Browser impacts in Mapungubwe National Park , South Africa : Should we be worried ?

Elephants can have major, positive ecological effects on savanna dynamics (Campos-Arceiz & Blake 2011; Kerley & Landman 2006; Owen-Smith 1989), but they may also cause loss or extinction of woody and succulent plant species, especially in fenced conservation areas (De Beer et al. 2006; Gandiwa et al. 2011; Guldemond & Van Aarde 2008; Kerley et al. 2008; Landman et al. 2014; O’Connor, Goodman & Clegg 2007; Shannon et al. 2008).

Introduction elephant impacts on vegetation are further complicated when the effects of fire and other herbivores are included (Holdo 2007;Holdo, Holt & Fryxell 2009;Moncrieff, Kruger & Middgley 2008;Midgley, Lawes & Chamaillé-Jammes 2010;Shannon et al. 2011;Valeix et al. 2011). While fire is not an important ecological driver in arid areas, herbivores are important agents of top-down control in savanna systems (Bond 2005). Browsing and mixed-feeding herbivore species (Table 1) may, either individually or in combination, significantly affect the woody vegetation dynamics in Mapungubwe.
This study assesses the impact of elephants and other herbivores on the woody component of the main vegetation types of Mapungubwe. Besides the measured levels of impact serving as a useful benchmark for future studies, we also wanted to evaluate the effect of specific localities, vegetation types and species on woody vegetation. The hypotheses that we tested were: (1) the impact of elephants and other browsers will be more severe for certain plant species within each vegetation type, (2) mesoherbivores contribute to the high levels of browser impact on the vegetation, (3) the impact of elephants and other browsers will be higher in certain vegetation types, with mopane-dominated vegetation being either more preferred because it is easily accessible on flat plains or less preferred because it grows in association with sandy and nutrient-poor soils and (4) vegetation on ridges will be less impacted as ridges will act as refugia from elephants. We could not measure whether elephants and other browsers prefer certain vegetation types or species as we did not measure rates (e.g. searching time, feeding time) of utilisation, instead we used direct impact as a proxy for preference. We made the assumption that the preference for a vegetation type was reflected by greater impact levels on it.

Study site
Mapungubwe National Park (~280 km 2 ) is situated on the southern banks of the Limpopo River in South Africa with Botswana and Zimbabwe on its northern border ( Figure 1). Mapungubwe forms part of the Greater Mapungubwe Transfrontier Conservation Area (GMTFCA), an area of 3065 km 2 which creates a continuous conservation area connecting South Africa, Botswana and Zimbabwe. The region is semi-arid with low and unpredictable rainfall of approximately 350 mm -400 mm per annum (O'Connor & Kiker 2004). The mean annual temperature is 21.6 °C with monthly minimum averages of 7.2 °C -20.3 °C and maximum averages of 24.7 °C -32 °C (Mucina & Rutherford 2006;O'Connor & Page 2014). Mucina and Rutherford (2006) classified three main vegetation types in Mapungubwe: Subtropical Alluvial Vegetation, Musina Mopane Bushveld and Limpopo Ridge Bushveld, while riparian areas on the Limpopo River are classified as Lowveld Riverine Forest. These broad vegetation types were previously classified as seven separate vegetation types (see Table 2; maps in Scientific Services data repository, based on South African Defence Force 1986a, 1986b).
The area's geology comprises Archaean Beit Bridge Complex, younger basalts of the Karoo Supergroup and sandstones of the Clarens Formation (Mucina & Rutherford 2006). Recent alluvial deposits with deep, fine-structured sandy to loamy soils are found along the rivers. The Limpopo River forms the northern boundary of Mapungubwe and several artificial watering points in the park provide water throughout the year. All our sampling sites were within 5 km of perennial water supplies and the average distance to water was 2.6 km (Scientific Services, unpublished data).

Data collection
In January and October 2012, we sampled 80 transects, each 100 m long by 4 m wide, covering a total area of 3.2 ha. For each tree and shrub taller than 0.5 m within a transect, we noted its species, height and width of the canopy and browser impact classified as: (1) the percentage of branches browsed in four different categories (0 mm -10 mm, 10 mm -20 mm, 20 mm -50 mm, > 50 mm diameters) and (2) branches broken (> 10 mm), twigs browsed (diameter < 5 mm), bark peeled, branches stripped (the end of branches higher than 2 m stripped of leaves in a manner typical of giraffe), trees toppled (pushed over) and pollarded (snapping of the main stem). This method is adapted from previous elephant impact surveys, including Shannon et al. (2008). Except for toppling, only damage that was estimated to have occurred during the past year or two (i.e. scars were still fresh) was recorded. Despite measurements being taken by the same observer to avoid observer bias, actual damage may have been consistently underestimated or overestimated. Browsing measurements were taken to canopy heights of 2 m and 5 m for small herbivores to mesoherbivores (twig scars and branches < 10 mm diameter).
Transects were placed in six vegetation types (Table 2) excluding riparian forest (Faidherbia albida -Panicum maximum Closed Woodland) which was assessed by previous impact studies. The Vachellia (Acacia) stuhlmannii Open Shrubland vegetation type was underrepresented because it occurs on old farmlands currently dominated by two unpalatable species with negligible browsing impact.

Data analysis
Statistical analyses were performed using R (R Core Team 2012). We used lme4 (Bates et al. 2015) to perform linear mixed effects analyses of the relationship of various indicators of browsing impact, that is, broken branches, twigs browsed and stem damage (both pollarded and pushed over) with vegetation characteristics. Only vegetation types with at least three sites sampled were retained in the regression analyses, that is, Commiphora species -Sesamothamnus lugardii Closed Shrubland, Colophospermum mopane -Brachiaria deflexa Closed Woodland, Hexalobus monopetalus -Bulbostylis hispidula Open Shrubland and Senegalia (Acacia) senegal -Barleria sinensis Closed Shrubland. The entire site (all four transects) was amalgamated into one sample to avoid pseudoreplication during statistical analyses. As fixed effects, vegetation type, height and whether plants were single-stemmed or multistemmed were entered into the model and site was included as a random effect. The most parsimonious model was selected by first fitting the maximum model, removing nonsignificant interaction terms and comparing models using ANOVA (analysis of variance) and Akaike Information Criterion (Crawley 2012). As none of the fixed effects contributed significantly to explaining variability in bark stripping and trees pushed over, species was included as a random effect for these two categories.
We determined which of the most common species occurred on both ridges and flats (see Figure 1-A1 for examples of these landscapes) and tested for differences in the utilisation of two branch size classes (0 mm -10 mm and 20 mm -50 mm) for these species at the two localities using a Wilcoxon Signed-Rank Test. We assumed that smaller browsers were responsible for breaking 0 mm -10 mm branches, and elephants for breaking 20 mm -50 mm branches.

Results
We encountered a total of 1740 plants of 65 different species. Six species accounted for 52% of all the plants encountered (Table 1-A1). We excluded all species for which we sampled less than 10 individuals from the analyses, leaving a total of 34 species and 1609 individuals (listed in Table 2-A1). All species showed obvious signs of herbivore impact except for Mundulea sericea (cork-bush), Catophractes alexandri (gabbabos) and Rhigozum zambesiacum (mopane rhigozum). Evidence of twig browsing was found in 83% of plants, while 71% had broken branches, 4% had stripped branches, 6% had stripped bark, 3% had main stems broken or pollarded and 9% were toppled.
Only 11 species had less than 60% of trees with broken branches (Figure 2a). Many species, and Commiphora species and Sterculia rogersii (star-chestnut) in particular, experienced high rates of toppling (Figure 2b). Elephants and other browsers appeared to target very specific tree species for debarking, with only three species having more than 10% of bark damage (data not shown, Lannea schweinfurthii [false marula] 0.20, Sesamothamnus lugardii [sesame-bush] 0.75 and Sterculia rogersii 0.29). Although not confirmed, it appeared that the bark was removed by animals other than elephants for some of these species (e.g. eland or kudu for Sesamothamnus lugardii).
Twig browsing, or browsing impact attributable to animals other than elephants, was high for many species, with only nine species having less than 70% of individuals impacted (Figure 3a). Giraffe tended to target specific species  (Figure 3b).
All browsing indicators were uniformly high across vegetation types. More broken branches were encountered in the middle height class (2.5 m -5 m) than in other height classes and in multi-stemmed plants than in single-stemmed plants (see Table 3-A1 for results from glmer, Figure 4a and b). Although stemmedness overall did not affect twig browsing, in the 2.5 m -5 m height class, higher proportions of multistemmed than single-stemmed plants were browsed (Figure 5a and b). The probability of bark stripping and trees being pushed over could not be predicted by vegetation type, stemmedness or height, but some transects were more impacted than others (data not shown). Including species as a http://www.koedoe.co.za Open Access random effect significantly improved the fit of the model (Table 3-A1).
Twelve species commonly occurred in ridges and flats. Utilisation of 0 mm -10 mm branches was significantly higher (p < 0.0001, Figure 6a) on the ridges for these 12 species than it was in flat areas, but there was no difference between areas in the 20 mm -50 mm branch size class (p = 0.17, Figure 6b).   (2014) found that both mopane-dominated woodlands and Commiphoradominated woodlands were heavily utilised by elephants.

Discussion
Elephants have been shown to increase their nutrient intake by selecting browse from nutrient-rich patches (Pretorius et al. 2011). We therefore expected elephants and other browsers in Mapungubwe to have a lower impact on the mopane-dominated vegetation type associated with nutrientpoor Kalahari sands (Wigley, unpublished data). Surprisingly, this was not found to be the case as our results did not show this, with high levels of utilisation being evident across all vegetation and soil types.
Current, high utilisation levels of certain species by elephants in Mapungubwe are unlikely to be sustainable. Elephants have toppled a third of the combined populations of Commiphora glandulosa (tall firethorn corkwood), Commiphora mollis (velvetleaved corkwood) and Commiphora tenuipetiolata (satin-bark corkwood) and these, once toppled, are likely to die as these species' inability to resprout makes them vulnerable, with few Commiphora seedlings or saplings being found in our surveys. Elephants have been shown to be responsible for reducing Commiphora species in other studies. For example, a 67% reduction of a population of Commiphora ugogensis (river corkwood) was recorded over a 6-year period in Ruaha National Park, Kenya (Barnes 1985), while elephants also caused huge decreases in the cover of Commiphora woodlands in Tsavo National Park, Tanzania (Leuthold 1996)    Reserve, including Vachellia (Acacia) nilotica (scented-pod thorn), Vachellia (Acacia) tortilis (umbrella thorn), Albizia anthelmintica (worm-cure albizia), Cadaba termitaria (greyleaved wormbush), most Commiphora species, Ximenia americana (sourplum) and Ziziphus mucronata (buffalo-thorn). Vachellia (Acacia) xanthophloea woodlands recovered after elephants were fenced out in the semi-arid Amboseli National Park in Kenya (Western & Maitumo 2004). Similarly, Wigley et al. (2014) showed that population demographics of many tree species remained healthy when protected from elephants (but not fire) within enclosures in the KNP, South Africa.
We tested whether steep slopes and rocky outcrops were acting as refuges and recruitment hotspots for plants.
Despite the numerous potential areas in Mapungubwe that may harbour rare vegetation and plants, our results suggest that the vegetation growing on ridges compared with flat areas is equally or more heavily utilised by browsers.
Although we attempted to sample communities on ridges that were either difficult or impossible for elephants to access, we failed to find significant numbers of recruits or plants with low levels of herbivore impact in these areas. This may be because vegetation on the rocky ridges has more access to water than does vegetation on the flat areas during dry periods and thus appeals to many browsers. Vegetation on bedrock outcrops elsewhere has been shown to access rainwater stored in fissures and crevices and spring water (see Nie et al. 2011 and sources therein). Wall, Douglas-Hamilton and Vollrath (2006) showed that elephant densities were much lower on minor hills in northern Kenya and that they avoided steep hills altogether. Another Kenyan study showed that elephants avoided steep hills and high elevations in summer but preferred high elevations in winter when evergreen forests provide more food than do the deciduous shrub-dominated lowlands (Ngene et al. 2009). Broken branches in the larger 20 mm -50 mm branch class did not differ between vegetation in flat compared with ridged areas, which may indicate that elephant impact is less important than that of the smaller herbivores on ridges. Although

Conclusion
Whether the high rates of utilisation and associated changes in vegetation occurring at present are an artefact of higher historical recruitment rates when elephants were absent from this area or the result of a wetter climate remains an enigma. Elephants are also funnelled into Mapungubwe from Botswana and Zimbabwe because of increased hunting (including illegal hunting) of elephants and ongoing developmental activities in those countries. As suggested by O'Connor and Page (2014) and O'Connor (2010), elephants cannot co-exist with certain plant species. Our findings suggest that without the creation of botanical reserves that exclude elephant and other browsers, structural diversity, if not species diversity, is likely to be lost in Mapungubwe in the near future. Whether this is seen as a problem or simply as a return to a previous alternative state will depend on current and future management paradigms.