Strong mixture effects among four species in fertilized agricultural grassland led to persistent and consistent transgressive overyielding

Authors

  • Daniel Nyfeler,

    1. Agroscope Reckenholz-Tänikon Research Station ART, Reckenholzstrasse 191, 8046 Zürich, Switzerland
    2. Institute of Plant Sciences, ETH Zürich, Eschikon 33, 8315 Lindau, Switzerland
    Search for more papers by this author
  • Olivier Huguenin-Elie,

    Corresponding author
    1. Agroscope Reckenholz-Tänikon Research Station ART, Reckenholzstrasse 191, 8046 Zürich, Switzerland
      *Correspondence author. E-mail: olivier.huguenin@art.admin.ch
    Search for more papers by this author
  • Matthias Suter,

    1. Agroscope Reckenholz-Tänikon Research Station ART, Reckenholzstrasse 191, 8046 Zürich, Switzerland
    2. Institute of Integrative Biology, ETH Zürich, Universitätstrasse 16, 8092 Zürich, Switzerland
    Search for more papers by this author
  • Emmanuel Frossard,

    1. Institute of Plant Sciences, ETH Zürich, Eschikon 33, 8315 Lindau, Switzerland
    Search for more papers by this author
  • John Connolly,

    1. Environmental and Ecological Modelling Group, School of Mathematical Sciences, University College Dublin, Dublin 4, Ireland
    Search for more papers by this author
  • Andreas Lüscher

    1. Agroscope Reckenholz-Tänikon Research Station ART, Reckenholzstrasse 191, 8046 Zürich, Switzerland
    Search for more papers by this author

*Correspondence author. E-mail: olivier.huguenin@art.admin.ch

Summary

  • 1Increasing plant species richness often increases biomass production in nutrient-poor semi-natural grasslands. If such positive diversity–productivity effects also apply to nutrient-rich agricultural grasslands, mixtures could improve resource-use efficiency in the vast area used for forage production. We therefore quantified the diversity–productivity effects in nutrient-rich agricultural grasslands using four-species grass–legume mixtures.
  • 2The sown overall density and species proportions of Lolium perenne, Dactylis glomerata, Trifolium pratense and Trifolium repens were varied in a 3-year field experiment to investigate the effects of species richness (1, 2, 4 species) and species proportion (0, 3, 10, 25, 40, 50, 70, 90, 100% sowing proportion) on productivity under a nitrogen fertilization of 50, 150 or 450 kg N ha−1 year−1.
  • 3The four-species mixtures reached up to twice the yield of the average of the four species’ monocultures (overyielding up to 106%), predominantly due to combining grass and legume species. Mixtures were up to 57% more productive than the most productive monoculture (transgressive overyielding). Both these diversity–productivity effects appeared across a broad range of species proportions and persisted at the two lower levels of N fertilization for 3 years.
  • 4Mixtures fertilized with 50 kg N ha−1 year−1 produced yields comparable to grass monocultures fertilized with 450 kg N ha−1 year−1, if the legume proportion was about 50 to 70%. Diversity–productivity effects were reduced at the highest level of N fertilization, where they virtually disappeared in the third year. Increased N fertilization also accelerated the observed general trend towards D. glomerata dominated and legume-poor swards.
  • 5Synthesis and applications. Diversity–productivity effects led to consistent transgressive overyielding in intensively managed grasslands, suggesting a highly increased resource-use efficiency in mixtures. Performance better than monocultures can be achieved with grass–legume mixtures that have a low number of species, across a wide range of species proportions and in nutrient-rich conditions. Processes such as niche complementarity and positive interspecific interactions leading to diversity effects proved to be highly relevant and widely applicable for intensive forage production. Such diversity–productivity effects could allow reduced inputs of N fertilizer without loss of productivity in different grassland production systems.

Introduction

Biodiversity experiments that varied plant species richness in nutrient-poor grassland systems have shown that species richness can enhance biomass production (Hector et al. 1999; Hooper & Dukes 2004; Roscher et al. 2005). Plant communities with a higher species richness are expected to better utilize available resources due to species niche complementarity and positive interspecific interactions (Loreau et al. 2001; Hooper et al. 2005). But fundamental differences in ecosystem processes, plant-to-plant interactions (competition for nutrients vs. light) and plant species (slow vs. fast growing, Lambers & Poorter 1992) between nutrient-poor semi-natural and nutrient-rich agricultural ecosystems make it impossible to extrapolate findings on diversity–productivity responses from the former to the latter. In natural habitats of high fertility, the most productive plant communities are typically characterized by low species diversity (Grime 2001). This suggests that the diversity–productivity relationship may be less important, and/or may saturate at a low number of species under fertile agricultural conditions. Indeed, monocultures of highly productive key grass species produce high forage yields of 12 to 17 t ha−1 year−1 when strongly fertilized (Daepp et al. 2000). But the synthesis of mineral nitrogen (N) fertilizer is energy-intensive and its application in large quantities is associated with negative environmental impact. The exploitation of diversity–productivity benefits for more resource efficient grassland production systems is highly relevant in Europe where agricultural grasslands cover more than 40% of utilized agricultural area. However, the need for energy- and protein-rich forage in dairy and intensive meat production systems urges farmers to harvest grasslands at an early phenological stage (Herrmann et al. 2005). Such high defoliation frequencies combined with high fertilization rates (intensive management) strongly decrease plant species richness (Kirchner 1977; Gough et al. 2000). Thus, we tested the extent to which diversity–productivity effects persist under fertile agricultural conditions and how they respond to a large gradient of N fertilizer input. The aim was to ascertain whether diversity–productivity effects can be exploited in alternative, productive forage production systems with lower N fertilizer input.

In order to be the most efficient system at a given level of growth resources, a mixture has to perform not only better than the average of all the species grown in monoculture (overyielding), but at least as well as the monoculture of the highest yielding species (transgressive overyielding, Trenbath 1974). This is especially relevant in agricultural systems where the selection of the monoculture cropped is not random but is the best-performing monoculture. In a meta-analysis, Cardinale et al. (2007) found that the most diverse mixture of 44 biodiversity experiments on average only achieved 88% of the biomass of the most productive monoculture. Significant transgressive overyielding occurred in only 12% of the cases and the meta-analysis suggests that it takes about 5 years before the most diverse mixture exhibits transgressive overyielding. However, the inferential methods that have been used to test for transgressive overyielding are subject to serious criticism of bias in favour of monocultures (Schmid et al. 2008). In practice, both the low rate of occurrence of transgressive overyielding and the long time span involved may be sufficient to deter farmers from using the mixtures.

The persistence of diversity–productivity effects over a period of several years requires the maintenance of diversity over time. This may be a problem in fertile agricultural systems where competitive exclusion of key species is an important issue (Faurie, Soussana & Sinoquet 1996; Guckert & Hay 2001). In addition to species richness, diversity–productivity effects may increase with increasing species evenness (a function of species proportions in the mixture) (Wilsey & Potvin 2000; Kirwan et al. 2007). Thus, a key factor in achieving significant diversity–productivity effects may be for mixtures to maintain relatively even species proportions during the cultivation period. Therefore, in this study, we focussed on the effects of species proportions on biomass production and on changes in species proportions over 3 years cultivation time.

On a fertile, productive agricultural grassland, we quantified the influence of species richness and species proportion on diversity–productivity effects. Special emphasis was placed on the influence of N fertilization, the occurrence and persistence of (transgressive) overyielding and on the evolution of the species proportions in the mixture over 3 years.

Materials and methods

establishment of the experiment

In August 2002, monocultures and mixtures were sown on plots of 3 × 6 m on a eutric cambisol at Zurich-Reckenholz (47°26′N, 8°32′E, 491 m above sea level, mean annual temperature: 9·4 °C, mean annual precipitation: 1031 mm). In April 2003, all plots were fertilized once with 30 kg N ha−1 to enable the swards to establish under similar conditions.

experimental design

The experiment was designed to provide (i) a wide range of grass and legume proportions in the sward, (ii) two levels of overall sown densities, and (iii) three levels of N fertilization. The grass species Lolium perenne L. cultivar (cv.) Lacerta and Dactylis glomerata L. cv. Accord, and the legume species Trifolium pratense L. cv. Merviot and Trifolium repens L. cv. Milo were selected as model species. L. perenne and T. pratense were expected to be fast-establishing species, D. glomerata and T. repens were expected to have a slow establishment but to be more persistent. Thus, these species served as a model system for different functional groups (grasses vs. legumes and fast vs. slow establishing) in productive agricultural systems. The monocultures of the four species and 21 different mixtures were set up following a simplex design (see Supporting Information,Table S1; Cornell 2002; Kirwan et al. 2007). The 21 mixtures were binary stands (50% of two species), equal stands (25% of each of four species), moderately dominant stands (70% of one species, 10% of three others), extremely dominant stands (90% of one species, 3·3% of three others), or co-dominant stands (40% of two species, 10% of two others). All mixtures and monocultures were sown at two levels of overall sown density, with the high level being the recommended seed weight (100%) under Swiss conditions, and the low level being 60%. At the high sown density, monocultures of L. perenne, D. glomerata, T. pratense and T. repens were sown with 35, 25, 12 and 15 kg seeds ha−1, respectively.

The plots were fertilized with 50, 150 or 450 kg N ha−1 year−1 as NO3NH4 (called N50, N150 and N450 hereafter); the recommendation for intensively managed agricultural grassland in Switzerland is 150 to 180 kg N ha−1 year−1. Because the first N application after sowing was the same for all plots (30 kg N ha−1), the effective total N fertilization in the first experimental year (2003) was 70, 150, and 390 kg N ha−1. All types of monocultures and mixtures were included in the N150 treatment (50 plots, Supporting Information, Table S1), whereas the N50 and N450 treatments only included the monocultures, the equal stands and the moderately dominant stands (18 plots). The 86 plots were arranged in a randomized design.

maintenance and measurements

All plots received 40 kg P ha−1 year−1 (as superphosphate) and 220 kg K ha−1 year−1 (as potassium sulphate) in early spring. According to recommendations for intensively managed agricultural grasslands in the Swiss lowland (5 to 6 cuts year−1) all plots were cut five times annually (2003, 2004 and 2005) at 5 cm above-ground surface using an experimental plot harvester (Hege 212). Annual N fertilization was split into five equal applications in mid-April, and 3 to 15 days after the first four defoliations depending on weather conditions. The total yield of the central 1·5 × 6 m of each plot was determined at each harvest by drying subsamples to constant weight (65 °C, 48 h). The biomass proportions of the four sown and unsown species were measured at the first, third and fifth harvest by manually separating plant samples from permanent sub-plots (0·8 × 0·3 m).

data analysis

The analysis of simplex designs is based on multiple linear regression and replicates of community compositions are not essential for estimation of coefficients or their standard errors. The estimate of the yield of a specific community (monocultures included) derives from all of the communities in the design. The biological application of this modelling approach to test for different patterns of interspecific interactions are described in Kirwan et al. (2009) with a detailed example of application. The influence of the overall sowing density, the species proportion and the N fertilization on the total yield was tested as described in Kirwan et al. (2007, 2009; see Supporting Information, Appendix S1 and Table S2). Following this procedure, we modelled total yield (y) as:

ŷ = αM + β1GLp + β2GDg + β3LTp + β4LTr + β5BGL + β6WGL + β7GLp · N1 + β8GDg · N1 + β9LTp · N1 + β10LTr · N1 + β11BGL · N1 + β12WGL · N1 + β13GLp · N2 + β14GDg · N2 + β15LTp · N2 + β16LTr · N2 + β17BGL · N2 + β18WGL · N2(eqn 1)

with

N1 = 1, if fertilization level = N150, and N1 = 0 otherwise,

N2 = 1, if fertilization level = N450, and N2 = 0 otherwise, and

where M is the overall sowing density, GLp is the proportion of grass Lolium perenne, GDg is the proportion of grass Dactylis glomerata, LTp is the proportion of legume Trifolium pratense, LTr is the proportion of legume Trifolium repens, BGL is the sum of the pairwise products between each grass species and each legume species proportion (GLp · LTp + GDg · LTr + GLp · LTr + GDg · LTp) to estimate the interaction effects between grasses and legumes, and WGL is the sum of the pairwise products of the two grass species proportions and the two legume species proportions (GLp · GDg + LTp · LTr) to estimate the interaction effects within grasses or legumes.

The sowing proportions served as regression predictors for the first year, and the analysis over all 3 years. The biomass proportions of the first year served as predictors for the second, and the proportions in the second year were used as predictors for the third. Equation 1 provided yield estimates for any mixture at each N fertilization level. Yield predictions based on this model were highly reliable, indicated by high correlations between predicted and observed values and normally and independently distributed residuals (r2 = 0·84 on average, see Supporting Information, Fig. S1). The calculation of the significance range for transgressive overyielding was developed in acknowledgement of the difficulties inherent in the generally used procedures (Supporting Information, Appendix S2).

We used the relative growth rate difference (RGRD) method (Connolly & Wayne 2005; Suter et al. 2007) to investigate the change in species proportions over time. This method is based on differences in relative growth rates between the species, and models the changes in species proportions as linear functions of treatments such as N fertilization and the initial biomass of each species. The species’ biomass in the first year (initial biomass) served as a predictor for changes in the species proportions in the second year, and the second year's biomass served as a predictor for changes in the third year. Further details on the RGRD analysis are provided in Supporting Information, Appendix S3. All analyses were performed using the statistical software r (r Development Core Team 2008).

Results

The overall sown density had no effect (Supporting Information, Table S2) and results are presented as the mean over both overall sown densities.

sward productivity: effects of species proportions and n fertilization

Over the 3 years and at N50, legume monocultures yielded between 8 and 15·2 t DM ha−1 year−1, whereas grass monocultures yielded only between 4·6 and 7·9 t DM ha−1 year−1 (t DM = ton Dry Matter; Table 1). N fertilization strongly increased the yield of grass monocultures up to 17·2 t DM ha−1 year−1 but generally did not affect the yield of legume monocultures. As a result, the highest yielding species at N50 and N150 were T. pratense (years 1, 3 and over all 3 years) and T. repens (year 2), while at N450, T. pratense (year 1) and D. glomerata (years 2, 3 and overall) performed best. Monoculture yields of L. perenne and T. repens were in the range observed in other experiments under comparable growth conditions (Hebeisen et al. 1997).

Table 1.  Predicted yield of all four monocultures, and predicted yield and overyielding of the equal stand mixture for the three N fertilization levels and the 3 years
  MonoculturesEqual stand mixture
 Overyielding
L. perenneD. glomerataT. pratenseT. repensYieldBGL effectWGL effect
(t DM ha−1 yr−1)(t DM ha−1 yr−1)(t DM ha−1 yr−1)(%)(t DM ha−1 yr−1)(%)
  1. Overyielding is maximal for equal stand mixtures (i.e. equal proportions of all four species) and is calculated as the difference between the equal stand mixture yield and the average yield over the four monocultures (). Overyielding is split into the effect of mixing grasses with legumes (between grasses and legumes, BGL) and the grass–grass and legume–legume mixing effect (within grasses or legumes, WGL). Predictions are based on multiple linear regressions (see equation 1; Supporting Information, Table S3). The calculation of the level of significance is based on regression analysis (for details, see Supporting Information, Appendix S1 and Table S2); ns, not significant; *P ≤ 0·05, **P ≤ 0·01, ***P ≤ 0·001. t DM, ton Dry Matter.

Year 1N50 7·6 7·215·212·010·515·74·6*** 430·6* 6
N150 8·6 8·915·212·011·216·44·6*** 410·6* 6
N45010·911·215·212·012·317·54·6*** 370·6* 5
SE 0·83 0·83 0·46 0·46 0·34 0·38    
Year 2N50 4·6 7·9 8·211·8 8·116·78·6***106 −
N150 9·711·311·111·811·017·46·5*** 59 −
N45014·317·212·011·813·818·54·7*** 34 −
SE 0·91 1·14 0·96 0·62 0·48 0·63    
Year 3N50 5·4 6·410·2 8·0 7·512·74·1*** 551·1**14
N150 8·6 9·610·2 8·0 9·114·13·9*** 431·1**12
N45013·216·010·2 8·011·914·41·5ns 131·1** 9
SE 0·92 0·97 1·01 0·67 0·50 0·71    
Year 1–3N50 6·7 7·411·910·6 9·214·84·9*** 530·8* 9
N150 9·2 9·611·910·610·316·04·9*** 470·8* 8
N45012·013·411·910·612·017·74·9*** 410·8* 7
SE 0·93 0·93 0·51 0·51 0·38 0·43    

Strong overyielding was observed in each year and over all 3 years at N50 (Table 1 and Supporting Information, Table S3). The estimated overyielding due to combining grasses and legumes in equal stand mixtures (BGL effect in Table 1) was 43%, 106%, and 55% in the 3 years. Additionally, there was also a small but significant positive ‘within grass or legume’ overyielding (WGL effect) in the first and third year (6% and 14%). The BGL effect was only slightly reduced with an increased N fertilization in the first year. However, N fertilization decreased the BGL effect from 106 to 34% in the second, and from 55 to 13% in the third year (Table 1 and Supporting Information, Table S3). Nevertheless, overyielding due to interaction between grasses and legumes was significant at N450 in year 1, year 2 and over the 3 years.

A first key finding of our study was that the yield of the most productive mixture exceeded that of the most productive monoculture (transgressive overyielding) at all N levels and all years, although not always significantly (Table 2). Transgressive overyielding declined with increased N fertilization: for individual years it was up to 57%, 53% and 19% at N50, N150 and N450, respectively. Yield of the mixtures and, thus, the level of transgressive overyielding, significantly depended on the legume proportion in the sward (Fig. 1). Nevertheless, it was significant over a broad range of legume proportions at N50 and N150. A second key finding was that the diversity effects were so strong that the mixtures producing the highest yield at N50 were at least as productive as the highest yielding monoculture at N450 in the first, second and over all 3 years (Table 2 and Fig. 1).

Table 2.  Predicted yield of the highest yielding monoculture, and predicted yield, transgressive overyielding and legume proportion of the mixture with the highest yield for the three N fertilization levels and the 3 years
  Monoculture with highest yieldMixture with highest yield
SpeciesYieldYieldTransgressive overyieldingLegume proportion
(t DM ha−1 yr−1)(t DM ha−1 yr−1)(t DM ha−1 yr−1)(%)(%)
  1. Transgressive overyielding is calculated as the difference between the highest mixture yield and the highest monoculture yield. Predictions are based on multiple linear regressions (see equation 1; Supporting Information, Table S3). ns, not significant; *P ≤ 0·05, **P ≤ 0·01, ***P ≤ 0·001; significance level based on regression analysis of equation 1 (for details, see Supporting Information, Appendix S2). Mixtures with highest yield were determined numerically. t DM, ton Dry Matter.

Year 1N50T. pratense15·216·81·6ns1170
N150T. pratense15·217·22·0**1465
N450T. pratense15·218·12·9***1958
SE  0·46 0·390·49  
Year 2N50T. repens11·818·56·7***5756
N150T. repens11·818·06·2***5351
N450D. glomerata17·219·72·5ns1436
SE  1·01 0·870·91  
Year 3N50T. pratense10·213·12·9ns2963
N150T. pratense10·214·34·1ns4150
N450D. glomerata16·017·01·0ns 6 0
SE  1·01 0·860·99  
Year 1–3N50T. pratense11·915·23·3**2763
N150T. pratense11·916·14·2***3556
N450D. glomerata13·417·94·5**3346
SE  0·93 0·460·77  
Figure 1.

Predicted mixture yield for increasing legume proportions and the three N fertilization levels in the second year (a) and over all 3 years (b). Predictions are based on regression analysis (equation 1 and Supporting Information, Table S3), and are displayed for mixtures with equal proportions of the two legume species and equal proportions of the two grass species1. Highest monoculture yield is indicated by horizontal lines2 (above the x-axis). Horizontal lines below the x-axis indicate the range of significant transgressive overyielding (P ≤ 0·05), that is, mixture yield being significantly higher than the highest monoculture yield (for details on the calculation, see Supporting Information, Appendix S2).
1Thus, the shape of the curve visualizes the between grass–legume overyielding (BGL effect in Table 1).
2In the second year, T. repens at N50 and N150 and D. glomerata at N450; over all 3 years T. pratense at N50 and N150 and D. glomerata at N450 (Table 2).

At N50, the highest mixture yield was predicted for legume proportions of 70% for the first, 56% for the second, and 63% for the third year (Table 2). Increased N fertilization reduced the legume proportion needed to achieve the highest yield, especially in year 3. The effect of the presence of legumes was therefore negligible to achieve maximal yield at N450 in year 3 (Table 2 and BGL effect in Table 1).

species proportions: shift over time and effects of n fertilization

In the first year, fast-establishing species (L. perenne, T. pratense) were present in proportions above or close to the sowing proportions, whereas the converse was true for the two slow-establishing species (Fig. 2-I). At N50, the observed legume proportion averaged over all four-species mixtures was 42% in the first year and increased to 56% in the second year. In the third year, the average legume proportion declined to only 24% and D. glomerata became the dominant species (Fig. 2; Supporting Information, Table S1 and α-coefficients in Table S4).

Figure 2.

Change in species proportions from one year (x-axis) to the following year (y-axis) at three nitrogen fertilization levels. Values above the 1 : 1 equilibrium line indicate an increase in species proportion, values below signify a decrease. (I) represents the mean observed proportions of a species during the first year depending on the sowing proportions. Error bars (SE) refer to overall means. (II) and (III) represent predicted proportions based on the analysis of the relative growth rate differences for the second and third year (Supporting Information, Table S4). Horizontal lines below the curves indicate the range of values for which vertical differences between the equilibrium line and the predicted proportions are significant (P ≤ 0·05).

Species proportions were significantly affected by N fertilization, which favoured grasses over legumes (Fig. 2). In the first year, average legume proportion over all four-species mixtures at N450 reached 32%, not too much below that of N50. However, at N450, the legume proportion decreased to 24% in the second year and to 5% in the third year (Fig. 2-I and Supporting Information, Table S1). The rapid and strong decline of legumes at N450 was mainly the result of a strong suppression of T. repens by both grasses (γ-coefficients in Supporting Information, Table S4). In the third year, D. glomerata became the dominant species in all N treatments (Fig. 2 and α-coefficients in Supporting Information, Table S4). Shifts of species proportions over time were mainly driven by the species’ identity (α-coefficients), i.e. their ability to produce biomass and by their reaction to N fertilization (γ-coefficients).

Discussion

strong diversity–productivity effects lead to robust transgressive overyielding

Over the 3 years, equal stand mixtures produced 1·6, 1·55 and 1·5 times the biomass yield of the average of the four species monocultures when fertilized with 50, 150 and 450 kg N ha−1 year−1, respectively (Table 1). These results demonstrate that productive agricultural grassland systems can reach the same range of overyielding (factor of 1·7) as found in the meta-analysis of biodiversity–productivity experiments on temperate grasslands, tundra, estuaries and bryophyte assemblages by Cardinale et al. (2007). In contrast to Cardinale et al. (2007), who reported mixture yield to be on average only 88% of the most productive monoculture, high levels of transgressive overyielding were reached under our agricultural conditions: the highest mixture yield was 1·27 to 1·35 times the yield of the most productive species’ monoculture (Table 2, average over 3 years). In addition, we observed significant transgressive overyielding in the first year, while Cardinale et al. (2007) found that it takes on average about 5 years before transgressive overyielding is achieved.

In the meta-analysis of Cardinale et al. (2007), the yield of the mixture was 1·7 times that of the average monoculture but 0·88 times that of the highest yielding monoculture. It follows that the yield of the highest yielding monoculture was 93% above the yield of the average monoculture (1·7 : 0·88 = 1·93), which requires very strong overyielding before transgressive overyielding is achieved. In our experiment, however, the difference in yield between the highest yielding and the average monoculture was only 42% at N50 and 28% at N450 (Table 1, average of the difference in each of the 3 years), which reduces the step from overyielding to transgressive overyielding. The reason for the exceptionally small yield differences among the four species’ monocultures in our experiment is the deliberate selection of the four most important grassland species for intensive forage production in the temperate climate. Thus, our findings of strongly increased resource efficiency due to mixing effects as niche complementarity and positive interspecific interactions are highly relevant for designing productive agricultural systems applicable across vast areas. Additionally, the transgressive overyielding was robust as it occurred (i) in individual years and over all three experimental years, (ii) over a wide range of N fertilization levels, and (iii) over a wide range of species proportions. Remarkably, significantly lower mixture yield than the highest monoculture yield occurred only in a small range of species proportions (range of significance not shown but only below 3% legumes in Fig. 1a). Thus, under our experimental conditions, growing a mixture was almost always more or at least as productive as the best-performing monoculture.

components of diversity–productivity effects

A dominant component of the strong BGL effects is likely to be related to the legume's access to the unlimited N source of the atmosphere. In comparable systems, the importance of N input through symbiotic N2 fixation was demonstrated by Zanetti et al. (1997) who found that up to 300 kg N ha−1 year−1 of the harvested N was derived from symbiotic N2 fixation in T. repensL. perenne mixtures. Symbiotic N2 fixation can only play an important role in diversity effects if symbiotic N2 fixation of legumes performs well and if nitrogen is a main growth-limiting resource. Symbiotic N2 fixation must have performed well in our experiment as the legume monocultures did not respond to the very strong increase in N fertilization from N50 to N450 (Table 1), and thus, did not depend on mineral N. Additionally, N was a main growth-limiting resource for the grass-dominated swards at N50 and N150, since the increase of N fertilization to 450 kg N ha−1 year−1 strongly increased their yield. The high level of N fertilization must strongly have reduced the limitation by mineral N (Daepp, Nösberger & Lüscher 2001) and consequently, the BGL effect was lower at N450 than at N50 or N150.

The BGL-effect for individual years still reached between 13% and 37% at N450, although N can be supposed to be nearly non-limiting at this fertilization level. This indicates that differences in plant traits other than the ability to fix atmospheric N2 contributed to the BGL diversity effect. Such other traits must also be the reason for the significant diversity effects within grasses or legumes (WGL-effect) and they compare with diversity effects reported from experiments in grassland also in the absence of legumes (van Ruijven & Berendse 2003; Roscher et al. 2008). Several such traits may be relevant in our experiment. Firstly, the deep tap root of T. pratense may have been relevant for nutrient and water uptake from deeper soil horizons in contrast to the other three shallow rooting species. Secondly, characteristic within-season growth patterns favouring the grasses in spring during reproductive growth (Daepp et al. 2001) and the legumes in summer when temperatures are high (Lüscher, Fuhrer & Newton 2005) could lead to temporal niches within the growing season that would contribute to the BGL-effect. Thirdly, a strong shift over the 3 years in the species’ performance from the fast-establishing but less persistent legume T. pratense to the slow-establishing but persistent grass D. glomerata was observed. Over the 3 years, this development in the mixtures led to a temporal niche differentiation between T. pratense and D. glomerata, being a component of the BGL-effect. For the monocultures, however, this led to a change in the most productive species between the years, which explains why transgressive overyielding at N450 reached 33% over the 3 years but was only 6–19% in individual years (Table 2).

species proportions

The magnitude of overyielding was strongly influenced by the species’ proportions as derived from equation 1 and as shown for the legume proportion in Fig. 1. Equal stand mixtures overyielded more than mixtures dominated by one species. This compares with results of the combined analysis by Kirwan et al. (2007) and the results of Wilsey & Potvin (2000). In contrast to Kirwan et al. (2007), overyielding was in our experiment not best described by a model based on evenness, because the pairwise interspecific interactions significantly differed in their effect on overyielding (equation 1, Table 1 and Supporting Information, Table S2). The model that best fitted our data predicts the highest overyielding at the highest level of evenness (evenness = 1, equal stands) at all levels of N fertilization, and further predicts that, at a given evenness level < 1, overyielding is larger at equal proportions of grasses and legumes (GLp + GDg = LTp + LTr).

The proportion of the legumes in the mixtures significantly decreased over the 3 years and the mixtures ended up as D. glomerata-dominated, especially at the highest N level (Fig. 2). We expected the proportion of the non-persistent T. pratense to decline over the 3 years. However, the persistent legume T. repens was not able to expand in the sward and contribute to a high legume proportion (Fig. 2 and Supporting Information, Table S1). With its stoloniferous growth habit, T. repens is ineffective in placing its leaves in the upper levels of mixed canopies to compete successfully for light especially at high N fertilization (Schwank, Blum & Nösberger 1986; Faurie et al. 1996). The predominant role of competition for light in fertile agricultural grasslands is evident from the high leaf area index (LAI) of 5 to 9 achieved in these systems (Daepp et al. 2001; Suter, Nösberger & Lüscher 2001). This strongly contrasts to LAI values in low productive ecosystems (van Wijk, Williams & Shaver 2005). Competition for light is asymmetric (Berntson & Wayne 2000), meaning that light intercepted by taller plants is not available for smaller plants. This suggests that in stands with high LAI, niche complementarity for light should be small. Consequently, our highest N fertilization may have affected diversity effects in two ways. Firstly it reduced the ‘potential’ diversity effect by shifting the system from N-limited with a large BGL-effect due to niche complementarity for N to a light-limited system with little niche complementarity for light. Secondly, by favouring grasses at the expense of legumes, N fertilization reduced the ‘realized’ diversity effect by reducing the legume proportion achieved in the mixture far below that needed for highest overyielding.

strategies for agricultural systems

By documenting transgressive overyielding in highly fertilized and productive grassland, our results complement studies showing positive diversity–productivity effects in nutrient-poor grasslands (Hooper et al. 2005). Positive diversity–productivity effects can therefore be expected under a wide array of agricultural situations, and thus, for large areas used for forage production. Mixtures with as few as four key species at optimized species proportions resulted in great benefit and even transgressive overyielding. An advantage of systems with relatively low species numbers is that species (and genotypes) with specific traits can be deliberately combined to maximize diversity effects. Such a combination of species and genotypes might contribute to maximizing gains through diversity effects in mixtures in two ways (Luescher & Jacquard 1991): selection for ‘combining ability’ would result in better resource exploration through niche separation (Hill 1990), while selection for more balanced ‘competitive abilities’ would result in more balanced and stable mixtures (Lüscher, Connolly & Jacquard 1992). For example, substituting a less aggressive species for D. glomerata might result in a more stable mixture without loss of productivity.

The potential of grass–legume mixtures compared to grass monoculture is considerable: in our experiment, mixtures fertilized with 50 kg N ha−1 year−1 often produced yields as high as grass monocultures fertilized with 450 kg N ha−1 year−1, if the legume proportion was between about 50% and 70%. Thus, mixing effects such as niche complementarity and positive interspecific interactions strongly increased resource efficiency, which allows saving large amounts of fertilizer N without yield reduction in intensive forage production systems. Unfortunately, our sward management failed to maintain such high proportions of legumes over 3 years. The maintenance of optimal legume proportions is regarded as a key problem for using legumes in fertile agricultural systems (Faurie et al. 1996; Guckert & Hay 2001). The RGRD analysis showed that this problem could be solved only unsatisfactorily by sowing legumes in higher than optimal proportions for highest yields since even swards with a high sown proportion of legumes had low legume proportions in the third year (Fig. 2 and Supporting Information, Table S1). The agricultural use of the positive diversity–productivity effect with grass–legume mixtures should therefore not only concentrate on the determination of adapted levels of N fertilization but also consider management strategies to increase the stability of the species proportions (e.g. different frequencies of defoliation as well as combinations of other species or cultivars). On the other hand, the benefits of mixtures for biomass production and productivity–diversity effects did not depend on a narrow range of species proportions over the 3 years. Mixing effects showed enough robustness and flexibility for their use in such fertile and productive grassland systems. These findings contribute to the development of resource-efficient agricultural grassland systems.

Acknowledgements

We thank the Swiss State Secretariat for Education and Research for funding this research. We were supported by the EU Commission through COST Action 852.

Ancillary