Seedling establishment in an Australian tropical savanna: effects of seed supply, soil disturbance and fire


  • S. A. Setterfield

    Corresponding author
    1. Faculty of Science, Information Technology and Education, Northern Territory University, Darwin, Northern Territory 0909, Australia
    Search for more papers by this author

S. A. Setterfield, Faculty of Science, Information Technology and Education, Northern Territory University, Darwin, Northern Territory 0909, Australia (fax + 61 88946 6847, e-mail


  • 1Australia's savannas typically are burnt every 1–3 years. Although there are concerns about the effect of frequent fire on recruitment of Australian savanna species, there is a lack of information. This research aimed to determine whether seed or microsite availability limits seedling recruitment of the overstorey tree Eucalyptus miniata and the midstorey shrub Acacia oncinocarpa, whether seed or microsite availability is affected by frequent fire, and the consequent effect on seedling recruitment.
  • 2Quadrats were established in unburnt areas. Experimental manipulations were addition of seed and/or disturbance of the soil surface to increase the number of microsites suitable for germination. In a second experiment, seed was added to quadrats established in three fire regimes after the annual burning event (unburnt, burnt early in the dry season, burnt late in the dry season).
  • 3In unburnt areas, seedling regeneration was limited by both seed supply and microsite availability. Both burning regimes reduced seedling emergence, possibly because the reduced canopy cover caused unfavourable microclimate, the increased grass and forb ground cover increased competition for resources, and there was increased loss of seed to seed harvesters.
  • 4The results indicate that sexual regeneration of these common species is disadvantaged by current burning practices because both seed supply and the number of microsites are reduced. Thus, long-term changes in savanna floristic structure seem likely unless fire managers aim to increase the fire-free intervals. The relative abundance of species able to reproduce vegetatively may increase under frequent fire regimes. Such a change may take a long time to detect given the fire-resistant and long-lived nature of overstorey species, and the capacity for vegetative regeneration in many savanna species. The impacts of fire on limiting seedling regeneration will require savanna managers to consider fire-free intervals of several years for effective recruitment of dominant woody species.


Fire is an important landscape-scale disturbance in many of the world's ecosystems, particularly tropical savannas. In Australia, Africa, Asia and South America tropical savannas are intentionally burnt every 1 to 3 years (Eiten & Goodland 1979; Trollope 1982; Stott 1990; Russell-Smith et al. 1997) for land clearing and livestock management, for protection of property, for cultural and spiritual purposes and for biodiversity conservation (Andersen et al. 1998). The dominance of fire as a land management tool has resulted in many debates over the ecological consequences of imposed fire regimes (Bowman 1992; Lonsdale & Braithwaite 1992). An important but little studied consequence of frequent fire is the constraint on plant reproduction and its consequences for processes of plant recruitment, woody plant structure and species composition. For example, in the savannas (cerrado) of Brazil, studies on woody plant recruitment suggest that frequent fire is causing a shift in species composition, favouring species that can reproduce vegetatively (Hoffmann 1998, 1999). Due to the dominance of vegetative regeneration, and low growth rates of cerrado species, any negative effects of fire on reproductive success and consequent attrition in plant populations could take decades to be detected in the adult plant density (Hoffmann 1999). A similar situation exists in northern Australia, where sexual reproduction of woody species in the Australian savannas has been described as rare (Lacey 1974), and vegetative reshoots from underground organs are the most conspicuous form of regeneration (Lacey & Whelan 1976; Fensham & Bowman 1992).

The Australian savannas cover the northern quarter of the continent, and are dominated by an overstorey of Eucalyptus species, a shrub layer of deciduous and evergreen species, including Acacia species, and an understorey of annual and perennial C4 grasses. The dry season is the peak period for reproductive activity in the majority of woody species in Australian savannas (Setterfield & Williams 1996; Williams et al. 1999), as it is in other areas in the seasonal tropics (Sarmiento & Monasterio 1983). Flowering and seed production therefore coincide with the main fire season in the region. Recent studies have shown that fire has a significant effect on floral phenology patterns in common woody species (Setterfield 1997a; Williams et al. 2001). Frequent low intensity fires early in the dry season cause a significant reduction in flower and seed production in some species (Setterfield 1997a,b), but have no significant effect on flowering in other species (Williams et al. 2001). Higher intensity fires that occur late in the dry season cause a significant reduction in the abundance of flowers, fruit and seeds relative to unburnt savanna for all species that have been studied (Setterfield 1997a,b; Williams 1997).

Whether the reduced seed production due to fires ultimately reduces seedling establishment will depend on the factors that limit seedling recruitment. There are two basic factors that can limit seedling recruitment: the availability of seeds, and the availability of microsites where seed can germinate and establish (Harper 1977; Crawley 1990; Eriksson & Ehrlen 1992). Increasing the seed supply can increase seedling recruitment (Shaw & Antonovics 1986), but only if suitable microsites are available (Wellington & Noble 1985; Andersen 1989). It has been suggested that for most plant species, recruitment is limited primarily by microsite availability rather than seed availability (Crawley 1990). However, manipulative experiments on a wide range of species in the deciduous and coniferous woodlands of Sweden showed that seed supply was also important for most of the species considered (Eriksson & Ehrlen 1992). Thus, Erikkson & Ehrlen (1992) concluded that the importance of seed limitation in plant populations has been underestimated.

Fire can affect both seed supply and the availability and quality of microsites. As described above, fire can reduce potential seed supply by reducing plant fecundity and killing ovules (Setterfield 1997a; Hoffmann 1998). Elsewhere, fire causes a substantial increase in seed supply of serotinous species following fire, when the heat from fires triggers massive seed release (Lamont et al. 1991). The number of suitable microsites can be altered by fire by removing competing vegetation, changing light quality and the surface soil structure (Lamont, Witkowski & Enright 1993; Tyler 1995). Fire can promote germination in many species. For example, the heat from fires can break seed dormancy of legumes by rupturing the hard seed coat (Cavanagh 1980), and fire-related cues, such as charred wood (Keeley et al. 1985) and smoke (Dixon, Roche & Pate 1995) can promote germination in some species. Fire can also effect seedling establishment by changing the abundance of seed harvesting ants (Andersen 1991). In northern Australia, the abundance of seed predator species of Monomorium is increased by frequent fire (Andersen 1991). However, the subsequent effect of increased seed predators on seedling recruitment is unknown.

This study investigates the factors that influence seedling establishment of two common woody species in the northern Australian savannas: the canopy dominant Eucalyptus miniata Cunn. Ex Schauer, and understorey shrub Acacia oncinocarpa Benth. Both species flower and produce seed during the dry season, and seed fall of both species is substantially reduced by fire (Setterfield & Williams 1996; Setterfield 1997a). This study determines the impact of fire regimes on recruitment by addressing the following questions: (i) is recruitment of E. miniata and A. oncinocarpa limited by seed or microsite availability, or a combination of the two; and (ii) does the fire regime influence the success of seedling emergence and survival?


study site

The study was part of a landscape-scale fire experiment (Andersen et al. 1998) conducted at the CSIRO Kapalga Research Station in Kakadu National Park, Northern Territory, Australia (132°25′ E, 12°40′ S). The study was undertaken in tall (to 25 m) open forest dominated by Eucalyptus miniata and Eucalyptus tetrodonta F.Muell. The mid-layer (4–10 m) included a number of Acacia species. The ground layer was dominated by a mixture of annual and perennial grasses (up to 3 m), particularly Sorghum spp. and Heteropogon spp. Woody vegetative sprouts formed a conspicuous component of the ground layer (Lacey & Whelan 1976). These included vegetative reshoots originating from lignotubers or the buried stem base, and vegetative suckers that originate from rhizomes and root buds at a distance from the parent plant. The climate is monsoonal, characterized by high temperatures throughout the year and a highly seasonal rainfall. Average mean monthly maximum temperatures vary from 31 °C in June and July, to 36 °C in January. Approximately 90% of the 1300 mm annual rainfall falls between December and March (Taylor & Tulloch 1985).

The study area was subjected to fires annually or biennially until 1987, and then remained unburnt until the commencement of a fire experiment in 1990. Fire treatment units were 15–20 km2 catchments, each based on seasonal creek-lines. This study considered three fire regimes that had been applied to catchments since 1990: (i) ‘unburnt’; (ii) ‘early’, burnt annually early in the dry season (May/June); and (iii) ‘late’, burnt annually late in the dry season (September). Three replicate catchments of each regime were used. Although there was considerable spatial and annual variation, fires lit early in the dry season (April–June) were typically low in intensity (approximately 2100 kW) and patchy (Williams, Griffiths & Allan 2001), whereas fires that occurred late in the dry season (July–November) are usually more intense (approximately 7700 kW m−1), often scorching the canopy completely and burning extensive areas (Williams 1995). These fire regimes are typical of those occurring across the broader region that supports mesic savanna vegetation (Russell-Smith, Ryan & Durieu 1997).

study species

Eucalyptus miniata Cunn. ex Schauer is one of the most common overstorey trees (15–30 m high) in the wetter, subcoastal regions of northern Australia (Boland et al. 1992). Acacia oncinocarpa Benth. is a small spreading shrub (2–4·5 m high) that is common in the open forest, woodland and shrubland of northern Australia. Both species can regenerate from seed and they can vegetatively regenerate new shoots from buds that escape fire or other damage (Lacey & Whelan 1976). Neither species can reproduce new individuals vegetatively.

For both E. miniata and A. oncinocarpa, floral bud burst occurs early in the dry season, and ovule development and seed fall are completed within the 8-month dry season. Seed release occurs upon ripening in both species, and typically occurs between July and October for A. oncinocarpa, and between September and December for E. miniata (Setterfield & Williams 1996; Setterfield 1997a). Given the extremely high frequency of fires in the region, seed is often released onto a seedbed that has been recently burnt.

microsite and seed supply as limiting factors

This experiment was undertaken in three replicate unburnt catchments. A site was randomly located within each catchment, and at each site 24, 75 × 75 cm quadrats were established. The experiment was a 2 × 2 factorial design with the manipulated factors being microsite density and seed addition. This study assumed that disturbance of the soil surface would increase microsite availability (see Eriksson & Ehrlen 1992). Microsites were manipulated by first removing the leaf litter from the quadrat, and then scarifying the soil surface with a hand cultivator to a depth of 5 cm. The leaf litter was then replaced. Seed was added at two densities: low density quadrats were sown with 20 seeds, high density quadrats were sown with 200 seeds. Seed was scattered evenly over the quadrat area. The germinability of seed used in this and the following experiment was 95% and 51% for E. miniata and A. oncinocarpa, respectively.

The following combination of treatments was applied to three replicates for both species at each of the three sites:

  • • soil and litter disturbed, low density of seed applied;
  • • soil and litter disturbed, high density of seed applied;
  • • soil and litter undisturbed, low density of seed applied; and
  • • soil and litter undisturbed, high density of seed applied.

Sites were located in areas where both study species were present. Microsite and seed manipulation occurred in November 1993. A 5-cm fence was constructed around each quadrat to ensure that the seed was not washed from the site during wet season rains. The fence consisted of four strips of galvanized wire mesh covered in flywire screen. Sticks and leaf litter were pushed up against the fences to encourage access to ants. Free movement by ants into and out of the plots was observed during the experiment. The number of seedlings was initially counted on 10 January 1994, and each was marked with a wooden skewer. The quadrats were rechecked during February, March, April, May, July and October 1994 to determine a cumulative total of emergent seedlings, and survival during the dry season. E. miniata seeds have soft seed-coats, and seed longevity trials have shown that the sown seed would have germinated or died by February (S. Setterfield, unpublished data). A. oncinocarpa has a hard seed-coat, and seed longevity trials have shown that most of these seeds would have germinated or died by July. A small proportion of A. oncinocarpa seeds are maintained dormant in the seedbank.

The effects of the treatments on seedling emergence were compared using a four-way mixed model anova with site (random), species (fixed), seed density (fixed) and microsite manipulation (fixed). Data were log-transformed prior to analysis to meet the assumptions of anova.

effects of fire regime on seedling emergence

The experiment was undertaken in three replicate catchments of each of three experimental fire regimes: unburnt, early and late. Two experimental sites were located within each catchment, and six 75 × 75 cm quadrats were randomly located at each site. In November 1993, 200 Eucalyptus miniata seeds were sown into each of three of the quadrats, and 200 Acacia oncinocarpa seeds were sown in the remaining three quadrats at each site. Therefore, in the early and late catchments, the seeds were sown onto a seedbed that had been burnt during the dry season. Each quadrat was bordered by a fly wire mesh fence (described above).

The number of seedlings was initially counted on 10 January 1994, and each was marked with a wooden skewer. Seedling emergence was monitored in February, March, April, May, July (post-early burn) and October 1994 (post-late burn). Sites were censused again in June 1995 (i.e. at the end of the second wet season) by which time seedlings were considered to be established. The cumulative total of seedlings that emerged in 1994 was compared with four-way mixed model anova with the factors species (fixed), fire regime (fixed), catchment (random and nested) and site (random and nested). The data were log-transformed prior to anova following an assessment of Cochran's test and probability plot. In September 1994, a wildfire entered one of the unburnt catchments, killing all marked seedlings. Therefore, data on seedling survival in October 1994 and June 1995 are from the two remaining unburnt catchments.

The ground flora in the savanna changes dramatically throughout the year, with annual plants and above-ground stems of perennial herbaceous species dying off during the dry season. At the onset of the wet season, many perennial herbaceous species sprout new leaves, and the seeds of annual species germinate and seedlings establish. The wet season is the main period of flowering and growth of herbaceous species (Wilson et al. 1996). As these changes could affect both emergence and establishment success of E. miniata and A. oncinocarpa seedlings, the grass, forb, woody sprout and leaf litter cover in each quadrat were estimated at each seedling census time. Canopy cover can effect germination and seedling establishment, so canopy cover over each quadrat was measured using a hemispherical forest densiometer (P.E. Lemmon, Arlington, Virginia, USA) held at a height of 1 m. Four readings were taken over each quadrat (each one facing the four directions on a compass), and the readings were averaged for the quadrat. The specific effect of shading was not determined in this study because the low leaf area index (< 1; O’Grady et al. 2000) of the tall open E. miniata/E. tetrodonta dominated forests results in relatively uniform shading over the landscape, and shading factors are approximately 0·3 (i.e there is 70% transmission of radiation to the forest floor; S. Setterfield, unpublished data). Differences in canopy cover are more likely to be attributable to differences in soil moisture availability (Stoneman & Dell 1994) and therefore germination success. Differences in the canopy and ground cover between regimes were determined using a three-way mixed model anova, with fire regime (fixed), catchment (random and nested) and site (random and nested). The data were arcsine transformed to satisfy the assumptions of anova.

To determine whether differences in seedling establishment could be due to differences in seed lost to ants, a seed removal experiment was established in three replicate catchments of each experimental fire regimes. Losses to seed removal were assessed by recording the loss of seeds from seed caches (Andersen & Ashton 1985) located at four random sites within each replicate catchment. At each site, caches of 10 seeds of each species were placed out in a 3-4-3 grid pattern. To ensure that the seed could be easily relocated, each seed was placed on a small amount of white sand and the seed within a cache was separated by 30 cm. Seeds were checked daily for 3 days. The trial was undertaken twice for each species; between 28 and 31 October, and 10 and 13 November 1994 for A. oncinocarpa, and between 10 and 13 November, and 24 and 27 November for E. miniata. Results were analysed separately for each species using a three-way mixed model anova, with factors: trial (fixed), fire (fixed) and catchment (random).


effects of seed and microsite addition on seedling emergence

The number of emergent seedlings was increased significantly by both increasing the number of seeds (Fig. 1; F1,2 = 61·0, P < 0·05) and manipulating microsites (F1,2 = 32·1, P < 0·05). Increasing seed density from 20 seeds to 200 seeds on unscarified soil resulted in approximately 20 times the number of E. miniata seedlings and seven times the number of A. oncinocarpa seedlings. Similarly, microsite disturbance resulted in a two- to fourfold increase in the number of seedlings establishing. A combination of seed and microsite creation resulted in approximately 75 times the number of E. miniata seedlings and 20 times the number of A. oncinocarpa seedlings compared with the undisturbed soil, low seed density treatment (Fig. 1). The higher number of E. miniata seedlings that established compared with A. oncinocarpa is probably due to the higher germinability of the E. miniata seed (95% compared with 51% for A. oncinocarpa). There were no significant interactions between experimental factors.

Figure 1.

The cumulative number of seedlings (+ 1 SE) that established in unburnt catchments after four treatments were applied: (i) undisturbed, high density of seed; (ii) undisturbed, low density of seed; (iii) disturbed, low density of seed; (iv) disturbed, high density of seed. Results represent means of three replicates of each treatment from three sites.

effects of fire on seedling establishment

All of the E. miniata seedlings that emerged did so in the first wet season following seed sowing in 1994, whereas some A. oncinocarpa seedlings (< 1% of seed applied) emerged from the seedbank during the second wet season. The total number of seedlings that established decreased with increasing intensity of the fire season (Fig. 2). For E. miniata, about 16 seedlings established per quadrat in the unburnt regime compared with 13 and 7 seedlings in the early and late regimes, respectively. For A. oncinocarpa, the figures were 10, 6 and 3, respectively. This variation between regimes was significant (F2,6 = 5·38, P < 0·05) and Ryan's test showed that the number of seedlings varied significantly from unburnt > early > late. There was no significant difference between species or between sites.

Figure 2.

Mean (+ 1 SE) cumulative number of E. miniata and A. oncinocarpa seedlings in three fire regimes. Results for each species represent three replicates at two sites within three replicate catchments.

At the time of seed sowing (November), there were substantial differences in the seedbed microenvironment between fire regimes (Table 1). Grass cover varied significantly between regimes (F2,6 = 5·69, P < 0·05, Ryan's test, late > early > unburnt), with approximately 7% cover in the unburnt, 11% in the early regime and 15% in the late regime. The cover of regenerating sprouts of woody species also varied significantly between regimes (F2,6 = 5·76, P < 0·05, Ryan's test, late > early = unburnt), with c. 12% cover in the late regime compared with only 3% and 2% in the unburnt and early regimes. Although mean forb cover also increased with increasing fire regime intensity, this difference was not significant. Projected canopy cover also decreased significantly with increasing fire intensity (F2,6 = 6·45, P < 0·05, Ryan's test, unburnt = early > late). Leaf litter cover varied significantly between regimes, with an 85% cover in the unburnt, and approximately 35% cover in the early and late regimes (F2,6 = 5·89, P < 0·05, Ryan's test, unburnt > late = early).

Table 1.  Percentage of (a) canopy cover, (b) grass cover, (c) forb cover, (d) woody sprout cover, and (e) leaf litter cover, at the time of seed sowing. Means (and standard errors) are from six quadrats (0·75 × 0·75 cm) from two sites in three catchments in each fire treatment. Different superscript letters denote characteristics that varied significantly between regimes according to Ryan's multiple comparisons test
Canopy71·7 (2·2)A65·6 (2·7)A45·5 (2·5)B
Grass 6·9 (0·9)A11 (1·8)B15 (1·8)C
Forb 4·4 (0·6) 6·1 (0·7) 8 (1·2)
Woody sprout 2 (2·6)A 2·6 (0·9)A12·4 (3·6)B
Leaf litter85·8 (4·2)A35·2 (6·2)B31·4 (7·3)B

Considerable variability also existed between sites with respect to the seed microenvironment. Significant differences between sites occurred in grass cover (F9,90= 6·01, P < 0·001), canopy cover (F9,90 = 7·78, P < 0·001) and woody sprout cover (F9,90 = 5·2, P < 0·001).

As the wet season progressed, the grasses and forb species flourished (Fig. 3). By late in the wet season (March), the grass cover in the burnt regimes was > 25% compared with 13% in the unburnt regime (Fig. 3a). Forb cover was also consistently higher in the burnt regimes compared with the unburnt during the wet season (Fig. 3b). By contrast, mean woody sprout cover in each regime remained relatively constant throughout the wet season (Fig. 3c). The leaf litter cover increased slightly in all regimes over the wet season (Fig. 3d). The cover of grass, forb and leaf litter was substantially reduced in the early regime by the experimental fires in June (Fig. 3).

Figure 3.

Change in percentage cover of (a) grass cover, (b) forb cover, (c) woody sprout cover, and (d) leaf litter cover, from the time of seed sowing in November 1993 to July 1994. Estimates are averages from six quadrats (0·75 × 0·75 cm) from two sites in three catchments in each fire treatment.

Figure 4 presents survival curves for E. miniata and A. oncinocarpa in the unburnt regime. The results are not presented for the burnt regimes because the very low number of seedlings that emerged provided an unreliable estimate of survival. The increase recorded for A. oncinocarpa seedlings between January and March is due to new seedlings emerging during these months. Approximately 25% of E. miniata seedlings and 35% of A. oncinocarpa seedlings that established in the unburnt regime in the first wet season survived to the end of the following dry season (October 1994). Survival had dropped to 11% and 33%, respectively, by the middle of the following dry season (June 1995).

Figure 4.

Survival data for E. miniata (open triangle) and A. oncinocarpa (closed square) in the unburnt regime at Kapalga. Results represent mean ± 1SE.

Approximately 40% of E. miniata seedlings in the early regime survived to June (7 months after sowing), immediately prior to the early dry season fires. Following the fires, only 3% of the seedlings were alive, and these died in the following month. In the late regime, 35% of the E. miniata seedlings survived to September, but these were all killed by the late season fires.

Fire also killed most of the A. oncinocarpa seedlings. However, 5% of seedlings in the early regime, and 10% of seedlings in the late regime lost their above-ground stem during the fires, but the seedlings resprouted in the wet season. In addition, a few new A. oncinocarpa seedlings germinated from the seed bank during the wet season.

The number of seeds removed from caches after 72 h varied significantly between fire regimes for both E. miniata (F2,6 = 9·01, P < 0·05) and A. oncinocarpa (F2,6 = 5·2, P < 0·05) with more than double the number of seeds removed from the two burnt regimes compared with the unburnt regime (Fig. 5). About 18% of E. miniata seeds were removed from the unburnt catchments compared with approximately 40% and 50% in the early and late regimes, respectively. For A. oncinocarpa, 35% of seeds were removed from the unburnt regime compared with c. 70% in the burnt regimes. This difference was significant, although Ryan's test did not detect a difference in the rate of seed removal between the burnt regimes. There was no significant difference in the amount of seeds removed between trials for either species and no significant interactions between factors. The amount removed varied significantly between catchments for A. oncinocarpa (F6,54 = 4·47, P < 0·001).

Figure 5.

Mean percentage (+ 1 SE) of E. miniata and A. oncinocarpa seed removed from seed caches in the three fire regimes. Results for each species represent four replicate caches within three replicate catchments.


Whereas many studies have viewed limitations to recruitment as a dichotomy of seed vs. microsite, Eriksson & Ehrlen (1992) argued that the recruitment of most species would be shown to be limited by the availability of both seed and microsites if the species were adequately studied. This study supports their view, because a combination of both seed and microsite availability limited population recruitment of A. oncinocarpa and E. miniata. Artificial increases in seed availability and microsite availability have increased the number of seedlings in other communities, such as calcareous grasslands (Zobel et al. 2000), deciduous and coniferous woodlands (Eriksson & Ehrlen 1992) and boreal forests (Eriksson & Fröborg 1996). The results from this study provide an important context in which to evaluate the effects of fire regimes on seed availability and microsite availability, since a reduction in the availability of either will reduce the chance of seedling recruitment for both species. For example, in light of the knowledge that seedling recruitment is limited by seed supply, the negative effect of burning on seed production (Setterfield 1997a,b) can be viewed as an important modifier of population dynamics in these species.

The reduction in seed availability caused by burning is amplified by differences in the intensity of seed removal between regimes, with seed removal in the early and late regimes more than double that of the unburnt regime. Most of the seed removed can be attributed to harvester ants. Andersen & Lonsdale (1990) reported that the rothsteini ant group of the Monomorium removed the majority of the Eucalyptus tetrodonta seeds in this region. Exclusion trials have shown that post-dispersal seed losses of E. miniata and A. oncinocarpa to vertebrates were insignificant (S. Setterfield, unpublished data). The impact of other invertebrates is also likely to be insignificant (Andersen & Lonsdale 1990). Seeds removed from the caches were assumed to be ‘lost’ to the recruitment process. Previous studies have shown that Eucalyptus seeds are eaten either in situ (Wellington 1989), or carried to the nest, stored and then may be eaten later. However, the fate of Acacia seeds is less certain, with some studies showing that seeds are dispersed by ants and the aril removed, but the seed is left intact (Berg 1975; Auld 1986). Nevertheless, Ireland & Andrew (1995) speculated that most Acacia papyrocarpa seed removed by ants is then eaten.

The difference in seed removal between burnt and unburnt areas reflects differences in the abundance and composition of the ant fauna between regimes at Kapalga. There was a marked reduction in ant abundance in the unburnt compartments after the imposition of fire regimes (Andersen & Muller 2000), and this pattern of decline has been documented elsewhere in the tropical savannas (Andersen 1991) and in other habitats of Australia (Andersen 1988). Fire regimes have a major effect on composition of ant communities in the Australian savannas, with several seed harvesting species occurring predominantly in burnt areas (Andersen 1991). Differences in seed removal rates may also be due to the differences in the foraging efficiency of ants, which is influenced by ground cover habitat structure (Andersen & Muller 2000). For example, the presence of litter significantly reduced removal of Eucalyptus baxteri seeds in southern Australia (Andersen & Ashton 1985). In the current study, the foraging efficiency of ants was probably higher in the burnt regime, where the ground layer at seed sow was mainly bare ground with some grass cover, whereas the unburnt ground was covered by a thick leaf litter layer (Table 1).

The low and variable rates of seedling establishment documented in this study (Fig. 2) are not a product of seed removal rates alone. Table 2 presents the germination success of 200 E. miniata and A. oncinocarpa seeds, calculated as a proportion of seed available for germination and establishment after seed removal has been taken into account. Establishment of less than 11% of available seed for both species demonstrates a lack of suitable microsites for establishment in all fire regimes. In addition, fire regime influenced the availability of microsites, with lower establishment rates in the burnt regimes compared with the unburnt (with the exception of the E. miniata, which had a similar rate of establishment in both the early and unburnt regime, Table 2). The differences may have resulted from an indirect effect of the changes in the vegetation structure on the seedbed conditions (Table 1 and Fig. 3). For example, lower seedling emergence in the late regime may be related to the effects of reduced canopy cover on soil water availability. Canopy cover is significantly less in the late regime compared with the unburnt and early regimes due to the reduction in live tree basal area and canopy fullness after fire (Williams et al. 1999). The reduced canopy cover causes fluctuations in soil water availability, which can be lethal if water supply from the irregular rainfall at the start of the wet season is sufficient for germination to be induced but insufficient for seedling establishment (Augspurger 1979). Seeding trials have shown that seedling establishment success in the Australian savannas is determined largely by the consistency of the first rains of the wet season (Setterfield 1997b). The greatly reduced canopy cover (Table 1), and the higher soil temperatures in the late regime (mean soil surface temperature was 8 °C higher in the late regime compared with the unburnt in December; Setterfield 1997b), are more likely to result in fluctuating soil water availability, and hence higher water stress, compared with the unburnt and early regime. It has been suggested that recruitment in savannas elsewhere is controlled by rainfall (Higgins, Bond & Trollope 2000). In addition, seedling establishment was generally greater under tree canopy for a range of neotropical savanna species (Hoffmann 1996). The impact of fluctuations in water availability is more likely to affect E. miniata than A. oncinocarpa because eucalypt seeds germinate within days of the soil becoming moistened and before consistent rainfall has commenced in the monsoonal areas of Australia (Setterfield 1997b). Acacias have hard seed coats and consequently germinate during the period of regular wet-season rains.

Table 2.  The fate of 200 seeds of (a) E. miniata, and (b) A. oncinocarpa that were applied to 0·75 × 0·75 m2 quadrats. Results show (i) the initial number of seeds applied to the quadrats; (ii) the proportion of seeds lost from caches; (iii) the number of seeds available for recruitment after accounting for seed removal; (iv) the mean number of seedlings that established per quadrat; and (v) the germination success of remaining seeds, which was calculated as the percentage of seeds available for recruitment (from iii) that resulted in established seedlings
Species  UnburntEarlyLate
(a) E. miniata(i)Number of seeds200200200
(ii)Proportion lost (%) 18 40 50
(iii)Number of remaining seeds164120100
(iv)Number of seedlings 16 13  7
(v)Germination success 10 11  7
(b) A. oncinocarpa(i)Number of seeds200200200
(ii)Proportion lost (%) 35 70 70
(iii)Number of remaining seeds130 60 60
(iv)Number of seedlings 13  4  2
(v)Germination success 10  6·7  3·3

The difference in seedling emergence between regimes may also be attributed to differences in the ground cover between regimes (Fig. 3). Although burning temporarily clears the ground flora, the perennial grasses soon reshoot, and annual grasses and herbaceous species germinate and establish at the onset of the wet season rains. A. oncinocarpa and E. miniata seeds germinate during the early to mid wet season and the establishing seedlings would experience greater competition for available space, water and nutrients in the burnt regimes compared with the unburnt regime.

In contrast to the results of this study, in which between 4% and 35% of E. miniata seeds emerged in the seeding experiments (Figs 1 and 2), a previous seed sowing trial using E. miniata resulted in no emergence following seed addition to both cultivated and uncultivated soil (Wilson & Bowman 1994). Although reasons for such a dramatic difference between these studies are difficult to determine, it may be related to site differences such as fire history, or differences in patterns of early wet season rainfall. In addition, placement of the seed in the seedbed varied between studies, with the seed in this study scattered over the quadrat, allowing the seed to move naturally to the soil surface. Wilson & Bowman (1994) pressed the seed into the soil and covered it with < 2 mm of soil. Depth of sowing affects the germination and emergence of many Eucalyptus species (Free 1951; Cremer 1965) and may be an explanation for differences in establishment rates between these studies. For example, a few species of Eucalypt have been shown to require light for germination (Boland et al. 1992) and, although the effect of light on germination of E. miniata is unknown, darkness may have inhibited germination in Wilson & Bowman's study.

A high rate of seedling mortality has been recorded in many studies (Andersen 1989; De Steven 1991), and reflects the vulnerability of this stage in the population recruitment of plants (Harper 1977). Various studies have shown that the mortality of seedlings is lower in a post-fire environment compared with an unburnt area (Whelan & Main 1979; Burrows et al. 1980; O’Dowd & Gill 1984). However, a similarly high rate of E. miniata and A. oncinocarpa seedling mortality was recorded in all regimes for the first 6 months after establishment. A large proportion of the seedlings died in fires during their first dry season after establishment. By contrast, Fensham's (1992) study at Melville Island showed that a dry season fire had little effect on the survivorship of Eucalyptus tetrodonta and E. miniata seedlings during their first year. These Eucalypt species are typical of many savanna species in that the seedlings rapidly develop a lignotuber that allows survival during fire (Oliveira & Silva 1993; Hoffmann 2000). The proportion of seedlings that survive a fire will depend upon fire characteristics (e.g. patchiness), as well as characteristics of the seedlings, such as lignotuber development. This study suggests that root development of A. oncinocarpa is rapid following establishment as up to 10% of A. oncinocarpa seedlings that had their aerial stems burnt off during fires in their first year were able to resprout during the following wet season. The length of the fire-free period after establishment is obviously an important factor affecting seedling recruitment of E. miniata and A. oncinocarpa.

The effects of fire regime on seedling emergence are summarized in Table 3, with seed budgets calculated from measurements of seedfall beneath and adjacent to fecund canopies in 1994, which was a year of high seed production for both species (Setterfield & Williams 1996; Setterfield 1997a), and therefore represents the upper level of expected seedling recruitment. The fate of A. oncinocarpa seed was not calculated for the late regime because mature stands of this species did not occur in this regime, and seed production could not be calculated. In the absence of fire at Kapalga, seedling establishment of E. miniata and A. oncinocarpa was of the order of 1 m−2 and 4 m−2, respectively, beneath and adjacent to the canopy of fecund trees. This density of E. miniata seedlings is similar to that reported by Fensham (1992). The data suggest that there will be limited, if any, sexual reproduction occurring in savanna subject to annual burning. Frequent fire reduced the availability of seed for recruitment by decreasing seed production and increasing seed losses to harvesters, and reducing the number of microsites for germination and establishment. In addition, the chance of seedlings surviving fire in their first year after emergence is low.

Table 3.  Seed and seedling survival budgets calculated for (a) E. miniata, and (b) A. oncinocarpa. The lines in these budgets represent the density of natural seedfall under the canopy of a seeding individual in 1994 (Seedfall), the percentage survival of 200 experimentally sown seeds on 0·5 m2 plots after one wet season (Establishment), the density of established seeds after one wet season (Density; = Seedfall × Seedling establishment), and the survival of established seeds after 1 year (one wet season and the subsequent dry season; Survival). Data from Setterfield (1997a,b)
(a) E. miniataSeedfall (seeds m−2)18·11·51·8
Establishment (proportion of 200 sown seeds) 0·080·0650·035
Density after 6 months (seedlings m−2) 1·40·10·06
Survival after 1 year (seedlings m−2) 0·200
(b) A. oncinocarpaSeedfall (seeds m−2)605 
Establishment (proportion of 200 sown seeds) 0·0650·02 
Density after 6 months (seedlings m−2) 3·90·1 
Survival after 1 year (seedlings m−2) 1·20·03 

This study suggests that current fire management regimes will disadvantage population recruitment of species reliant on sexual regeneration in northern Australian savannas. Data on the fire regimes for Kakadu National Park between 1980 and 1994 indicated that approximately 65% of Eucalyptus dominated woodland and 50% of open forest was burnt annually over the 15-year period (Russell-Smith et al. 1997). Approximately 60% of this occurred in the early dry season. Similarly, fire history mapping over the broader landscape indicated that 50% of the mesic savannas in Australia's Northern Territory were burnt at least 3 years in 6 (Williams et al. 2001). Where plant populations are subject to such frequent fire, limited, if any, sexual reproduction will occur. Although many Australian savanna species have the capacity to reshoot from lignotubers, only a small number, such as Eucalyptus tetrodonta and Eucalyptus porrecta, can reproduce vegetatively (i.e. from rhizomes or root suckering at a distance from the parent plant, Lacey & Whelan 1976). For species reliant on sexual reproduction, such as E. miniata and A. oncinocarpa, the current fire regimes will be disadvantageous, and, in time, a change in the vegetation composition will become evident. The change is likely to be very gradual, as suggested by Fensham & Bowman (1992), whose assessment of stem size class showed that stand structure is perpetuated under current fire regimes, but indicated that the relative dominance of the canopy species will shift from E. miniata to Eucalyptus tetrodonta.

The importance of understanding the effect of fire on the process of regeneration should not be underestimated because the consequences may take decades to detect through monitoring of stand composition and structure (Hoffmann 1998). Indeed, comparison of burnt and unburnt savanna vegetation over 5 years at Kapalga detected little effect of annual early fires on the woody structure, although annual late fires did result in a 20% reduction in basal area. Assessment after a 13-year period at the Munmarlary experimental area (also in Kakadu National Park) showed little evidence of an effect of annual fire on floristic and structural differences in open forests (Bowman et al. 1988), although this finding must be considered cautiously in light of the low intensities of the experimental fires (< 2500 kW m−1; Russell-Smith et al., in press). The long-term nature of effects on sexual regeneration are due to the fire-resistant and long-lived characteristics (greater than 100 years; Werner 1986) of the overstorey trees in this region and because the woody sprouts that form a conspicuous part of the ground storey also live for a long time in a cycle of resprouting after fire (Fensham 1992). Short-term population maintenance is therefore not dependent on frequent sexual reproduction. The lack of sexual recruits into this population will limit the input of genetic variation into the layer of woody sprouts, and lead to an increase in the average age of woody sprouts in species reliant on sexual regeneration. It is these woody sprouts that represent the future recruits into the mid-storey and canopy layer, and it has been suggested that after some years, the sprouts lose the ability to move into the next size category, even if given a fire-free period, due to an inability to allocate resources effectively because of competition between buds and shoots (Fensham & Bowman 1992).

By evaluating factors affecting woody recruitment, this study has clearly demonstrated the important role of fire management practices in determining stand structure in Australia's savannas. Frequent burning is likely to have a gradual but significant effect on the population structure of Australia's savannas, and fire managers need to consider fire-free intervals of several years for effective recruitment of dominant woody species. The results of this study concur with those in savannas elsewhere (Hoffmann 1998, 1999) in demonstrating the importance of fire management in altering savanna stand composition.


Many volunteers assisted with fieldwork; particular thanks to Michael Douglas, Teena Press-Brennan, Barbara Setterfield and Gordon Setterfield. The CSIRO and Parks Australia North provided access to Kapalga Research Station. Keith McGuinness provided advice on experimental design and statistical analysis. I am grateful to Alan Andersen, Dick Williams and Gordon Duff for advice throughout the study and to Dave Bowman, Michael Douglas and Ed Witkowski for improvements on the draft manuscript. This study was supported by an APRA scholarship, Greening Australia NT and Parks Australia North.