Plant Diversity and Productivity Experiments in European Grasslands

Abstract

At eight European field sites, the impact of loss of plant diversity on primary productivity was simulated by synthesizing
grassland communities with different numbers of plant species. Results differed in detail at each location, but there was
an overall log-linear reduction of average aboveground biomass with loss of species. For a given number of species, communities
with fewer functional groups were less productive. These diversity effects occurred along with differences associated with
species composition and geographic location. Niche complementarity and positive species interactions appear to play a role in generating diversity-productivity relationships
within sites in addition to sampling from the species pool.

Full text

Plant Diversity and Productivity
Experiments in European
Grasslands
A. Hector,
1
* B. Schmid,
2
C. Beierkuhnlein,
3
M. C. Caldeira,
4
M. Diemer,
2
P. G. Dimitrakopoulos,
5
J. A. Finn,
6
†H. Freitas,
4
‡
P. S. Giller,
6
J. Good,
6
R. Harris,
6
P. Ho¨gberg,
7
K. Huss-Danell,
8
J. Joshi,
2
A. Jumpponen,
7,8
C. Ko¨rner,
9
P. W. Leadley,
9
§
M. Loreau,
10
A. Minns,
1
C. P. H. Mulder,
7,8
㛳G. O’Donovan,
6
¶
S. J. Otway,
1
J. S. Pereira,
4
A. Prinz,
3
D. J. Read,
11
M. Scherer-Lorenzen,
12
E.-D. Schulze,
12
A.-S. D. Siamantziouras,
5
E. M. Spehn,
9
A. C. Terry,
11
A. Y. Troumbis,
5
F. I. Woodward,
11
S. Yachi,
10
#J. H. Lawton
1
At eight European field sites, the impact of loss of plant diversity on primary
productivity was simulated by synthesizing grassland communities with dif-
ferent numbers of plant species. Results differed in detail at each location, but
there was an overall log-linear reduction of average aboveground biomass with
loss of species. For a given number of species, communities with fewer func-
tional groups were less productive. These diversity effects occurred along with
differences associated with species composition and geographic location.Niche
complementarity and positive species interactions appear to play a role in
generating diversity-productivity relationships within sites in addition to sam-
pling from the species pool.
Because species differ in their ecological at-
tributes, the loss of biodiversity from local
communities may be detrimental to the eco-
system goods and services on which humans
ultimately depend (1). This issue has been the
subject of major recent research efforts using
experimental plant assemblages (2–6). How-
ever, differences in aims and approaches, and
the fact that experimental manipulations of
diversity have been restricted to single local-
ities, limit the ability of ecologists to make
generalizations and predictions. The design,
analysis, and interpretation of these experi-
ments are also complex (7), and the view that
the loss of plant species can be detrimental to
ecosystem functioning remains contentious
(8–11). In particular, the mechanisms under-
lying the relationship between species rich-
ness and ecosystem functioning are still the
subject of debate because of the difficulty in
identifying and interpreting the importance of
niche complementarity versus “sampling ef-
fects” (8,12,13). Here we report patterns of
aboveground plant biomass from the most
extensive experiment to date in terrestrial
ecosystems, and we examine the underlying
mechanisms.
We used standardized protocols to estab-
lish experimental assemblages of grassland
species (grasses and forbs) that varied in
species richness, and we measured above-
ground plant biomass production at two locali-
ties in the United Kingdom and at single sites in
Germany, Ireland, Greece, Portugal, Sweden,
and Switzerland (14,15). Sites differed widely
in climate and other major environmental
factors (Table 1). We simulated the loss of
plant species by removing the existing vege-
tation and seedbank and reestablishing plant
communities from seed (16). At each site, we
established five levels of species richness,
ranging from monocultures of grasses or
forbs to higher-diversity assemblages that ap-
proximately matched background levels of
diversity in comparable unmanipulated semi-
natural grasslands at each site (Table 2). If
reducing the number of species reduces pro-
ductivity because of a decrease in functional
diversity and therefore the amount of niche
space occupied in the resulting depauperate
community (2,4,6,17), then we expect, for
a given number of species, that productivity
will also be lower in communities with fewer
functional groups. To test this, we catego-
rized species into three functional groups:
graminoids (grasses), nitrogen-fixing legumes,
and other herbaceous species (herbs) and estab-
lished communities containing one, two, or
three of these groups. To replicate plant diver-
sity, each level of species richness and function-
al group richness was represented by several
different plant assemblages at each site (18).
Each assemblage contained a different species
or mixture of species. We used constrained
random selection from the local pool of grass-
land species (14,15) to form experimental plant
assemblages where all polycultures contained at
least one grass. To investigate the effects of
species composition, each assemblage was rep-
licated in a minimum of two plots including
monocultures of many of the species involved.
In total, the experiment comprised 480 plots
and 200 different plant assemblages (19).
Aboveground biomass patterns. Above-
ground plant biomass in the second year of the
experiment (an estimate of net annual
aboveground primary production) differed sig-
nificantly between sites [F
7,185
⫽24.73, P⬍
0.001 (Table 3)]. The productivity of plots with
eight species (the highest richness common to
all sites) ranged from 337 g m
⫺2
in Greece to
802gm
⫺2
in Germany (Table 1) and was
driven by environmental differences among
sites. Extreme northern and southern locations
in Sweden, Portugal, and Greece, where grow-
ing seasons are short and productivity is often
limited by temperature and water (20,21), had
the lowest biomass.
Species richness and functional group
richness had highly significant effects on
aboveground biomass; overall, assemblages
with lower diversity were less productive on
average [combined effect of species richness
and functional group richness: F
12,185
⫽7.01,
P⬍0.001 (Table 3)] (22). Because there was
no location-by-species richness interaction,
differences in slopes between sites were not
1
Natural Environmental Research Council (NERC)
Centre for Population Biology, Imperial College at
Silwood Park, Ascot, Berkshire, UK, GB-SL5 7PY.
2
In-
stitut fu¨r Umweltwissenschaften, Universita¨t Zu¨rich,
Winterthurerstrasse 190, Zu¨rich, Switzerland, CH-8057.
3
Lehrstuhl Biogeographie, Universita¨t Bayreuth, Bay-
reuth, Germany, D-95440.
4
Departmentos de Engen-
haria Florestal e de Botaˆnica, Universidade Tecnica de
Lisboa, Tapada da Ajuda, Lisboa, Portugal, PT-1399.
5
Biodiversity Conservation Laboratory, Department of
Environmental Studies, University of the Aegean, Kara-
doni 17, Mytilene, Lesbos, Greece, GR-811 00.
6
Depart-
ment of Zoology and Animal Ecology, University College
Cork, Lee Maltings, Prospect Row, Cork, Ireland.
7
Depart-
ment of Forest Ecology, Swedish University of Agricul-
tural Sciences, Umeå, Sweden, SE-90183.
8
Crop Science
Section, Department of Agricultural Research for North-
ern Sweden, Box 4097, Swedish University of Agricul-
tural Sciences, Umeå, Sweden, SE-90403.
9
Institute of
Botany, University of Basel, Schoenbeinstrasse 6, Basel,
Switzerland, CH-4056.
10
Laboratoire d’Ecologie, UMR
7625, Ecole Normale Supe´rieure, 46 Rue d’Ulm, F-72530
Paris Cedex 05, France, FR-75230.
11
Department of An-
imal and Plant Sciences, University of Sheffield, South
Yorkshire, U.K., GB-S10 2TN.
12
Max-Planck-Institute for
Biogeochemistry, Postfach 10 01 64, Jena, Germany,
D-07701.
*To whom correspondence should be addressed. E-
mail: a.hector01@ic.ac.uk
†Present address: Department of Agriculture, The
University of Reading, Earley Gate, Post Office Box
236, Reading, Berkshire, UK, GB-RG6 6AT.
‡Present address: Departamento de Botaˆnica, Univer-
sidade de Coimbra, 3000 Coimbra, Portugal.
§Present address: Ecologie des Populations et Com-
munaute´s, Universite´ Paris Sud XI, URA CNRS 2154,
Baˆtiment 326, Orsay Cedex, France, FR-91405.
㛳Present address: Victoria University of Wellington,
School of Biological Sciences, Post Office Box 600,
Wellington, New Zealand.
¶Present address: Department of Environmental Re-
source Management, University College of Dublin,
Belfield, Dublin, Ireland.
#Present address: Centre for Ecological Research,
Kyoto University, Kyoto 606-01, Japan.
RESEARCH ARTICLES
www.sciencemag.org SCIENCE VOL 286 5 NOVEMBER 1999 1123

significant [F
22,29
⫽1.25, P⫽0.287 (Table
3)]. The overall effect of decreasing species
richness was best described by a linear rela-
tion between productivity and the natural log-
arithm of the number of plant species
[F
1,185
⫽55.13, P⬍0.001 (Fig. 1A)], which
is similar to patterns reported from previous
single-location experiments (4,5) and pre-
dicted by theory (17,23). The log-linear re-
lation corresponds to an initially weak but
increasing reduction of productivity with de-
creasing species richness. Each halving of the
number of plant species reduced productivity
by approximately 80 g m
⫺2
on average.
Plant cover was reduced by loss of plant
species richness (F
1,185
⫽3.84, P⬍0.001).
Cover and aboveground biomass are likely to
be correlated, and biomass patterns may not
occur after controlling for differences in cov-
er (8,24). However, highly significant reduc-
tions in aboveground biomass with declining
plant species richness remained in multiple
regressions which included cover as a covari-
ate, and when plots with less than 80% cover
were excluded.
For a given number of species, assem-
blages with fewer functional groups were less
productive [F
2,185
⫽6.34, P⬍0.01 (Fig.
1B)]. A multiple regression using the (un-
transformed) number of functional groups,
after accounting for species richness (Table
3), revealed that the omission of a single
functional group reduced productivity by ap-
proximately 100 g m
⫺2
on average.
Importance of scale. When all sites were
analyzed together, the lack of a significant
location-by-species richness interaction de-
termined that the log-linear regression with
parallel slopes provided the best overall mod-
el (Table 3 and Fig. 1A). However, when the
data for individual sites are plotted separate-
ly, they look different, and when analyzed
alone, produce a variety of different models
(Fig. 2) corresponding to alternative qualita-
tive relationships between species richness
and ecosystem processes (25). There are two
explanations for this result: (i) all sites con-
form to the same underlying model, and ap-
parent differences between sites are due to
the lower sample sizes and statistical power
at each site; (ii) sites differ in their responses,
but the analysis is not powerful enough to
Table 1. Details of the eight field sites, including location [site, country,
degrees of latitude and longitude, and altitude above sea level (asl)]; climate
(mean January and July temperatures and annual precipitation); previous land
use (arable crops, horse grazing, fallow land, or none); method of site
preparation (methyl bromide fumigation or steam sterilization of the soil,
hand weeding only, or use of a sterile sand substrate); number of biomass
harvests; and mean aboveground biomass of the plant assemblages with eight
species. Aboveground biomass comprised all living and standing dead plant
material above 5 cm, harvested in two quadrats 20 cm by 50 cm once or
twice each season around the times of peak biomass (where two harvests
were taken, the values reported are the sum totals per plot). For brevity, we
refer to accumulated net annual aboveground biomass as productivity but
note that it provides only an estimate of the aboveground component of this
process. All vegetation was cut to a height of 5 cm at the times of harvest and
the clippings were removed. Plots received no fertilizer during the first 2 years
of the experiment.
Site Country Latitude Longitude Altitude
(m asl)
January
mean (°C)
July
mean (°C)
Annual
rain (mm)
Previous
land use
Site
preparation Harvests Biomass
(g m
⫺2
)
Bayreuth Germany 50°N 12°E 350 ⫺0.1 18.2 630 Arable Steam 2 802.2
Riverstick Ireland 52°N 08°W 75 5.4 15.6 1130 Grazing Bromide 2 767.5
Silwood UK 51°N 01°W 50 3.7 16.9 652 Grazing Bromide 1 683.4
Sheffield UK 53°N 01°W 137 4.5 17.2 788 None Sand 1 675.0
Lupsingen Switzerland 47°N 08°E 439 0.7 18.3 1046 Arable Hand 2 605.5
Lezirias Portugal 39°N 09°W 25 9.4 26.1 588 Grazing Heat 1 432.3
Umeå Sweden 64°N 20°E 12 ⫺7.0 16.1 600 Arable Hand 1 402.9
Mytilini Greece 39°N 27°E 30 9.6 26.5 682 Fallow Bromide 1 336.5
Table 2. The experimental design at each location, showing numbers of plots per species richness level
and for each level of functional group richness. Plant assemblages (where an assemblage is a particular
species or mixture of species) were replicated in two plots at each site, with the same assemblage
sometimes occurring at more than one site.
Species richness
1 2 3 4 8 111214161832
Germany 20 14 10 10 6
Ireland 20 16 4 20 10
Silwood 22 12 12 10 10
Sheffield 24 10 8 8 4
Switzerland 20 14 16 10 4
Portugal 28 10 10 4 4
Sweden 24 12 12 6 4
Greece 14 12 10 8 8
Functional richness
1 172 32 14 6 2
26840184
3 44442484684
Table 3. Summary of the analysis of second-year aboveground biomass. We present the combined effect
of the two richness terms and partition the separate species and functional group richness effects from
initial analysis of variance (ANOVA) into a linear contrast (regression) and a deviation from linearity; that
is, the quadratic and higher order polynomial terms (shown indented). Our experiment has multiple error
terms: Diversity terms are tested against the plant assemblage term, the site differences and the
assemblage term against the assemblage-by-location interaction, and the assemblage-by-location inter-
action against the overall residual. Nonsignificant block effects and locality-by-diversity interactions are
omitted.
Source of variation d.f. s.s. % s.s. m.s. FP
Locality 7 12,413,386 28.3 1,773,341 24.73 1.32 ⫻10
⫺10
Total richness effect 12 7,769,673 17.7 647,473 7.01 1.93 ⫻10
⫺10
Species richness 10 6,599,591 15.1 659,959 7.15 1.72 ⫻10
⫺7
Log-linear contrast 1 5,089,222 11.6 5,089,222 55.13 4.01 ⫻10
⫺12
Deviation 9 1,510,369 3.5 167,819 1.82 0.0675
Functional group richness 2 1,170,082 2.7 585,041 6.34 0.0022
Linear contrast 1 966,878 2.2 966,878 10.47 0.0014
Deviation 1 203,204 0.5 203,204 2.20 0.1396
Assemblage 185 17,079,328 38.9 92,321 1.29 0.2133
Locality ⫻assemblage 29 2,079,864 4.8 71,719 3.77 1.83 ⫻10
⫺8
Residual 235 4,465,007 10.2 19,000
Total 468 43,807,258 100.0 93,605
RESEARCH ARTICLES
5 NOVEMBER 1999 VOL 286 SCIENCE www.sciencemag.org1124

reveal a significant location-by-species rich-
ness interaction when sites are analyzed to-
gether. Much of the individual site deviation
from the overall log-linear model may be due
to lower within-site replication. There may
also be transient effects at this early stage of
the experiment that largely disappear by the
following year (26). For these reasons, and
for parsimony, we favor the more general and
powerful combined analysis, which shows
that differences between locations are not
significant and suggests that there may be a
single general relationship between species
richness and diversity across all sites.
Our results highlight the importance of
considering scale when studying relation-
ships between diversity and productivity
(14), as predicted by theory (23). Despite
large differences in productivity between lo-
cations and no clear relationship between
productivity and maximum within-site spe-
cies richness (Fig. 1A), within a site, produc-
tivity generally declines as species are lost,
reconciling apparent contradictions in the lit-
erature (27).
Multiple influences on productivity.
Our experiment reveals the relative roles of
richness, location, and species composition as
determinants of productivity; these key vari-
ables explained approximately 18, 28, and
39% of the total sums of squares, respectively
(Table 3). Although it accounted for a large
amount of the total variation, species compo-
sition was not statistically significant
[F
185,29
⫽1.29, P⫽0.21 (Table 3)] (28).
However, when we tested the presence in an
assemblage of a particular plant species or
functional group (29), of the 71 more com-
monly occurring species, 29 had significant
(P⬍0.05) effects on productivity, although
virtually all these effects were small (Fig. 3).
Only one species, the nitrogen-fixing Trifoli-
um pratense, had particularly marked effects.
On average, the omission of this species re-
duced productivity by approximately 360 g
m
⫺2
. We also found highly significant effects
from the presence of legumes and herbs when
considered collectively as functional groups.
Evidence for niche complementarity
and positive species interactions. There are
three processes through which the loss of
plant species richness could decrease produc-
tivity: (i) the “sampling effect” (17) or “se-
lection probability effect” (8), in which more
Fig. 1. Productivity de-
clines with the loss of
plant diversity. (A)
Overall log-linear re-
duction of above-
ground biomass with
the simulated loss of
plant species richness.
(B) Linear reduction
with the loss of func-
tional group richness
within species richness
levels. Points in (A) are
total aboveground bio-
mass for individual
plots; lines are slopes
from the multiple regression model using species richness on a log
2
scale. Silwood and Sheffield are labeled together as UK. In (B), assemblages with 11 species occurred
only at Silwood, whereas assemblages with 2, 4, and 8 species are represented at all sites, including the more productive, and therefore have a higher average biomass.
Table 4. Summary of regression analyses of the aboveground biomass of individual species across the
species richness gradients. Slopes are from simple regressions analyzing change in estimated biomass per
individual sown of a species with increasing log
2
number of species, after adjusting for differences
between blocks and sites. “Plots” gives the sample size for each regression, and “sites” gives the number
of locations from which they were derived.
Species Sites Plots (n) Slope SE
Achillea millefolium 4 35 0.44 0.277
Agrostis capillaris†368⫺0.15 0.087
Alopecurus pratensis†5 63 0.01 0.025
Anthoxanthum odoratum 5 54 0.10 0.079
Arrhenatherum elatius†3 74 1.17* 0.220
Dactylis glomerata 5 61 0.42* 0.113
Festuca rubra 356⫺0.12 0.057
Holcus lanatus†5 86 0.37* 0.134
Lolium perenne 2 37 0.47* 0.135
Lotus corniculatus 5 59 0.49* 0.212
Plantago lanceolata†6 92 0.56* 0.114
Rumex acetosa 4 43 0.02 0.046
Trifolium pratense†4 41 0.60* 0.180
Trifolium repens 6 85 0.28* 0.066
*Significant change in aboveground biomass, with species richness P⬍0.05. †Significant location-by-species
richness interaction, P⬍0.05.
Fig. 2. Biomass patterns at each site (displayed with species richness on a log
2
scale for comparison
with Fig. 1A). Best-fit models from individual sites based on adjusted R
2
are as follows: log-linear
in Switzerland and Portugal; linear (untransformed species richness) in Germany and Sweden;
quadratic in Sheffield; ANOVA with five species richness levels (significant treatment effects with
no simple trend) in Ireland and Silwood; and no significant effect in Greece.
RESEARCH ARTICLES
www.sciencemag.org SCIENCE VOL 286 5 NOVEMBER 1999 1125

diverse synthesized plant communities have a
higher probability of containing, and becoming
dominated by, a highly productive species (10,
23); (ii) niche complementarity, where ecolog-
ical differences between species lead to more
complete utilization of resources in intact com-
munities relative to depauperate versions (2,4,
6,17,23); (iii) a reduction in positive mutual-
istic interactions between species in simplified
assemblages (6). Distinguishing (ii) from (iii)
will require detailed local experiments, nor is
separating the sampling effect from comple-
mentary and positive interactions straightfor-
ward (6,12,13,30). However, the sampling
effect predicts that the dominance of some spe-
cies in high-diversity mixtures should be com-
pensated for by reductions in the biomass of
subordinate species. In contrast, fewer species
declined in performance in polycultures than
increased, which is consistent with a reduction
of competition in mixtures of species relative to
monocultures due to niche complementary or
positive species interactions or both (Table 4)
(31). Similar results have been reported else-
where (5).
Only niche complementarity and positive
species interactions can generate “overyield-
ing,” where the total biomass of a mixture of
species exceeds the monoculture biomass
achieved by the highest yielding of the com-
ponent species (32). We adapted (13) well-
established techniques from agricultural and
plant competition experiments (33) and used
data from our replicated monocultures to test
for overyielding in individual plant assem-
blages. As in the productivity analysis, over-
yielding differed significantly between sites
(F
7,136
⫽6.59; P⬍0.001), but there was a
consistent average decrease in overyielding
with the simulated loss of species richness
(slope ⫽⫺0.021, SE ⫽0.0075, F
1,126
⫽
6.09, P⬍0.05) and with declining number of
functional groups within species richness lev-
els (slope ⫽⫺0.143, SE ⫽0.0544, F
1,126
⫽
5.08, P⬍0.05). These results are again
consistent with the occurrence of comple-
mentary and positive interactions within our
mixtures of plant species and provide a sec-
ond line of evidence indicating that our pro-
ductivity patterns cannot be explained solely
by the sampling effect.
Our results demonstrate multiple control
of the productivity of experimental plant
communities by geographic location and by
the richness and composition of plant species
and functional groups. Biomass patterns pre-
dict a log-linear decline in productivity with
the loss of plant species richness, in which
reductions of niche complementary or posi-
tive species interactions or both appear to
play a role.
References and Notes
1. Nature’s Services, G. C. Daily, Ed. (Island Press, Wash-
ington, DC, 1997).
2. S. Naeem, L. J. Thompson, S. P. Lawler, J. H. Lawton,
R. M. Woodfin, Nature 368, 734 (1994).
3. S. Naeem, K. Håkansson, J. H. Lawton, M. J. Crawley,
L. J. Thompson, Oikos 76, 259 (1996); A. J. Symstad,
D. Tilman, J. Willson, J. M. H. Knops, Oikos 81, 389
(1998).
4. D. Tilman, D. Wedin, J. Knops, Nature 379, 718
(1996).
5. D. Tilman et al., Science 277, 1300 (1997).
6. D. U. Hooper and P. M. Vitousek, Science 277, 1302
(1997); D. U. Hooper, Ecology 79, 704 (1998); D. U.
Hooper and P. M. Vitousek, Ecol. Monogr. 68, 121
(1998).
7. G. W. Allison, Am. Nat. 153, 26 (1999).
8. M. A. Huston, Oecologia 110, 449 (1997).
9. J. P. Grime, Science 277, 1260 (1997).
10. L. W. Aarssen, Oikos 80, 183 (1997).
11. J. G. Hodgson, K. Thompson, P. J. Wilson, A. Bogaard,
Funct. Ecol. 12, 843 (1998).
12. A. Hector, Oikos 82, 597 (1998).
13. M. Loreau, Oikos 82, 600 (1998).
14. J. H. Lawton, S. Naeem, L. J. Thompson, A. Hector,
M. J. Crawley, Funct. Ecol. 12, 848 (1998).
15. M. Diemer, J. Joshi, C. Ko¨rner, B. Schmid, E. Spehn,
Bull. Geobot. Inst. ETH 63, 95 (1997); C. P. H. Mulder,
J. Koricheva, K. Huss-Danell, P. Ho¨gberg, J. Joshi, Ecol.
Lett. 2, 236 (1999); E. M. Spehn, J. Joshi, B. Schmid, M.
Diemer, C. Ko¨rner, Funct. Ecol., in press.
16. Field experiments were established in spring 1995 in
Switzerland, in autumn 1996 in Portugal, and in
spring 1996 at all other sites. Plots2mby2mwere
seeded with 2000 seeds per square meter, divided
equally between the number of species in each plant
assemblage. Seeds were collected locally as far as
possible or otherwise were purchased from national
commercial sources, avoiding agricultural cultivars.
Plots were regularly weeded to remove unwanted
species.
17. D. Tilman, C. L. Lehman, K. T. Thomson, Proc. Natl.
Acad. Sci. U.S.A.94, 1857 (1997).
18. T. J. Givnish, Nature 371, 113 (1994); D. Tilman,
Oikos 80, 185 (1997).
19. The numbers of functional groups in our assemblages
were exactly as planned. A few plant species failed to
establish, particularly in the higher-diversity assem-
blages, but realized richness was highly correlated
with initial number of species sown. In the second
year, realized species richness was 10% lower on
average than planned richness (slope ⫽0.9, SE ⫽
0.007, n⫽480, R
2
⫽0.95, P⬍0.001). The analyses
reported here use the planned number of species.
Analyses using actual numbers of species present in
the second year of the experiment are not presented
but also reveal highly significant effects of species
richness and functional group richness.
20. Correlations of environmental parameters with aver-
age productivity per site for assemblages with eight
species support reduced productivity in northern and
southern sites by revealing a significant quadratic
effect of latitude (linear term: F
1,5
⫽3.19, P⫽0.134;
quadratic term: F
1,5
⫽40.47, P⫽0.001, R
2
⫽0.86).
When included in a model with the linear effect of
latitude, July temperature also had a significant ef-
fect on productivity (July temperature: F
1,5
⫽11.64,
P⫽0.019; temperature and latitude model: R
2
⫽
0.73).
21. H. A. Mooney, in Ecosystems of the World 11. Medi-
terranean-type Shrublands, F. di Castri, D. W. Goodall,
R. L. Specht, Eds. (Elsevier Scientific, Amsterdam,
1981); F. di Castri and H. A. Mooney, Eds., Mediter-
ranean-type Ecosystems: Origin and Structure
(Springer-Verlag, Berlin, 1973).
22. Because species richness and functional group rich-
ness are unavoidably correlated, in statistical analy-
ses it is impossible to unequivocally distinguish their
relative effects [G. W. Allison, Am. Nat. 153,26
(1999)]. We present the sequential analysis deter-
mined by our experimental design and a priori hy-
potheses, which addressed the effects of (i) species
richness and (ii) functional group richness for a given
number of species. Analyses used sequential back-
ward selection of terms from the maximal model,
which included sites, blocks within sites, species rich-
ness, functional group richness (within-species rich-
ness levels), plant assemblage, the locality-by-assem-
blage interaction, and the overall residual variation
between plots within the above treatments. Locality-
by-diversity interactions, the interaction of species,
and functional group richness were also included but
were never statistically significant and, for brevity,
are not reported here.
23. M. Loreau, Proc. Natl. Acad. Sci. U.S.A. 95, 5632
(1998).
24. The percent of plant cover in each plot was visually
estimated several times during each growing season
and by the presence or absence of rooted individuals
in 50 cells of a permanent quadrat measuring1mby
0.5 m. Productivity patterns could be associated with
poor cover in low diversity assemblages, which may
arise from poor plant establishment [M. A. Huston,
Fig. 3 Percentages of the total sums of squares explained by the effects of individual species and
functional groups. Twenty-nine species had significant effects (P⬍0.05); the 15 most highly
significant species (P⬍0.001) are shown.
RESEARCH ARTICLES
5 NOVEMBER 1999 VOL 286 SCIENCE www.sciencemag.org1126

Oecologia 110, 449 (1997)]. The analysis and eco-
logical interpretation of this issue are complex, but
low levels of cover in the unmanipulated reference
plots at some of our sites (sometimes 50% at the
annual-dominated Portuguese field site) provide ev-
idence against the automatic exclusion of plots with
low cover.
25. J. H. Lawton, Oikos 71, 367 (1994); S. Naeem, J. H.
Lawton, L. J. Lindsey, S. P. Lawler, R. M. Woodfin,
Endeavour 19, 58 (1995); F. Schla¨pfer, B. Schmid, I.
Seidl, Oikos 84, 346 (1999).
26. Data from the third year of the experiment have been
processed for all sites except Portugal. Although there
still appears to be no response in Greece, much of the
other variation shown in Fig. 2 has disappeared, and the
overall pattern appears to match the general log-linear
relation more closely than in year two.
27. D. A. Wardle, O. Zackrisson, G. Ho¨rnberg, C. Gallet,
Science 277, 1296 (1997); D. Tilman et al.,Science
278, 1866 (1997).
28. Composition effects are a combination of the effects
of particular species and of interactions between
species in polycultures. An assemblage-by-location
interaction indicated that where the same species or
mixture of species occurred at more than one site,
they generally achieved significantly different bio-
masses at different locations [F
29,235
⫽3.77, P⬍
0.001 ( Table 3)].
29. Each species or functional group was added individ-
ually to the multiple regression models in Table 3.
Fitting each species or group separately meant that
the effect attributed to each was maximized. Our
ability to test the effects of the grass functional
group was limited, because most of the assemblages
included grasses.
30. E. Garnier, M.-L. Navas, M. P. Austin, J. M. Lilley, R. M.
Gifford, Acta Oecol. 18, 657 (1997).
31. We performed simple regressions of the estimated
per-plant biomass of 14 species across a gradient of
increasing species richness (on a log
2
scale) after
adjusting for differences between locations and
blocks by taking residuals from analyses with these
terms [seven of the species showed significantly dif-
ferent responses at different sites ( Table 4)]. We
expected slopes of zero where intraspecific competi-
tion was equal to competition with other species and
expected positive and negative slopes where it was
more and less intense, respectively. Under the sam-
pling hypothesis, we expected approximately equal
distributions of positive and negative slopes. In con-
trast, 12 of the 14 species had slopes that were
positive; 8 significantly so. Two species had slopes
that were negative, but neither was significantly dif-
ferent from zero.
32. J. L. Harper, Population Biology of Plants (Academic
Press, London, 1977).
33. We identified the species with the highest biomass in
each plant assemblage and, where data were avail-
able (271 of the 308 polycultures), compared their
average monoculture biomass with the biomass of
the total assemblage using the overyielding index
Dmax [M. Loreau, Oikos 82, 600 (1998)], where
Dmax ⫽(total biomass of a plant assemblage–aver-
age monoculture biomass of the dominant species in
that assemblage)/average monoculture biomass of
the dominant species in that assemblage. We ana-
lyzed Dmax after transformation using natural loga-
rithms (after adding 1 to make all values positive) to
meet the assumptions of parametric analyses.
34. The BIODEPTH project is funded by the European
Commission within the Framework IV Environment
and Climate program (ENV-CT95-0008). Many col-
leagues too numerous to list have assisted with the
project; in particular, we thank P. Heads and E. Bazeley-
White. We thank J. Nelder for advice on statistical
analyses.
20 May 1999; accepted 14 September 1999
REPORTS
Hydrogen Storage in
Single-Walled Carbon
Nanotubes at Room
Temperature
C. Liu,
1
Y. Y. Fan,
1
M. Liu,
1
H. T. Cong,
2
H. M. Cheng,
1
*
M. S. Dresselhaus,
3
*
Masses of single-walled carbon nanotubes (SWNTs) with a large mean diameter
of about 1.85 nanometers, synthesized by a semicontinuous hydrogen arc
discharge method, were employed for hydrogen adsorption experiments in their
as-prepared and pretreated states. A hydrogen storage capacity of 4.2 weight
percent, or a hydrogen to carbon atom ratio of 0.52, was achieved reproducibly
at room temperature under a modestly high pressure (about 10 megapascal)
for a SWNT sample of about 500 milligram weight that was soaked in hydro-
chloric acid and then heat-treated in vacuum. Moreover, 78.3 percent of the
adsorbed hydrogen (3.3 weight percent) could be released under ambient
pressure at room temperature, while the release of the residual stored hydrogen
(0.9 weight percent) required some heating of the sample. Because the SWNTs
can be easily produced and show reproducible and modestly high hydrogen
uptake at room temperature, they show promise as an effective hydrogen
storage material.
Hydrogen (H
2
) has attracted a great deal of
attention as an energy source. Once it is
generated, its use as a fuel creates neither air
pollution nor greenhouse gas emissions.
However, no practical means for H
2
storage
and transportation have yet been developed.
Of the problems to be solved for the utiliza-
tion of hydrogen energy, how to store H
2
easily and cheaply has been given high pri-
ority on the research agenda.
Recently, carbon nanotubes and carbon
nanofibers were reported to be very promising
candidates for H
2
uptake. Dillon et al. (1) first
measured the H
2
adsorption capacity of an as-
prepared soot containing only about 0.1 to 0.2
weight % SWNTs at 133 K, from which they
extrapolated an H
2
adsorptivity for pure
SWNTs of 5 to 10 weight % (the weight of H
2
adsorbed divided by the weight of SWNTs plus
the H
2
adsorbed by the SWNTs), and predicted
that SWNTs with a diameter of between 1.63
and 2 nm would come close to the target H
2
uptake density of 6.5 weight %. Ye et al. (2)
reported that a ratio of H to C atoms of about
1.0 was obtained for crystalline ropes of
SWNTs at a cryogenic temperature of 80 K and
pressures ⬎12 MPa. Instead of SWNTs, Cham-
bers et al. (3) claimed that tubular, platelet, and
herringbone forms of carbon nanofibers were
capable of adsorbing in excess of 11, 45, and 67
weight % of H
2
, respectively, at room temper-
ature and at a pressure of 12 MPa. More recent-
ly, Chen et al. (4) reported that a high H
2
uptake of 20 and 14 weight % can be achieved
for Li-doped and K-doped multi-walled carbon
nanotubes (MWNTs) in milligram quantities,
respectively, under ambient pressure. The K-
doped MWNTs can adsorb H
2
at room temper-
ature, but they are chemically unstable, whereas
the Li-doped MWNTs are chemically stable,
but require elevated temperatures (473 to 673
K) for maximum adsorption and desorption of
H
2
.
We measured the H
2
storage capacity of
SWNTs synthesized by a hydrogen arc-dis-
charge method, with a relatively large sample
quantity (about 500 mg) at ambient temperature
under a modestly high pressure. A H
2
uptake of
4.2 weight %, which corresponds to a H/C atom
ratio of 0.52, was obtained by these SWNTs
with an estimated purity of 50 weight %. Also,
⬃80% of the adsorbed H
2
can be released at
room temperature. These results indicate that
SWNTs are highly promising for H
2
adsorption
even at room temperature.
1
Institute of Metal Research, Chinese Academy of
Sciences, 72 Wenhua Road, Shenyang 110015, China.
2
State Key Lab for Rapidly Solidified Non-equilibrium
Alloys, Institute of Metal Research, Chinese Academy
of Sciences, 72 Wenhua Road, Shenyang 110015,
China.
3
Department of Physics and Department of
Electrical Engineering and Computer Science, Massa-
chusetts Institute of Technology, Cambridge, MA
02139, USA.
*To whom correspondence should be addressed. E-
mail: cheng@imr.ac.cn (H.M.C.) and millie@mgm.mit.
edu (M.S.D.)
RESEARCH ARTICLES
www.sciencemag.org SCIENCE VOL 286 5 NOVEMBER 1999 1127