Introduction

Vertebrate herbivores can have strong impacts on grassland plant species richness (Huntly 1991; Pacala and Crawley 1992; Collins et al. 1998; Ritchie and Olff 1999). However, the extent and course of their effects can vary widely depending on the variety of habitat involved and the plant and herbivore characteristics (Milchunas et al. 1988; Milchunas and Lauenroth 1993; Proulx and Mazumder 1998; Olff and Ritchie 1998; Frank 2005; Díaz et al. 2007; Symstad and Jonas 2011). Among these, primary productivity is particularly important, as it determines standing biomass and grazing characteristics (e.g., grazing pressure and regime, herbivore types), modulates plant interactions, and is linked to community composition and structure (Waide et al. 1999; Grace and Jutila 1999). It is widely accepted that large herbivores can increase species richness in habitats of high productivity by preventing competition for light by tall dominant species, whereas they do not affect or reduce species richness in low-productivity habitats (Milchunas and Lauenroth 1993; Huisman and Olff 1998; Olff and Ritchie 1998; Proulx and Mazumder 1998).

The impact of herbivores on grassland plant species richness may also be a function of herbivore size (Olff and Ritchie 1998; Olofsson et al. 2004; Bakker et al. 2006; Denyer et al. 2010). Body size differences may affect the abilities of herbivores to use plants that differ in fiber content and other secondary metabolites, thus determining herbivore foraging selectivity and food quality requirements (Demment and Van Soest 1985). In particular, large herbivores are usually less selective grazers and can increase plant species richness when they reduce the abundance of dominant plants (Milchunas and Lauenroth 1993). In contrast, smaller mammal herbivores with more specialized diets can select more nutritious seedlings and less abundant plants or plant parts (Hulme 1994; Edwards and Crawley 1999; Milchunas 2011), and may reduce plant species richness.

A number of studies have highlighted the importance of the interaction between herbivory and productivity on the impact of vertebrate herbivores on plant diversity, although most of them have focused on large herbivores (Milchunas and Lauenroth 1993; Osem et al. 2002; Frank 2005). However, most natural and semi-natural grasslands worldwide are grazed by assemblages of herbivores with a wide range of body sizes. Although it is often assumed that large herbivores dominate the dynamics of plant communities, strong effects of small mammal herbivory have also been observed; for example, in Africa savanna plant communities, where the abundance of small herbivores depends on that of large ungulates (Keesing 2000). Furthermore, wild small mammals are usually present in managed grassland ecosystems or in nature reserves where livestock is used to maintain species-rich vegetation (Pakeman et al. 2003). Understanding the combined influences of site productivity and herbivore body size on the impact of herbivores on plant species richness is of fundamental and practical importance for the management of grassland ecosystems, as this information can identify the ecosystems in which different-sized herbivores may play a key role in biodiversity. Following this line, Bakker et al. (2006) studied the combined effect of different-sized herbivores grazing across a gradient of primary productivity at seven perennial-dominated grassland sites in North America and Europe. They found that assemblages including large herbivores increased plant diversity at high productivities but decreased diversity at low productivities, whereas small mammals did not have consistent effects along the productivity gradient. This seems to be similar—in terms of the impact of only large herbivores—to the results of Osem et al. (2002), who studied the impacts of sheep grazing and variation in primary productivity on the diversity of a Mediterranean annual plant community, and to the predictions of Milchunas et al.’s (1988) model. However, we still lack a clear understanding of how grasslands dominated by annual species respond to grazing by small mammalian herbivores and to grazing by herbivore assemblages of differing body sizes. This matters, since a large proportion of the world’s grasslands are located in semiarid regions where annual species can be an important component of the vegetation. Besides, we expect annual-and perennial-dominated plant communities to respond differently to grazing by small herbivores. According to Osem et al. (2002), annual grasslands—in contrast to perennial ones—lack temporal continuity in their competitive interactions, while at the same time the window of relaxed competitive interactions after disturbance may depend more on seedbank dynamics and seedling establishment than regrowth capabilities. Hence, it can be hypothesized that small mammalian herbivores with highly selective feeding habits have a greater impact on plant species richness in arid and semiarid communities dominated by annuals where seedlings and species dependent on reproduction by seeds are more abundant (Milchunas 2011). Significant impacts on plant community structure have been found in studies of interactions between granivore rodents and annual plant communities (Brown and Heske 1990; Maron and Simms 2001; Valone and Schutzenhofer 2007), but the response of these communities to grazing by widespread selective small folivores, such as lagomorphs, has received much less attention. The response to rabbits and other small mammalian herbivores is difficult to predict due to uncertainty about the effect of selective grazing on the dynamics of the seedbank. The expected negative effects of small herbivores on richness could instead be neutral in less-productive habitats, due to greater control over annual plant dynamics and germination by abiotic factors (e.g., the amount and timing of annual precipitation; Espigares and Peco 1993; Pérez-Camacho et al. 2012) compared to biotic ones. In contrast, the expected negative effects could be positive in more-productive habitats (with denser herbaceous plant cover) if grazing prevents competitive exclusion of some plants by tall dominant species or litter accumulation. Further, a large proportion of Mediterranean Basin grasslands dominated by annuals are also characterized by a long history of grazing by large herbivores and disturbance (Noy-Meir and Seligman 1979; Le Houérou 1981). This has likely promoted the existence of many plant species that are well adapted to grazing (Milchunas et al. 1988; Perevolotsky and Seligman 1998), and this should be taken into account when assessing plant responses to grazing.

In this paper, we ask if the interaction between plant richness, primary productivity and mammal herbivore body size that is known to exist in perennial grasslands also occurs in annual grasslands, using Iberian semiarid grasslands as a model system. Since annual-dominated grasslands of the Mediterranean Basin are among the most biologically diverse ecosystems of Europe (Verdú et al. 2000), they offer an excellent opportunity to study this question, as their biodiversity may be at risk if productivity or herbivore community composition changes. In these semiarid ecosystems, primary productivity is usually constrained by the seasonality of soil resources, primarily water and nitrogen (Seligman and van Keulen 1989). Further, interannual rainfall variability and the slope valley systems typical of this Mediterranean region often result in a highly patchy distribution of these limiting resources (Noy-Meir 1973; Peco et al. 1998; Osem et al. 2002), causing wide spatiotemporal variations in primary productivity.

We investigated the effects of large and small mammal grazing for six years in annual plant communities with different primary productivities in the central Iberian Peninsula. The topology of the study area allowed us to study sites with different primary productivities—uplands (low productivity) and lowlands (high productivity)—under the same semiarid climate regime. We selected two herbivores of contrasting size: European rabbits (Oryctolagus cuniculus) and free-ranging sheep (Ovis aries). Sheep are ruminants and have been classified as generalist herbivores (Schwartz and Ellis 1981), whereas rabbits are hindgut fermenters and have been classified as selective grazers (Bhadresa 1977).

In the Iberian Peninsula, traditional livestock management has influenced pastures for millennia (Le Houérou 1981). Further, these plant communities have co-evolved with the wild European rabbit, given that this species is native to the peninsula and can reach high local densities (Thompson and King 1994; López-Martínez 2008). To get a sense of the history of the rabbit–vegetation relationship in these environments, it is believed that the species Oryctolagus laynensis, considered the direct ancestor of the modern rabbit, appeared in the middle Pliocene (around 3.5 million years ago) in central and southern Iberia, associated with savanna-like ecosystems (López-Martínez 2008). High densities of wild rabbits were also described by Greeks and Romans several millennia ago (García-Bellido 1945).

We expect that in this semiarid region, changes in plant richness in response to herbivory depend on the combined effects of grazing by different-sized mammal herbivores and spatial (topography) and temporal (year) variations in productivity. We also expect that large generalist herbivores have a positive effect on species richness at high-productivity sites (lowlands) and a negative or null effect at low-productivity sites (uplands), whereas small selective herbivores have a negative effect on species richness at both high- and low-productivity sites. Herbivore-mediated control of plant communities may be exerted directly through the removal of plant biomass, or indirectly through the modification of competitive interactions among plants and litter accumulation (Huntly 1991). We examined changes in other plant community characteristics such as evenness and species turnover as well as green and litter cover and plant height to better understand the effects of the herbivores. Our experimental design allowed mixed rabbit and sheep grazing, rabbit grazing only, or no grazing, which were replicated five times within the two sites with different productivities. Species richness was measured during a period with highly variable precipitation: from 2002 to 2007; 2005 was the driest year in the last fifty years.

Materials and methods

Study area

The study was carried out in a 330 ha “dehesa” located in central Spain (40°23′N, 4°12′W). Mean altitude is 690 m a.s.l. The climate is semi-arid continental Mediterranean, with a mean annual temperature of 12.6 °C and a mean annual rainfall of 432.6 mm (1932–1980), no rain in summer, and wide interannual variations in rainfall. Annual precipitation and precipitation during the growing season period (Oct–May) varied during the course of the six sampling years (from 2002 to 2007) (Table 1). The substrate is sandy to sandy loam and lies upon a fractured bedrock of granite. The vegetation is typical of dehesa: evergreen oak tress (Quercus ilex) scattered in an open, savannah-like pasture. The herbaceous layer is very diverse (143 species found during the study), and mainly consists of winter annual species. The study area is characterized by a gentle undulating topography promoting water and soil fertility gradients from uplands to lowlands that lead to marked differences in primary productivity (see Table 2). Consequently, herbaceous vegetation can be divided into low-productivity pastures composed mainly of short and sparse annual plants on uplands, and higher-productivity pastures composed mainly of taller annual plants with dense canopies and some perennial species (e.g., Agrostis castellana and Cynodon dactylon) on lowlands. The dehesa is grazed primarily by a dense population of wild European rabbits (Oryctolagus cuniculus) (about 8.3 warrens/ha) and a flock of 600 free-ranging sheep (about 2 sheep/ha) from October to June; no other mammal herbivores are responsible for a significant amount of grazing (personal observation). The dehesa is maintained through traditional management practices, with periodic ploughing of the slopes to eliminate shrubs and improve pasture, summer mowing of the most productive pastures in favorable years, and small game hunting.

Table 1 Total annual and growing-season precipitation for the six sampling years
Full size table
Table 2 Plant and soil variables (mean ± SE) in uplands and lowlands in 2002, and statistical comparisons of the sites
Full size table

Experimental design and sampling

In August 2001, we established five replicate blocks with three herbivore body size treatments at both upland and lowland areas (30 plots in total). The treatments were: (1) unfenced plots that included all herbivores, i.e., the control condition (L + S = large plus small herbivores); (2) plots fenced to exclude large herbivores (sheep) but allow access to smaller species (rabbits) (S = small herbivores); and (3) plots fenced to exclude sheep and rabbits (U = ungrazed). All plots measured 36 m2 (6 × 6 m). Fenced plots were constructed with a 1 m tall chicken mesh (mesh size of 2.5 cm). The mesh in plots which excluded large herbivores was lifted 20 cm above ground level to allow rabbit access but not sheep. The mesh in plots which excluded sheep and rabbits was buried 30 cm into the soil, forming an “L” shape to prevent rabbits burrowing underneath it. The minimum distance between upland and lowland plots was 900 m, and within each block the three grazing treatment plots were adjacent. Data on green canopy cover, litter cover, plant height, species richness, and species cover were gathered in each plot at peak standing crop in April for uplands and May–June for lowlands for six years (period 2002–2007), in seven 20 cm × 20 cm quadrats randomly placed each year inside the plots. Sampling was prevented when prior efforts had already been undertaken. This quadrat size is commonly used to study Mediterranean herbaceous communities rich in species (Montalvo et al. 1993; Osem et al. 2002). Species richness was calculated as the number of species per quadrat. Species cover was estimated as canopy cover, i.e., the percentage of aerial visual surface area covered by each plant species. Likewise, green cover was visually estimated as the percentage of the herbaceous layer that was green, whereas litter cover was visually estimated as the percentage of litter (dead vegetation from the previous year). Plant height was measured as the maximum height of the vegetation (excluding flowers and seed heads).

The cover of each plant species (a measure of abundance) was used to calculate several indices related to plant community. Species evenness, a diversity index which quantifies how equal the species are numerically, was calculated as Pielou’s (1966) evenness index (J′ = H′/H′max), where H′ is the Shannon–Wiener diversity index. This evenness index ranges between 0 and 1, where 1 indicates that all species are equally abundant in the community (Magurran 1988). Shifts in plant community composition under grazing were expressed using the Sorensen (Bray–Curtis) index of dissimilarity (Bray and Curtis 1957), which is based on species abundance data. The Sorensen index is defined as D = 1 – 2W/(A + B), where A and B are the abundances of species in samples A and B, and W is the lowest abundance of species shared by the two samples. When species composition is exactly the same and every species is present in the same proportion, the index is 0, whereas at the maximum difference the index is 1. To compare patterns of species shifts under different-sized herbivore grazing, we calculated dissimilarity values among quadrats for the different herbivore body-size treatments within each topographical position (upland/lowland).

Data analysis

To analyze the impact of different-sized mammal herbivores on the response variables, we combined fixed effects with nested levels of random effects in general linear mixed-effects models (LME). This enables the analysis of repeated measures and spatially nested data without succumbing to the problems of nonindependence and pseudoreplication (see Crawley 2007; Zuur et al. 2009). Note that in LME models, fixed effects influence only the mean of the dependent variable, whereas random effects influence the variance–covariance structure of the response variable (Pinheiro and Bates 2000). Sampling year (from 2002 to 2007), site (upland and lowland), and herbivore grazing treatment (L + S, S and U) were modeled as fixed effects, and we included all interaction terms. Quadrats were nested within each block and treated as a random effect. To simplify the assessment of whether herbivores altered species richness and the other response variables, we analyzed the data independently for uplands and lowland sites and evaluated the effects of different herbivore assemblages on the response variables. In this case, year and herbivore treatment were used as fixed factors and quadrats nested within blocks as a random factor. After initial model fitting, the distribution of residuals and homogeneity of variance were tested using the Shapiro–Wilk goodness of fit test and Levene’s test. In general, species richness data followed normality, but the other variables required square root- or log-transformation to meet normality requirements. All statistical tests were implemented in R (version 2.11.1; R Development Core Team 2010) using the nlme package (Pinheiro et al. 2009).

Results

The impact of herbivores on the response variables depended on the interaction between grassland primary productivity and herbivore assemblage, and was strongly affected by sampling year, as shown by a significant interaction of these three factors (Table 3). All variables were also affected by sampling year and, except for species evenness, by herbivore treatment and productivity site as independent factors.

Table 3 Results of linear mixed models (LME) of factors affecting plant species richness, green canopy cover, litter cover, plant height, species evenness, and dissimilarity in plant species composition
Full size table

Species richness

Species richness ranged between 6 and 20 species per quadrat. In the control plots (grazed by both rabbits and sheep), species richness was higher in lowlands than in uplands in five of six years (except in 2005) (F 1, 404 = 61.9, P < 0.0001) (Fig. 1a). In the low-productivity uplands, species richness increased in the S and U treatments [i.e., when sheep were not in the assemblage (F 2, 608 = 11.6, P < 0.001)], although it was only significant in two years: 2004 and 2007 (P < 0.05). At high-productivity lowlands, species richness decreased in those plots that were not grazed (U) or were only grazed by rabbits (S) (F 2, 608 = 83.4, P < 0.001) (Fig. 1b). In 2003, species richness was higher in plots grazed by rabbits alone (S) as compared to ungrazed plots (U) (P < 0.05).

Fig. 1
figure 1

Variations in a, c species richness (number of species per 20 × 20 cm quadrats) and b, d evenness (based on species cover in the 20 × 20 quadrats) in the upland and lowland sites in response to three herbivore treatments: L + S = sheep plus rabbits, S = rabbits alone, U = ungrazed. Values are mean ± SE (n = 5 per treatment, site productivity, and year)

Full size image

Herbaceous green cover, litter cover, and plant height

In uplands, green canopy cover was always higher in the ungrazed plots (U), but also increased significantly in 2004, 2006, and 2007 in plots grazed only by rabbits (S) (F 2, 608 = 60.3, P < 0.001) (Fig. 2a). In contrast, in lowlands, green cover decreased when sheep were excluded (S), and even more when both herbivores were removed (U) (F 2, 608 = 168.1, P < 0.001) (Fig. 2b). Note that in lowlands, green cover remained relatively stable and close to 100 % in plots grazed by both herbivores.

Fig. 2
figure 2

Mean cover of living herbaceous vegetation (a, b), litter cover (c, d), and plant height (e, f). Presentation as in Fig. 1

Full size image

In uplands, grazing by rabbits alone (S) did not affect litter cover, whereas protection from both herbivores (U) promoted an increase in litter (F 2, 608 = 69.1, P < 0.001) (Fig. 2c). Litter cover accumulation was always greater in lowlands, where it ranged between 0 and 67 % depending on the year and treatment (Fig. 2d). At these high-productivity sites, protection from sheep (S) significantly increased litter cover (F 2, 608 = 78.0, P < 0.001). In some years, this increase was stronger when both herbivores were removed (U) (Fig. 2d). In both uplands and lowlands, we observed a peak in the increase of litter in 2005 as a result of the intense drought. This was generally parallel to a decrease in green cover and plant heights in all treatments.

In uplands, plant height increased significantly only when both herbivores were excluded (U) (F 2, 608 = 268.5, P < 0.001) (Fig. 2e). In lowlands, grazing protection from sheep (S) or both herbivores (U) increased plant height compared to that seen in plots grazed by both herbivores (L + S) (F 2, 608 = 110.8, P < 0.001) (Fig. 2f).

Evenness

The response of species evenness to herbivore body size treatments was complex in both sites (Fig. 1c, d). In uplands, species evenness was higher (i.e., a decrease in plant dominance) when rabbits grazed alone (S) (F 2, 608 = 8.2, P < 0.001), although it was only significant in the last two years (2006 and 2007). When both herbivores were excluded (U), species evenness showed an unclear pattern that varied with sampling year. In contrast, rabbit grazing (S) in lowlands decreased species evenness (F 2, 608 = 14.1, P < 0.001). When both herbivores where removed (U), there was an unclear pattern that again varied with sampling year.

Dissimilarity in species composition

Analysis of vegetation dissimilarity based on Sorensen’s quantitative index showed significant changes in plant species composition with grazing in both upland (F 2, 608 = 124.4, P < 0.001) and lowland (F 2, 608 = 25.09, P < 0.001) plots. In the uplands, a comparison of plots grazed by both herbivores and plots grazed by rabbits alone yielded a low and relatively constant dissimilarity index that significantly increased from 2002 to 2003, decreasing slightly afterwards (Fig. 3a, see “L + S vs. S,” black symbols). This indicates that, 2002–2003 excluded, plant community composition changed little between plots grazed only by rabbits and plots grazed by both herbivores (i.e., sheep exclusion had little effect on species composition). Further, a comparison of plots grazed by both herbivores or by rabbits alone against ungrazed plots indicated higher values of dissimilarity (Fig. 3a, see “L + S vs. U,” white symbols and “S vs. U,” gray symbols), with a tendency to increase with year for “L + S vs. U.” In lowlands, a comparison of plots grazed by both herbivores and plots grazed by rabbits alone showed a significant increase in dissimilarity in species composition with time, mainly from 2003 to 2004 and at the end of the experiment (2006–2007) (Fig. 3b, see “L + S vs. S,” black symbols); i.e., sheep exclusion caused a change in species composition. In general, dissimilarity in species composition between the three treatments increased with time.

Fig. 3
figure 3

Dissimilarity in plant species composition (Sorensen index) between grazed and ungrazed subplots (L + S vs. U and S vs. U) and between sheep plus rabbits and rabbit-grazed subplots (L + S vs. S) in the uplands and lowlands. Bars denote SEs

Full size image

Taken together, these results suggest that in both uplands and lowlands, herbivore exclusion causes large changes in the species composition of the plant community. However, in uplands, excluding rabbits has the strongest effect, whereas excluding both herbivores has the strongest effect in lowlands.

Discussion

We found that the impact of herbivores on plant species richness depended on habitat primary productivity and year. Herbivores barely affected species richness at low-productivity sites (uplands), but when they did—in only two of the sampled years (2004 and 2007)—their impact was negative. In contrast, herbivores had a large positive effect on species richness at high-productivity sites (lowlands). In both cases, we only detected noteworthy effects when sheep (the larger herbivore) was included in the assemblage.

This reversal of grazing impact on plant species richness observed among the two productivity sites is consistent with expectations for low-productivity semiarid and high-productivity subhumid grasslands dominated by perennials (see Milchunas et al. 1988 for a global large-scale model) or annuals (see Osem et al. 2002 for a topographic small-scale model), with a long history of grazing by large generalist herbivores. The starting premise for both models is that plant–soil interactions in dry grasslands are fundamentally different from plant–soil interactions in subhumid grasslands. At lower productivity, plant growth is primarily limited by soil resources (water and minerals), whereas at higher productivity competition between plants is primarily for canopy resources (light) (Tilman 1982, 1988). In the low-productivity grasslands, the strong limitation of soil water leads to the complete occupation of the soil by roots but insufficient resources to support continuous aboveground plant cover. In contrast, in the high-productivity grasslands, in which soil resources are not a limiting factor for plant growth, plant cover is relatively continuous, and the major force in plant–soil interactions is related to feedbacks among plant biomass production, litter quality, and nutrient availability (see Burke et al. 1998). Therefore, since canopy development and aboveground-to-belowground ratios increase with increasing productivity, canopy removal by large herbivores is expected to have a greater effect in high-productivity than in low-productivity environments (Milchunas and Lauenroth 1993; Osem et al. 2002). Also, adaptations by plants to frequent loss of organs from drought in semiarid conditions (low-productivity sites in our study) can be similar to those needed to resist or tolerate herbivory (Milchunas et al. 1988). Hence, under these semiarid conditions, canopy structure, species diversity, and compositional changes in response to grazing are predicted to be minimal, since a long grazing history over evolutionary time results in greater capacities for regrowth following herbivory, and favors prostrate growth forms, small leaves, basal meristems, and annual life cycles—traits useful for minimizing the impact of the loss of plant organs (Milchunas et al. 1988; Milchunas and Lauenroth 1993).

Low-productivity sites (uplands)

Based on the above predictions, we suggest that plant communities of low-productivity sites (uplands) have responded to the convergent selection pressures of semiaridity and herbivory, and consequently species richness remains largely unaffected by grazing (Fig. 1a). It is possible that the lack of grazing effects in these low-productivity areas could be also attributed to a lower herbivore grazing pressure at uplands compared to lowlands. However, a second study in the same area in 2002 found that grazing was more intense in the uplands: 47 % of the aboveground biomass was consumed at uplands versus 34 % at lowlands (Rueda et al. 2010). Interestingly, in our study, the effect of herbivores on green canopy cover and especially on vegetation height in uplands (Fig. 2a, e) did not impact on species richness. We assume that this response is mediated by the predominance of plant competition for soil resources and mediated by the characteristics of the semiarid Mediterranean climate. The intense summer drought and the typically sparse and variable precipitation plus a long history of grazing have likely favored an annual plant strategy over evolutionary time that, unlike perennials, is mainly focused on giving seeds quickly rather than in a rapid regrowth following defoliation that is probably impeded by limiting soil resources. Hence, changes in canopy cover or plant height are not expected to cause changes in species richness, because these changes do not modify the competitive relations between plants.

Despite the lack of effects on species richness, we found large changes in species composition with grazing in uplands, which seems to be primarily due to rabbits (Fig. 3, uplands). Other cases in the literature based on large herbivores in semiarid plant communities report marked changes in plant composition in response to grazing, although species richness barely changes (Noy-Meir 1990; Milchunas and Lauenroth 1993; Jauffret and Lavorel 2003). Generally, grazing might cause greater changes in plant species composition than in richness when herbivores create a balance between rates of local colonization and extinction, so that the numbers of species lost and gained are equal. This might occur when vertebrate herbivores induce extinctions by the selective consumption of palatable species, but change resource availability to favor different functional groups of plants, such as grasses or forbs (Anderson et al. 2007). In uplands we observed, for example, that cover of Ornitopus compressus, Trifolium arvense, and Lupinus hispanicus, three typical legume species of semiarid Mediterranean grasslands of the central Iberian Peninsula, virtually disappeared from the rabbit-grazed plots while increasing in the ungrazed plots (M. Rueda, unpublished data). This implies that not all plants that evolved in low-productivity environments with a long history of grazing are necessarily resistant to grazing. Concerning this, Cingolani et al. (2005) proposed modifications of the original generalized model of Milchunas et al. According to these authors, the selective pressure of wild and domestic herbivores on systems with a long history of grazing has fluctuated over time due to epidemics, migration, weather, and wars, allowing the development of two pools of plant species adapted to low or high grazing intensities in semiarid or infertile systems: (1) a grazing-resistant pool that increases in periods of high grazing intensity but competes less effectively for soil resources, and (2) a less grazing-resistant pool, better adapted to soil resource capture (water or minerals), that increases in periods of lower grazing intensity. As grazing intensity increases, decreases in the latter set of species may be compensated by increases in the former, so that richness remains almost constant, as predicted by Milchunas et al. (1988), even if species composition changes considerably. It is possible, therefore, that two species pools coexist in our system, so that when grazing intensity decreases, some palatable species like legumes—also known by their ability to capture soil nutrients—increase, changing plant composition but not richness. These short-term changes in species composition are likely possible due to the seedbanks of some species in Mediterranean annual-dominated grasslands (Peco et al. 1998), which can persist in grazed areas as well as in natural refuges protected from grazing by rock outcrops and shrubs (Osem et al. 2002).

As a general rule, these arguments have been used to explain the lack of effects on plant species richness of large generalist herbivores in semiarid environments (Milchunas et al. 1988; Osem et al. 2002). However, given our results, we propose that the herbivory–semiaridity co-evolution in our system was probably not only promoted by wild or domestic large generalists, but mainly by the small European wild rabbit. We justify our logic by noting two aspects. First, species richness remained high and unchanged over time in the plots grazed only by rabbits. This was not due to low levels of rabbit grazing; they consumed on average 37 % of the aboveground biomass versus 47 % that was consumed by rabbits plus sheep (Rueda et al. 2010). Second, a long history of grazing by rabbits could explain why this small mammal has no negative effect on species richness in these Mediterranean environments, whereas it adversely affects the richness in other semiarid environments outside of the wild rabbit’s native range (Cooke et al. 2010).

High-productivity sites (lowlands)

In high-productivity lowlands, species richness was higher under herbivore grazing, whereas richness decreased when sheep were excluded (Fig. 1b). Higher richness with grazing intensity could be partly a response to larger soil resources, which in this case allowed plant regrowth (see Burke et al. 1998)—note that despite the herbivore-driven reduction of vegetation height, green canopy cover was always close to 100 % (except in 2005, the driest year) in the plots grazed by both herbivores. Sheep may also increase plant species richness under productive conditions when they alleviate light competition through biomass removal and control of the tall dominant plant species (Huisman and Olff 1998; Huisman et al. 1999). Simultaneously, an increase in light availability can enhance germination rates and seedling survival (Jutila and Grace 2002). Accordingly, we found that species dominance increased when sheep were excluded (Fig. 3, lowlands). However, this was only significant in a few years despite the large differences in species composition found between grazed and ungrazed plots, suggesting the existence of alternative mechanisms controlling species richness. Although resource competition is undoubtedly important in limiting the diversity of many high-productivity plant communities, litter accumulation likely plays an important role in our high-productivity sites. Across the whole exclosure, the increase in litter cover paralleled the decline in green cover and species richness. Litter accumulation can influence all stages of recruitment of plants by changing the physical, chemical, and biological microenvironments of seed and seedlings (Carson and Peterson 1990; Facelli and Pickett 1991). In Mediterranean annual communities, litter accumulation reduces the total number of emerged seedlings and seedling survival (Rebollo et al. 2001), reducing green cover. Furthermore, these negative effects are inversely related to seed mass, with litter accumulation being more limiting to small-seeded species (Gulmon 1992; Rebollo et al. 2001), which would explain decreases in species richness.

We found no effects of rabbits on plant species richness at high-productivity sites, although increases in the dissimilarity values when comparing rabbit grazing alone and ungrazed plots indicate that they promote changes in species composition in the more productive areas. We found that in high-productivity lowlands, rabbits did not visit the small-herbivore grazing plots as frequently as expected. In fact, biomass consumption by rabbits in plots exposed only to small herbivores dropped from 37 % in uplands to 16 % in lowlands (Rueda et al. 2010). This could be related to the fact that higher soil moisture at lowlands limits rabbit burrowing. As a central-place forager, the scarcity of burrows constrains rabbit access to the high-productivity sites (Rueda et al. 2008). Furthermore, rabbits usually show a strong preference for shorter swards, where they can spend more time foraging and less time scanning for predators (Iason et al. 2002). We have evidence (dropping counts, see also Rueda et al. 2008) that except in spring, when the vegetation is tallest, rabbits used the large plus small herbivore plots in lowlands, suggesting that sheep grazing/trampling might facilitate rabbit access to these plots, as has been demonstrated in other studies (Bakker et al. 2009; Denyer et al. 2010).

In sum, our results are consistent with those found in perennial-dominated grasslands. However, at low-productivity sites, there were virtually no effects on plant species richness in the assemblage including the large herbivore, which is in accord with a long evolutionary history of grazing and semiarid conditions. In this respect, we would like to draw attention to the potential role—not considered historically—that abundant and native small herbivores may have played in the evolutionary history of semiarid grasslands. Although rabbits did not affect plant species richness at the plot scale, large changes in species composition promoted by this herbivore suggest that rabbits can increase plant heterogeneity at a larger scales (Gálvez-Bravo et al. 2011). Thus, we conclude that the conservation of different-sized herbivores and the maintenance of traditional pastoralism, in the absence of wild large herbivores, are likely crucial to preserving high plant diversity in semiarid Mediterranean pastures.