Diverse marsh plant communities are more consistently productive across a range of different environmental conditions through functional complementarity


Correspondence author. E-mail: bastiansteudel@aol.com


1. Understanding the influence of biodiversity on ecosystem functionality is crucial in modern ecosystem management, especially with regard to the resistance and resilience of ecosystems to future environmental changes. In this study, we assessed the effects of three different environmental regimes on the relationship between diversity and biomass production among marsh plants in comparison with a control treatment to elucidate the underlying classes of proximate mechanisms.

2. We subjected assemblages of up to 23 marsh plant species to four different treatments (control, drought, salt, shade) for 4 months. We examined the treatment effect on the relationship between species diversity and biomass production and explored the underlying mechanism.

3. Biomass production in the manipulated treatments showed a stronger positive effect of biodiversity than the control because of greater declines of biomass production in low diversity mixtures. This effect was owing to an increasingly positive complementarity effect, i.e. a benefit of most species, with increasing diversity, particularly in shade treatment. The selection effect, i.e. a benefit for few species at the expense of the others, was increasingly negative with increasing diversity and dominance by species with lower than average monoculture biomass. The variability of biomass production decreased with increasing species richness in all treatments.

4.Synthesis and applications. We show that the productivity of diverse marsh plant communities is more consistent across a range of environmental conditions than that of depauperate communities and that this unexpectedly resulted from complementarity rather than selection effects. Our results demonstrate that loss of vegetation diversity reduces the average biomass production across a range of environmental conditions and emphasizes the importance of maintaining species-rich biotic assemblages, especially in the face of global change.


In the face of changing environmental conditions and the continuing widespread loss of biodiversity, the effect of biodiversity on ecosystem resilience has become a central issue in ecology (McCann 2000; Pfisterer & Schmid 2002; Hooper et al. 2005; Ives & Carpenter 2007; Duffy 2009; Griffin et al. 2009). Biodiversity can improve ecosystem functioning and reduce spatial or temporal variability (e.g. the insurance hypothesis, Yachi & Loreau 1999; the portfolio effect, Doak et al. 1998; Tilman, Lehman & Bristow 1998). Here, we focus on biomass production with increasing biodiversity in systems with environmental variation. These effects have to be distinguished from biodiversity effects at a given point in time in a single system, i.e. changing ecosystem function with increasing biodiversity in situations without environmental variation or without contrastingly different environmental conditions. Recently conducted meta-analyses on biodiversity effects conclude that there is a significant general biodiversity effect in most grassland and marine ecosystems (Hooper et al. 2005; Cardinale et al. 2006, 2007; Hector et al. 2010), but that these effects are weak under environmental variation (Balvanera et al. 2006). This may be because a variable environment per se need not have negative implications for an ecosystem (e.g. Reich et al. 2001; Ji et al. 2009).

Although many different factors can vary in ecosystems, the majority of empirical studies on effects of the consistency of ecosystem functions under environmental variation have considered only a single environmental parameter (or closely correlated parameters), in most cases drought and increased temperatures (B. Steudel and M. Kessler unpublished data). However, three studies have studied more than one factor. Caldeira et al. (2005) studied an artificial grassland system for 3 years of which the third was drier and had more frost days than the previous years but they did not distinguish between the two environmental factors in their analysis. Sankaran & McNaughton (1999) studied the effects of grazing and fire regimes on the composition stability of savanna grasslands, but excluded nondisturbed (control) plots from the analysis so that interpretation of their results is difficult. They nonetheless concluded that the impact of environmental variation on savanna grassland can be ‘best explained by the ecological history and species characteristics of communities rather than by species diversity in itself’ (Sankaran & McNaughton 1999). A more recent study compared a ‘natural’ ecosystem where a diversity gradient was established by removing species, with ‘artificial’ systems in which species were sown, and concluded that the effect of species diversity on productivity is higher in natural ecosystems (Flombaum & Sala 2008).

Often the contribution of species richness or biodiversity to ecosystem functioning is evaluated by measuring features such as mean biomass production. However, a further possible effect of biodiversity on ecosystem functioning is the stabilization of a certain function, so that the variability of this function decreases with increasing species richness (Tilman 1999; Cottingham, Brown & Lennon 2001). Mixed results have been reported in the literature (Petchey et al. 2002; Caldeira et al. 2005; Steiner 2005; Tilman, Reich & Knops 2006; Zhang & Zhang 2006), although the way in which variability is measured is crucial. Some studies have used the relative rate of plant community biomass change (Rodriquez & Gomezsal 1994; Tilman & Downing 1994; Tilman 1996) but most have advocated the use of the coefficient of variation, which expresses the variability relative to the means (Griffin et al. 2009). A full assessment of the impact of biodiversity on ecosystem functioning under varying environmental conditions will ideally consider both the means and one or several measures of variability.

Here, we used experimental assemblages of one to 23 species of freshwater marsh plants to assess the diversity dependence of biomass production in relation to different abiotic treatments. Our main research questions were: Does this experimental system respond in different ways to different environmental treatments? What are the underlying mechanisms determining possible effects of biodiversity on consistency under environmental variation or ecosystem resilience?

We compared a control with plant assemblages subject to three different treatments, namely (i) limited water availability and high temperatures (‘drought’); (ii) low light levels (‘shade’) and (iii) salty water (‘salt’). These treatments represent changes in environmental conditions that may result from global change. We then compared shifts in plant species composition and biomass production in the four treatments and used the partitioning method of Loreau & Hector (2001) to determine the influence of selection and complementarity effects on the observed biodiversity and stabilizing effects. We also assessed the influence of species numbers on the stability of the study systems.

Materials and methods

Experimental design

We selected freshwater marsh plant species with a grass-like growth form (i.e. plants without branches or with narrowly linear blades) so that our experimental assemblages approached the maximum species richness of comparable natural communities, and because the plants grow quickly and many different species can be cultivated in a small area. The experimental species belonged to different taxonomic, functional and ecological groups, e.g. ferns, grasses and sedges (Table S1 Supporting Information). Although all species are marsh plants that naturally grow in more or less water-saturated soils, many of them never occur together in nature because their distributions do not overlap or because their ecological requirements are different. Some species were annuals but the majority were perennial. Natural plots of the size of the pots used in this experiment supported five to ten species on average but we included more species so as to include maximum levels of diversity. We cultivated 60 species (22 grown from seeds and 38 from shoots or cuttings) in multiplate pots in glasshouses obtained from the holdings of the Old Botanical Garden of the University of Göttingen and from wild populations in northern and central Germany. At the start of the pot experiment, in July 2006, only 23 species (Table S1) were available in sufficient numbers of vigorous individuals of similar size for inclusion in the experiment. Accordingly, we set up six levels of diversity with these 23 species, cultivating each in a monoculture, in 12 randomly selected combinations of two species and in eight combinations of 4, 7, 11 and 23 species under the conditions described. This resulted in a total of 67 pots for each of the four treatments, giving a grand total of 268 pots. We were unable to replicate the monoculture treatments because of space limitations in the greenhouse. Each pot was filled with 1·5 L of a mixture of sand and clay heated for 24 h to kill any alien seeds. The planting surface was 205 cm2, and in each pot, we planted 23 individuals, irrespective of their size or weight. After planting the species combinations on 08–09 July 2006, all pots were kept for 1 week in partial shade with ample water and for another 4 weeks at optimal conditions of full sun and water levels continuously 1–2 cm above the soil surface. As individual plants could not be accurately weighed prior to the experiment because of attached soil particles and water, to assess the average initial weight of each species, we randomly selected five additional individuals per species that were washed, separated into roots and shoots, dried at 80 °C for 24 h and weighed to an accuracy of 0·1 mg on a BP161P-OCE scale, Sartorius, Göttingen (Germany).

The different treatments were initiated on day 37 of the experiment (14 August 2006). The control treatment was kept under ambient conditions throughout the experiment. In the drought treatment, pots were placed in a greenhouse, a hole was drilled through the bottom of each pot, and 50 mL of water was added once a week. In the shade treatment, we placed the pots in a shaded outside area without direct sunlight and kept water levels at 1–2 cm above the soil surface. In the salt treatment, we added 0·2 g of salt (NaCl) per pot, but otherwise kept the plants in the sun and with water levels at 1–2 cm above the soil surface. After salt application, we observed an effect on the plants (changes in colour and dried tips of the leaves) and therefore added no further salt. A comparison with ambient or extreme soil and water conditions in the field was not possible because of the wide variation in their natural habitat. Within each treatment, pots were repositioned randomly on a weekly basis. Plants were harvested on days 107–108 (23–24 October 2006). For each pot, above- and below-ground biomass was harvested separately, and above-ground parts were separated by species. This plant material was placed in individual paper bags, initially dried at ambient conditions for storage and subsequently dried and weighed as detailed earlier.

Statistical analyses

We used initial (planted) species richness for the analysis because it defined treatments in the experimental design. The decline in species richness in the assemblages is shown in Fig. S1 Supporting Information. We first analysed the effect of biodiversity on biomass production. Secondly, we partitioned the overyield, i.e. that communities are more productive than expected based on the monoculture yields of the constituent species, among the complementarity effect, i.e. that most or all species in a polyculture perform better than in their respective monocultures, and the selection effect, i.e. that few species benefit at the expense of the others, following Loreau & Hector (2001). In both cases, we fitted linear mixed-effects models (Pinheiro & Bates 1996) using the lme function from the nlme library for R 2.9.2 (R Development Core Team 2009) with species richness as an explanatory variable for the different treatments and the identity of the assemblage as random factor. Thirdly, we visualized patterns of species composition in the different treatments at the diversity level of 23 species by Principal Component Analysis (PCA). Finally, we analysed the variability over the different treatments as standard deviation and coefficient of variation with linear models.


At harvest after three and a half months (107–108 days), biomass per pot had increased 5–12-fold on average. The patterns of above-ground and below-ground biomass production were similar to each other (Fig. S2 Supporting Information), so we present only total biomass in detail. A linear mixed-effects model with species richness as the explanatory variable for the different treatments and the identity of the assemblage as a random factor showed significantly lower intercepts and higher slopes for regressions of overall biomass production as a function of species richness for all three manipulated treatments compared with the control (significant species richness x treatment interaction, P = 0·023; Fig. 1, Table 1). In the control, no significant increase in biomass production was detectable (slope versus ln species richness with 95% CI = 0·98 (−0·55–2·52). However, the experimental systems showed a stronger increase in biomass production with species richness than the control (change in slope with 95% CI = 0·77 [0·11–1·42], 0·67 [0·02–1·33], 0·98 [0·33–1·64], for the drought, salt and shade treatment, respectively), indicating that there was a positive effect of biodiversity under changed environmental conditions (Fig. 1 and Table 1). Total biomass in the experimental treatments relative to the control averaged 15–45% lower in assemblages of 1 or 2 species, but only 5–15% lower in assemblages of 23 species (Fig. 1).

Figure 1.

 Total biomass production (above-ground + below-ground biomass) to the log-transformed species richness in experimental pots containing 1–23 species of marsh plants for the four different treatments.

Table 1.   Effect (total biomass, complementarity effect, selection effect, standard deviation (SD) and coefficient of variation (CV) of experimental drought, salt addition and shade. Results are shown as mean and 95% CI
 SourceTotal biomassComplementarity effectSelection effectSDCV
  1. The effects are reported as the value for the control and the differences (in italics) between control and the other treatments.


Our analysis used above-ground biomass only because the below-ground root mass could not be separated by species. When we partitioned the net effect of biodiversity (see Fig. S3 Supporting Information) into complementarity and selection effect components following the method from Loreau & Hector (2001), the effect on overyielding differed significantly between treatments (significant species richness x treatment interaction, P = 0·003 for the complementarity effect and P = 0·004 for the selection effect; Fig. 2; Table 1). The strengths of both the complementarity and selection effects were not related to increasing species richness except for the shade treatment. For this treatment, the complementarity effect increased with increasing species richness (slope with 95% CI = 1·33 [0·61–2·06]), while the selection effect decreased (slope with 95% CI = −0·93 [−1·54 to −0·32]; Fig. 2; Table 1). An outlier in the shade treatment with a value of 49·3 for the complementarity effect and a value of −48·2 for the selection effect was removed from the analysis. The results were conservative and unaffected with or without this outlier but we removed it for the clarity of the results.

Figure 2.

 Linear regression slopes and 95% confidence intervals for the relationships between complementarity and selection components in response to log-transformed species richness in experimental pots containing 1–23 species of marsh plants with a grass-like growth form. Grey line in manipulated treatments shows the slope in the control treatment for easy comparison. To be conservative an outlier in the shade treatment with a value of 49·3 for the complementarity effect and a value of −48·2 for the selection effect was removed from the analysis.

When we analysed changes in species composition because of the different treatments at the diversity level of 23 species by PCA, we found no directional shifts in the relative biomass of the plant species when comparing the control to all manipulated treatments or when comparing the three treatments under variation (Fig. S4 Supporting Information). Axis one explains 19·1% and axis two 12·6% of the observed variation. The variation of the species Scirpus holoschoenus, Equisetum scirpoides and Festuca arundinacea is highly influential along axis one. When the plant assemblages at the beginning of the experiment were included in a PCA, all treatment plots were clustered together. The composition at the start of the experiment was located at the opposite side of the graph, indicating that all species changed their biomass in similar ways in all treatments (not shown).

Analysing the variability over the different treatments as standard deviation (SD) and coefficient of variation (CV), the fitted linear models showed similar patterns of above-ground, below-ground and total biomass production (Fig. S5 Supporting Information) and for clarity we focus on the variability of the total biomass production (Fig. 3). The SD of total biomass for each treatment showed a significant decrease with increasing species richness with no difference between the treatments (slopes with 95% CI = −1·74 [−2·24 to −1·23]; Fig. 3a; Table 1). The SD of the control treatment (intercept = 7·89) was not significantly different from that of the drought (change in intercept with 95% CI = −1·27 [−2·76–0·22]) and salt treatments (change in intercept with 95% CI = −0·85 [−2·34–0·64]) but differed significantly from that of the shade treatment (change in intercept with 95% CI = −2·15 [−3·64 to −0·65]; Fig. 3; Table 1). The CV for each treatment also declined significantly with increasing species richness with no differences between the treatments (slopes with 95% CI = −30·34 [−36·09 to −24·60]). Again, only the CV of the shade treatment differed from the control (change in intercept with 95% CI = 17·12 [0·16 – 34·07]; Fig. 3b; Table 1). In contrast to SD, the values of CV were lower in the control than in the manipulated treatments, although these differences were only significant for the shade treatment.

Figure 3.

 Variability of biomass production as standard deviation (a) and coefficient of variation (b) to differences in species richness in experimental pots containing 1–23 species of marsh plants.


This is the first study to compare the functionality of ecosystems subject to different environmental conditions to a control and to provide an assessment of the underlying mechanisms producing consistency of biomass production under environmental variation. Unexpectedly, we found that biomass production in our experimental plant assemblages did not significantly increase with species richness under ambient conditions (control). This contradicts the commonly documented link between biodiversity and productivity, although some previous studies have found similar results (Hooper et al. 2005; Balvanera et al. 2006). However, a trend towards increasing biomass with increasing biodiversity was visible in the control; the statistical weakness of the relationship probably reflects the high variability in biomass at low levels of diversity. The latter may reflect the wide range of weights for different species at the beginning of the experiment, ranging from 0·0014 g in Pilularia globulifera to 0·2563 g in Carex cf. gracilis mean dry biomass per individual. The species with the highest biomass in monoculture (Carex cf. gracilis) also had the highest biomass in all mixed assemblages. This might lead one to conclude that a monoculture of this species might produce more biomass than a mixed assemblage. However, this species also had the largest individuals at the beginning of the experiment, and its relative biomass increase was not exceptional.

Importantly, we found that the increase in biomass production was more pronounced in the manipulated systems relative to the control, as also found by Mulder, Uliassi & Doak (2001). In other words, the positive effect of biodiversity on productivity was stronger when there was environmental variation, in this case availability of water and light. This was especially true for the shade treatment, providing clear evidence for a stabilizing effect of biodiversity under environmental variation, i.e. that these systems performed relatively better with increasing biodiversity compared to the control system. Such effects have been difficult to document, with results from other experimental studies ranging from negative to positive (Griffin et al. 2009), leading Balvanera et al. (2006) to conclude that stabilizing effects of biodiversity are weak at best.

We set out to study the response of a set of artificial plant assemblages of different species richness to environmental variation. Our aim was not to compare the quantitative response of the assemblages to a specific variation, as this would have required the same intensity of disturbance across all treatments and it is impossible to quantify the disturbance caused by salt application versus drought, for example. Indeed, the intensity of abiotic variation can only be defined through the functional response of the organisms (Villarreal-Barajas & Martorell 2009), and a similar intensity of environmental change would thus by definition results in a similar functional response. Furthermore, with the exception of the salt treatment, our experimental design included combinations of different factors in each treatment, rather than single, isolated factors. For example, by conducting the drought treatment in a greenhouse, we also changed air temperatures and perhaps even ambient gas concentrations. Similarly, in the shade treatment, the plants not only experienced lower insolation but also lower temperatures than in the control. Our aim was to subject the plants to three ecologically meaningful treatments that were as dissimilar as possible. For convenience, we call these treatments ‘drought’, ‘shade’ and ‘salt’ but we are aware that they include complex combinations of conditions differing from the control. Furthermore, our results might have been influenced by the density of plants in each pot (Garnier et al. 1997; Huston et al. 2000) or by the relatively short duration of the experiment (van Ruijven & Berendse 2003; Spehn et al. 2005). In particular, we cannot know whether the observed effects were because of early competition among individuals, interspecific or intraspecific interactions. However, theoretical predictions about biodiversity-mediated resistance under varying environmental conditions have not distinguished between these aspects either (Doak et al. 1998; Tilman, Lehman & Bristow 1998; Yachi & Loreau 1999).

Unexpectedly, we found that the observed effects did not depend on shifts in species composition, and hence the commonly assumed selection effect (McNaughton 1977; Yachi & Loreau 1999; Loreau et al. 2001), but rather resulted from a complementarity effect. This was significant for the shade treatment and visible as trends in the other treatments. It was slightly counteracted by the selection effect but explained the positive effect of biodiversity on productivity (net effect) in this treatment. The partition of selection and complementarity effects following the method of Loreau & Hector (2001) does not directly address the underlying mechanisms determining their effects. Rather, they point to groups of mechanisms that may all result in similar effects. Thus, resource use complementarity or interspecific benefits such as nitrogen fixation through legumes may both lead to a complementarity effect. Although we cannot pinpoint the precise biological mechanism responsible for the observed increased consistency of biomass production under environmental variation with increasing biodiversity, we can still conclude that this mechanism favours the growth of the majority of species in our experiment rather than a few selected species benefitting at the expense of most other species. The results of the additive partitioning analysis and the relative increase of the complementarity effect in the stress treatments are consistent with decreased intraspecific competition in the mixed treatments.

Turning to the variability of biomass production, we found that both the standard deviation (SD) and the coefficient of variation (CV) decreased with increasing species richness. Because the assemblages in the different treatments were identical, in this case a direct comparison between the variability of the control treatment and the treatments under environmental variation was possible (Fig. 3). We were not so much interested in the decline itself but in how consistent it was across treatments. The decreasing variability with diversity may be influenced by several factors. First, the increasing similarity in the experimental assemblages with increasing species richness will by itself decrease the observed functional variability. This is typical for experimental studies of this kind with monocultures at one extreme and all species combined at the other end of the diversity gradient. Secondly, the relationship of the CV with species richness is affected by the fact that the variance is higher when the mean is smaller for assemblages with low species richness and smaller when the mean is higher for assemblages with high species richness. Beyond this general trend, however, we found subtle differences between the control and the experimental systems. The SD of biomass production was lower in the manipulated systems, while the CV was higher. The reason for this apparent discrepancy lies in the fact that the mean biomass production was lower in the experimental systems compared with the control. The SD, which measures the variability in absolute terms, was therefore lower in the manipulated systems, although relatively speaking their variability was higher, as expressed by the CV. This suggests that the changes in environmental conditions in our experiment destabilized the systems to some degree. Importantly, this comparison shows that the use of different measurements of variability may result in apparently contradictory results, and that a comparison of different studies should carefully consider the variability measures that were used.

In conclusion, we found significant support for complementarity-mediated consistency of biomass production under environmental variation in an experimental pot study of freshwater marsh plants. Importantly, the responses in the three different treatments were qualitatively similar, suggesting that the stabilizing effect of biodiversity is a general community attribute and does not depend on the type of abiotic variation. Surprisingly, we found that the effects of biodiversity in the manipulated treatments resulted not from strong selection for species that did best under these conditions but rather from complementarity. Although comparative studies from other systems are needed before general conclusions can be drawn, our study suggests that the effects of environmental and climate changes will be less evident in assemblages with high diversity. This is positive in the sense that such assemblages will remain more stable but it may also be misleading if changes in assemblage composition are used to assess the intensity of environmental changes in differently rich ecosystems. Overall, our study shows that maintenance of high local biodiversity, as opposed to preserving high diversity only in selected sites, is necessary to buffer ecosystem functioning in the face of changing environmental conditions.


We thank C. Körner for inspiring this study, M. Becker, G. Jacob and M. Schwerdtfeger of the Old Botanical Garden of the University of Göttingen for help in the cultivation of the specimens, K. Adler, M. Bos, F. Lipfert, M. Müller, H. Muth and S.G. Sporn for help with the experiments, and S. R. Gradstein, M. Jones and A. R. Smith for comments on the manuscript. Further, we thank two anonymous referees for their comments on a previous version of the manuscript.