The savanna-grassland ‘treeline’: why don’t savanna trees occur in upland grasslands?

Authors

  • Julia L. Wakeling,

    Corresponding author
    1. Department of Botany, University of Cape Town, Private Bag X3, Rondebosch, Cape Town 7701, South Africa
    Search for more papers by this author
  • Michael D. Cramer,

    1. Department of Botany, University of Cape Town, Private Bag X3, Rondebosch, Cape Town 7701, South Africa
    2. School of Plant Biology, Faculty of Natural and Agricultural Sciences, The University of Western Australia, 35 Stirling Highway, Perth, WA 6009, Australia
    Search for more papers by this author
  • William J. Bond

    1. Department of Botany, University of Cape Town, Private Bag X3, Rondebosch, Cape Town 7701, South Africa
    Search for more papers by this author

Correspondence author. E-mail: juliawakeling@gmail.com

Summary

1. Treeless grasslands with climates that can support tree growth are common in upland regions around the world. In South Africa, the upland grasslands are adjacent to lowland savannas in many areas, with an abrupt boundary between them that could be termed a savanna-grassland ‘treeline’. Both systems are dominated by C4 grasses and burn regularly, yet fire-tolerant savanna trees do not survive in the grasslands. The upland grasslands experience lower temperatures throughout the year and frost in winter, compared with the warmer savannas.

2. We tested whether frost in the dormant season or slow growth in the growing season in conjunction with frequent fires may explain the tree-less state of grasslands. We measured Acacia seedling growth for a year in a transplant experiment at ten sites across an altitudinal gradient (42–1704 m) from savannas to grasslands. The effect of frost on seedlings was scored during the following winter.

3. Across all species, height (t = −6.04, d.f. = 471, < 0.001), biomass (t = −4.56, d.f. = 228, < 0.001) and height increase (t = −3.40, d.f. = 471, < 0.001) were significantly higher at savanna sites. As the plants were irrigated and initially supplied with nutrients, the main factor affecting growth was likely to be growing season temperature.

4. Saplings that experience slow growing conditions will take longer to reach a height above the flame zone and will therefore have a lower probability of reaching adult tree height and surviving fires. Day length may be the most important cue for the end of the growing season in savanna trees, as growth decreased with shortening day length in February–March while temperatures were still high and plants were not water limited.

5.Synthesis. Savanna trees grew more slowly in cooler upland grassland sites compared with lower elevation warm savanna sites and, under frequent fire regimes, would be prevented from reaching maturity. This may be true globally for similar grasslands where tree growth can occur and could partly explain the lack of trees in grasslands.

Introduction

Treeless grasslands are common in upland regions from the tropics and sub-tropics to temperate regions, often forming mosaics with closed forests (Fig. 1). Examples include the prairies (Knapp et al. 1999) and Appalachian balds of North America (Mark 1958; Lindsay & Bratton 1979), the campos grasslands of South America (Behling et al. 2007; Overbeck et al. 2007), the balds of south eastern Australia (Webb 1964; Fensham & Fairfax 1996, 2006; Fairfax et al. 2009), and upland grasslands throughout sub-Saharan Africa (Acocks 1953; White 1983) and Madagascar (Bond et al. 2008). The climate of the grasslands has the potential to support forests (Bond, Midgley & Woodward 2003b; Bond, Woodward & Midgley 2005) and indeed, forest patches are common in these landscape mosaics (Fig. 1). Given that many of these grasslands are similar to savannas in that they are dominated by C4 grasses and burn regularly (Acocks 1953), it is puzzling that fire-tolerant savanna trees, such as Acacia karroo (Schutz, Bond & Cramer 2009), do not occur within them. Much work has investigated the co-existence of grasses and trees in savannas (Walter 1971; Walker 1987; Scholes & Archer 1997; Higgins, Bond & Trollope 2000), yet little experimental work has investigated the reason for the lack of trees in grasslands. This is surprising, given the extent and economic value of grasslands.

Figure 1.

 Upland C4 grasslands in Brazil, South Africa, Malawi and Madagascar.

Tree density is highly variable in savannas (Sankaran et al. 2005; Sankaran, Ratnam & Hanan 2008). Non-savannas include, at one extreme, grasslands with no trees and at the other, closed woody vegetation (including forests) with little or no grass cover. Grasslands and forests provide an interesting challenge for tree-grass coexistence models. The root-niche separation model (Walter 1971) assumes that trees utilise deeper rooting spaces compared with grasses and predicts that trees would disappear in arid areas where the low precipitation does not penetrate beyond the grass layer. Montane grasslands are mesic, not arid, and should therefore support dense tree cover according to this hypothesis. The ‘escape hypothesis’ argues that tree cover in savannas is limited primarily by demographic bottlenecks to recruitment, especially the growth of saplings to mature tree sizes (Bond & van Wilgen 1996; Scholes & Archer 1997; Higgins, Bond & Trollope 2000). In mesic savannas, fire prevents the formation of high densities of juvenile trees and the growth of these juveniles into adults (Bond & van Wilgen 1996). Juveniles may be burnt to the ground many times, and resprout repeatedly from below-ground root stocks, before they become adults (Bond & van Wilgen 1996; Gignoux, Clobert & Menaut 1997; Scholes & Archer 1997; Higgins, Bond & Trollope 2000; Wigley, Cramer & Bond 2008). Once trees reach a large enough stem size and ‘escape’ height above the flame zone (c. 2–4 m), they are relatively immune to stem kill by fire and have ‘escaped’ the fire trap (Higgins, Bond & Trollope 2000). According to this hypothesis, trees could be excluded either by very frequent intense fires, or by slow sapling growth rates or both.

In South Africa, lower-altitude savannas and higher-altitude grasslands are adjacent in many areas (Acocks 1953; Mucina & Rutherford 2006) (Fig. 1). The boundary between the tree-less grasslands and the wooded savannas is often narrow and abrupt, analogous to an alpine treeline in some areas (Mills et al. 2006), and could be termed a savanna ‘treeline’. The savanna treeline marks the loss of savanna trees from a C4 dominated grassy layer. In contrast to forest-grass boundaries with no fire vs. frequent fire (Bowman 2000), the fire regime is continuous across the savanna-grassland treeline.

Several hypotheses have been proposed to explain the tree-less grasslands in South Africa, including anthropogenic disturbances, frost, fire and soil properties. These have been reviewed by O’Connor & Bredenkamp (1997) and Mills et al. (2006). Of the various types of frost (Sakai & Larcher 1987), air and radiation frosts are relevant to the Highveld grasslands. Several studies suggest that trees are prevented from growing in other grassland areas because of annual frosts or infrequent extreme frost events (Silberbauer-Gottsberger, Morawetz & Gottsberger 1977; Osmond et al. 1987; Fensham & Kirkpatrick 1992; Ball et al. 1997; Coop & Givnish 2007). We tested the hypothesis proposed by Acocks (1953) that frost excludes trees from the upland grasslands of the Highveld. To do so, we planted seedlings in a transplant experiment across a large altitudinal gradient from savannas into grasslands and scored the effects of frost on seedlings in a winter season when plants were dormant.

We also propose a new hypothesis regarding the lack of trees in the grasslands of the Highveld, which is based on the escape hypothesis used to explain variation in tree cover in savannas (Bond & van Wilgen 1996; Scholes & Archer 1997; Higgins, Bond & Trollope 2000). We suggest that, due to the cooler and shorter growing season of these grasslands, tree saplings cannot grow fast enough within the intervals between fires to escape the flame zone. This implies that savanna trees are not limited by growing conditions alone, but by the interaction between fire and growth, and should be able to grow under the climate of the Highveld grasslands if fire is excluded as has been shown in several fire exclusion studies (Westfall, Everson & Everson 1983; Ellery, Mentis & Scholes 1992; Titshall, O’Connor & Morris 2000). To test this, we measured seedling growth for 1 year in a seedling transplant experiment from savannas into grasslands. To supplement this altitudinal experiment, a common garden experiment to follow the phenology of growth for a complete growing season was set up in one location. To help evaluate the ecological significance of relative differences in growth rates, we used the experimental growth rates, adjusted to natural sapling growth, to compare times to escape height in grasslands and savannas relative to fire frequencies in the region of study.

Methods

Species selection

Six species of Acacia, a widespread genus in Southern Africa and elsewhere, were selected to represent the full altitudinal range of the genus, from both warm and cool regions: Acacia karroo (Bloem), A. karroo (Hlu), A. gerrardii, A. sieberiana, A. tortilis and A. mearnsii, an Australian species. Two varieties of Acacia karroo were used, as this species is the most widely distributed Acacia species in southern Africa and varieties may differ in their growth and survival at different temperatures. ‘A. karroo (Bloem)’ was collected from the Bloemfontein area (c. 1500 m, mean annual temperature (MAT) = c. 14–16 °C), and ‘A. karroo (Hlu)’ from eastern KwaZulu Natal, including the Hluhluwe-iMfolozi Game Reserve (c. 100–400 m, MAT = c. 18–21 °C). Coates Palgrave (Coates-Palgrave 2002) split A. karroo into six species, naming the Hluhluwe variety A. natalitia and the Bloemfontein variety remains A. karroo (Swartz 1982). Acacia gerrardii is morphologically similar to A. karroo (Hlu), and yet its altitudinal distribution is limited to lower elevations (Boon 2010). In our study area, A. tortilis grows at low elevations in hot areas and A. sieberiana occurs at mid to high elevations in warm to cool areas (Boon 2010). Acacia mearnsii is an Australian species, naturally occurring between 34 and 44 °S, and altitudes up to 885 m in south eastern Australia (Sherry 1971). It is an invasive species throughout savannas and grasslands at a range of elevations in South Africa (Bromilow 2010). As it naturally occurs in areas cooler than the range of most southern African Acacia species, it may grow better in cool climates and was used as a comparison with indigenous species.

Sites

South African grasslands and savannas are characterised by summer rainfall and winter drought (Mucina & Rutherford 2006). Annual rainfall varies between 400 and 2500 mm in grasslands and 200 and 1350 mm in savannas (Mucina & Rutherford 2006; Rutherford & Westfall 1986). Ten relatively flat sites were selected across a broad altitudinal gradient (c. 40–1700 m above sea level), from savannas to grasslands, encompassing a wide range of MATs and frost days (Schulze 1997) (Table 1). All sites were fenced to protect the seedlings from mammal herbivory. Ten seedlings of each taxon were planted 1 m apart at each of the ten sites between 27 October and 9 November 2006, giving a total of 600 trees planted and transported to sites over a distance of 500 km.

Table 1.   Characteristics of sites in the transplant experiment. Alt = Altitude; SHU = Summer Heat Units, where the sum of degree days above a 13 °C threshold below which growth does not take place, and summer is October to March, from Schulze (1997); # Growth Days = the number of days between 7 November 2006 and 30 April 2007 that had an average temperature of 13 °C or more, calculated from temperature data collected; # Frost days = predicted number of days that receive frost (Schulze 1997); MAP = Mean Annual Precipitation; MAT = Mean Annual Temperature; G = grassland; S = savanna; CB = Indian Ocean coastal belt; Lat = latitude; Long = Longitude. Sites were classified into Biome according to Mucina & Rutherford (2006). Weather data was extracted from Schulze (1997)
SiteBiomeAlt (m)SHU (° days)# Growth days# Frost daysMAP (mm)MAT (°C)Long (°S)Lat (°E)
  1. *Site data was not used in analyses as sites failed.

ErmeloG17041369179396991426.49029.967
WakkerstroomG16481472187377541527.40029.917
AtholeG15701539193208331526.60030.500
HleloG13531634191238751626.91730.550
CominsG1078187121827201727.71731.233
FordS808196321618411828.00031.100
MalanS554228923116231927.56731.567
Goss*S360253523216712127.55031.750
iMfolozi*S170265323316892128.23031.799
KirkwoodCB422500232010082128.41732.317

Each seedling was planted in 15-L bags of sandy clay loam from the Centenary Centre (S28.28956, E31.99089) in the Hluhluwe-iMfolozi Game Reserve and supplied with 70 g of slow-release fertiliser shallowly buried in the soil. The fertiliser contained 15% total N (7.1%inline image and 7.9%inline image), 9% P2O5, 9% K2O, 3.0% MgO, 0.02% B, 0.047% Cu, 0.40% Fe, 0.06% Mn, 0.020% Mo and 0.015% Zn (Osmocote Exact 15 + 9 + 9 + 3 MgO+Te 12–14M; Scotts, Bramford, Suffolk, UK). The bags were buried level with the soil surface and had drainage holes. Grass was mechanically cleared before planting and sites were weeded monthly. The plots were irrigated for 15 min every 2nd day and twice daily in the extremely hot months with a Gardena Water Computer (Basic T 1020; Gardena, Ulm, Germany) and a sprinkler. Where this was not possible, or when they failed, plants were watered manually three times a week. The need to irrigate varied with rainfall. Sites were classified into biomes according to Mucina & Rutherford (2006). Site air temperatures were recorded hourly at c. 1.3 m above the soil surface with DS1921G Thermochron iButton data loggers (Maxim, Sunnyvale, CA, USA) in a ventilated radiation shield. Summer heat units for each site were extracted from Schulze (1997), which are calculated from the sum of degree days between October and March above a 13 °C threshold, below which growth of fruit trees does not take place. Data on the growth of subtropical fruit trees in the study area were used as the closest equivalent to that of acacias, for which there is no data. The number of growth days was calculated as the number of days between 7 November 2006 and 30 April 2007 that had an average temperature of 13 °C or more, calculated from temperature data collected.

Growth measurements

Seedling height was measured in October or November 2006 and January, March and June 2007. The above-ground biomass of half the plants was harvested at the end of the growing season in June 2007. Leaves and stems of the cut material were separated, oven dried at 70 °C to constant weight, and weighed. Relative height growth was calculated from the slope of the logarithm of height over time.

Frost survival measurements

Survival and the degree of topkill by frost were recorded for the ten sites at the end of the cold season in October 2007. Complete topkill was used here to refer to death of the entire above-ground biomass. Different degrees of topkill were scored categorically as: ‘No topkill’, ‘a few tips killed (c. 25% of above-ground biomass)’, ‘most branches killed (c. 75% of above-ground biomass)’ or ‘complete topkill (100% above-ground biomass)’.

The ecological significance of variation in growth rates

We wanted to evaluate the ecological relevance of the variation in growth rates between individuals grown at our grassland and savanna sites. To do this, we used differences in growth to estimate the time it would take saplings to reach a height above the flame zone of fires and become adults (Higgins, Bond & Trollope 2000). We used biomass data from our experiment as a measure of seedling growth (Balfour & Midgley 2006; Schutz, Bond & Cramer 2010). The difficulty was that our transplanted seedlings were grown in controlled experimental environments and we needed to relate the growth of the experimental plants to sapling growth in the field. We made the assumption that if seedlings at our savanna sites were grown under natural conditions, they would achieve escape height in the same time as those recorded in a natural savanna environment within the study region (HiP plots, see Staver et al. 2009). Only the fastest-growing individuals are expected to escape the fire trap, and for this exercise, we use the fastest 20% of individuals (Wakeling, Staver & Bond 2011). We therefore assumed that the growth of the fastest c. 20% of our savanna seedlings [average biomass of the two largest individuals at each savanna site, B(sav avg)] is equivalent to the growth of the top 20% of individuals in HiP and hence, also equivalent to the time it would take the naturally grown HiP individuals to reach an escape height of 3 m [T(esc) HiP]. The proportion of the biomass of every individual in the transplant experiment to B(sav avg) is then used to scale T(esc)HiP to result in the expected time to escape height [T(esc)] for each plant, as follows:

image

T(esc) was compared between savanna and grassland sites using the average of the T(esc) estimates for all individuals in each biome. Species-specific T(esc)HiP estimates were used for A. karroo (Hlu), A. gerrardii and A. tortilis, with the A. karroo (Hlu) estimate also used for A. karroo (Bloem), A. mearnsii and A. sieberiana, which do not occur in HiP. T(esc)HiP for A. karroo (Hlu) was 4–5 and 9–10 years for A. gerrardii and A. tortilis.

Seasonality pot experiment

To supplement the altitudinal experiment, a common garden experiment was set up in one location and regular growth measurements were taken to follow the phenology of growth for a complete growing season. Seedlings of A. karroo (Hlu) and A. sieberiana were planted in 10 L bags of soil in November 2007 at the nursery at Hilltop research station in Hluhluwe Game Reserve (28°4.6′S, 32°2.5′E). Soils were collected from undisturbed areas near each of the ten sites. Plants were watered daily or as necessary depending on rainfall, and their heights were measured once a month from November 2007 to April 2008 (see also Wakeling, Cramer & Bond 2010). Average day length was calculated from sunrise and sunset times (Russell, Miller & Rind 1995

Statistical analyses

Statistical analyses were performed using statistica software (Ver. 8; StatSoft, Inc., Tulsa, OK, USA). Growth data were log transformed, and t-tests were used to compare grassland and savanna height and biomass in June and relative height increases from November to January and November to June for each species. T-test significance levels were corrected using the False Discovery Rate technique (Benjamini & Hochberg 1995).

Results

Two of the sites (Goss and iMfolozi) were discarded from growth analyses because of poor plant growth due to shading and a lack of reliable water supply, respectively. The lowest altitude site, Kirkwood (42 m above sea level), is in the Indian Ocean Coastal Belt Biome, a mosaic of edaphic grasslands, savannas and forest on Holocene sandy soils. It falls within the savanna biome in analyses of tree growth that compare grassland with savanna sites.

Climate differences between grassland and savanna sites

Sites in the higher-elevation grasslands had lower average monthly temperatures, fewer growth days and lower summer heat units than the lower-altitude savanna sites (Table 1, Fig. 2). Mean annual temperature of grassland and savanna sites were 15.4 and 20.0 °C respectively. Grassland sites experienced more frost days than savanna sites (Table 1). Average daily minimum temperatures in winter, calculated for the winter months June, July and August, ranged from 1 to 13 °C (Ermelo and iMfolozi sites respectively; Fig. 2). The lowest minimum temperatures recorded during the experiment occurred during a cold ‘snap’ in May 2007 (21st–24th), and ranged from −8 to 6 °C (Ermelo and Malan sites respectively). According to Agricultural Research Council long-term data, the minimum temperature at Ermelo, the highest coldest site, fell to −8 °C, or below, 16 times between 1950 and 2000, on average every 3 years (Schulze & Maharaj 2003).

Figure 2.

 Minimum temperatures recorded at each site during the cold period between 22 and 26 May 2007; winter average daily temperatures per site, calculated from recorded data; and the number of growth days for sites in savanna and grassland biomes along an altitudinal gradient. Sites are classified into biome according to Mucina & Rutherford (2006), and the lowest altitude site from the Indian Ocean Coastal Belt biome is grouped into the savanna biome. Site altitudes are shown with site codes (the first letter of each site name (see Table 1)).

Seedling growth

More growth occurred at savanna sites than grassland sites measured as total above-ground biomass, height and relative height growth (Fig. 3). Average above-ground biomass after one growing season (in June) for all species at savanna sites was twice that at grassland sites (t = −4.56, d.f. = 228, < 0.001). Biomass accumulation ranged between 1.4 (A. mearnsii) and 5 times (A. sieberiana) greater at savanna than at grassland sites; however, this difference was not significant for A. gerrardii and A. mearnsii (Fig. 3). Seedling heights at the start of the experiment (November 2006) were not significantly different between grassland and savanna sites (all species together: t = −0.21, d.f. = 478, > 0.05). However, at the end of the growing season, mean height was significantly greater at savanna than grassland sites, and for all species combined (t = −6.04, d.f. = 471, < 0.001) and for each species individually, except A. gerrardii (Fig. 3). Relative height growth was also significantly greater in savanna than grassland sites for all species combined (t = −3.40, d.f. = 466, < 0.001), but not for A. karroo (Bloem) and A. gerrardii (Fig. 3).

Figure 3.

 End of growth season (June) height and biomass, and relative height increase (November to January), between grassland and savanna sites for Acacia karroo (Hlu) (AK), A. karroo (Bloem) (AR), A. gerrardii (AG), A. tortilis, (AT), A. sieberiana (AS) and A. mearnsii (AM). Bars represent standard errors about the mean. Values above bars are the ratios of averages between savanna and grassland sites for each species. Stars indicate significant differences between grassland and savanna values for each species corrected by the False Discover Rate technique (Benjamini & Hochberg 1995) (***< 0.001, **< 0.01, *< 0.05).

Phenology of growth

Growth in the transplant experiment declined between January and March at both grassland and savanna sites while temperatures were still warm and plants were being watered (data not shown). Stem elongation of A. karroo (Hlu) and A. sieberiana in the pot experiment increased from November, peaked in January, and decreased sharply in February with negligible growth in March and April (Fig. 4). Average air temperature at this site increased from November to February, and only decreased in March and April (Fig. 4). Average day length increased from November to December and then decreased through to April (Fig. 4).

Figure 4.

 Average temperature (a) and day length (b) and relative stem elongation of the mean of two Acacia species (Acacia karroo (Hlu) and A. sieberiana) grown in a pot experiment from November 2007 to April 2008. See also Wakeling, Cramer & Bond (2010) and Schutz, Bond & Cramer (2009).

Frost damage and survival

Frost only impacted plants at the four highest altitude sites (Ermelo, Wakkerstroom, Athole and Hlelo; Fig. 5). Acacia gerrardii and A. sieberiana were the most frost damaged at the coldest sites, followed by A. karroo (Hlu), A. mearnsii and then A. tortilis. Acacia karroo (Bloem) was least affected by frost. At least one individual of each species survived at each site (Fig. 5). Lowest survival occurred at the coldest site, Ermelo, where all species had very high mortality (70–90%), except for A. karroo (Bloem), which had only 20% mortality. Survival increased rapidly with increasing temperature, and all species showed at least 80% survival (20% mortality) at Hlelo.

Figure 5.

 Frost effects on seedlings of six Acacia species along a gradient of decreasing temperature. The degree of topkill and percentage mortality were recorded after winter (in October) for each species at each site. There were 10 individuals per species per site. Bars with different shading represent proportions of individuals that showed different degrees of topkill: 100% topkilled (dark grey bars), 75% topkilled (light grey bars), 25% topkilled (white bars). Site codes represent the first letter of each site name (see Table 1). K, I, G, M and F are savanna sites, and C, H, A, W and E are grassland sites.

The ecological significance of variation in growth rates

When the biomass accumulated in this experiment was translated into the time it would take saplings to escape fires, A. karroo (Hlu) saplings in warm savanna sites would reach an escape height of 3 m (Trollope 1984) in 6 years, whereas those in cooler grassland sites would take 12 years (Table 2). For other species, saplings grown in the grasslands would take between 4 and 33 years longer (but only 1.7 years longer for A. mearnsii) to reach escape height compared with saplings grown in the savannas (Table 2).

Table 2.   Proportional differences in plant biomass between grassland and savanna sites in the transplant experiment were used to calculate proportional changes in sapling growth rates to escape height relative to savanna acacias measured at Hluhluwe-iMfolozi Park (Staver et al. 2009)
SpeciesGrasslandSavannaDifference in time to escape height between grassland and savanna systems (years)
Average time to escape height (years)SEAverage time to escape height (years)SE
Acacia karroo (Hlu)122.3 61.4 6
A. gerrardii  91.1 50.8 4
A. tortilis 4811.3143.834
A. karroo (Bloem)143.7 50.8 9
A. sieberiana 265.4 82.218
A. mearnsii  71.3 51.5 2

Discussion

Demographic bottlenecks have been shown to be important for determining tree densities in savannas (Higgins et al. 2007; Staver, Bond & February 2011). Slow growth rates decrease the success of plants at escaping these bottlenecks (Bond, Midgley & Woodward 2003a). We showed that growth was lower in the cooler grasslands than the warmer savannas. This would translate to a substantially longer time for saplings to grow to a height above the flame zone and hence, a lower probability of escaping to adult tree height in the interval between fires. We also showed that frost negatively affected seedling survival, but only at the coldest sites high above the savanna-grassland treeline.

Frost had an impact on the seedlings in our study mainly through topkill. The low temperatures experienced in our study occur approximately every 3 years in the study region, and thus, frost is not a rare event. Infrequent severe frosts are thought to exclude trees from other grasslands (Osmond et al. 1987; Fensham & Kirkpatrick 1992) and may help exclude trees from the most frost-prone grasslands of the Highveld. We recorded the effect of winter frosts on above-ground biomass lost and survival of seedlings. Mortality was infrequent at all but the highest sites, which were well above the savanna-grassland transition. As large areas of grassland occur where frost is rare or absent, we concluded that frost alone does not define the savanna treeline. All the species showed at least some survival at all sites and frost sensitivity was unrelated to the geographic distribution of the species. Acacia tortilis, for example, was among the most frost tolerant species but occurs at warm, low elevations in our study area. As A. gerrardii was completely topkilled at sites that experienced frost, this species’ distribution could be influenced by frost. Acacia sieberiana and A. mearnsii naturally occur in cool high-elevation regions, so it is surprising that they did not show better survival in frosted areas. The two varieties of A. karroo were the only examples where results were consistent with climatic distributions; the Hluhluwe variety, naturally occurring in warm frost-free areas, experienced more topkill and less survival than the Bloemfontein variety, indigenous to colder frosty areas. Acacia karroo (Bloem), evidently well adapted to cold, had only 20% mortality at the coldest site, Ermelo, relative to 70% or more for other species.

If frost is a contributing factor in the exclusion of trees from grassland areas in South Africa, it is likely be due to a loss of biomass, which renders a plant less likely to escape the fire trap. Where it occurs, frost may contribute to the fire trap (Holdo 2006). Larger trees with thick stems are less likely to be killed by frost (Sakai & Larcher 1987). Seedlings, which are less hardy than older plants (Nobel 1984; Bannister, Colhoun & Jameson 1995) and do not have the thermal buffering that thicker-stemmed individuals do (Sakai & Larcher 1987), have to endure the cold temperatures close to the ground (Bannister 1984; Nobel 1984; Sakai & Wardle 1978; Sakai & Larcher 1987; Bannister, Colhoun & Jameson 1995; Ball et al. 2002) and may experience annual loss of above-ground biomass due to frost. Frost-damaged seedlings must resprout from ground level which would increase the time needed to grow to a height which allows them to escape the effects of frost and fire. Seedlings growing near rocks (Nobel 1984) or under adult trees (Childes 1989; Holdo 2006) are sheltered from the full effects of frost, but it may be difficult for seedlings to establish in open grasslands without shelter of large trees. Seedlings in our study were not sheltered, yet survival was still high at all but the coldest sites.

Plants grown at grassland sites were significantly smaller, both in height and biomass, than savanna sites, when assessed across all species. The main factor that appeared to influence the difference in growth was growing season temperature. The lack of difference in growth of A. gerrardii between grassland and savanna sites may be due to the poor growth of this species across all sites. Growth comparisons between species are complicated by differing initial plant sizes, but across all species, there was a general trend for slower growth at upland grassland sites. Using these experimental growth rates adjusted to natural sapling growth, our results show that for all African Acacia species in this study, saplings in warm savanna climates would grow fast enough to allow them to escape the fire trap under intermediate to long fire return intervals, but those in cool grassland climates would grow too slowly to reach escape height under typical fire regimes. In comparison, A. mearnsii showed little difference in time to reach escape height between savanna and grassland sites and may be able to escape fires in grassland areas in long fire return intervals. This alien invasive is currently encroaching in many grassland areas in South Africa.

In savannas, the transition of saplings into adults is rare (Bond & van Wilgen 1996) and saplings may be trapped by fire for decades (Staver et al. 2009). Sapling transitions often occur in cohorts linked to local effects such as patches that do not burn, and landscape effects such as severe droughts when grass biomass is low, reducing competition and fuel for fires (Staver, Bond & February 2011). The length of a fire interval required to allow sapling release depends on sapling growth rates. Slow-growing plants require longer intervals between fires. Given the growth rates we recorded, saplings in grassland areas would require longer fire return intervals to escape than saplings in savannas. Savanna and grassland fire regimes are similar, with more variability between high and low rainfall areas within savannas and grasslands than between them (Tainton & Mentis 1984; Van Wilgen et al. 2000; Balfour & Howison 2001). Mean fire return intervals are 2.9–3.8 years respectively for mesic and arid savanna regions with rare longer intervals of up to 10 years in Hluhluwe-iMfolozi Park South Africa (Balfour & Howison 2001). In South African grasslands, annual and biennial burning is common in higher rainfall areas (>650 mm), but less frequent in lower rainfall areas (Tainton & Mentis 1984), and if fire is excluded for long periods, there is a shift from grassland to forest (Granger 1976; Westfall, Everson & Everson 1983). We have shown that saplings are unlikely to escape the frequent fires in grasslands, yet if fire were to be excluded for long enough to allow saplings to escape, the grassland would change to forest.

The phenology of the growing season is such that most stem elongation occurred before February in both the pot and transplant experiments. Other work has also shown savanna tree stem elongation to occur mainly in the early part of the growing season, rapidly declining from February or March, for Acacia species in southern Africa (Chidumayo 2001; Schutz, Bond & Cramer 2010), and a range of species in northern Australian savannas (Prior, Eamus & Duff 1997; Prior, Eamus & Bowman 2004). In temperate regions, both temperature and day length have been shown to be cues for cessation of growth at the end of a growing season (Greer & Warrington 1982; Greer, Wardle & Buxton 1989; Bannister, Colhoun & Jameson 1995; Fernández et al. 2003; Korner 2003). In savannas, phenology and inter-annual variation in phenology, have been related to moisture availability, temperature and photoperiod in different species and in different locations (Childes 1989; Williams et al. 1997; Jolly & Running 2004; Do et al. 2005; Archibald & Scholes 2007). Here, we have shown experimentally that savanna tree growth decreased in February–March, while plants were being watered and temperatures were still high. The cessation of growth did not appear to be related to temperature, but may be linked to the shortening day length.

Temperate trees stop growth in order to cold harden before cold temperatures begin (Sakai & Wardle 1978; Bannister, Colhoun & Jameson 1995; Fernández et al. 2003; Korner 2003). However, savanna trees do not need to be adapted to survive such low temperatures. Deciduousness and the associated lack of savanna tree growth has mostly been attributed to dry season drought (Prior, Eamus & Duff 1997; Jolly & Running 2004; Do et al. 2005). Prior, Eamus & Duff (1997) postulated that the lack of above-ground growth, despite high carbon assimilation during the latter part of the wet season and early dry season, was because plants were allocating resources below ground. Allocation of resources below ground is important for survival and sprouting after disturbances such as fire (Bond & van Wilgen 1996; Gignoux, Clobert & Menaut 1997; Wigley, Cramer & Bond 2008). Perhaps, the early cessation of above-ground growth is not simply due to a lack of moisture, but so that plants can allocate reserves below ground before dry season fires.

For some species in our study, biomass accumulation at savanna sites was five times greater than at cool grassland sites. This was greater than the effect of differing soils on growth, in which growth in savanna sites was up to 1.8 times greater than the same species growing in grassland sites (Wakeling, Cramer & Bond 2010); and the effect of variation in rainfall, which had a 0.5- to 1.5-fold impact on growth (Botha 2006). Biomass accumulation was more similar to both modelled and experimentally shown differences in growth between half and double the current concentrations of atmospheric CO2 (Bond, Midgley & Woodward 2003a,b; Kgope, Bond & Midgley 2010). Even though a common garden experiment in which acacias were grown in the soils from each site showed statistically significant growth differences, the effect size of soil differences was much smaller than climate-related site differences reported here (Wakeling, Cramer & Bond 2010), and unlikely to be great enough to prevent trees from escaping the fire trap in nutrient-poor grassland areas. A study of multiple factors together would be useful to tease out the most important ones. Sankaran et al. (2005), Sankaran, Ratnam & Hanan (2008) used regression analyses to highlight the most important factors that determine tree densities in savannas, but unfortunately did not include grassland areas in their study. In reality, many factors influence sapling growth, and the time it would take saplings to reach escape height in grasslands would be even longer than predicted in this study.

Conclusion

Upland C4 grasslands occur globally which suggests there may be a general explanation for their lack of trees. Unlike alpine treelines, the savanna-grassland pseudo treeline is not climatically controlled. Upland C3 grasslands may show similar dynamics (Norman & Taylor 2005; Fensham & Fairfax 2006; Fairfax et al. 2009) but are more likely to be climatically controlled (Rochefort & Peterson 1996). In the present study, we suggest that the interaction between cold temperatures and fire could account for the treeless state of upland grasslands in South Africa. To be globally applicable, it implies that all upland grasslands are maintained by fire (for example Norman & Taylor (2005) for North American temperate grasslands). However, this may not be true in all upland grasslands of the world. This study assumed that the sapling-to-adult demographic bottleneck is the most limiting and showed that a combination of slow plant growth and fire excludes trees. However, it is possible that an earlier life-history stage, seedling recruitment, is limiting and that grass competition alone can explain treeless grasslands. Grasses are known to be formidable competitors for tree seedlings (Knoop & Walker 1985; Partel & Wilson 2002; Jurena & Archer 2003), but grasses typically slow tree seedling growth rather than excluding recruitment altogether (Cramer et al. 2007; Cramer, Van Cauter & Bond 2009). Many studies of recruitment limitation have compared forest and savanna tree species (Bowman & Panton 1993; Hoffmann 2000; Hoffmann, Orthen & Franco 2004; Gignoux et al. 2009). We know of none that have compared recruitment in savannas versus grasslands. If grass competition was indeed sufficiently intense to exclude seedlings, the puzzle then would be to explain why upland grasslands exclude tree seedlings entirely, whereas savanna grasslands do not. We suggest that further study is needed to help reveal the mechanisms underpinning variation in tree cover in grassy ecosystems at this least-studied end point; where trees disappear all together.

Acknowledgements

We would like to thank EKZN Wildlife and a number of dedicated landowners for hosting this study; the staff of the Zululand Tree Project for field assistance and the Andrew Mellon Foundation for funding.

Ancillary