Elephants (Loxodonta africana (Blumenbach 1797)) can have a major transforming effect on savanna structure through felling, debarking or uprooting trees (Dublin et al. Reference DUBLIN, SINCLAIR and MCGLADE1990, Laws Reference LAWS1970, Mapaure & Campbell Reference MAPAURE and CAMPBELL2002). However, it is difficult to separate their influence from that of other causes of tree mortality, including wind storms (Spinage & Guinness Reference SPINAGE and GUINNESS1971), drought (Lewis Reference LEWIS1991, van de Vijver et al. Reference VAN DE VIJVER, FOLEY and OLFF1999), fire (Higgins et al. Reference HIGGINS, BOND and TROLLOPE2000), and in some situations frost (Childes & Walker Reference CHILDES and WALKER1987, Holdo Reference HOLDO2006), especially when interactions among them may occur (de Beer et al. Reference DE BEER, KILIAN, VERSVELD and VAN AARDE2006, Laws et al. Reference LAWS, PARKER and JOHNSTONE1975, Pienaar et al. Reference PIENAAR, VAN WYK and FAIRALL1966). Furthermore, the consequences for woodland dynamics depend on the size classes of the trees affected, as well as on how the disturbance is concentrated in time and space. Mortality of canopy trees has a much greater and longer-lasting impact than losses among the regenerating stages of these trees. However, the consequences may be less adverse for ecosystem function and biodiversity if the disturbing effects are locally concentrated, generating a patch mosaic of stands at different stages of regeneration (Remmert Reference REMMERT and Remmert1991).

The observations reported here were obtained fortuitously during a 3-y study aimed at assessing factors governing the selective impact of elephants on woody vegetation in northern Botswana. Routine recording showed that feeding by elephants was concentrated on shrubs or saplings of tree species at levels of utilization that appeared sustainable (Chafota Reference CHAFOTA2007). However, we encountered three situations where numerous big trees had been pushed over by elephants in particular localities. Each occurrence was evidently associated with a different causal influence, and hence unreplicated. Episodic severe damage of this nature is difficult to document within any study of limited duration. Hence we report the patterns observed as examples to be augmented by further observations.

Our study was conducted within a section of Chobe National Park adjoining the Chobe River together with neighbouring forest reserves and hunting concessions. The region is generally underlain by Kalahari sand, and the mean annual rainfall recorded in Kasane town in the north-east is 700 mm, falling mainly between November and March. Extremely high concentrations of elephants develop near the Chobe and Linyanti rivers during the dry season, amounting to 7–12 animals km⁻² (Craig Reference CRAIG and Hancock1990, Gibson et al. Reference GIBSON, CRAIG and MASOGO1998). Impacts by these animals have led to the transformation of much of the riparian woodland adjoining the Chobe River into a shrubland with standing dead trunks of the former trees remaining (Mosugelo et al. Reference MOSUGELO, MORE, RINGROSE and NELLEMANN2002), and a similar process of woodland conversion was apparently underway in the riparian woodland adjoining the Linyanti River further west (Wackernagel Reference WACKERNAGEL1993).

The first case, associated with the influence of fire, occurred about 30 km south of the Chobe River, on the border between Chobe National Park and Kasane Forest Reserve Extension (location 25°6′E, 18°9′S). Severe tree damage on the forest reserve side of the boundary was discovered in the second week of November 1992, when we tracked movements of elephants away from the river after rain showers had filled pools in the interior. An extensive fire had burned through the forest reserve during the last week of September, but was stopped from penetrating the national park by a fire-break marking the boundary.

The area within the forest reserve where tree felling was evident was sampled within 22 belt transects each 10 m wide × 400 m long located about 200 m apart. Ten transects were sampled similarly on the unburned national park side of the border. All woody plants taller than 2.5 m with a main stem diameter > 5 cm at breast height were assessed for frequency of occurrence of recent uprooting, trunk snapping, or bark removal amounting to at least 50% of trunk circumference. About one third of all Brachystegia boehmii and one quarter of all Burkea africana trees with trunk diameters > 10 cm had been felled (either uprooted or with the main stem snapped) in the burned area, while substantially lower levels of such damage were recorded on the unburned side of the border (Table 1; tree nomenclature follows Gibbs Russell et al. Reference GIBBS RUSSELL, GERMISHUIZEN, HERMAN, OLIVIER, PEROLD, REID, RETIEF, SMOOK, VAN ROOY and WELMAN1984). From the size of the associated tracks, most of the tree felling had been done by female elephants.

Table 1. Severe damage to canopy trees inflicted by elephants in the Kasane Forest Extension during November 1992 following a fire.

A second observation, related to frost damage, was made following extremely low minimum temperatures recorded in Kasane in June (2 °C) and July (1 °C) 1994. Local felling of canopy trees was noticed in July 1994 while driving through the Kazuma Forest Reserve (location 25°31′E, 18°25′S). Nearby pans held pools of water, enabling some male elephants to remain in this region through the dry season. Recent severe damage inflicted by elephants was recorded in October 1994 within 11 belt transects 10 m × 400 m placed systematically to the west of the tarred road, plus another five transects east of the road. Frost damage to shrubs was recorded within randomly located patches of 10-m radius along each transect, noting the proportion of terminal stems of each shrub that appeared dead as a result of the frost.

Within the area sampled, over 50% of all B. africana and B. boehmii trees and over 40% of Terminalia sericea trees had recently been felled (Table 2). From 65% to 85% of the terminal stems of the majority of woody species in the shrub layer had been killed by frost, but with less damage recorded for Combretum apiculatum (39%), Colophospermum mopane (16%), T. sericea (7%) and B. africana (< 1%).

Table 2. Severe damage to canopy trees inflicted by elephants in the Kazuma Forest Reserve following frosts during the dry season of 1994.

A third instance of localized severe damage to trees was associated with unusually low rainfall through March–April 1993, causing elephants to concentrate near the Linyanti River earlier than usual in the dry season. Observations were made in August 1993 in an area extending westwards of Linyanti Safari Camp (23°35′–23°47′E, 18°28′S). Within the riparian woodland, recent severe damage by elephants to three relatively common tree species was recorded within 20 belt transects measuring 10 m × 400 m, placed systematically 1 km apart perpendicular to the river. A further four sites were sampled within the adjoining mopane woodland where concentrated tree felling was noted. At each site observations were made within five 400 m × 15-m belt transects placed systematically 100 m apart. The occurrence of recent severe damage was recorded for each tree > 5 cm in stem diameter.

In the riparian woodland, 79% of Acacia nigrescens trees, 60% of Acacia erioloba trees and 61% of Terminalia prunioides trees showed recent severe bark removal, but no recent felling was recorded in the transect sample because most trees were too large to be easily toppled. Nevertheless, we observed 11 trees being felled by male elephants over 6 d during September 1993. The felled trees included Acacia luederitzii, Berchemia discolor, Boscia albitrunca, C. mopane, Peltophorum africanum and T. prunioides, with stem diameters of 32–57 cm. Within local sites of heavy impact in the mopane woodland, over 50% of all C. mopane trees had been felled, including 25% of those 40–50 cm in stem diameter.

In all three situations, damage likely to result in the death of the tree affected 25–80% of canopy trees, and was inflicted within 1–10 wk. In savanna woodlands with elephants, rates of large tree mortality amounting to 1.5–6% y⁻¹ have been sustained over periods of a decade or longer (Buechner & Dawkins Reference BUECHNER and DAWKINS1961, Croze Reference CROZE1974, Tafangenyasha Reference TAFANGENYASHA1997, van de Vijver et al. Reference VAN DE VIJVER, FOLEY and OLFF1999), reaching as high as 12–20% y⁻¹ in specific localities (Buechner & Dawkins Reference BUECHNER and DAWKINS1961, Field Reference FIELD1971). Although our study sites showed much greater tree mortality being imposed within less than a year, they represented small sections of the regional landscape.

The burn that occurred in the Kasane Forest Extension was the only instance of a fire during the 3-y duration of our study. We did not observe any other occurrences of canopy tree felling or frost damage either within our study area or along the road to Kasane which we drove repeatedly.

The common feature in two cases was a reduction in forage available within the shrub layer, which is normally favoured by elephants for feeding (Chafota Reference CHAFOTA2007). Fire frequently kills much of the above-ground material of plants 2.5 m or less in height, but has little direct effect on larger trees, except in conjunction with bark removal by elephants (Laws et al. Reference LAWS, PARKER and JOHNSTONE1975) or porcupines (Yeaton Reference YEATON1988). Pienaar et al. (Reference PIENAAR, VAN WYK and FAIRALL1966) noted that elephants may push over trees after a fire before leaving the area, while Laws (Reference LAWS1970) suggested that the combined influences of elephants and fire on the shrub layer could lead to increased ring-barking of mature trees by elephants. Top-kill of shrubs by frost has been documented elsewhere in this region of south-central Africa (Childes & Walker Reference CHILDES and WALKER1987, Holdo Reference HOLDO2006). In our study site, around 50% of the shoots that would otherwise have supplied forage for the elephants through the late dry season were eliminated. Nevertheless, some of the toppled trees showed no signs of feeding by elephants on the branch tips brought within reach.

In the Linyanti region, the early cessation of rain probably caused pools of water in the mopane-dominated hinterland to dry out earlier than usual. The consequently prolonged dependency of elephants on access to water from the river exacerbated the depletion of forage in the shrub layer of the riparian woodland. Heightened elephant impacts on woody plants near water during years with low rainfall have been recorded at Etosha in northern Namibia (de Beer et al. Reference DE BEER, KILIAN, VERSVELD and VAN AARDE2006). In our study site, much of the tree damage took the form of bark removal, and feeding occurred on all of the trees seen to be felled, indicating that food shortage rather than social display was responsible.

Localities where damage sufficient to cause the death of many trees had been inflicted by elephants, apparently within a brief period, are sometimes noted (Bell Reference BELL2003, Owen-Smith pers. obs.), but with the cause remaining uncertain. Suggestions that these occurrences could in some cases result from social interactions among male elephants (Hendrichs Reference HENDRICHS and Hendrichs1971, Midgley et al. Reference MIDGLEY, BALFOUR and KERLEY2005) remain unsubstantiated. Our findings suggest that in some cases these events could arise via interactions with physical agents of disturbance, drastically reducing food availability in the shrub layer normally favoured by elephants. Holdo (Reference HOLDO2006) found that prior damage by frost had no consistent influence on the likelihood of these shrubs incurring subsequent damage from elephants, because species differed in their relative susceptibility to these agents. Furthermore, previously damaged stems were more likely to survive following new breakage by elephants than undamaged stems. However, the consequences of leaf or stem damage following frosts for the surrounding canopy trees were not reported. A simulation model developed using these findings suggested that elephants are the primary drivers of community structure and composition in Kalahari sand woodlands, and in interaction with fire and frost could promote the conversion of woodland into a shrub-dominated state (Holdo Reference HOLDO2007).

However, neither the model developed by Holdo (Reference HOLDO2007) nor the approach by Baxter & Getz (Reference BAXTER and GETZ2008) emphasizing elephant effects interacting with stochastic fire and rainfall conditions addressed consequences of spatial and temporal heterogeneity in these processes. Instead of leading to a progressive decline in canopy trees, lethal disturbances to canopy trees that are spatially concentrated while sufficiently widely spaced in time could promote a mosaic of patches in different stages of recovery, with rather different consequences for ecosystem diversity and function (Gilson Reference GILSON2004). Our observations suggest some mechanisms whereby a mosaic cycle of savanna woodland dynamics might be developed with elephants as the prime disturbing agent. A fundamental template for these interactions is provided by the distribution of surface water, constraining the regions of the broader landscape that elephants can occupy in high local densities under different conditions (Chammaille-Jammes et al. Reference CHAMMAILLE-JAMMES, FRITZ, VALEIX, MURINDAGOMO and CLOBERT2008). Further case studies are needed to establish how widely prevalent the episodic lethal disturbance to local stands of canopy trees that we apparently observed might be. Under certain conditions, the consequence could be greater regional diversity in woodland structure and composition than would prevail in the absence of elephants as a disturbing agent.