Explaining grassland biomass – the contribution of climate, species and functional diversity depends on fertilization and mowing frequency

Authors

  • Markus Bernhardt-Römermann,

    Corresponding author
    1. Institute of Ecology, Evolution and Diversity, Goethe-University Frankfurt am Main, Max-von-Laue-Str. 13, 60438 Frankfurt am Main, Germany
    Search for more papers by this author
  • Christine Römermann,

    1. Institute of Physical Geography, Goethe-University Frankfurt am Main, Altenhöferallee 1, 60438 Frankfurt am Main, Germany
    Search for more papers by this author
  • Stefan Sperlich,

    1. Department of Econometrics, University of Geneva, Bd du Pont d’Arve 40, 1211 Genève, Switzerland
    Search for more papers by this author
  • Wolfgang Schmidt

    1. Department Silviculture and Forest Ecology of the Temperate Zones, Georg-August University Göttingen, Büsgenweg 1, 37077 Göttingen, Germany
    Search for more papers by this author

Correspondence author. E-mail: bernhardt-m@bio.uni-frankfurt.de

Summary

1. Grassland ecosystems are often used to generate biomass in temperate regions of the world. It is well known that biomass is influenced by climate and biodiversity, but the relative importance of these two factors in relation to management has not been widely studied. To recommend management treatments maximizing biomass yields we aim to quantify the relative effects of climate species and functional diversity on biomass in differently managed grasslands.

2. We studied the development of biomass yields over the last 37 years on a grassland site in Germany, with mowing at five frequencies (one to eight times per year), each with and without fertilization. We measured plant species richness (SR) and functional richness (FR) (the diversity of functional species properties) using presence–absence indices. We also measured species evenness (SE), functional evenness and functional divergence (FD) using abundance weighted indices. Climate was included as the mean temperature and sum of precipitation during the growing period. By relating biomass to the above-mentioned climatic and biodiversity parameters, we extracted the contribution of these to biomass yields.

3. Biomass changed over time for all treatments and was maximal at intermediate mowing frequencies. Temporal changes in biomass were partly explained by climate and different aspects of biodiversity, although this differed significantly between treatments. The relative importance of precipitation was highest at high mowing frequencies; the contribution of temperature was highest on less disturbed, unfertilized plots. FR and SR influenced biomass changes in the most intensive disturbance regimes on unfertilized and fertilized plots respectively. FD was most important on intensively disturbed, fertilized plots. SE influenced biomass at low mowing frequencies.

4.Synthesis and applications. Climate, species and functional diversity influence annual grassland biomass yields but their importance depends on nutrient status and management frequency. Our results indicate that management treatments with intermediate disturbance regimes will maximize biomass yields in temperate environments. This recommendation may become even more important in the context of climate change: at intermediate mowing frequencies the influence of climatic variables on biomass is less important by comparison to different aspects of biodiversity.

Introduction

The way in which different environmental factors influence ecosystem services is an important question in vegetation ecology. In grassland ecosystems human land-use often focuses on yearly biomass yields. It is well-known that biomass yields are largely constrained by water availability (Rosenzweig 1968; Churkina & Running 1998), which is driven by edaphic and climatic factors. On a continental or biome scale, ecosystem productivity is predictable from temperature and precipitation (Knapp & Smith 2001), but at local scales such models are not able to explain the variability in biomass very well (Lauenroth & Sala 1992; Knapp & Smith 2001). In semi-natural grasslands in temperate zones, with an adequate water supply and intensive land-use, it is likely that fertilization and frequency of disturbance may overcome the influence of climatic conditions on biomass yields (Bradford et al. 2006).

In general, fertilization increases biomass of grasslands (Tilman 1988). Additionally, it has been shown that biomass yields are highest at intermediate mowing frequencies, but interactions between disturbance intensity and ecosystem productivity strongly determine biomass production (Huston 1979, 1994; Kondoh 2001). These interactions can be explained by functional adaptations of the vegetation to the current environment: if productivity is high, strong competitors are favoured. If disturbance is high, good colonizers (resprouters) increase in frequency. With high disturbance and low productivity strong competitors cannot survive. If disturbance decreases and productivity increases, strong competitors out-compete good colonizers (Grime 1979; Kondoh 2001). To recommend management treatments maximizing biomass yields of temperate grasslands these interdependencies underline the importance of directly analysing the combined effects of fertilization, disturbance frequency, climatic factors and functional adaptations.

In this paper we relate two different aspects of species and functional diversity to biomass yields: one includes counts of species and trait attributes, the other is based on species and trait abundances. Both aspects show significant impacts on ecosystem services like biomass yields (for species diversity see Cardinale et al. 2007; for functional diversity see Díaz & Cabido 2001 and Villéger, Mason & Mouillot 2008).

The first aspect [species and functional richness (FR)] refers to the probability that a species containing trait attributes aligned with the requirements of the actual management treatment occurs in the local species pool. At high species or FR this probability is increased. Multispecies communities are often more productive than monocultures (Hector et al. 1999; Cardinale, Palmer & Collins 2002). Species or FR is especially important in highly disturbed ecosystems as only species optimally adapted to frequent disturbance may survive.

Species and functional evenness (FE) measures potential niche partitioning and facilitation, and the relative contribution to ecosystem function (‘mass ratio’ hypothesis according to Grime 1998). Under low intensive management without strong filters for adaptations to frequent disturbance, biomass is maximized by the co-existence of species enabling a more efficient resource uptake (Hector et al. 1999; Cardinale et al. 2007).

This papers aims is to quantify the relative effects of climatic factors as well as floristic and functional diversity indices on annual variations in biomass yields at different mowing frequencies and fertilization. To our knowledge, no comparable analysis exists, even though these relationships are of principal importance for sustainable land-use in the context of global climate change. For the current study we used a 37-year data set from a permanent plot experiment located in Germany. This experiment allows long-term analyses of the development of biomass in relation to climate and biodiversity parameters at different management regimes including several disturbance and fertilization treatments. The treatments were applied on grasslands which have developed on former arable-fields since they were abandoned 37 years ago. Yearly vegetation samplings at all treatments were used to calculate indices describing functional and species diversity (based on presence/absence and abundance weighted vegetation data).

Materials and methods

The permanent plots used in the current study are located in the Experimental Botanical Garden of the University of Göttingen, Germany, (180 m a.s.l., 51°34′0″N, 9°56′60″E; mean annual temperature: 8·5 °C; mean annual rainfall: 635 mm; Schmidt 2006). The plots were established in 1969 on a former arable field cultivated until 1968. The soil is deep, calcareous and fertile alluvial loamy brown earth. All experimental sites were adjacent, with identical environmental conditions at the time of experimental set-up. Management treatments were set up on 125 m2 paired plots, one with and one without fertilization. The fertilization treatment started in 1970 and was applied twice yearly by adding mineral fertilizer to replace nitrogen, phosphate and potassium (N, P, K) removed by mowing the previous year (Schmidt 1993, 2006). At each harvest and for each mowing treatment, a biomass sample was analysed for N-, P-, and K-contents. By summing harvests for each year, the yearly amount of nutrients removed by mowing was estimated.

Different mowing treatments represent a disturbance gradient of different timing and frequency, each on fertilized (F-) and unfertilized (U-) plots:

  • 1 Mowing once per year in autumn (at the end of the growing season; U-1A, F-1A).
  • 2 Mowing once per year in spring (at the beginning of the growing season; U-1S, F-1S).
  • 3 Mowing twice per year (end of May and end of July; U-2, F-2).
  • 4 Mowing four times per year (May, June, July and September; U-4, F-4).
  • 5 Mowing eight times per year (2 × May, 2 × June, July, August, September and October; U-8, F-8).

In comparison to mowing once in the spring, mowing in the autumn was assumed to be less intensive, as the yearly growing cycle was complete. Management intensity was taken as the number of annual mowing events; mowing once in autumn was taken as 0·5. Although only one plot was mown for each treatment, this experiment is unique due to its duration and the high number of different mowing treatments.

From 1972 to 2006 (with a 3 years gap between 1999 and 2003), percentage canopy coverage of all vascular plant species was estimated visually on each plot. A list of the most dominant species per treatment is given in Table S1 (Supporting Information). Appendix S1 provides further information on the method.

The mown aboveground biomass per plot was determined at every harvest and the sum per year was calculated. The total living and dead biomass was weighed fresh and 5–10 mixed samples were gathered and dried at 105 °C to give aboveground dry matter.

Climatic data

Monthly temperature and precipitation data were available from the meteorological station in Göttingen, provided by the German Meteorological Service (station number 1564 in 6 km distance away from study site; Fig. 2 in Appendix S2). We included data from the April to October growing season and used mean temperature and the sum of precipitation for this time period in each year for further analyses.

Figure 2.

 Importance of climate and biodiversity parameters for the explanation of variations in biomass. The results are reported by management frequency and fertilized (solid lines) vs. unfertilized (dotted lines) treatments. As a measure of importance we took the relative contribution of climate and biodiversity parameters (a–f) to the total variance of the generalized linear models (GLMs). The relative contribution was deduced from separate GLMs shown in Table 4 by dividing each estimator by the sum of all estimators. These values were multiplied by the r2 of the current GLM to present the relative contributions converted into per cent of total variance explained by the models. Management treatments were merged according to Table 2.

Selection of functional traits related to management intensity

Initially, we used 14 traits representing a broad spectrum of processes determining vegetation development such as competitive ability (including persistence and regeneration), long-distance seed dispersal and crucial stages in the life history of the species (Table 1). We used the multivariate ordination technique RLQ to analyse which traits are related to management intensity. We investigate the relationship between the trait data (Q-table) and the environmental characteristics (R-table) by the use of a third, connecting table (L-table; species-relevés matrix). The resultant RLQ-axes display maximum covariance between the traits and environment. Bernhardt-Römermann et al. (2008) extended this method by applying an optimization process to select those traits which are optimal for describing the environmental gradient under investigation (here mowing intensity). For further details on this method and the used traits see Appendix S1.

Table 1.   Traits used for the trait approach. Trait names, their ecological function (Com: Competitive ability – P, mainly in relation to persistence; R, mainly in relation to regeneration; Dis, Dispersal ability (here traits with a capacity for long-distance dispersal; >100 m); life history (traits on time or processes of crucial life-history stages), trait attributes and data sources (BiolFlor: Klotz, Kühn & Durka 2002; SID: Flynn et al. 2006; CLOPLA: Klimešová & Klimeš 2007; LEDA: Kleyer et al. 2008)
TraitEcological functionTrait attributesSource
Max. canopy height (cm)Com (P); plantContinuousLEDA
Specific leaf area (SLA, in mm² mg−1)Com (P); leafContinuousLEDA
Leaf anatomyCom (P); leaf1, Helomorphic; 2, mesomorphic; 3, scleromorphic; 4, hygromorphicBiolFlor
Leaf persistenceCom (P); leaf1, Persistent green; 2, over-wintering green; 3, summer greenBiolFlor
Leaf distribution along the stemCom (P); stem1, Rosette plant; 2, hemirosette plant; 3, regularBiolFlor
Type of reproductionCom (P & R); plant1, Seeds; 2, vegetativeBiolFlor
Plant life spanCom (R); plant1, Annuals; 2, bi-annuals; 3, perennialsLEDA
Lateral spreadCom (R); stem0, No lateral spread; 1, <0·01 m; 2, 0·01–0·25 m; 3, >0·25 mCLOPLA
Clonal growth organ (CGO)Com (R); stem0, No CGO; 1, above ground runners; 2, above ground plant parts; 3, below ground runners; 4, below ground storage organsCLOPLA
Seed mass (mg)Com (R) & Disp; seedContinuousSID
Dispersal typeDisp; seed1, Nautochory; 2, anemochory; 3, hemerochory; 4, zoochoryLEDA
Duration of floweringLife history1, Short (1–2 months); 2, medium (3–4 months); 3, long (≤4 months)BiolFlor
Start of floweringLife history1, January–April; 2, May–June; 3, July–SeptemberBiolFlor
Type of pollinationLife history1, Abiotic; 2, by animals; 3, selfingBiolFlor

To characterize the change in relevance of the selected functional traits with management frequency, we used the covariance structure between the gradient in management intensity and functional traits. This relationship was tested for significance using randomization tests (compare Römermann et al. 2009).

Diversity indices

Species diversity was described by species richness (SR) which is the number of species; and species evenness (SE) which is a measure for the equal distribution of species abundances in space. As a measure for SE we used the Evar-index being independent from SR as proposed by Smith & Wilson (1996) (for further information see Appendix S1).

Functional diversity was calculated using the pre-selected traits within the framework presented by Villéger, Mason & Mouillot (2008): FR reflects the range of the trait attributes present in the analysed community. FE describes the evenness of abundance distributions in a functional trait space. It quantifies the regularity with which functional space is filled by species weighted by their abundances. The index of functional divergence (FD) represents how species abundance is distributed within the volume of the functional trait space occupied by species. High FD is caused by the clustering of species with high abundances at the edges of the functional trait space. It indicates the degree of resource differentiation or a predominance of extreme species (Mason et al. 2005). FE and FD were both shown as unbiased by SR and FR (Villéger, Mason & Mouillot 2008). A detailed description of the used indices is presented in Appendix S1.

We calculated all diversity measures for each treatment plot within each year. The minimal species number of the plots was 31. All explanatory variables were uncorrelated (r² < 0·3, n.s.).

Changes in biomass over time

To assess changes in biomass over time (1972–2006), generalized linear models (GLMs) were applied to search for differences in biomass production between the management treatments. We corrected the covariance structure of the regression model using heteroscedasticity and autocorrelation consistent (HAC) covariance estimators. First we fitted a model containing time as a continuous variable and management as a factor. In the course of model simplification, we merged management treatments showing no significant differences in the estimators for intercepts and slopes until the resulting model obtained a minimal Akaike information criterion (AIC; Crawley 2007).

To explain the observed changes in biomass over time, we calculated separate GLMs (with HAC correction to account for temporal autocorrelation) for each management treatment (treatments not showing significant differences in biomass development were merged). Climate (temperature and precipitation), and biodiversity indices (species and FR, species and FE, FD) were used as explanatory variables. We simplified the maximal model via backward selection of least significant variables until the final minimal adequate model contained significant terms only (P-value < 0·05) and a minimal AIC was obtained. From this final model we extracted for each term its importance for explaining biomass as its relative contribution to total explained variance: All variables were normalized prior to analyses by scaling between zero and one. Each estimator was divided by the sum of all estimators for each model, giving its relative contribution to total explained variance.

All statistical analyses were done in R 2.9.2 (R Development Core Team 2008), with the additional packages sandwich, ape and vegan. For calculations of the functional diversity indices we used the R-code provided by Villéger, Mason & Mouillot (2008).

Results

Changes of biomass over time

Biomass increased overtime for all management treatments, except for the unfertilized plots mown four and eight times per year where a slightly decreasing trend was found Fig. 1, Table 2, Fig. 1 in Appendix S1). For both unfertilized and fertilized plots there was no difference in changes of biomass over time between mowing four and eight times per year; for the unfertilized plots no difference was found between mowing once in the spring and mowing twice a year.

Figure 1.

 Development of biomass yields from 1972 to 2006. Absolute values plus generalized linear models (GLMs) corrected for heteroscedasticity and autocorrelation. For complete summary statistics see Appendix S2 (Supporting Information). All models were significant (< 0·05). (a–c) Unfertilized plots; (d–g) fertilized. Management treatments were presented together, when the GLMs presented in Table 2 do not differ significantly in intercept and slope.

Table 2.   Temporal changes in biomass for the different management treatments. Treatments that do not show a significant difference in intercept and slope were merged. The generalized linear model (GLM) was corrected for heteroscedasticity and autocorrelation (time series). The final model has null deviance 93·47 on 350 d.f. and residual deviance 5·55 on 336 d.f. (r² = 0·94***). Results of the GLM are in Appendix S2 (Supporting Information). Abbreviations: the first letter refers to the fertilization treatment (U, unfertilized; F, fertilized), the numbers behind the hyphen refer to the management frequency (1A: mowing once in the autumn, 1S: mowing once in the spring, 2: mowing twice per year, 4: mowing four times per year, 8: mowing eight times per year). We tested the intercepts and slopes for differences by using the re-levelling procedure described by Crawley (2007). Different symbols indicate significant differences from random distribution with ***< 0·001, **< 0·01, *< 0·05
 ValueSEz-valueP-value
Intercepts
 U-1A−11·764·279−2·750·006**
 U-1S & U-2−13·083·026−4·320·000***
 U-4 & U-86·163·0262·040·042*
 F-1A−38·224·279−8·930·000***
 F-1S−45·564·279−10·650·000***
 F-2−44·744·279−10·450·000***
 F-4 & F-8−20·203·026−6·680·000***
Slopes
 U-1A0·0060·0022·800·005**
 U-1S & U-20·0010·0030·280·779
 U-4 & U-8−0·0090·003−3·410·001***
 F-1A0·0130·0034·400·000***
 F-1S0·0170·0035·650·000***
 F-20·0170·0035·530·000***
 F-4 & F-80·0040·0031·680·094

On the fertilized plots (Table 2, Fig. 1 in Appendix S1), biomass was highest at the intermediate disturbance frequency mowing twice per year (F-2). Mowing once in the spring (F-1S) had a comparable slope but smaller intercept. Lower biomass was observed for those plots mown once in the autumn (F-1A) and four and eight times per year (F-4, F-8). For the unfertilized plots, biomass was highest for the treatments mown once in the spring and twice a year (U-1S, U-2), but the difference was less marked than on the fertilized plots. In comparison to these, mowing in the autumn (U-1A) again showed an increasing trend but on a lower level, whereas for the unfertilized treatments, mowing four and eight times per year (U-4, U-8) led to a reduction in biomass over time.

Functional traits related to management intensity

On unfertilized plots nine traits were clearly related to management frequency: maximal canopy height, specific leaf area, mesomorphic leaf anatomy, scleromorphic leaf anatomy, leaf persistence, leaf distribution along the stem, lateral spread, duration of flowering and type of pollination. On fertilized plots six traits were linked to management frequency: maximal canopy height, leaf distribution along the stem, plant life span, clonal growth organ, seed mass and zoochorus dispersal (Appendix S1 and Table 3). For both fertilized and unfertilized plots several traits were related to management intensity including decreasing canopy height and regular leaf distribution. For a detailed overview of trait responses to management frequency see Table 3.

Table 3.   Covariance between the gradient in management intensity and functional traits for the fertilized and unfertilized plots. Covariance increases with the intensity of the relationship between the time gradient and the species traits through the species abundance table. For details on the selection procedure of the relevant traits see Appendix S1
 Management intensity
UnfertilizedFertilized
Max. canopy height−0·38***−0·46***
Specific leaf area (SLA)−0·13**
Leaf distribution: rosette plant0·200·42***
Leaf distribution: hemirosette plant0·060·12*
Leaf distribution: regular−0·30**−0·60**
Mesomorphic leaf anatomy0·24***
Scleromorphic leaf anatomy0·11*
Leaf persistence: persistent green0·47***
Leaf persistence: summer green−0·41***
Leaf persistence: over-wintering green0·09
Lateral spread: no−0·09
Lateral spread: <0·01 m−0·08
Lateral spread: 0·01–0·25 m−0·06
Lateral spread: >0·25 m0·33*
Duration of flowering: short (1–2 months)−0·04
Duration of flowering: medium (3–4 months)−0·43***
Duration of flowering: long (≤4 months)0·39***
Type of pollination: animals0·12**
Type of pollination: selfing−0·29**
Type of pollination: abiotic0·21*
Plant life span: annuals0·45***
Plant life span: bi-annuals−0·17
Plant life span: perennials−0·03
Clonal growth organ: no0·15
Clonal growth organ: above ground runners0·92***
Clonal growth organ: above ground plant parts−0·25
Clonal growth organ: below ground runners−0·12
Clonal growth organ: below ground storage organs−0·62
Seed mass−0·02
Dispersal type: zoochory0·34***

Relating changes in biomass to climate, biodiversity and management

The models relating yearly biomass to climate, species and functional diversity parameters gave highly significant results for all management treatments (Table 4 and Appendix S3). In none of the final GLMs did FE remain. All other factors were included in some of the final models, though their effects differed considerably between management treatments.

Table 4.   Estimates and standard errors (SE) of the generalized linear models (GLMs) fitted to biomass yields at different management treatments. The covariance estimates in the regression model were corrected for heteroscedasticity and autocorrelation; all estimates were significant (compare Appendix S3. Abbreviations for estimates: T, temperature; P, precipitation; SR, species richness; FR, functional richness; SE, species evenness; and FD, functional divergence. Abbreviations for management treatments according to Table 2
 U-1AU-1S & U-2U-4 & U-8F-1AF-1SF-2F-4 & F-8
Est.SEEst.SEEst.SEEst.SEEst.SEEst.SEEst.SE
T1·080·560·320·1          
P    0·210·03    0·330·120·270·06
SR−1·330·44−0·380·11  −0·980·23−1·600·48−0·570·210·310·14
FR  0·440·160·230·05        
SE0·950·31    1·690·282·590·47−0·510·29  
FD          1·420·330·210·09
r²0·920·920·860·920·890·920·87

The parameter estimates of these models were used to calculate the importance of climate, species and functional diversity parameters to explain biomass over the last 37 years (Fig. 2). In the unfertilized plots we detected a decreasing importance of temperature and SR with increasing management frequency, while for precipitation and FR an increasing trend was found. SE explained biomass changes in the plot mown once per year in autumn only.

For the fertilized plots there was a decreasing trend in importance for SE to explain biomass changes with increasing management frequency, while for precipitation and FD an increasing trend was found. SR influenced biomass for all fertilized management treatments.

For the unfertilized plots (Table 4, Fig. 3) we found higher biomass at higher mean temperature during the growing period (U-1A, U-1S, U-2), and at higher rates of precipitation (U-4, U-8). Increased biomass was found where there were more species with different functional attributes (U-1S, U-2, U-4, U-8) as well as at more even distributions of plant cover values (U-1A). The opposite result was found for SR: plots mown in spring or twice a year yielded higher biomass at higher species numbers, whereas for the plot mown in autumn a slightly decreasing relationship was found (Fig. 3).

Figure 3.

 Relationship between climate and biodiversity parameters and biomass yields of the unfertilized plots. Management treatments were merged according to Table 2. Relationships are presented for those parameters detected as important for the explanation of differences in yearly biomass (Table 4). Species richness refers to the number of species per 125 m2. Mean temperature and sum of precipitation refer to the main growing season from April to October.

A positive relationship between precipitation and biomass (F-2, F-4, F-8), and between SE and biomass (F-1A, F-1S, F-2), was found for the fertilized plots (Table 4, Fig. 4). For the plots mown once a year (F-1A, F-1S) biomass decreased with increasing species numbers, while an increasing relationship was found for the more frequently mown plots (F-2, F-4, F-8). For the plots mown four or eight times per year higher biomass was found where there was greater FD. In contrast, for the plot mown twice annually biomass was highest where FD was low.

Figure 4.

 Relationship between climatic and biodiversity parameters and biomass yields of the fertilized plots. Details as for Fig. 3.

Discussion

For all treatments biomass changed over time; it was maximal at intermediate mowing frequencies. Both climate and diversity parameters contributed to temporal changes in biomass, but their relevance differed between treatments. All aspects of diversity, including species and functional identity, contributed to our understanding of the functional adaptations to disturbance frequency and fertilization.

Biomass and functional adaptations to disturbance frequency and fertilization

Grassland biomass was influenced by disturbance intensity and fertilization. For all plots mown once or twice per year, both fertilized and unfertilized, there were comparable increases in biomass development over time. This may be due to ongoing adaptations of species to the available resources when annual nutrient removal by mowing does not lead to a decrease in nutrient availability (Tilman 1988). On the fertilized plots, nutrient removal is compensated for by fertilization; on unfertilized plots some nutrients may be replaced by deposition, which may be sufficient to compensate for losses due to low mowing frequencies (Bernhardt-Römermann et al. 2007). In contrast, for the fertilized and unfertilized plots mown four and eight times per year, nutrient replacement and removal by hay-making were no longer in equilibrium (Fig. 1c). This may explain the decreasing productivity of unfertilized frequently mown agricultural lands and old-field communities (Tilman 1988; Huberty, Gross & Miller 1998). Furthermore, on frequently mown sites the influence of intensive disturbance results in functional adaptations such as increasing numbers of annual and rosette species as well as species able to spread laterally; while taller species, species with a regular leaf distribution and higher specific leaf area (SLA, and hence higher growth rates, Cornelissen et al. 2003) decreased, indicating a shift towards lower aboveground competition (Römermann et al. 2008, 2009). Such species are, however, less productive in comparison to those typical for less frequently disturbed sites (Collins, Wein & Philippi 2001; Thompson et al. 2001; Schmidt 2006).

Comparing all fertilized plots, the highest biomass was found when the plots were mown twice per year, in agreement with the findings of Huston (1979), Kondoh (2001) and Haddad et al. (2008), who predicted highest biomass at intermediate disturbance regimes. At an intermediate disturbance intensity, an equilibrium between nutrient removal from the system, which favours species with lower competitive strength, and nutrient replacement by deposition and from the soil, which favours strong competitors, can develop (Kondoh 2001; Haddad et al. 2008). Co-existence of several species is possible. Biomass is maximized by species-specific adaptations to the current environmental conditions allowing an efficient use of the available nutrients (Bernhardt-Römermann et al. 2010). A comparable pattern was found for the unfertilized plots, although we could not detect any difference between treatments mown once a year in the spring and those mown twice a year, indicating that both treatments represent comparable disturbance levels.

Relating biomass to climate and biodiversity patterns

It is striking that the importance of temperature decreased with increasing mowing frequency for the unfertilized plots, whereas the importance of precipitation increased for both the unfertilized and the fertilized plots. In all cases higher biomass was observed at higher temperature and precipitation. These findings accord with the literature: it is well-known that grassland biomass increases with greater water availability and higher temperature as long as there is not a water deficit (Knapp & Smith 2001).

For the frequently mown fertilized plots nutrient availability does not constrain growth but, as for the unfertilized plots, biomass may be enhanced by additional water supply because plant nutrient uptake and water availability are closely related (e.g. Ellenberg & Leuschner 2010). Such effect is even more pronounced during regrowth of vegetation following frequent disturbances by mowing. Thus, the biomass on sites where hay-making is frequent is greater when precipitation is not limiting. There was a decrease in the relative contribution of temperature to biomass with increasing mowing frequency in the fertilized plots. Plants typical for less disturbed plots are mostly strong competitors (see above), with fast growth rates related to greater photosynthetic capacity, which is influenced by temperature (Reich et al. 1999).

Species and FR, SE and FD were also important in explaining variations in biomass. FR had an increasing impact on biomass with increasing disturbance intensity for the unfertilized plots: only species functionally adapted to high disturbance may survive (Kondoh 2001; Dölle et al. 2008). By contrast, SR was important in all management treatments except mowing four and eight times per year on unfertilized plots, but we detected both positive and negative relationships with biomass: For the less frequently mown plots (U-1A, U-1S, F-1A) SR is negatively related to biomass; for all other management treatments there was a positive relationship. The positive relationship between SR and biomass at high disturbance intensities is linked to the probability of optimally adapted species being found in the local species pools (Ozinga et al. 2007). Less clear are the negative relationships between biomass and SR at low mowing frequencies. For the unfertilized plots mown in autumn it should be acknowledged, however, that our results were dependent on just two observations of high biomass; when refitting the models without these data points the importance of SR disappears. For the fertilized plots mown once per year, there appeared to be a slight negative relationship between biomass and SR only. However, the vegetation composition of the plots mown once a year revealed that only few species contribute high cover values to biomass (i.e. Solidago canadensis L. and tall grasses like Arrhenatherum elatius (L.) P. Beauv. ex J. Presl & K. Presl). For such treatments biomass may be better explained by abundance based diversity indices like SE.

The importance of SE on biomass decreased for both fertilization treatments with increasing disturbance. Under less intensive management, species with a high efficiency in resource-use (strong competitors) can dominate grassland communities (Wilson & Keddy 1986; Kahmen & Poschlod 2004) whereas highly specialized plant species are less important (Kondoh 2001; Haddad et al. 2008). This is described by species complementarity, in which biomass is enhanced due to niche partitioning (either in space or time, Tilman 1999; Bernhardt-Römermann et al. 2010) or through interspecific facilitation (Cardinale, Palmer & Collins 2002). Enhanced niche partitioning or interspecific facilitation both relate to dominance structures (see ‘mass ratio’ hypothesis, Grime 1998), and can be detected by increased SE.

In the more frequently mown fertilized plots FD was important in explaining annual biomass yields. This index gives a measure for the clustering of the functional identity of most abundant species (Mason et al. 2005; Villéger, Mason & Mouillot 2008). FD refers to dominance structures of functional adaptations, but characterizes the functional identities of the most dominant species in relation to all occurring functional strategies: If the most abundant species have trait attributes located further from the centre of the functional trait space FD is high, meaning that the most abundant species have extreme functional identities in comparison to the functional identities of all plants present. In other words, if FD is high, the most abundant species are assumed to be specialists. In the plots mown four or eight times per year higher biomass was detected at higher FD, reflecting dominance by more specialized plants, i.e. only species well adapted to such extreme management regimes can survive and dominate the vegetation (Kondoh 2001; Dölle et al. 2008). In contrast, at intermediate disturbance frequencies (F-2) biomass was highest if no specialist plants dominated. Biomass is maximized when species that use the available resources most efficiently can co-exist, i.e. neither plant species specialized to survive frequent disturbance nor strong competitors dominate the vegetation community.

Conclusions

The importance of climate, species and functional diversity parameters in biomass production clearly depends on the frequency of disturbance and fertilization. Our results emphasize the importance of the interaction between nutrient status and management frequency when analysing ecosystem services such as biomass yields of grassland ecosystems. Our results indicate that management treatments with intermediate disturbance regimes will maximize biomass yields. This recommendation may become even more important in the context of climate change: at intermediate mowing frequencies the influence of climatic variables on biomass is less important by comparison to different aspects of biodiversity. More specific management recommendations will depend on the precise fertilization and mowing regime.

Acknowledgements

We thank our field assistants and A. Parth for data handling. We thank two anonymous reviewers and the editors for useful comments on a previous version of the manuscript. We thank GRADE for its offered English language service and proof reading by a native speaker. The research was partly supported by the European Science Foundation (ESF) under the EUROCORES program EURODIVERSITY, through contract No. ERAS-CT-2003-980409 of the European Commission, DG Research, FP6 (Project ASSEMBLE).

Ancillary