Introduction
Key symbols and abbreviations.
FNQ
Far North Queensland
N
Nitrogen
P
Phosphorus
V
Vegetation type
F
Forest
S
Savanna
Φ
Plant functional type
Amax
Rate of CO2 assimilation, light and CO2 saturated
AN
Photosynthetic nitrogen use efficiency
AP
Photosynthetic phosphorus use efficiency
gs
Stomatal conductance to CO2 diffusion
Jmax
Light saturated potential rate of electron transport
Vcmax
Maximum carboxylation velocity
d
Leaf (lamina) thickness
ρ
Leaf density
Ma
Leaf mass per unit area
ξ
Leaf dry matter content
Subscripted “a”
Per unit leaf area
Subscripted “m”
Per unit leaf dried mass
Forests and savannas dominate the tropical vegetated regions
covering 15–20 % of Earth's surface (Torello-Raventos et al., 2013).
At a broad scale, it has been long recognised that the distribution of these
two biomes, each with its own structural characteristics and species
composition, is principally governed by precipitation and its seasonality
(Schmimper, 1903), but with soil chemical characteristics also important
(Lloyd et al., 2008; Lloyd et al., 2009; Lehmann et al., 2011). Edaphic
conditions are especially influential in regions where the two biomes
intersect – often referred to as “ecotones” or “zones of (ecological)
tension” (ZOT) – both forest and savanna existing as discrete “patches”
under similar climatic conditions (Cochrane, 1989; Ratter, 1992; Thompson et
al., 1992; Hoffmann et al., 2009; Lehmann et al., 2011; Saiz et al., 2012;
Veenendaal et al., 2014). The patchiness of the ZOT mosaic at small spatial
scales has led some to argue that disturbances, principally fire, must
interact with climatic/edaphic boundaries in determining the transition between
the two alternative vegetation types (e.g. Lehmann et al., 2011; Hoffmann et
al., 2012). Whatever the drivers, feedbacks associated with changes to
distributions of these biomes in response to anthropogenic climate change
have the potential to substantially modify the rate of future global warming
(e.g. Malhi et al., 2009).
The underlying causes of variation in photosynthetic carbon acquisition
across and within these two biomes remain, however, poorly understood. There
is, nevertheless, accumulating evidence that for tropical forest species
phosphorus (P) availability may limit photosynthetic rates and productivity
(Vitousek, 1984; Domingues et al., 2010; Mercado et al., 2011; Quesada et
al., 2012). Whereas in savanna ecosystems nitrogen (N) may be more important
as a limiting nutrient (Lloyd et al., 2009). Soils in Australia are generally
highly weathered with the consequence that plant performance, even in the
subtropical and temperate regions, is often considered more limited by the
supply of P than of N (Beadle, 1962, 1966; Webb, 1968). Nevertheless, in Far
North Queensland (FNQ), where almost all the Australian tropical forest
occurs, recent volcanic activity (0.01–4.5 Ma BP – before present) has produced some very
young soils. These young basaltic soils cover about 60 % of FNQ's land
area (Whitehead et al., 2007) and display higher levels of organic matter and
total P when compared with other parent material groupings such as granitic
or metamorphic (Spain, 1990). Direct links from soil P status to measures of
forest productivity are not straightforward, however, and interspecific
variations in P use efficiency are likely to have contributed to the varied
composition of local plant communities (Gleason et al., 2009). Likely
selection pressures, on infertile soils, for enhanced P use efficiency
coupled with FNQ's recent volcanic history mean that primacy for P as
the major limiting nutrient to photosynthetic capacity is still hypothetical.
The forest and savanna vegetation types (Vs) have very few plant
species in common (Torello-Raventos et al., 2013) and the edaphic
determinants of the ZOT are of particular interest in Australia (Beadle,
1962, 1966; Russell-Smith et al., 2004). The savannas of FNQ are distinctive
globally being dominated by eucalypts (Myrtaceae). Here, species of the
closely related genera Eucalyptus and Corymbia are
characterised by sclerophyllous (hard) leaves with relatively low leaf [N],
but a high oil content and correspondingly high heat of combustion (Beadle,
1966) – traits that contribute to a highly flammable leaf litter. In
contrast to the tree species of the moist forests, such evergreen savanna
species are expected to be able to withstand periods of water shortage and
high water vapour pressure deficit. Sclerophylly imparts both structural and
physiological leaf traits but, to date, most research has focused on the
structural aspects: leaf thickness and density combining in the ratio of leaf
mass per unit area (Ma, g m-2). Sclerophyllous leaves are
often amphistomatous (i.e. with stomata abundant on both the adaxial and
abaxial surfaces) displaying an isobilateral mesophyll distribution (Burrows,
2001) – characteristics thought to be associated with both high
photosynthetic potentials (Mott et al., 1982) and high-insolation
environments (Pyykko, 1966; Parkhurst, 1978). Such ecophysiological
associations are expected on theoretical grounds, especially under conditions
of low water availability (Buckley et al., 2002) and it may be that
sclerophyllous eucalypt leaves display an “investment strategy” that at once
combines resilience with high photosynthetic return (Cernusak et al., 2011).
Plot coordinates after Torello-Raventos et al. (2013), vegetation
classification V, elevation a.s.l. EV, mean annual
temperature TA, mean annual precipitation PA,
0.95 quantile upper stratum canopy height H*, upper stratum canopy area
index CU, soil pH, soil exchangeable cations, soil extractable
phosphorus and World Reference Base (WRB) soil classification for the study
site's soil values represent the top 0.3 m of soil.
Plot
Lat
Long
V
EV
TA
PA
H*
CU
pH
[Al]e
[Ca]ex
[K]ex
[Mg]ex
[Na]ex
[P]ex
WRB soil classification
(m)
(∘C)
(m)
(m)
(m2m-2)
(µgg-1)
mmoleqkg-1
CTC-01
16.103∘ S
145.447∘ E
Tall forest
90
25.2
3.20
38.9
2.36
5.56
0.48
17.94
0.71
7.65
0.65
208
Haplic Cambisol (Hyperdystric, Alumic, Skeletic)
KBL-01
17.764∘ S
145.544∘ E
Tall forest
761
20.5
1.75
38.0
1.45
4.79
0.27
4.71
0.35
2.58
0.36
952
Haplic Regosol (Siltic,Hyperdystric)
KBL-03
17.685∘ S
145.535∘ E
Tall forest
1055
19.1
1.34
35.8
2.30
4.38
4.84
0.90
0.47
1.82
0.62
227
Haplic Nitisol (Hyperdystric, Rhodic)
KCR-01
17.107∘ S
145.604∘ E
Tall forest
813
20.5
1.96
44.0
2.21
5.40
0.60
16.11
0.78
5.71
0.39
165
Haplic Cambisol (Dystric, Alumic)
DCR-01
17.026∘ S
145.597∘ E
Tall savannawoodland
683
21.2
1.45
26.2
1.63
5.65
0.90
8.78
0.71
6.93
0.71
79
Haplic Cambisol(Orthodystric, Alumic)
DCR-02
17.021∘ S
145.584∘ E
Tall savannawoodland
653
21.3
1.46
22.1
0.70
5.52
1.27
8.82
0.55
5.74
0.33
56
Arenic Cambisol(Epieutric)
KBL-02
17.849∘ S
145.532∘ E
Tall savannawoodland
860
20.1
1.43
28.1
0.77
5.28
0.26
4.78
0.17
7.66
0.77
216
Geric Acrisol(Hyperdystric, Rhodic)
Broad overlap in leaf traits has been reported for savanna and forest tree
species, but, due to different soils and the drier conditions typical of
savannas, we might expect differences between the two vegetation types in
leaf N and P content, rates of photosynthesis, morphology and longevity
(Wright et al., 2001; Buckley et al., 2002; Meir et al., 2007; Domingues et
al., 2010). It remains unclear, however, whether such differences persist
within the relatively narrow boundaries of a ZOT. In addition, within
individual tropical forest stands, leaves can vary markedly in their
physiological and structural properties depending on canopy position and the
availability of gaps (Popma et al., 1992; Lloyd et al., 2010). Indeed
tropical forest tree species are often grouped according to their degree of
shade tolerance and/or light requirement (e.g. Swaine and Whitmore, 1988; but
see Poorter, 1999). Shade-tolerant species in the understory may receive less
than 2 % of full light at the canopy crown (Chazdon, 1992) whilst pioneer
species typically require high light exposure for germination and survival and depend
on the creation of canopy gaps (Turner, 2001). As a result, species
characteristic of differing light niches have commonly been considered to
display distinctive photosynthetic traits linked to nutrient investment,
allocation and leaf architecture (Niinemets, 1997; Carswell et al., 2000;
Wright et al., 2001). In the context of P availability, a study in the forest
of Guyana, on relatively infertile Ferralsols, found that pioneer tree
species exhibited higher photosynthetic nitrogen and phosphorus use
efficiencies than neighbouring climax species (defined here as those species
whose seeds can germinate and establish in the shade) (Raaimakers et al.,
1995). Fyllas et al. (2012), in describing forest tree species of the Amazon
Basin, derived four discrete PFTs (plant functional types) aligned with the species' stature, canopy
position and pioneering ability. It remains to be seen whether such an
attractively simple system can be applied to tropical forest species of FNQ.
In this study we contrast leaf photosynthetic traits for tree species from
forest and savanna communities of northern Australia addressing the
following questions.
Are there differences in photosynthetic capacity and nutrient use efficiency
between adjacent forest and savanna vegetation types?
And, if so, are these distinctions associated with systematic differences in
leaf structural traits?
Is there evidence of a greater role for P rather than N (or vice versa) in
determining photosynthetic capacity across both sites and species?
Can a simple classification system based on light requirement and adult
stature help to describe observed variation in photosynthetic traits of
tropical forest trees?
Materials and method
Sites and species
A series of sites was selected in FNQ, Australia, in an arc from the Atherton
Tablelands, inland from Cairns, to Cape Tribulation, north of the Daintree
River. The series, which forms part of the Tropical Biomes in
Transition (TROBIT) network, was designed to provide a contrast of
vegetation types, specifically forest (F) and savanna
(S), and are located on diverse soils (Table 2). Further
descriptions of all sites and the rationale (both structural and floristic)
for our distinction between F and S are available in
Torello-Raventos et al. (2013). Seven sites were visited in 6 weeks of
fieldwork during April and May 2009 and measurements were performed on 125
trees representing 30 species. A full list of species by site is presented in
Table S1 (Supplement).
Gas exchange measurements
Leaf gas-exchange measurements were performed using a portable photosynthesis
system (Li-Cor 6400, Li-Cor, Lincoln, NE, USA) on young, fully developed
leaves. During the measurements, chamber conditions were set with block
temperature (mean 27 ∘C) held slightly above ambient air temperature
to avoid problems of condensation; relative humidity remained close to
ambient (mean = 67 %). The rate of air flow to the sample cell was
held constant at 500 µmol s-1 but, exceptionally, when
faced with very low stomatal conductance this was reduced (minimum
250 µmol s-1). Light (A↔Q) curves were
generated for each tree species to determine the saturating light level for
adoption in subsequent CO2 response curves (Aa↔Ci curves). Those saturating light levels ranged from 500 to
2000 µmol m-2 s-1. Measurements of light-saturated net
CO2 assimilation per unit leaf area (Aa) were then obtained
for a range of intercellular CO2 concentrations (Ci) by
varying chamber CO2 concentration (Ca). The
Aa↔Ci curves provided area-based values
of light-saturated photosynthesis under ambient and elevated atmospheric
[CO2] (Asat,a and Amax,a, respectively). For
the purposes of modelling photosynthetic capacity we focus on variations in
Amax – preferred over Asat in this context as it is less
susceptible to limitations of stomatal conductance (gs).
In the absence of cranes or suspended walkways, branches had to be cut from
trees. Sun-exposed branches for short trees were reached using handheld
secateurs or forestry shears on telescopic poles; for taller tree branches
were pulled down using a weighted line shot from a catapult. Trees of
subcanopy species were rarely found growing in full sunlight; therefore, their
leaves, although sampled from upper branches free of self-shading,
developed in a relatively low-light environment. Once detached, the stem was
recut under water in order to re-establish the xylem water column (Domingues
et al., 2010). Performing gas exchange measurements on excised branches can
affect subsequent calculations where stomatal conductance is heavily
depressed (Santiago and Mulkey, 2003). The Aa↔Ci curves were reviewed for such instances and where necessary the
data excluded from all further analysis (n=11). A further check on data
integrity proposed by Kattge et al. (2009) rejects those measurements where
Asat / [N]m is < 2 µmol
CO2 g-1N s-1; any such curves were likewise excluded
(n=2).
Leaf morphology and nutrient analysis
At the completion of gas-exchange measurements, the leaf (gas leaf) was cut
from the branch and leaf thickness (d, µm) taken by averaging
repeated measurements (Mitutoyo dial thickness gauge, n=6) alternating back
and forward across the mid-vein and proceeding down the lamina from tip to
base. A series of discs (6.6 mm, diameter) was then punched from the leaf
avoiding veins and necrosis or other damage. The discs with the remainder of
the leaf (petiole and mid-vein discarded) were oven dried at 70 ∘C
for a minimum of 48 h before their dried mass was recorded. The combination
for the discs of known area and dried mass allowed calculating leaf
mass per unit area (Ma, g m-2). Leaf density (ρ,
g cm-3) was estimated using the equation
ρ=Ma/d.
In addition to the gas leaf,
the opposing leaf was also cut from the branch, petiole and mid-vein
discarded, and placed in a ziplock plastic bag with moist cotton wool until
fresh mass could be measured that evening (or exceptionally the next day).
The leaf was then placed in an envelope, oven-dried as above and dried mass
recorded. The ratio of the leaf's dried to fresh mass is termed leaf dry
matter content (ξ). All subsequent references to ξ relate to
opposing and not gas leaves. Logistical constraints imposed by repeated
changes of base camp and lack of electricity supply meant that delays were
experienced between harvesting the leaves and oven drying (maximum delay of 30
days).
Oven-dried material was used for determination of total leaf [N] and [P]:
dried ground leaf material was acid-peroxide digested before colorimetric
analysis using a segmented flow analyser (Skalar San+System, Breda,
the Netherlands). The photosynthetic efficiency of nutrient use was
estimated as the maximal rate of carbon gain per unit of leaf N and P
(AN and AP, respectively).
Statistical analyses
All statistical analysis and modelling was conducted using the open-source
statistical environment R (R Development Core Team, 2011). As initial data
exploration revealed wide variation in many trait values across the different
sites, non-parametric Kruskal–Wallis tests were used to test for differences
among the categorical factors of site and V using the coin package (Hothorn
et al., 2008). Where significant, differences among factor levels were
assigned using Tukey's HSD (honest significant difference) post hoc tests (p< 0.05) applied to data
rankings. After exclusion of poor Aa↔Ci curves (n=13, described above) and replicates with other
missing values (n=3), the revised data-set of 109 leaf measurements
contained many more observations for F (n=85) than S (n=24); therefore,
there is an element of imbalance in the test specification where V is
adopted as the fixed factor. Bivariate relationships were described using
standardised major axis (SMA) line fits using smatr-3 (Warton et al., 2006).
Relationships between replicated foliar traits (photosynthesis, nutrient
content) and site-dependent variables (soil, climate) were quantified using
Kendall's non-parametric rank-order correlation (tau, τ); especially
appropriate in cases with many replicated response values for each value of
the predictor variable (Legendre and Legendre, 2012).
Paired boxplots of key leaf traits (untransformed data) by site and
V. The two V classes are
distinguished by colour: green for forest and brown for savanna. Site
abbreviations are laid out in Table 2. Leaf traits are photosynthetic
capacity (a, b) per unit leaf area and (c, d) per unit leaf
dried mass; (e, f) leaf mass per unit area; (g, h) leaf
dried matter content; (i, j) leaf thickness; (k, l) ratio
of total leaf nitrogen to phosphorus; total leaf nitrogen (m, n) per
unit leaf area and (o, p) per unit leaf dried mass; total leaf
phosphorus (q, r) per unit leaf area and (s, t) per unit
leaf dried mass. The box and whiskers show the median result as a thick
horizontal band, the ends of the box denote the interquartile range; the
whiskers extend 1.5 times the interquartile range or to the most extreme
value, whichever is smaller; any points outside these values are shown as
outliers. The grey dashed line in plot k represents a mean N : P ratio of
18.8 reported for tropical forests by Reich et al. (2009).
Mixed-effects linear model of photosynthetic capacity
The study involved replicated measurements of tree species within and across
forest and savanna plots. Such a design introduced the strong likelihood that
measurements within the same site would be influenced by spatial proximity.
In specifying a model that attempted to explain differences in photosynthetic
capacity between V, it was necessary to recognise this hierarchical
structure in order to avoid systematic variation in the residuals leading to
potentially biased interpretation (Zuur et al., 2009). The sites and species
selected, rather than considered of primary interest per se, were viewed as
representative of a wider population and focus was placed on their variance. The
model's random component therefore included the categorical variables of
species nested within sites. Unfortunately, because not all tree species at
all sites were measured with replication (see instances of n=1 in Table S1), convergence problems meant that the random component of the model
could not accommodate differing slopes as well as intercepts for species
within a site.
The final model (fitted using the nlme package in R) may be expressed as
Amax,a[ijk]=α+β1bjk+β2[N]a[ijk]+β3[P]a[ijk]+ak+aj|k+εijk.
Here response variable Amax,a[ijk] denotes the maximum rate of
area-based photosynthesis for observation i of species j at site k with
b being a categorical variable taking a base value of 0 for species in
plots classified as F and a value of 1 for
S. The term ak is a random intercept and allows for variation among
sites. The term aj|k allows for interspecific variation at the same
site. The term εijk is the residual (unexplained) error and
represents the within-site variation, i.e. variation among plants of the same
species and measurement error. Each of the variation terms is assumed to be
normally distributed with mean zero. With the independent covariates centred
(i.e. zeroed on the population mean), the fitted intercept term α
thus represents the predicted forest tree Amax,a at the (F+S)
population mean [N]a and [P]a. The term β1b
represents the difference in predicted Amax,a between the two V values (in
this case Amax,a[S]-Amax,a[F]). The predicted S tree
Amax,a at the population mean [N]a and [P]a
values is therefore α+ β1.
Scatterplots of the relationships between leaf phosphorus and leaf
nitrogen (a) on an area basis and (b) on a mass basis. Plot
(c) shows the relationship between leaf nitrogen on an area basis
and leaf mass per unit area; plot (d) shows the equivalent
relationship for leaf phosphorus. Each point corresponds to a single tree and
vegetation types are distinguished by colour: green for F and
brown for S. SMA-fitted lines are shown
for the two vegetation types only where the bivariate relationship proved
significant p<0.05. Pearson correlations testing the assumption of
linearity are given in Table S4 together with the likelihood ratio and Wald
statistics testing the H0 of common slope, elevation and axis shift for
the two V classes. Intercept, slope and r2 values for the
SMA-fitted lines are given in Table 3. In plot (a) a third fitted line
(grey, dashed) displays a slope based on a mean N : P ratio of 18.8 typical
for tropical forests as reported by Reich et al. (2009) and passing through
the origin. In plot (b) there is a common slope to the
[P]m ↔ [N]m relationship for both
vegetation types (black line). In reviewing residual plots of initial SMA
fits (not shown), four data points were identified as outliers (crossed
circles). The four outliers have been excluded from the rerun SMA fits shown
here.
Plant functional types
F species were assigned to one of four plant functional types (Φ)
depending on their adult stature and light requirements for recruitment
(Veenendaal et al., 1996). Three of the authors (D. M. Crayn, A. Ford and D. J. Metcalfe), each with
an extensive knowledge of Australian tropical forest trees, made independent
Φ assignments before the combined results were consolidated and minor
discrepancies resolved. The Φ descriptions are provided in Table S2
and the relevant species designations in Table S1. All of the
tropical moist forest species in this study are obligate
evergreen.
Results
Key leaf traits: forest versus savanna trees
There was a tenfold range across the data-set in photosynthetic capacity per
unit leaf area (Amax,a) from 4.9 to
52.0 µmol m-2 s-1 (Fig. 1a). Mean values (treating each
sampled tree as an independent variable) differed among sites
(p < 0.0001) and were significantly higher in the S plots
(Fig. 1b; p < 0.0001). When expressed per unit leaf dried mass,
photosynthetic capacity (Amax,m) was also highly variable with
significant plot-to-plot differences (Fig. 1c; p < 0.0001). Overall
there was, however, no difference in mean Amax,m between
F and S (Fig. 1d; p=0.11). There were striking
intersite differences in leaf mass per unit area (Ma; Fig. 1e)
which was highly variable, spanning a fivefold range from lowest to highest
observations. Overall, Ma was higher for S than for
F (Fig. 1f; p < 0.0001). That contrast in Ma
derived chiefly from variable leaf dry matter content ratios (ξ) with a
threefold variation in ξ observed. Of note were pronounced differences
among sites (Fig. 1g; p < 0.0001) and, like Ma, higher mean
values for S than for F (Fig. 1h; p= 0.0035). By
contrast, there was no difference between S and F for
measures of leaf thickness (d) (Fig. 1j; p= 0.70) which was also much
less variable.
Both area- and mass-based leaf nitrogen levels were highly variable and this
was most pronounced within the F plots. On an area basis,
differences among sites (Fig. 1m; p < 0.0001) produced higher mean
[N]a values for S than for F (Fig. 1n;
p= 0.0002) with this intersite N variability even more pronounced on a
mass basis (Fig. 1o; p < 0.0001). Overall, mean [N]m was
higher for F than for S (Fig. 1p; p < 0.0001).
Broadly similar trends were observed for total leaf phosphorus with higher
mean [P]a for S versus F (Fig. 1r;
p= 0.003), with that ranking also reversed when expressed on a mass
basis (Fig. 1t; p= 0.0004). Foliar ratios of [N] : [P] ranged from 10.1
(Neisosperma poweri, KBL-01) to 39.1 (Symplocos hayesii,
KBL-03) but, on average, did not differ across plots (Fig. 1k; p= 0.29) and
showed no systematic differences between F and S
(Fig. 1l; p= 0.74).
Considering the data-set as a whole (i.e. F and S trees
combined), significant correlations of Amax,a with
environmental variables such as elevation, temperature and soil cation status
were found. However, these are mirrored by significant correlations of the same
sign for both leaf [P]a and [N]a (Table S3). In
investigating the underlying sources of our data-set's tree-to-tree variation
in photosynthetic properties, we therefore focussed (using the mixed-effects
model) on associated tree-to-tree variations in leaf-based nitrogen and
phosphorus concentrations; checking for any edaphic or climatic effect beyond
that through an examination of model residuals in relation to the
site-associated climate and soil covariates.
Coefficients for SMA bivariate
relationships. Vegetation contrasts: forest and savanna; n, number of
observations; r2, correlation coefficient and associated p value;
intercept; slope and 95 % confidence interval (CI).
Response
Bivariate
Vegetation
n
r2
p
Intercept
Slope
Slope:
Slope:
class
low 95 % CI
high 95 % CI
[P]a
[N]a
Forest
81
0.82
< 0.0001
-11.01
47.92
43.58
52.68
Savanna
24
0.51
< 0.0001
-70.96
75.86
55.87
102.99
[P]m
[N]m
All
105
0.81
< 0.0001
-0.097
0.046
0.042
0.050
[N]a
Ma
Forest
81
0.43
< 0.0001
0.299
0.018
0.015
0.021
Savanna
24
0.08
0.1848
[P]a
Ma
Forest
81
0.38
< 0.0001
3.31
0.84
0.71
1.00
Savanna
24
0.00
0.8207
Amax,a
[N]a
Forest
81
0.47
< 0.0001
-11.43
17.01
14.46
20.00
Savanna
24
0.17
0.0442
-54.48
42.15
28.47
62.40
Amax,a
[P]a
Forest
81
0.47
< 0.0001
-7.52
0.35
0.30
0.42
Savanna
24
0.09
0.1522
Amax,a
Ma
All
105
0.30
< 0.0001
-6.03
0.29
0.25
0.34
Amax,m
[N]m
Forest
81
0.63
< 0.0001
-192.81
18.47
16.12
21.16
Savanna
24
0.31
0.0049
-100.47
21.87
15.26
31.35
Amax ,m
[P]m
Forest
81
0.61
< 0.0001
-148.85
404.67
352.06
465.15
Savanna
24
0.30
0.0061
-14.60
394.68
274.50
567.47
Amax ,m
Ma
Forest
81
0.08
0.0132
607.15
-4.95
-6.13
-4.00
Savanna
24
0.30
0.0056
628.00
-2.73
-3.93
-1.90
Vcmax ,a
[N]a
Forest
81
0.41
< 0.0001
-50.19
59.99
50.57
71.17
Savanna
24
0.23
0.0168
-183.24
142.76
97.82
208.35
Vcmax ,m
[N]m
Forest
81
0.66
< 0.0001
-0.94
0.071
0.063
0.081
Savanna
24
0.35
0.0025
-0.35
0.075
0.053
0.107
Vcmax ,a
[P]a
Forest
81
0.41
< 0.0001
-36.40
1.25
1.06
1.48
Savanna
24
0.17
0.0471
-49.69
1.88
1.27
2.79
Vcmax ,m
[P]m
Forest
81
0.65
< 0.0001
-0.77
1.56
1.37
1.78
Savanna
24
0.35
0.0022
-0.05
1.36
0.96
1.92
Jmax ,a
[N]a
Forest
81
0.51
< 0.0001
-52.66
76.12
65.16
88.92
Savanna
24
0.15
0.0660
Jmax ,m
[N]m
Forest
81
0.65
< 0.0001
-0.89
0.083
0.073
0.095
Savanna
24
0.37
0.0017
-0.38
0.094
0.066
0.133
Jmax ,a
[P]a
Forest
81
0.53
< 0.0001
-35.17
1.59
1.36
1.85
Savanna
24
0.08
0.1873
Jmax ,m
[P]m
Forest
81
0.65
< 0.0001
-0.69
1.82
1.59
2.07
Savanna
24
0.33
0.0033
-0.01
1.69
1.19
2.41
φ
Ma
Forest
85
0.40
< 0.0001
-0.063
0.009
0.007
0.010
Savanna
24
0.30
0.0056
0.086
0.005
0.003
0.007
d
Ma
Forest
82
0.26
< 0.0001
107.77
3.91
3.23
4.72
Savanna
24
0.74
< 0.0001
152.77
1.86
1.48
2.32
ρ
Ma
Forest
82
0.45
< 0.0001
0.040
0.0020
0.0017
0.0024
Savanna
24
0.66
< 0.0001
0.141
0.0014
0.0011
0.0018
AN
Ma
Forest
81
0.00
0.6122
Savanna
24
0.00
0.9268
Top panels (plots a, b and c):
scatterplots of the area-based relationships between photosynthetic capacity
(Amax,a) and (a) leaf nitrogen, (b) leaf
phosphorus, and (c) leaf mass per unit area (Ma). Bottom
panels (plots d, e and f): the equivalent plots are
expressed per unit leaf dried mass. Each point represents a single leaf; dots
denote F sites and square symbols denote S sites;
individual sites are distinguished by colour: CTC-01 black, DCR-01 red,
DCR-02 green, KBL-01 royal blue, KBL-02 turquoise, KBL-03 pink, and KCR-01
yellow. SMA-fitted lines are shown where
significant: F (green), S (brown). Pearson
correlations testing the assumption of linearity are given in Table S4
together with likelihood ratio and Wald
statistics testing the H0 of common slope, elevation and axis shift for
the two V classes. In plot (c) there was no difference in
slope between the two Vs and so a common line is fitted (black).
SMA intercept, slope and r2 values are given in Table 3.
Leaf nutrient relationships
There was a strong positive linear relationship between leaf [P]a
and [N]a for both F and S (Fig. 2a), but
with a steeper slope observed for S. The shallower slope for
F differs (p=0.031) from the mean relationship for tropical
forests as suggested by Reich et al. (2009), depicted by the dotted line
passing through the origin in Fig. 2a (slope = 103/18.8). When
expressed on a mass basis, there was a single, common P ↔ N relationship for both Vs (Fig. 2b). Leaf nutrient investment
on an area basis showed positive relationships with Ma for
F only (Fig. 2c, d).
To test for differences in the photosynthesis ↔ nutrient relationships between
the two Vs, a series of SMA analyses was
undertaken with photosynthetic capacity (Amax) as the response
variable and leaf chemistry ([N] and [P]) as the explanatory bivariate
(Table 3). For the combined data-set, linear relationships were strong for
both nutrients irrespective of whether variables were expressed on a mass or
area basis (r values ranging from 0.63 to 0.70; Table S4).
The Amax,a ↔ [N]a association as
shown in Fig. 3a suggests two important differences between the two
Vs. First, across the (pooled) data-set the lowest 0.3 fraction
of [N]a is confined to F-associated trees (as can also
be inferred from Figs. 1 and 2). Second, for the lowest [N]a for
S-associated trees (ca. 1.6 g m-2), similar
Amax,a values are observed for both S and F, but
as [N]a increases beyond that point Amax,a for
S rises with a sensitivity nearly 3 times that observed for
F. There was no difference between the two Vs in either
the slope or the intercept of the Amax,a↔Ma association (Table S4) and a single line (r2 = 0.3)
describes the common positive relationship (Fig. 3c).
Boxplots of photosynthetic N use efficiency (AN) and P use efficiency (AP) by ϕ. Boxes which share the same letter
correspond to means that were not significantly different (Tukey's HSD on
ranked values). Boxplot construction is explained in Fig. 1.
Stepwise selection process for the fixed component of the linear
mixed-effect model: photosynthetic capacity (Amax,a) as
response variable. Categorical variable b has two levels: forest and
savanna for the contrasting vegetation types. Continuous explanatory
variables are [N]a, [P]a, leaf mass
per unit area (Ma), leaf dry matter content (ξ) and leaf
thickness (d). The effect of dropping sequential terms was tested by
comparing the nested model variants. Model variants were all run using the
maximum likelihood method; the model's random component was identical in all
variants. Test parameters and statistics are df, degrees of freedom; AIC,
Akaike information criteria; BIC, Bayesian information criteria; logLik,
maximum likelihood; the likelihood ratio statistic and associated p value.
Models with the same degrees of freedom are not nested one in the other.
Model
Fixed component
df
AIC
BIC
logLik
Test
L. ratio
p value
1
b + [N]a+[P]a+Ma+d+ξ
11
735.51
765.12
-356.76
2
b + [N]a+[P]a+Ma+d
10
733.52
760.44
-356.76
1 vs. 2
0.011
0.915
3
b + [N]a+[P]a+Ma
9
731.94
756.16
-356.97
2 vs. 3
0.417
0.519
4
b + [N]a+[P]a
8
730.58
752.11
-357.29
3 vs. 4
0.640
0.424
5
b + [N]a
7
732.96
751.80
-359.48
4 vs. 5
4.383
0.036
6
[N]a+[P]a
7
738.39
757.23
-362.20
4 vs. 6
9.814
0.002
7
[N]a+[P]a+Ma+d+ξ
10
742.94
769.86
-361.47
1 vs. 7
9.432
0.002
8
b + [N]a+[P]a+ b : [N]a+ b : [P]a+[N]a : [P]a+ b : [N]a : [P]a
12
734.73
767.02
-355.36
4 vs. 8
3.852
0.426
9
b + [N]a+[P]a+ b : [N]a+ b : [P]a
10
733.15
760.07
-356.58
7 vs. 9
2.426
0.297
10
b + [N]a+[P]a+ b : [N]a
9
731.37
755.59
-356.68
8 vs. 10
0.213
0.645
11
b + [N]a+[P]a+ b : [P]a
9
731.49
755.71
-356.75
8 vs. 11
0.339
0.561
12
b + [N]a+[P]a+[N]a : [P]a
9
732.53
756.75
-357.26
4 vs. 12
0.050
0.823
Nutrient use efficiency and leaf structure
Elevated rates of photosynthesis per unit N yielded higher photosynthetic use
efficiencies (AN) for S species (Fig. 4a). Of the
F trees only the tall pioneers showed an equivalent AN
to the S species and there was a significant difference between
tall pioneers and shade-tolerant species. Very similar patterns were observed for AP (Fig. 4b).
Scatterplots of the relationship with leaf mass per unit area for
(a) leaf thickness (measured fresh), (b) leaf
density (derived as Ma/d), (c) the ratio of leaf dry
mass to water content and (d) photosynthetic N use efficiency. Each
point represents one tree and separate ϕ values are distinguished by colour.
SMA-fitted lines are shown in the top panel for the
two vegetation types: F in green and S in brown.
Outlying values for Wilkiea angustifolia (crossed circles) were excluded from
the SMA analyses of plots (a) and (b). In plot (c)
a separate fitted line is shown for each ϕ. Pearson's correlations
testing the assumption of linearity are given in Table S4 together with
the likelihood ratio and Wald statistics testing the H0 of common slope,
elevation and axis shift for the two V classes. SMA intercept,
slope and r2 values are given in Table 3. There was no relationship
AN↔Ma for individual levels of
V or ϕ.
The higher Ma values for the S trees (Fig. 1f)
suggested underlying differences in leaf structure between the two
Vs and the range of Ma values for separate Φ
classes showed distributions centred at different points along the
Ma axis (Fig. S1). Whilst we found positive relationships for
Ma with d and ρ, the slopes of those relationships were
V-dependent (Fig. 5a, b) and the association was much stronger
for the S group. Over common ranges of Ma, the
F trees displayed the thicker leaves (measured fresh) – a result
heavily influenced by the upper canopy group. Indeed, Ma was
seldom greater than 75 g m-2 for either the small pioneer or
subcanopy classes but, for any given Ma, the ratio of leaf dry
mass to water content (φ, as an index of sclerophylly) was much
higher for the small pioneers (Fig. 5c). The slope of the relationship
φ↔Ma also differed among the Φ
classes (p < 0.0001) – being shallowest for the S trees (an
increased ratio of 0.005 per unit Ma) and steepest for the small
pioneers (an increase of 0.019). Such structural differences among classes of
V and Φ appeared unrelated, however, to our measure of
photosynthetic N allocation: AN was independent of d (p=0.46)
and an overall correlation with Ma (p= 0.0009) disappeared
upon controlling for V and Φ (Fig. 5d).
Modelling photosynthetic capacity
In attempting to model variation in Amax,a the starting, or
maximal, fixed component of our linear mixed-effects model (model 1; Table 4)
included, along with a V-dependent term, those continuous
variables spanning leaf morphology and chemistry suggested by pair-wise
correlation plots in Fig. S2. The optimal fixed term, on AIC and
likelihood ratio criteria, was provided by model 4: with vegetation type a
fitted categorical variable (through the β1b term) along with the
continuous variables [N]a and [P]a. Model performance
was not improved by adding interaction terms (see model variants 8–12; Table 4). Also of note is model 6, which in dropping the vegetation term
(b) produces a significantly inferior model (p= 0.002). That is to say,
we could not account for the intrinsically higher area-based photosynthetic
capacities of S-affiliated trees (Fig. 1b) through systematic
F-S differences in any of our measured foliar traits.
Comparisons against a model variant (model 7; Table 4), excluding the
vegetation term, but reinstating available traits relating to leaf chemistry
and structure confirmed that vegetation type could not be substituted in this
way.
The full model, fixed and random terms combined, explained 82 % of
variation in the observed values of Amax,a and with all four
forest Φ values reasonably well predicted (Fig. 6a). A plot of model residuals
against fitted values raised no concerns for the model assumptions (Fig. S3).
These model results also suggest, other things being equal, that
photosynthetic capacities are intrinsically higher for S than
for F species (Fig. 1b) – with estimated values in the data-set trait
means of 29 and 18 µmol CO2 m-2 s-1 (Table 5).
Despite the relatively low t value for the fitted [P]a term,
the predictive power of the overall model was improved by retaining
[P]a as a fixed term (compare models 4 and 5 in Table 4). The
greater significance attached by the final model to the [N]a term
is confirmed when the two slope coefficients are standardised to allow for
the difference in underlying units: β2′ [N]a=0.294 and
β3′ [P]a=0.172 (interpreted as the relative effect on
Amax,a of 1 standard deviation change in the independent
variable).
Model output indicated that, after controlling for vegetation type (b) and
leaf nutrient levels, less than 10 % of variation in Amax,a
was attributable to site effects. The environmental influence on
photosynthetic capacity noted above (Table S3) was, however, adequately
captured by our mixed model's fixed term (which incorporates leaf N and P),
as shown by the absence of any relationship between model residuals and those
same site variables relating to climate and soil conditions (Table S5,
Fig. S4).
Broadly similar results were obtained when the mixed modelling approach was
repeated for mass-based leaf traits (Table S6). The fixed component of the
final parsimonious model again included vegetation type (b),
[N]m and [P]m, but this time with a significant N × P
interaction. The preferred random term was unchanged from that presented
above.
Output from linear mixed-effects model (Eq. 2): (a)
scatterplot of observed photosynthetic capacity (Amax,a)
against the model-fitted values as an indication of goodness of fit and
(b) boxplot of model random intercepts (Amax,a ↔ [N]a) by ϕ for the F subset;
boxes which share the same letter correspond to means that were not
significantly different (Tukey's HSD on ranked values). Boxplot construction
is explained in Fig. 1.
Output of the linear mixed-effects model (Eq. 2): fixed effects
(top) and random effects (bottom). The top section shows fixed-effect
parameter estimates and associated standard error (SE), degrees of freedom, test
statistic and associated p value. The final “optimal” model is compared
against a simpler “null” model that includes only vegetation type (b) in
the fixed component, but has an identical random term of species nested
within site. Continuous explanatory variables were centred on their
respective means (i.e. zero reset to the trait average).
Final model
Null model
Fixed effect
Estimate
SE
df
t value
p value
Estimate
SE
df
t value
p value
Forest (if other variables were zero)
18.06
1.57
73
11.537
< 0.0001
17.08
2.47
75
6.925
< 0.0001
Savanna (vegetation contrast)
β1
11.18
3.08
5
2.076
0.0151
14.15
4.54
5
3.119
0.0263
Leaf [N]a
β2
6.66
2.07
73
3.472
0.0020
Leaf [P]a
β3
0.07
0.03
73
1.618
0.0502
Random effect
Variance
% of total
Variance
% of total
Intercept variance: among sites
ak
3.56
9.4 %
12.06
13.9 %
Intercept variance: among species
aj|k
26.77
70.8 %
66.39
76.2 %
Residual (within species, within sites)
εijk
7.49
19.8 %
8.64
9.9 %
37.82
100.0 %
87.10
100.0 %
AIC
726.7
746.5
Likelihood ratio test
-355.4
-367.2
Plant functional types
The area-based model's nested random component, which recognises the grouping
of species within sites, allows the variance of the response
(Amax,a) to be partitioned among the available terms. The
fraction attributable to variance among plots (9 %) is dwarfed by that
among species (71 %). Variation in Amax,a around the
population mean was therefore influenced much more by a tree's taxonomy than
the plot in which it was growing. One factor that may help explain this
interspecific variation is plant functional type (Φ, related to a
tree's growth strategy and light requirements as described in Table S2). We
examined model output for any Φ-related pattern in random intercepts
for the different species. We found that tall pioneers displayed higher
intercepts than subcanopy species (p= 0.0326) as is shown in Fig. 6b.
Thus, at any given [N]a and [P]a tall pioneers
typically achieved a higher Amax,a than shade-tolerant forest
trees – as confirmed by their higher AN and AP
(Fig. 4). Small pioneers and shade-tolerant canopy species were intermediate
between these two extremes and showed intercepts close to the population
mean.
Discussion
The main aim of this study was to compare photosynthetic traits for the tree
species typical of adjacent tropical moist forest and savanna plots – a
dynamic boundary potentially sensitive to changes in global climate (Sitch et
al., 2008; Booth et al., 2012; Gloor et al., 2012; Huntingford et al., 2013).
Our findings include higher photosynthetic capacity and nutrient use
efficiencies for the savanna species, but our prediction of a primary
photosynthetic role for P rather than N across the forest sites was not
supported. Our preferred area-based model retained only three fixed terms,
vegetation type, leaf N and P, yet accounted for 82 % of variation in
Amax,a. Model output revealed a stronger relationship
of A ↔ N than of A ↔ P and found the variability
among species much more pronounced than among sites. For F there
was qualified support for the expectation that pioneer species would show
higher photosynthetic traits of Amax,a and AN
compared to late successional shade-tolerant species (Raaimakers et al.,
1995).
Forest and savanna comparisons
Values reported here for key leaf traits such as Amax,a,
Ma and levels of foliar N and P fell within previously published
ranges for F and S trees (e.g. Medina, 1984; Prior et
al., 2005; Harrison et al., 2009; Cernusak et al., 2011). There were,
however, significant differences among sites and between F and
S in all these traits (Fig. 1). In particular, a recently
cyclone-affected F site south of Ravenshoe (KBL-01) stood out as
high in leaf nutrients and photosynthetic capacity when expressed on a mass
basis. Due to lower Ma, however, that prominence was all but
removed when area-based traits were examined.
Our study included measurements of 30 tree species across seven sites; these
sites and species were viewed as representative of wider populations and our
modelling treatment of those terms therefore focused on their influence on
the variance of the photosynthetic response rather than mean values. The
linear mixed-effects model (Table 5), through its random component of species
nested within sites, showed that most of the variance in the data occurred
among species (71 %). Once levels of leaf N and P had been included in
the model, variation among sites represented less than 10 % of total
variation. This corresponds with the findings of other Australian studies
where within-site variation has proved much larger than that across sites
(e.g. Wright et al., 2004; Asner et al., 2009). For this study, it could be
argued, however, that the climatic and topographical ranges spanned by the
seven sites (Table 2) were rather modest – mean annual precipitation, for
example, is nowhere lower than 1.3 m and the range in mean annual
temperatures is only 6 ∘C.
Lower mass-based leaf nutrient values for S species have
traditionally been linked to their higher Ma associated with
contrasts in leaf longevity and economic strategy. In the Australian
literature, these species are widely described as sclerophyllous,
characterised by tough leaves and adaptations to limit water loss. We argue,
however, that on theoretical grounds it is area- rather than mass-based
concentrations of N (and presumably also P, where relevant for
photosynthetic
carbon gain) that should be modulated by differences in water availability
(Buckley et al., 2002). With declining precipitation, therefore, an increase
in area-based photosynthetically important nutrients (in our case S>F) seems to be the general case (Buckley et al., 2002; Cernusak et
al., 2011; Domingues et al., 2014; Schrodt et al., 2014). It is
non-systematic variations in Ma with precipitation, in turn
probably attributable to differences in rainfall seasonality, that produce
contradictions in mass-based N ↔ precipitation relationships
(Schrodt et al., 2014).
Linking leaf structure to metabolism
At the leaf level, AN is dependent upon a number of factors
including N allocation, conductance and Rubisco kinetics and no single cause
has been found to explain observed interspecific differences (Hikosaka, 2004;
Hikosaka and Shigeno, 2009). The idea that species with high Ma
exhibit low AN due to greater structural investment (e.g.
Takashima et al., 2004) has been countered by later studies that found no
relationship between AN and the proportion of leaf N allocated to
cell walls (Harrison et al., 2009; Hikosaka and Shigeno, 2009). Indeed, our
general positive association between area-based photosynthetic capacity and
Ma – also observed by Domingues et al. (2014) – challenges the
general notion that thick sclerophyllous leaves should be characterised by
low photosynthetic rates and/or low photosynthetic nutrient use efficiencies
(Wright et al., 2004; Westoby and Wright, 2006). Certainly, it has long been
known that typically sclerophyllous eucalypt species can have exceptionally
high photosynthetic rates (Larcher, 1969), with Denton et al. (2007) also
finding very high nutrient use efficiencies for numerous xeromorphic
Proteaceae species that exhibit some of the very highest Ma
values worldwide. Maximov (1929) noted that “the drier the habitat, the smaller and
more leathery the leaves, and the higher their rate of transpiration”.
Interestingly, our best-fit photosynthesis model was not improved by the
inclusion of morphological traits such as Ma, leaf thickness or
the ratio of leaf dry to fresh mass. Although Ma was much higher
for the S plots, there was no difference in leaf thickness between
S and F sites (p= 0.95), suggesting that most of the
difference in Ma between the two vegetation types was
attributable to a higher leaf density for S. But, as noted above,
with no adverse effects on photosynthetic nutrient use efficiencies.
The role of phosphorus
It has long been considered that vegetation differentiation in Australia is
strongly influenced by edaphic constraints and specifically soil P status
(Webb, 1968; Russell-Smith et al., 2004). The widely observed positive
correlation of leaf [N] ↔ [P] (e.g. Wright et al., 2001) is
evident here as well (Fig. 2a), but the slope of the relationship differed
between F and S. Despite their situation in the
Atherton basalt province (Whitehead et al., 2007), the mean foliar N : P
ratios for all of the sites visited in this study were far above thresholds
believed to constitute P deficiency (Townsend et al., 2007; Cernusak et al.,
2010). This is particularly striking since the forests studied here, even
after excluding the subcanopy trees, had median [N]a of only
1.63 g m-2. Such N levels are low compared to other tropical forests
for which extensive data have recently become available – see Table 2 of
Domingues et al. (2014). Values here are lower, for example, than Cameroon
(2.12 g m-2) and lower even than for trees growing on soils of low
nutrient availability in the Amazon Basin (1.90 g m-2). Foliar P
levels, however, were lower still despite concentrations of total soil
extractable P (Pex) for our forest plots being relatively high
(165–958 µg g-1; Table 2). For what have been classed “high
nutrient” soils in the Amazon Basin, for example, Fyllas et al. (2009)
reported median foliar P values of 1.11 mg g-1 and 0.11 g m-2 with
equivalent soil Pex ranging from 52 to 422 µg g-1
(Quesada et al., 2010). In our study, however, upper canopy forest trees
displayed median P values of only 0.76 mg g-1 and 0.08 g m-2,
much closer to Amazon forest trees growing on “low nutrient status” soils for
which soil Pex ranges from 14 to 87 µg g-1 (Quesada
et al., 2010) with foliar P median values of 0.7 mg g-1 and
0.06 g m-2 (Fyllas et al., 2009).
This apparent “inability” of Australian forest trees to utilise high soil P
availabilities could perhaps be related to their unique evolutionary history.
Essentially of Gondwanan origin (Crisp et al., 1999), today's forests
represent remnants of more temperate moist forests that covered much of the
continent until the mid-Miocene (Adam, 1992). Presumably arising from a flora
already adapted to the characteristically P-limited soils of much of
Australia (McKenzie et al., 2004), it may be that many of the species
occurring within the Australian tropical forest region suffer from an
“evolutionary hangover” lacking the ability to utilise high levels of soil P
even where available. There is in addition, especially for the lowlands, a
considerable Asia-derived element in the Australian tropical forest flora
(Sniderman and Jordan, 2011; Crayn et al., 2014), as many soils of the Asian
lowland tropical forest region are also of relatively low nutrient status
(Acres et al., 1975; Tessins and Jusop, 1983; Ohta and Effendi, 1992; Banin
et al., 2014). By comparison, despite the generally lower P status of the
savanna soils (Table 2), the savanna trees in our study had a slightly higher
median [P]a than those of the forest (S=0.08 g m-2 and F=0.06 g m-2) and this was true also for [N]a (S=2.09 g m-2 and F=1.62 g m-2). This finding for the
Australian species contrasts with previous work in western Africa and Cameroon
where area-based N and P concentrations were lower for savanna than for
forest species (Domingues et al., 2014; Schrodt et al., 2014).
Despite these differences in area-based nutrient concentrations, there are
notable consistencies between our results and the African studies mentioned
above. First, albeit with different model constructs, there is clear evidence
of a role for both N and P in the modulation of photosynthetic rates in the
field. Second, other things being equal, it seems that savanna trees have
higher N use efficiency than their forest counterparts. This higher
AN (Fig. 4) may reflect differences in leaf construction linked
to the higher light environment. Earlier studies have suggested that lower
AN values for sclerophytes may be caused by limitations to
internal conductance caused by leaf structural factors linked to greater leaf
longevity (e.g. Warren, 2008). Mesophyll conductance(gm) is,
however, the complex and variable product of at least three phases acting in
series (Flexas et al., 2008): conductance through intercellular air spaces
(gias), through cell walls (gw) and through the
liquid and membrane phases inside cells (gliq). The most
important constraint on gm is thought to be gliq,
which is the phase least affected by leaf structure.
Overall, our results suggest a complex effect of P on photosynthetic capacity
for these Australian tree species. Foliar [P]a was only
marginally significant in the preferred linear mixed-effects model (Table 5),
but its inclusion improved the overall predictive power. When examined on a
mass basis, P did, however, appear more critical and with a N × P interaction
term also included in the optimal model (Table S6). This mass- versus
area-based inconsistency in the apparent importance of P as a modulator of
photosynthetic rates was also noted by Domingues et al. (2014), who likewise
found their mass-based models to include a N × P interaction term not
present in the area-based version. The sudden appearance of apparently
significant terms when transforming area-based entities to a mass basis is,
however, to be expected (Lloyd et al., 2013).
Describing trait variation using plant functional types
Where possible in the F sites, tree species were selected in order
to provide a contrast of light environment as described in the assigned
categories of plant functional type (Φ). Such categorisation is often
problematic and especially in the setting of boundaries from one group to
another (e.g. Poorter, 1999). For Australian tropical moist forests, Webb
developed a classification system of 20 structural vegetation types along
climatic and edaphic gradients (1968). Faced with such complexity and
subjectivity, many authors have instead argued for a spectrum of vegetation
types or habits (e.g. Coste et al., 2005). The current study used Φ to
attempt to explain residual patterns in the data after controlling for
V (Fig. 6b). As hypothesised, there was evidence that pioneer
trees of the F showed higher photosynthetic capacity and nutrient
use efficiency than those shade-tolerant species which persist in the
understory (Fig. 4).