The study was conducted in three savanna habitats, Mopane woodland (Mopane area), Acacia savanna (Satara area) and Terminalia woodland (Pretoriuskop area). Ant sampling was carried out on experimental burn plots that form part of the long-term burning experiment. Although there were four replicates within each vegetation type, because the fourth replicate differed in soil type from the other three it was considered unrepresentative (Venter 1999). Therefore, this study only focused on three replicates, each situated 10–20 km apart. The replicates in the Mopane area were Tsende (23°41′S 31°31′E), Mooiplaas (23°34′S 31°27′E) and Dzombo (23°26′S 31°22′E); in the Satara area N’wanetsi (24°26′S 31°51′E), Marheya (24°32′S 31°46′E) and Satara (24°24′S 31°45′E); and in the Pretoriuskop area Kambeni (25°15′S 31°26′E), Numbi (25°13′S 31°20′E) and Shambeni (25°12′S 31°23′E). Each replicate was divided into 12–14 plots laid out in a strip. Each plot measured approximately 380 × 180 m (7 ha) and represented a different burning regime (season and frequency combination; Trollope et al. 1998a). Fire-break roads separated the plots in each replicate. In the Mopane and Satara areas, sampling was carried out on the following burn plot treatments: August annual, August biennial, August triennial, April biennial, April triennial and, importantly, an unburnt control that had remained unburnt since 1954. The exception to this was one of the three Satara area control plots, which was accidentally burnt in April 2001. August burns represent late winter burns at the end of the dry season, while April burns represent autumn burns at the end of the wet season. In the Pretoriuskop area, because of sampling constraints, ants were sampled on two burn plot treatments only: the unburnt control plot and an August annual plot.
Additional sampling was also carried out adjacent to each replicate in the general landscape to assess the effect of the general burning regime (usually spring fires) that had been applied by Kruger National Park management; these were referred to as ‘variable’ plots but were not considered controls. They were neither prevented from burning nor subjected to a specific experimental burning regime. Fire records for the Mopane and Satara areas (Kruger National Park Scientific Services, unpublished data) indicated that the variable plots were last burnt at least 4 years prior to sampling and had burnt very infrequently (i.e. twice in the last 40 years). Two variable plots (adjacent to the Dzombo and Satara replicates) were older than 20 years. The variable plot for the N’wanetsi replicate burnt in 2001.
The Mopane plots were situated in the Mopane shrubveld, a mopane (Colophospermum mopane J. Kirk ex J. Léonard)-dominated habitat with few other woody species (Low & Rebelo 1996). The altitude of the area ranges from 300 to 340 m a.s.l., and mean annual rainfall from 450 to 500 mm (Gertenbach 1983). The Satara plots were situated in a mixed knobthorn Acacia nigrescens Oliv. and marula Sclerocarya birrea (A. Rich. Hochst.) savanna in the sweet lowveld bushveld (Low & Rebelo 1996). Mean annual rainfall here is 550 mm, and altitude ranges from 240 to 320 m a.s.l. (Gertenbach 1983). Satara and Mopane areas share basalt-derived clay soils (Gertenbach 1983). The vegetation of the Pretoriuskop area is sour lowveld bushveld (Low & Rebelo 1996). This is an open tree savanna dominated by silver clusterleaf Terminalia sericea Burch. ex DC but also with bushwillow Combretum collinum Fresen. Soils are sandy and granitic-derived. The area has a mean annual rainfall of 700 mm and altitude ranges from 560 to 640 m a.s.l. (Gertenbach 1983).
Epigaeic ants were collected by pitfall trapping during two sampling periods: November–December 2000, and January–February 2002 (hereafter sampling periods 2000 and 2002). In the Pretoriuskop area, sampling was carried out in January and February 2002 only. In the summer rainfall region of South Africa, ants are most active and abundant at this time (Swart, Richardson & Ferguson 1999). Collections could not be made in November and December 2001 because heavy rains prevented sampling. Pitfall trapping was determined to be the optimum sampling method based on a pilot study comparing winkler and pitfall sampling (Parr & Chown 2001).
On each plot, 20 pitfall traps (Ø 62 mm) were laid out in a grid (5 × 4) with 10-m spacing between traps. All grids were situated at least 50 m from the plot edge to reduce the possibility of ants from adjacent areas being collected in the traps, and to reduce edge effects. Pitfall traps contained 50 mL of a 50% solution of propylene glycol, which neither significantly attracts nor repel ants (Adis 1979). The pitfalls were left to settle to reduce the ‘digging-in’ effect (Greenslade 1973) and then all traps were opened for a period of 5 days. A pilot study indicated that this period was sufficient for reasonably complete sampling, without collecting excessive numbers of ants. There was no rain during the periods that the pitfall traps were open, and the weather during both sampling periods was hot and humid. Pitfall samples that had been disturbed by animals were excluded from the analyses.
Samples were washed and sorted in the laboratory. Whenever possible ants were identified to species, otherwise they were assigned to morphospecies. Voucher specimens are held at the Iziko Museum of Cape Town, Cape Town, South Africa.
To assess the overall effect of fire regime on vegetation, vegetation foliage height profiles (for the 2002 sampling period) were determined based on the methods discussed by Rotenberry & Wiens (1980) and Bestelmeyer & Wiens (1996). Vegetation height was measured at four points located 90° apart on a 1·5-m radius centred on each pitfall trap. At each point, a 1·5-m long pole was placed vertically, and the number of times vegetation came into contact with the pole in each height class (1 = 0–0·25 m, 2 = 0·26–0·50 m, 3 = 0·51–1·00 m, 4 = 1·00–1·50 m) was recorded. Ground cover was estimated on each plot by placing a 1-m2 quadrat next to each pitfall and estimating the percentage cover of grass, bare ground, litter and dead grass and forbs.
analyses: ant assemblages
Total species richness and abundance were compared between and within habitat types for both sampling periods using analysis of variance (anova), and for each habitat type species richness and abundance were also compared between sampling periods. anovas were used to determine if there were any significant differences in species richness and abundance between replicates and burn plot treatments for each sampling period separately, and combined. This was done for the Mopane and Satara areas separately. Because the use of anova requires that data are normally distributed, data were log transformed where necessary. For data that could not easily be transformed, non-parametric tests, Mann–Whitney U and Kruskal–Wallis anova, were applied to the data. Sequential Bonferroni corrections were applied to adjust the statistical significance for multiple tests (Rice 1989).
In addition to the richness values obtained from the sampled ants, the non-parametric incidence-based coverage estimator (ICE) provided in EstimateS (Version 5; Colwell 1997) was used to improve the estimate of species richness per plot. ICE is a promising, and reliable, estimator of species richness (Chazdon et al. 1998; Longino, Coddington & Colwell 2002) because it stabilizes fairly well, and provides an estimate independent of sample size. ICE is based on the number of species found in 10 or fewer sampling units (Lee & Chao 1994). While estimators are valuable tools they should not be viewed uncritically, and yield minimum estimates of species richness (Longino, Coddington & Colwell 2002).
Multivariate community analyses were undertaken using primer v.5.0 (Clarke & Gorley 2001) to assess overall changes in ant assemblage composition. Cluster analyses using group averaging and Bray–Curtis similarity measures were used to determine whether ant assemblage structure varied between years, and within- and between- habitats. Data were fourth-root transformed prior to analyses to reduce the weight of common species. Analyses of similarity (anosim) were used to establish if there were significant differences in the ant assemblages on plots that differed in burn season (August, April, control and variable), frequency (annual, biennial, triennial, control and variable) and age (i.e. time since fire). The anosim procedure of primer is a non-parametric permutation procedure applied to rank similarity matrices underlying sample ordinations (Clarke & Warwick 2001). anosim produces a global R-statistic, which is an absolute measure of distance between groups. An R-value approaching one indicates strongly distinct assemblages, whereas an R-value close to zero indicates that the assemblages are barely separable. These R-values were used to compare ant assemblages between habitat types, and burn plot treatments within and between sampling periods. R-values may occasionally be very low, indicating that assemblages are barely separable, but these values may also be significantly different from zero. This reflects a high number of replicates or samples, and the fact that R is inconsequentially small is of greater importance (Clarke & Warwick 1994). The converse may also be found, where R-values may be very high (indicating that assemblages are almost completely different) but these values are not significant. This situation occurs when the sample size is small, and in such instances the R-value is of greater importance (Clarke & Gorley 2001). The relationships between habitat types and burn plot treatments for both sampling periods combined were displayed using non-metric multidimensional scaling (nMDS) ordinations. These were iterated several times from at least 10 different starting values to ensure that a global optimum was achieved (indicated by no decline in the stress value) (Clarke & Gorley 2001). Although the experiment was designed originally to test season and frequency aspects of the fire regime, the effect of individual fires cannot be ignored as different post-fire fuel ages may result in successional effects (York 1994). For age analyses, plots were classified as follows: young = 4–5 months since fire; intermediate = 8–16 months since fire; old = 20–28 months since fire; unburnt = control and variable.
The effects of frequency, season and age were initially assessed using a series of pair-wise anosim tests. First, the age of the plots was varied (young, intermediate and old) while controlling for season and frequency. Ant assemblage composition did not differ with age for these pair-wise tests. Additional pair-wise anosim tests were necessary to determine whether it was possible to combine all frequencies or both seasons when doing subsequent analyses. Thus, frequency of burn was controlled for while season was altered (pair-wise tests: August biennial vs. April biennial and August triennial vs. April triennial), and season of burn was controlled for while frequency of burn was altered (pair-wise tests: August biennial vs. August triennial, August annual vs. August biennial, August annual vs. August triennial and April biennial vs. August triennial). There were no significant differences in ant composition between any of the burn treatment pairs. Therefore, for all further analyses, both seasons were combined when testing for frequency, and frequencies were combined when testing for effects of season.
Finally, ant species characteristic of the three habitat types, and of burnt and unburnt plots in each area, were identified using the Indicator Value method (Dufrêne & Legendre 1997; http://mrw.wallonie.be/dgrne/sibw/outils/indval/home.html). This method assesses the degree to which a species fulfils the criteria of specificity (uniqueness to a particular habitat) and fidelity (frequency of occurrence). A high indicator value (IndVal, expressed as percentage) indicates that a species can be considered characteristic of a particular habitat or site. This method can derive indicators for hierarchical and non-hierarchical site classifications, and is robust to differences in the numbers of sites between site groups (McGeoch & Chown 1998). Indicator values for each species were calculated based on a species abundance matrix, and Dufrêne & Legendre's (1997) random reallocation procedure of sites among site groups was used to test for the significance of IndVal measures for each species. Species with significant IndVals > 70% were considered as species characteristic of the site or habitat in question (subjective benchmark; McGeoch, Van Rensburg & Botes 2002).
Differences in vegetation structure were assessed for each area by comparing the mean number of hits per plot in each foliage height category using anova. Vegetation structure was compared between replicates within an area, and between different burn plot treatments. Percentage cover for each habitat cover component was summed across all quadrats on each burn plot treatment for each replicate. Kruskal–Wallis tests were used to determine whether total habitat cover differed significantly between burn plot treatments. The bioenv procedure in primer was used to examine the relationship between habitat cover on the plots and the ant assemblages (Clarke & Gorley 2001). A single, among-site species similarity matrix was constructed using Bray–Curtis similarity measures, while several similarity matrices were constructed for each of the possible combinations of the specified habitat cover variables also using Bray–Curtis similarity measures because the measurement scale for all variables was percentage cover. Spearman's rank correlation coefficients (ρ) were then calculated for the species matrix and each of the possible habitat matrices. The variable or set of variables that have the highest ρ-value are those that best explain the species data (Clarke & Gorley 2001). Unfortunately this ρ-value does not produce an associated significance value.
In addition, anosim analyses were carried out for the Mopane and Satara areas to determine the effect of plot age (single fire effect) on habitat cover composition. The effect of season and frequency of burn (i.e. for fire regime) on habitat cover could not be investigated because anosim tests revealed that there were significant differences in vegetation cover between burn treatments of different ages. Vegetation analyses did not include Pretoriuskop because data from that area were only available for one sampling period.