Content uploaded by David A Wedin
Author content
All content in this area was uploaded by David A Wedin on Dec 13, 2013
Content may be subject to copyright.
Content uploaded by David A Wedin
Author content
All content in this area was uploaded by David A Wedin
Content may be subject to copyright.
Diversity and Productivity in a
Long-Term Grassland
Experiment
David Tilman,
1
* Peter B. Reich,
2
Johannes Knops,
3
David Wedin,
4
Troy Mielke,
1
Clarence Lehman
1
Plant diversity and niche complementarity had progressively stronger effects
on ecosystem functioning during a 7-year experiment, with 16-species plots
attaining 2.7 times greater biomass than monocultures. Diversity effects were
neither transients nor explained solely by a few productive or unviable species.
Rather, many higher-diversity plots outperformed the best monoculture. These
results help resolve debate over biodiversity and ecosystem functioning, show
effects at higher than expected diversity levels, and demonstrate, for these
ecosystems, that even the best-chosen monocultures cannot achieve greater
productivity or carbon stores than higher-diversity sites.
Recent demonstrations that greater plant di-
versity can lead to greater productivity (1–5)
have generated considerable debate (6–14).
This has been fueled by uncertainty about
which of many alternative hypotheses (6,9,
15–20) is operating in nature and about the
number of species required to maintain eco-
system functioning (6,7,12,14). We report
results of a long-term experiment that allows
tests of these alternative hypotheses.
It has been hypothesized that productivity
may be greater at higher diversity because of
“niche complementarity” among particular
combinations of species and the greater
chance of occurrence of such combinations at
higher diversity (15–20). Niche complemen-
tarity, which results from interspecific differ-
ences in resource requirements and in spatial
and temporal resource and habitat use, or
from positive interactions (21), is predicted to
allow stable multispecies coexistence and
sustainably greater productivity at higher di-
versity (17,18). Alternatively, it has been
hypothesized that reported diversity effects
might be short-lived transients caused solely
by the presence of some species with high
growth rates (6); be experimental artifacts
resulting solely from species pools containing
some low-viability species (6); or result from
the most productive species being the best
competitor (6,9,17). These “sampling ef-
fects” all result from the greater chance of
any given species being present at higher
diversity and from dynamics that cause a
single species to dominate and determine
ecosystem functioning (6,9,17,19).
Sampling and complementarity have dif-
ferent signatures (6,9,12,17–20).Sampling
effects limit the maximal productivity of
higher-diversity plots to that of the best mono-
culture, giving an upper bound of variation in
community performance that is independent
of diversity. With niche complementarity, the
upper bound increases with diversity because
no monoculture is as productive as some
combinations of two species and no combi-
nation of Nspecies is as productive as some
combinations of N⫹1 species.
In a 7-year experiment [(4), supplement A
(22)], we controlled one component of diversi-
ty, the number of plant species, in 168 plots,
each9mby9m.Weseeded the plots, in May
1994, to have 1, 2, 4, 8, or 16 species, with 39,
35, 29, 30, and 35 replicates, respectively. The
species composition of each plot was chosen by
random draw from a pool of 18 grassland pe-
rennials that included four C4 (warm-season)
grasses, four C3 (cool-season) grasses, four le-
gumes, four nonlegume forbs, and two woody
species. All species occurred in monoculture,
and all but three were in at least two monocul-
ture plots, allowing comparison of responses of
each species in monoculture to higher-diversity
combinations of these same species. We do not
use 76 additional plots that had functional
group compositions drawn from an augmented
species pool or 46 plots planted to 32 species
(4) because the additional species were not
grown in monoculture, and combining results
from different species pools could introduce
bias. We focus analyses on species number,
because it was directly controlled, and function-
al group composition (23), because of its hy-
pothesized importance. Other measures of di-
versity, including number of species per func-
tional group and the presence or absence of
species or functional groups, are highly corre-
lated with species number and show similar
responses.
In our grasslands, plant aboveground liv-
ing biomass, because it is all produced within
a growing season, is an index of primary
productivity. In contrast, total biomass
(aboveground plus belowground plant bio-
mass) measures carbon accumulated in living
tissues. Both aboveground and total biomass
increased highly significantly with species
number each year (Fig. 1, A and B, and Table
1), and functional group composition ex-
plained a highly significant amount of the
residual variation (Table 1). Moreover, when
the effect of each variable was determined
after controlling for effects of the other (type
III regressions), effects of functional group
composition predominated in the early years
[as for (24,25)], but species number had
highly significant positive effects on both
aboveground and total biomass by 1999 and
2000, showing the simultaneous importance
of species number and functional group com-
position in the long-term (Table 1).
The initial saturating dependence of
aboveground and total biomass on species num-
ber (Fig. 1, A and B) became, by 2000, a linear
increase for species number ⱖ2. In 2000,
16-species plots had 22% greater aboveground
biomass total and 27% greater total biomass
than 8-species plots (both differences signifi-
cant; ttests: P⫽0.018, P⫽0.002, respective-
ly). The dependence of biomass on species
number and functional group composition be-
came progressively stronger, explaining about
one-third of variance in 1997 and two-thirds in
2000 (Table 1). This strengthening of the effect
of diversity and the increasingly steep and lin-
ear trends (Fig. 1, A and B) fail to support the
hypothesis (6) that diversity effects were short-
lived transients. Comparable and significant
(P⬍0.01) dependences of total and
aboveground biomass on diversity and compo-
sition were observed when analyses used the
actual number of planted species observed in
each plot (Fig. 1C) or the Shannon diversity
index [supplement B (22)].
We tested the low-viability sampling hy-
pothesis by identifying the five species that
attained least total biomass in monoculture in
2000 and excluding from analysis plots con-
taining any combinations of just these spe-
cies. Total biomass was still significantly de-
pendent on species number and functional
group composition in the remaining 131 plots
[general linear model (GLM) type III regres-
sion: F
OVERALL
⫽8.39, P⬍0.001; F
DIV
⫽
11.6, F
COMP
⫽4.36, P⬍0.001 for each].
Similar results occurred when we excluded
from analysis of aboveground biomass plots
containing any combinations of the five spe-
cies with least aboveground biomass in mono-
culture (GLM type III regression: F
OVERALL
⫽7.08, P⬍0.001; F
DIV
⫽10.6, P⫽0.0014;
F
COMP
⫽3.84, P⬍0.001). In another anal-
ysis, we excluded the 30 plots with the lowest
total biomass in 2000 (total biomass ⬍400 g
1
Department of Ecology, Evolution and Behavior, Uni-
versity of Minnesota, St. Paul, MN 55108, USA.
2
De-
partment of Forest Resources, University of Minneso-
ta, St. Paul, MN 55108, USA.
3
School of Biological
Sciences, University of Nebraska, Lincoln, NE 68588,
USA.
4
School of Natural Resource Sciences, University
of Nebraska, Lincoln, NE 68583, USA.
*To whom correspondence should be addressed. E-
mail: tilman@umn.edu
REPORTS
www.sciencemag.org SCIENCE VOL 294 26 OCTOBER 2001 843
m
⫺2
). Species number and composition still
had highly significant effects on total bio-
mass in the remaining plots (GLM type III
regression: F
OVERALL
⫽5.05, P⬍0.001;
F
DIV
⫽12.3, P⬍0.001; F
COMP
⫽2.82, P⬍
0.001). Similarly, when we excluded the 31
plots with lowest aboveground biomass
(⬍100gm
⫺2
) in 2000, effects of composi-
tion and species number remained highly sig-
nificant (GLM type III regression: F
OVERALL
⫽3.77, P⬍0.001; F
DIV
⫽10.5, P⬍0.0016;
F
COMP
⫽2.28, P⫽0.0015). Similar results
occurred with lower and higher cutoffs (in-
cluding 50% higher) for aboveground and
total biomass. In total, the dependence of
biomass on species number and composition
was not explained solely by sampling effects
for a species pool containing some poorly
performing species.
We tested the sampling hypothesis that the
most productive species determined the effects
of diversity (6,9,17) by retaining in analyses of
year 2000 results only plots containing at least
one of the nine species with the highest mo-
noculture total biomass in 2000. Total biomass
remained significantly dependent on species
number and functional group composition in
these 145 plots, and in the subset of 95 plots
that contained at least two of these nine species
[type III regressions; supplement C (22)]. Sim-
ilar results occurred for aboveground biomass
in 2000 [supplement C (22)]. These analyses
fail to support the sampling hypothesis. Anoth-
er test comes from examining performance of
higher-diversity plots relative to the best mo-
noculture (Fig. 2). In 1999 and 2000, many
higher-diversity plots had greater aboveground
and total biomass than the single best-perform-
ing monoculture (Fig. 2). The percentage of
such plots was an increasing function of diver-
sity, on average for 1999 and 2000, with about
half of the 16-species plots having greater
aboveground or total biomass than the best
monocultures (Fig. 1D). The strength and re-
peatability of this increasing upper bound in
both aboveground and total biomass support the
importance of niche effects and refute the hy-
pothesis that sampling effects were the sole
explanation for the long-term effects of diver-
sity. The coexistence of most species, with
about 12 planted species per 2 m
2
persisting in
each 16-species plot (Fig. 1C), further supports
niche complementarity. However, in earlier
years, such as 1997, few high-diversity plots
had greater biomass than top monocultures, and
the percentage was independent of species
number (Fig. 1D), which is consistent with the
sampling hypothesis (6,7,12).
The increasing importance of complemen-
tarity and the increasingly linear effects of
species number raise another question. Did
complementarity occur among most spe-
cies—i.e., did most species contribute to in-
creasing community biomass—or is there a
smaller set of species with complementary
interactions, with this set being increasingly
likely to co-occur at higher diversity (19)?
We used analysis of variance (ANOVA) to
determine the simultaneous effects of the pres-
ence or absence of each species (entered as
main effects) on aboveground or total biomass
(one test per year, with 4 years for total bio-
mass, 5 years for aboveground biomass). Three
or four species had significant (P⬍0.05) pos-
itive effects on aboveground or total biomass in
most years. Among legumes, Lupinus perennis
had significant effects in all nine tests, Lespe-
Fig. 1. The dependence of (A) plant aboveground biomass and (B) total biomass (aboveground plus
belowground living plant mass) on the number of planted species. Data are shown as the mean ⫾
SE. (C) The relation between the number of species planted in a plot and the actual number
(mean ⫾SE) of planted species visually observed ina2m
⫺2
area of each plot. (D) The percentage
of all plots of a given planted diversity level, on average for 1999 and 2000 combined, or on average
for 1997, that had greater biomass than the single monoculture plot with the greatest biomass.
Fig. 2. The dependence of aboveground (Aand B) and of total (Cand D) biomass of each plot on
planted species number for 1999 and 2000. The broken line shows the biomass of the top
monoculture for a given year. The solid line is a regression of biomass on the logarithm of species
number. Logarithm of species number was used in the figure because it gave slightly better fits, but
was not used in Table 1 because it often gave slightly lower R
2
values than species number.
REPORTS
26 OCTOBER 2001 VOL 294 SCIENCE www.sciencemag.org844
deza capitata in six tests, and Petalostemum
purpureum in two tests. Schizachyrium scopa-
rium and Sorghastrum nutans, both C4 grasses,
were significant in five tests each. These are
five of the six most abundant species in mix-
tures. A rarer forb also had a significant effect.
Similarly, when plots were characterized by the
presence or absence of functional groups in
ANOVAs, in 2000 there were significant posi-
tive effects of legumes (P⬍0.001), forbs (P⬍
0.05), and C4 grasses (P⬍0.01) on
aboveground biomass, and significant positive
effects of legumes (P⬍0.001) and C4 grasses
(P⬍0.001) on total biomass. For aboveground
biomass, only the legume ⫻C4 grass interac-
tion was significant (P⬍0.05), and it was
positive. For total biomass, the legume ⫻C4
grass interaction was marginally significant
(P⫽0.068) and biased toward positive (26),
suggesting complementarity or facilitation
among legumes and C4 grasses (4). However,
even after controlling for the presence or ab-
sence of all functional groups, there were pos-
itive (P⬍0.02) effects of species number on
both aboveground and total biomass in 2000,
indicating that biomass also depended on spe-
cies number rather than on just the presence of
functional groups.
Although these analyses suggest that the
presence of about five dominant species
might explain much of the effects of diversi-
ty, there may be small but additive effects of
rarer species. To test this possibility, we
ranked all species on the basis of their abun-
dance ( percentage cover) in 16-species plots
in 2000, and created 17 new diversity indices.
Each index states how many of the Nmost
abundant species (N⫽2, 3,. . .18) had been
planted in each plot. We then determined
which diversity index (log transformed) ex-
plained the most variance in aboveground or
total biomass in 2000. Aboveground biomass
was most dependent on how many of the 9 to
13 most abundant species were planted in
each plot [supplement D (22)], showing that
many rarer species contributed detectably to
aboveground biomass. However, total bio-
mass in 2000 was most dependent on how
many of the four most abundant species were
planted [supplement D (22)], likely because
three of these four are C4 grasses, the species
that accumulate the greatest root mass.
In summary, diversity effects were neither
transients, nor explained in the long-term solely
by other sampling-effect hypotheses, nor solely
by the presence of legumes on a low-Nsoil.
Rather, niche complementarity contributed sig-
nificantly. Plant species number [as in (1–5,20,
21)] and functional group composition [as in (4,
5,24,25)] became simultaneously and approx-
imately equally important in our long-term ex-
periment. Compared with the average of the
single best species in monoculture, our 16-spe-
cies plots had 39% greater aboveground bio-
mass and 42% greater total biomass on average
for 1999 and 2000. Moreover, 16-species plots
in 1999 and 2000 had 2.7 to 2.9 times greater
aboveground and total biomass than the average
for all species in monoculture (Fig. 1A). The
nonsaturating increase in aboveground biomass
with diversity likely reflects niche effects among
about 9 to 13 species and their greater chance of
co-occurrence at higher diversity (19), whereas
such effects among about four species seem to
account for total biomass responses.
The demonstration that diversity effects
strengthened through time and were not the
result solely of sampling effects or functional
group composition should resolve aspects of
the biodiversity debate (6–14). Moreover,
our results suggest, for ecosystems in which
niche complementarity occurs, that even with
the wisest choices, monocultures may be less
productive and accumulate less living carbon
than many higher-diversity species combina-
tions. Our results show that ecosystem pro-
cesses are simultaneously influenced by di-
versity and composition, but long-term work
in additional ecosystems is needed to deter-
mine the generality of our results, to better
understand the effects of nonrandom commu-
nity assembly and disassembly, and to better
determine the implications of biodiversity for
ecosystem management.
References and Notes
1. S. Naeem, L. Thompson, S. Lawler, J. Lawton, R.
Woodfin, Nature 368, 734 (1994).
2. 㛬㛬㛬㛬 ,Philos. Trans. R. Soc. London Ser. B 347, 249
(1995).
3. D. Tilman, D. Wedin, J. Knops, Nature 379, 718
(1996).
4. D. Tilman, J. Knops, D. Wedin, P. Reich, M. Ritchie, E.
Siemann, Science 277, 1300 (1997).
5. A. Hector et al.,Science 286, 1123 (1999).
6. M. Huston, Oecologia 110, 449 (1997).
7. J. Grime, Science 277, 1260 (1997).
8. E. Garnier, M.-L. Navas, M. Austin, J. Lilley, R. Gifford,
Acta Oecol. 18, 657 (1997).
9. L. Aarssen, Oikos 80, 183 (1997).
10. J. Hodgson, K. Thompson, P. Wilson, A. Bogaard,
Funct. Ecol. 12, 843 (1998).
11. D. Wardle, Oikos 87, 403 (1999).
12. D. Wardle et al.,Bull. Ecol. Soc. Am. 81, 235 (2000).
13. J. Kaiser, Science 289, 1282 (2000).
14. L. Guterman, The Chronicle of Higher Education (13
October 2000), A24.
15. J. Vandermeer, The Ecology of Intercropping (Cam-
bridge Univ. Press, New York, 1989).
16. M. Swift, J. Anderson, in Biodiversity and Ecosystem
Function, E.-D. Schulze, H. Mooney, Eds. (Springer
Verlag, Berlin, Germany, 1993), pp. 15–41.
17. D. Tilman, C. Lehman, K. Thomson, Proc. Natl. Acad.
Sci. U.S.A. 94, 1857 (1997).
18. C. Lehman, D. Tilman, Am. Nat. 156, 534 (2000).
19. M. Loreau, Oikos 91, 3 (2000).
20. M. Loreau, A. Hector, Nature 412, 72 (2001).
21. C. P. H. Mulder, D. D. Uliassi, D. F. Doak, Proc. Natl.
Acad. Sci. U.S.A. 98, 6704 (2001).
22. Supplementary Web material is available on Science
Online at www.sciencemag.org/cgi/content/full/294/
5543/843/DC1.
23. There are 31 possible combinations of five functional
groups taken 1 to 5 at a time (2
5
⫺1), of which 28
combinations occurred in the 168 plots. The combi-
nation that occurred in a plot is referred to as its
functional group composition.
24. D. Hooper, P. Vitousek, Science 277, 1302 (1997).
25. 㛬㛬㛬㛬 ,Ecol. Monogr. 68, 121 (1998).
26. Adjusted mean total biomass in 2000, from ANOVA:
342gm
⫺2
for plots with neither C4 grasses nor
legumes; 832 g m
⫺2
for C4’s but not legumes; 564 g
m
⫺2
for legumes but not C4’s; 1234 g m
⫺2
for both.
27. We thank the National Science Foundation (NSF/
DEB 0080382 and NSF/DEB 9629566) and the
Andrew Mellon Foundation for support; J. Fargione,
S. Pacala, S. Levin, A. Dobson, and J. Reichman for
comments; and N. Larson, C. Bristow, and L. John-
son for assistance.
5 March 2001; accepted 20 August 2001
Table 1. Analyses, using general linear models, of the effects of number of
planted species (continuous variable; entered first using type I SS) and of
functional group composition (categorical variable; entered second) on
total biomass and on aboveground biomass, showing results for each year
when measured. N⫽168. Overall model df ⫽28 and error df ⫽139.
Species number df ⫽1, composition df ⫽27. The last columns show type
III effects of species number (entered second, after functional group
composition).
Variable analyzed Year
Overall Species number
(entered first)
Functional group
comp. (entered
second)
Species number
(entered second)
R
2
Fvalue PFvalue PFvalue PFvalue P
Total biomass 1997 0.32 2.26 0.001 9.80 0.002 2.06 0.004 3.88 0.051
Total biomass 1998 0.47 4.37 ⬍0.001 43.8 ⬍0.001 2.91 ⬍0.001 2.38 0.13
Total biomass 1999 0.60 7.31 ⬍0.001 94.2 ⬍0.001 4.09 ⬍0.001 7.11 0.009
Total biomass 2000 0.68 10.5 ⬍0.001 152. ⬍0.001 5.27 ⬍0.001 12.3 ⬍0.001
Aboveground biomass 1996 0.41 3.28 ⬍0.001 15.5 ⬍0.001 2.83 ⬍0.001 3.80 0.053
Aboveground biomass 1997 0.39 3.02 ⬍0.001 12.3 ⬍0.001 2.60 ⬍0.001 0.52 0.47
Aboveground biomass 1998 0.49 4.80 ⬍0.001 31.8 ⬍0.001 3.81 ⬍0.001 2.56 0.11
Aboveground biomass 1999 0.56 6.27 ⬍0.001 90.7 ⬍0.001 3.15 ⬍0.001 14.9 ⬍0.001
Aboveground biomass 2000 0.61 7.80 ⬍0.001 111. ⬍0.001 3.97 ⬍0.001 10.7 ⬍0.001
REPORTS
www.sciencemag.org SCIENCE VOL 294 26 OCTOBER 2001 845
