SEARCH

SEARCH BY CITATION

Keywords:

  • climate–vegetation relationships;
  • dynamic global vegetation models;
  • fire ecology;
  • global biomes;
  • plant biogeography

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix
  • • 
    This paper is the first global study of the extent to which fire determines global vegetation patterns by preventing ecosystems from achieving the potential height, biomass and dominant functional types expected under the ambient climate (climate potential).
  • • 
    To determine climate potential, we simulated vegetation without fire using a dynamic global-vegetation model. Model results were tested against fire exclusion studies from different parts of the world. Simulated dominant growth forms and tree cover were compared with satellite-derived land- and tree-cover maps.
  • • 
    Simulations were generally consistent with results of fire exclusion studies in southern Africa and elsewhere. Comparison of global ‘fire off’ simulations with landcover and treecover maps show that vast areas of humid C4 grasslands and savannas, especially in South America and Africa, have the climate potential to form forests. These are the most frequently burnt ecosystems in the world. Without fire, closed forests would double from 27% to 56% of vegetated grid cells, mostly at the expense of C4 plants but also of C3 shrubs and grasses in cooler climates.
  • • 
    C4 grasses began spreading 6–8 Ma, long before human influence on fire regimes. Our results suggest that fire was a major factor in their spread into forested regions, splitting biotas into fire tolerant and intolerant taxa.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix

It is generally believed that climate exerts the key control over the distribution of the world's major ecosystems. On different continents, with distantly related floras, similar vegetation formations occur under similar climatic conditions (Schimper, 1903). The distribution of the major biomes of the world – desert, tundra, grasslands, savannas and forests (tropical, temperate and boreal) – can be broadly predicted from temperature and precipitation (Holdridge, 1947; Whittaker, 1975) and correlate well with water balance (Woodward, 1987; Stephenson, 1990). Ecosystem properties, such as biomass, leaf area and net primary productivity, also vary along gradients of temperature and moisture. Total plant biomass increases with temperature, for example increasing from 650 g m−2 in Arctic tundra to 8300 g m−2 in boreal forests to 26 700 g m−2 in temperate forests (Chapin et al., 2002).

Here we argue that several of the world's major biomes owe their distribution and ecological properties to the fire regime. Fire is under-appreciated as a global control of vegetation structure, even though fires are a common and predictable feature of the world's grasslands, savannas, mediterranean shrublands and boreal forests (e.g. Goldammer, 1993; Archibold, 1995). Together these fire-prone formations cover some 40% of the world's land surface (Chapin et al., 2002). Fires always reduce plant biomass and, depending on their frequency and severity, can also replace trees with shrublands or grasslands. The implication therefore is that some flammable ecosystems may be far from physiognomic limits set by climate. There is direct and indirect evidence that this is often the case. First, patches of forest are common in many landscapes dominated by fire-prone grasslands and shrublands suggesting a mismatch between climate and vegetation (Sarmiento, 1983; Bond & van Wilgen, 1996; Midgley et al., 1997; Bowman, 2000). Second, experimental exclusion of fire has often led to biome switches from flammable formations to forested ecosystems (Pickett & White, 1985; Bond & van Wilgen, 1996). Third, anthropogenic fires introduced to island ecosystems have transformed forests to flammable shrublands and grasslands (e.g. Hawaii, D’Antonio & Vitousek, 1992, New Zealand, Ogden et al., 1998, Madagascar, Koechlin, 1972). Finally, plantation forestry, and the invasion of nonnative trees into flammable grasslands and shrublands (Richardson, 1998) shows that tree biomass in these ecosystems is far from the limit set by climate. These observations suggest that fire may be a primary factor in determining biome distributions, promoting flammable ecosystems where the climate can support forests.

In this paper, we provide the first global assessment of the importance of fire in determining world biome distribution. We do so by asking: how different the distribution of global biomes would be if we could ‘switch fire off’; and to what extent would global vegetation change if fires were suppressed and succession allowed to proceed until the growth forms present were limited only by climate? To address these questions, we used simulation models to predict global ecosystem structure and growth form composition as if plant growth were limited only by climate. Until recently, analyses of determinants of the global distribution of vegetation have been largely correlative. Correlative methods cannot discriminate between the roles of climate and fire. In the last decade, process-based models for predicting global vegetation have become available. Dynamic Global Vegetation models (DGVMs) are designed to simulate vegetation responses to changing climates. DGVMs ‘grow’ plants according to physiological processes (Woodward et al., 1995; Haxeltine & Prentice, 1996; Cramer et al., 2001). They simulate carbon and water dynamics and structure of vegetation using input data of climate, soil properties, and atmospheric CO2 (Woodward et al., 1995; Beerling & Woodward, 2001; Cramer et al., 2001). The models generate predictions of the composition and structure of vegetation for a given climate in terms of relatively few plant functional types (PFTs, e.g. Woodward et al., 1995; Haxeltine & Prentice, 1996). Several DGVMs include fire modules (Cramer et al., 2001). No mechanistic model to generate fire on a global scale exists. Instead DGVMs simulate fire from empirical relationships between moisture content of plant litter (which can be simulated from climate) and fire return intervals (Thonicke et al., 2001; Venevsky et al., 2002). The fire modules assume that ignition is not limiting (Woodward et al., 2001).

DGVMs provide a useful biogeographical tool for exploring potential vegetation. Because they are based on an understanding of the first principles of plant photosynthesis, carbon allocation and growth, DGVMs allow the simulation of global ecosystem structure and growth form composition as if plant growth were limited only by climate. The real global biome distribution can then be compared with the simulated climate potential vegetation to ascertain the importance of fire in determining global biome distribution. In this paper, we use the Sheffield Dynamic Global Vegetation Model (SDGVM; Woodward et al. 1995, 2001) to investigate the importance of fire vs climate as determinants of global biome distribution. The SDGVM is a global-scale model that includes a fire module. Output of the SDGVM has been tested against measured ecosystem properties over a wide range of climates worldwide and gives a satisfactory fit (Cramer et al., 2001; Woodward et al., 2001). The DGVM is particularly useful for our purpose because the model is mechanistic and not based on correlations of existing vegetation with climate. We were therefore able to separate effects of climate from those of fire by ‘switching off’ the fire module in the simulations. We used long-term fire experiments to test model simulations of woody biomass and dominant plant functional type. By comparing model simulations with global maps of landcover, tree cover and the distribution of fires, we could assess the extent of fire-controlled vs climate-controlled global biome distribution.

Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix

The basic workings of the DGVM are described in Appendix 1. Climate data for DGVM simulations were taken from the University of East Anglia global data set for the 20th century. The DGVM incorporates soil depth and texture from a global database (FAO, 1998). It assumes soils are freely drained. Model output includes ecosystem properties, such as plant biomass, and also the cover of several major growth forms. Stem biomass (above-ground woody biomass) indicates relative dominance of trees and is therefore a pointer to biome type. To test model output, we compared simulated above-ground biomass with measured above-ground biomass reported for five long-term fire experiments.

There are difficulties in using long-term fire-exclusion experiments to test ‘climate potential’ in terms of biomass. Although there are many such experiments, results are generally reported as changes in cover or density of trees and other growth forms, rather than as changes in biomass. A second problem is that many decades of fire exclusion may be needed before woody plants colonise a site and grow to their climate-limited potential biomass. For much of the data available, the effects of fire exclusion are best measured qualitatively as a tendency for increased woody cover. Three qualitative outcomes can be expected from long-term fire exclusion experiments:

  • 1
    no change (vegetation is climate limited)
  • 2
    increased density or size of woody plants but no change in species (climate-limited, fire modified)
  • 3
    increased density and size of woody plants and successional tendency to forest with invasion of fire-sensitive trees and shrubs (fire-limited).

The third case indicates meta-stable vegetation with alternate fire-dependent or climate-dependent states. As an additional test of the utility of the SDGVM for exploring determinants of vegetation, we simulated plant functional types (PFTs) with ‘fire on’ and ‘fire off’ for the southern African region. The region is relatively arid, with semiarid shrublands in the west, grasslands and savannas in the east, and Mediterranean shrublands in the south-west. Small patches of forest occur throughout the higher rainfall regions suggesting the potential for biome switches (Midgley et al., 1997; O’Connor & Bredenkamp, 1997). We compare the simulations with results from numerous long-term fire exclusion studies in the region (Bond et al., 2003a).

We simulated the dominant growth form based on both cover and biomass for global comparisons. Cover, in mixed tree/grass ecosystems, emphasises grasses, while biomass emphasises trees because of the large amount of biomass contained in tree stems. As indicators of major biomes we used model output for four key growth forms: gymnosperm trees (mostly conifers, deciduous and evergreen), angiosperm trees (deciduous and evergreen), temperate grasses or shrubs with C3 photosynthesis (‘C3’) and tropical grasses with C4 photosynthesis (‘C4’). Areas of low cover or biomass are indicated as ‘bare’. We compared simulated global vegetation with ‘fire off’ to a map of observed vegetation. Producing a global map of vegetation is not without its own problems of interpretation. A number of vegetation maps are available (e.g. Matthews, 1983; Olson et al., 1983; Haxeltine & Prentice, 1996; Hansen et al., 2000). We used the land cover map produced by ISLSCP (Meeson et al., 1995), which shows dominant functional types similar to those simulated by the DGVM. The land cover map is primarily determined from the annual variations in a satellite-derived vegetation index, Normalized Difference Vegetation Index, NDVI, for each 1° × 1° pixel of the terrestrial surface. The approach (De Fries & Townshend, 1994) builds on previously established techniques of analysis and classification of NDVI data (Los et al., 1994; Sellers et al., 1994). In addition, the classifications based on the NDVI data have been trained, and therefore constrained, by established vegetation maps, such as those of Matthews (1983) and Olson et al. (1983). The ISLSCP map includes land cover modified by agriculture and so is an attempt to map actual vegetation. The map derived by the SDGVM is for potential vegetation and does not account for any human impacts on vegetation.

Since reduction in tree cover is one of the major effects of fire, we also compared median tree cover for the 20th century, simulated with and without fire, with a satellite derived map of global tree cover (FAO, 2001). The FAO tree cover map was derived from satellite imagery for the period from 1995 to 1996 obtained from the Advanced Very High Resolution Radiometer (AVHRR), and archived in the Global Land Cover Characteristics Database (GLCCD). This imagery consists of five calibrated AVHRR bands and a NDVI band. The preliminary map was reviewed by experts from around the world and tested against International Geosphere Biosphere Programme (IGBP) validation points and full land-cover data sets from the governments of USA and China. The evaluations showed that the average accuracy of the maps for all tree cover classes is about 80% with greatest accuracy for closed forest (FAO, 2001).

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix

Biomass change

Where fires are frequent, woody biomass should be reduced relative to climate potential. Fig. 1 compares stem biomass in frequently burnt savannas with plots where fire has been excluded for at least 30 yr in experiments from Africa, the USA and Venezuela. These studies show very large differences in woody biomass between frequently burnt and unburnt treatments in the more humid sites. The biomass differences are comparable in magnitude with differences between major biomes such as tundra and temperate forests (Chapin et al., 2002). The DGVM simulations with ‘fire off’ showed a good fit between observed and simulated maximum biomass for fire exclusion treatments, although biomass was somewhat underestimated for two of the African savannas. The ‘fire on’ simulations showed a good fit between median biomass and observed data at relatively arid sites but overestimated woody biomass at more mesic sites. In part, this is because the fire module generated too few fires in more humid climates. The simulated fire return intervals (fri) for the 20th century were 200 yr for the North America site and 73 yr for the Venezuela site, both of which burnt at intervals of 2–5 yr. By contrast, the African sites had simulated fris of 3–5 yr, close to the actual fri of 2–5 yr.

image

Figure 1. Above-ground woody biomass in savannas with ‘fire on’ (burnt every 2–3 yr) and ‘fire off’ (unburnt for 40+ yr). (a) Measured biomass in g m−2 dry weight; (b) simulated biomass using the Sheffield Dynamic Global Vegetation Model (SDGVM; maximum values for fire off, median values for fire on). Sites are Kruger National Park (Shackleton & Scholes, 2000); Zimbabwe 1, Matopos (Kennan, 1972, P. Frost pers. comm., 2003), Zimbabwe 2, Marondera (Barnes, 1965; Tsvuura, 1998, and P. Frost pers comm., 2003); Venezuela (San Jose et al., 1998); Cedar Creek, USA (Tilman et al., 2000).

Download figure to PowerPoint

Regional biome simulations

Fig. 2 shows simulated tree cover (angiosperms) for southern Africa with and without fire. The simulations of ‘fire on’ are consistent with the actual vegetation, which is grassland with very low tree cover except near the eastern sea-board (savanna vegetation), and in the south-west, which supports evergreen forests. The ‘fire off’ simulation shows a striking contrast with trees dominating all the higher rainfall regions of the east. The simulations imply that most of this region would be forest in the absence of burning. The figure also shows the locality of a number of fire exclusion studies and whether exclusion treatments resulted in biome switches (to fire-intolerant forest) or merely structural or no change as defined above (Bond et al., 2003a). The simulations are generally consistent with the results of fire exclusion studies (Bond et al., 2003a).

image

Figure 2. Tree cover for southern Africa simulated with and without fire by the Sheffield Dynamic Global Vegetation Model (SDGVM). Fire exclusion studies are indicated on the ‘without fire’ simulation. Black rimmed circles indicate sites in which there was a successional trend to closed forest; white rimmed circles indicate sites that showed no trend to forest (see Bond et al., 2003b for sources).

Download figure to PowerPoint

Global biome simulations

Functional types  The regional test of the SDGVM gives some confidence in the use of the model for simulating global biome distribution as affected by fire. Fig. 3 shows the ISLSCP landcover map (Meeson et al., 1995) of the world using similar broad growth form categories to the DGVM output (see Table 1 for ISLSCP and DGVM map units). The locations of several long-term fire exclusion studies are also indicated on the map and the successional trends reported for the experiments are shown in Table 2. Fig. 4 shows the dominant growth form, as measured by relative cover, simulated with fire ‘off’. Simulations of the dominant growth form as measured by biomass produced similar results with only slightly larger areas of C4 and C3 cover and are not shown here. A map of fires in 1998, derived from satellite imagery, is shown in Fig. 5 to give an indication of the global distribution of fires in a single year. A large proportion of C4 grassy ecosystems burn on an annual basis relative to other biome types.

image

Figure 3. ISLSCP Landcover according to dominant functional types. See Table 1 for conversion of landcover classes to dominant functional types. Squares indicate the location of long-term fire exclusion studies listed in Table 2. Source: ftp://daac.gsfc.nasa.gov/data/inter_disc/biosphere/land_cover

Download figure to PowerPoint

Table 1.  ISLSCP Mapping units and plant functional type (PFT) equivalents in the dynamic global vegetation model (DGVM) simulations used in Fig. 3
ISLSCP numberLand cover typeDGVM PFT
1Broadleaf evergreen forestAngio
2Broadleaf deciduous forest and woodlandAngio
3Mix of 2 and coniferous forestAngio
4Coniferous forest and woodlandGymno
5High latitude, deciduous forest and woodlandGymno
6Wooded C4 grasslandC4
7C4 grasslandC4
9Shrubs and bare groundBare
10TundraC3
11Desert, bare groundBare
12CultivationCultivation
13IceBare
14C3 wooded grassland, shrublandsC3
15C3 grasslandC3
Table 2.  Long-term fire-exclusion studies in savannas (see Fig. 3)
SiteLocalityYears of fire protectionTrendMAP mmT °CSource
  1. In all studies, tree density increased with fire protection. Successional trends were obtained from the listed sources. Forest, change to closed tree canopies with no grass understorey, shift to forest tree species; woodland, increase in tree cover with canopies closing but sufficient grass to carry a fire, increase in fire-intolerant species; savanna, no tendency to closed tree cover, no change in tree composition. Y, years of fire protection; MAP, mean annual precipitation; T, mean annual temperature.

1USA   35Forest 790 6.0Tilman et al. (2000)
2Venezuela> 100Forest124927.9San Jose et al. (1998), San José & Farinas (1983)
3Brazil   18Woodland149121.3Moreira (2000)
4South Africa   45Savanna 54812.0Shackleton and Scholes (2000)
5Zimbabwe   46Woodland 58118.2Kennan (1972); P. Frost (pers. comm. 2003)
6Zimbabwe   46Woodland 92417.2Barnes (1965); Tsvuura (1998); P. Frost (pers. comm. 2003)
7Zambia   22Forest120025.6Trapnell (1959)
8Ghana   32Forest119034.5Swaine et al. (1992)
9Australia   15Woodland/forest142028.2Bowman and Panton (1995)
image

Figure 4. Distribution of dominant functional types measured by cover and simulated with ‘fire off’.

Download figure to PowerPoint

image

Figure 5. Global distribution of fire in 1998 mapped by ATSR-2 World Fire Atlas (European Space Agency). Note that most fires occur in C4 grass ecosystems. Source: ATSR World Fire Atlas web page: http://shark1.esrin.esa.it/FIRE/AF/ATSR/.

Download figure to PowerPoint

Tree cover Fig. 6(a,b) shows simulated angiosperm tree cover with ‘fire off’ and ‘fire on’. The FAO map of tree cover is shown in Fig. 6(c) for comparison. The DGVM simulated greater tree cover than that recorded in the FAO map probably, in part, because of the underestimation of fire frequency in humid C4 grasslands. However areas with closed forest in the FAO classification (60–100% tree cover) show a good correspondence with areas with a simulated tree cover of 80–100% in the DGVM simulations. Fire has a significant effect on the extent of global forest cover. According to the simulations, forest cover (80–100% tree cover) would double from 26.9% of vegetated grid cells to 56.4% in the absence of burning. More than half (52.3%) of grid cells with C4 grasses present (> 20% cover) in the ‘fire on’ simulation would change to closed angiosperm forest in the absence of burning. None would be replaced by gymnosperm forest. Ecosystems with C3 grasses or shrubs (> 20% cover) are somewhat less dependent on fire with 41% being lost to forest in the ‘fire off’ simulations. Of these, 53% would be replaced by gymnosperm, 34% by angiosperm, and 13% by mixed forests

image

Figure 6. Tree cover (%) (a) simulated with ‘fire off’ (b) simulated with ‘fire on’ (c) observed tree cover derived from satellite imagery in 2000 (FAO, 2001). Simulated cover values are median tree cover for 20th century simulations. Observed tree cover classes are: 40–100%, closed forest with no grass understorey; 10–40% more closed forms of savanna and other types of ‘forest’; 5–10% scattered trees. The map does not discriminate between natural forests and plantations.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix

The aim of our study was to evaluate the extent to which fire determines the global distribution of vegetation after climate limitations are accounted for. Where would global vegetation change if fires were suppressed and succession allowed to proceed until the growth forms present were limited only by climate? Comparisons of the ISLSCP land cover map (Fig. 3) with fire-off simulations (Fig. 4) suggest that biomes of large parts of the world are far from their climate potential supporting flammable formations (Fig. 5) such as grasslands and savannas. We label these fire-dependent ecosystems. They are fire-dependent in the sense that the dominance of grasses or shrubs (measured as biomass or cover) depends on burning. FDEs are meta-stable. In the absence of burning, FDEs would be replaced by forest. FDEs, according to the simulations, are of much greater extent in the tropics and southern hemisphere than the temperate and boreal north (Figs 3–5). Vast areas of Africa and South America, and smaller areas of Australia are dominated by C4 grassy ecosystems, which have the climate potential to support woodlands and forest. Satellite maps of fire occurrence show that these biomes experience by far the most extensive fires in the world (Fig. 6; Dwyer et al., 1998; Barbosa et al., 1999; Dwyer et al., 2000; Tansey et al., 2004). In the northern hemisphere, outside the tropics, these systems are of much smaller extent, occurring as prairies in North America, and in relatively small areas of savanna in India, south-east Asia and northern China. It is perhaps not coincidental that fire is barely mentioned in general ecological textbooks written in the northern hemisphere (Bond & van Wilgen, 1996). By contrast, fire ecology has been a central theme of ecologists in Australia (e.g. Gill, 1975; Whelan, 1995; Bradstock et al., 2002) and Africa (Phillips, 1930; Booysen & Tainton, 1984; Bond & van Wilgen, 1996). In South America, too, there is an extensive literature on fire in savannas (e.g. Beard, 1944; Coutinho, 1982, 1990; Sarmiento, 1983; Hoffmann, 1999; Hoffmann & Franco, 2003) but a larger literature on forests.

Four of the world's major biomes experience frequent and/or predictable fire: temperate and tropical grasslands and savannas, Mediterranean shrublands, and boreal forest (Archibold, 1995). Our results point to grasslands and savannas, dominated by C4 grasses, as by far the most extensive FDEs. Most tropical and subtropical grasslands and savannas are not at their climate potential according to these simulations and would be replaced by woodlands and forests in a ‘fire off’ world. However the simulations also point to significant areas of climate-limited C4 ecosystems in the more arid regions of the world, which are not dependent on burning (Fig. 4). Mediterranean shrublands are of much smaller extent but also have the climate potential to be forest, not shrublands, according to the simulations. The implication is that fire, and not just a combination of winter rainfall and summer drought (Specht, 1969; Mooney, 1977), is responsible for the peculiar dominance of shrubs. Mediterranean climate regions are characterised by steep climate gradients and the global simulation averages conditions over heterogeneous landscapes. At finer scales of resolution, a mosaic of fire-maintained and climate maintained shrublands seems more likely. The third major fire-prone biome, boreal forests, are often dominated by fire-adapted trees with serotinous cones that release seeds only after crown fires (Johnson, 1992; Keeley & Zedler, 1998). However, by our measure of fire dependence, the dominance of the gymnosperm tree growth form does not depend on burning according to the simulations. If fire dependence were measured by changes in species composition, rather than broad functional type, large areas of boreal forest (and other ecosystems?) might be considered ‘fire-dependent’.

Testing the simulations

‘Fire-off’  DGVMs are relatively new tools for exploring biogeographical questions. How valid are the simulations? The ‘fire-off’ simulations of stem biomass gave a reasonable fit to observed biomass in vegetation where fire had been excluded for long periods (Fig. 1). Simulations of tree cover, a functional type which varies greatly with fire, also gave a satisfactory fit to southern African vegetation. The ‘fire on’ simulations predicted low tree cover (Fig. 2), consistent with the grassy and shrubby vegetation that dominates most of the region (Acocks, 1953; Cowling & Richardson, 1997). ‘Fire-off’ simulations are strikingly different with forests dominating the more humid eastern and south-western areas. Studies of successional trends following long-term fire exclusion support these predictions (Fig. 2; Bond et al., 2003a). The locations of several experimental studies from other parts of the world are shown in Fig. 3. All of these are consistent with the ‘fire-off’ simulations, with those in more humid sites showing successional trends to woodland or closed forest (Table 2). We suspect there are many more fire-exclusion experiments around the world that could be used to test the DGVM simulations but no global compilation is available.

A major problem with using fire exclusion experiments to test the potential for biome switches is the time lag before all potential forest trees have colonised an area and grown to maturity. However there are many ‘natural experiments’ where, in landscapes dominated by flammable ecosystems, forest patches persist in fire refugia adjacent to water bodies, in deep ravines, and at the edges of barren or rocky areas (Sarmiento, 1983; Furley et al., 1992; Bond & van Wilgen, 1996; Bowman, 2000). Since forest patches often differ in other environmental factors too, there has been much debate on which of them limit forest distribution (e.g. Manders, 1990; Furley et al., 1992; Bowman, 2000). Bowman (2000) has recently comprehensively reviewed the evidence on factors limiting rainforest distribution in Australia. He concluded that intolerance of recurrent fire, rather than climate or soil, is the only factor that consistently explains the distribution of Australia's archipelago-like rainforest patches in a sea of flammable vegetation.

Potential for afforestation also provides clues on the extent of climate control of vegetation. Many grasslands and shrublands in the southern hemisphere have been planted up to conifers and eucalypts or have been invaded by these trees (Richardson, 1998). Replacement of these systems by trees of much greater biomass is one indication that the native vegetation is not at climate potential. In the Cape region of South Africa, Le Maitre et al. (1996) reported above-ground biomass for fynbos, a flammable Mediterranean shrubland, of c. 1500–3500 g−2 where the SDGVM predicted 1000–2600 g−2 for ‘fire-on’. In the same landscapes, they reported biomass of invasive conifer forests of 11 500–18 600 g−2 close to the SDGVM prediction of 12 000–22 000 gm2 for ‘fire off’ for these localities. Both empirical and simulated data support the idea that these shrublands are fire-maintained and not at their climate potential.

Fire-on simulations  Although there is clearly room for refinement, the SDGVM predictions of climate-limited (‘fire off’) biome properties are well supported by available evidence. The ‘fire on’ simulations are more problematic, especially for humid grassy areas. The DGVM simulated very low fire return intervals (200 and 73 yr, respectively) for the North American and Venezuelan sites (Fig. 1) over the 100-yr simulation period. In reality, both sites burnt at intervals of 2–5 yr (San Jose et al., 1998; Tilman et al., 2000). Other attempts to simulate fire for DGVMs also underestimate fire frequency in humid savannas. Thonicke et al. (2001) predicted fire return intervals of 50 to > 200 yr for humid savanna regions of Brazil, Africa and tall grass prairies in North America. In practise, these C4 grass-dominated systems may burn every year and at least several times in a decade. Current versions of fire modules used in DGVMs are therefore likely to greatly overestimate tree cover in humid savannas (Fig. 1; Venezuela site). This is apparent in comparisons of simulated and observed tree cover (Fig. 6b, c). While the tree cover classes are difficult to equate, the SDGVM cover class of > 90% tree cover for the ‘fire on’ simulation is largely coincident with the FAO ‘closed forest’ cover class. However the SDGVM simulated high tree cover in, for example, the Brazilian cerrados. This vast (> 2 million km2) region of humid savannas (mean annual rainfall of 800–2000 mm) has strikingly low tree cover, especially when compared with the patches of closed forest that occur as isolated patches within cerrado (Sarmiento, 1992; Ratter et al., 1997; Oliveira-Filho & Ratter, 2002). Since humidity is a major driver of current fire modules in global DGVMs, it is not surprising that simulated fire frequencies are greatly underestimated in humid savannas. More work is needed on developing fire models, perhaps by including differences in flammability among PFTs, to improve simulations of the very high fire frequencies of humid tropical grassy biomes. However our aim was to determine which parts of the world support vegetation very different from the potential set by climate. To this end, the ‘fire-off’ simulations of DGVMs are currently the best available tool for exploring the biogeography of a world without fire. It is clear that humid savannas are the prime candidate.

Alternative explanations for nonforested ecosystems

Our study indicates large differences between climate potential and actual vegetation, especially in tropical grasslands and savannas. But is fire really the culprit? There has long been a debate on determinants of savanna distribution (Frost & Robertson, 1987; Scholes & Archer, 1997). Savannas generally occur in seasonally dry climates. But, as revealed by the DGVM simulations, fire exclusion studies, the presence of closed forests in savanna landscapes, and extensive conversion of humid grassy ecosystems to plantation forestry, seasonally dry climates can also support closed forests. In South America, forests occur over the entire rainfall range of savannas (Sarmiento, 1992; Oliveira-Filho & Ratter, 2002). Soil factors are frequently invoked to explain the absence of tree-dominated vegetation. Seasonally waterlogged soils are a common feature of bottomlands and lower slopes of soil catenas in many tropical savanna landscapes of low relief. Grasslands typically dominate on these soils with few, widely scattered trees. The Pantanal, a vast South American wetland (c. 400 000 km2), is probably the only area, at a global scale, where seasonal waterlogging is so extensive that it can account for the sparse tree cover (Eiten, 1975).

Low soil nutrients in ancient weathered landscapes, including most of Australia and large areas of South America and Africa, have also been invoked as an explanation for the lack of forest in humid regions (e.g. Cole, 1986 for savannas; Specht & Moll, 1983 for shrublands). However long-term fire exclusion studies and growth experiments suggest that, although soil nutritional properties may influence the rate of tree invasion into grasslands and shrublands, they do not prevent tree incursion (e.g. Kellman, 1984; Manders, 1990; Bowman & Panton, 1993). Forest trees tend to accumulate nutrients more than savanna trees and such enrichment may sometimes (but not always; e.g. Hoffmann & Franco, 2003) be an essential precursor to their invasion of fire-protected savannas (Bowman & Fensham, 1991). Where fires are frequent, any factor that slows tree growth will tend to favour dominance by fire-tolerant shrubs or grasses (Kellman, 1984).

Other than fire, and anthropogenic activities, few (if any) disturbance agents reduce tree biomass at a global scale. The extent to which herbivores control ecosystem structure has been long debated (Hairston et al., 1960; Polis, 1999). Fire, as an alternative consumer of plants, has not been part of this debate. Yet Africa, despite having the largest extant diversity of ungulate herbivores, also has the most frequent and extensive fires in humid grassy ecosystems (Fig. 5; Barbosa et al., 1999; Dwyer et al., 2000). Grasses of the humid tropics are typically coarse and inedible and support low grazer biomass (Bell, 1982). In Africa, large elephant populations confined in protected areas have major impacts on tree cover (e.g. Cumming et al., 1997). They may (or may not) have been influential in reducing tree cover over larger areas in the past. Stand die-back from insect out-breaks is a common feature of higher latitude forests and can limit tree biomass, especially of conifers (Kurz & Apps, 1999). These and other disturbance agents may be of local importance in limiting tree biomass. None have the global extent and influence of biomass burning (Fig. 5).

Origins of fire-dependent biomes

The vast extent of flammable biomes, especially in the tropics and subtropics, has often been attributed to anthropogenic burning. Although anthropogenic fires have undoubtedly extended areas of flammable vegetation, there is now abundant evidence that natural fires occurred long before humans (Scott, 2000) and that flammable ecosystems predate anthropogenic burning by millions of years. C4 grassy ecosystems first began to form a distinct vegetation type some 6–8 Ma according to isotope evidence from fossil bone and soil carbonate (Cerling et al., 1997). Their appearance has been attributed to decreasing atmospheric [CO2], which favours the C4 photosynthetic mechanism (Ehleringer et al., 1997, but see Keeley & Rundel, 2003), but increased fire frequencies must have been a major factor in their rapid spread at the expense of forests (Sage, 2001; Bond et al., 2003b; Keeley & Rundel, 2003). It is interesting to note that these grassy ecosystems were even more extensive at the last glacial maximum (Harrison & Prentice, 2003) when anthropogenic effects were minor but fires continued to burn (Scott, 2002).

The very extensive flammable formations in Australia (woody and grassy) seem, from fossil charcoal and palynological records, to have begun carving out forests from the Miocene (Bowman, 2000; Kershaw et al., 2002; Hassell & Dodson, 2003). Mediterranean shrublands, whose origin has usually been attributed to the onset of mediterranean-type climates, seem also to have expanded in the late Tertiary more or less coincidentally with flammable C4 grassy biomes (California Axelrod, 1989, Europe Herrera, 1992, Australia Hassell & Dodson, 2003; South-west Africa Linder, 2003).

If the extent of fire-dependent ecosystems were an anthropogenic artefact, then the biota should reflect a very recent origin with just a few widespread species profiting from burning along with some fire-tolerant survivors. Flammable grassy ecosystems with just such characteristics occur on islands such as Madagascar and Hawaii altered by relatively recent human settlement (D’Antonio & Vitousek, 1992). However the biota of flammable ecosystems on continents show evidence of greater antiquity. Among grasses, the Androgoponeae dominate many humid grassy ecosystems with the climate potential to form forests worldwide (Hartley, 1958; Barkworth & Capels, 2000). They have several characteristics that promote frequent fires (Bond et al., 2003a). The sudden appearance of C4 grass-fuelled fire regimes in the late Tertiary must have presented a formidable obstacle to tree recruitment and survival, especially in the context of falling CO2 levels (Bond et al., 2003b). Tree floras of flammable formations would be expected to have evolved independently on different continents with little time for dispersal across ocean barriers between continents. Evidence for this is that dominant tree taxa differ from one savanna region to the next and have diversified greatly within regions. Examples include Eucalyptus (Myrtaceae) in Australia (Ladiges et al., 2003), Caesalpiniaceae and Acacia in Africa (White, 1983), Dipterocarpaceae in savannas of south-east Asia (Stott, 1988; Stott et al., 1990), and diverse lineages in the vast cerrados of Brazil (Sarmiento, 1983; Ratter et al., 1997; Hoffmann & Franco, 2003). From the few studies available, savanna tree species are not a fire-tolerant subset of forest tree species. They are generally endemic to flammable formations with an entirely different suite of species occurring in closed forests (Prance, 1992; Sarmiento, 1983 for South America; Bowman, 2000 for Australia; White, 1983 for Africa). However, at least in some genera, sister taxa in forests and savannas have apparently arisen independently several times (Prance, 1992; Hoffmann & Franco, 2003) and few genera, and no families appear to be endemic to fire-dependent grassy biomes consistent with the relatively recent origin of flammable floras. Our point is that the global extent of fire-dependent ecosystems is not merely an artefact of recent anthropogenic burning. They have existed long enough to evolve distinctive biotas.

Conclusion

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix

Although the importance of fire in determining vegetation structure and composition has been extensively studied in many parts of the world, this is the first integrated report on the global extent of fire-dependent ecosystems. It is made possible by the recent development of DGVMs, which, for the first time, allow an evaluation of the mismatch between climate potential and actual world vegetation measured, largely, by the importance of trees. The reduction of trees by fire has resulted in the evolution of some of the most biodiverse ecosystems in the world, and facilitated the rise of essentially modern C4 grass-dominated floras and associated faunas. The great extent of apparently fire-dependent grasslands and shrublands raises a number of questions. What limits the occurrence of fire and what determines particular fire regimes? What caused changes in fire regimes in the past initiating the spread of FDEs? How do humans alter fire regimes, often promoting, but also consciously or inadvertently, suppressing fires? How should fire be incorporated in global change scenarios, not only through atmospheric impacts of biomass burning but also as a major determinant of global ecosystem structure and composition? Answers to these questions will help fill the large gaps in our current understanding of global ecology and biogeography.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix

Financial support was provided by the National Research Foundation of South Africa and the Andrew Mellon Foundation. We thank Mark Lomas for help with the DGVM simulations, Peter Frost for providing the Zimbabwe fire data, Braulio Dias for helpful advice on South America, Jeremy Midgley and Sally Archibald for constructive criticism, Dave Bowman, Ross Bradstock, Jon Keeley and CJ Fotheringham for fruitful discussions on fire and biogeography.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix
  • Acocks JPH. 1953. Veld types of South Africa. Memoirs of the Botanical Survey of South Africa. Pretoria, South Africa: Government Printer, 28: 1192.
  • Archibold OW. 1995. Ecology of World Vegetation. London, UK: Chapman & Hall.
  • Axelrod DI. 1989. Age and origin of chaparral. In: KeeleySC, ed. The California Chaparral: Paradigms Re-Examined. Natural History Museum of Los Angeles County, no. 34, Science Series.
  • Barbosa PM, Grégoire JM, Stroppiana D, Pereira JMC. 1999. An assessment of fire in Africa (1981–91): burnt areas, burnt biomass and atmospheric emissions. Global Biogeochemical Cycles 13: 933950.
  • Barkworth ME, Capels KM. 2000. Grasses in North America: a geographic perspective. In: JacobsSWL, EverettJ, eds. Grasses: Systematics and Evolution. Melbourne, Australia: CSIRO. 331350.
  • Barnes DL. 1965. The effect of frequency of burning and mattocking on the control of coppice in the Marandellas sandveld. Rhodesian Journal of Agricultural Research 3: 5556.
  • Beard JS. 1944. Climax vegetation in tropical America. Ecology 25: 127158.
  • Beerling DJ, Woodward FI. 2001. Vegetation and the Terrestrial Carbon Cycle: Modelling the First 400 Million Years. Cambridge, UK: Cambridge University Press.
  • Bell RHV. 1982. The effect of soil nutrient availability on community structure in African ecosystems. In: HuntleyBJ, WalkerBH, eds. Ecology of Tropical Savannas. Berlin, Germany: Springer, 193216.
  • Bond WJ, Midgley GF, Woodward FI. 2003a. What controls South African vegetation – climate or fire? South African Journal of Botany 69: 7991.
  • Bond WJ, Midgley GF, Woodward FI. 2003b. The importance of low atmospheric CO2 and fire in promoting the spread of grasslands and savannas. Global Change Biology 9: 973982.
  • Bond WJ, Van Wilgen BW. 1996. Fire and plants. Population and Community Biology Series 14. London, UK: Chapman & Hall.
  • Booysen P de V, Tainton NM, eds. 1984. Ecological effects of fire in South African ecosystems. Ecological Studies 48. Berlin, Germany: Springer Verlag.
  • Bowman DMJS. 2000. Australian Rainforests: Islands of Green in a Land of Fire. Cambridge, UK: Cambridge University Press.
  • Bowman DMJS, Fensham RJ. 1991. Response of a monsoon forest–savanna boundary to fire protection, Weipa, northern Australia. Australian Journal of Ecology 16: 111118.
  • Bowman DMJS, Panton WJ. 1993. Factors that control monsoon rainforest seedling establishment and growth in north Australian Eucalyptus savanna. Journal of Ecology 81: 297304.
  • Bowman DMJS, Panton WJ. 1995. Munmarlary revisited: response of a north Australian Eucalyptus tetrodonta savanna protected from fire for 30 years. Australian Journal of Ecology 20: 526531.
  • Bradstock RA, Williams JE, Gill AM, eds. 2002. Flammable Australia: the Fire Regimes and Biodiversity of a Continent. Cambridge, UK: Cambridge University Press.
  • Cerling TE, Harris JM, MacFadden BJ, Leakey MG, Quade J, Eisenmann V, Ehleringer JR. 1997. Global vegetation change through the Miocene/Pliocene boundary. Nature 389: 153158.
  • Chapin FS, Matson PA, Mooney HA. 2002. Principles of Terrestrial Ecosystem Ecology. New York, USA: Springer.
  • Cole MM. 1986. The Savannas. Biogeography and Geobotany. London, UK: Academic Press.
  • Coutinho LM. 1982. Ecological effects of fire in Brazilian Cerrado. In: HuntleyBJ, WalkerBH, eds. Ecology of Tropical Savannas. Ecological Studies, 42. Berlin, Germany: Springer Verlag, 273291.
  • Coutinho LM. 1990. Fire in the ecology of the Brazilian Cerrado. In: GoldammerJG, ed. Fire in the Tropical Biota. Ecological Studies, 84. Berlin, Germany: Springer Verlag, 82105.
  • Cowling RM, Richardson D, eds. 1997. Vegetation of Southern Africa. Cambridge, UK: Cambridge University Press.
  • Cramer W, Bondeau A, Woodward FI et al. 2001. Global response of terrestrial ecosystem structure and function to CO2 and climate change: results from six dynamic global vegetation models. Global Change Biology 7: 357373.
  • Cumming DH, Fenton MB, Rautenbach IL, Taylor RD, Cumming GS, Cumming MS, Dunlop KM, Ford AG, Hovorka MD, Johnston DS, Kalcounis M, Mahlangu Z, Portfors CVR. 1997. Elephants, woodlands and biodiversity in southern Africa. South African Journal of Science 93: 231236.
  • D’Antonio CM, Vitousek PM. 1992. Biological invasions by exotic grasses, the grass/fire cycle, and global change. Annual Review of Ecology and Systematics 23: 6387.
  • De Fries RS, Townshend JRG. 1994. NDVI-derived land cover classification at global scales. Int. Journal of Remote Sensing 15: 35673586.
  • Dwyer E, Gregoire JM, Malingreau JP. 1998. A global analysis of vegetation fires using satellite images: Spatial and temporal dynamics. Ambio 27: 175181.
  • Dwyer E, Pereira JMC, Gregoire JM, DaCamara CC. 2000. Characterization of the spatio-temporal patterns of global fire activity using satellite imagery for the period April 1992 to March 1993. Journal of Biogeography 27: 5769.
  • Ehleringer JR, Cerling TE, Helliker BR. 1997. C4 photosynthesis, atmospheric CO2, and climate. Oecologia 112: 285299.
  • Eiten G. 1975. The vegetation of the Serra do Roncador. Biotropica 7: 112135.
  • FAO. 1998. World reference base for soil resources. FAO, Rome. http://www.fao.org/ag/agl/agll/prtsoil.stm.
  • FAO. 2001. Global forest resources assessment 2000: Main report. FAO Forestry Paper 140. Rome, Italy: FAO. http://www.fao.org/forestry/site/8952/en.
  • Frost PGH, Robertson F. 1987. The ecological effects of fire in savannas. In: WalkerBH, ed. Determinants of Tropical Savannas. Miami, FL, USA: ICSU Press, 93140.
  • Furley PA, Proctor J, Ratter JA, eds. 1992. Nature and Dynamics of Forest-Savanna Boundaries. London, UK: Chapman & Hall.
  • Gill AM. 1975. Fire and the Australian flora: a review. Australian Forestry 38: 425.
  • Goldammer JG. 1993. Wildfire management in forests and other vegetation: a global perspective. Disaster Management 5: 310.
  • Hairston NG, Smith FE, Slobodkin LB. 1960. Community structure, population control, and competition. American Naturalist 94: 421425.
  • Hansen M, De Fries RS, Townshend JRG, Sohlberg R. 2000. Global land cover classification at 1 km spatial resolution using a classification tree approach. International Journal of Remote Sensing 21: 13311364.
  • Harrison SP, Prentice CI. 2003. Climate and CO2 controls on global vegetation distribution at the last glacial maximum: analysis based on palaeovegetation data, biome modelling and palaeoclimate simulations. Global Change Biology 9: 9831004.
  • Hartley W. 1958. Studies on the origin, evolution and distribution of the Gramineae. I. The tribe Andropogoneae. Australian Journal of Botany 6: 116128.
  • Hassell CW, Dodson JR. 2003. The fire history of south-west Western Australia prior to European settlement in 1826–1829. In: AbbottI, BurrowsN, eds. Fire in Ecosystems of South-West Western Australia: Impacts and Management. Leiden, the. Netherlands: Backhuys, 7196.
  • Haxeltine A, Prentice IC. 1996. BIOME3: an equilibrium terrestrial biosphere model based on ecophysiological constraints, resource availability, and competition among plant functional types. Global Biogeochemical Cycles 10: 693709.
  • Herrera CM. 1992. Historical effects and sorting processes as explanations for contemporary ecological patterns: character syndromes in Mediterranean woody plants. American Naturalist 140: 421446.
  • Hoffmann WA. 1999. Fire and population dynamics of woody plants in a Neotropical savanna: matrix model predictions. Ecology 80: 13541369.
  • Hoffmann WA, Franco AC. 2003. Comparative growth analysis of tropical forest and savannas woody plants using phylogenetically independent contrasts. Journal of Ecology 91: 475484.
  • Holdridge LR. 1947. Determination of world plant formations from simple climatic data. Science 105: 367368.
  • Johnson EA. 1992. Fire and Vegetation Dynamics: Studies from the North American Boreal Forest. Cambridge, UK: Cambridge University Press.
  • Keeley JE, Rundel PW. 2003. Evolution of CAM and C4 carbon-concentrating mechanisms. International Journal of Plant Sciences 164: S55S77.
  • Keeley JE, Zedler PH. 1998. Evolution of life histories in Pinus. In: RichardsonDM, ed. Ecology and Biogeography of Pinus. Cambridge, UK: Cambridge University Press, 219249.
  • Kellman M. 1984. Synergistic relationships between fire and low soil fertility in neo-tropical savannas: a hypothesis. Biotropica 14: 158160.
  • Kennan TCD. 1972. The effects of fire on two vegetation types at Matopos, Rhodesia. Proceedings of the Annual Tall Timbers Fire Ecology Conference 11: 5398.
  • Kershaw AP, Clark JS, Gill AM, D’Costa DM. 2002. A history of fire in Australia. In: BradstockRA, WilliamsJE, GillAM, eds. Flammable Australia: the Fire Regimes and Biodiversity of a Continent. Cambridge, UK: Cambridge University Press, 325.
  • Koechlin J. 1972. Flora and vegetation of Madagascar. In: BattistiniR, Richard-VindardG, eds. Biogeography and Ecology in Madagascar. The Hague, the Netherlands: Junk, 145190.
  • Kurz WA, Apps MJ. 1999. A 70-year retrospective analysis of C fluxes in the Canadian forest sector. Ecological Applications 9: 526547.
  • Ladiges PY, Udovicic F, Nelson G. 2003. Australian biogeographical connections and the phylogeny of large genera in the plant family Myrtaceae. Journal of Biogeography 30: 989998.
  • Le Maitre DC, Van Wilgen BW, Chapman RA, McKelly DH. 1996. Invasive plants and water resources in the Western Cape Province, South Africa: modelling the consequences of a lack of management. Journal of Applied Ecology 33: 161172.
  • Linder HP. 2003. The radiation of the Cape flora, southern Africa. Biological Reviews 78: 597638.
  • Los SO, Justice CO, Tucker CJ. 1994. A global 1 degree by 1 degree NDVI data set for climate studies derived from the GIMMS continental NDVI data. International Journal of Remote Sensing 15: 34933518.
  • Manders PT. 1990. Fire and other variables as determinants of forest/fynbos boundaries in the Cape Province. Journal of Vegetation Science 1: 483490.
  • Matthews E. 1983. Global vegetation and land use: new high resolution data bases for climate studies. Journal of Climate and Applied Meteorology 22: 474487.
  • Meeson BW, Corprew FE, McManus JMP et al. 1995. ISLSCP Initiative 1. Global Data Sets for Land Atmospheric Models, 1987–88, Vols 1–5, published on CD by NASA, Goddard Space Center, USA.
  • Midgley JJ, Cowling RM, Seydack AHW, Van Wyk GF. 1997. Forest. In: CowlingRM, RichardsonD, eds. Vegetation of Southern Africa. Cambridge, UK: Cambridge University Press, 278299.
  • Mooney HA, ed. 1977. Convergent Evolution in Chile and California Mediterranean Climate Ecosystems. Stroudsberg. Pennsylvania, PA, USA: Dowden, Hutchinson & Ross.
  • Moreira AG. 2000. Effects of fire protection on savanna structure in Central Brazil. Journal of Biogeography 27: 10211029.
  • O'Connor TG, Bredenkamp GJ. 1997. Grassland. In: CowlingRM, RichardsonD, eds. Vegetation of Southern Africa. Cambridge, UK: Cambridge University Press, 215257.
  • Ogden J, Basher L, McGlone MS. 1998. Fire, forest regeneration and links with early human habitation: evidence from New Zealand. Annals of Botany 81: 687696.
  • Oliveira-Filho AT, Ratter JA. 2002. Vegetation physiognomies and woody flora of the cerrado biome. In: OliveiraPS, MarquisRJ, eds. The Cerrados of Brazil. New York, USA: Columbia University Press, 91120.
  • Olson JS, Watts J, Allison L. 1983. Carbon in live vegetation of major world ecosystems. TN USA, Oak Ridge, USA: US Department of Energy, Oak Ridge National Laboratory.
  • Parton WJ, Scurlock JMO, Ojima DS et al. 1993. Observations and modelling of biomass and soil organic matter dynamics for the grassland biome worldwide. Global Biogeochemical Cycles 7: 785809.
  • Phillips JFV. 1930. Fire: its influence on biotic communities and physical factors in South and East Africa. South African Journal of Science 27: 352367.
  • Pickett STA, White PS, eds. 1985. The Ecology of Natural Disturbance. New York, USA: Academic Press.
  • Polis GA. 1999. Why are parts of the world green? Multiple factors control productivity and the distribution of biomass. Oikos 86: 315.
  • Prance GT. 1992. The phytogeography of savanna species of neotropical Chrysobalanaceae. In: FurleyPA, ProctorJ, RatterJA, eds. Nature and Dynamics of Forest-Savanna Boundaries. London, UK: Chapman & Hall, 295330.
  • Ratter JA, Ribeiro JF, Bridgewater S. 1997. The Brazilian cerrado vegetation and threats to its biodiversity. Annals of Botany 80: 223230.
  • Richardson DM. 1998. Forestry trees as invasive aliens. Conservation Biology 12: 1826.
  • Sage RF. 2001. Environmental and evolutionary preconditions for the origin and diversification of the C4 photosynthetic syndrome. Plant Biology 3: 202213.
  • San José JJ, Farinas MR. 1983. Changes in tree density and species composition in a protected Trachypogon savanna, Venezuela. Ecology 64: 447453.
  • San José JJ, Montes RA, Farinas MR. 1998. Carbon stocks and fluxes in a temporal scaling from a savanna to a semi-deciduous forest. Forest Ecology and Management 105: 251262.
  • Sarmiento G. 1983. The savannas of tropical America. In: BourliereF, ed. Ecosystems of the World 13: Tropical Savannas. Amsterdam, The Netherlands: Elsevier, 245288.
  • Sarmiento G. 1992. A conceptual model relating environmental factors and vegetation formations in the lowlands of tropical South America. In: FurleyPA, ProctorJ, RatterJA, eds. Nature and Dynamics of Forest-Savanna Boundaries. London, UK: Chapman & Hall, 583601.
  • Schimper AFW. 1903. Plant Geographyraphy on a Physiological Basis. Oxford, UK: Clarendon Press.
  • Scholes RJ, Archer S. 1997. Tree–grass interactions in savannas. Annual Review of Ecology and Systematics 28: 517544.
  • Scott AC. 2000. The pre-Quaternary history of fire. Palaeogeography, Palaeoclimatology, Palaeoecology 164: 297345.
  • Scott L. 2002. Microscopic charcoal in sediments: Quaternary fire history of the grassland and savanna regions in South Africa. Journal of Quaternary Science 17: 7786.
  • Sellers PJ, Los SO, Tucker CJ et al. 1994. A global 1*1 degree NDVI data set for climate studies. Part 2: the generation of global fields of terrestrial biophysical parameters from satellite data. Journal of Climate 9: 706737.
  • Shackleton CM, Scholes RJ. 2000. Impact of fire frequency on woody community structure and soil nutrients in the Kruger National Park. Koedoe 43: 7581.
  • Specht RL. 1969. A comparison of sclerophyllous vegetation characteristics of mediterranean type climates in France, California and southern Australia. I. Structure, morphology and succession. Australian Journal of Botany 17: 277292.
  • Specht RL, Moll EJ. 1983. Mediterranean-type heathlands and sclerophyllous shrublands of the world: an overview. In: KrugerFJ, MitchellDT, JarvisJUM, eds. Mediterranean Type Ecosystems: the Role of Nutrients. Ecological Studies 43. Berlin, Germany: Springer Verlag, 4165.
  • Stephenson NL. 1990. Climatic control of vegetation distribution: the role of water balance. American Naturalist 135: 649670.
  • Stott PA. 1988. The forest as Phoenix: towards a biogeography of fire in mainland South East Asia. Geographyraphical Journal 154: 337350.
  • Stott PA, Goldammer JG, Werner WL. 1990. The role of fire in the tropical lowland deciduous forests of Asia. In: Goldammer, JG, ed. Fire in the Tropical Biota. Ecological Studies 84. Berlin, Germany: Springer Verlag, 3244.
  • Swaine MD, Hawthorne WD, Orgle TK. 1992. The effects of fire exclusion on savanna vegetation at Kpong, Ghana. Biotropica 24: 166172.
  • Tansey K, Gregoire J-M, Stroppiana D, Sousa A, Silva J et al. 2004. Vegetation burning in the year 2000: global burned area estimates from SPOT VEGETATION data. Journal of Geophysical Research 109: D14S03.
  • Thonicke K, Venevsky S, Sitch S, Cramer W. 2001. The role of fire disturbance for global vegetation dynamics: coupling fire into a dynamic global vegetation model. Global Ecology and Biogeography 10: 661678.
    Direct Link:
  • Tilman D, Reich P, Phillips H, Menton M, Patel A, Vos E, Peterson D, Knops J. 2000. Fire suppression and ecosystem carbon storage. Ecology 81: 26802685.
  • Trapnell CG. 1959. Ecological results of woodland burning experiments in northern Rhodesia. Journal of Ecology 47: 129168.
  • Tsvuura V. 1998. Effects of long-term burning on some vegetation and soil characteristics on a dry miombo region at Marondera, Zimbabwe. MSc Thesis, University of Zimbabwe, Zimbabwe.
  • Venevsky S, Thonicke K, Sitch S, Cramer W. 2002. Simulating fire regimes in human dominated ecosystems: Iberian peninsula case study. Global Change Biology 8: 984998.
  • Whelan RJ. 1995. The Ecology of Fire. Cambridge, UK: Cambridge University Press.
  • White F. 1983. The Vegetation of Africa. Paris, France: UNESCO.
  • Whittaker RH. 1975. Communities and Ecosystems. London, UK: Collier MacMillan.
  • Woodward FI. 1987. Climate and Plant Distribution. Cambridge, UK: Cambridge University Press.
  • Woodward FI, Lomas MR, Lee SE. 2001. Predicting the future productivity and distribution of global vegetation. In: JacquesR, SaugierB, MooneyH, eds. Terrestrial Global Productivity. New York, USA: Academic Press, 521541.
  • Woodward FI, Smith TM, Emanuel WR. 1995. A global land primary productivity and phytogeography model. Global Biogeochemical Cycles 9: 471490.

Appendix

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusion
  8. Acknowledgements
  9. References
  10. Appendix

Appendix. Description of the Sheffield Dynamic Global Vegetation Model (SDGVM)

The SDGVM is a generalised global-scale model that predicts vegetation structure and dynamics from input data of climate, [CO2] and soil texture. The basic processes and assumptions in the SDGVM are outlined in Cramer et al. (2001). The SDGVM requires input data of climate, [CO2] and soil texture. The climate data are monthly mean and minimum temperatures, water vapour pressure deficit, precipitation and cloudiness. The physiology and biophysical module simulates carbon and water fluxes from vegetation (Woodward et al., 1995) with water and nutrient supply defined by the water and nutrient flux module. The soil module incorporates the Century soil model of carbon and nitrogen dynamics (Parton et al., 1993), with a model of plant water uptake. Eight soil carbon and nitrogen pools are modeled: surface and soil structural material, active soil organic matter, surface microbes, surface and soil metabolic material, slow and passive soil organic material.

The carbon and nitrogen dynamics are described by linear autonomous differential equations with parameters that depend on soil texture, temperature, precipitation, humidity, soil moisture, water flow, potential evapotranspiration and litter. These variables are held constant over a given period and the differential equations are solved by standard means for these conditions. The set of parameters is updated at each successive period, and the carbon calculation is advanced using the final state in the previous period as the initial state in the current period. These equations are solved each month. The organic nitrogen flows are equal to the product of the carbon flow and the nitrogen to carbon ratio of the state variable that receives the carbon (Parton et al., 1993). The carbon to nitrogen ratios of the soil state variables receiving the flow of carbon, are linear functions of the mineral nitrogen pool. The mineral nitrogen pool is an additional pool, which stores surplus nitrogen. The dynamics imposed by the linear functions ensure that this pool is always positive.

Water fluxes are modelled using a ‘bucket’ model. The model is composed of four buckets: one thin (5 cm) layer at the surface and three buckets of equal depth, which make up the remainder of the soil layer. The depth of the total soil layer is set to a default of 1 m. The effects of bare soil evaporation, sublimation, transpiration and interception (each of which represents a loss of water available to the vegetation system) are incorporated into the model.

The primary productivity model simulates canopy CO2 and water vapour exchange and nitrogen uptake and partitioning within the canopy. Nitrogen uptake is linked directly with the Century soil model, which simulates the turnover of carbon and nitrogen in plant litter, of differing ages and depths within the soil, in addition to soil water status.

The primary productivity model determines the assimilated carbon available for the growth of plant leaves, stems and roots. The plant structure and phenology module defines the vegetation leaf area index and the vegetation phenology. Leaf phenology is defined by temperature thresholds for cold deciduous vegetation and by drought duration for drought deciduous vegetation (Cramer et al., 2001).

The vegetation dynamics module (Cramer et al., 2001; Woodward et al., 2001) simulates the establishment, growth, competition and mortality of plant functional types (evergreen and deciduous broad leaved and needle leaved trees, grasses with the C4 photosynthetic metabolisms and C3 grasses and shrubs). Functional types of plants compete for light and soil water and all suffer random mortality that increases with age. The densities (plants per unit area), heights and ages of all of the functional types, except grasses, are simulated at the finest spatial resolution (pixel) of the model. A fire module, based on temperature and precipitation, burns a fraction of the smallest pixel of study (Woodward et al., 2001). The fire model simulates disturbance by fire for a small fraction of the pixel. It is assumed that 80% of above-ground carbon and nitrogen are lost as a consequence of the fire and fire only occurs when, in effect, leaf litter reaches a critical point of dryness, at which point fire will occur at a random time and for a random subset of the pixel (Woodward et al., 2001).

The SDGVM simulations start from a soil, defined by texture and depth, climate and atmospheric CO2 concentration. Therefore there is a necessary initialisation stage in which the soil carbon and nitrogen storage of the soil is determined, with the appropriate vegetation for the simulated climate. The model initialisation is determined by running with a repeated and random selection of annual climates from 1901 to 1920. The soil carbon and nitrogen values are first determined by solving Century analytically. Then the model is run until the vegetation structure is at equilibrium, typically after, at most 500 yr. When initialisation is completed the SDGVM then simulates vegetation for the whole climate series. Fire was simulated from the same initial values as the fire-off simulation for the 20th century.