation within a single hyperdominant tree species. Cavender-Bares et al. (2016) similarly demonstrated separability of common-garden Quercus oleoides from populations across Central America where gene flow was limited due to the geographic separation of populations. Here we demonstrate spectral differentiation at scales much smaller than several hundred kilometers as monodominant stands of different M. polymorpha variants can exist directly adjacent to one another – a promising first step toward landscape-scale mapping of this species. Further, the leaf reflectance and derived trait data may reflect the differential adaptation of the four varieties of M. polymorpha to contrasting environmental niches in accordance with the spectral variability hypothesis (Palmer et al. 2000, 2002), as is discussed below.

The spectral signatures of the four varieties of M. polymorpha on Hawaii Island were separable in pairwise comparisons, except for those of the two successional varieties, incana and glaberrima. Despite their distinct leaf phenotypes, pubescent incana and glabrous glaberrima are the most weakly genetically differentiated pair of varieties on Hawaii Island (DeBoer and Stacy 2013; Stacy et al. 2014) and Oahu (Stacy et al. 2020). Weak differentiation is consistent with their likely multi-million-year history of alternating periods of isolation by selection on new (incana) and old (glaberrima) lava flows and periods of hybridization on intermediate-aged flows (Corn and Hiesey 1973; Drake and Mueller-Dombois 1993; Kitayama et al. 1997; Stacy et al. 2017; Stacy and Sakishima 2019). Glaberrima and incana were not differentiable when all four varieties were included in the analysis; however, they were separable in the analysis comprising just these varieties and their hybrids. This result suggests that classifying glaberrima and incana using airborne imaging spectroscopy will be possible, but it may require the training of a secondary classification model on these varieties alone. In addition to their lack of separability via reflectance data, these varieties had a similar degree of spectral overlap (SSI) with the other varieties. Notably, incana had the lowest CV of reflectance of any variety. This low variation may be due to lower genetic variation among sampled incana due to purifying selection (Cvijović et al. 2018) in the harsh abiotic environments of new lava flows or due to the narrow sampling of incana for this study (i.e., from a single population) relative to the other varieties.

The spectral signatures and leaf traits recorded for polymorpha were consistent with values expected for high-elevation plants. Polymorpha dominates forests above ~ 1400 m and exhibits many traits associated with high-elevation plants, such as slow growth, compact form, and highly pubescent leaves (Dawson and Stemmermann 1990; Homeier et al. 2010; King et al. 2013; Yang et al. 2008). The high LMA observed in polymorpha compared to the other varieties is consistent with expectations, as thicker leaves are often associated with high-elevation plants (Read et al. 2014). Lignin is associated with tensile strength (Trupiano et al. 2012; Zhang et al. 2014), and high lignin in polymorpha may be an adaptation to the mechanical stress of wind at high elevations (Zaborowska et al. 2023). High LMA and chlorophyll a + b in polymorpha are consistent with the relatively higher total chlorophyll and lower leaf surface area observed in common-garden trees of M. polymorpha derived from high-elevation, open-pollinated seeds (Martin et al. 2007). High chlorophyll a + b in polymorpha may be related to leaf pubescence, as pubescent polymorpha leaves self-shade to reduce damage to photosystems (Martin et al. 2007). Low peak reflectance in the visible spectrum is supported by the high chlorophyll content and suggests that polymorpha captures more light than the other varieties (Martin et al. 2007).

The CV of the reflectance spectra was high for polymorpha, which was unexpected given the lower genetic variation of polymorpha relative to other varieties of M. polymorpha on Hawaii Island (Stacy et al. 2014). Moreover, despite its high genetic differentiation relative to incana and glaberrima (DeBoer and Stacy 2013), polymorpha had the greatest spectral overlap with these taxa. Although lower trait variability at high elevations has been observed using imaging spectroscopy data in Peru (Asner et al. 2016) and Hawaii (Seeley et al. 2023), polymorpha had high variability in its reflectance spectra. This appears to be due primarily to high variation in NSC and LMA, which may be a result of growing plants adapted to high elevations in low elevations or to sampling from young potted plants as opposed to full-grown trees. Comparisons of high- versus low-elevation M. polymorpha in situ revealed lower CV of canopy reflectance and trait variability at high elevations (Martin and Asner 2009; Seeley et al. 2023), supporting the conclusion that the greenhouse growing conditions affected polymorpha.

Reflectance spectra for newellii were generally consistent with isolation of small populations in separate riparian environments. Newellii is restricted to small, linear populations along riparian corridors on east Hawaii Island (Dawson and Stemmermann 1990; Ekar et al. 2019). The relatively strong genetic isolation of newellii from the other varieties (mean pairwise FST between newellii populations and populations of all other varieties = 0.13; max = 0.25; pairwise FST between glaberrima and polymorpha = [0.040, 0.137], incana and glaberrima = [0.029, 0.117], and incana and polymorpha = [0.051, 0.079] on young and old substrates, respectively; (Stacy et al. 2014) likely explains the low degree of spectral overlap observed in the SSI. Newellii populations are significantly diverged from each other due to genetic drift (Stacy et al. 2014), and individuals included in this study originated from different populations. Structural flexibility reduces drag in water and is a common adaptation in plants contending with flowing water (Dittrich et al. 2012). As lignin adds rigidity to foliage (dos Santos Abreu et al. 1999), the low lignin observed in leaves of newellii is consistent with adaptation of this variety to high river discharge events (Ekar et al. 2019) as well as the lignification suppression observed in the roots of flood-stressed soybeans (Komatsu et al. 2010). The relatively high total N in newellii leaves may indicate that newellii has higher protein concentration than the other varieties. While some riparian plants use specialized proteins to withstand flooding events (Xue et al. 2020), this has yet to be investigated in newellii. Further, reflectance of visible light was greatest for newellii, which may be a means of photoprotection. Newelli leaves, like those of glaberrima which had the second highest reflectance in the visible, are typically glabrous and therefore do not self-shade via pubescence.

Spectral patterns in hybrids

Patterns of spectra and leaf traits varied across the four F1 genotypes. The glaberrima-incana, incana-polymorpha, and newellii-polymorpha F1s largely showed levels intermediate to those of the parent varieties, whereas the glaberrima-polymorpha F1s did not. Leaf traits of glaberrima-polymorpha often ranged higher or lower than those for either parent, although few of the differences were significant. Interestingly, these same patterns match those observed in the phenotypes of 2-year-old seedlings of these same four F1 genotypes, which were intermediate for all F1s except glaberrima-polymorpha (Stacy et al. 2016, unpub. data).

Phylogenetic signal in reflectance spectra has been demonstrated in multiple genera (Blonder et al. 2020; Cavender-Bares et al. 2016; Czyż et al. 2020; Madritch et al. 2014; McManus et al. 2016; Meireles et al. 2020). Here, we show inheritance patterns of reflectance spectra in intraspecific F1 hybrids of M. polymorpha. Through this study, we hope to understand the applicability of imaging spectroscopy in classifying hybrids in landscape-wide mapping efforts. Of the four F1 genotypes included in this study, only glaberrima-incana and incana-polymorpha were separable from one of the parent varieties using leaf spectra. In the case of glaberrima-incana, spectra could distinguish the hybrid from glaberrima, whereas individual leaf traits could not. For this hybrid and its parents, leaf traits were not distinct enough to discriminate the genotypes. For the other F1 genotypes, at least one of the leaf traits differed significantly between the hybrid and one parent variety. These results indicate that classifying hybrids using airborne imaging spectroscopy may be possible with an increased sample size but will likely require both PCA and leaf trait estimations from spectral data to capitalize on all the information present in the data.

Comparing populations across islands

We assessed whether spectra of M. polymorpha var. incana from islands of differing ages are differentiable and consistent with their contrasting environments. As found in other studies of conspecific populations sampled across a broad spatial scale (Cavender-Bares et al. 2016; Madritch et al. 2014), we found that the populations of incana from Hawaii Island and Oahu had distinct spectral signatures. Further, the SSI indicated that Oahu incana were more distinct spectrally than any of the Hawaii varieties. These results are consistent with a higher genetic similarity of populations within islands than among islands (Choi et al. 2021; Percy et al. 2008; Stacy and Sakishima 2019). Although the spectra of Oahu and Hawaii Island incana are separable, they share a characteristic shape in the CV between 500 and 750 nm. This shape was also present in all incana hybrids and includes a rounded peak around the red wavelengths as well as a sharp peak at the red edge. As the red edge defines the inflection point between red and infrared and has been linked to chlorophyll content, mesophyll structure, and leaf water content (Collins 1978; Horler et al. 1983), it is likely that variability patterns for one or all of these traits are present in incana and inherited by incana hybrids. Within-variety variation of leaf spectra was greater for Oahu incana, which may be due to the weaker purifying selection there relative to that on new lava flows on volcanically active Hawaii Island, the relatively narrow sampling of Hawaii Island incana in this study, or simply the older age of Oahu. Oahu, being approximately 3 million years older than Hawaii Island, has more available nitrogen in its soils (Vitousek et al. 1997), which results in greater trait variability (Asner et al. 2016; Ordoñez et al. 2009) and therefore spectral variability (Seeley et al. 2023).

Conclusion

Using the highly variable, landscape-dominant tree species, M. polymorpha, grown in a common garden on Hawaii Island, we used leaf reflectance spectra and derived leaf traits to distinguish four ecologically diverged varieties and their hybrids with varying degrees of success. Further, we discussed the possible associations between the reflectance spectra and leaf trait data with local adaptation of the four varieties to their respective environments. The intersection of genetic analyses and geographical information systems (GIS) has been important in informing biogeographical research and conservation decisions that seek to protect genetic diversity (Koskela et al. 2013; Zonneveld et al. 2012); however, spatial genetic data of forests are limited. This study demonstrates that reflectance spectra can discriminate genotypes of M. polymorpha, suggesting that while the varieties and hybrids can be spatially mapped using airborne imaging spectroscopy, further investigations are necessary to determine if the resolution from canopy-level data will be stronger or weaker relative to leaf-level data (Jacquemoud et al. 2009; Jacquemoud and Baret 1990). As we plan for the use of imaging spectroscopy in biodiversity studies, the M. polymorpha model system will help us incorporate genetic variation rather than land-cover or morpho-taxonomic variation and pattern into conservation science and management.