Taxonomical and functional diversity turnover in Mediterranean grasslands: interactions between grazing, habitat type and rainfall

Authors


Correspondence author. E-mail: carlos.perez@uam.es

Summary

  1. Changes in livestock grazing regimes are among the most important drivers of species loss and decrease in functional diversity world-wide. However, taxonomic and functional diversities (TD and FD) can respond differently to changes in grazing regime or productivity.
  2. We surveyed plant communities from 67 sites under different grazing regimes (from heavy grazing to grazing abandonment) in wet and dry habitats, in both wet and dry years. We tested the influence of grazing intensity, habitat type and rainfall on TD, FD and the relationship between them. We also partitioned diversity to examine the effects of grazing on TD and FD across scales (within communities, within grazing levels and between grazing levels).
  3. The effect of grazing within and across communities was modulated by water availability, with grazing showing the strongest effects in dry habitats. The relationship between FD and TD varied between habitat types and years and revealed high functional similarity between species (i.e. redundancy) in dry habitats. TD was reduced in the driest conditions across all the observation levels, contrasting with the high temporal stability of FD, suggesting that FD was decoupled from TD, especially in dry habitats. However, despite the high temporal and spatial stability of FD, results show that under severely limited water availability, high grazing pressure can reduce FD, revealing a convergence in traits under the combined effect of grazing and drought conditions.
  4. Synthesis and applications. Results highlight the dependence of functional diversity on the combined effect of water availability and grazing regime. Under severely limited water availability, grazing intensification reduced the functional diversity of these grasslands. Because of the foreseeable reduction in water availability in Mediterranean environments, we recommend the adoption of flexible grazing management schemes that take species and functional diversities into account simultaneously and adapt the level of grazing pressure to water availability.

Introduction

Land-use change is one of the most important drivers of biodiversity loss and ecosystem services changes world-wide (MEA 2005). In many areas of the world, changes in livestock management are leading to profound changes in plant biodiversity patterns (Bakker et al. 2006). Grazing and water stress are often considered convergent selective forces selecting for similar plant traits (Milchunas, Sala & Lauenroth 1988; Quiroga et al. 2010). Consequently, it is generally expected that changes in grazing management would have a more pronounced effect on biodiversity in wetter areas (Cingolani, Noy-Meir & Díaz 2005).

Although taxonomic diversity (TD) has received most attention in the literature, biodiversity includes multiple components, beyond the number of species. Among them, functional diversity (FD, i.e. the extent of trait differences between species in a community; Díaz & Cabido 2001) can reveal why biodiversity changes in response to environmental change, as well as how biodiversity influences ecosystem functioning (Díaz & Cabido 2001). As a result, there is increasing interest in the study of FD and its dependence on environmental conditions and management (de Bello, Lepš & Sebastià 2006; Mayfield et al. 2010).

European pastures, particularly in Mediterranean environments (Sluiter & Jong 2006), are under a dual process of abandonment in the less productive areas and intensification in the more productive or accessible ones (Stoate et al. 2009). Grazing abandonment affects floristic composition, functional traits such as canopy height, leaf dry weight, onset of flowering, life-form and seed mass (Peco et al. 2005, 2012; Díaz et al. 2007), the overall FD of communities (de Bello, Lepš & Sebastià 2006) and many environmental aspects such as light quality and intensity and soil characteristics (Peco, Sánchez & Azcárate 2006). In contrast, intensification processes have been found to reduce FD, although these results depend on the studied site or taxonomic group (Flynn et al. 2009). Plant diversity in the Mediterranean areas of Europe is expected to be greatly impacted because of the predicted reduction in rainfall (Thuiller et al. 2005). Nevertheless, relatively few studies have related the response by FD to changes in land use (e.g. Flynn et al. 2009; Mayfield et al. 2010) or climate (Thuiller et al. 2006), and even fewer have assessed how the functional turnover within and between management types is affected by environmental or management conditions (de Bello et al. 2009). In particular, it is uncertain whether the changes in TD will be mirrored by changes in FD within and among communities.

There are a number of reasons to suspect that TD and FD do not always covary in response to land-use changes (Mayfield et al. 2010), thus characterizing their relationship provides insights into the consequences of different disturbance regimes or management strategies on community assembly and ecosystem functioning (Sasaki et al. 2009b). For example, assemblages that have functional redundancy contain species that have similar functional traits, and thus, the addition of these species does not increase FD (Petchey et al. 2007). On the contrary, the extinction of a functionally unique species can compromise the stability of a given community (Micheli & Halpern 2005). Thus, functional redundancy can be examined through the relationship between TD and FD – namely if TD increases and FD remains constant. Factors such as productivity (Sasaki et al. 2009b) or disturbance regime (Biswas & Mallik 2011) modulate this relationship because they can select for, or against, convergent types of traits (Grime 2006). For example, grazing can increase or reduce grassland diversity, depending on productivity, grazing history and intensity (de Bello, Lepš & Sebastià 2006; Bakker et al. 2006).

These factors affect diversity within but also between communities (Milchunas, Sala & Lauenroth 1988). Consequently, to understand its effects across habitats, it is necessary to partition the total diversity of the studied system (γ) into its α (within communities) and β (among communities) components (Whittaker 1972) and examine their temporal variability. Partitioning diversity also allows us to understand how different community assembly rules act simultaneously at different scales (de Bello et al. 2009). In this study, we study how environmental gradients affect community structure and examine how the mechanisms that control coexistence in plant communities are regulated by the productivity and disturbance levels. We also want to provide an insight into the potential effects of land-use change, combined with the foreseeable reduction in water availability, on Mediterranean grasslands at a landscape level. In particular, we focus on how habitats with different levels of productivity, linked mainly to differences in water availability, affect both TD and FD across a gradient of grazing pressure. We also study how these parameters are affected by interannual rainfall variability by comparing two years that differed notably in the amount of precipitation. Finally, we examine whether, and to what extent, water availability determines the spatial partitioning of both TD and FD in a landscape under several grazing intensities. Specifically, our hypotheses were as follows: (1) because of a stronger effect of habitat filtering, FD should be lower in dry than in wet conditions, (2) because grazing and aridity select for similar traits, differences in diversity among different levels of grazing should be higher in wet conditions than in dry ones and (3) γ diversity, as well as the proportion of diversity explained by the variability between grazing levels, should be higher in wet than in dry conditions.

Materials and methods

Study area

The study area (5 × 4 km) is 35 km North of Madrid, Spain (40°38′ N; 3°70′W; mean elevation 860 m). Climate is Mediterranean, with dry summers and maximum precipitation in spring and autumn. Annual rainfall is 540 mm, with large interannual fluctuations, and mean temperature is 13 °C. The landscape is characterized by gentle slopes (<5%), shallow acidic soils over a gneiss substratum and many rocky outcrops. The area is covered by dehesas, open woodlands (ca. 40 trees ha−1) of Quercus ilex subsp. ballota L. and Juniperus oxycedrus L, with a grassland understorey with a high proportion (ca. 70%) of annual species.

The area has been grazed extensively for centuries, but recent abandonment of grazing has created a mosaic of grazing pressure (Peco et al. 2005). We selected sites in four categories of increasing grazing intensity: (1) areas abandoned by ranchers and no longer grazed, (2) areas only grazed occasionally, with no permanent grazing, (3) areas with a permanent presence of livestock but away from points of livestock concentration (water points and points where animals are provided with feed) and (4) areas around points of livestock concentration, with permanent high levels of grazing. All the selected areas have maintained their current grazing status for at least 30 years. Within each of these levels, we identified two different habitat types: (1) wet habitats, at the bottom of slopes and depressions, with deep soils, water and nutrient inflows; and (2) dry habitats along the upper slopes, with shallow soils and nutrient and water outflows. Previous studies in the same area found differences in soil water availability, clay percentage, total nitrogen and soil organic matter between these two habitat types (Peco, Sánchez & Azcárate 2006). We selected 8–9 independent sites in each of the eight grazing by habitat levels for a total of 67 sites. The average distance between adjacent sites was 106 m (minimum of 45 m), and the slope was always under 5%. To estimate differences between soil water content between habitat types, in April 2012, a soil sample (cylinder of 5 cm and 98·17 cm3) from each site was collected and oven-dried. Its water content was defined as the ratio of the mass of water and the dry weight of the sample. As expected, there were great differences in soil water content between wet (33·04% ± 1·39; mean ± SE) and dry habitats (14·60% ± 1·09).

Vegetation and functional traits sampling

In each site, we placed three sampling quadrats (20 × 20 cm), always in the same relative positions (1 m N, E and W from the site centre). Species cover was estimated using six classes: (0) absent; (1) cover < 1%; (2) 1–12%; (3) 12–25%; (4) 25–50% and (5) >50%. After assigning to the species the median value of its cover class, we averaged their cover across the three quadrats in each site. Vegetation was monitored in exactly the same quadrats in the springs of a dry (2009, 50% of average spring precipitation) and a wet year (2010, 153%). A total of 177 species were found in the surveys.

We collected data for eight functional traits related to species strategies in response to disturbance and productivity for a sufficient number of species (Pakeman & Quested 2007). Canopy height (distance between the plant base and the highest photosynthetic leaf) was measured on 10 nongrazed mature individuals of each species, at least 25 m from each other, in the areas where the species was more abundant. Specific leaf area (SLA; mm2 mg−1) was measured in the same individuals, dividing the leaf area by its oven-dried mass. Seed mass and the presence of dispersal structures were obtained in most cases from Azcárate et al. (2002), and new measurements were taken for the species not included in that study, following the same protocol (30 dry seeds per species). Information on the remaining traits (Life-form, Growth form, Longevity, and Clonality) was taken from literature (Valdés, Talavera & Fernández-Galiano 1987).

Taxonomical and Functional diversities calculation

Quantitative traits were log-transformed and standardized to a 0–1 scale. For each trait, we calculated a matrix of distances between all possible species pairs and then tested the correlation between traits by performing Mantel tests between their dissimilarity matrices (not shown). For FD calculations, we selected four traits with a very low level of correlation (Canopy height, SLA, Seed mass and Clonality). These traits have been described as indicators of ecosystem functions, plant dispersal, establishment, persistence and response to grazing (Weiher et al. 1999; Díaz et al. 2007) and represent an extension of the LHS scheme (leaf, height, seed traits; Westoby 1998) with the inclusion of clonal trait information (Klimešová & de Bello 2009). With these traits, we computed a new matrix of species functional dissimilarities using Gower distance and used it to calculate FD.

We calculated TD and FD at different spatial scales to have a comparable spatial partitioning of diversity for both of them. FD for each sampling site was calculated on the basis of the Rao index of diversity (Lepš et al. 2006):

display math(eqn 1)

The parameter dij expresses the dissimilarity between each pair of coexisting species i and j and varies from 0 (two species with exactly the same traits) and 1 (two species with completely different traits). In the case of binary or categorical traits, when the species has the same trait value, then dij = 0; otherwise, dij = 1. FDRao is a generalization of the Simpson index of diversity (Lepš et al. 2006), that is, when all distances between species are equal to 1, FDRao is the Simpson index of diversity (i.e. 1−Simpson dominance, which was used to characterize TD). FDRao is among the few indices, if not the only one, that allow a comparable diversity partitioning of taxonomical and functional diversities with the same mathematical formulation, while taking species relative abundances into account (de Bello et al. 2010).

We computed α- (within sampling sites; expressed in eqn (eqn 1)), β- (between sites) and γ-diversity (within habitat type and grazing intensity) for both TD and FD. To calculate γ-diversity, we pooled samples of habitat type, grazing intensity and year. β-diversity for both TD and FD was specifically expressed in percentage over γ, as β = (γ-mean α)*100/γ (details in de Bello et al. 2010). According to this formulation, β-diversity summarizes the proportion of between-sites diversity with respect to the total diversity within a grazing intensity zone, habitat type and year. In another analysis, we added a new level and partitioned diversity by considering the diversity between grazing intensity zones (within year and habitat type). We thus had three levels for each year and habitat type: within sites, between sites (within grazing levels) and between grazing intensity levels. To calculate the abundance of species at different spatial scales, we applied the methods described by Villeger & Mouillot (2008) and de Bello et al. (2010). Finally, using the Rao index, we also calculated β-diversity between all possible pairs of the 67 sampling sites, for both TD and FD using the R function Rao (de Bello et al. 2010). This resulted in two dissimilarity matrices expressing the taxonomical and functional turnover between all possible pairs of sites. To partition α and β diversity correctly, we applied a correction used in the context of the Rao index (de Bello et al. 2010), which involves first calculating the diversity for single sites (α) according to eqn (eqn 1). For TD, α is expressed as 1 over dominance (where dominance is computed by inline image) and for FD, it is equal to 1/(1−FDRao).

Statistical analyses

We fitted repeated-measures anovas (type III sum of squares) to examine the relationships of α-TD and α-FD with grazing level (categorical with four levels), habitat type and year, using year as the repeated-measures factor. Whenever we found significant values for the interactions between the variables, we fitted separated models for each habitat type and year, using only grazing level as explanatory variable, followed by Tukey's HSD tests in the case of a significant effect of grazing.

We then checked whether grazing, habitat and year influenced the dependence of α-FD on α-TD. We fitted a repeated-measures ancova in which α-FD of the sites was used as response variable and the explanatory variables were α-TD, α-TD2 (to account for nonlinear relationships), grazing level, year (repeated-measures factor) and habitat type, as well as the interactions of α-TD with year and habitat type. In the case of a significant effect of the interactions, a different model for each year and habitat type was fitted using α-TD, α-TD2, grazing level and the interaction α-TD × grazing level. These models were simplified by selecting significant variables through backward stepwise regressions.

We then analysed the effect of grazing and habitat type on the taxonomical and functional turnover (β) between sites. We calculated the taxonomical and functional dissimilarities between each pairs of sites using the Rao index and analysed the resulting dissimilarity matrix with permanova (R package vegan; Oksanen et al. 2011) in which habitat type, grazing intensity and their interaction were used as explanatory variables. To visualize these analyses, we performed a Nonmetric Multidimensional Scaling (NMDS) for each year based on the pairwise sites dissimilarities for both TD and FD. The same distance matrices were used to calculate the distance of each site to the centre of its class centroid (permdist; Oksanen et al. 2011) and calculate the homogeneity of sites within each grazing intensity level and habitat type and year. We then analysed the PERMDIST results with repeated-measures anovas as for α-TD and α-FD (for each site, we considered as response variable its distance to the centroid within each grazing intensity level, habitat type and year). Finally, for each year and habitat type, we calculated and plotted the partitioning of TD and FD, as explained in the previous section. All analyses were performed using R (R Development Core Team 2011).

Results

The response of α-TD and α-FD to grazing was different between the different habitat types (significant grazing x habitat type interaction; Table 1a). For α-TD there was a stronger response to grazing in the more humid conditions (wet habitats, wet year; i.e. the anovas for each habitat type and year revealed significant differences between levels of grazing in wet conditions; Fig. 1a). In these conditions, α-TD increased with higher grazing pressure (Fig. 1a). Conversely, grazing affected α-FD negatively in the driest conditions (i.e. dry habitats, dry years), but it had no effect in the other scenarios. Interannual variability of α-TD was much higher in dry habitats, where α-TD exhibited a pronounced increase in the wet year compared with the dry one. In contrast, α-FD remained rather stable between years (Table 1a; Fig. 1a).

Figure 1.

Response to grazing pressure of α and β taxonomical and functional diversities for each year and habitat type. Error bars show mean ± SE. Different letters indicate significant differences between grazing levels (Tukey's test, < 0·05).

Table 1. Results of the repeated-measures anovas. Grazing level, habitat type and year as well as their interactions were used as explanatory variables and year as the repeated measure. α and β taxonomical and functional diversities were the response variables. Significant results (< 0·05) are in bold
 α-TDα-FDβ-TDβ-FD
d.f.FFFF
Between factors
Grazing level32·800·552·8210·33
Year10·750·150·510·84
Habitat10·1515·4317·8326·22
Grazing × Habitat34·505·063·042·41
Residuals58    
Within factors
Year146·220·9825·574·15
Grazing × Year30·171·021·044·48
Year × Habitat145·431·4219·540·70
Grazing × Year × Habitat32·431·732·830·29
Residuals58    

The relationship of α-FD with α-TD at the site level depended strongly on habitat type (α-TD x habitat type interaction = 0·004) and year (α-TD × year interaction = 0·005), and thus, we fitted models for each combination of habitat type and year. Neither grazing nor the interaction α-TD × grazing was significant in any of the scenarios, and was consequently excluded from the final models. During the dry year, the selected models revealed a nonlinear relationship between α-TD and α-FD indicating that beyond a given threshold, the addition of new species did not increase FD at the site level, but rather increased the functional similarity between species, especially in the dry habitat (Fig. 2a). In the wet year, the relationship between α-FD and α-TD was linear, either positive in the wet habitats or negative in the dry habitats (Fig. 2b).

Figure 2.

Relationship between α and taxonomical and functional diversities. In each scenario, the most parsimonious model for each year and humidity level is represented. Point size reflects grazing intensity (bigger points indicate higher grazing levels). Each plot shows the proportion of variance in FD explained by each of the explanatory variables (TD, taxonomical diversity; TD2, squared taxonomical diversity) and its significance level (*< 0·05; **< 0·01; ***< 0·001).

The NMDS and permanova analyses showed that habitat type was the main factor controlling both taxonomical and functional turnover between sites (Fig. 3). Regarding the taxonomical dissimilarity, we observed an important differentiation between dry and wet habitats in the 2 years, and a remarkable increase in the differences between sites in the dry habitats during the wet year compared with the dry one, when these sites were much more alike. This implies that differences between dry and wet habitats are reduced in a wet year. The effect of grazing was small compared with that of habitat type and differed between wet and dry habitats in both years, especially during the dry year. The significant interaction between habitat type and grazing observed for TD was also detected for FD in the dry year, when the sites of the wet and dry habitats were functionally more similar for lower than for higher levels of grazing. In contrast, during the wet year, the functional composition of both habitat types was similarly affected by grazing. Increased grazing also increased functional homogeneity, for both wet and dry habitats, but did not affect the taxonomical β diversity of wet sites (Table 1b). In contrast, the effect of grazing during the wet year was less important, and we only found a small, nonlinear effect of grazing on the β-TD of dry habitats (Fig 1b).

Figure 3.

Results of the nonmetric multidimensional scaling, showing the effects of grazing and humidity on the taxonomical and functional compositions in dry and wet years. Arrows show the effect of grazing for dry sites (continuous line) and wet sites (discontinuous line). Point size reflects grazing intensity (bigger points indicate higher grazing levels). Each plot shows the results of the permanova analysis (9999 permutations): proportion of explained variance by the explanatory variables (H, habitat; G, grazing; H*G, habitat × grazing) and its significance level (*< 0·05; **< 0·01; ***< 0·001).

The overall partition of TD revealed a remarkably higher total (γ) diversity in the dry habitat during the wet year (11·95) compared with the dry year (3·56; Fig. 4). TD was more evenly distributed between observation levels in wet than in dry habitats, where most of the taxonomical variability could be observed at the site level. Nevertheless, there was a shift in the proportion of the variability that was observed at each level in dry habitats: while in the dry year most of the TD in the dry habitats was observed within sites (78·7% over the total diversity within habitat and year), this proportion was smaller during the wet year (60·7%), mainly because of the increase in the variability within grazing zones (12·5% in the dry year, 28·6% in the wet one). In contrast, most of the observed variability in FD (around 93% for dry habitats and 88% for wet habitats) took place at the within-site scale, regardless of habitat type and year. The proportions of FD explained by each level were much more independent of conditions (habitat and year) than those of TD. In this way, it is remarkable that the huge increase in γ-TD in the dry habitat during the wet year was not accompanied by any increase in γ-FD in these areas, but rather a small decrease (1·66 vs. 1·61; Fig. 4).

Figure 4.

Partition of the effective number of species (taxonomical diversity) and extent of index dissimilarity within communities (functional diversity) at different levels of the sampling hierarchy. Note that the scale of the functional diversity plot is log-transformed to facilitate its visualization.

Discussion

Our main objective was to examine the combined effect of grazing and water availability on the diversity of Mediterranean grasslands, along with the processes that determine these effects. Water availability was the major determinant of both taxonomical and functional compositions, confirming results reported in many other studies (e.g. Cingolani et al. 2003; Sasaki et al. 2009a). In these semi-arid ecosystems, plants fitness is strongly affected by reduced water availability, making the relative importance of disturbance less evident (Kikvidze, Suzuki & Brooker 2011). However, the effect of grazing was significant in all the analysed scenarios, corroborating the importance of grazing in the diversity and composition of these systems at multiple spatial scales (Peco et al. 2005; Golodets, Kigel & Sternberg 2011).

Water availability also modulated the effect of grazing on TD and FD as well as the relationship between them. Thus, our results suggest that the effect of grazing on various diversity components was contingent on water availability. Increased water stress (dry year in dry habitats) resulted into a negative effect of grazing on α-TD, while this effect was positive when water availability was high (wet year in wet habitats), confirming that the response of α-TD to grazing varies along a gradient of water availability (Milchunas, Sala & Lauenroth 1988; de Bello, Lepš & Sebastià 2006; Bakker et al. 2006). This observation supports the idea that in more productive conditions, grazing reduces the relative importance of competitively dominant species and promotes species diversity through an increase in subordinate species, while in less productive conditions, grazing reduces species diversity by increasing plant mortality, for example excluding the most palatable species (Bakker & Olff 2003). This is consistent with predictions for areas with a long history of grazing (Milchunas, Sala & Lauenroth 1988; Cingolani, Noy-Meir & Díaz 2005) and shows that not only regional but also local differences in productivity within a climatic region can affect the relationship between TD and grazing (Sasaki et al. 2009a). Local differences in productivity can be particularly important in semi-arid systems like oak savannas, because small changes in productivity can change the direction of the effect of grazing on diversity (Bakker et al. 2006). However, our results contrast with previous studies in the same area, which found no differences in the species richness between grazed and ungrazed conditions (Peco, Sánchez & Azcárate 2006; Peco et al. 2012). The inclusion of more grazing levels in this study, especially high grazing pressure, may have caused these differing results. Climatic conditions not only affect species composition of these communities, but also modify the effect of grazing on TD, as shown in this study. Consequently, studies analysing the effects of grazing on plant diversity should take into account interannual differences in water availability.

Contrary to our expectations, and to the response of α-TD, the effect of grazing on α-FD was greater in dry than in wet conditions. Grazing reduced α-FD under conditions of increased water stress (dry habitats in dry year), contrasting with its lack of effect in wetter conditions (Fig. 1). Although grazing and aridity are often considered convergent selective forces (Milchunas, Sala & Lauenroth 1988; de Bello, Lepš & Sebastià 2006), this pattern is possibly not universal (e.g. plant height in Quiroga et al. 2010). Rather than supporting the role of disturbance as the principal factor creating and sustaining divergence in trait values (Grime 2006), our results agree with those of Pakeman, Lennon & Brooker (2011), considering disturbance (grazing in this case) as a filter that can cause trait convergence within plant communities and suggest that its strength as a filter for trait values depends on productivity. In particular, the combined effect of stress and disturbance might reduce the functional strategies available in a site.

Grazing intensity effects, while different for α-FD and α-TD, did not affect the slopes of the relationship between them in any of the studied cases, in contrast with the findings of Biswas & Mallik (2011). Contrary to the conclusions of studies that evaluated this relationship as positive and linear, regardless of environmental conditions (Micheli & Halpern 2005; Sasaki et al. 2009b), we found that both the shape and the strength of the relationship depended on water availability. While this relationship was positive and linear over most of its range in wet habitats, in dry habitats, it was negative and linear during the wet year and hump-shaped during the dry year. Our results highlight the importance of considering both FD and TD in the assessment of ecosystem responses to disturbance and stability (Mayfield et al. 2010). Nevertheless, care should be taken when comparing our results with other studies, because different FD indices can be influenced differently by species richness (Poos, Walker & Jackson 2009) and can have different ecological interpretations (Mouchet et al. 2010). Such comparisons are currently also limited by the small number of studies that consider both TD and FD (Cadotte, Carscadden & Mirotchnick 2011).

Negative and hump-shaped relationships between α-FD and α-TD have been previously reported in animal communities (e.g. Mason et al. 2008) and plant communities after changes in land use (Mayfield et al. 2010). Above a determined level of α-TD, we observed increasing levels of redundancy, that is, similarity in functional traits of the species that result in a weaker or nonpositive, relationship between α-FD and α-TD (Petchey et al. 2007). Below that level, reduced redundancy of the communities (Petchey et al. 2007) would explain the abrupt reduction in α-FD. This was especially evident in the driest conditions, in which communities under high grazing pressure presented a very strong relationship between α-FD and α-TD (Fig. 2a), resulting from the effect of grazing on the more palatable species (Díaz et al. 2007; Sasaki et al. 2009a). The nonlinear relationship was only observed during the dry year because great increases in α-TD in dry sites during the wet year masked the segment of the relationship dominated by reduced redundancy. This functional convergence is consistent with the notion of habitat filtering (Cornwell & Ackerly 2009; Pakeman, Lennon & Brooker 2011), indicating that only a subset of the possible trait values were found in the harshest conditions. Dry and wet habitats differed in the levels of α-TD at which α-FD peaked (higher in wet habitats), indicating higher functional redundancy in dry habitats because of a stronger role of habitat filtering. Nevertheless, we are aware of the potential confounding effects that the existence of a legacy effect of drought on the growth of perennial species could have on our results (de Vries et al. 2012).

Partition of diversity revealed important differences between TD and FD. As expected, the proportion of the total (γ) TD because of differences between grazing levels was remarkably higher in wet than in dry habitats. This was especially evident during the dry year, when the harsh conditions imposed a great reduction in γ-TD in the dry habitats in comparison with the wet ones and minimized the proportion of γ-TD because of differences between grazing levels and between sites at the same grazing level (Fig. 4). Small differences between grazing levels in these conditions indicated that aridity and grazing operate as convergent forces (Quiroga et al. 2010). However, grazing reduced taxonomical differences between sites within the same grazing level (Fig. 1b), which contradicts the former observation. This suggests that among the small number of species that are able to survive under severe water restrictions, most but not all of them can cope with high grazing pressures (Cingolani, Noy-Meir & Díaz 2005). On the other hand, both γ-FD and its spatial partition were less dependent on climatic conditions and did not vary among years, suggesting that FD is largely independent of TD in dry habitats because of the high functional redundancy. As expected, the proportion of FD explained by within-site variability was smaller in wet than in dry habitats, but this difference was smaller than for TD. Most functional variability was found to be at the site level, which is consistent with previous studies (Bello et al. 2009; Cornwell & Ackerly 2009) and with the hypothesis of limiting similarity as a mechanism promoting trait divergence within communities (Wilson 2007). Our results support the idea that habitat filtering and limiting similarity are not mutually exclusive, but mechanisms that can take place hierarchically in a community (de Bello et al. 2012).

Conclusions

The results indicate the importance of both functional differentiation and functional redundancy in species coexistence (Fukami et al. 2005) as well as their dependence on the combined effect of water availability and grazing regime. Despite great interannual variations in species diversity, functional diversity was often decoupled from species diversity and rather stable over time and space, showing the temporal functional stability of the system. However, under more limited water availability, grazing intensification reduced the functional diversity of these grasslands. The predicted reduction in rainfall in the Mediterranean area increases the risk of high grazing levels causing dramatic declines in the functional diversity of Mediterranean grasslands, probably compromising their stability and resilience over time. In the light of these results, we suggest that overgrazing could have serious consequences for the functionality of these ecosystems and therefore recommend the adoption of flexible grazing management schemes that should take species and functional diversities into account and adapt the level of grazing pressure to water availability.

Acknowledgements

The Spanish Ministries of Science and Education (Projects CGL2007-63382 and CGL2011-24871 and grants FPI BES-2008-009821 for CPC and Salvador de Madariaga PR2011-0491 for BP) and the Madrid Government (REMEDINAL-S0505/AMB-0335 and REMEDINAL2-S2009/AMB-1783) provided financial support. We thank Catherine Levassor for her fieldwork and expert knowledge, and Marc Cadotte and two anonymous reviewers for helpful comments on an earlier version of the manuscript.

Ancillary