Volume 104, Issue 3 p. 864-875
Standard Paper
Free Access

Different-sized grazers have distinctive effects on plant functional composition of an African savannah

Fons van der Plas

Corresponding Author

Fons van der Plas

Conservation Ecology, Groningen Institute for Evolutionary Life Sciences, 428 University of Groningen, PO Box 11103, 9700 CC Groningen, The Netherlands

Institute of Plant Sciences, Plant Ecology Group, University of Bern, Altenbergrain 21, CH-3013 Bern, Switzerland

Correspondence author. E-mail: [email protected]Search for more papers by this author
Ruth A. Howison

Ruth A. Howison

Conservation Ecology, Groningen Institute for Evolutionary Life Sciences, 428 University of Groningen, PO Box 11103, 9700 CC Groningen, The Netherlands

Search for more papers by this author
Nokukhanya Mpanza

Nokukhanya Mpanza

Conservation Ecology, Groningen Institute for Evolutionary Life Sciences, 428 University of Groningen, PO Box 11103, 9700 CC Groningen, The Netherlands

Search for more papers by this author
Joris P. G. M. Cromsigt

Joris P. G. M. Cromsigt

Department of Wildlife, Fish, and Environmental Studies, Swedish University of Agricultural Sciences, Skogsmarksgränd, 901 83 Umeå, Sweden

Centre for African Conservation Ecology, Department of Zoology, Nelson Mandela Metropolitan University, PO Box 77000, 6031 Port Elizabeth, South Africa

Search for more papers by this author
Han Olff

Han Olff

Conservation Ecology, Groningen Institute for Evolutionary Life Sciences, 428 University of Groningen, PO Box 11103, 9700 CC Groningen, The Netherlands

Search for more papers by this author
First published: 29 January 2016
Citations: 28

Summary

  1. Grazing ungulates play a key role in many ecosystems worldwide and can form diverse assemblages, such as in African savannahs. In many of these ecosystems, present-day ungulate communities are impoverished subsets of once diverse assemblages. While we know that excluding all ungulates from grasslands can exert major effects on both the structure and composition of the vegetation, how different individual ungulate species may have contrasting effects on grassland communities remains poorly understood.
  2. Here, we performed a long-term ‘Russian doll’ grazing exclosure experiment in an African savannah to test for the effects of different size classes of grazers on grassland structure and composition. At five sites, grazer species of decreasing size class (ranging from white rhino to scrub hare) were excluded using four fence types, to experimentally create different realized grazer assemblages. The vegetation structure and the grass functional community composition were characterized in 6 different years over a 10-year period. Additionally, animal footprints were counted to quantify the abundance of different ungulate species in each treatment.
  3. We found that while vegetation height was mostly driven by total grazing pressure of all species together, ungulate community composition best explained the functional community composition of grasses. In the short term, smaller ungulate species (‘mesoherbivores’) had strongest effects on vegetation composition, by shifting communities towards dominance by species with low specific leaf area and low nutritional value. In the long term, large grazers had stronger but similar effects on the functional composition of the system. Surprisingly, the largest ‘mega-herbivore’, the white rhinoceros, did not have strong effects on the vegetation structure or composition.
  4. Synthesis. Our results support the idea that different size classes of grazers have varying effects on the functional composition of grassland plant communities. Therefore, the worldwide decline in the diversity of ungulate species is expected to have (had) major impacts on community composition and functioning of grassland ecosystems, even if total grazing pressure has remained constant, for example, due to replacement by livestock.

Introduction

Grasslands cover a major part of the terrestrial world and ungulates play a central role in their structure and functioning (Hobbs 1996). Many grasslands historically harboured highly diverse communities of ungulates (Olff, Ritchie & Prins 2002), but are now dominated by only a few species due to extinctions and land use change (Owen-Smith 1989; Lorenzen et al. 2011). Such changes in ungulate communities are ongoing, reflected by the 63 ungulate species that are red listed (IUCN 2013) and by the transformation of many natural grasslands to pastures (FAO 2008) dominated by a one or a few livestock species. This worldwide loss of large herbivores may be one of the most underestimated drivers of global change (Dirzo et al. 2014; Ripple et al. 2015). Hence, it is crucial to understand the consequences of ungulate diversity loss for the functioning of terrestrial ecosystems, particularly in understudied developing countries.

Several experiments have demonstrated that the exclusion of complete ungulate communities can either increase or decrease structural heterogeneity of grasslands (Adler, Raff & Lauenroth 2001) and cause changes in plant species composition (Milchunas, Sala & Lauenroth 1988) and species richness (Olff & Ritchie 1998). In addition, more recent work has shown that grazers can alter the functional community composition of grasslands (Díaz et al. 2007), as plant species with certain traits may be better able to resist or tolerate grazing. We thus have strong evidence that grazing pressure per se highly affects both (i) vegetation structure and (ii) the species and functional composition of grass communities. However, we poorly understand how changes in ungulate community composition, such as declining populations of individual species or replacements of one species by another, may affect the vegetation.

If traits enable a plant species to resist or tolerate herbivory of some ungulate species, but not others, then different ungulate species may have different effects on plant communities. For example, smaller ungulates favour plant species with high leaf nitrogen content, while many larger ungulate species are less selective and also eat species of lower nutritional value (Kleynhans et al. 2011; Treydte et al. 2013). Having low nitrogen content may thus be an effective way to prevent consumption by smaller ungulates and might lead to an increase in relative dominance if competing species with high nitrogen content are eaten. However, if the herbivore community is dominated by large, less selective species, this trait would be less effective. In line with this idea, some studies have demonstrated that grazer species differentially affect species compositions of plant communities (Veblen & Young 2010; Young et al. 2013). However, a mechanistic functional trait-based framework that explains such contrasting effects of different grazers is still largely missing. Such a framework would allow for generalizations of the individual effects of herbivore types on the functional composition of plant communities to a wide array of ecosystems.

Here, we studied the long-term effects of different ungulate size classes on the structure and functional trait composition of plant communities in an African savanna. African savannahs are among the few ecosystems on earth which still harbour diverse communities of ungulates (Prins & Olff 1998), making them an ideal system to study for the effects of different ungulate species on plant communities.

Several studies have demonstrated that ungulates in general (Belovsky 1997), and those from African savannahs (Demment & van Soest 1985; Kleynhans et al. 2011), vary in their food preferences, where differences are often explained by body size. Small ungulate species [e.g. warthog (Phacochoerus africanus) or impala (Aepyceros melampus)] generally prefer plants with high nutrient contents, while larger species [e.g. buffalo (Syncerus caffer)] are less selective (Belovsky 1997; Kleynhans et al. 2011). One very large grazer species, the white rhinoceros, forms an exception and prefers grasses with high nitrogen contents, making its diet similar to those of much smaller grazer species (Kleynhans et al. 2011).

Hence, we hypothesize that both small ungulate species (‘mesoherbivores’ hereafter) and white rhinoceros (or ‘megagrazers’) could be expected to shift the vegetation towards dominance of grasses with low nutritional quality (Fig. 1a) if grasses that resist being eaten by ungulates indirectly profit from the presence of grazers (‘grazing resistance hypothesis’). Alternatively, if the ability to regrow after ingestion (with plants profiting from high nutrient input of, e.g. defecation) is an important strategy to cope with grazers (McNaughton 1984), it might be that meso- or megaherbivores promote fast-growing grasses, even if those also have high nitrogen content (‘grazing tolerance hypothesis’; Fig. 1b). Unlike mesoherbivores, white rhinoceros hardly suffers from predation (Owen-Smith & Mills 2008) and might therefore be food- rather than predation-limited (Owen-Smith 1988). Therefore, it might have even stronger top-down effects on the vegetation than mesograzers (Owen-Smith 1988; Waldram, Bond & Stock 2008; Cromsigt & te Beest 2014). In contrast to meso- or megagrazers, grazers of intermediate size [‘large grazers’ hereafter, such as buffalo and zebra (Equus quagga)] eat the majority of grass species (Kleynhans et al. 2011). As shorter grass species are expected to be restricted by light competition under ambient conditions, large grazers could release them by reducing vegetation height. Hence, large grazers could promote short species with a high ability to regrow after defoliation (‘grazing tolerance hypothesis’; Fig. 1b). Alternatively, if grazing tolerance strategies do not differ much between species, one could expect that large grazers promote plants with low nitrogen content, by eating (and hence removing) more nutritious species (‘grazing resistance hypothesis’; Fig. 1a). However, this scenario might be less likely given the low preferences of large grazers. Despite diet differences between ungulates, they all have in common that by eating, they reduce vegetation height. Therefore, we expect that not specific size classes of grazers, but total grazing pressure, will have strongest effects on vegetation structure.

Details are in the caption following the image
Different size classes of grazers are expected to have different effects on the vegetation. Direct effects are shown using arrow with continuous borders, while indirect effects are shown using arrows with dashed borders. (a) If grazing resistance is most important to cope with ungulates, then short grasses with high specific leaf area (SLA) and high leaf content are expected to be eaten more, so that grazers indirectly promote tall grasses with low SLA and leaf N content. These effects are expected to be most pronounced for more selective meso- and megagrazers. (b) If grazing tolerance is most important to cope with ungulates, then short grasses with high SLA and high leaf content are expected to be better able to regrow after ingestion, so that they are indirectly promoted by ungulates. These effects are expected to be most pronounced for less selective, large grazer species.

We tested this idea using a 10-year experiment with treatments that varied grazing pressure by both meso-, large and megagrazers. In a South African savannah, we excluded grazing ungulate species in a nested design. Within both control plots and in different exclosure types, we characterized both grazing pressure of different herbivores and the species composition of the grass layer. In addition, we used data on three traits: canopy height, related to the ability to compete for light (Weiher et al. 1999); specific leaf area, related to growth rate and shade tolerance (Westoby et al. 2002) and leaf nitrogen content, related to nutritional quality (Behmer 2009) measured for each grass species to quantify community-weighted trait means in each plot.

Materials and methods

Study site

Our study was done in Hluhluwe-iMfolozi Park (HiP), an 89 600 ha park in South Africa (28°00′–28°26′ S, 31°41′–32°09′ E). HiP contains a variety of habitat types, ranging from closed forests, woodland, bunch grasslands and grazing lawns (Whateley & Porter 1983; Stock, Bond & van de Vijver 2010). At large scales, this heterogeneity is caused by gradients in rainfall (Figure S1a in Supporting information), fire frequency and soil factors, while at smaller spatial scales, ungulates can create and maintain structural vegetation mosaics (Cromsigt & Olff 2008).

Annual rainfall in HiP ranges from 490 mm in the south to 920 mm in the north (KZN wildlife, unpublished data). Rain falls mostly between October and April, while in the dry season (May–September), rainfall is below 50 (south) or 200 (north) mm. Fires are usually ignited by park managers as managed burns, with fire return intervals between 2 and 6 years in the area of our study sites. The park has a high diversity of large mammal species, with the most common species that include a significant amount of grass in their diet being white rhinoceros [body weight ranging from 1200 to 3600 kg (average = 2250 kg); park abundance = several thousands (information too conservation-sensitive to provide more details)], cape buffalo (250–850 kg; μ = 550 kg; 4789 individuals), wildebeest (Connocheates taurinus; 140–290 kg; μ = 214 kg; 3002 individuals), plains zebra (175–322 kg; μ = 242 kg; 2749 individuals), nyala (Tragelaphus angasi; 62–150 kg; μ = 98 kg; 4082 individuals), impala (40–76 kg; μ = 53 kg; 14 054 individuals) and warthog (45–150 kg; μ = 83 kg; 1531 individuals; Kingdon 1997; unpublished Earthwatch Institute Field Reports 2004, 2010).

Experimental design

In early 2000, a ‘Russian Doll’ experiment was set up. Five sites, differing in geology (Stock, Bond & van de Vijver 2010), were chosen (Figure S1a) in the central nQmeni section of the park (minimum distance between sites: 2.3 km) to investigate effects of different grazer species on vegetation structure and composition. At each site, four 40 × 40 m plots were fenced (Table 1; Figure S1b): the first plot (‘rhino fence plot’) was fenced with a single cable at a height of 50 cm, with the aim to exclude white rhinoceros while still allowing smaller grazer species. The second plot (‘zebra fence plot’) was fenced with two cables at a height of 0.7 and 1.0 m, in order to exclude zebras and larger species. The third plot (‘impala fence plot’) consisted of a 2 m high fence with a mesh size of approximately 23 × 30 cm, in order to exclude impala and larger grazer species, while still allowing hares and smaller antelope species (Cephalophinae spp.). The fourth plot (‘hare fence plot’) consisted of a 2-m high fence with a mesh size of 1.3 × 1.3 cm, to exclude all grazer species of the size of a scrub hare (Lepus saxatilas) and larger. Finally, at each site, there was an adjacent, unfenced 40 × 40 m ‘control plot’, allowing all grazer species (Table 1; Figure S1b). After setting up exclosures, all sites were burnt biennially (2000, 2002, 2004, 2006 and 2008, in August or September) in order to keep fire frequency the same in all sites and within the range of other sites in the park.

Table 1. The different exclosure types with the different herbivores they are intended to allow in
Control Rhino fence Zebra fence Impala fence Hare fence
White rhino
Buffalo Buffalo
Zebra Zebra
Wildebeest Wildebeest
Impala Impala Impala
Nyala Nyala Nyala
Duikers Duikers Duikers Duikers
Hare Hare Hare Hare
Mice Mice Mice Mice Mice
  • Control plots are accessible to all grazers, rhino fence plots were intended to exclude white rhinos, zebra fence plots were intended to additionally exclude buffalo, zebra and wildebeest, impala fence plots were intended to additionally exclude impalas and nyalas, while hare fence plots were intended to additionally exclude duikers and hares.

Grazer data

Although the fence types we used tended to exclude different herbivore size classes, they did not exclude all the individuals from species that had difficulties crossing them. In addition, our experiment was set up in five different locations within HiP, varying in ambient herbivore densities. As a result, our experiment created continuous variation in the densities of different grazer size classes. Therefore, rather than focusing our analysis on the effects of exclosure types on the vegetation, we were mainly interested in the effects of experimentally manipulated, continuous variation of grazing pressure of different size classes of grazers (or: realized grazing pressure) on vegetation properties. To assess local ungulate densities in both control and fenced plots, footprints, identified to species level, were characterized from March 2001 until September 2005 in each plot. For this, three separate 1 × 4 m strips were dug out to 5 cm depth towards three sides of each plot, and filled with fine loamy (dolorite-derived) sand. This substrate was selected so as to retain footprints also after drying out and/or after subsequent rain. Every 14 days, we counted tracks, identified them to species level and then raked over until smooth. As a relative measure of abundance, we divided each track strip into four subplots, and for every species and within each 40 × 40 m treatment plot, we counted in how many of the twelve (3 strips × 4 subplots) 1 × 1 m subplots tracks of the species was observed. We then averaged (across the bimonthly monitoring sessions) footprint counts per plot and per year for the seven most common herbivores: African buffalo, blue wildebeest, impala, nyala, warthog, white rhinoceros and zebra. To estimate animal densities from footprint counts, we assumed that average footprint counts across control plots reflected average species densities across HiP (as reported by the Earthwatch Institute Field Reports 2004, 2010). As such, we could estimate animal densities in each plot by multiplying the ratio of footprints between a focal plot and the average control plot with the average, park-level density, so that plots with higher than average footprint counts also had higher than average density and vice versa: urn:x-wiley:00220477:media:jec12549:jec12549-math-0001, in which Di,p,s is the density (per hectare) of species i and in plot p, Fi,p is the average (across monitoring sessions) number of footprints of species i in plot p, urn:x-wiley:00220477:media:jec12549:jec12549-math-0002 is the average number of footprints of species i across control plots, Ai is the abundance of species i across HiP (based on Earthwatch Institute Field Reports 2004, 2010) and 89 600 is the number of hectares in HiP [hence (Ai/89 600) is the average density of species i per hectare in HiP]. These estimated animal densities were then used to calculate ‘daily energy expenditure’ (DEE), by assuming that the consumption of a herbivore species scales with its average body mass scaled with a power of 0.75. We adapted the equation from Demment & van Soest (1985) on body size – basic metabolic rate relationships: urn:x-wiley:00220477:media:jec12549:jec12549-math-0003. Here, DEEi,p is the daily energy expenditure (in kJ day−1 ha−1) of herbivore species i in plot p and Mi is the average adult body mass of species i according to Kingdon (1997). The right side of the equation from Demment & van Soest (1985) (urn:x-wiley:00220477:media:jec12549:jec12549-math-0004) was multiplied with 2 (assuming that actual metabolic rate is approximately twice the basic metabolic rate) and multiplied with the animal density, sensu Bakker et al. (2004). DEE estimates per species were then used to calculate (i) total DEE of all grazers combined, (ii) DEE of mesograzers, (iii) large grazers and (iv) megagrazers. We considered impala, nyala, wildebeest and warthog (average body mass < 225 kg) as ‘mesograzers’, buffalo and zebra (225 kg < average body mass < 1000 kg) as ‘large grazers’ and the white rhinoceros (> 1000 kg) as the only ‘megagrazer’. DEE values were calculated in two ways: (i) for individual years, in those years where footprints were counted previous to vegetation characterization (2002; 2003; 2004; see ‘Vegetation data’), these are called ‘short-term DEE’ hereafter and (ii) for the whole period in which footprints were counted (March 2001–September 2005), called ‘long term DEE’ hereafter. This way, we could analyse the effects of both short- and long-term grazing pressure on the vegetation.

Vegetation data

In April 2000–2004 and 2010, at the end of the wet season, vegetation structure and composition were characterized using a point-intercept method. In all 25 plots, a grid spanning 18 × 34 m was laid out, with individual grid cells measuring 2 × 2 m (Figure S1c). At each intersection point [(18/2) + 1] × [(34/2) + 1] = 180 per plot, the most dominant plant species of the grass layer was identified using Van Oudtshoorn (2002). In addition, with a disc pasture metre (diameter: 46 cm, mass: 460 g) vegetation height was measured at each intersection point. Using these data, we calculated variables related to vegetation structure (average vegetation height and spatial vegetation heterogeneity) and functional plant community composition (community-weighted mean (CWM) values of canopy height, specific leaf area and leaf nitrogen content).

Average vegetation height was calculated as the average height across all 180 intersection points of each plot. Spatial vegetation heterogeneity was quantified as the coefficient of variation of all vegetation height measurements in a plot. CWM values (Violle et al. 2007) were calculated with the help of published, in situ measured trait values (Van der Plas & Olff 2014) of all dominant grass species (together accounting for 97.9% of grass observations of this study): urn:x-wiley:00220477:media:jec12549:jec12549-math-0005 where CWMi is the CWM of plot i, s is the species richness in plot i, aj is the relative abundance of the j-th species and tj is the average trait value of the j-th species. CWM values were calculated for three continuous traits: canopy height, that is the height of the base of the highest leaf of a plant and related to the ability to compete for light (Weiher et al. 1999), specific leaf area (SLA), that is the fresh leaf area divided by its dry weight and related to growth rate (Westoby et al. 2002) and leaf nitrogen content, defined as the percent of nitrogen in a leaf and related to nutritional quality for herbivores (Behmer 2009). In line with other studies (Wright et al. 2004), these traits were moderately correlated (canopy height – SLA: = −0.510; canopy height – leaf N content: = −0.660; SLA – leaf N content: = 0.734; Figure S2) with each other, albeit weakly enough to investigate their independent responses to grazing. For details on trait measurements, we refer to Van der Plas & Olff (2014). We also calculated the proportion of grasses that were horizontally growing, stolon-forming lawn grasses (McNaughton 1984; Hempson et al. 2015), in order to investigate whether continuous functional traits provide insights that cannot be picked up by data on classical functional groups of grasses.

Rainfall data

At each of the five sites, from January 2001 until December 2007, rain gauge stations at each site were used to measure the amount of rainfall each day. With these data, average annual amount of rainfall was calculated.

Data analysis

Using a forward model selection procedure with general linear mixed models (LMMs), we analysed how vegetation structure and plant functional community composition were affected by different size classes of grazers and by environmental factors (rainfall, fire). Specifically, we ran LMMs with variables related to vegetation structure or plant functional community composition as a response variable, site and plot (nested within site) as random variables and as fixed factors, variables related to grazing (total/meso-/large or megagrazer DEE), rainfall (annual amount of rainfall) and fire (site burnt in the dry season previous to vegetation sampling or not). To investigate both long- and short-term effects of grazing on the vegetation, we ran separate LMMs with either (i) short-term DEE values (only available for 2002, 2003 and 2004) or (ii) long-term (across-year average) DEE values as predictors. To avoid problems of multicollinearity, we started with a simple LMM only including the fixed factor which predicted the given response variable most significantly. Then, step by step, we added additional, most significant fixed factors that had a variance inflation factor (VIF) below 2.5 with respect to the fixed factors that were already included in the LMM. Additionally, we added significant interaction effects. This model selection procedure was repeated until no significant fixed factor with VIF below 2.5 could be added. To investigate how much variation was explained by final models, we calculated both the marginal and conditional R2 following Nakagawa & Schielzeth (2013). To investigate whether finally selected models (‘main models’) were truly explaining more variation than ‘alternative models’ with other grazer size classes as predictors, for each main model, we ran two alternative models, in which the grazer predictor of the main model was replaced by a grazing predictor from an alternative size class. We then compared the proportion of variation explained, significance and standardized effect sizes between main and alternative models. Finally, we investigated whether for those grazer variables that affected the functional community composition of grasses, how they did so, by (i) investigating with LMMs [site and plot (nested within site) as random variables; and with significant covariates included] for each dominant grass species (>250 observations) to which extent it was affected by the given grazer variable and by (ii) investigating how the effect sizes estimated by those LMMs correlated with average trait values of species.

Additionally, we used LMMs with site as a random factor and ‘fence type’ as a fixed factor to investigate to which extent different fence types decreased short-term DEE values of all grazers combined, meso-, large and megagrazers. When differences between fence types were found, we ran a post hoc Tukey test to investigate which fence types did or did not differ from each other in grazer DEE. LMMs were run with the ‘lme’ function from the nlme package (Pinheiro et al. 2013) in r.

Results

Effectiveness of fences

Fences reduced short-term, total DEE by grazers (F4,66 = 93.244, < 0.001). Short-term, total, short-term DEE by grazers was highest in control plots, intermediate in rhino fence and zebra fence plots and virtually zero in impala and hare fence plots (Fig. 2). Fences affected short-term DEE by mesograzers similarly (F4,66 = 49.851, < 0.001), with values being intermediate in rhino fence and zebra fence plots and virtually zero in impala and hare fence plots (Fig. 2). In addition, fences affected short-term DEE by large grazers (F4,66 = 15.955, < 0.001) with DEE by large grazers highest in control plots and close to zero in all other fence types (Fig. 2). This was also true for DEE of megagrazers, (F4,66 = 39.276, < 0.001) which only had high DEE in control plots (Fig. 2). Fences thus had differential effects on different functional groups, thereby experimentally creating unique grazing assemblages. To investigate whether this allowed us to independently unravel the effects of grazing pressure of meso- versus large versus megagrazers on vegetation structure and communities, we tested whether short-term grazing pressure of these ungulates was correlated. Correlations appeared to be rather low: meso-large grazers: R2 = 0.122; meso-mega grazers: R2 = 0.276; large-mega grazers: R2 = 0.285, confirming the independence of herbivore groups.

Details are in the caption following the image
Short-term consumption by grazers in different fence types. Letters above bars indicate significant differences (post hoc Tukey tests).

Grazing and vegetation structure

Fencing generally increased vegetation height, and these effects were already visible after 1 year (Figure S3). Short-term (effects of the previous year of grazing) and long-term (average, across-year grazing pressure) effects of grazing were similar: combined grazing pressure decreased vegetation height and increased heterogeneity in vegetation structure (Table 2; Figs 3 and 4). In addition, vegetation height was lower but heterogeneity was higher in post-fire years (Table 2; Figs 3 and 4). Heterogeneity in vegetation structure resulted more from the reduction of mean vegetation height than the standard deviation of vegetation height to grazing (Table 2; Table S2).

Table 2. Best fitting models for each response variable (in bold) with significant predictors in normal font
Variable Mar. R2 Cond. R2 d.f. F P Standardized effect size
Short-term grazing effects
Vegetation structure
Vegetation height (cm) 0.528 0.856
Post-fire year 1,48 134.344 < 0.001 −0.537
DEE all grazers 1,48 28.164 < 0.001 −0.375
Spatial heterogeneity 0.570 0.706
Post-fire year 1,47 12.606 < 0.001 0.243
DEE all grazers 1,47 5.434 0.024 0.319
Post-fire year × DEE all grazers 1,47 7.830 0.007 0.341
Vegetation communities
CWM canopy height (cm) 0.000 0.713
CWM SLA (cm2 g−1) 0.158 0.667
DEE mesograzers 1,49 13.810 < 0.001 −0.415
CWM leaf N content (%) 0.136 0.659
DEE mesograzers 1,49 10.734 0.002 −0.391
Proportion of lawn grasses 0.020 0.662
Post-fire year 1,49 2.044 0.046 0.139
Long-term grazing effects
Vegetation structure
Vegetation height 0.163 0.543
DEE all grazers 1,19 24.965 < 0.001 −0.356
Post-fire year 1,124 12.883 < 0.001 −0.201
Spatial heterogeneity 0.114 0.235
DEE all grazers 1,19 17.713 < 0.001 0.304
Post-fire year 1,124 4.234 0.042 0.147
Vegetation communities
CWM canopy height (cm) 0.000 0.590
CWM SLA (cm2 g−1) 0.141 0.498
Post-fire year 1,124 10.538 0.002 −0.186
DEE large grazers 1,19 6.243 0.022 −0.321
CWM leaf N content (%) 0.180 0.612
Post-fire year 1,124 14.870 < 0.001 −0.196
DEE large grazers 1,19 7.070 0.016 −0.375
Proportion of lawn grasses 0.000 0.567
  • For the final model, both marginal (Mar.) and conditional (Cond.) R2 (Nakagawa & Schielzeth 2013) are given. Results based on short-term DEE are shown in the upper part, results on long-term DEE values in the lower part. CWM, community-weighted mean; DEE, daily energy expenditure; SLA, specific leaf area.
Details are in the caption following the image
Long-term effects of grazing. A high, across-year, grazing pressure by all grazers combined decreased vegetation height (top, left) and increased vegetation heterogeneity (top, right). A high, across-year, grazing pressure by large grazers decreased community-weighted mean values of specific leaf area (SLA) and leaf N content.
Details are in the caption following the image
Short-term effects of grazing. A high, last-year grazing pressure by all grazers combined decreased vegetation height (top, left) and increased vegetation heterogeneity (top, right), thus mimicking long-term grazing effects (Fig. 3). A high, last-year grazing pressure by mesograzers decreased community-weighted mean values of specific leaf area (SLA) and leaf N content. Lines show expected values based on linear models with fire years (in case of vegetation height/heterogeneity) and total grazing pressure/grazing pressure by selective/megagrazers as predictor variables.

Grazing and plant community composition

Already 1 year after putting up exclosures, there were visible differences in plant species composition across fence types (Figure S4). Analyses of CWM trait values demonstrate that grazing pressure variation also led to differences in functional composition of grass communities. On the short term, grazing by mesograzers decreased SLA and leaf N content (Table 2; Fig. 4). Effects of large grazers were more pronounced at the long term: across-year average DEE of large grazers decreased both SLA and leaf N content (Table 2; Fig. 3). Comparisons between these finally selected models and alternative models, with other grazer size classes as predictors, confirmed that the grazer size classes in final models explained a higher proportion of the variation than other size classes, with stronger effect sizes and higher significance (Table S1). Although vegetation height decreased with total grazing pressure, this was not primarily caused by shifts in community composition, as grazing pressure did not cause shifts in CWM values of plant canopy height (Table 2).

Follow-up analyses showed that among the common (>250 observations) grass species, those which responded negatively to long-term DEE by large grazers on average had high SLA (LM: F1,5 = 7.768; = 0.0237; R2 = 0.4927) and leaf N content (LM: F1,5 = 8.484; = 0.0195; R2 = 0.5147) but average canopy height (> 0.05; Fig. 4). In addition, grass species which responded negatively to short-term grazing by mesograzers also tended to have high SLA (LM: F1,8 = 5.265; = 0.0509; R2 = 0.3215) and leaf N content (LM: F1,8 = 4.361; = 0.0702; R2 = 0.3528) but average canopy height (> 0.05; Fig. 5). Finally, among the abundant grasses, stolon-forming lawn grass species were similar to bunch grasses in terms of canopy height, SLA and leaf N content (Fig. 5), explaining why responses of these typically defined functional groups (McNaughton 1984) were not mainly responsible for species differences in their response to grazing (Table 1).

Details are in the caption following the image
Response of different grass species to increased grazing pressure as a function of their traits. Responses of grass species to grazing pressure were estimated using linear mixed models with site and plot (nested within site) as random factors, grazing pressure of either large grazers or mesograzers as focal fixed factor and, in the models containing large grazers as predictor, also post-fire year as an additional covariate. Continuous or dotted lines are shown when simple models demonstrated that effect sizes significantly or borderline-significantly (0.05 < < 0.10) correlated with species trait values. Lawn grass species are shown in red; bunch grass species are shown in blue.

Discussion

Our long-term exclosure experiment with different fence types worked well in generating distinct grazer assemblages, with pronounced impact on the vegetation structure and community composition. Data on ungulate abundances were used to calculate continuous variables that reflected grazing pressure by meso-, large and megagrazers, which allowed the testing of a novel set of hypotheses on the relation between grazer assemblages and plant communities. Our approach in quantifying herbivore assemblages contrasts with many previous exclosure studies where differences in herbivory between treatments are generally assumed, but actual herbivore visitation is not measured. In such approaches, treatments reflect potential differences in visitation instead of actual (realized) differences in herbivore assemblages. Due to differences in ungulate densities across sites (unrelated to experimental treatments) and because fence types neither excluded all individuals they were intended to exclude, nor were fully open to all individuals they were supposed to be open to (Fig. 2), we think that this approach is more accurate and realistic than traditional (categorical) analysis methods for exclosure experiments. As we found that grazing pressure of different types of grazers did not correlate strongly with each other, we are confident that we managed to separate the effects of the different functional groups of grazers on the vegetation.

Our main results show that while it was mostly total grazer consumption that affected vertical vegetation structure, specific size classes of grazers appeared to have strongest effects on grass community composition, as we hypothesized (Fig. 1). Mesograzers exerted the strongest short-term effects, by shifting grass communities towards dominance by species with low leaf nitrogen content and low specific leaf area. On the long term, large grazers as zebra and buffalo had strongest effects. Their effects were similar to the short-term effects of mesograzers and in line with the grazing resistance hypothesis: a high, across-year average grazing pressure by large grazers shifted communities towards dominance by species with low leaf nitrogen content and low specific leaf area, presumably because those species were less eaten (Fig. 4). Surprisingly, despite their reputation as keystone species or ecosystem engineers (Waldram, Bond & Stock 2008), white rhinos did not have detectable effects on the community composition of grasses, even though our study area hosts among the highest white rhino densities of the world, with rhino grazing pressure contributing to approximately one-third of the total ungulate grazing pressure in unfenced plots (Fig. 2).

Total grazing pressure by grazers, rather than the grazing pressure by certain size classes, had the strongest, negative effects on vegetation height and positive effects on structural heterogeneity. Interestingly, short-term effects of grazers on heterogeneity were most pronounced when there had been a fire in the dry season previous to vegetation measurements (Fig. 4), possibly because (i) in unburnt vegetation, productivity is too high to allow grazers to create variation in vegetation height or (ii) because grazers maintain the heterogeneity that is created by fire or (iii) palatability differences between different plant functional types decline over time after a fire. Our results are thus in line with the idea that grazing can increase spatial heterogeneity when grazers revisit sites that were grazed before (McNaughton 1984; Adler, Raff & Lauenroth 2001), although such effects may be less strong in ecosystems that do not burn as frequently as the one we studied.

We found that different size classes of grazers have different effects on the functional composition of grasslands (Table 2). Short-term grazing pressure by mesograzers, such as impala or warthog, shifted grass communities to dominance of low-quality grass species with low specific leaf area. Mesograzers are known to prefer grasses of high nutritional value (Kleynhans et al. 2011), and this selective removal of higher quality grasses may have had a positive effect on the relative dominance of ungrazed, lower quality grasses, in line with the grazing resistance hypothesis (Fig. 1a). This finding was somewhat of a surprise, as a high SLA is associated with a high growth rate (Westoby et al. 2002) and hence with a high capacity to regrow after defoliation. In line with this (but in contrast to our findings), others found positive responses of high SLA species to grazing (Vesk, Leishman & Westoby 2004; Cingolani, Posse & Collantes 2005; but see Díaz, Noy-Meir & Cabido 2001). One explanation for this surprising result might be that other, unmeasured traits, also related to regrowth, such as a plant's ability to store resources, were affected by grazing pressure. Hence, it is possible that grazing tolerance affected species’ response to grazing, but that we did not capture this. Another potential explanation for the decrease in SLA with grazing is that this has resulted from a trade-off between grazing resistance and grazing tolerance, as both in global data sets (Wright et al. 2004) and in our study (Figure S2), leaf nitrogen content (a measure for grazing attractiveness, the opposite of grazing resistance) and SLA are positively correlated. Hence, while in our study, grazing resistance strategy appeared more important than grazing tolerance for coping with mesograzers, future research on additional traits might also provide evidence for a role of tolerance.

Over the long term (average grazer densities across years), large grazers as buffalo and zebra appeared to exert strongest effects on grassland communities (Fig. 4). Similarly to the short-term effects of mesograzers as impala, they shifted grass communities towards dominance of low-quality grass species with low specific leaf area. Again, this could be interpreted as an indirect positive effect on low-quality species (grazer resistors) through the ingestion of high-quality species. This came as a surprise, as we expected that due to the lack of strong diet preferences in this group (Kleynhans et al. 2011), large grazers should defoliate all grass species, thereby indirectly favouring grazing tolerant species which were released from light competition (short stature) or which were able to regrow after defoliation (high SLA; Vesk, Leishman & Westoby 2004; Cingolani, Posse & Collantes 2005; Díaz et al. 2007). A potential explanation is that even though large grazers are not as selective as mesograzers, they still have moderate preferences for higher quality species, so that grazing-resistant plans were promoted by large grazers, in line with the grazing resistance hypothesis.

It came as a surprise that megagrazers (white rhino) did not have detectable effects on the vegetation structure or functional community composition, as white rhinos have been coined ‘ecosystem engineers’ (Waldram, Bond & Stock 2008) with previously shown pronounced effects on their surroundings (Owen-Smith 1988; Waldram, Bond & Stock 2008; Cromsigt & te Beest 2014; but see Van der Plas & Olff 2014). One potential explanation could be the duration of our study. Although with 10 years our study lasted longer than most other exclosure experiments, other authors have noted that effects of rhinos on their surroundings might even take longer to manifest (Cromsigt & te Beest 2014). For example, if white rhinos mainly affect their surroundings through non-trophic effects such as trampling and compacting the soil (Van der Plas et al. 2013; Veldhuis et al. 2014), it might take a very long time before effects manifest, but also a long time before effects are reversed. Experiments that last even longer than ours and where non-trophic mechanisms of vegetation change, such as changes in soil properties, are also taken into consideration could shed more light on this. Nevertheless, our exclosure study already revealed clear effects of other size classes of grazers, suggesting that even if rhinos do affect their surroundings, on short or intermediate time scales, their effects are weaker than those of other ungulates and not strong enough to statistically detect. Indeed, in another reserve, Cromsigt & te Beest (2014) demonstrated that rhinos can affect both the vegetation structure and plant community composition, but these effects were moderate. Previous studies, such as Waldram, Bond & Stock (2008) that did find stronger effects, focused on hotspots of rhino use (wallows). Our results are thus likely to be more representative of the whole savannah landscape and show that across landscapes, on short to mid-term time scales, rhino effects on the vegetation are weaker than those of other grazers.

It is worth noting that the main patterns that we found were not primarily caused by shifts in dominance of different growth forms of grasses. Ecologists have for a long time recognized that savannah grasses can be categorized in two main growth forms: short, stolon-forming ‘lawn grasses’ and tall, vertically growing ‘bunch grasses’ (McNaughton 1984; Cromsigt & Olff 2008; Stock, Bond & van de Vijver 2010; Van der Plas et al. 2013). Studies have shown that at least in some circumstances, these growth forms reflect differences in grazing tolerance (Coughenour, McNaughton & Wallace 1985; Ruess, McNaughton & Coughenour 1983; but see Anderson et al. 2013). However, in our study, we did (i) not detect differences between lawn and bunch grasses in grazing responses and (ii) we found that traits which are related to grazing response (SLA and leaf nitrogen content) do not highly differ among lawn and bunch grasses (Fig. 5). Hence, in line with Hempson et al. (2015), we emphasize that even while the classic lawn-bunch grass dichotomy may provide some insight in vegetation responses to grazing, the more continuous variation that exists among savannah grass species better captures reality.

Overall, our analyses show that meso-, large and megagrazers have different effects on savannah grass communities. The fact that consumption by different size classes of grazers better predicted various aspects of grass communities than did overall grazer consumption emphasizes the point that ungulate species should not be seen as equivalent, but that due to their differences in diet, grazer species can highly differ in their effects on the vegetation (Pringle et al. 2014). Our results thereby complement other recent studies focusing on grazing effects on species composition in grasslands. Young et al. (2013) showed that effects of livestock grazing on plant communities in a Kenyan savannah were stronger than effects of wild herbivores. In the same research area, Veblen & Young (2010) showed that due to diet differences, the grazing by large herbivores (mostly zebra) led to a vegetation dominated by an early successional lawn grass species, while the presence of megaherbivores (elephant and giraffe) led to a vegetation dominated by a late successional bunch grass species. Our study builds upon this working by investigating the effects of even more size classes of grazers and by focusing on the functional responses of plant communities. Due to this focus on functional traits, it may be possible to extrapolate some of our findings to other systems. For example, we found that while large grazers mainly had long-term effects on plant communities, mesograzers mostly had short-term effects: only their ‘previous-year’ grazing pressure had strong effects on grass communities and caused communities to be dominated by species with low SLA and low nitrogen content. This suggests that in systems which are dominated by mesoherbivores (larger ungulates locally extinct), population fluctuations of ungulates will also cause strong fluctuations in the functional properties of the vegetation. In contrast, vegetation properties in grasslands dominated by larger ungulate species (often more pristine systems, as larger species are often the first to go extinct) are expected to be more stable, even if the populations of the ungulates fluctuate.

We therefore suggest that the different effects that different functional groups of grazers have on the different plant functional groups contributes to our understanding of local habitat heterogeneity in savannahs and the consequences of size-specific species loss. While high grazer densities are almost ubiquitous in HiP (Cromsigt, Prins & Olff 2009), grazing patterns by different types of grazers may to a large extent be responsible for the high, local heterogeneity that is observed in both our study system (Cromsigt & Olff 2008) and others (DuToit, Rogers & Biggs 2003). As a consequence, while a high total grazer pressure might be important for promoting structural heterogeneity in savannahs, the species or functional diversity of grazers might promote the diversity of functional properties of plant communities across larger scales. Due to cascading effects (Pringle et al. 2011), understanding the role of different size classes of grazers might help to understand the diversity of other trophic levels (e.g. insects, birds) as well.

Acknowledgements

We thank the staff of the SABRE and ZLTP and Georgette Lagendijk, Jeroen Kusters, Matt Waldram, Mariska te Beest, Nicole Hagenah-Shrader, Julia Wakeling and William Bond for helping setting up this experiment and maintaining data collection. Figure drawings were provided by Isabel Martinez Isabel (zebra) and Mathew Hall (rhino) from the Noun Project. This study was funded by the Dutch Scientific Organisation (NWO-Pionier to HO). JPGM Cromsigt was supported by an EU FP7 Marie Curie Career Integration Grant ‘HOTSPOT’ (PCIG10-GA-2011-304128).

    Data accessibility

    Raw electronic data, except those related to footprints, are available from the Dryad Digital Repository http://dx.doi.org/10.5061/dryad.512m0 (Van der Plas et al. 2016). Footprint data have not been made publicly available, as these contain conservation-sensitive information, although those who are interested can contact the first author for more information.