Herbivores shape woody plant communities in the Kruger National Park: Lessons from three long-term exclosures

The role of herbivory in structuring plant communities and determining community compositions is well recognised (see reviews by Augustine & McNaughton 1998; Huntly 1991). For example, native ungulate browsers have been found to have major impacts on shrub dynamics in East African savannas (Augustine & McNaughton 2004). Different feeding behaviours and food preferences of browser species have major impacts on South African (Owen-Smith & Cooper 1987) and West African savanna woody communities (Jachmann & Croes 1991). These impacts are often attributed to selective feeding related to food quality (Owen-Smith & Cooper 1987). As in savannas (e.g. Staver et al. 2009), herbivores have been found to regulate recruitment rate and species composition of trees in temperate forests (Ammer 1996; Kriebitzsch et al. 2000; Van Hees, Kuiters & Slim 1996) and grassland (Anderson & Briske 1995).


Introduction
The role of herbivory in structuring plant communities and determining community compositions is well recognised (see reviews by Augustine & McNaughton 1998;Huntly 1991).For example, native ungulate browsers have been found to have major impacts on shrub dynamics in East African savannas (Augustine & McNaughton 2004).Different feeding behaviours and food preferences of browser species have major impacts on South African (Owen-Smith & Cooper 1987) and West African savanna woody communities (Jachmann & Croes 1991).These impacts are often attributed to selective feeding related to food quality (Owen-Smith & Cooper 1987).As in savannas (e.g.Staver et al. 2009), herbivores have been found to regulate recruitment rate and species composition of trees in temperate forests (Ammer 1996;Kriebitzsch et al. 2000;Van Hees, Kuiters & Slim 1996) and grassland (Anderson & Briske 1995).
Many of the studies on browser-plant interactions in savannas have tended to focus on interactions between browsers and specific woody species.For example, Coetzee et al. (2008), Gadd (2002) and Helm and Witkowski (2012) looked at browser interactions with Sclerocarya birrea, whilst Fornara and Du Toit (2007) documented the response of Acacia nigrescens to ungulate browsing.These two species are important browse species in southern African savannas.As savanna woody species tend to be highly variable in form and function, these studies usually do not provide a better understanding of savanna dynamics at the community level.Augustine and McNaughton (2004) emphasised the need for replicated experiments that assess the effects of native browsers on shrub dynamics in African savannas, as our understanding of the importance of browser effects on woody plant dynamics remains unclear.A better understanding of how browsers and woody plants interact at the community level and how these interactions affect savanna dynamics will be invaluable in managing browser populations.
The existence of three long-term herbivore exclosures in the Kruger National Park, South Africa, provides a good opportunity to empirically determine the role of browsers in determining woody species composition and population structures of woody communities growing at the three study sites.Previous studies have shown evidence for taller canopies and higher tree and shrub densities inside exclosures at Nhlangwini (Asner et al. 2009;Levick et al. 2010;Trollope et al. 1998), Makhohlola (Levick et al. 2009;Trollope et al. 1998) and N'waxitshumbe (Asner et al. 2009;Levick & Rogers 2008;Trollope et al. 1998).In these studies, the analyses were generally based on remote sensing and ignored species identity.Differences in woody community compositions and plant traits between exclosure treatments are therefore still largely undocumented.
Differences in species compositions between treatments would be highly informative as the species found inside but not outside, are likely to be vulnerable to herbivory and therefore no longer found outside of the exclosures or found at reduced densities.The traits of these species can be used to provide us with clues as to what makes them vulnerable to herbivory.For example, Díaz et al. (2007) found that at the global scale there is a spectrum of plant responses to herbivory with high levels of herbivory favouring annual over perennial plants, short plants over tall plants, prostrate over erect plants and stoloniferous plants over those with tussock architecture.This article aims to determine how browsers and plant traits interact and how these interactions affect woody plant population dynamics and community assemblages.We aim to determine if there are differences in species compositions and key plant functional traits between the woody communities found growing inside and outside the three long-term exclosures in the Kruger National Park (KNP).Plant traits are important in determining both the type of and degree of vertebrate herbivory, whilst they also can reflect plant responses to herbivory.We measured a number of plant traits that are likely to impact and respond to browsing; these included leaf nitrogen (N) and phosphorus (P), leaf carbon-nitrogen ratio (C:N), specific leaf area (SLA), leaf size, leaf dry matter content (LDMC), leaf tensile strength, bite size index (BSI) and levels of secondary compounds.High leaf N and leaf P are associated with higher nutritional quality for herbivores (Cornelissen et al. 2003).Leaves with lower C:N ratios are much more attractive to herbivores as these leaves have higher N content (Cornelissen et al. 2003).The SLA of a species is usually a good correlate of its potential growth rate.Lower SLA values tend to correspond with relatively high investments in defences, both structural and chemical, and long leaf life span, all of which affect nutritional value (Cornelissen et al. 2003).Environmental nutrient stress and disturbance, as well as phylogenetic factors, play a role in determining leaf size (Cornelissen et al. 2003).Smaller leaves with higher SLA are typically more attractive to herbivores.Leaves with high LDMC values tend to be tough and therefore more resistant to disturbances such as herbivory or wind than leaves with low LDMC.Species with low LDMC are usually associated with productive and highly disturbed areas (Cornelissen et al. 2003).Leaf tensile strength is a good indicator of the relative carbon investment in structural protection of the photosynthetic tissue.Physically stronger leaves are better protected from abiotic (winds, storms) and biotic mechanical damage (herbivory).Higher physical strength usually results in longer leaf span; however, this is usually combined with other physical and chemical defences such spines and tannins (Cornelissen et al. 2003).BSI is a measure of how easily leaves can be browsed by a simulated browser (a human), whilst stem specific density broadly trades off against relative growth rate and stem defences against pathogens, herbivores and mechanical damage (Cornelissen et al. 2003).Condensed tannins and polyphenols are organic N-free chemical defences that bind with protein, reducing N availability to herbivores, thereby decreasing preference by mammal herbivores (Bergvall & Leimar 2005;Haslam 1988;Owen-Smith 2005).The concentrations of the cell wall constituents (neutral detergent fibre [NDF], acid detergent fibre [ADF] and acid detergent lignin [ADL]) determine the digestibility of leaf material for herbivores.Detergent analysis has been used to compare fibre digestibility across different plant types (e.g.see Codron et al. 2007 and references therein).
The following questions are specifically addressed using the three exclosures.Firstly, has the long-term exclusion (> 40 years) of herbivores led to differences in woody plant species compositions at the three exclosures?Secondly, do the structures, densities and abundances of the woody species populations differ inside and outside of the exclosures?Thirdly, are there differences in plant leaf and stem traits between the species only found inside versus common species found both inside and outside?Fourthly, has the long-term removal of herbivores from the exclosures had any effects on soil nutrient cycling?Fifthly, do the exclosures help us to predict under what conditions herbivores play an important role in determining the structure and composition of woody plant communities in savannas?

Study sites
Three long-term herbivore exclosures (> 40 years old) exist in the KNP.These include the Nhlangwini exclosure near Pretoriuskop, the Makhohlola exclosure near Crocodile Bridge and the N'waxitshumbe exclosure on the northern plains (Table 1).At each exclosure two treatments were sampled, one inside the exclosure and one outside the exclosure at close proximity.
The 220.0 ha Nhlangwini exclosure was established in 1973 and occurs in the south-western portion of the KNP near Pretoriuskop rest camp.This is the wettest part of KNP with mean average precipitation close to 750 mm.The exclosure is situated within the broad-leaved bushveld vegetation Page 2 of 12 type that occurs on sandy soils derived from granitic rocks, common species include Terminalia sericea and S. birrea (Venter, Scholes & Eckhardt 2003).Mean fire return interval is similar inside and out, with fire occurring on average every 3.6 years inside (Asner et al. 2009) and less than every 4 years outside (Smit et al. 2012) (2012) showed that the area surrounding the N'waxitshumbe exclosure has a mean fire return interval of less than 4 years.
The faunal assemblage in the park includes all the major mammal species typical of the region.

Community composition
The distribution, abundance and density of the species growing both inside and outside of each of the exclosures was determined by walking four 100 m long by 4 m wide transects for each treatment at each site (1600 m 2 ) in March 2011.The four transects in each treatment were taken parallel to each other and 50 m apart.Inside and outside pairs were situated in as close a proximity as fences and firebreaks permitted.Care was also taken to ensure that the paired transects were on similar positions along catenal gradients.At Nhlangwini, the paired transects were taken perpendicular to the northern fence of the exclosure and parallel to the drainage in the area, outside transects were taken to the north-west of the exclosure.At Makhohlola, the paired transects were taken perpendicular to the western fence and perpendicular to the drainage at the site, outside transects were taken to the north of the exclosure.Owing to the limited size of the Makhohlola exclosure (100 m × 240 m) two of the four transects were taken in the no-burn part of the fire experiment (since 2004) and two were taken in the part that burns in conjunction with the surrounding landscape.At N'waxitshumbe, the paired transects were taken perpendicular to the southern fence in the south-western portion of the exclosure and parallel to the drainage in that area.
Species identity was recorded for each woody plant encountered in each transect using a personal digital assistant (PDA) with CyberTracker software.The abundance data were used to create an inside abundance index, whereby the proportion of a species found in the inside compared to the outside treatment was calculated.This index was calculated by dividing the number of individuals of each species found inside by the total number of individuals found both inside and outside.Thus a value of 1 indicates the species was only found inside, whilst a value of 0 indicates the species was only found outside.The inside abundance index was then plotted against a suite of leaf traits to test whether species growing on the inside had high values of these traits.This index can be used as a proxy to determine which plant species are preferentially selected for or 'favoured' by browsers relative to their abundance.

Trait sampling
The dominant woody plant species making up 80% -90% of the woody biomass inside and outside the three exclosures in the KNP were determined (Table 2).Fourteen different plant traits relating to the leaves and stems were sampled for each species.All trait data (except BSI) were measured and recorded according to the methods explained in Cornelissen et al. (2003).For each species, leaf material was collected from five individuals and analysed for N, P and carbon (C) content using a Leco TruSpec CN Analyser (LECO Corporation, St. Joseph, MI).SLA and average leaf area were measured using four healthy and complete sun-exposed leaves from five individuals (20 leaves per species).The BSI of a species is the total dry weight of the leaves removed from two human bites taken from five individuals (total of ten bites).An attempt was made to remove the maximum amount of leaf material with each bite.The BSI for all species was measured by the same person in order to control for potential differences between individual recorders.Total condensed tannins were calculated according to the methods described by Hattas and Julkunen-Tiitto (2012), whilst total polyphenols (TP) were calculated according to the methods described by Hattas et al. (2005).Cell wall constituents, that is, NDF and ADF, were determined in an Ankom fibre analyser (Ankom Technology Corp., Fairport, NY), whereas ADL was determined in accordance with the acid detergent lignin in beakers method.Ash was determined by incinerating the filter bag containing plant residue in a muffle furnace at 525 °C for 3 h.All cell wall constituents were determined consecutively as outlined in the Ankom Technology procedures (Ankom Technology 2010).

Soil sampling
Five replicate soil samples were collected both inside and outside of each exclosure at three depths (0 cm -10 cm, 10 cm -20 cm and 40 cm -50 cm) using a soil auger.Care was taken to sample away from the canopies of trees.Soil bulk density measurements were taken at these same depths.The methods used for soil sampling and analyses are described in detail by Wigley et al. (2013).

Statistical analyses
All analyses were performed using R (R Project 2012).Chisquare tests were used to test for differences in tree densities between treatments.Non-parametric Wilcoxon tests were used to test for differences in trait measurements as the low replication resulting from the low species diversity at each site precluded the use of statistical tests founded on the normal distribution.Analyses of variances were used to test for differences in soil nutrients between treatments using pooled data from all three depths, as the conditions for homoscedasticity were not violated (fligner test: p > 0.05; Conover, Johnson & Johnson 1981).Principle components analyses (PCA) were used to explore the relationships between treatments, sites and measured plant traits.The function 'PCAsignificance' available in the BiodiversityR package (Kindt & Coe 2005) in R was used to test for significant PCA axes using the broken-stick distribution.

Community composition data
At the Nhlangwini exclosure, there were a number of differences between the two communities.The four most common species for both treatments were T. sericea, Dichrostachys cinerea, S. birrea and Acacia gerrardii.However, T. sericea was found to be much more abundant (nearly 80% of all trees) inside compared to outside (c.35%) of the exclosure (Figure 1).Dichrostachys cinerea showed the opposite pattern, with a higher relative abundance outside (30%) compared to inside (15%) the exclosure (Figure 1).Sclerocarya birrea and A. gerrardii showed similar abundances between treatments (Figure 1).Both treatments had a few unique species, with Strychnos madagascariensis, Catunaregam spinosa and Searsia leptodictya only found inside, whilst Antidesma venosum, Combretum hereroense, Albizia harveyi, Philenoptera violacea and Piliostigma thonningii were only found outside.Species diversity was found to be higher on the outside of the exclosure, with a total of 16 species compared to 12 species inside.
There were distinct differences in woody communities growing inside and outside of the Makhohlola exclosure (Figure 1).Acacia nigrescens was found to be more abundant inside the exclosure (32%) but was still relatively abundant outside (22%).Flueggea virosa was common inside (15%) but rare outside (4%).Gymnosporia senegalensis was found to have a much higher abundance outside (33%) compared to inside (12%).Lannea schweinfurthii and S. birrea had lower abundances outside (7% and 1%, respectively) compared to inside (12% and 7%, respectively).Dichrostachys cinerea and A. harveyi were relatively common outside (13% and 9%, respectively) but rare inside (3% for both).Dalbergia melanoxylon, Ehretia rigida and Euclea undulata were only found inside the exclosure, making species diversity higher on the inside (13 species) than outside (10 species).
There were also distinct differences in the woody communities growing inside and outside of the N'waxitshumbe exclosure.At this site, C. mopane was found to be the most common species both inside (43%) and outside (40%, Figure 1).The second most common species inside the exclosure (37%), D. melanoxylon, was not found outside of the exclosure.
Combretum imberbe and P. violacea were more common outside (14% for both) compared to inside (7% and 1%, respectively).Sclerocarya birrea and G. senegalensis were relatively rare both inside and outside of the exclosures.Dalbergia melanoxylon, Grewia monticola and Ozoroa obovata were unique to inside, whilst D. cinerea, A. harveyi and A. nigrescens were only found outside of the exclosure.Thus species diversity was the same inside and outside of the exclosure, but species composition differed.

Population structure
Total woody plant densities were significantly higher for the inside treatments at all three sites (Figure 2).These large differences were determined by the most common species growing at each site At Makhohlola, major differences were evident between treatments for each of the four most common species (Figure 4).The size class distributions of L. schweinfurthii and A. nigrescens showed an approximately inverse J-shaped curve for the populations growing inside the exclosures but not for the outside populations.The inside populations of these two species both had a much higher number of individuals in the smaller (< 2 m) size classes than the outside populations (Figure 4).The outside population of G. senegalensis was more uniform in height (1 m -2 m) than    the inside population (0 m -3 m).There were many more and taller F. virosa plants growing on the inside of the exclosure compared to the outside (Figure 4).

A l b i z i a h a r v e y i A l b i z i a h a r v e y i A l b i z i a h a r v e y i P h i l e n o p t e r a v i o l a c e a P h i l e n o p t e r a v i o l a c e a P h i l e n o p t e r a v i o l a c e a P h i l e n o p t e r a v i o l a c
At N'waxitshumbe, the two most common species C. mopane and C. imberbe had similar size class distributions in each treatment (Figure 5).However, there were higher densities in each size class for the inside treatment for C. mopane, whilst C. imberbe had similar numbers in each size class for both populations.The most striking difference was found for D. melanoxylon, where a population spanning the full range of size classes was found inside the exclosure, whilst not one individual was found outside of the exclosure in the sampling area (Figure 5).

Leaf traits of unique inside species versus common outside species
There were no significant differences (p > 0.05; Wilcoxon test) in leaf nutrient concentrations -N, C, C:N, P, Calcium (Ca), Magnesium (Mg), Sodium (Na) and Potassium (K)between each of the inside and outside treatments (Table 3).There were, however, some consistent trends; the lack of statistically significant differences was likely a result of the small sample sizes resulting from low woody species diversity at the sites.Mean leaf N (%) was noticeably higher for the inside treatment at all three sites (Table 3).The higher N and similar C concentrations inside of the Nhlangwini exclosure resulted in a lower mean C:N ratio, whilst the other two sites had similar C:N ratios (Table 3).Leaf P (%) was similar between treatments at all three sites (Table 3).
There were no significant differences (p > 0.05; Wilcoxon test) in SLA, leaf size, BSI and tensile strength between treatments.
Mean SLA was consistently higher for species restricted to the inside treatments (Table 3).Average leaf area, which was highly correlated to BSI, were both higher inside the exclosures at the two nutrient rich sites on basalts (Makhohlola and N'waxitshumbe), whilst the opposite pattern was true for the Nhlangwini site which is situated on nutrient poor granite derived soils (Table 3).Mean leaf tensile strength was higher outside exclosures at the two basalt sites, but no difference was evident between treatments at the granite site (Table 3).
The pooled data showing the inside abundance index of each species plotted against leaf traits and BSI showed a weak positive relationship between the inside abundance index and leaf N and P (Figures 6a and 6b).Almost no relationship was evident between the inside abundance index and SLA and BSI (Figures 6c and 6d; p > 0.05 for all correlations).

Chemical defence
No significant differences (p > 0.05; Wilcoxon test) in total phenolics, condensed tannins and fibre were evident between treatments.However, an interesting pattern emerged; mean leaf phenolic concentrations were consistently lower for species restricted to the inside treatments compared to those in the outside treatment, with the biggest difference evident at the Nhlangwini site (Table 3).Mean condensed tannin concentrations showed no consistent patterns between treatments and sites (Table 3).The three types of fibre analyses showed different patterns.NDF, which includes hemicellulose, cellulose and lignin, was slightly lower for the outside treatments at Makhohlola and N'waxitshumbe, whilst the opposite pattern was found at Nhlangwini (Table 3).ADF, which includes cellulose and lignin, showed no consistent patterns (Table 3).ADL, which only includes lignin, was consistently lower for the inside treatments at all three sites (Table 3).

Soil nutrients
Although the pooled data for all depths of soil N and P showed a pattern of higher values for the three inside treatments, the only statistically significant differences were evident at N'waxitshumbe (p < 0.05; Figures 7a and 7c).Soil C and K also showed a pattern of being consistently higher for the inside treatments at Makhohlola and N'waxitshumbe, with no differences evident at Nhlangwini (Figures 7b-7d).Similar patterns were evident for N, C and P when total N, nitrogen; P, phosphorus; SLA, specific leaf area; BSI, bite size index.In Out nutrient stocks were calculated using soil bulk density data (data not shown).

Multivariate analyses
The PCA performed on the data using 13 traits relating to the leaves and stem (Figure 8) showed that the first two principal components together explained 44.0% of the total variation amongst species.The most important traits separating species along the first PCA axis were NDF, ADL and TP.The most important traits separating species along the second PCA axis were leaf N, leaf C:N and stem density (Figure 8).The PCA showed that the species from the inside tended to more variable along PCA axis 2 (higher leaf N and stem density), whilst the species from the outside tend to be more variable along PCA axis 1 (higher fibre and TP values).The variance explained by the first PCA axis was found to be insignificant, as the percentage of explained variance (24.3%) was not larger than the corresponding percentage of variance of the broken-stick distribution (24.5%), whilst the second PCA axis was significant as the percentage of explained variance (20.3%) was larger than the corresponding percentage of variance of the broken-stick (16.8%) distribution (Kindt & Coe 2005).

Discussion
Clear differences were found in both the species composition and population structure of the woody plant communities growing inside and outside of three herbivore exclosures in the KNP.Although the three exclosures were situated in very different vegetation types, some similarities exist in species' responses to low herbivory.Some species were frequently encountered both inside and outside of the exclosures (e.g.T. sericea and D. cinerea at Nhlangwini, A. nigrescens and G. senegalensis at Makhohlola, C. mopane at N'waxitshumbe).Smaller size classes tended to be better represented inside, and larger size classes outside, for species which were found in large numbers both inside and outside of the exclosures.
There were some exceptions though; G. senegalensis for instance, had similar population structures both inside and out.This could be because this species is not a preferred browsing species as a result of the high condensed tannin and ADL concentrations in the leaves (see Shrader et al. 2011).We argue that savanna woody species are either well adapted to browsing (such as D. cinerea), well defended (such as G. senegalensis) or both (such as A. nigrescens).For example, Fornara and Du Toit (2007) showed that heavily browsed A. nigrescens trees develop tolerance traits, such as high regrowth rates and extensive branching, as well as resistance traits, such as close thorn spacing, suggesting that it is both well adapted and well defended.In fact, a metaanalysis by Carmona, Lajeunesse and Johnson (2011) has shown that secondary metabolites are less important in plant defence when compared to other defensive plant traits such as physical resistance, gross morphology and growth rates.
Some species were found more frequently outside of the exclosures, such as D. cinerea, which was found in higher numbers outside of the exclosure at Nhlangwini, and A. harveyi at all exclosures.These species are most likely stimulated by disturbance, for example D. cinerea shrubs rootsucker, and expand significantly when stimulated by fire and browsing (Wakeling & Bond 2007).More frequently, species were only found inside the exclosures (e.g.D. melanoxylon, G. monticola and O. obovata at N'waxitshumbe).Dalbergia melanoxylon had the highest leaf N of all the species in this study and we assume that the species found only on the inside of the exclosures are mostly very palatable and targeted by browser species outside of the exclosures.Dalbergia melanoxylon was found outside of the N'waxitshumbe exclosure in a previous study, but in lower lying areas than at our study area (Levick & Rogers 2008).It is uncertain whether D. melanoxylon has since disappeared outside the exclosure or whether this species may be preferentially targeted by browsers at specific positions along the catena.
Why do species become locally extinct outside of exclosures?Impala and kudu are highly selective when foraging in nutrient-poor broadleaf savannas (Cooper & Owen-Smith 1985;Owen-Smith & Cooper 1987).This selectivity is attributed to differences amongst species in the level of defensive chemical compounds.Alternatively, these species may also be predisposed to Total tree densities were significantly higher for the inside treatments at all three sites, usually with more individuals falling into the smaller size classes inside of the exclosures.These findings are in strong agreement with previous studies from the same study sites.Trollope et al. (1998), Levick and Rogers (2008) and Asner et al. (2009) reported higher tree densities and different population structures between treatments.Moncrieff et al. (2011) reported different tree allometries between trees growing inside and outside of the same herbivore exclosures in the KNP, with heavily browsed trees shorter for a given stem diameter.Our results are therefore in agreement with previous studies and provide strong evidence that herbivory has a major impact on savanna woody plant diversity, species population structure and overall woody plant densities.
The next step was to determine if there were differences between treatments for some of the key plant functional traits of each woody community.There was a trend in leaf N concentration, with higher mean N for the species unique to the inside treatments compared to the species found growing outside.No obvious patterns or trends were evident for leaf C:N or P. Leaf structural traits, SLA, average leaf size and BSI, which are all correlated, showed similar trends with higher values inside the exclosures at the two fertile sites and lower values inside compared to outside at the infertile site.The opposite pattern was evident for tensile strength where the leaves were generally tougher on the outside at the two fertile sites with no difference at the infertile site.Apart from leaf N, total phenolic content and lignin content were the only traits that showed consistent patterns between treatments with higher values for species outside of the exclosures.Increases PCA, principal components analysis; LP, leaf phosphrus; SD, stem density; LN, leaf nitrogen; CT, condensed tannins; SLA, specific leaf area; ADL, acid detergent lignin; NDF, neutral detergent fibre; ADF, acid detergent fibre; LCN, leaf carbon to nitrogen ratios; ALA, average leaf area; TP, total phenolics; BSI, bite size index.
Neutral detergent fibre, acid detergent lignin and total phenolics were most influential along PCA axis 1 (24% of total variation), whilst leaf nitrogen, leaf carbon-nitrogen ratio and stem density were most influential along PCA axis 2 (20% of total variation).The locations of each species on the first two PCs are shown as well as the groupings by treatment (inside versus outside). of carbon-based structural metabolites with decreases in leaf N have been shown with high browsing intensity in previous work (Bryant et al. 1991;Wessels, Van der Waal & De Boer 2007).An alternative theory for increases in lignin and leaf N inside the exclosures may be that herbivores preferentially select species with higher leaf N and that species with lower total phenolic and lignin content are also targeted by herbivores.
In general, the removal of browsers led to higher soil nutrients, the response was the greatest at the two basalt sites.At N'waxitshumbe differences in soil N and P were higher with browsers absent, whilst soil C and K were also higher inside of the exclosures.The same pattern was evident at Makhohlola, whilst the nutrient-poor site (Nhlangwini) showed almost no differences between treatments for all measured soil nutrients.Although herbivores have been found to increase soil nutrients (e.g.Augustine & McNaughton 2006;Frank 2008), Ritchie, Tilman andKnops (1998) found that herbivores can actually depress soil N by indirectly decelerating N cycling through decreasing the abundance of plant species with N-rich tissues.The increased soil fertility inside of the exclosures could possibly be explained by differing litter dynamics.The expectation would be that leaf litter with higher N and lower fibre content would result in faster cycling and improved soil fertility (e.g.Reich et al. 2001).Furthermore, the higher density of woody plants would increase the amount of leaf litter reaching the soil.Thus the higher soil nutrients found inside of the exclosures could possibly be a result of the indirect effect of herbivores, as their removal resulted in higher tree densities.Similarly, fire indirectly affects nitrogen cycling in the KNP by decreasing woody densities (Coetsee, Bond & February 2010).The similar soil nutrient concentrations found inside and outside of the Nhlangwini exclosure can most likely be attributed to the overall low litter quality and lower densities of herbivores resulting from the nutrient poor granitic soils at this site.

Conclusion
This study has demonstrated the important role browsers play in savanna dynamics.Browsers directly impact species distributions, densities and population structures by actively selecting for species with favourable traits; this study suggests that species with high leaf N, low total phenolic content and low ADL are favoured.Forty years of no browsing impact was insufficient to allow the recruitment of the dominant species into larger size classes.This supports the idea that cohorts of trees recruit simultaneously, they would however need a break from both herbivory and fire.This study also suggests that browsers have important indirect effects on savanna functioning through their impact on soil nutrient cycling.The magnitude of these indirect effects on soil nutrient cycling appears to depend on inherent soil fertility and climate.The effect of browsers on both vegetation and soils was highest at the low rainfall site with high soil fertility, with a somewhat lesser effect at the higher rainfall basalt site, with almost no effect at the infertile high rainfall granite site.This work highlights the importance of herbivore exclosures in protected areas and some of the challenges faced by researchers when studying browser effects on savanna ecosystems because of the paucity of replicated long-term herbivory exclosures.Future work should identify other possible sites in savannas from around the globe that could be used to corroborate the findings from this study.
. The size class distributions for the two most common species growing at the Nhlangwini exclosure (T.sericea and D. cinerea) show very different patterns for each treatment.The size class distribution of T. sericea inside the exclosure shows an inverse J-shaped curve, with many plants in the smaller size classes less than 3 m and fewer plants in the large size classes.The size class distribution for outside the exclosure shows much lower numbers of plants in the smaller size classes and fewer plants in the very large size classes (Figure 3).A different pattern was evident for D. cinerea, with most plants falling into size classes between 2 m and 4 m inside the exclosure, whilst plants on the outside were more evenly distributed in a greater range of size classes (Figure 3).

FIGURE 2 :FIGURE 1 :
FIGURE 2: Total tree densities for all woody plants greater than 0.5 m in height inside and outside the three long-term herbivore exclosures in the Kruger National Park.
l e a n a t a l e n s i s

FIGURE 3 :FIGURE 4 :
FIGURE 3: Size class distributions of Terminalia sericea populations growing (a) inside and (b) outside and Dichrostachys cinerea populations growing (c) inside and (d) outside of the Nhlangwini exclosure in the Kruger National Park.

FIGURE 5 :
FIGURE 5: Size class distributions of Dalbergia melanoxylon populations growing (a) inside, Combretum imberbe populations growing (b) inside and (c) outside and Colophospermum mopane populations growing (d) inside and (e) outside of the N'waxitshumbe exclosure.No Dalbergia melanoxylon plants were found growing on the outside of the exclosure in the Kruger National Park.

FIGURE 6 :
FIGURE 6: Data for, (a) leaf nitrogen, (b) leaf phosphorus, (c) specific leaf area and (d) bite size index plotted against the proportion of species found on the inside compared to the outside of three exclosures in the Kruger National Park.A value of 1 indicates that the species was only found inside, whilst a value of 0 indicates the species was only found outside the exclosures.
; C, carbon; P, phosphorus; SB, Sum of Bases; in, inside treatment; out, outside treatment.N = 15 for each treatment.

FIGURE 7 :
FIGURE 7: Mean (± s.e.) for, (a) soil nitrogen, (b) carbon, (c) phosphorus and (d) potassium concentrations inside and outside of three exclosures in the Kruger National Park.Plotted means are calculated from five replicates taken at three depths.

FIGURE 8 :
FIGURE 8: Principal components analysis of trait measurements from 20 species sampled inside and outside of three exclosures in the Kruger National Park.

TABLE 1 :
Site names, locations and descriptions of the exclosures in the Kruger National Park where sampling took place.

TABLE 2 :
Dominant woody species accounting for > 80% of standing biomass for inside and outside treatments at the three Kruger National Park exclosures.
In, inside treatment; Out, outside treatment.

TABLE 3 :
Mean (± s.e.) values of the measured leaf traits for the inside and outside treatments of three exclosures in the Kruger National Park.