1Top-down regulation of ecosystems by large herbivores is a topic of active debate between scientists and managers, and a prime example is the interaction between elephants Loxodonta africana and trees in African savannas. A common assumption among wildlife managers is that a local reduction in elephant numbers will ultimately allow woodland to self-restore to a desired former state. Such regeneration is, however, dependent on the survival of seedlings of impacted tree species. We conducted a field experiment to investigate seedling predation in the elephant-transformed Chobe riparian woodland of northern Botswana.
2We planted seedling gardens in (i) complete exclosures that excluded all herbivores except small rodents and invertebrates, (ii) semi-permeable exclosures that excluded ungulates but included primates, lagomorphs, all rodents, gallinaceous birds, etc, and (iii) completely open plots. Seedlings were of two tree species decreasing in the area (Faidherbia albida and Garcinia livingstonei) and two that are increasing (Combretum mossambicense and Croton megalobotrys).
3After 9 months, seedling survival ranged from >75% for all species in the complete exclosure to <20% for Faidherbia albida in the open plots. Survival of all seedlings except C. megalobotrys declined precipitously in open plots during the dry season when invertebrates are largely dormant but when impalas Aepyceros melampus (locally abundant ungulates) increase the browse components of their diets.
4Seedling survival in the open plots was negatively related to local impala density but unrelated to that of any other browser.
5Synthesis and applications. Our findings relate to the current debate about managing elephants to restore southern African savanna landscapes to desired historical states. Various seedling predators, including the ubiquitous impala Aepyceros melampus, regulate the regeneration of trees from seedlings, and our experiments support the hypothesis that tall closed-canopy woodlands originate during episodic windows of opportunity for seedling survival. To artificially recreate such a window would require the decimation of seedling predators as well as elephants, which is impractical at the landscape scale.
Savanna systems are characterized by a mixture of trees and grasses where the relative dominance of woody cover is determined by abiotic factors like water (Amundson, Ali & Belsky 1995), nutrients (Walker & Langridge 1997) and fire (Trollope 1984), and biotic factors like herbivory (Scholes & Archer 1996). However, when high densities of ungulates consume large proportions of the biomass, fire becomes less important in determining the balance between woodland and grassland (McNaughton, Ruess & Seagle 1988). African elephants Loxodonta africana Blumenbach are usually regarded as the main factor determining savanna woody cover, because of its very large body size enabling elephants to kill mature trees (Owen-Smith 1988; Dublin 1995). However, the local persistence of woody species is not only dependent on the survival of mature individuals, but also on recruitment from germinating seedlings. A recent study from East Africa has shown that elephants exert minimal browsing damage to acacias in smaller size classes (<2·5 m) (Augustine & McNaughton 2004), yet low recruitment of the seedlings of certain woody species is a major factor affecting the structure and composition of African savanna woodlands (e.g. Laws 1970; Pellew 1984; Dublin, Sinclair & McGlade 1990; Ruess and Halter 1990). While elephants generally browse on trees, ungulates commonly select seedlings (Hobbs 1996; Augustine & McNaughton 1998) because they are more palatable than older plants (Crawley 1983), and browsing by medium-sized ungulates has been suggested as the controller of woodland regeneration in East African savannas (e.g. Belsky 1984; Mwalyosi 1990; Prins and van der Jeugd 1993; Sinclair 1995). In Serengeti National Park, Tanzania, Belsky (1984) found that browsing antelopes such as impala Aepyceros melampus Lichtenstein, Grant's gazelle Gazella granti Brooke, Thomson's gazelle G. thomsoni Günther, eland Taurotragus oryx Pallas and dikdik Madoqua kirkii Günther controlled woody plants <1 m in height. Prins & van der Jeugd (1993) showed that even-aged stands of acacia woodland observed in Lake Manyara National Park, Tanzania, were established during years of low impala populations.
While previous studies have debated the effects of elephants and smaller ungulates on woody cover, recent experimental studies have pointed to invertebrates (Shaw, Keesing & Ostfeld 2002) and rodents (Goheen et al. 2004) as important acacia seedling predators in central Kenya. Indeed, Goheen et al. (2004) found that the net seedling survival was approximately twice as high in areas where large mammals were present compared to excluded areas in central Kenya, which is clearly contrary to the findings of other studies. These increases in seedling survival in areas with large mammals were attributed to the increase of rodents and invertebrates in plots excluding large mammals (Goheen et al. 2004).
The riparian habitat of the Chobe riverfront (Chobe National Park, northern Botswana) has been gradually fragmented by elephants over the past half century. At present, only isolated remnants of the once continuous belt of tall mature trees remain (Mosugelo et al. 2002). The previously dominant riparian species are disappearing and being replaced by previously minor species (Simpson 1975; Moroka 1984; Rutina 2004). A well-developed continuous belt of tall mixed gallery woodland previously dominated by what are now ‘decreasers’–Faidherbia albida (Delile) A. Chev., Garcinia livingstonei T. Anderson, Diospyros mespiliformis Hochst. ex A.DC, Combretum imberbe Wawra, Acacia nigrescens Oliv., A. tortilis (Forssk.) Hayne and Philenoptera violacea (Klotze) Schrire (Child 1968) – has been replaced by shrubs and thickets dominated by what are now ‘increasers’–Croton megalobotrys Pax, Capparis tomentosa Lam., and Combretum mossambicense (Klotzsch) Engl. (Simpson 1975, 1978; Sommerlatte 1976; Moroka 1984; Addy 1993). Most of the decreasers show little signs of regeneration over the past 30–40 years (Sommerlatte 1976; Moroka 1984; Rutina 2004), suggesting either lack of germinable seeds or low seedling survival.
The Chobe River represents the only permanent supply of drinking water for wildlife in >10 000 km2 of dystrophic savanna, with the result that exceptionally high ungulate biomass densities occur in and around the Chobe floodplain in the dry season (Skarpe et al. 2004). With the locally high ungulate biomass, caused by the attraction of drinking water, there is a sparse herbaceous layer, and consequently, the fire frequency is low (Skarpe et al. 2004), which suggests the lack of regeneration by decreasers is not attributable to fire (cf. Dublin, Sinclair & McGlade 1990). Elephants have increased at 6% per annum in this area since 1987 (DWNP 1997; Gibson, Craig & Masogo 1998), and because it is well known that elephants can change mature woodland into shrubland and grassland (Caughley 1976; Dublin, Sinclair & McGlade 1990; Cumming et al. 1997), previous studies in Chobe have focused on elephants as the main agent of change (Child 1968; Simpson 1975, 1978; Sommerlatte 1976; Moroka 1984; Mosugelo et al. 2002). However, while elephants have probably been the main cause of mortality in mature trees, a high elephant population does not explain the virtual absence of seedlings of decreasers. This is supported by a previous study in Chobe that found elephants to browse predominantly in the 1–3 m height zone, which is well above the level of seedlings (Stokke and du Toit 2000).
Apart from elephants, most other ungulate populations have been either declining or stable along the Chobe Riverfront during the past decades (DWNP 1997). One exception is the impala population that has increased substantially since the 1960s (Sheppe & Haas 1976; Rutina 2004), and some previous studies have either suggested (e.g. Prins & Van Der Jeugd 1993) or experimentally shown (Sharam, Sinclair & Turkington 2006) that browsing antelopes are important regulators of woodland regeneration.
In view of the previously contradictory findings on the effects of large mammals on seedling survival, we aimed to investigate the impact of seedling predation on woody plant regeneration along the Chobe riverfront. In addition to identifying the main herbivores involved in seedling predation, we also tested whether seedling predation was higher on those woody species decreasing than on those increasing in abundance in woodlands heavily impacted by elephants. In essence, our study examined the reversibility of an elephant-driven shift in woody plant community structure in an ecosystem characterized by low fire frequency but high seedling predation.
Materials and methods
The study was conducted within Chobe National Park in northern Botswana. The study area was a 20-km strip (~40–70 m wide) along the southern bank of the Chobe River. The study area is part of the vast (2·5 million ha) nutrient-poor Kalahari sand system extending into Angola, Namibia, South Africa, Zambia and Zimbabwe (Scholes et al. 2002). The key determinants of the Kalahari sand vegetation are rainfall, nutrients, fire and herbivory (Scholes & Walker 1993). Rainfall, generally increasing (from 200 to 1000 mm) from south to north in the Kalahari system, is intermediate in Chobe with an annual rainfall of ~600 mm while mean minimum and maximum monthly temperature range is 4–30 ºC in July and 14–39 ºC in October (Botswana Meteorological Service Department, unpublished data). Soils in the study area can be grouped into two distinct classes, the nutrient-poor Kalahari sand dominating the Kalahari sand system and a more nutrient-rich alluvial or alkaline soil in areas earlier covered by floodplain (Simpson 1975). The rainy season lasts from November to May and most plant growth takes place during this time (Omphile 1997). A detailed description of the study system is given in Skarpe et al. (2004).
A recent study has shown that the area along the Chobe floodplain was covered with a continuous belt of riparian woodland as late as in 1962 (Mosugelo et al. 2002). The riparian woodland is now fragmented with only a few isolated remnants where the gradient down to the floodplain is too steep for elephants to negotiate or in areas with high human disturbance. Woody species composition and structure have changed from a dominance of large Acacia and Combretum trees and a dense shrub layer to Capparis and Combretum shrubland (Simpson 1975, 1978; Moroka 1984; Addy 1993; Rutina 2004), with low recruitment in most of the previously dominant woody species (Moroka 1984; Rutina 2004).
During the rainy season, most ungulates move into the woodland dominated by Zambezi teak Baikiaea plurijuga Harms, where they drink seasonal water located in pools and pans (Child 1968), but they congregate along the Chobe River during the dry season. Apart from elephants, giraffe Giraffa camelopardalis L., greater kudu Tragelaphus strepsiceros Pallas and impala, common ungulate species in the area include hippopotamus Hippopotamus amphibious L., warthog Phacochoerus africanus Pallas, puku Kobus vardoni Livingstone, roan Hippotragus equinus Desmarest, sable H. niger Harris and waterbuck Kobus ellipsiprymnus Ogilby. In addition to the more conspicuous large mammals, several smaller herbivore species of primates, gallinaceous birds, lagomorphs and large rodents (notably the spring hare Pedetes capensis Forster), are present in the area.
experimental design and sampling
A total of 12 experimental sites (each 20 × 30 m) were selected along the Chobe riparian zone. Six sites were located on areas heavily impacted by elephants (where the original woodland had disappeared) and six sites in remnants of the original woodland (henceforth termed high and low impact areas, respectively). On high impact areas, three sites were located on nutrient-poor Kalahari sand and three on the nutrient-rich alkaline soil. Similarly, on the low impact areas, three sites were located on nutrient-poor Kalahari sand and three on the nutrient-rich alkaline soil. At each site, three different temporary plots were randomly positioned: (i) complete exclosure, fenced on all four sides (2 × 1·2 m) and on top (1·5 m above-ground) with diamond-mesh wire (5 cm diameter mesh), allowing only insects and small rodents to enter the fenced area; (ii) semi-permeable exclosure with diamond mesh on all sides and on top like the complete exclosure but with the sides open up to a height of 0·5 m above-ground, allowing access to gallinaceous birds, primates, lagomorphs and large rodents, but excluding ungulates; (iii) open plots, with no fencing at all.
After removing the existing woody vegetation in each plot, seedlings (grown in a local nursery for 2 weeks after germination) of four woody species were planted; two species that have increased in abundance in the area –Combretum mossambicense and Croton megalobotrys– and two species that have decreased –Faidherbia albida and Garcinia livingstonei. Seedlings were planted in February–March 2001. Each species was replicated five times per treatment for all plots, apart from C. megalobotrys, which was only replicated three times per treatment and planted in nine plots, due to an inadequate stock of seedlings. Within each plot, 20 positions (4 species × 5 replicates) were evenly distributed and seedlings were randomly allocated to a position in the plot.
The number of surviving seedlings for each species was monitored monthly in each plot until July and then in September and November 2001. Ungulate stem browsing (seedling shoot torn leaving a rough stump) was recorded on surviving seedlings.
In November 2001, 9 months after planting, all surviving seedlings were counted, checked, and uprooted from the experimental plots. Since the plots were small, they could be cleared of all seedlings. Then, to increase the robustness of our results, we monitored naturally occurring seedling establishment (from the seed bank or seed rain) in the three treatments and compared these results to those from the planted seedlings. All vegetation in the plots was manually uprooted and the plots were left for 15 months, after which all plots were carefully inspected and all naturally regenerating seedlings of woody plants were counted and removed for herbarium identification.
To test for relationships between seedling predation and ungulate abundance, fixed-width (100 m) transect surveys of all ungulate species were conducted along the riverfront three times a month from June 2000 to July 2001, with a transect segment corresponding with each experimental site. Successive surveys of each transect were conducted at different times of the day (early morning, noon, late afternoon) during each month. At each ungulate sighting, the species, number of animals, GPS coordinates, and odometer distance from the transect starting point were recorded.
For each treatment on each of the 12 sites, the number of dead and living seedlings was monitored as described above. For analysis of survival, we used the lmer function which is part of the lme4 library in the statistical package r (http://cran.r-project.org/). This linear mixed-effect model assumes binomial errors (with the logit link function) (Crawley 2007). We used the combined variable of seedlings alive and dead in each plot in each month as the binomial response variable (Crawley 2007). To predict the seedling survival, we used a repeated measure analysis with month as the repeated random variable, and treatment (complete exclosures, semi-permeable exclosures, open plots), species (Combretum mossambicense, Croton megalobotrys, Faidherbia albida, Garcinia livingstonei), elephant impact (high, low) and soil (alkaline, Kalahari sand) as fixed effects. When selecting the most parsimonious model, one of the categorical variables is used as a reference. The references used for treatment, species, elephant impact and soil type in our model were: complete exclosure, Combretum mossambicense, high elephant impact, and alkaline soil.
The most parsimonious model was selected by first fitting the maximum model, testing for overdispersion, removing non-significant interaction terms and comparing models using anova (analysis of variance) and Akaike Information Criterion (Crawley 2007). Only first-order interactions were included in the model.
Mean local population densities of the most common browsers (elephants, impalas, kudus) in the area of each experimental site were calculated from the fixed-width transect data. A linear regression was performed to test for relationships between seedling survival and local population densities (in November, at the transitional between dry and wet seasons) of elephant, impala and kudu.
Naturally regenerating seedlings of woody plants collected from the cleared plots after 15 months were identified to species and then pooled into either ‘increaser’ or ‘decreaser’ categories, based on whether the species is characteristic of the transformed shrubland (increaser) or the disappearing tall-canopy woodland (decreaser). Total numbers of increaser or decreaser seedlings were counted across all experimental sites within each treatment type (complete exclosure, semi-permeable exclosure, or open plot). Frequency distributions of increasers and decreasers were compared across treatments using a 3 × 2 contingency table and the chi-square statistic.
The seedling survival (mean ± SE) was highest in the complete exclosures (0·78 ± 0·04), followed by the semi-permeable exclosures (0·65 ± 0·04) and the open plots (0·32 ± 0·04) (Table 1 and Fig. 1). Faidherbia albida had substantially lower survival in the open plots compared with the other species while the survival of Croton megalobotrys was relatively high (Table 1 and Fig. 1). The survival of F. albida seedlings in the open plots was found to be significantly low when analysed with comparison to the references (complete exclosure and Combretum mossambicense) (Table 1 and Fig. 1).
Table 1. The most parsimonious model (d.f. = 27) for the effects of treatment (complete exclosure (reference in the model), open plots and semi-permeable exclosure), species (Combretum mossambicense (reference in the model), Croton megalobotrys, Faidherbia albida and Garcinia livingstonei) on seedling survival along the Chobe floodplains in northern Botswana. Elephant impact (high and low) and soil type (alkaline soil and Kalahari sand soil) did not improve the model
Open plots (Open)
Semi-permeable exclosures (Semi)
Croton megalobotrys (Cromeg)
Faidherbia albida (Faialb)
Garcinia livingstonei (Garliv)
Open × Cromeg
Semi × Cromeg
Open × Faialb
Semi × Faialb
Open × Garliv
Semi × Garliv
The proportion of seedlings surviving in the open plots was strongly negatively related to local impala population density (linear regression, R2 = 0·59, P = 0·004) (Fig. 2). Neither kudu nor elephant densities were negatively related with seedling survival (P > 0·05).
The proportion of surviving seedlings that were twig-browsed peaked in June–September and Croton megalobotrys was browsed significantly less than the other species (LSD, P < 0·05 in all cases), except in September (Fig. 3). F. albida was browsed more than the other species in April and May (Fig. 3). July had the highest proportion of browsed seedling for Combretum mossambicense, F. albida and Garcinia livingstonei (Fig. 3).
natural regeneration of seedlings
After 15 months following the termination of the seedling garden experiment and complete clearing of all vegetation from all plots, we found that seedlings of eight woody species were naturally regenerating in complete exclosures, five in the semi-permeable exclosures and four in the open plots (Table 2). As expected, we found a significant difference between the frequency distributions of increaser and decreaser seedlings across experimental treatments, with species dominance shifting from decreasers in the complete exclosures to increasers in the open and semi-permeable plots (χ2 = 14·6, d.f. = 2, P < 0·001).
Table 2. Seedlings of woody plants sprouting from naturally introduced seeds within experimental plots that had been cleared of all vegetation and then left for 15 months (December 2001–March 2003) at 12 sites along the Chobe riverfront. At each site, there were three plots (2·4 m2), each with a separate treatment: complete exclosure from large herbivores; semi-permeable exclosure (open for 0·5 m from ground-level); open plots (no exclosure)
Frequency of occurrence per treatment
Total (no. of species represented)
Our results demonstrate that ungulate browsing reduces woody seedling survival substantially along the Chobe riverfront, supporting what has been inferred from previous studies elsewhere in Africa (Belsky 1984; Pellew 1984; Prins & van der Jeugd 1993; Dublin 1995; Sinclair 1995; van de Vijver, Foley & Olff 1999; Barnes 2001). Nine months after planting, mean seedling survival was as high as 78% in the complete exclosures compared with only 32% in the open plots, indicating that damage by insects and small mammals contributes comparatively little to seedling mortality in this area. Results from a similar experimental study on Acacia drepanolobium Sjost. seedlings in Kenya (Shaw, Keesing & Ostfeld 2002) concluded that invertebrates are the most important mortality factors in that area, while a later study from the same area (Goheen et al. 2004) found that rodents also played an important role in seedling predation. Although the planted seedlings in the complete exclosures suffered higher mortality compared to the open plots, there was no difference between semi-permeable exclosures and open plots in seedlings surviving after the natural regeneration experiment. Thus, primates, lagomorphs and large rodents may also play a role in consuming seeds and young seedlings of the decreaser species. However, since invertebrates were not important agents of seedling mortality in our study area, it is apparent that the primary agents of seedling predation vary across African savanna ecosystems presumably because of variation in the compositions, and interactions, of woody plant and herbivore communities. Also, while comparatively unimportant at the seedling stage in our study, we do not discount the potentially significant effect that invertebrates have as functional predators of dormant seeds (Lewis & Gripenberg 2008).
Browsing by ungulates was particularly important in Chobe in the dry season (Figs 1 and 3), when the availability of green grass was low. In May, at the start of the dry season, shoot browsing by medium-sized ungulates started to increase in the open plots. The proportion of seedlings browsed by ungulates was particularly high towards the end of dry season from June to September. Seedling predation was relatively low in the rainy season (November–May), when invertebrates may have been potentially important. Rignous & Young (2007) have shown that grass cover may also affect woody species both positively and negatively in that browsing damage to saplings is far higher in areas were grass is removed, but in areas of grass removal, saplings grow to more than twice the size of control trees.
Faidherbia albida had considerably lower survival rates in semi-permeable exclosures, compared with complete exclosures, indicating that gallinaceous birds, primates, lagomorphs and large rodents probably contribute to seedling mortality in this species. Seedling survival was considerably reduced for all species in the open plots, thus pointing to the importance of large mammals in seedling predation. Both Combretum mossambicense and Croton megalobotrys are increasing in the area. However, only C. megalobotrys had a relatively high proportion of surviving seedlings in the open plots (53% vs. 29%, 16%, and 29% for C. mossambicense, F. albida and Garcinia livingstonei, respectively) in November. Similarly, C. megalobotrys was the least-browsed species in the experiment (Fig. 3). The poor utilization of C. megalobotrys has also been shown by other studies concluding that despite its high nutritive values, this species has shoot diameters that exceeds common bite size of ungulate browsers (Makhabu et al. 2006).
It was not possible to quantify the relative contribution of each large browser species to seedling mortality. However, the relationship between local impala density and seedling mortality along the riverfront (Fig. 2) strongly suggests that impala is the main species involved, since no such relationship was found with the other common browsers (kudu and elephant). The impala population along the Chobe riverfront has expanded rapidly during the past few decades; from being a rare species during the early 1970s (Sheppe and Haas 1976), impalas now reach local densities of around 52 animals km−2 (Rutina 2004). This rapid increase could be attributed to changes in vegetation caused largely by elephants, with shrubland replacing gallery forest and thus increasing browse availability to impala (Mosugelo et al. 2002). A previous study from the same area (Rutina, Moe & Swenson 2005) found that impalas feed in Capparis and Combretum shrubland while Baikiaea woodland is avoided. The preferred shrubland habitats have increased while the woodland has receded (Mosugelo et al. 2002).
Results from the exclosure experiment in which we monitored natural regeneration from the seed bank were conclusive in demonstrating that ‘decreasers’ will readily regenerate in the absence of vertebrate herbivores (Table 2). While seedlings of both functional types suffer high mortality in the presence of browsing, the ‘increasers’ survive relatively better and dominate over ‘decreasers’ in these conditions. We suggest that these results support the hypothesis that the riparian forest developed along the Chobe floodplain from a pulse of seedling regeneration during and shortly after the period when rinderpest decimated the artiodactyl populations around the end of the 19th century (Walker 1986; Skarpe et al. 2004).
When testing whether the thickets in Serengeti establish simultaneously due to favourable conditions (low fire frequency, low browsing impact and low grass density), or if thicket species enter as a pioneer species and then promote forest succession, Sharam, Sinclair & Turkington (2006) drew conclusions similar to this study in that the present thickets in Serengeti was a result of the rinderpest epizootic during the 1890–1920s. Present conditions in Serengeti with high fire frequency, high browsing and dense grass cover have not permitted the re-establishment of thickets (Sharam, Sinclair & Turkington 2006).
conclusions and management implications
The disappearing riparian woodland along the Chobe riverfront (Mosugelo et al. 2002) has received considerable media attention during the past decade, and stakeholders frequently call for elephant culling to restore the once-spectacular woodlands. However, the management implication of our study is that even severe culling of elephants will not restore the system to its mid-20th century state. As long as the impala density remains high in the area, few seedlings, especially of key riparian species like F. albida, will be able to regenerate. The local impala density has increased dramatically (cf. Sheppe & Haas 1976) following the change from tall dense woodland to shrubland dominated by Combretum mossambicense, Croton megalobotrys and Capparis tomentosa.
Companion studies to ours have shown that the tall closed-canopy riparian forest along the Chobe riverfront was a temporary unstable state that developed following a combination of intensive ivory hunting ~150 years ago and the rinderpest pandemic ~100 years ago (Skarpe et al. 2004). The return of the system to the shrubland state that must have prevailed in the 19th century, when elephant densities were very high (Campbell 1990), provides an example of how a complex adaptive system can self-restore as long as there is no major loss of biodiversity (Walker, Kinzig & Langridge 1999; Folke et al. 2004; Walker et al. 2004). Stakeholders now have the unique opportunity of adjusting their value systems to recognize the multiple ecotourism opportunities to be gained from a locally high density of elephants within a vast wildlife area that is not (yet) constrained by fencing or human settlement patterns.
This study was conducted as part of the Botswana–Norway Institutional Co-operation and Capacity Building Project (BONIC), jointly funded by the Department of Wildlife and National Parks (DWNP), Botswana and NORAD. We highly appreciate the constructive comments of Jacob R. Goheen and an anonymous referee on earlier drafts of this manuscript.