LETTER
Functional identity is the main driver of diversity effects in
young tree communities
Cornelia M. Tobner,
1
Alain
Paquette,
1
* Dominique Gravel,
2,3
Peter B. Reich,
4,5
Laura J. Williams
6
and
Christian Messier
1,7
Abstract
Two main effects are proposed to explain biodiversity–ecosystem functioning relationships: niche
complementarity and selection effects. Both can be functionally defined using the functional diver-
sity (FD) and functional identity (FI) of the community respectively. Herein, we present results
from the first tree diversity experiment that separated the effect of selection from that of comple-
mentarity by varying community composition in high-density plots along a gradient of FD, inde-
pendent of species richness and testing for the effects of FD and community weighted means of
traits (a proxy for FI) on stem biomass increment (a proxy for productivity). After 4 years of
growth, most mixtures did not differ in productivity from the averages of their respective mono-
cultures, but some did overyield significantly. Those positive diversity effects resulted mostly from
selection effects, primarily driven by fast-growing deciduous species and associated traits. Net
diversity effect did not increase with time over 4 years.
Keywords
Biodiversity and ecosystem functioning, biodiversity experiment, complementarity, functional
diversity, functional identity, functional traits of trees, IDENT, overyielding.
Ecology Letters (2016) 19: 638–647
INTRODUCTION
The idea that increasing diversity leads to increased ecosystem
functioning has been proposed at least since the 19th century.
However, it was not until the late 20th century with the con-
siderable and accelerating change in the earth’s biota that a
systematic and concerted search for the effects of biodiversity
on ecosystem functioning (BEF) began. After more than two
decades of BEF research, the hypothesis that increased pro-
ducer diversity leads to increased producer productivity has
been accepted with high confidence for a variety of systems
(Cardinale et al. 2012; Balvanera et al. 2014).
To explain positive BEF relations, two main effects have
been proposed: complementarity and selection. Complemen-
tarity effects include niche partitioning, positive interactions
(i.e. facilitation) and positive feedbacks on resource supply,
whereas selection effects are due to dominant species driving
ecosystem functioning (Roscher et al. 2012). These mecha-
nisms have been shown to work together in a variety of sys-
tems, with complementarity often explaining more of the
variance (Reich et al. 2001; Cardinale et al. 2007; Fargione
et al. 2007; Cardinale et al. 2011), especially with the advance
of time (Cardinale et al. 2007; Fargione et al. 2007; Allan
et al. 2011; Reich et al. 2012). Both mechanisms hinge on the
role of functional traits. Indeed, functional aspects of diversity
have been shown to have greater explanatory power on
ecosystem functioning than SR (species richness) alone (Dı
az
& Cabido 2001; Mokany et al. 2008; Gravel et al. 2011),
although recent work highlights that the approach taken in
such analyses can lead to the opposite conclusion (Venail
et al. 2015).
Two main approaches to measure functional aspects of a
community have been used. One approach measures distances
among species in a multidimensional trait space (Lalibert
e&
Legendre 2010; Mouchet et al. 2010) to characterise func-
tional trait diversity (FD). Since functional traits should be
related to a species’ resource-based niche, greater differentia-
tion among species in functional traits ought to reflect greater
resource-use complementarity and reduced competition
(Hooper 1998; Petchey 2003). Another approach assesses the
community weighted mean trait value of all species present in
a mixture (CWM, Mokany et al. 2008; Roscher et al. 2012).
Conceptually, CWMs are based on the ‘mass ratio hypothesis’
stating that the functional traits of the dominating species in
a community drive ecosystem functioning (Grime 1998). As a
consequence, CWMs are closely linked to the selection effect
(SE) and are a direct measure of the functional identity (FI)
of a species assemblage (Mokany et al. 2008; Roscher et al.
2012). In mature forests, CWM canopy traits explained a
large majority of the variance in productivity across > 100
1
Center for Forest Research, Universit
eduQu
ebec
a Montr
eal, PO Box 8888,
Centre-ville Station, Montr
eal, QC H3C 3P8, Canada
2
Universit
e de Sherbrooke, 2500 Boulevard de l’Universit
e, Sherbrooke, QC
J1K 2R1, Canada
3
Qu
ebec Center for Biodiversity Science, McGill University, 1205 Dr. Penfield
Avenue, Montr
eal, Qu
ebec, Canada H3A 1B1
4
Department of Forest Resources, University of Minnesota, St. Paul, MN
55108, USA
5
Hawkesbury Institute for the Environment, Western Sydney University,
Penrith, NSW 2753, Australia
6
Department of Ecology, Evolution and Behavior, University of Minnesota, St.
Paul, MN 55108, USA
7
Institut des sciences de la for
^
et temp
er
ee (ISFORT), Universit
eduQu
ebec en
Outaouais (UQO), 58 rue Principale, Ripon, QC J0V 1V0, Canada
*Correspondence: E-mail: alain.paquette@gmail.com
© 2016 John Wiley & Sons Ltd/CNRS
Ecology Letters, (2016) 19: 638–647 doi: 10.1111/ele.12600
sites (Reich 2012). De facto, both approaches have been
shown to be good predictors of ecosystem functioning in
herbaceous (Mokany et al. 2008; Roscher et al. 2012) and tree
systems (Nadrowski et al. 2010).
Among natural assemblages, SR and FD are inevitably
linked, which makes it difficult to disentangle their respective
influence on ecosystem functioning (Naeem 2002). In addition,
the relationship between SR and FD is likely to be nonlinear.
While at low SR, each species addition may linearly increase
FD, the relationship may plateau at higher levels of SR due
to functional redundancy. As a consequence, it has been
hypothesised that the relationship between FD and ecosystem
functions ought to be positive and more linear than between
SR and ecosystem functions (Tobner et al. 2014). However,
very few published studies so far have manipulated FD varia-
tion and tested for its effect on ecosystem functioning inde-
pendently of SR (Reich et al. 2004), and even fewer have
done so with trees (but see www.treedivnet.ugent.be).
We conducted a common garden experiment using high-
density tree communities to assess the relationship between
FD and stem biomass increment (a proxy for productivity;
hence this term is used thereafter), independently of SR (Tob-
ner et al. 2014). The core of this experiment consists of species
mixtures of identical SR varying in FD. Here, we present the
first results of this experiment, 4 years after its establishment.
We investigated the following hypotheses:
(1)FD and CWMs both explain tree mixture productivity, but
with the former being a stronger predictor, and
(2)Net diversity effects increase over time (i.e. 4 years), driven
primarily by increases in complementarity.
METHODS
Site description
The study site is located at Ste-Anne-de-Bellevue, near Mon-
treal, Qu
ebec, Canada (45°26
0
N, Long 73°56
0
W, 39 m.a.s.l.).
Mean annual temperature is 6.2 °C with a mean annual pre-
cipitation of 963 mm (climate.weatheroffice.gc.ca). The study
site is a flat former agricultural field that has been intensively
managed for decades. The soil consists of a 20–70 cm deep
sandy layer (91 3.7% sand, 6 3.0% clay, 3 2.1% silt,
mean standard deviation) overtopping clay.
In spring 2009, an area of 0.6 ha was cleared of corn debris
before trees were planted with seedlings 1 or 2 years old (Tob-
ner et al. 2014) (Appendix S1). The species pool comprised of
12 North American temperate forest species, namely five
broad-leaf species and seven conifers (Appendix S1). This
experiment is part of the ‘International Diversity Experiment
Network with Trees’ (IDENT) that includes several sites in
North America and Europe (Tobner et al. 2014).
Experimental design
Trees were planted in square plots of 8 9 8 individuals, with
50 cm spacing among trees and 1.25 m between plots to allow
movement and minimise interplot interactions. For the latter
reason, tree roots were also sliced vertically 30 cm deep
around each plot in the third and fourth growing seasons
(2011 and 2012). Plot types were monocultures of all 12
species, 14 combinations of two-species mixtures, 10 combina-
tions of four-species mixtures and one mixture including all
12 species (Tobner et al. 2014). Each community was repli-
cated four times in a randomised block design for a total of
148 plots and 9472 trees. Within plots, species were planted at
random and equal proportions with some restrictions: in two-
species mixtures, at least two of the eight neighbours had to
be different species. In four-species mixtures, at least two of
the eight neighbours had to be from two different species. In
the 12 species mixture, all species had at least five individuals,
and four had six (for a total of 64).
The distribution of trees within plots was identical in all
four blocks. However, the distribution of plots was rando-
mised for each block. Around the outermost rows of the
experiment, three rows of trees at 50 cm distance were planted
to minimise edge effects and a resource for replacing dead
trees – 52 trees died and were replaced after the first year. A
fence to protect against herbivory surrounded the experiment
and all plots were regularly weeded manually to minimise
herbaceous competition.
Calculation of functional identity and diversity
The effects of FD and identity (CWMs) on productivity were
tested. To calculate FD, we chose functional dispersion (FDis,
Lalibert
e & Legendre 2010; Tobner et al. 2014). FDis is the
mean distance of each species to the centre of mass of all spe-
cies in a multidimensional trait space and produces indices
similar to Rao’s quadratic entropy but offers advantages per-
tinent to this study: species’ abundances can be used to weight
distances; in addition to accepting any number and type of
trait (i.e. continuous to qualitative), FDis resolves for commu-
nities with as few as two species, and allows weighting of indi-
vidual traits (Lalibert
e & Legendre 2010).
Functional dispersion was calculated twice: once prior to
planting only for constructing the initial FD gradient (Fig. 1),
and once more for post-planting data analyses with updated
trait data from the literature and in situ measurements. Prior
to planting, FDis were therefore calculated for all possible
combinations of two and four species from the 12 species pool
and 14 two-species and 10 four-species mixtures were chosen
to comprise the initial FD gradient (Fig. 1) (Tobner et al.
2014). Those initial FDis indices were computed using 12
above- and belowground traits to capture plant resource-use
strategies linked to a wide range of ecosystem processes, and
weighted simply using planted abundances. For post-planting
analyses FDis were weighted by species’ relative importance
measured using stem volume (diameter
2
9 height)
(Appendices S2 and S3 for details).
Community weighted means of trait values (CWM) were
computed for each trait following Lavorel et al. (2008) as the
mean trait value of all species present weighted by their rela-
tive importance (Appendices S2 and S3). Stem volume per
plot and species, FDis and CWM were calculated including
all 8 9 8 individuals in a plot as they impact on the inner
6 9 6 trees used for computing response variables, whereas
plot-level productivity (response variable) was assessed for the
© 2016 John Wiley & Sons Ltd/CNRS
Letter Diversity effects in tree communities 639
inner 6 9 6 only to minimise edge effects from neighbouring
plots.
Aboveground stem productivity
Tree height (H) and diameter (D, at 5 cm from ground) were
measured at the end of each growing season from 2009 to
2012. An approximation of stem biomass was calculated for
each year as D
2
9 H 9 wood density. Mortality was negligi-
ble over the first 4 years. In 2012, 39 out of the 5328 trees of
the inner 6 9 6 had died and 122 showed signs of crown dam-
age, mostly due to insect herbivory or snow. In 2011, wasp
nests prevented the measurement of 35 trees. To accommo-
date for these missing values, biomass per plot was calculated
as the species mean biomass of measured trees multiplied by
the number of planted individuals of each species, summed
for all species, all within the same 6 9 6 subplot. Net diversity
effects (NE), complementarity effects (CE) and selection
effects (SE) were calculated with these values.
Biodiversity effects
Several metrics are used to partition the effects of diversity on
ecosystem functioning (most often yield). The common aspect
to all of those metrics is the comparison between a species’
yield in mixture and its yield in monoculture, also called rela-
tive yield (RY, De Wit 1960). Relative yield can be calculated
for each component species or for the whole community (rela-
tive yield total, RYT, Vandermeer & Goldberg 2003). For a
two-species mixture of species A and B, RYT is calculated as
follows (dimensionless):
RYT ¼
yield
A
ðmixtureÞ
yield
A
ðmonocultureÞ
þ
yield
B
ðmixtureÞ
yield
B
ðmonocultureÞ
ð1Þ
where yield is expressed as unit per area (e.g. ton biomass per
hectare). RYT = 1 indicates no diversity effect (i.e. the perfor-
mance of species in mixture is equal to their performance in
monoculture). RYT < 1 indicates a negative and RYT > 1a
positive mixture effect.
However, the RY approach does not allow distinguishing
the mechanisms underlying diversity effects. A related
approach is also based on the calculation of a general devia-
tion of yield in mixture (observed yield, Y
O
) from that in
monoculture (expected yield, Y
E
). This (net) diversity effect
(NE - dimensionless) can be partitioned into complementarity
(CE) and selection effects (SE) (Loreau & Hector 2001,
eqn 2).
NE ¼ Y
O
Y
E
¼ CE þ SE
¼ N
DRY M þ N covðDRY; MÞð2Þ
Both constituent effects (CE and SE) hinge on the
calculation of RY, in this case expressed as ΔRY:
DRY
A
¼
yield
A
ðmixtureÞ
yield
A
ðmonocultureÞ
P
A
ð3Þ
where P is the proportion of the species in mixture (i.e. in
terms of density).
CE is the mean of the change in species’ relative yields
(
DRY) multiplied by the mean of species’ monoculture yields
(
M) times the number of species (N) (eqn 2, Loreau & Hector
2001). CE averages positive and negative diversity effects of
all species present in the mixture. SE is calculated as the
covariance between species’ relative (ΔRY) and monoculture
yields (M) multiplied by the number of species present in the
community (N, eqn 2). SE is positive when more productive
species in monoculture perform better in mixture than in
monoculture, and alternatively negative when less productive
species perform better in mixtures than in monoculture. In
case of overyielding of both high and low productive mono-
culture species, SE can be positive or negative, depending on
the stronger effect (see Appendix S4 for illustrations).
Using NE, CE and SE offers the advantage of linking diver-
sity effects to underlying mechanisms. However, some
caveats have to be kept in mind, especially when working with
Species richness
1
2
4
12
Funconal diversity
PruPre
BaQr
BpQr
PgPs
AbAr
AsTo
AsBa
PgTo
LlPs
QrTo
AbAs
PgPru
AbBp
LlPg
BaPru
PrePs
AbPg
PruPre
All
LlPs
PrePs
Mono-
cultures
AbAs
ArPs
QrTo
PgPru
PrePs
ArBa
BpQr
1
2
3
4
5
6
7
8
LlPg
ArTo
ArBa
BpPs
AsBp
PgPs
AsLl
Figure 1 Schematic representation of the experimental design (replicated four times). Communities are implemented along a gradient of species richness (SR)
and functional diversity (FD). FD was calculated as functional dispersion (FDis) (Lalibert
e & Legendre 2010) based on 12 above- and belowground traits
(see section FD calculation). Smaller superposed squares indicate the replication of different communities with similar FD resulting in a total of 14 two-
species mixtures and 10 four-species mixtures. Species codes are Ab, Abies balsamea; Ar, Acer rubrum; As, Acer saccharum; Ba, Betula alleghaniensis; Bp,
Betula papyrifera; Ll, Larix laricina; Pg, Picea glauca; Pru, Picea rubens; Pre, Pinus resinosa; Ps, Pinus strobus; Qr, Quercus rubra; To, Thuja occidentalis.
© 2016 John Wiley & Sons Ltd/CNRS
640 C. M. Tobner et al. Letter
trees. NE, CE and SE are sensitive to absolute values
(i.e. monoculture yields) and strongly weight the contribution
of higher yielding species (Fridley 2003, Appendix S4 for illus-
tration). To assess the relative importance of the diversity
effect, we also calculated RYT at the plot level and RY for
each species.
Data analysis
We first used a simple random effect model with REML esti-
mation to test for the effects of species richness (SR; 1, 2, 4
and 12) on aboveground stem biomass accumulated after
4 years, with block and plot[SR] as random factors (noted R),
and an error term ɛ:
logBiomass ¼blockðRÞþSR þ SR blockðRÞ
þ plot½SRðRÞþe
ð4Þ
Plot[SR] (i.e. the different assemblages of species within a
SR level), was added to account for the large differences in
the sources of variations between the four different SR levels.
We calculated NE (i.e. positive NE equalling overyielding) to
test whether mixtures performed better than expected com-
pared with their respective monocultures. NE, CE and SE
were calculated for each plot. Two-tailed t-tests were applied
(n = 4 blocks) to test when diversity effects were significantly
different from zero.
A second model was applied to test for the relationships of
functional diversity (FDis) and identity (CWM) on response
variables (i.e. biomass and diversity effects, Y
0
) including two-
and four-species mixtures only, and microtopography (the dif-
ference in elevation between plot centres measured on site;
range = 36 cm) to account for slight differences in drainage:
Y0¼microtopography þ blockðRÞþFDis
1
þþFDis
n
þ CWM
1
þþCWM
n
þ e ð5Þ
where ‘FDis’ is functional dispersion, and ‘CWM’ community
weighed means. Models included two- and four-species mix-
tures only. Residuals were checked for heteroscedasticity and
the data were log transformed in one case (CWM of leaf longev-
ity). FDis and CWMs to include into the model were chosen
using visual estimation of relatedness to Y
0
through redundancy
analyses (see Appendix S3 for full list of traits), as follows. Step-
wise regression was applied to select FDis and CWMs with sig-
nificant effects (p < 0.05) on the response variable. Constraints
were added to the stepwise selection: i) CWM of wood density
was omitted from analyses since that information was used in
biomass calculations and ii) only one trait from suites of traits
with known correlations in leaves (e.g. LMA, leaf area, leaf
nitrogen) or roots (SRL, fine root diameter, branching inten-
sity) was used. CE was tested against FDis and SE against
CWMs only, as conceptually suggested. We also tested if there
was a relationship between the residuals of the model and SR
but no significant results were found.
We tested if the effect of diversity changed over the 4 years
with repeated measures
ANOVA including microtopography,
year and SR as treatments and relative yield totals as response
variable. In case of significant effect of year, differences in
RYT between years were tested against zero in a one-sample
t-test. A correction for multiple comparisons similar to the
one used in Tukey HSD tests was applied.
RESULTS
Averaged over blocks, aboveground stem biomass ranged
from 14.5 kg to 98.1 kg per plot (11.8 and 80.1 Mg ha
1
)
after the fourth year of the experiment (Appendix S5 and S8)
but did not show significant variation among species richness
levels (Fig. 2). Monocultures of L. laricina produced the high-
est biomass, followed by mixtures that included B. papyrifera
(Appendix S5). Although biomass was highest in the 12 spe-
cies plots in absolute values, and lowest in monocultures,
sources of variations were large by design in two- and four-
species plots (where composition varied over a FD gradient),
which is illustrated by the large coefficient of determination
(R
2
= 0.93) obtained by accounting for plot identity (eqn 4).
In other words, the observed variation in productivity was
due for the most part to species composition, not their
number.
Transgressive overyielding (i.e. where mixture yield exceeds
the highest monoculture yield) was not observed. NE were
mostly not significantly different from zero but included some
significant positive mixture effects. Positive NE occurred in
two two-species mixtures, two four-species mixtures and
in the 12-species mixture (Fig. 3, Appendix S8). NE ranged
from 3.6 kg (2.9 Mg ha
1
) for the A. rubrum and B. al-
leghaniensis mixture to +44 kg (35.9 Mg ha
1
) for the A. sac-
charum, B. papyrifera, P. glauca and P. strobus mixture
(Fig. 3).
Positive mixture effects, when present, were largely due to
significant selection effects (SE) (Appendix S8). Although
SE had no significant negative values and many positive
(9), more than half (i.e. 14) of the communities had nega-
tive absolute values for CE, however, none were significant
Figure 2 Stem biomass measured in 2012 by species richness (means and
standard deviation, across all blocks and plots). Random effect model
(eqn 4) did not find a significant effect of species richness (P < 0.001)
(P = 0.56; R
2
= 0.93).
© 2016 John Wiley & Sons Ltd/CNRS
Letter Diversity effects in tree communities 641
(Fig. 3; Appendix S8). SE was larger than CE in seven of
the eight mixtures with greatest overyielding (including the
12-species plots). A total of 190 kg (155 Mg ha
1
) were
overyielded in the two-and four-species mixtures with signif-
icant positive effects, of which 79% (123 Mg ha
1
) was due
to SE and the other 21% (32 Mg ha
1
) to CE. Almost
90% of the overyielding occurred in eight mixtures, includ-
ing three two-species and four four-species mixtures, as well
as the one with all 12-species (Fig. 3). All were mixtures of
deciduous and evergreen species (except B. papyrifera,
Q. rubra) and all but one contained either or both B. pa-
pyrifera and L. laricina.
Diversity indices
In general, mean trait values (CWMs) explained a larger pro-
portion of variance of stem biomass and NE than trait varia-
tion (FDis). Six indices were selected by the stepwise method
that together with microtopography explain more than 90%
of variation in aboveground stem biomass. The diversity
indices with the greatest F-ratios include CWMs of leaf long-
evity, rooting depth and fine root branching intensity (Fig. 4).
The sign for the first two indices is negative, indicating that
greater aboveground biomass was achieved in communities
with shorter leaf life spans and shallower root systems. The
positive sign for branching intensity indicates that greater
aboveground stem biomass was achieved in communities with
dominance of highly branched fine roots (Fig. 4). Other sig-
nificant predictors of aboveground stem biomass included
CWMs of seed mass and leaf nitrogen and a small but signifi-
cant effect of variation in seed mass (Fig. 4).
Indices selected to predict NE largely overlap those for
aboveground stem biomass. Those with greatest explanatory
power on NE were CWMs of fine root branching intensity
(positive sign) and rooting depth (negative sign) as well as
trait variation (FDis) in seed mass (positive sign). Together
with CWM of leaf longevity and seed mass, FDis of leaf
nitrogen and microtopography, 47% of variation in NE was
explained (Fig. 4). Looking at the two aspects of NE sepa-
rately, CWMs of shade tolerance, leaf nitrogen and branching
intensity together with microtopography explained 31% of
variation in SE. The strongest effect on SE was related to
community trait means of shade tolerance followed by CWMs
of leaf nitrogen and fine root branching intensity (Fig. 4).
However, no FDis index had a significant effect on CE.
Microtopography alone explained 24% of variation in CE
(Fig. 4).
Diversity indices (RY) over time
RYT was greater or equal to one in all years when aver-
aged over all mixtures (i.e. 2009 through 2012). RYT
increased significantly from 2009 to 2010 but decreased in
2011 and 2012 to levels similar to 2009, with no significant
difference between SR two and four (Fig. 5). Relative yield
for a given species varied from 0.5 for T. occidentalis to 3.1
for B. papyrifera (both in four-species mixtures). Diversity
had a significant effect on most species, which was mostly
negative to no effect (RY ≤ 1) for coniferous species and
no effect to positive (RY ≥ 1) for broad-leaved species
(Fig. 6, Appendix S6). Changes in RY over time were spe-
cies specific. While RY for B. papyrifera and L. laricina sig-
nificantly increased, RY for A. balsamea, P. glauca,
P. strobus and T. occidentalis significantly decreased from
2009 to 2012 (Appendix S6). RY for all the other species
did not change significantly over time.
–5
5
15
25
35
45
55
65
75
85
ArBa
AsTo
PrePs
AsLl
ArTo
LlPs
PruPre
PgPs
BaQr
AbAs
AbAr
BpQr
LlPg
BpPs
AbPgPruPre
PgPruPrePs
ArPsQrTo
AsBaPgTo
ArBaBpQr
AbAsPgPru
BaPruPrePs
LlPsQrTo
AbBpLlPg
AsBpPgPs
All
Biodiversity effects 2012 (kg)
Tree community
2-species mix
4-species mix
12-species mix
*#
*#
*#
*#
*
#
SE CE
NE =
+
*
#
#
#
#
#
Figure 3 Biodiversity effects by tree community. Net biodiversity effect (NE, filled circles) + standard deviation (for the four blocks) and its two
components: complementarity effects (CE, light red bars) and selection effects (SE, dark red bars). Biodiversity effects were calculated on mean species
biomass in each mixture in the fourth year of the experiment, 2012. Communities with significant NE are annotated with *, whereas significant selection
effects are noted using # (P < 0.05). See Fig. 1 for species codes.
© 2016 John Wiley & Sons Ltd/CNRS
642 C. M. Tobner et al. Letter
