Ecosystems

, Volume 15, Issue 3, pp 504–517

Sources of Canopy Chemical and Spectral Diversity in Lowland Bornean Forest

Authors

    • Department of Global EcologyCarnegie Institution for Science
  • Roberta E. Martin
    • Department of Global EcologyCarnegie Institution for Science
  • Affendi Bin Suhaili
    • Forest Department Sarawak
Article

DOI: 10.1007/s10021-012-9526-2

Cite this article as:
Asner, G.P., Martin, R.E. & Suhaili, A.B. Ecosystems (2012) 15: 504. doi:10.1007/s10021-012-9526-2

Abstract

Sources of variation among the chemical and spectral properties of tropical forest canopies are poorly understood, yet chemical traits reveal potential ecosystem and phylogenetic controls, and spectral linkages to chemical traits are needed for remote sensing of functional and biological diversity. We analyzed 21 leaf traits in 395 fully sunlit canopies, representing 232 species and multiple growth forms, in a lowland mixed dipterocarp forest of Sarawak, Malaysia. Leaf traits related to light capture and growth (for example, photosynthetic pigments, nutrients) were up to 55% lower, and defense traits (for example, phenols, lignin) were 15–40% higher, in the dominant family Dipterocarpaceae and in its genus Shorea, as compared to all other canopy species. The chemical variation within Dipterocarpaceae and Shorea was equivalent to that of all other canopy species combined, highlighting the role that a single phylogenetic branch can play in creating canopy chemical diversity. Seventeen of 21 traits had more than 50% of their variation explained by taxonomic grouping, and at least 16 traits show a connection to remotely sensed spectroscopic signatures (RMSE < 15%). It is through these chemical-to-spectral linkages that studies of functional and biological diversity interactions become possible at larger spatial scales, thereby improving our understanding of the role of species in tropical forest ecosystem dynamics.

Keywords

canopy chemistrymixed dipterocarp forestleaf chemistryMalaysia remote sensingSarawakspectranomics

Introduction

Humid tropical forests are comprised of thousands of plant species, of which a substantial fraction can acquire upper canopy position, thereby accessing direct-beam solar radiation. These canopy species play a major role in the uptake of carbon from and release of water to the atmosphere, as well as myriad biogeochemical interactions. They also provide sub-canopy conditions and habitat for the entire food web.

Canopy species diversity and ecosystem function are linked through a variety of pathways; one in particular is canopy foliage, and the chemicals contained with leaves. Leaf chemicals can be partitioned functionally into major groups related to light capture and growth, longevity and defense, and maintenance and metabolism. Leaf nitrogen (N), phosphorus (P), water, and chlorophyll-a and -b (chl-a, chl-b), as well as leaf mass per area (LMA), adjust to regulate physiological processes including carbon (C) fixation (Field and Mooney 1986; Poorter and Evans 1998). Secondary metabolites such as lignin, cellulose, phenols, and tannins contribute to foliar defense and longevity (Dudt and Shure 1994). Maintenance-metabolism elements are those required in small quantities to support and mediate multiple functions within the leaf (Schlesinger 1991). Along with carbon fractions such as lignin, foliar concentrations of N, P, and base cations (Ca, K, Mg) are tied to ecosystem-level nutrient cycling and decomposition rates (Vitousek 1984; Aerts 1997). Although regional variation in climate, soils and other factors impart variation in many foliar traits (Vitousek and Sanford 1986), species composition is another determinant of spatial variation in tropical forest canopy chemistry (Townsend and others 2008). Despite the suggestion of a link between biological and chemical trait diversity in tropical forest systems, the nature of that connection remains unclear, because floristic composition also varies within and among different forest types, which may affect the degree to which species dictate canopy chemical patterns.

In recent work from lowland Amazonia, leaf chemical traits ranging from growth-related nutrients, to defense and maintenance compounds, were explicitly linked to variation in species composition across contrasting sites (Fyllas and others 2009; Asner and Martin 2011). Both studies considered substrate controls over canopy chemical traits and LMA, finding that site fertility is expressed in foliar chemical concentrations, but that taxonomy remains an important factor within a site. However, these studies were Neotropical, and thus may not be the representative of other regions or of distant phylogenies. Tropical forest canopies of SE Asia and Oceania are often dominated by species in the Dipterocarpaceae (Ashton 1987; Ashton and others 1988), a family not found in the Neotropics. In a study of lowland Bornean rainforest, Paoli (2006) had uncovered a differential effect of environment and phylogeny on leaf chemical and trait variation: Within the common Dipterocarpaceae genus Shorea, foliar P and LMA variation was influenced more by soil fertility than was foliar N, which more closely tracked phylogeny. Like many studies, however, this one included plants in their sapling stage, and shade variation in the forest understory imparts major foliar chemical variation that, although an important contribution to ecosystem processes, can trump phylogenetic- or environment-based chemical patterns, observed when lighting conditions are held constant (Dudt and Shure 1994; Kitajima and others 2005; Poorter and others 2009). In Sarawak, Kurokawa and Nakashizuka (2008) did control for canopy illumination conditions, while seeking linkages between foliar herbivory and decomposition rates, finding that total N and C display a significant degree of phylogenetic structure. Furthermore, Kenzo and others (2004) found that inter-specific differences in leaf traits were on par with variation linked to light availability determined by tree height (Kenzo and others 2006). Other than a few studies focused on inventories of some canopy chemicals in Bornean forests (Breulmann and others 1998; Breulmann and others 1999), we are aware of no work that evaluates sources of multichemical variation among canopy species in the region, and this information is needed to improve our understanding of whole-canopy and ecosystem function.

Ecological patterns in leaf properties may be measureable well beyond chemical traits, such as in the spectral optical properties of the foliage (Curran 1989; Jacquemoud and others 1995; Sanchez-Azofeifa and others 2009). Imaging spectroscopy, which can resolve contiguous optical reflectance signatures (for example, 400–2,500 nm) of vegetation from aircraft and spacecraft, has proven the most effective means to remotely estimate canopy chemical properties (reviews by Kokaly and others 2009; Ustin and others 2009). However, the strength of a chemical-to-spectral link remains unknown for most vegetation types, and has proven particularly challenging to ascertain in biologically diverse canopies of the humid tropics. Even as remote sensing of leaf traits stands to provide an essential tool to extend chemical information to spatial scales relevant to ecosystem dynamics, we are aware of no studies to quantify relationships between foliar chemical and spectral traits among Bornean forest canopy taxa, which might subsequently advance the role of remote sensing for use in studies of these ecosystems.

Here, we report on the chemical and spectral variation among 395 lowland humid tropical forest canopies in Borneo. We sought to quantify variation in leaf traits within and among canopy species, and to develop a quantitative link between those traits and remotely sensed data. We considered variation in 20 chemical properties, ranging from photosynthetic pigments to carbon compounds and micro-nutrients, as well as LMA, within species, among plant growth forms, and taxonomically. We then determined the relationship between these traits and canopy reflectance spectroscopy using a combination of measurement and modeling techniques. We focused effort at the top of canopy where sunlight control could be applied. We did not measure leaf traits within the canopy vertical profile because changes in leaf properties generally follow light gradients within canopies, and although these relationships vary across species, they can be accounted for when modeling of canopy chemistry and reflectance based on top-of-canopy foliar traits (Poorter and others 1995; Bondeau and others 1999; Asner 2008).

Materials and Methods

Site Description

The study was conducted in the Lambir Hills National Park in the Malaysian state of Sarawak, on the Island of Borneo. Vegetation is classified as lowland mixed dipterocarp forest (Ashton 2005), and as moist lowland tropical forest in the Holdridge system. Soils are classified as Ultisols, with sub-order variation from sandy humults to clayey udults based on highly localized terrain variation on the order of meters (Seng Lee and others 2004). Mean annual precipitation and temperature are 2,450 mm y−1 and 26.3°C, respectively.

Sample collection was carried out over an area of about 100 ha neighboring the Lambir 52-ha plot, that is, part of the Center for Tropical Forest Science (CTFS) network (Condit and others 2005). The canopy ranges in height from 30 to 60 m. The 52-ha plot census data show the canopy to be dominated by individuals in the Dipterocarpaceae, representing more than 40% of the stand-level basal area for all stems larger than 1 cm in diameter (Seng Lee and others 2004). Based on CTFS plot data, we estimate that 300 tree species occupy the upper, sunlit canopy in and around the 52-ha plot, with an additional 25–50 liana species likely to be present (Putz and Chai 1987).

To capture the diversity of sunlit canopies throughout the site, while also maintaining statistical power for replication at the family, genus, and species levels, we sampled 395 individual canopies (tree = 377; liana = 18) from 232 unique species. The 232 species were partitioned into 108 genera and 49 families, including 96 samples for 36 species in 6 genera within the dominant family Dipterocarpaceae and 16 species in 8 genera from the second most important family Euphorbiaceae (Table S1). Of the total, 86 species were selected for replication. Different levels of replication were based on the requirements of the various statistical analyses employed in the study (Supplemental Information online). Overall, our sampling structure captured the most common species, with repeat measurements among them (19% of the species with three or more representatives), as well as the less common species, and as a result, our partitioning approximates the relative abundance of canopy individuals on a basal area basis (Seng Lee and others 2004).

Strict measures were taken to collect samples from fully sunlit canopies. Vouchers were collected from all selected individuals and matched by local expert taxonomists to type specimens kept at the Sarawak Herbarium, Forest Research Centre. We also matched genus names to information provided by Kew Botanic Gardens and revised the family-level taxonomy to follow the Angiosperm Phylogeny Group III (Stevens 2010). Project reference vouchers are kept at the Carnegie Institution facility, and reference photos of all specimens can be viewed at http://spectranomics.ciw.edu.

Leaf Chemical Traits

Leaf collections were conducted using tree climbing techniques and the Sarawak canopy crane facility (Kumagai and others 2004). Only fully sunlit branches of mature leaves were taken and transported to a local site for processing within 15 min of collection. A chemical profile, including 20 chemical elements and compounds as well as LMA, was developed for each sample. Statistical analyses included general linear modeling, nested random-effects analysis of variance (ANOVA) modeling, principal components analysis (PCA), and stepwise linear discriminant analysis (LDA). Detailed methodologies are provided in the Supplementary Information online, and laboratory protocols are downloadable from the Carnegie Spectranomics website (http://spectranomics.ciw.edu).

Leaf and Canopy Spectroscopy

Hemispherical reflectance and transmittance spectra spanning the 400–2,500 nm wavelength range were measured on 12 leaf surfaces immediately after acquiring each branch in the field. The spectral measurements were taken at or close to the mid-point between the main vein and the leaf edge, and approximately half-way from petiole to leaf tip. Care was taken to avoid large primary or secondary veins, while allowing for smaller veins to be incorporated into the measurement. The spectra were collected with a field spectroradiometer (FS-3 with custom detectors and exit slit configuration to maximize signal-to-noise performance; Analytical Spectra Devices, Inc., Boulder, Colorado USA), an integrating sphere designed for high-resolution spectroscopic measurements, and a custom illumination collimator (Supplemental Information). Twenty-five spectra per sample were averaged and calibrated for dark current and stray light, then referenced to a calibration block within the integrating sphere (Spectralon, Labsphere Inc., Durham, New Hampshire USA).

We projected the measured leaf reflectance and transmittance spectra to the canopy level using a radiative transfer model described by Asner and Martin (2008) and in the Supplemental Information online. The model simulates top-of-canopy spectral reflectance based on the measured leaf spectra and incorporates variation of leaf area index (LAI), leaf angle distribution, and other crown geometric-optical properties as they are distributed throughout the canopy. For each field-based sample, a randomly-selected combination of canopy structural parameters based on growth habit was used to generate a canopy reflectance signature. Ranges for LAI and other structural parameters for each growth habit (tree, liana, hemi-epiphyte, palm, vine) were taken from Asner and others (2003) and Asner and Martin (2008). This was repeated 250 times per sample, and the mean reflectance signatures were recorded for subsequent analyses.

Results

Canopy Chemical Variation

We recorded wide ranging values for nearly all leaf chemicals and LMA in the upper canopy of this forest (Table 1). Foliar Ca and Mn had the largest coefficients of variation (CV = 106–122%), whereas variation in total C and water was limited to 6 and 12%, respectively. Photosynthetic pigments varied by 31–36% among species, and C fractions by 27–38%.
Table 1

Descriptive Statistics for Leaf Traits, Arranged by Broad Functional Group, Collected from Sunlit Canopies in Lambir Hills National Park, Sarawak

 

M

SD

CV

Min

Max

Light capture and growth

 Chl-a (mg g−1)

4.34

1.43

32.87

0.91

10.08

 Chl-b (mg g−1)

1.63

0.58

35.69

0.36

4.09

 Carotenoids (mg g−1)

1.28

0.40

31.46

0.24

2.95

 LMA (g m−2)

118.8

40.7

34.3

51.4

244.1

 Water (%)

55.6

6.5

11.6

41.1

77.3

 N (%)

1.44

0.40

28.05

0.72

3.53

Longevity and defense

 C (%)

50.1

2.9

5.9

40.6

56.1

 Soluble C (%)

41.6

11.2

27.0

18.2

70.0

 Hemi-cellulose (%)

10.0

3.8

37.9

0.00

25.0

 Cellulose (%)

21.9

6.2

28.3

6.7

41.7

 Lignin (%)

26.0

9.9

38.1

5.7

61.6

 Phenols (mg g−1)

108.7

47.2

43.4

5.4

358.2

 Tannins (mg g−1)

55.8

28.3

50.7

0.00

173.0

 P (%)

0.06

0.02

37.15

0.02

0.15

Maintenance and metabolism

 K (%)

0.75

0.39

51.27

0.15

2.62

 Ca (%)

0.49

0.52

106.38

0.04

4.47

 Mg (%)

0.24

0.11

48.07

0.05

0.71

 Zn (μg g−1)

11.1

10.1

91.0

3.5

95.7

 Mn (μg g−1)

276.2

336.9

122.0

6.6

2359.0

 B (μg g−1)

33.2

20.7

62.4

6.8

149.0

 Fe (μg g−1)

48.2

25.2

52.2

20.8

219.8

LMA = leaf mass per unit area; M = mean chemical concentration; SD = standard deviation; CV = coefficient of variation; Min = minimum; Max = maximum.

Elements and compounds differed by growthform (Figure 1): Lianas had a 14% higher mean concentration of chlorophylls and carotenoids, and an average 6–10% higher concentration of N, water and P, as compared to trees. Base cation and micronutrient concentrations were elevated in lianas compared to trees, with an average 78 and 90% higher concentration of Ca and Mn, respectively. In contrast, concentrations of lignin, phenols and tannins averaged 8–21% lower in lianas than in tree canopies.
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-012-9526-2/MediaObjects/10021_2012_9526_Fig1_HTML.gif
Figure 1

Mean percentage difference between lianas and trees for each leaf trait. LMA is leaf mass per unit area.

Comparisons among taxa indicated that dipterocarps maintain chemical properties that are significantly different from other trees found in the canopy (Figure 2; Table S3). Specifically, they had much lower concentrations of base cations and micronutrients, with Ca being the most suppressed at an average 57% lower than in other trees. Photosynthetic pigment concentrations, along with soluble C fractions, were also highly suppressed among dipterocarps. In contrast, LMA and tannin concentrations were elevated by 23% and 35%, respectively. Similar results were found comparing the genus Shorea to all non-dipterocarp species (Figure 2; Table S3), further indicating that this genus heavily contributes to the patterns described for the Dipterocarpaceae family as a whole.
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-012-9526-2/MediaObjects/10021_2012_9526_Fig2_HTML.gif
Figure 2

Mean percentage difference between Shorea (S; grey bars) or Dipterocarpaceae (D; black bars) and non-dipterocarp “Other” trees for each leaf trait. LMA is leaf mass per unit area.

Intra-specific Variation

The coefficients of variation reported here are for species with three or more individuals (n = 43 species). Although this represents a subset of all canopy species found at the site, it is equal to approximately 50% of the total number of samples collected, whether among all taxa or among dipterocarps or Shorea, thereby providing a good representation of the community without biasing of results caused by taxonomic grouping. The median of the distribution of intra-specific variation in leaf properties ranged from a low of 0.1% for total C to a high of 12% for Mn (Figure 3). The distributions were highly skewed left in all cases, and whereas median values for within-species variability were extremely low, variances reached 40–60% in some cases and more than 100% for Mn. The patterns were similar when considering only the Dipterocarpaceae (Figure 3B), and when only Shorea was considered (Figure 3C); however, median intra-specific CV values were at times up to 20% higher within these sub-groups.
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-012-9526-2/MediaObjects/10021_2012_9526_Fig3_HTML.gif
Figure 3

Distributions of coefficients of variation (CV) for 20 chemical traits and leaf mass per unit area (LMA) for A non-dipterocarp species, B species in the family Dipterocarpaceae, and C species in the genus Shorea. Chemical data are mass-based; number of species = 26, 16, and 9 for nondipterocarp species, Dipterocarpaceae, and Shorea, respectively, with three or more replicates per species.

Inter-relationships Among Leaf Properties

Chemical traits, and the variations among them, are not generated individually within a leaf, but instead are maintained as a constellation of properties meeting multiple physiological and life strategy requirements (Evans and Seemann 1989; Chapin 1991). Thus, it is more useful to consider the chemical portfolio of species, also known as ‘chemical signatures’, by combining leaf traits. However, the potential diversity of chemical signatures among species is also a function of inter-relationships among these traits. Principal components analysis (PCA) showed that a single linear combination of traits (PC1) explained just 30% of the variation among all 21 leaf properties, and 14 orthogonal combinations were required to explain 95% of the overall variation (Table 2). A similar pattern in all 21 traits held when considering only samples taken from the Dipterocarpaceae. PC1 from the all-trait PCA was positively correlated with pigments, N and P (r = 0.66–0.71; Pearson correlation, p < 0.05), and negatively correlated with LMA (r = −0.69). PC2 was positively linked to soluble C (r = 0.73) and negatively correlated with cellulose (r = −0.59) concentrations. Finally, PC3 was positively correlated with phenol and tannin concentrations (r = 0.74–0.79).
Table 2

Principal Components Analysis (PCA) Results for Different Combinations of Leaf Properties

 

Leaf properties

Variance explained (%)

Raw

Log-transformed

(i)

Chl a, Chl b and carotenoids

96.4

96.7

(ii)

LMA, N, and P

69.7

70.1

(iii)

N, P, Ca, K, and Mg

49.1

49.7

(iv)

Soluble C, cellulose, hemi-cellulose, and lignin

58.8

51.6

(v)

i + ii + iii + iv above

37.7

35.2

(vi)

Remote sensing properties1

32.7

28.7

(vii)

All leaf properties

30.7

28.1

1Fifteen leaf properties selected for remote sensing analysis including water, C, N, leaf mass per unit area (LMA), chlorophyll a, cellulose, soluble C, phenols, tannins and hemi-cellulose. The percentage variance explained by the first principal component is shown for raw and ln-transformed data.

We also considered other leaf trait combinations, beginning with photosynthetic pigments alone, for which PC1 accounted for more than 96% of the variation (Table 2). Among three of the ‘leaf economics spectrum’ traits of LMA, N and P (Reich and Oleskyn 2004; Wright and others 2004), we found that PC1 accounted for nearly 70% of the variability. However, PCA indicated weaker explanatory power among combinations of C fractions (58%). Again, similar patterns were found when limiting the analysis to the Dipterocarpaceae (data not shown). Correlation analyses supported the PCA results, indicating weak correlations, with r values less than 0.40 for more than 90% of the trait pairs (Table S4). Major exceptions included inter-correlations among photosynthetic pigments (r = 0.95–0.97), which were expected based on our PCA results (Table 2), and which are well understood biologically (Sims and Gamon 2002).

Taxonomic Partitioning of Chemical Traits

An average 57% of the foliar trait variation was explained by nested taxonomic assignment to families, genera and species, including contributions from intra-specific variation (Figure 4A). Factors other than taxonomy contributed to a residual, which includes unaccounted variation from micro-environment such as the substantial local-scale variation in soil texture at this site (Seng Lee and others 2004), growth form (tree vs. liana), sample selection and analytical error. Yet the residuals averaged just 43% of the total variance among chemical traits. The overall pattern became clearer when considering individual chemical traits: Nested taxonomy accounted for 74–75% of the variance in foliar Zn, C and water concentrations (Figure 4). Species accounted for 55–66% of the variation observed in macronutrients—N, P and the base cations Ca, K, and Mg. In contrast, chl-a, chl-b and carotenoids displayed relatively weak taxonomic organization (32–40%).
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-012-9526-2/MediaObjects/10021_2012_9526_Fig4_HTML.gif
Figure 4

Nested random-effects ANOVA results for canopy taxa from A all taxa sampled at Lambir Hills National Park, B all taxa but without Dipterocarpaceae, C all taxa without Shorea.

Families alone accounted for an average 27% variation in chemical traits, although at this deeper phylogenetic level, we found weak family-level organization of pigments, LMA, and some C fractions, and oppositely, strong accounting of Zn, K and water variation (Figure 4A). When excluding Dipterocarpaceae from the analysis, nested taxonomic organization accounted for an average 55% of the variance among chemicals, with a high of 76% for cellulose. Family-level organization varied by only a small amount, and the strongest control was found for Zn, K, and water (Figure 4B). Removal of the genus Shorea led to an increase in the genus-level organization of carotenoids and LMA, as well as soluble C, cellulose, hemi-cellulose, lignin and phenols.

Regression analyses supported results from the nested ANOVA tests (Table S5): Family, genus and species accounted for an average 20, 29 and 52%, respectively, of the variation in leaf chemicals and LMA (adjusted-r2, p < 0.01). A total of 14 of 21 traits had species-level regression adjusted-r2 values exceeding 0.50, and the strongest results were 0.70 for cellulose, 0.68 for water, and 0.67 for total C. Rerunning the analyses without the Dipterocarpaceae or without Shorea indicated the same overall pattern, at times increasing or decreasing the power of the regression by small amounts (Table S5).

Chemical Classification

Stepwise linear discriminant analysis (LDA) was performed on a 242 sample subset of the data (86 species) containing two or more species representatives to understand how multi-chemical foliar signatures relate to taxonomic classification. This analysis was performed to assess the potential for classification at the species, genus and/or family level when more chemical data are available via ecosystem-scale remote sensing collections. LDA indicated that 81% of the species were correctly classified using their full chemical signatures (Figure 5). A combination of C, hemi-cellulose, phenols, tannins, LMA and Zn classified more than 50% of the species correctly (Table 3); adding other elements and compounds yielded less than 5% more power per chemical to the classification until the signature saturated at about 18 constituents. Family- and genus-level analyses performed on the same dataset generally mirrored species results (Figure 5), although with lower overall classification accuracy due to the increased variation among leaf traits and the unavoidable, concurrent increase in the number of samples per class at these lower taxonomic levels.
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-012-9526-2/MediaObjects/10021_2012_9526_Fig5_HTML.gif
Figure 5

Increasing accuracy of chemically-based classification of canopy species using stepwise linear discriminant analysis (LDA): chemical signature composition steps (1–21) map to specific leaf traits added at each increment. The traits are listed for family (short-dashed), genus (long-dashed line), and species (black line) in Table 3.

Table 3

The Cumulative Order in which Leaf Traits Enter a Stepwise Linear Discriminant Analysis (LDA) for Canopy Taxonomic Classification

Step

Species (86)

Genus (46)

Family (27)

1

C

C

C

2

Hemi-cellulose

Hemi-cellulose

Tannins

3

Phenols

Tannins

Hemi-cellulose

4

Tannins

Phenols

Phenols

5

LMA

Chl-b

B

6

Zn

Carotenoids

Carotenoids

7

Fe

Chl-a

Chl-a

8

Chl-a

Ca

N

9

N

Water

Fe

10

Mg

N

Cellulose

11

Ca

K

Water

12

Carotenoids

Mg

K

13

Cellulose

Zn

LMA

14

P

LMA

Mg

15

Chl-b

B

Ca

16

Soluble C

Lignin

Chl-b

17

B

Fe

P

18

K

Cellulose

Soluble C

19

Water

Soluble C

Zn

20

Mn

P

Mn

21

Lignin

Mn

Lignin

Chl = Chlorophyll; LMA = leaf mass per unit area. The analysis was repeated at species, genus and family levels, always holding the number of samples constant at 242. LDA modeling results are shown in Figure5. The number of unique taxa at each taxonomic level is provided in parentheses.

Canopy Spectroscopy

Variation in modeled canopy reflectance, derived from measured leaf optical properties among 395 individual canopies, is shown in Figure 6A. The largest spectral coefficients of variation were found in the shortwave-infrared (1,300–2,500 nm) (Figure 6B). This suggests that, although the overall reflectance magnitude is lower in the shortwave-infrared, variation within this spectral region is high and should be indicative of chemical variation among canopies. Although intra-specific variation in canopy reflectance also varied by wavelength to a maximum of about 10% in the shortwave-infrared (Figure 6C), this was still on average about 50% lower than the inter-specific variation (Figure 6B). Spectral variance was similar when considering only Dipterocarpaceae or Shorea (Figure 6D), and like the site-level variance (Figure 6B), was double that of intra-specific variation (Figure 6C).
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-012-9526-2/MediaObjects/10021_2012_9526_Fig6_HTML.gif
Figure 6

A Minimum and maximum values of modeled spectral reflectance for canopies in the Lambir Hills National Park, Sarawak. B Spectral coefficients of variation (CVs) for all canopies (n = 395). C Mean intraspecific CV of canopy spectral reflectance. D Spectral CV of all samples included in the genus Shorea (dotted line; n = 62) and in the family Dipterocarpaceae (solid line; n = 96). The canopy spectra were derived from leaf spectra using radiative transfer modeling (Asner and Martin 2008).

Partial least squares regression (PLSR) is particularly well suited for linking spectroscopic reflectance signatures to chemical data because it uses the full spectrum to build predictive models that later can be applied and tested with image spectra. The PLSR model results suggested that a number of leaf properties could be estimated using high-fidelity spectroscopy at the canopy scale (Table 4). Validation r2 values ranged from a low of 0.08 for Mn to a high of 0.94 for LMA. Of 21 traits tested, 13 had r2 values of 0.70 or higher. Root mean square error (RMSE) values expressed as a percentage of the mean for each chemical trait (equivalent to the standard error of prediction for an independent dataset) ranged from a low of 5.2% for water to a high of 50.2% for Mn. Chemical traits found to be critical to taxonomic classification (Table 3), such as C, hemi-cellulose, phenols and tannins, had prediction RMS errors of 8.2, 10.1, 7.7, and 17.0%, respectively (Table 4). Taking a cut-off RMSE below 20% and r2 above 0.50, or combining these as a Remote Sensing Use Index (RS Use Index) value (7.5), we found that 18 of 21 leaf properties are most useful for taxonomic classification based on remotely sensed spectroscopy. Considering that chl-a, chl-b and carotenoids are highly correlated (Table S3), our list of leaf properties must be reduced to 16 for subsequent remote sensing classification analyses. Testing for correlation among these remotely sensible traits with PCA, we found that about 33% of the trait ensemble was expressed in the first axis of variation (Table 3). Eleven principal components were subsequently required to account for 95% of the variance among these 16 traits. Finally, a second LDA analysis with the 16 remotely sensed traits indicated that 76% of species can be correctly classified (Figure S1), with C, hemi-cellulose, phenols and tannins accounting for almost half of the classification model (48% total). Genera and families were classified to 48 and 33% accuracy, respectively (Figure S1).
Table 4

Spectroscopic Estimation of Leaf Chemical Properties and Leaf Mass per Unit Area (LMA) Using Modeled Canopy Reflectance Signatures (Figure 6A)

Leaf property

Latent vectors

RMSE

RMSE (%)

r2

RSUI

LMA

32

10.23

5.2

0.94

0.33

Water

33

1.70

5.4

0.92

0.40

Cellulose

32

1.84

5.3

0.91

0.47

N

39

0.14

6.7

0.86

0.92

Chl a

31

0.58

7.6

0.84

1.25

C

38

1.16

8.2

0.84

1.30

Chl b

31

0.24

8.0

0.83

1.36

Carotenoids

31

0.18

8.1

0.81

1.53

Phenols

30

20.36

7.7

0.80

1.56

P

25

0.01

8.6

0.77

1.97

Soluble C

38

5.56

9.4

0.76

2.31

Lignin

36

4.99

9.4

0.75

2.39

Ca

48

0.36

11.2

0.70

3.35

Hemi-cellulose

34

2.81

10.1

0.66

3.47

Fe

19

17.02

11.4

0.61

4.50

Mg

28

0.07

14.8

0.63

5.51

K

17

0.24

17.2

0.56

7.59

Tannins

20

18.65

17.0

0.55

7.69

B

16

17.25

24.1

0.39

14.79

Zn

12

9.11

27.2

0.26

20.15

Mn

5

350.49

50.2

0.08

46.17

RMSE = root mean square error; RSUI = remote sensing use index.

Discussion

Sources of Chemical Diversity

This dipterocarp forest harbors co-existing species with very wide ranging values for most leaf traits, but how does this variation compare to published compilations? LMA ranged from 51.4 to 244.1 g m−2 in the upper canopy at Lambir, which nearly equals the range reported for global humid tropical forests (Asner and others 2011b) and other biomes (Poorter and others 2009). A similar result held for N, which varied from 0.72 to 3.53% at Lambir, approaching the global range reported by Wright and others (2004). However, we also found that foliar P was held to a much narrower range in the Lambir canopy, 0.02–0.15%, compared to Wright and others’ (2004) global compilation. Foliar N:P ratios are often used as an index of site fertility (Hedin 2004), and we found that canopies at Lambir have a mean N:P of 25.6 (s.d. = 5.5), which is two standard deviations greater than the threshold of 14–16 said to indicate substrate-driven P limitation (McGroddy and others 2004). P limitation has a cascading effect on patterns of plant nutrient supply and demand, as well as decomposition processes, resulting in ecosystem-level feedbacks that can promote or limit primary production (Vitousek 1982; Vitousek and Sanford 1986; Vitousek and Walker 1987). Despite broad patterns of canopy chemistry associated with site fertility, there also often exists taxonomic pattern in chemical traits (Townsend and others 2008), suggesting an additional role of evolution in maintaining biogeochemical processes.

Where exactly does the chemical trait variability reside in the Lambir canopy? For many traits, we found major differences by growthform (Figure 1). Concentrations of P and other rock-derived nutrients (Ca, K, Mg, Zn, Mn, B and Fe) were 14–82% higher in lianas, as were pigment concentrations (17–18%). Although we did not assess whether localized variation in soil nutrients were correlated with liana presence-absence (Schnitzer and Bongers 2011), Paoli and others (2006) did find associations between soil conditions and both canopy chemistry and composition among dipterocarp trees. We thus suspect that there are similar niche partitioning processes at work for lianas. Nonetheless, there is a clear growthform effect by which lianas concentrate growth chemicals (nutrients, pigments), while maintaining lower concentrations of longevity and defense chemicals (LMA, lignin, phenols, tannins).

Beyond the pattern based on growthforms, we found that the Dipterocarpaceae maintains systematically lower concentrations of nutrients and photosynthetic pigments, and higher concentrations of secondary metabolites, than do trees in other families (Figure 2). We found a similar pattern within the genus Shorea (Figure 2). Moreover, intra-specific variation in most leaf traits within the Dipterocarpaceae, and even within the genus Shorea, was equivalent to what is found among all other species in the forest (Figure 3). That a single family or genus can strongly diverge in foliar chemical make-up from other canopy species is surprising, but given the observed dominance of Shorea and the dipterocarps at Lambir and in similar Bornean forests (Ashton 1987; Condit and others 2005), the results suggest that these chemical trait divergences are advantageous in a life strategy context affecting growth, longevity and defense. An alternative explanation might be that limits to migration (biogeography) and evolutionary processes have produced a forest dominated by a family that has subsequently undergone chemical trait radiation based on community-scale interactions such as competition and plant-pest dynamics (sensu Janzen 1970; Coley and Barone 1996). This will be difficult to assess prior to the development of a quantitative phylogeny for dipterocarps and other species found in the system, as well as an improved knowledge of host-specific interactions across trophic levels.

Taxonomic Patterns

Taxonomic partitioning of chemical traits is ultimately a function of intra-specific variation and phylogenetic distance among taxa. We did not consider distance in the formal sense because we do not have genetic (for example, barcode) data. Instead, we focused on relative distance for each trait, which is an approach used by others to assess broad taxonomic patterns in plant traits at regional (Fyllas and others 2009) and global scales (Chave and others 2009). With nested random-effects ANOVA models that incorporate intra-specific variation, we found that taxonomic assignment (family-genus-species) accounts for a low of 32% (chl-a) to a high of 75% (Zn) of the chemical variance among canopies (Figure 4). However, the relative contribution of family, genus and species organization varied widely for each trait. For example, pigment variation, although the weakest in terms of taxonomic partitioning of variance, was almost entirely organized at the species level (Figure 4). In contrast, several micronutrients (K, Mg, and Zn) fell under mostly family-level organization. Still others such as the secondary metabolites (soluble C, cellulose, and lignin) displayed mostly genus- and species-level organization, without a pattern among families. Chemical partitioning at differing taxonomic levels suggests differential patterns and controls over chemical trait evolution. This also suggests that traits may be queuing to different selective forces over time, such as Zn with a deeper and older evolutionary history versus chlorophyll with a more recent species-level radiation.

What effect, if any, does a dominating family or genus have on patterns in taxonomically-grouped chemical traits throughout a forest canopy? Although species in the Dipterocarpaceae or Shorea maintain chemical concentrations that are very different from other taxa in the canopy (Figure 2), the stand-level taxonomic partitioning of chemicals is nearly invariant whether or not this most common family or genus is included in the ANOVA (Figure 4B, C). This suggests a common set of evolutionary processes at work in creating the observed chemical partitioning among all canopy taxa, which would go against the argument that dipterocarps have somehow undertaken a different pathway of trait evolution.

Environmental constraints, such as strong P limitation, may affect phylogenetic partitioning of canopy chemical traits in a system such as Lambir. Compared to a site in western Amazonia with relatively high P availability (N:P = 18.1 ± 5.4) but in similar climate conditions (Asner and Martin 2011), the Bornean canopy had up to 117% lower concentrations of leaf P and Ca. Foliar longevity and defense traits including LMA, lignin, phenols and tannins were up to 22% higher in Borneo. Although the Bornean site maintains a canopy of lower nutrient concentrations and greater investment in tougher foliage than in the Amazonian site, taxonomy still explains a substantial fraction of the chemical trait variability at both sites—72% in the Amazon and 57% in Borneo. The main difference is that a maximum of 6% of the floristic composition is explained by a single family in Amazonia, whereas more than 40% is driven by dipterocarps at Lambir. So although strong P infertility may limit the diversity of families found at a site like Lambir, both the chemical variation and the taxonomic partitioning of that variation appears to be maintained independent of the number of dominating families present. This supports the hypothesis that chemical trait variation is driven by evolutionary radiation based on species interactions (for example, competition, allelopathy) and host-specific defense (Fine and others 2004; Kursar and others 2009).

Linking Chemical and Biological Diversity

Although individual canopy traits may display differing levels of intra-specific variation, and have differing degrees and patterns of taxonomic organization, combining them into chemical signatures tends to improve the differentiation of taxa on a chemical basis. However, the dimensionality of the foliar chemical signatures depends upon the degree of correlation among leaf properties. We found that most leaf properties at Lambir are uncorrelated, and that at least 14 axes of variation exist among 21 leaf traits (Table 2, Table S4). Although LMA, N and P forms a highly inter-correlated, growth-related leaf ‘economics spectrum’ (Reich and others 1997; Wright and others 2004), our data indicate that adding chemical constituents, many of which are acquired and synthesized by plants to manage metabolism, longevity and defense, creates a rather unique portfolio for many canopy species. This is strongly evidenced in the LDA results, which showed that 81% of the multi-chemical trait variation (including intra-specific variation) was explained at the species level. About 42% of the variation was explained at the family level. These results suggest the existence of a wide variety of strategies among coexisting canopies in humid tropical forests, expressed in chemical signatures.

Prospects for Remote Sensing of Biological Diversity

Until recently, the possibility of remotely sensing canopy functional or biological diversity in tropical forests seemed out of reach, but advances in imaging spectroscopy have opened new doors that may allow for an improved mapping of canopy traits (Castro-Esau and others 2004; Clark and others 2005; Carlson and others 2007; Sanchez-Azofeifa and others 2009; Papeş and others 2010). These and other studies provide novel links between spectral data and species composition, yet a more general approach to combining canopy taxonomy and spectroscopy might best be made via the chemical properties of canopies. Asner and Martin (2009) developed the spectranomics concept to link spectroscopic remote sensing to canopy diversity via differences in the leaf chemical signatures found among many species. The approach was tested in a lowland Amazon forest (Asner and Martin 2011), revealing that the majority of species, spread among a large number of families, had unique chemical and thus spectral signatures. However, intra-specific variation in chemical attributes and the diversity of chemical signatures among co-existing canopy taxa each influences whether spectroscopic signals might indicate the richness and abundance of species (Asner and others 2009). Our remote sensing goal in Borneo was thus to assess whether these chemical-to-spectral linkages made in a community dominated by one family would be different from that of the Amazon site harboring a more even distribution of families.

Compared to the Amazon site and to most other tropical forest systems we have assessed (Asner and others 2011a), we found the Lambir forest to have high inter-specific and low intra-specific spectral variability (Figure 6). This variability was about the same among the Dipterocarpaceae, and within the genus Shorea (Figure 6D), as it was at the site level among hundreds of species (Figure 6B). The strong spectroscopic variability within this dominant family or genus is commensurate with the pronounced chemical variation measured among their chemical traits. The formal link between spectral and chemical properties was made using PLSR analysis, a technique which treats the spectrum as a single, contiguous measurement, rather than as a series of spectral bands. In the past, particular spectral regions or bands have been singled out for analysis of canopy chemical traits (Curran 1989). However, the entire spectrum can now be analyzed as a contiguous measurement, which greatly improves accuracy and transferability of the new relationships to other ecosystems (Boulesteix and Strimmer 2006; Feilhauer and others 2010). And new evidence suggests that treatment of the spectrum as a contiguous signal is more in line with the way that plants invest in and make trade-offs among the multitude of chemical compounds affecting vegetation–light interactions (Ustin and Gamon 2010; Ollinger 2011). With current computing technology, it is no longer necessary to discard spectral information by selection of narrow wavelength regions for processing, and an increasing number of studies are showing the advantage of using the full spectrum for remotely-sensed chemical analysis (for example, Martin and others 2008; Skidmore and others 2010; Knox and others 2011).

With this in mind, our PLSR results establish that 16 leaf traits can be remotely estimated in the Lambir canopy (Table 4), including traits characterized in past studies—pigments, N, water, and C fractions such as cellulose and lignin (see reviews by Kokaly and others 2009; Ustin and others 2009). Moreover, we found that about 76 and 33% of species and families, respectively, can be classified with these remotely sensible chemical signatures (Figure S1). The subset of chemical traits most important to developing the classification included C, hemi-cellulose, phenols and tannins (Table S6). Our analyses indicated that these four chemical traits have particularly unique values within the Dipterocarpaceae, as compared to other taxa, hence their important contributions to these combined chemical signatures. And finally, we found that phenols were a key contributor to the remotely sensed classification of species at the Bornean site, highlighting the value of defense compounds in the development of ecological remote sensing approaches for biological diversity.

Conclusions

This study provides new insight into sources of spectral and chemical variation among canopy species in lowland Bornean forests. We drew a connection between chemical and biological diversity, and between chemical and spectral properties among canopies. Our approach is unique because it allows for quantitative linkage between these otherwise disparate areas of study, yet we still view these techniques as needing refinement. Ripe areas for increased effort range from improving models of canopy reflectance based on sunlight foliage and vertical light attenuation gradients, to integrating quantitative remote sensing and phylogenetic methods (for example, barcodes and distance matrices). Such efforts will be essential to upscaling what are ultimately limited field-based measurements to scales commensurate with ecosystem-level processes.

Despite the limitations of our current techniques, we found the chemical diversity of lowland mixed dipterocarp forest to be very high, and that this high chemical diversity is expressed through a suite of elements and compounds regulating growth, defense, longevity, and other vital plant and ecosystem functions. Some of the chemical ranges observed among full sunlight canopies in this single Bornean forest nearly matched the global variability within and among different biomes. However, we also found that much of the chemical diversity resides in the dominant family Dipterocarpaceae, and in the genus Shorea, highlighting the role that a single phylogenetic branch can play in creating chemical variation. These findings suggest that chemical diversity is not simply a function of biological diversity at low taxonomic levels (for example, families or genera present at the site), but rather it reflects the variation among co-existing species. This may indirectly suggest over-dispersion (low phylogenetic signal) of foliar chemical traits (Kursar and others 2009), a topic warranting further research. Independent of the underlying evolutionary causes and the present distribution of species, chemical diversity in canopies such as those found in lowland Borneo are measurable using high-fidelity spectroscopic remote sensing techniques. Through these chemical-to-spectral linkages, studies of functional and biological diversity interactions will become increasingly possible at larger spatial scales, thereby improving our knowledge of how species mediate and are affected by ecosystem and evolutionary processes.

Acknowledgements

We thank D. Knapp, F. Sinca, L. Carranza, C. Anderson, M. Houcheime, K. Smith, and A. Enjah, and colleagues from the Forest Department Sarawak, University Malaysia Sarawak, and University Putra Malaysia for assistance with logistics, field work and laboratory analyses. We thank S. Davies and the Center for Tropical Forest Science (CTFS) for programmatic assistance. The Carnegie Spectranomics Project (http://spectranomics.ciw.edu) is supported by the John D. and Catherine T. MacArthur Foundation, and activities in the field campaign were made possible through the development project of the Ministry of Natural Resources and Environment (NRE), Malaysia.

Supplementary material

10021_2012_9526_MOESM1_ESM.doc (596 kb)
Supplementary material 1 (DOC 596 kb)

Copyright information

© Springer Science+Business Media, LLC 2012