The worldwide leaf economics spectrum
Ian J. Wright
1
, Peter B. Reich
2
, Mark Westoby
1
, David D. Ackerly
3
, Zdravko Baruch
4
, Frans Bongers
5
, Jeannine Cavender-Bares
6
,
Terry Chapin
7
, Johannes H. C. Cornelissen
8
, Matthias Diemer
9
, Jaume Flexas
10
, Eric Garnier
11
, Philip K. Groom
12
, Javier Gulias
10
,
Kouki Hikosaka
13
, Byron B. Lamont
12
, Tali Lee
14
, William Lee
15
, Christopher Lusk
16
, Jeremy J. Midgley
17
, Marie-Laure Navas
11
,
U
¨
lo Niinemets
18
, Jacek Oleksyn
2,19
, Noriyuki Osada
20
, Hendrik Poorter
21
, Pieter Poot
22
, Lynda Prior
23
, Vladimir I. Pyankov
24
,
Catherine Roumet
11
, Sean C. Thomas
25
, Mark G. Tjoelker
26
, Erik J. Veneklaas
22
& Rafael Villar
27
1
Department of Biological Sciences, Macquarie University, New South Wales 2109, Australia
2
Department of Forest Resources, University of Minnesota, St Paul, Minnesota 55108, USA
3
Department of Biological Sciences, Stanford University, Stanford, California 94305, USA
4
Departamento de Estudios Ambientales, Universidad Simo
´
n Bolivar, Caracas 1080, Venezuela
5
Forest Ecology and Forest Management Group, Department of Environmental Sciences, Wageningen University, PO Box 342, 6700 AH Wageningen,
The Netherlands
6
Smithsonian Environmental Research Center, PO Box 28, 647 Contees Wharf Road, Edgewater, Maryland 21037, USA
7
Institute of Arctic Biology, University of Alaska, Fairbanks, Alaska 99775, USA
8
Institute of Ecological Science, Department of Systems Ecology, Vrije Universiteit, De Boelelaan 1087, 1081 HV, Amsterdam, The Netherlands
9
Institute fu
¨
r Umweltwissensch, University of Zurich, Zurich, Switzerland
10
Departament de Biologia, Laboratori de Fisiologia Vegetal, Universidad de Illes Balears, 07122 Palma de Mallorca, Illes Balears (Spain)
11
Centre d’Ecologie Fonctionnelle et Evolutive, CNRS, UMR 5175, 1919, Route de Mende, 34293 Montpellier cedex 5, France
12
Department of Environmental Biology, Curtin University of Technology, Perth, Western Australia 6845, Australia
13
Graduate School of Life Sciences, Tohoku University, Aoba, Sendai 980-8578, Japan
14
Department of Biology, University of Wisconsin-Eau Claire, Eau Claire, Wisconsin 54702-4004, USA
15
Landcare Research, Private Bag 1930, Dunedin, New Zealand
16
Departamento de Bota
´
nica, Universidad de Concepcio
´
n, Casilla 160-C, Concepcio
´
n, Chile
17
Department of Botany, University of Cape Town, ZA-7701 Rondebosch, South Africa
18
Department of Plant Physiology, University of Tartu, Riia 23, Tartu 51011, Estonia
19
Polish Academy of Sciences, Institute of Dendrology, Parkowa 5, 62-035 Kornik, Poland
20
Nikko Botanical Garden, Graduate School of Science, University of Tokyo, 1842 Hanaishi, Nikko, Tochigi 321-1435, Japan
21
Plant Ecophysiology, Utrecht University, PO Box 800.84, 3508 TB, Utrecht, The Netherlands
22
School of Plant Biology, University of Western Australia, Crawley, Western Australia 6009, Australia
23
Key Centre for Tropical Wildlife Management, Charles Darwin University, Darwin, Northern Territory 0909, Australia
24
Ural State University, Yekaterinburg, Russia
25
Faculty of Forestry, University of Toronto, 33 Willcocks St, Toronto, Ontario M5S 3B3, Canada
26
Department of Forest Science, Texas A&M University, 2135 TAMU, College Station, Texas 77843-2135, USA
27
Area de Ecologı
´
a, Campus de Rabanales, Universidad de Co
´
rdoba, 14071 Co
´
rdoba, Spain
...........................................................................................................................................................................................................................
Bringing together leaf trait data spanning 2,548 species and 175 sites we describe, for the first time at global scale, a universal
spectrum of leaf economics consisting of key chemical, structural and physiological properties. The spectrum runs from quick to
slow return on investments of nutrients and dry mass in leaves, and operates largely independently of growth form, plant
functional type or biome. Categories along the spectrum would, in general, describe leaf economic variation at the global scale
better than plant functional types, because functional types overlap substantially in their leaf traits. Overall, modulation of leaf
traits and trait relationships by climate is surprisingly modest, although some striking and significant patterns can be seen.
Reliable quantification of the leaf economics spectrum and its interaction with climate will prove valuable for modelling nutrient
fluxes and vegetation boundaries under changing land-use and climate.
Green leaves are fundamental for the functioning of terrestrial
ecosystems. Their pigments are the predominant signal seen from
space. Nitrogen uptake and carbon assimilation by plants and the
decomposability of leaves drive biogeochemical cycles. Animals,
fungi and other heterotrophs in ecosystems are fuelled by photo-
synthate, and their habitats are structured by the stems on which
leaves are deployed. Plants invest photosynthate and mineral
nutrients in the construction of leaves, which in turn return a
revenue stream of photosynthate over their lifetimes. The photo-
synthate is used to acquire mineral nutrients, to support metabo-
lism and to re-invest in leaves, their supporting stems and other
plant parts.
There are more than 250,000 vascular plant species, all engaging
in the same processes of investment and reinvestment of carbon and
mineral nutrients, and all making enough surplus to ensure con-
tinuity to future generations. These processes of investment and
re-investment are inherently economic in nature
1–3
. Understanding
how these processes vary between species, plant functional types
and the vegetation of different biomes is a major goal for plant
ecology and crucial for modelling how nutrient fluxes and veg-
etation boundaries will shift with land-use and climate change.
Data set and parameters
We formed a global plant trait network (Glopnet) to quantify leaf
economics across the world’s plant species. The Glopnet data set
spans 2,548 species from 219 families at 175 sites (approximately
1% of the extant vascular plant species). The coverage of traits,
species and sites is at least tenfold greater than previous data
compilations
4–11
, extends to all vegetated continents, and represents
a wide range of vegetation types, from arctic tundra to tropical
rainforest, from hot to cold deserts, from boreal forest to grasslands.
Site elevation ranges from below sea level (Death Valley, USA) to
4,800 m. Mean annual temperature (MAT) ranges from 216.5 8Cto
27.5 8C; mean annual rainfall (MAR) ranges from 133 to 5,300 mm
per year. This covers most of the range of MAT–MAR space in which
higher plants occur
12
(Fig. 1). The broad coverage of the data set has
articles
NATURE | VOL 428 |22 APRIL 2004 | www.nature.com/nature 821
© 2004
Nature
Publishing
Group
allowed us to quantify the relationships of leaf economics to climate
at a scale not previously possible. Here we report some global
outcomes from our analyses.
We focus on six key features of leaves that together capture many
essentials of leaf economics. (1) Leaf mass per area (LMA) measures
the leaf dry-mass investment per unit of light-intercepting leaf area
deployed. Species with high LMA have a thicker leaf blade or denser
tissue, or both. (2) Photosynthetic assimilation rates measured
under high light, ample soil moisture and ambient CO
2
are here
called photosynthetic capacity (A
mass
) for brevity. Photosynthetic
capacity is influenced both by stomatal conductance and by the
drawdown of CO
2
concentration inside the leaf (carboxylation
capacity). (3) Leaf nitrogen (N) is integral to the proteins of
photosynthetic machinery, especially Rubisco
8,13
. The photosyn-
thetic machinery is responsible for drawdown of CO
2
inside the leaf,
a process also affected by leaf structure
14,15
. (4) Leaf phosphorus (P)
is found in nucleic acids, lipid membranes and bioenergetic mol-
ecules such as ATP. Phosphorus derives from weathering of soil
minerals at a site, in contrast to nitrogen, much of which may be
fixed from the atmosphere by plants. (5) Dark respiration rate
(R
mass
) reflects metabolic expenditure of photosynthate in the leaf,
especially protein turnover and phloem-loading of photo-
synthates
16
. (6) Leaf lifespan (LL) describes the average duration
of the revenue stream from each leaf constructed. Long LL requires
Figure 1 Mean annual rainfall (MAR) and mean annual temperature (MAT). Results for
the 175 sites from where leaf data were compiled (a), in relation to major biome types of
the world (b), following ref. 12. Biome boundaries are only approximate.
Figure 2 Three-way trait relationships among the six leaf traits with reference to LMA,
one of the key traits in the leaf economics spectrum. The direction of the data cloud in
three-dimensional space can be ascertained from the shadows projected on the floor and
walls of the three-dimensional space. Sample sizes for three-way relationships are
necessarily a subset of those for each of the bivariate relationships. a, A
mass
, LMA and
N
mass
; 706 species. b, LL, R
mass
and LMA; 217 species. c, N
mass
, P
mass
and LMA; 733
species. d, A
area
, LMA and N
area
; 706 species.
articles
NATURE | VOL 428 | 22 APRIL 2004 | www.nature.com/nature822
© 2004
Nature
Publishing
Group
robust construction in the form of high LMA
6,17,18
.
Ecophysiological attributes such as leaf N, leaf P, A
mass
or R
mass
can be expressed per leaf area, per leaf dry mass or per leaf volume. A
leaf-area basis reflects fluxes in relation to surfaces. It is a natural
basis for describing light capture and for expressing transactions
through surfaces such as trade-offs between carbon gain and water
transpiration
19,20
. On a mass basis, leaf economics are quantified in
terms of revenues and expenditures per unit investment, measured
as biomass, or C, N or P. Scaling up to whole plants, mass-based
expressions of leaf nutrient concentrations (N
mass
, P
mass
), A
mass
and R
mass
are more tightly correlated than area-based expressions to
relative growth rates of seedlings or to absolute height growth rates
of young trees
21,22
. By definition, area- and mass-based traits can be
interconverted via LMA (for example, N
area
¼ N
mass
£ LMA). Here
we report relationships among leaf traits using both area- and mass-
based formulations, and also analyses where the effect of LMA is
explicitly controlled.
We first quantify relationships among the six mass-based leaf
traits: A
mass
, R
mass
, LMA, LL, N
mass
and P
mass
. We find that these
economic traits covary tightly across all species. Trait relationships
are similar for species pooled by growth form, plant functional
group or biome, which indicates the existence of a single global
spectrum of leaf economic variation. Second, we treat photosyn-
thetic capacity, R
mass
and leaf nutrient concentrations on a leaf-area
basis. These relationships are not as strong as among mass-basis
traits, and we consider why. Third, we assess the influence of climate
on leaf trait relationships. We find that, in general, the influence is
modest, although particular traits and trait relationships show
substantial patterning with climate.
The leaf economics spectrum
Mass-based leaf traits
The six mass-basis leaf traits varied by one to two orders of mag-
nitude across the data set. LMA ranged from 14 to 1,500 g m
22
and
LL from 0.9 up to 288 months. A
mass
ranged from 5 to
660 nmol g
21
s
21
; dark respiration from 2.2 to 65 nmol g
21
s
21
.
N
mass
ranged from 0.2 to 6.4%; P
mass
from 0.008 to 0.6%. Con-
sidered pairwise, all leaf traits were highly correlated (Table 1).
These correlations have been reported previously from smaller data
sets
6–10,23
. Here we have generalized the patterns over many more
species, sites and vegetation types.
We moved beyond pairwise consideration of traits to determine
the extent to which leaf economic traits covary in multidimensional
trait space. This covariation can be quantified as the proportion of
total trait variation explained by the first principal axis in a principal
components analysis. In two-trait space, the principal axis is the
long axis of the ellipse resulting from two correlated traits. In three-
trait space, the principal axis is the long axis of an ellipsoid. In multi-
dimensional trait space, the principal axis describes the main axis of
variation through a hyperellipsoid
24
. A remarkable 82% of all
variation in A
mass
, LMA and N
mass
across species lay along the
first principal axis in three-trait space (Fig. 2a). Because some of the
residual 18% must be measurement variation, 82% represents a
minimum estimate of the dominance of this single spectrum in
explaining variation across plant species worldwide. Further three-
dimensional subsets of the six-dimensional data set are shown in
Figs 2b and c. Multi-dimensional analyses including from four to all
six of the traits similarly showed the large majority of variation
explained with a single axis (Table 2). With the six traits included,
74% of all variation lay along the first principal axis.
The extent to which each trait contributed to the principal axis of
variation is indicated by a loading (or weight) assigned to each trait.
The directionality of these loadings (Table 2) indicates that the axis
can be thought of as a leaf economics spectrum. This spectrum runs
from species with potential for quick returns on investments of
nutrients and dry mass in leaves to species with a slower potential
rate of return. At the quick-return end are species with high leaf
nutrient concentrations, high rates of photosynthesis and respira-
tion, short leaf lifetimes and low dry-mass investment per leaf area.
At the slow-return end are species with long leaf lifetimes, expensive
high-LMA leaf construction, low nutrient concentrations, and low
rates of photosynthesis and respiration.
Within growth forms or functional groups the principal axes of
variation had the same directionality of trait correlations as for the
total data set (Table 2). Similarly, species grouped by major biome
type (Fig. 1), or by MAT or MAR classes, yielded the same pattern
(data not shown). The concordance of these results is of special
significance, indicating a coordination of these key leaf traits that is
consistent across major plant functional types, growth forms and
biomes. The amount of variation captured by the principal axis in
the different species groupings was also similar to that across all
species in most cases.
The main exception was among deciduous trees and shrubs
where, as expected, there was substantially less variation in LL
(5-fold versus 100-fold), and where LL–LMA relationships were
partially uncoupled. Still, whereas different growth forms and
functional groups were differentiated along the leaf economics
spectrum when trait means were considered, the overlap between
species groups was large (data not shown). Evergreen trees and
shrubs had longer mean LL and higher LMA than deciduous
species, but evergreens had much wider ranges for both traits,
extending to LLs almost as short as for the shortest-LL deciduous
species, and to similarly low LMA. Similarly, on average, shrubs and
trees had higher LMA and longer LLs but lower N
mass
, A
mass
and
Rd
mass
than herbs and grasses, yet trees and shrubs spanned almost
the entire range of any of these leaf traits. Another example: N
2
-
fixing species had higher mean N
mass
than non N
2
-fixing plants, yet
the range of N
mass
was larger and extended higher in non N
2
-fixing
species.
Allometries among traits
Slopes on log–log axes, or ‘scaling exponents’, indicate the propor-
tionality of pairwise trait relationships. Most slopes were signifi-
cantly different from þ1.0 or 21.0 (Table 1, above diagonal). In
other words, the traits showed allometric relationships rather than
scaling in direct proportion with one another (‘isometry’). A tenfold
increase in P
mass
corresponded with a 4.7-fold increase in N
mass
(scaling slope 0.67), indicating that N:P ratios decline as one moves
towards the end of the spectrum that represents quick returns on
investments of carbon and nutrients. The stoichiometry between
Table 1 Mass basis of bivariate relationships between the leaf traits
log LL log LMA log A
mass
log N
mass
log P
mass
log R
mass
...................................................................................................................................................................................................................................................................................................................................................................
log LL 1.71 (1.62, 1.82) 21.38 (21.45, 21.31) 22.26 (22.39, 22.14) 21.06 (21.19, 20.94) 21.67 (21.82, 21.53)
log LMA 0.42 (678) 20.75 (20.79, 20.72) 21.28 (21.32, 21.24). 20.82 (20.86, 20.78) 20.96 (21.05, 20.88)
log A
mass
0.68 (512) 0.50 (764) 1.72 (1.63, 1.81) 1.03 (0.91, 1.16) 1.18 (1.09, 1.27)
log N
mass
0.42 (706) 0.57 (1958) 0.53 (712) 0.66 (0.64, 0.69) 0.70 (0.64, 0.76)
log P
mass
0.24 (207) 0.55 (739) 0.16 (212) 0.72 (745) 1.04 (0.87, 1.25)
log R
mass
0.60 (217) 0.45 (274) 0.59 (259) 0.55 (267) 0.34 (78)
...................................................................................................................................................................................................................................................................................................................................................................
Standardized major axis slopes with 95% confidence intervals are given in the upper right section of the matrix (y variable is column 1, x variable in row 1). Coefficients of determination (r
2
) and sample
sizes are given in the lower left section of the matrix. All relationships were highly significant, P ,, 0.0001. Further details allowing calculation of predictive regression equations for each pair of leaf
traits are given in Supplementary Information.
articles
NATURE | VOL 428 | 22 APRIL 2004 | www.nature.com/nature 823
© 2004
Nature
Publishing
Group
leaf N and P has begun to attract increasing interest as an index of
soil nutrient limitation (taken across sets of co-occurring species)
25
,
and also because it relates to plant growth strategies and influences
plant–herbivore interactions in food webs
26
.
The slope of the LL–LMA relationship was 1.7, meaning that
tenfold greater dry mass invested per unit leaf area coincided with
50-fold longer LL. All else being equal, this implies that the measure
of light-intercepting leaf area (in mm
2
) £ duration (in months) per
gram leaf was greater for high-LMA than for low-LMA species. If
this measure translated directly into fitness benefit, this might lead
to runaway selection towards ever-increasing LMA and LL
27
. But
continuing ecological success of low-LMA species shows that all else
is not equal. On average, a tenfold decrease in LMA, for example,
coincided with a 21-fold increase in photosynthetic capacity.
Further, low LMA, high A
mass
and generally faster turnover of
plant parts permit a more flexible response to the spatial patchiness
of light and soil resources
28
, as well as conferring advantages via a
compound interest effect, whereby carbon fixed earlier can be
reinvested in new leaves sooner
27,29
. On the other hand, high
A
mass
requires high N
mass
, and the combination of high LMA and
high N
mass
may increase vulnerability to herbivory as well as
increasing energy losses via respiration
9
, which can be detrimental
in situations where energy gain is low owing to low resource
availability, such as under low light conditions
30
.
Area versus mass basis of expression
Because N
mass
¼ N
area
/LMA, we considered whether the relation-
ship between N
mass
and LMA should be thought of as arising from
N
area
values that do not vary greatly across species, divided through
by highly varying LMA. The evidence contradicts this interpret-
ation. First, N
area
varied more widely than did N
mass
(35-fold versus
26-fold; data for 1,958 species). Second, the 21.28 slope of the
N
mass
–LMA relationship was significantly steeper (Table 1) than the
21.0 slope expected if N
area
was independent of LMA.
N
area
was indeed correlated with LMA, but positively, and more
weakly than were N
mass
and LMA (r
2
¼ 0.34 versus 0.57; Table 3).
Gradients of leaf N on a mass versus an area basis hence represent
fundamentally different multiple-trait gradients because of their
different patterns of covariation with LMA
31
. The leaf economics
spectrum of species from low to high N
mass
also constitutes a
spectrum of decreasing LMA and LL, and of increasing A
mass
,
R
mass
and P
mass
. But a spectrum of leaf types in terms of N
area
would be less informative. A given N
area
can result from low N
mass
combined with high LMA, high N
mass
combined with low LMA, or
from combinations in between. In general, low N
mass
with high
LMA represents a species with long-lived leaves and low A
mass
, while
high N
mass
with low LMA represents the opposite. As a result, a
wider variety of leaf types may be found at a given N
area
than at a
given N
mass
31–33
. Here for example, LMA varied approximately
20-fold at the grand mean of N
area
versus tenfold at the grand
mean of N
mass
.
Because of the covariation between leaf N and LMA, relationships
between leaf N and other traits changed substantially when
expressed on an area rather than on a mass basis. As seen pre-
viously
7–9,31,33
, relationships between leaf N and A
mass
or dark
respiration were weaker when considered on an area basis, as were
relationships between leaf P and other traits (Table 3). By also
including LMA in analyses, we can quantify the independent effects
of leaf structure and nutrient content on A
mass
and dark respira-
tion
32,34
. A
mass
increased with increasing N
mass
at any given LMA,
and decreased with increasing LMA at any given N
mass
(partial
regression coefficients for LMA and leaf N, P , 0.001; regression
details in Supplementary Information). Similarly, A
area
increased
with increasing N
area
at any given LMA, and decreased with
increasing LMA at any given N
area
(Fig. 2d and Supplementary
Information). That is, both leaf structure and nitrogen concen-
tration affect photosynthetic capacity
5,6,8,32
. The independent LMA
effect is most probably due to leaves with high mass per area having
longer diffusion paths from stomata to chloroplasts or greater
internal shading of lower chloroplasts, limiting the A
mass
possible
for a given leaf protein content
14,15,35
. Also, less of the N may
be invested in photosynthetic versus non-photosynthetic leaf com-
ponents in high-LMA species
36,37
. Similarly, N
mass
and LMA showed
independent effects on R
mass
(partial regression coefficients both
Table 2 Principal components analyses of global leaf trait data
All species Trees* Shrubs Herbs* Grasses* N
2
-fixers* Non
N
2
-fixers
C3 C4† Broad-leaved
shrubs
and trees
Needle-leaved
shrubs
and trees*
Evergreen
shrubs
and trees
Deciduous
shrubs
and trees*
...................................................................................................................................................................................................................................................................................................................................................................
Variation explained (%) 74.4 77.2 72.0 67.8 70.7 79.6 75.0 73.0 68.7 72.4 58.0 64.0 48.5
Leaf trait Loadings
LL 20.85 20.90 20.83 20.73 20.86 20.86 20.84 20.88 20.84 20.79 20.74 20.43
LMA 20.88 20.84 20.84 20.72 20.91 20.87 20.87 20.84 20.84 20.84 20.73 20.82 20.28
N
mass
0.91 0.82 0.91 0.89 0.71 0.88 0.92 0.90 0.71 0.90 0.71 0.87 0.81
A
mass
0.86 0.91 0.82 0.88 0.79 0.96 0.86 0.88 0.87 0.83 0.82 0.76 0.84
R
mass
0.88 0.93 0.83 0.89 0.93 0.89 0.89 0.90 0.88 0.84 0.74 0.82 0.90
P
mass
0.79 0.87 0.80 0.72 0.85 0.78
...................................................................................................................................................................................................................................................................................................................................................................
The principal axis or component explained 74.4% of variation in the total data set. LL and LMA were negatively correlated with this primary axis of variation while the other traits were positively correlated with
it, both in the total data set and in data subsets defined by growth form or functional group. The same directionality of trait loadings and similarly high percentage of variance was explained by the principal
axis with species grouped by site temperature, rainfall or altitude, or with sites grouped into major biome type following Fig. 1 (data not shown), demonstrating the broad generality of the coordinated
spectrum of leaf economics. All data were log
10
-transformed before analyses.
*P
mass
excluded due to too few data.
†P
mass
and LL excluded due to few data. C3, C4 indicate species with C3 and C4 photosynthetic pathways, respectively.
Table 3 Area basis of bivariate relationships between the six leaf traits
log LL log LMA log A
area
log N
area
log P
area
log R
area
...................................................................................................................................................................................................................................................................................................................................................................
log LL 1.71 (1.62, 1.82) 22.12 (22.30, 21.96) 22.36 (22.19, 22.54) NA (P ¼ 0.862) NA (P ¼ 0.211)
log LMA 0.42 (678) NA (P ¼ 0.152) 1.54 (1.48, 1.59) 1.23 (1.14, 1.32) 1.20 (1.08, 1.34)
log A
area
0.13 (512) 0.003 (764) 1.21 (1.13, 1.31) 0.66 (0.58, 0.75) 0.93 (0.83, 1.04)
log N
area
0.04 (706) 0.34 (1,958) 0.13 (722) 0.68 (0.64, 0.72) 0.89 (0.80, 0.98)
log P
area
0.0002 (202) 0.01 (739) 0.02 (223) 0.35 (750) 1.26 (1.04, 1.54)
log R
area
0.01 (217) 0.14 (274) 0.19 (259) 0.34 (267) 0.26 (78)
...................................................................................................................................................................................................................................................................................................................................................................
Standardized major axis slopes with 95% confidence intervals are given in the upper right section of the matrix (y variable is column 1, x variable in row 1). Coefficients of determination (r
2
) and sample
sizes are given in the lower left section of the matrix. NA, not applicable: in these cases the correlation was clearly non-significant (P values given in parentheses). While a standardized major axis can
still be fitted in such cases, its slope is essentially meaningless. All other relationships were highly significant, P ,, 0.0001, with the exception of those between log P
area
and each of log LMA
(P ¼ 0.029) and log A
area
(P ¼ 0.034). Further details allowing calculation of predictive regression equations for each pair of traits are given in the Supplementary Information.
articles
NATURE | VOL 428 | 22 APRIL 2004 | www.nature.com/nature824
© 2004
Nature
Publishing
Group
P , 0.001), although the physiological basis for an independent
LMA effect on R
mass
is less clear than for photosynthetic capacity
9
.
Relating to the weaker and somewhat different pattern of area-
based trait relationships, the principal axis of a principal com-
ponents analysis involving LMA, LL, A
area
, R
area
, N
area
and P
area
explained a lower proportion of total trait variation than the mass-
based equivalent and had a different pattern of trait loadings. This
principal axis explained 43% of variation (versus 74% on a mass
basis), and largely reflected a spectrum of increasing N
area
(the trait
with the strongest trait loading; Supplementary Information). The
five other traits showed positive loadings with the principal axis,
with these loadings clearly weaker than in the mass-based analysis.
About half the residual variation was explained by a second
principal axis, which expressed essentially the same trait corre-
lations as the first axis from the mass-based analysis. Together, these
first two axes explained 72% of total trait variation in the area-based
data set, marginally less than the first principal axis in the mass-
based analysis. Area-based analyses for species grouped by growth
form, functional type or biome showed patterns broadly concor-
dant with those across all species, though less clearly than for the
mass-based analyses (results not shown).
In summary, the coordination among leaf traits appears to be
stronger and simpler on a mass basis than an area basis. This is not
because area-basis traits are less varied among species. Rather it is
because the LMA–LL spectrum is related to mass-based nutrient
concentrations and assimilation and respiration rates in a simpler
way than on an area basis.
Climate influence on leaf investment
Plant ecologists have emphasized broad relationships between leaf
traits and climate for at least a century. In particular, a general
tendency for species inhabiting arid and semi-arid regions to have
leathery, high-LMA leaves has been reported
4,10,38–40
. Building high-
LMA leaves needs more investment per unit leaf area. Construction
cost per unit leaf mass varies relatively little between species: leaves
with high protein content (typically low-LMA leaves) tend to have
low concentrations of other expensive compounds such as lipids or
lignin, and high concentrations of cheap constituents such as
minerals
41
. Leaf traits associated with high LMA (for example,
thick leaf blade; small, thick-walled cells) have been interpreted as
adaptations that allow continued leaf function (or at least postpone
leaf death) under very dry conditions, at least in evergreen species.
We characterized sites by annual means of temperature, potential
evapotranspiration (PET), vapour pressure deficit (VPD) and solar
irradiance, and MAR. PET, VPD and irradiance were cross-
correlated with rainfall and temperature. Against expectations,
LMA showed only a very weak relationship with lower rainfall,
considered worldwide (r
2
¼ 0.002, P ¼ 0.032; 2,370 species from
163 sites). LMA was actually more strongly (positively) correlated
with MAT (r
2
¼ 0.10). Worldwide, precipitation is correlated with
temperature, cold high-latitude environments typically having low
precipitation (Fig. 1). Once variation in MAT was controlled in a
multiple regression, LMA did indeed increase as rainfall decreased
(Fig. 3). Similarly, LMA was more strongly (positively) correlated
with VPD (r
2
¼ 0.15), PET (r
2
¼ 0.15) or site irradiance
(r
2
¼ 0.18) than with MAT or site rainfall alone.
Despite the substantial overlap in leaf traits of evergreen and
deciduous species, these species groups varied somewhat in their
leaf trait–climate relationships. LMA and rainfall showed a strong
negative relationship in evergreen shrubs and trees, whereas in
deciduous species they were virtually unrelated (r
2
¼ 0.22 versus
0.002). Across all species, LL was positively correlated with PET,
VPD, MATand site irradiance (r
2
values ranging from 0.04 to 0.10).
In deciduous trees and shrubs these relationships were consistently
stronger (r
2
values ranged from 0.37 to 0.51). That is, LL of
deciduous species was shorter at colder sites where the growing
season was shorter. But in evergreen shrubs and trees, LL tended to
be longer at colder, lower-humidity sites (for example, MAT versus
LL, r
2
¼ 0.10; PET versus LL, r
2
¼ 0.18). This is consistent with
cold climate vegetation being typically N-limited and demonstrat-
ing ‘slow-return’ strategies
42
.
Trait coordination is largely independent of climate
A major aim of the Glopnet collaboration was to obtain enough
coverage of climate variation to dissect out effects of climate on
relationships between leaf economic traits. There were indeed
statistically significant effects of climate. Nevertheless, a major
finding from this project is that the influence of climate was, in
general, quite modest. How can this be, given that traits such as
LMA and LL vary systematically with MAR, MATand other climate
indices? The answer seems to have two elements.
First, much of the total leaf economic variation occurs among co-
existing species. The proportion of total variation in LL within sites
was 57%, the remaining 43% occurring between sites (variance
components analysis). For R
mass
, the proportion of within-site
variation was 67%, for A
mass
48%, for N
mass
38%, for LMA 36%,
Figure 3 LMA as a function of MAT and MAR at the study sites (data for 2,370 species
from 163 sites; rainfall and LMA are log
10
-scaled). The coefficients for MAT and log
rainfall were highly significant in a multiple regression (both P , 0.0001; further details
given in Supplementary Information).
Figure 4 LL as a function of LMA and MAR (all axes are log
10
-scaled). When viewed in
three dimensions, the two-dimensional LL–LMA cloud of points is spread along a sloping
surface. The slope of this surface is steeper in the LMA dimension than in the rainfall
dimension, reflecting the higher partial regression coefficient for LMA (1.23 versus 0.47).
Both coefficients were highly significant in a multiple regression (P , 0.0001;
r
2
¼ 0.51; data for 678 species from 51 sites).
articles
NATURE | VOL 428 | 22 APRIL 2004 | www.nature.com/nature 825
© 2004
Nature
Publishing
Group
