Diversity-dependent productivity in semi-natural grasslands following climate perturbations


  • A. KAHMEN,

    Corresponding author
    1. Max Planck Institute for Biogeochemistry, Hans-Knöll-Straße 10, D-07745 Jena,
      †Author to whom correspondence should be addressed. E-mail: akahmen@bgc-jena.mpg.de
    Search for more papers by this author
  • J. PERNER,

    1. Institute of Ecology, University of Jena, Dornburger Straße 159, D-07743 Jena, Germany
    Search for more papers by this author

    1. Max Planck Institute for Biogeochemistry, Hans-Knöll-Straße 10, D-07745 Jena,
    Search for more papers by this author
    • §

      Present address: Institute of Plant Sciences, ETH Zürich, Universitätsstraße 2, CH-8092 Zürich, Switzerland.

†Author to whom correspondence should be addressed. E-mail: akahmen@bgc-jena.mpg.de


  • 1The consequences of globally declining biodiversity and climate change for ecosystem functions are intensively debated topics in ecological research. However, few studies have investigated potential interactions, or the combined effects of both scenarios, for ecosystem functioning. In the work presented here we tested the hypothesis that increasing plant diversity acts as insurance for ecosystem functions during extreme weather events which are predicted by climate change scenarios.
  • 2We measured the effect of plant diversity on above- and below-ground productivity in semi-natural grasslands following experimentally induced early summer drought. To test the insurance hypothesis directly, we determined in each community the range of δ13C values of individual plant species as drought stress indicators.
  • 3Increasing plant diversity significantly enhanced below-ground productivity as a consequence of simulated drought, while above-ground productivity was reduced independently of plant diversity.
  • 4Plants shifting carbon allocation to below-ground compartments during drought maintain various aspects of ecosystem services and functions. Although we were not able to detect physiological evidence for the insurance hypothesis, we conclude from our below-ground results that plant diversity is an essential entity of ecosystems for maintaining ecosystem functions in a changing climate.


There is a long-standing debate in ecological research about the relationship between biodiversity and stability (MacArthur 1955; McNaughton 1977; Leps, Osbornovakosinova & Rejmanek 1982; Pimm 1984). The focus of the debate has shifted in recent years from population and food-web levels to the ecosystem level, where the influence of biodiversity on the stability of ecosystem functions is of interest. This is partly a consequence of the dramatic worldwide loss in species diversity (Chapin et al. 1997), and the predicted increase in extreme climatic events such as flood, frost, fire, storm and drought as a result of global climate change (IPCC 2001). Climate change scenarios, as well as the causes and consequences of decreasing diversity, are beginning to be well understood (Loreau et al. 2001). However, it remains unknown if ecosystems with high diversity can buffer the effects of climate change and if, consequently, ecosystems reduced in diversity are more sensitive to climate change. If so, climate change effects on ecosystem functions will become more severe if diversity is continuously lost.

According to the so-called ‘insurance hypothesis’, species diversity influences the stability or resistance of ecosystem functions against environmental perturbations (McNaughton 1977; Doak et al. 1998; Yachi & Loreau 1999). The hypothesis is based on the assumption that different species react differently to environmental change. With increasing species diversity, the range of species with different responses to environmental change will therefore also increase in an ecosystem. Consequently, a more diverse ecosystem has a higher likelihood of containing species that are adapted to a changed environment and can compensate for the decline of less adapted species, thus maintaining stable ecosystem functions.

Evidence that stability of ecosystem functions increases with species diversity has been given in theoretical studies (Doak et al. 1998; Loreau & Behera 1999; Yachi & Loreau 1999; Ives, Klug & Gross 2000), and in laboratory experiments (McGrady-Steed, Harris & Morin 1997; Naeem & Li 1997; Mulder et al. 1999). In field studies it was observed that interannual variability of productivity was reduced by increasing species diversity, suggesting a positive relationship between diversity and stability of ecosystem functions (Frank & McNaughton 1991; Dodd et al. 1994; Tilman & Downing 1994). The only field study that has addressed the relationship between diversity and stability experimentally found no effect of plant diversity on the stability of above-ground productivity during disturbance in grasslands (Pfisterer & Schmid 2002). Potential effects on below-ground productivity were, however, ignored in this study (Schmid & Pfisterer 2003).

In the work reported here, we investigated the role of plant diversity for ecosystem functions during extreme weather events which are predicted as a consequence of global climate change for central Europe (Schär et al. 2004). Specifically, we investigated (1) if plant diversity influences productivity above- and below-ground in semi-natural grasslands during experimentally induced early summer drought; and (2) if, in a grassland community, the range of species with different drought responses increased with increasing diversity, giving physiological evidence for the insurance hypothesis.


study sites and experimental design

The study was conducted in semi-natural grasslands in the Thüringer Schiefergebirge/Frankenwald, a plateau-like mountain range, in central Germany. In early spring 2002, 19 grassland sites were selected that differed in plant diversity and were similar in edaphic conditions. Only sites that were ungrazed, cut twice a year in late June and late August with no fertilizer application in the past 10 years, were selected.

At each site two 5 × 5 m plots were established. In the centre of each plot a 1 × 2 m area was established for vegetation, biomass and soil sampling. While one plot in each site served as control, the other plot was roofed from 23 April to 12 June 2003 to simulate drought. Roofs were located in the centre of the plot, covering an area of 3 × 3·5 m. Roofs were constructed with a steel frame and covered with transparent foil that permitted 90% penetration of photosynthetically active radiation (Cello Flex 4TT, Prosyn Polyane, St Chamond, France). To ensure air circulation, roofs were 2·3 m in height and tunnel-shaped with the two ends open. For true control of drought effects, ideally roofs should have been established for both control plots and drought treatments, with the collected water added back to the control plot. However, given the large amount of labour involved, we roofed only the treatments and tested for roof effects other than drought using meteorological measurements in six of the 19 sites. Specifically, air temperature in 60 cm height, soil temperature (5–10 cm below ground) and soil moisture (5–10 cm below ground) were measured continuously during the entire year in roofed and control plots.

sampling of vegetation, productivity, δ13c and environmental data

In a central 3 × 3 m area of each 25 m2 plot, we recorded number of plant species and percentage cover of each species at peak standing biomass. We harvested above-ground biomass twice a year, in mid-June 2003 and early September 2003, following the local management regime. In the central sampling area of each plot we collected eight 25 × 50 cm subsamples. Biomass was clipped 2 cm above ground, dried for 48 h at 60 °C and weighed thereafter. We sampled below-ground biomass using ingrowth cores (4·3 cm diameter, 10 cm length). In early spring, prior to the beginning of the vegetation period, four ingrowth cores were installed in the central sampling area of each plot. For each plot we sieved soil from the respective grassland to 10 mm and filled the cores with this root-free soil. At the end of the vegetation period in mid-September, ingrowth cores were collected and ingrown roots separated from soil in the laboratory. Thereafter the roots were dried for 48 h at 60 °C and weighed.

To gain physiological evidence for the insurance hypothesis, we investigated if, in a community, bulk leaf δ13C values of individual species were affected by experimental drought, and if the range of species with different drought responses increased with increasing diversity in a community. The isotopic composition of carbon in plant tissue gives evidence for leaf stomatal conductance during the time when the C was assimilated and can therefore be utilized as an integrated measure of photosynthetic water-use efficiency and drought stress of plants (Farquhar & Richards 1984). The number of plant species sampled for δ13C in control and roofed plots was proportional to the species richness in each plot, and covered two-thirds of the most dominant species in a site. Plant material was dried, ball milled and analysed for δ13C with an isotope ratio mass spectrometer (IRMS, Delta C Finnigan MAT, Bremen, Germany). For analysis of community δ13C in a site, the values of all species investigated in a site were averaged for each plot, and the differences between treatment and control were compared (see Statistical analyses). To test if, in a community, the range of species with different drought responses increased with increasing diversity, the difference between δ13C values of treatment and control was calculated individually for each species sampled, the difference averaged for each site, and the standard deviation of the averaged δ13C differences determined as a measure for within-site variability.

We collected soil samples in each plot to separate diversity effects from confounding background effects of edaphic variables. Soil was collected six times throughout the year 2002 and two times in 2003. In each plot, four soil subsamples were collected and pooled. One part of each soil sample (≈10 g) was extracted with 50 ml 1 m KCl for 60 min on the same day of sampling. KCl extracts were filtered and then frozen at −20 °C and later analysed using a continuous flow analyser (SAN Plus, Skalar, Erkelenz, Germany) for inline image and inline image, and an ICP-AES (Optima 3300 DV, Perkin-Elmer, Norwalk, USA) for Ca2+. The remaining soil was dried at 35 °C and extracted for 1 h using a 1 m calcium–acetate–lactate (CAL) solution. CAL extracts were analysed with ICP-AES for P, K+, Mg2+ and inline image. Soil pH was measured in a water extract. For determination of the soil C : N ratio and the total soil nitrogen (Ntot) and carbon (Ctot), dry soil was ground and analysed with an element analyser (Vario EL II, Elementar, Hanau, Germany). No seasonal trend was observed within soil variables. Therefore subsamples of variables sampled in 2002 and 2003 were averaged for each plot before they entered the statistical analyses.

quantifying diversity, species composition and soil properties

To test for the effect of plant diversity on productivity following drought, for each plot we calculated the effective diversity (heterogeneity or exponential Shannon–Wiener: N1 = eH; H′ = −Σ(pi)(ln pi); pi = species cover/sum of cover for all species) that corrects species richness for differences in evenness (for more details see Krebs 1999). In order to account for effects of community composition (plant composition and abundance), we performed non-metric multidimensional scaling (NMDS) ordination for control and treatment plots in the 19 grassland sites studied, and used the scores of the resulting NMDS axes as numeric values for community composition in consecutive analyses. The NMDS ordination was based on plant cover data to calculate a distance matrix. The plant cover data were square-root transformed to linearize the data set and to reduce the importance of extreme values, as suggested by Krebs (1999). As a distance measure, we used the Bray–Curtis coefficient (also known as Sørensen or Czekanowski coefficient), written in shorthand as 1–2 W/(A + B), where W is the sum of shared species cover in control and roof plots; and A and B are the sums for cover for all species in the control and roof plots. The Bray–Curtis coefficient is one of the most robust measures for this purpose. NMDS is an iterative search procedure that places objects of a distance matrix (e.g. different sampling sites) in a space of minimized dimensionality while preserving their rank order of distances as far as possible. The coordinates (scores) that position an object along the axes of the minimized dimensional space can then be used as numeric values in consecutive analysis, representing a solution with highly reduced dimensionality of the previously multivariate data set. For more details see Legendre & Legendre (1998). NMDS analyses were performed using the program pc-ord (McCune & Mefford 1997).

Soil variables were aggregated using principal component analysis (PCA) using the program canoco (ter Braak & Smilauer 2002). PCA summarizes the multivariate information of the soil variables as four major axes. As PCA axes are, by definition, orthogonal and independent of each other, this procedure creates composite independent variables and avoids the danger of spurious correlations (multicollinearity). Consequently these axes were used in all consecutive analyses as independent soil variables for the plots.

statistical analysis

To differentiate the diversity-dependent effects of experimental drought on the productivity of parameters above and below ground, as well as on community δ13C from edaphic background variation, in a first step we performed multiple regression analyses for each parameter with the PCA constructed variables soil1, soil2 and soil3 as independent variables. For each parameter, the residuals of the multiple regression analysis were then added to the parameters’ arithmetic mean to obtain an estimate of productivity above and below ground, as well as for community δ13C that was corrected for edaphic background variation. In a second step, we used these corrected values for each of the three parameters to calculate the difference between treatment and control for each site. In a final step, for each parameter we determined the effect of species composition (NMDS1 and NMDS2) as well as effective diversity on the difference between treatment and control using general linear models (type I sum of squares). In the analysis, the means between treatment and control plots of NMDS1, NMDS2 and effective diversity were entered as covariates in a fixed hierarchical sequence. We decided on a conservative approach and entered the composition variables NMDS1 and NMDS2 first, so that composition effects were removed from the model before the actual diversity effect was tested. To account for general differences in productivity among sites, above- and below-ground biomass data were log-transformed before the analysis, which corresponds to an analysis of relative changes in productivity and meets the assumptions of general linear models (SPSS Inc., 2001).


Plant species richness in the plots investigated varied between 13 and 38 species, and effective diversity between 6·5 and 23·3. Non-metric multidimensional scaling showed that a two-dimensional solution was sufficient (minimum stress values: 1st axis/dimension = 20·1, 2nd axis/dimension = 12·4, R2 = 0·90) to explain plant community composition. The scores of both NMDS axes described a gradual compositional change from Geranio-Trisetetum nardetosum grasslands (low NMDS scores) to Geranio-Trisetetum alopecuretosum grasslands (high NMDS scores) (Hundt 1964). For further analyses we used the scores of the first two NMDS axes as numeric variables for plant community composition (NMDS1 and NMDS2). In the PCA analysis, the first three axes explained 72% of the variance of all soil variables tested (Table 1). Therefore PCA axes 1–3 were used for consecutive analysis as variables soil1–3. While soil1 was mainly correlated with pH-related variables (pH, inline image, Ca2+), soil C, inline image and Nmin, soil2 was correlated with inline image, Nmin, K+ and soil N. Soil3 was most strongly correlated with Ptot (Table 1).

Table 1.  Eigenvalues and eigenvector coefficients (loadings) of the standardized principal component analysis of soil variables
Eigenvalue 0·4336 0·1706 0·1262 0·0957
inline image−0·7528−0·0894−0·4272−0·0366
inline image−0·4182−0·7174 0·0401 0·2269
Nmin−0·7226−0·5849−0·2032 0·1508
K+−0·1975 0·6178−0·4851−0·0893
Mg+−0·8665−0·2935 0·0039−0·0326
Na+ 0·4642−0·3608 0·3979 0·0323
Ptot−0·4465 0·1967−0·7175−0·0873
inline image 0·7965−0·2070−0·4131 0·2365
Ca2+−0·8087−0·2100 0·1300−0·4460
Soil C 0·6903−0·5007−0·3030−0·4060
Soil N 0·5911−0·6143−0·4045−0·1591
C : N 0·5161 0·0388 0·0842−0·7694
pH−0·8873−0·0188 0·2368−0·3366

Roof establishment in mid-April did not affect mean daily air temperature (Fig. 1a), but resulted in a significant reduction of mean daily soil moisture compared with control plots (Fig. 1b). Differences in soil moisture between treatment and control were highest in the phase of highest biomass productivity on control plots (Fig. 1b).

Figure 1.

Effects of experimentally induced drought (shaded area) on mean daily air temperature and mean daily soil moisture in one representative site of the study. No differences between treatment (grey line) and control plots (black line) were observed for air temperature (a, control behind treatment), while soil moisture was affected substantially by the drought treatment (b). The drought experiment was carried out during the time of highest above-ground biomass production in the control plots (b). For air temperature, data are missing for treatment and control between 15 and 28 February as well as for treatment between 23 June and 15 August.

Mean annual above-ground productivity in the control plots was 506·0 g dw m−2 152·6 SE and 412·5 124·0 in the roofed plots (Fig. 2a). Mean annual below-ground productivity in the control plots was 175·7 66·2 and 189·6 62·6 in the roofed plots (Fig. 2c). When background variation in productivity above and below ground was analysed in multiple regression analyses, above-ground productivity showed a significant relationship with soil variables, specifically soil1 (Table 2). In contrast, below-ground productivity was not significantly influenced by soil variables (Table 2). The effects of species composition and diversity following experimental drought on the changes in above- and below-ground productivity (corrected for edaphic background variation) were tested using general linear models. NMDS1 and NMDS2, as well as effective diversity, had no significant effect on changes in above-ground productivity (Table 3; Fig. 2b). In contrast, differences in below-ground productivity were significantly influenced by NMDS1, and highly significantly influenced by effective diversity (Table 3; Fig. 2d).

Figure 2.

Relationship between effective plant diversity and above- and below-ground productivity in control and treatment plots (a,c); relative change of above- and below-ground productivity caused by drought (b,d). (a,c) Full symbols, control plots; open symbols, treatment (roof) plots; (b,d) full symbols, change in productivity when corrected for edaphic background variation; open symbols, values predicted by general linear models (Table 3).

Table 2.  Multiple regression models for parameter productivity above and below ground, and for community δ13C. Separate models were calculated for each parameter with the variables soil1, soil2 and soil3 entered into each model
Independent parameterVariableDetails of multiple regression modelModel summary
Above-ground biomass   0·498<0·001
soil1−0·695 <0·001  
soil2−0·010  0·937  
soil3−0·123  0·317  
Below-ground biomass   0·055  0·582
soil1 0·168  0·321  
soil2−0·116  0·491  
soil3 0·116  0·491  
δ13C   0·389  0·001
soil2−0·157  0·249  
soil3−0·280  0·044  
Table 3.  Effects of species composition (NMDS1 and NMDS2) and plant diversity on drought-induced changes in above- and below-ground productivity (corrected for edaphic background variation) and for community δ13C, tested using general linear models (type I sum of squares)
Source of variationSum of squaresdfFP
Above-ground biomass
Model0·304 3 2·4340·105
NMDS10·125 1 3·0050·103
NMDS20·075 1 1·7970·200
Effective diversity0·104 1 2·4920·135
Below-ground biomass
Model2·663 3 7·9640·002
NMDS10·623 1 5·5890·032
NMDS20·050 1 0·4520·512
Effective diversity1·990 117·8520·001
Model0·688 3 1·6740·215
NMDS10·612 1 4·4670·052
NMDS20·066 1 0·4820·498
Effective diversity0·010 1 0·0730·790

Mean community δ13C values in the control plots was −27·32 0·32 and −26·99 0·46 in the treatment plots (Fig. 3a). Interestingly, soil variables had a significant effect on community δ13C values when tested in a multiple regression analysis (Table 2). However, changes in δ13C following the experimental drought treatment showed no significant relationship with species composition (but marginal significance for NMDS1, P = 0·052) or for effective diversity when tested using general linear models (Table 3; Fig. 3b). The range of species with different drought responses (standard deviation of average differences between δ13C in treatments and controls of individual species) did not correlate with effective diversity (Fig. 3c).

Figure 3.

Average differences of carbon isotopic composition of plant communities between control and treatment plots as related to effective diversity: (a) community δ13C values for control (full symbols) and treatment (open symbols) plots; (b) change in δ13C values when corrected for edaphic background variation (open symbols) and predicted δ13C values (full symbols) in general linear models (Table 3); (c) correlation between standard deviation of mean differences in δ13C values between control and treatment plots.


A positive effect of diversity on productivity has been detected in several previous experimental studies (Tilman, Wedin & Knops 1996; Hector et al. 1999; Loreau, Naeem & Inchausti 2002). The relationship observed in these studies was, however, largely influenced by very low species numbers. At the diversity levels typically found in natural plant communities, no direct relationship between plant diversity and productivity has been observed (Kahmen et al. 2005). The results we obtained in the control plots of our semi-natural grasslands are in line with these observations (Fig. 2a). It has been argued that, at high diversity levels, plant species are functionally redundant and thus no direct effect of diversity on productivity can be observed (Vitousek & Hooper 1993; Walker 1992). Few studies have tested if species diversity, which has little or no effect on ecosystem function in a stable environment, influences the stability of ecosystem functions following environmental perturbations (Loreau et al. 2001). The results from this study show that, independently of plant diversity, above-ground productivity was reduced by drought, while increasing plant diversity enhanced below-ground productivity during drought, thus maintaining more stable overall productivity of the respective community (Table 3; Fig. 2).

Drought can influence various aspects of plant productivity and C allocation, depending on the severity of water deficiency in the soil (Kramer & Boyer 1995; Lambers, Chapin & Pons 1998). As a first reaction to drought, plant roots sensing dry soil produce abscisic acid (ABA), which is transported to the leaves. In the leaves ABA reduces stomatal conductance, leaf expansion and eventually photosynthesis, leading to a reduction or cessation of productivity (Davies & Zhang 1991; Tardieu et al. 1992). At moderate water deficiency in the soil, however, below-ground plant parts are less sensitive to drought than leaves, as root growth is less affected than leaf expansion by ABA (Saab et al. 1990). Also, moderate water stress can reduce leaf growth prior to photosynthesis, resulting in a surplus of assimilates that are exported to the roots, and enhance root productivity (Boyer 1970). As a consequence, enhanced root production through shifts in C allocation in moderately water-stressed plants might exceed that of well watered plants and is presumably an adaptation to dry soils, allowing the exploitation of reduced soil moisture levels (Sharp & Davies 1979; Jupp & Newman 1987; Lambers, Chapin & Pons 1998).

Different plant species react differently to drought disturbance, depending on their specific water-use efficiency or drought tolerance (Molyneux & Davies 1983). Several studies have shown that, for some plant species, productivity above and below ground was reduced as a consequence of drought, while for other species total productivity was less affected, but a shift in C allocation caused reduced above-ground productivity but enhanced below-ground productivity (Foulds 1978; Stevenson & Laidlaw 1985; Carter, Theodorou & Morris 1997; Guenni, Marin & Baruch 2002). Similarly, we found in our study that the productivity of the grasslands investigated was affected differently by drought. While in some grasslands productivity was reduced above and below ground, at other sites a decrease in above-ground productivity was partly balanced by an increase in below-ground productivity (Fig. 2d). Interestingly, drought effects on below-ground productivity were significantly related to plant diversity, suggesting that a more diverse community has a higher likelihood of containing drought-tolerant species that allow C allocation to below-ground parts.

We used δ13C values of leaf tissue as drought stress indicators of individual plant species, and as a potential functional explanation of the insurance hypothesis. Specifically, we tested if the range of species with different drought responses (δ13C ratios of foliage) increased in communities with increasing plant diversity. Carbon isotope ratios (δ13C) of plant tissue are determined, in part, by the ratio of intercellular to ambient CO2 concentrations (ci/ca) (Farquhar, O’Leary & Berry 1982). Increasing δ13C values indicate low ci/ca ratios resulting from either high photosynthetic demand or low stomatal conductance (Ehleringer & Cooper 1988; Farquhar et al. 1989). Thus the C-isotope composition of plant tissue can be used as an integrating measure of photosynthetic water-use efficiency during the time when the respective C was assimilated (Farquhar & Richards 1984). We collected leaf tissue of individual species at the end of the experimental drought period and analysed the bulk leaf material for δ13C. The experimentally induced drought had only small effects on mean δ13C values of the leaf material analysed in the sites, with slightly less negative δ13C values for drought-stressed plants (Fig. 3a,b). Consequently, the variability of drought responses did not change with increasing diversity in a community (Fig. 3c). The experimentally induced reduction in soil moisture (Fig. 1b) and the reduced above-ground biomass production (Fig. 2a) were substantial in the experiment, suggesting drought effects on the grassland communities investigated. As a reduction in stomatal conductance is among the first responses of plants to drought, changes in the δ13C values of C assimilated during the drought experiment were to be expected. However, the small observed changes in δ13C values of leaf tissue, and the constant variability along the diversity gradient, are probably due to the relatively short duration of our experiment. It is likely that the time of drought treatment was not sufficient to alter the overall isotopic signature of bulk leaves, due to large fractions of leaf C assimilated before the drought experiment. Consequently, we view the lack of functional evidence for the insurance hypothesis based on δ13C values of leaf tissue as a result of the short duration of the experiment rather than a result of non-existent diversity effects.

Experimental design, data interpretation and potentially confounding ‘hidden treatments’ in studies investigating the effects of biodiversity on ecosystem functions have been highly debated in the literature (Aarssen 1997; Huston 1997; Hector et al. 2000; Wardle 2001; Schmid & Pfisterer 2003; Wardle & Grime 2003). To account for such hidden treatments in our study, we incorporated soil variables and community composition into our data analysis as covariables (Tables 2 and 3). Our results are consistent with previous studies that have determined the importance of species composition for the stability of ecosystem functions (MacGillivray et al. 1995; Grime et al. 2000; Wardle et al. 2000). However, species composition was only significant in the model for below-ground productivity when this covariable was entered in the analysis prior to effective diversity (Table 3). If the sequence was reversed and effective diversity was entered first, the significant effect of NMDS1 vanished, suggesting only a weak effect of composition (data not shown). In our analyses we used effective diversity as a measure of plant diversity. We also tested the effects of species richness on drought resistance of productivity. In general, the same statistical trends were observed when either species richness or effective diversity was used in the general linear models. However, effective diversity proved to explain more of the variability in the data, suggesting that it is important to consider not only species diversity, but also the evenness of species in the analyses of ecosystem functions.

Despite the lack of physiological evidence using δ13C values, the data from this study show strong diversity effects on productivity following drought. Following water limitation, drought-tolerant species enhance root productivity by shifting C allocation to below-ground parts and, as a result, maintain overall community productivity. Enhanced root production during drought will also positively influence many other ecosystem services and functions, such as nutrient cycling and retention, water-holding capacity during rainfall following drought periods, and maintenance of overall community stability (Daily 1997). Our study therefore provides strong evidence for the insurance hypothesis, suggesting that an increasing number of plant species leads to higher likelihood of a community containing drought-tolerant species that maintain stable ecosystem functions (Doak et al. 1998). Consequently, high biodiversity levels might buffer some of the anticipated consequences of a globally changing climate on ecosystem services by maintaining ecosystem functions during environmental perturbations (IPCC 2001).


This study was supported by grant LC0013 in the BIOLOG program of the German Federal Ministry for Education and Research (BMBF). We thank Marco Pöhlmann, Mathias Putze, Karin Zuber and Karin Sörgel for assistance in the field, and Heike Geilmann for analysing δ13C of plant material. We also thank Jens Schumacher for help with the statistical analyses, and Wolfgang Weisser for coordinating the project.