What controls the distribution of tropical forest and savanna?

Authors

  • Brett P. Murphy,

    Corresponding author
    1. Geographic Information Science Center of Excellence, South Dakota State University, Brookings SD 57007, USA
    2. School of Plant Science, University of Tasmania, Hobart, Tasmania 7001, Australia
      E-mail: brettpatrickmurphy@hotmail.com
    Search for more papers by this author
  • David M.J.S. Bowman

    1. School of Plant Science, University of Tasmania, Hobart, Tasmania 7001, Australia
    Search for more papers by this author

E-mail: brettpatrickmurphy@hotmail.com

Abstract

Ecology Letters (2012) 15: 748–758

Abstract

Forest and savanna biomes dominate the tropics, yet factors controlling their distribution remain poorly understood. Climate is clearly important, but extensive savannas in some high rainfall areas suggest a decoupling of climate and vegetation. In some situations edaphic factors are important, with forest often associated with high nutrient availability. Fire also plays a key role in limiting forest, with fire exclusion often causing a switch from savanna to forest. These observations can be captured by a broad conceptual model with two components: (1) forest and savanna are alternative stable states, maintained by tree cover-fire feedbacks, (2) the interaction between tree growth rates and fire frequency limits forest development; any factor that increases growth (e.g. elevated availability of water, nutrients, CO2), or decreases fire frequency, will favour canopy closure. This model is consistent with the range of environmental variables correlated with forest distribution, and with the current trend of forest expansion, likely driven by increasing CO2 concentrations. Resolving the drivers of forest and savanna distribution has moved beyond simple correlative studies that are unlikely to establish ultimate causation. Experiments using Dynamic Global Vegetation Models, parameterised with measurements from each continent, provide an important tool for understanding the controls of these systems.

Introduction

Forests and savannas dominate the tropical landscapes of Africa, Australia, South America and Asia, covering about 15 and 20% of the Earth’s land surface, respectively (Grace et al. 2006; Fig. 1). These biomes have strikingly different ecological structure and function. Tropical forests are characterised by dense tree cover, typically with a high diversity of trees, lianas, and epiphytes, and competition for light primarily drives dynamics and structural complexity of the vegetation. Savannas consist of an open tree layer with a continuous grassy ground layer, typically dominated by shade-intolerant species possessing the C4 photosynthetic pathway (Ratnam et al. 2011). The grass biomass of savannas often supports high densities of large grazers and provides fuel for frequent fires. Tropical forests and savannas are globally important centres of biodiversity, reflecting complex evolutionary histories involving co-evolution, niche differentiation, as well as vicariant speciation and convergent evolution amongst continents. The interface between these biomes has been postulated as an important theatre for evolutionary diversification, including our own ancestors (Bobe & Behrensmeyer 2004).

Figure 1.

 The distribution of (a) closed canopy forest and (b) mean annual rainfall, in the tropics. The vertical extent of the map corresponds to the latitudinal tropics. The vegetation and rainfall maps are derived from the GlobCover (version 2.2; Arino et al. 2008) and WorldClim (Hijmans et al. 2005) datasets.

Determining the environmental controls of tropical forest and savanna biomes is a prerequisite for a comprehensive understanding of the global carbon cycle. Tropical forests are one of the most carbon dense biomes on the planet, storing on average 320 t C ha−1, representing about 25% of the carbon stored in the world’s vegetation and soil organic matter (Grace et al. 2006). In contrast, savannas store much less carbon, averaging 120 t C ha-1, representing about 15% of the global total (Grace et al. 2006). Despite the large difference in carbon storage between the two biomes, they are similar in terms of total net primary productivity (NPP), each responsible for around 30% of the global terrestrial total (Grace et al. 2006). The lower carbon storage capacity of savannas, despite high NPP, is largely due to frequent fires rapidly returning captured carbon to the atmosphere. Tropical savannas are the most frequently and extensively burnt ecosystem on Earth, accounting for around 44% of global carbon emissions by fires (van der Werf et al. 2010). Although tropical forests are the least flammable of any biome, fires associated with tropical deforestation have contributed about 20% of anthropogenic radiative forcing since the industrial revolution (Bowman et al. 2009).

The purpose of this synthesis is to propose a novel conceptual model to explain the relative distribution of forest and savanna throughout the tropics. The ecological processes controlling these distributions have received much less attention than the processes controlling tree biomass within either biome (e.g. Cochrane & Schulze 1999; Sankaran et al. 2005). We borrow heavily from the savanna-based literature focusing on drivers of tree abundance within savannas, our understanding of which has expanded rapidly in recent years (e.g. Sankaran et al. 2004, 2005; Bond 2008). The question of what processes allow trees and grasses to co-exist (in a savanna), is fundamentally similar to the question of what processes allow trees to form a closed canopy and exclude grasses (in a forest); metaphorically, these questions represent opposites sides of the same coin. We also draw on alternative stable state theory that is increasing being applied to forest and savanna distributions, regionally (Warman & Moles 2009; Staver et al. 2011a) and globally (Hirota et al. 2011; Staver et al. 2011b). First, we present a simple intercontinental comparison of the climatic correlates of tropical forest distribution, highlighting that the transition from savanna to forest cannot be explained in terms of climate alone. We then evaluate other competing theories that emphasise the primacy of a single factor in controlling forest and savanna distribution. Broadly, these factors divide into resource-based (or ‘bottom-up’ ) and disturbance-based (‘top-down’ ) controls, including the role of soils (Askew et al. 1970), topography and drainage (Beard 1953), and fire (Bowman 2000; Bond et al. 2005). We show that such single factors cannot explain the distribution of forest and savanna biomes at local and continental scales. Rather, the controls are emergent from a web of feedbacks amongst biological and environmental variables. Building on our analysis and literature review, we propose a new conceptual model to explain the relative distribution of tropical forest and savanna, based on alternative stable state theory and tree growth-fire interactions. Although humans have had a massive impact on tropical forest distribution – by directly removing or degrading forest canopies (e.g. deforestation, logging), or modifying other controls such as fire activity and resource availability (e.g. nutrient deposition, CO2) – here we specifically focus on relatively natural, intact systems and deliberately avoid a detailed review of the impacts of humans on forest and savanna distribution.

Climate as the Primary Determinant of Tropical Forest and Savanna Distribution

Vegetation ecologists have long noted that, at large spatial scales, climate controls the distribution of tropical forest and savanna. Indeed, Schimper (1898), who coined the term ‘rainforest’ (from the German ‘Regenwald’), stressed the overriding control of forest and savanna by annual rainfall amount and its seasonal distribution, with forests typically associated with year-round high rainfall (Fig. 1). Generally, forest replaces savanna when annual rainfall exceeds about 1,500 mm (Cole 1986; Lewis 2006) and the length and intensity of seasonal drought is minimal (Nix 1983; Lewis 2006; Lehmann et al. 2011). Lehmann et al. (2011) pointed out that the effect of high rainfall seasonality is twofold, both reducing tree growth and increasing the probability of fire (that potentially excludes or destroys forest). Periodic drought, at the decadal scale, has been suggested to be a strong driver of forest distribution in certain regions, most notably in Amazonia (Hutyra et al. 2005). For instance, recent droughts in the tropics have caused high rates of forest tree mortality and biomass loss (Phillips et al. 2009), and increased susceptibility of forests to fire (Cochrane 2003).

The recent availability of satellite-derived tree cover datasets (e.g. Hansen et al. 2003), coupled with global climate datasets (e.g. Hijmans et al. 2005), provides globally consistent data to investigate climatic control of the relative distribution of forest and savanna in more detail than has been previously possible (e.g. Bucini & Hanan 2007; Hirota et al. 2011; Staver et al. 2011a,b). Hirota et al. (2011) and Staver et al. (2011b) recently demonstrated that annual precipitation is the primary control of the relative distribution of tropical forest and savanna globally. It is important to note that these analyses hinge on the interpretation of distinct peaks in the frequency of tree cover: around 0%, representing grassland; 20%, representing savanna; and 80%, representing forest. There was a notable absence of sites with intermediate levels of cover, around 5 and 60%, suggesting a clear natural disjunction between grassland, savanna and forest. Such a global approach necessarily ignores regional floristic and ecological variation, although it unlikely biases the results in any particular direction. For example, some dry tropical forests are more closely allied with savannas, given that they have relatively open canopies and support sufficient grass biomass to fuel frequent fires (Ratnam et al. 2011) and may be falsely classified as true closed canopy forests on the basis of tree cover alone. Conversely, some anthropogenically degraded forests may be incorrectly classified as savannas (Ratnam et al. 2011). Furthermore, substantial error may occur if heavily degraded tropical landscapes are not masked out of continental- and global-scale analyses (e.g. Hirota et al. 2011).

The analysis of Hirota et al. (2011) suggested that forests are more restricted in Australia than in parts of Africa and South America with similar annual rainfall. We further investigated this issue by statistically modelling the same MODIS tree cover dataset (Hansen et al. 2003) as used by those authors, albeit masking out deforested areas, to determine: (a) the climatic variables that are most closely related to forest occurrence (> 60% tree cover) and (b) if there are substantial differences in vegetation-climate relationships between Africa, South America and Australia (including New Guinea) (for methods, see Appendix S1 in Supporting Information). We found that the best predictor of forest cover on these three continents was an approximation of soil water content (%) through the six driest consecutive months (Fig. 2), based on a monthly soil water budget with inputs of monthly rainfall and potential evapotranspiration, and an assumed soil depth of 2 m (Trabucco & Zomer 2010; Appendix S1). The clear superiority of this variable over annual rainfall in our modelling highlights the importance of seasonal drought in limiting forest, consistent with earlier observations (e.g. Nix 1983; Lewis 2006; Lehmann et al. 2011).

Figure 2.

 The observed (bars) and predicted (lines) probability of forest occurrence, in relation to modelled dry season soil saturation (%), across the tropics of three continents: (a) South America, (b) Africa and (c) Australia (including New Guinea). A comparison of the relationships for the three continents is shown in (d). Forest is defined as tree cover > 60%, based on the 500 m resolution MODIS tree cover dataset (Hansen et al. 2003; for methods, see Appendix S1). The observed probability of forest occurrence represents the proportion of observations that were forest rather than savanna, in each dry season soil saturation class (100 × 1% intervals). For each continent, the range of dry season soil saturation values plotted is restricted to the 0.5 to 99.5% quantiles. The distribution of model residuals are presented in Fig. S3.

Though it is clear that forest distribution is closely tied to water availability, our analysis shows substantial intercontinental differences in forest-climate relationships. A striking feature of South America is the abundance of forest in climate zones that would rarely support forest in Australia or Africa (Fig. 2). In South America, forests cover the majority of the landscape at a dry season soil saturation of 45%, though similar climatic zones in Australia and Africa are strongly savanna-dominated, containing just 8 and 9% forest, respectively. If we applied the Australian pattern (i.e. the relationship shown in Fig. 2c) to South America, we would see a massive reduction in forest cover (Fig. 3).

Figure 3.

 Predicted distribution of closed canopy forest, based on dry season soil saturation (%), using regression models derived with data from: (a) South America, (b) Africa and (c) Australia (including New Guinea) (relationships shown in Fig. 2). The ‘closed canopy dominant’ class refers to ≥ 67% probability of forest; the ‘open canopy dominant’ class refers to ≤ 33% probability of forest; the ‘mixed’ class refers to 67–33% probability of forest. The cross-hatching indicates the observed distribution of forest (≥ 67% coverage of a 25 km search radius) in South America, Africa and Australia, based on the MODIS tree cover dataset. The vertical extent of the map corresponds to the latitudinal tropics.

Although Hirota et al. (2011) found that forests are more restricted in Australia than in parts of Africa and South America with similar annual rainfall, they suggested that there was little difference between Africa and South America. It is difficult to pinpoint the cause of the discrepancy between our results and those of Hirota et al. (2011), however, possible explanations include the inadequacy of annual rainfall as a predictor of forest occurrence, primarily because it ignores rainfall seasonality that can differ markedly between the continents, relative to annual precipitation amount (Lehmann et al. 2011; Staver et al. 2011b), and the fact that Hirota et al. (2011) did not exclude disturbed areas from their analysis, despite the presence of large deforested areas in the tropics, especially in South America.

Another critical feature of forest distribution in South America is that the dry season soil saturation index had lower predictive power than in Australia and Africa (R2 = 30% cf. 75 and 79%, respectively), suggestive of weaker climatic control of forest cover on that continent. A similar pattern is borne out in the pattern of tree cover in savannas in South America (assumed tree cover ≤ 60%; for methods, see Appendix S1). In Africa and Australia, there is a strong relationship between savanna tree cover and dry season soil water availability (R2 = 74 and 61%, respectively; Fig. 4), similar to the patterns reported extensively elsewhere for individual continents (e.g. Sankaran et al. 2005; Bucini & Hanan 2007). However, in South America this relationship is very weak (R2 = 18%), with very little apparent relationship between dry season soil saturation and savanna tree cover; at all but the driest sites, any level of tree cover appears possible. Thus, in terms of patterns of tree cover, South America appears to differ markedly from the other two continents. These patterns may reflect biogeographic differences amongst the continents, which we shall return to after we examine other resource- and disturbance-based controls of tropical forest and savanna distribution.

Figure 4.

 The relationship between savanna tree cover and dry season soil saturation (%), across the tropics of three continents: (a) South America, (b) Africa and (c) Australia (including New Guinea). Savannas are defined as tree cover ≤ 60%. The solid regression lines represent the modelled mean relationship for savannas only, and the dashed lines represent the modelled 99% quantile. The data points represent a randomly selected subsample of 5000 from each continent. Tree cover is derived from the 500 m resolution MODIS tree cover dataset (Hansen et al. 2003; for methods, see Appendix S1).

Other Resource-Based Controls

Earlier work generally assumed climate was the sole driver of biome distributions (Schimper 1898; Holdridge 1947). Though climate is undoubtedly an important driver of tropical forest distribution (Figs 1–2), the relationship often breaks down at local and regional scales. Within a single climate zone, complex mosaics of forest and savanna may exist, suggesting the ‘decoupling’ of climate and vegetation (Beard 1953; Bowman 2000), making it clear that controls other than climate are also important. Bond (2008) used the terms ‘bottom-up’ and ‘top-down’ to distinguish between resource-based (e.g. water and nutrient availability) and disturbance-based (e.g. fire and herbivory) controls on tree cover, respectively, and we briefly review the evidence for these, below.

Soils

Forest and savanna are often clearly differentiated by edaphic factors, such as soil depth (Furley 1999), texture (Askew et al. 1970), parent material (Ash 1988) and drainage (Beard 1953; Lloyd et al. 2008). Of edaphic factors, nutrient availability has probably received the most attention in the literature, largely driven by the widespread observation that forests are often associated with nutrient-rich soils (Ash 1988; Bowman 2000; Bond 2008) and savannas with deeply weathered, ancient soils (Kellman 1984). For example, Lloyd et al. (2008) pointed out that the Brazilian cerrado occupies an area where, on the basis of climate, forest would be expected to occur, if not for the low nutrient availability of the soil. Though the tendency for savannas to be associated with nutrient-poor soils is clear, Bond (2010) used a nutrient stock analysis to argue that it is not the size of the total nutrient pool that limits forest formation, with soils typically having adequate nutrient stocks to construct forest biomass. Rather, he suggested that it is the effect of nutrient limitation on tree growth rates, in combination with high fire frequencies, which prevents the formation of forest.

An explanation for at least some of the contrasting properties of forest and savanna soils is the ability of vegetation to directly influence soils. This vegetation–soil feedback relates to the fundamentally different nutrient cycles, ecophysiology, hydrology and soil biota between the two biomes. Forest soils typically have higher organic matter and nitrogen content, both largely originating from the vegetation itself, than would develop under savanna on equivalent parent materials, because of greater inputs of litter and higher density of deep-rooted trees which are able to access sub-soil nutrients (McCulley et al. 2004). In savannas, the dominance of shallow-rooted grasses and the higher frequency of fires and grazing results in more open nutrient cycles that are less efficient at accumulating nutrients and soil organic matter (Belsky 1994). Forest soil moisture may be increased through enhanced infiltration down root channels, and via hydraulic uplift in the dry season (Scheffer et al. 2005).

Disturbance-Based Controls

Fire

The role of fire in shaping forest distribution is most apparent where forest patches exist within a highly flammable savanna matrix. In such landscapes, forest is frequently restricted to topographic settings protected from fire and some ecologists have used this pattern to emphasise the primacy of fire in limiting forests (Russell-Smith et al. 1993; Bowman 2000). However, topography that confers fire protection is often highly confounded with ‘bottom-up’ factors, especially water availability, making it extremely difficult to attribute ultimate causation. For example, steep gullies may be fire protected, but also provide soils with greater water availability and less extreme microclimates (Bowman 2000).

Forest trees are generally considered more susceptible to both topkill (death of aboveground parts) and whole-tree mortality following fire than savanna trees, although many forest species can resprout following a single fire (Ash 1988; Hoffmann & Moreira 2002; Brando et al. 2011). Recently in Brazil, Hoffmann et al. (2009) found that forest and savanna trees adjacent to forest-savanna boundaries had similar rates of whole-tree mortality following fire, but forest trees were more likely to be topkilled than savanna trees of similar stem diameter. They attributed this effect to greater bark thickness for a given stem diameter in savanna trees, with the implication that forest trees would take much longer to reach fire-resistant sizes assuming similar growth rates.

Perhaps the most compelling evidence that fire controls the distribution of forest is the results of fire-exclusion experiments that have shown forest species invading fire-protected savanna, especially in high rainfall areas (San Jose & Farinas 1983; Moreira 2000; Woinarski et al. 2004), with a complete biome shift from grassy to closed canopy sometimes occurring within a few decades (Trapnell 1959; Swaine et al. 1992; Louppe et al. 1995). There are, however, exceptions to this generalisation, with some fire-exclusion experiments failing to produce shifts from savanna to forest after several decades, pointing to additional limiting factors (Bond et al. 2003; Higgins et al. 2007).

Fire has been used by humans since the Pleistocene and it was, and remains, an important land management tool for many hunter-gatherer and pastoral societies (Bowman et al. 2011). However, the extent to which prehistoric humans increased landscape fire activity and thereby reduced forest cover with the deliberate application of fire is hotly debated, and in only a few cases has loss of forest been clearly attributed to landscape burning by humans. In one of the clearest examples, McWethy et al. (2010) showed that when humans arrived on New Zealand’s South Island 800 years ago, there was an abrupt spike in fire activity, accompanied by a rapid loss of forest cover that failed to recover in low rainfall regions. However, in continental regions where fire is already frequent, such as savanna-dominated landscapes, the impact of humans on fire regimes and fire-sensitive vegetation has tended to be relatively minor and therefore difficult to detect in the palaeorecord (Bowman et al. 2011). In northeastern Australia, for example, despite the existence of a number of long, well-dated lake sediment cores, a sustained research effort over the last 30 years has not been able to establish conclusively that an increase in fire activity and a decrease in closed forest cover followed the arrival of humans around 45 000 years ago (Lynch et al. 2007). While some researchers have reported a clear human signal in proxies for fire activity since the Late Pleistocene (e.g. Haberle & Ledru 2001), an extensive recent study by Mooney et al. (2011) found no evidence of an increase in fire activity with the arrival of humans in Australia, and no relationship between fire and archaeological activity. Hence, the impact of landscape burning by prehistoric human societies on the distribution of tropical forest remains highly uncertain.

Herbivory

The role of large herbivores (> 50 kg) in limiting tropical forest remains poorly understood, although there is evidence that grazing and browsing exert a strong control on some aspects of vegetation structure. Grazing (consumption of grass) can favour woody vegetation by reducing grass biomass, thereby reducing both tree-grass competition and fire frequency and intensity, which have been shown to increase tree recruitment and growth (Werner 2005). Such a ‘grazing effect’ explains ‘bush encroachment’ following heavy overgrazing by domestic stock (van Langevelde et al. 2003). Conversely, browsing (consumption of woody plants) can limit woody plants, maintaining open, grassy systems. Indeed, browsing has been shown experimentally to reduce woody plant abundance in savannas (Augustine & McNaughton 2004), largely due to its strong negative effect on height growth, especially in conjunction with fire (Staver et al. 2009). Insights into the relative importance of grazing and browsing in savannas were provided by Asner et al. (2009), who used a high-resolution airborne remote sensing system (LiDAR) to compare vegetation structure inside and outside large-scale herbivore exclosures in Kruger National Park, in place for between 6 and 41 years. They found 66% less bare ground and 68% more woody cover in areas protected from both large grazing and browsing herbivores, suggesting any positive effects of grazing on woody cover (via suppression of herbaceous cover and fire activity) are outweighed by the negative effects of browsing.

Despite reports of herbivory strongly affecting woody biomass in savannas, and suggestions that introduced herbivores can destroy small forest patches in northern Australia (Russell-Smith & Bowman 1992), we suspect that large herbivores are not a critical factor in controlling forest distribution. This view is based on our observation that forest distribution and savanna tree cover are broadly similar between Africa and Australia (Figs 2 and 4). The large herbivore niche has been entirely empty in Australia for around 45 000 years (Bowman et al. 2010b), albeit now filled by exotic grazers such as horses, donkeys and swamp buffalo in the savannas. In contrast, a diverse herbivore assemblage remained abundant in African savannas until the 20th Century. Lehmann et al. (2011) suggested that large herbivores in arid parts of Africa promote savanna at the expense of treeless grassland, and although they didnot provide a mechanism, this would presumably be via the grazing effect described above. Importantly, Lehmann et al. (2011) did not posit an effect of herbivores on the mesic transition from savanna to forest. The role of large herbivores in controlling tropical forest-savanna boundaries remains to be clarified worldwide. Historical aerial photographs, for example, could be used to compare boundary dynamics inside African game parks containing intact herbivore assemblages with areas where herbivore populations have collapsed in recent decades due to overhunting.

Biogeographic effects

Biogeography, and the composition of the regional species pool in particular, could potentially be important and generally overlooked factors influencing the response of vegetation to disturbance. The biogeographic perspective may help to explain our finding that forests are more restricted in Australian than South America under comparable climates (Figs 2–3). Australia’s ubiquitous tree group, the eucalypt, is among the most fire-tolerant on Earth (e.g. Mills et al. 2006; Bond 2008) and is widely believed to allow fire to invade closed forests (Bowman 2000; Lehmann et al. 2011). Epicormic resprouting, that enables eucalypts to recover rapidly from fire, is thought to have appeared around 60 Ma (Crisp et al. 2011) and molecular phylogenetic evidence suggests that the savanna biome appeared at least 10 Ma (Crisp et al. 2010). In stark contrast to the extreme antiquity of fire in Australia, the tropical landscapes of South America have a much more recent history of fire. Recent phylogenetic work suggests that the cerrado tree flora diversified just 4 Ma in response to increased fire activity, and most of the cerrado tree flora is derived from forest species (Simon et al. 2009). Thus, it seems likely that South American savanna trees are less well adapted to co-existing with frequent fires than Australian savanna trees (primarily eucalypts). Clearly, dated phylogenies provide powerful frameworks for further testing hypotheses about historical relationships of fire adaptation of savanna and forest floras.

Co2 and Dynamic Forest–Savanna Boundaries

Analyses of sequences of aerial photographs, taken over the last 70 years, have revealed that forest boundaries are currently expanding globally, albeit at a variable rate, with numerous reports from Australia (Russell-Smith et al. 2004; Bowman et al. 2010a; Tng et al. 2012), Africa (Mitchard et al. 2009; Wigley et al. 2010), South America (Ratter et al. 1978; Durigan & Ratter 2006) and India (Mariotti & Peterschmitt 1994). This trend of forest expansion into savannas highlights the dynamic relationship between these biomes (Bond & Midgley 2012). Clearly, theories assuming strong resource-based control of forest distribution seem difficult to reconcile with high boundary dynamism, and those that assume strong control by disturbance seem more likely. However, some researchers have posited that elevated CO2 concentration may be the primary driver of the trend (Bond & Midgley 2000, 2012; Bowman et al. 2010a; Wigley et al. 2010). There are three mechanisms by which elevated atmospheric CO2 concentrations could promote invasion of savannas by trees (Bond & Midgley 2000). First, elevated CO2 would increase tree growth rates, allowing them to rapidly recover following disturbance such as fire. Second, though elevated CO2 increases carbon assimilation in plants possessing the C3 pathway (most woody plants), plants possessing the C4 pathway (most savanna grasses) may be relatively unresponsive. Third, elevated CO2 would increase whole plant water use efficiency, reducing transpiration by shallow-rooted species and increasing percolation of soil water to deeper soil layers, favouring establishment and persistence of deep-rooted woody plants. These effects would be expected to be diminished if soil factors strongly limit tree growth, or if nutrient cycling is affected by increasing tree biomass. For example, more of the available nitrogen pool could become sequestered in organic matter with increased tree biomass, unless there is a corresponding increase in nitrogen mineralisation (Johnson 2006).

The main point that we can draw from the widespread pattern of forest expansion in the tropics is that forest distribution is not strongly controlled by rigid edaphic constraints that impose a strong upper bound on tree biomass, such as total nutrient pools (sensuBond 2010). There is little reason to suspect a pantropical trend of increasing nutrient availability, other than isolated occurrences of nitrogen deposition (Wigley et al. 2010). If, as now widely assumed, forest expansion is driven by elevated CO2, the limitation imposed by total nutrient pools would apply equally under high CO2 conditions, and forest expansion into strongly nutrient limited areas could not occur. Indeed, Free Air CO2 Enrichment (FACE) studies have shown that elevated CO2 effects on plant biomass are strongly offset by nutrient limitation (e.g. Oren et al. 2001; Reich et al. 2006). This logic also applies equally to other site factors proposed to strongly limit trees, such as water-logging and frost; the global trend of forest expansion suggests that these factors must be of very localised importance.

Conceptual Model

Forest and savanna as alternative stable states

There is increasing evidence that tropical forest and savanna exist as fire-mediated ‘alternative stable states’ (ASS). The key premise of the ASS concept is that strongly contrasting ecosystem states can exist indefinitely under an identical set of exogenous environmental conditions, due to strong stabilising positive feedbacks between each state and processes that maintain that state (Scheffer et al. 2001). Though ASS theory has been invoked to explain states in a wide range of systems, from lake turbidity (Scheffer et al. 1993) to sea urchin ‘barrens’ within kelp forests (Konar & Estes 2003), it seems highly applicable to the forest-savanna dichotomy (Warman & Moles 2009). Hirota et al. (2011) and Staver et al. (2011b) recently used an analysis of satellite-derived tree cover throughout Africa, South America and Australia to show that the frequency distribution of tree cover is strongly tri-modal across a broad rainfall gradient, with peaks in the frequency of tree cover around 0%, representing grassland, 20%, representing savanna, and 80%, representing forest. Critically, there was a notable absence of sites with intermediate cover. This pattern suggests three separate ‘basins of attraction’ representing grassland, savanna and forest, with intermediate states tending to be unstable. It seems likely that this pattern could only be maintained by strong biological feedbacks, and Staver et al. (2011b) concluded that the main feedbacks responsible are the suppressive effect of tree cover on fire activity, and conversely, of fire activity on tree cover.

A number of key predictions of broader ASS theory appear to be consistent with observational and experimental studies of forest – savanna systems. First, we expect sharp spatial boundaries between alternative states (Schröder et al. 2005). Indeed, this is one of the most conspicuous and fascinating characteristics of forest–savanna boundaries. Boundaries often span just a few metres, accompanied by extremely abrupt changes in tree cover, light availability, temperature, grass abundance and fire activity (Bowman 2000; Hoffmann et al. 2009). Indeed, it is the absence of abrupt environmental changes, despite such striking changes in the vegetation, which has stimulated speculation on the controls of forest–savanna distribution.

Second, a key aspect of ASS theory is that alternative states are stabilised by strong biological feedbacks. In the case of forest and savanna, the feedbacks are numerous and well documented (Fig. 5). Foremost, closed forest canopies have a strong suppressive effect on fire, by: (1) limiting grassy fuel loads, by reducing light availability at ground level (Hoffmann et al. 2009) and (2) decreasing the severity of fire weather at ground level, by increasing relative humidity and decreasing temperature and wind speed (Cochrane 2003). In turn, fire has the effect of reducing tree cover, by killing individual stems (Hoffmann et al. 2009) and reducing tree growth rates (Murphy et al. 2010b). Additional feedbacks that would stabilise forest cover, include increased water infiltration rates (Scheffer et al. 2005) and nutrient accumulation (Belsky 1994), as well as maintaining higher relative humidity and lower temperature beneath dense canopies to prevent desiccation of seedlings (Scheffer et al. 2005). Additionally, dispersal limitation might impose a stabilising feedback on vegetation, by slowing the invasion of disturbed forest by grasses or the invasion of savanna by forest trees, rather than imposing rigid constraints on vegetation distribution in the longer term.

Figure 5.

 Relationships and feedbacks between tropical biome (forest or savanna) and several key hypothesised biome controls. The impacts of humans on various components of this system are shown with dotted lines. The width of each black arrow approximately indicates the assumed strength of the relationship/feedback at temporal scales relevant to global tropical forest distribution.

Third, and perhaps most importantly, ASS theory suggests that abrupt shifts between ecosystem states should be possible if stabilising feedback processes are interrupted. This is consistent with the dynamic nature of forest-savanna boundaries at a range of temporal scales (Mariotti & Peterschmitt 1994; Desjardins et al. 1996; Bowman et al. 2010a; Wigley et al. 2010). Furthermore, there is evidence from a number of multidecadal fire-exclusion experiments that, in the absence of fire (the presumed stabilising process in this ASS system), forest species invade savannas, increasing canopy cover, reducing grass biomass, and eventually resulting in a complete switch from savanna to forest (Trapnell 1959; Swaine et al. 1992; Louppe et al. 1995). However, it is important to note that in some cases, especially in more arid savannas, fire exclusion does not produce a switch to a forest biome even after many decades, although tree biomass generally increases (Bond et al. 2003; Higgins et al. 2007).

This alternative stable state model would operate at very localised spatial scales, in the order of a few tens of metres, reflecting the scale of the suppressive effect of closed forest canopies on grass biomass, fire frequency and intensity (Hoffmann et al. 2009; Trauernicht et al. in press). This is in strong contrast to the earlier alternative stable state model of Sternberg (2001), who proposed that landscapes dominated by either closed forest or savanna represent alternative stable states, with the persistence of forest-dominated landscapes promoted by the positive effect of forests on regional rainfall.

Fire-mediated demographic bottleneck

The idea that fire is critical to the maintenance of savannas has become widely accepted over the last decade. Higgins et al. (2000) proposed that tree densities in savannas are limited by the effects of fire on critical life stages – a so-called ‘fire-mediated demographic bottleneck’. In mesic savannas, the critical life-stages are thought to be saplings (Sankaran et al. 2004), and repeated fires can maintain juvenile trees (saplings and smaller) in a suppressed state, potentially for decades. Only by growing through the sapling stage, and attaining sufficient height (Bond & Midgley 2000) or bark thickness (Hoffmann et al. 2009; Lawes et al. 2011) are savanna trees able to escape the ‘fire trap’ and reach maturity. This concept can be extended to forest trees regenerating at forest–savanna boundaries or invading savanna. However, rather than merely attaining a height or bark thickness sufficient to survive subsequent fires, forest trees must also attain canopy closure sufficient to exclude light, grasses and reduce fire risk. Though mature savanna trees are generally able to survive repeated savanna fires, forest trees tend to be more susceptible to topkill and much less able to cope with repeated fires (Hoffmann & Moreira 2002; Hoffmann et al. 2009).

A related concept that has emerged over the last few decades (e.g. Kellman 1984; Wilson & Bowman 1994; Fensham et al. 2003), and has recently been more clearly articulated (Hoffmann et al. 2009; Lehmann et al. 2011), is that the likelihood of a closed forest canopy forming is a function of both fire frequency and tree growth rates. If fire intervals are sufficiently long for forest seedlings to establish, mature and form a closed canopy, then forest can replace savanna. Any factor that increases growth rates, including water and nutrient availability, and soil drainage, is likely to promote the formation of forest. So too would any factor that decreases fire frequency, such as topographic fire protection or low rainfall seasonality.

‘Tree growth-fire interaction’ model

We propose a broad conceptual model that we refer to as the ‘tree growth-fire interaction’ model, combining two key concepts: (1) forest and savanna are alternative stable states, maintained a strong negative feedback between the forest canopy and fire activity, (2) it is the interaction between tree growth rates and fire frequency that determines the likelihood of a forest canopy forming to displace flammable savanna. According to the model, any factor that increases tree growth rates, such as elevated CO2, will increase the likelihood of seedlings of forest trees maturing between subsequent savanna fires, forming a closed forest canopy, thereby excluding grasses and reducing the flammability of the vegetation. Additionally, any factor that decreases fire frequency is likely to increase the chance of forest formation, e.g. rockiness, topographic fire protection, insularity. The model is described in Fig. 6, presented from the perspective of resource- or disturbance-based controls of canopy closure, which can cause a switch from one vegetation state (forest or savanna) to another.

Figure 6.

 Representation of the combined effects of plant resource availability and fire activity on canopy cover in the tropics, based on the ‘tree growth–fire interaction’ model (e.g. Kellman 1984; Wilson & Bowman 1994; Hoffmann et al. 2009; Lehmann et al. 2011). In (a), canopy cover is shown as a function of resource availability, and in (b), canopy cover is shown as a function of fire interval, although these are both alternative representations of the same model. Because tropical vegetation tends to exist in either low (savanna) or high (forest) tree cover states (Hirota et al. 2011), we would always expect an abrupt transition between savanna and forest states, regardless of the types of resources or disturbance indicated by the x-axes.

The model seems highly consistent with the view that forest distribution is shaped by a ‘constellation of weak forces’ (Mills et al. 2006), given that it accommodates a multitude of factors contributing to tree growth rates and fire return times (Fig. 5). It can account for the vast majority of the observed local- and regional -scale patterns in forest distribution, as any factor that increases tree growth is likely to increase the chance of forest formation, e.g. high rainfall and soil moisture, high nutrient availability, sheltered topographic positions. It can also accommodate putative CO2-driven forest expansion because elevated CO2 tends to increase tree growth rates, albeit mediated by other limiting factors such as soil nutrients. Understanding the relative importance of these controls remains the key challenge to tropical vegetation ecologists.

Dynamic Global Vegetation Models – A Way Forward?

The strength of the feedbacks controlling tropical forest distribution (Fig. 5) mean that simple correlative and phenomenological studies have limited utility for revealing the processes that determine forest distribution, a point made in a recent study of the intercontinental distribution of savannas (Lehmann et al. 2011). For example, there is ample evidence that tree cover and fire frequency are negatively related in tropical landscapes (Bucini & Hanan 2007; Archibald et al. 2009; Murphy et al. 2010a). Despite this pattern representing a classic ‘chicken and egg’ problem, researchers have cited the correlation as evidence that frequent fires reduce tree cover (e.g. Bucini & Hanan 2007), or conversely, that high tree cover reduces fire frequency (e.g. Archibald et al. 2009). More realistically, strong feedbacks between tree cover and fire frequency mean that one can never be considered an ultimate cause of the other. Mills et al. (2006) pointed out that because factors such as fire activity are not independent of vegetation state, there is a danger of circular reasoning if they are used to explain the existence of the vegetation state; that is, they are proximate, rather than ultimate causes. Similar arguments would apply to other components affected by vegetation feedbacks, such as soil nutrient availability and even climate (Fig. 5), although these are probably not affected as strongly and immediately as is fire frequency. If simple correlative studies examining the distribution of forest and savanna in relation to environmental variables are unlikely to provide deep insights into the true controls of the distribution of the two biomes, where do we go from here? We propose that an important way forward is the use of process-based Dynamic Global Vegetation Models (DGVMs).

In their seminal paper, Bond et al. (2005) pioneered the use of DGVMs to explore the role of fire in driving the distribution of vegetation at a global scale. Within a DGVM framework, they ‘switched off’ fire and compared the global distribution of vegetation under ‘fire on’ and ‘fire off’ scenarios, and concluded that fire halved the global area of closed forest, mostly to the benefit of savannas. Scheiter & Higgins (2009) used a DGVM developed explicitly for fire-prone tropical vegetation to substantially downplay the importance of fire in driving tropical vegetation patterns. This approach provides an excellent example of a question that would be largely intractable using a conventional ecological approach, but can be addressed using realistic DGVMs. We envisage DGVMs being parameterised and validated using extensive field data, and underpinned by conceptual models such as the ‘tree growth-fire interaction’ model. Hence, the ability of such DGVMs to accurately reproduce vegetation patterns could be used to validate the underlying conceptual models. Predictions of DGVMs could also be used to explore the relative importance of various drivers, such as climate, fire, soil properties and biogeographic effects. An example of a biogeographic effect that could be incorporated into DGVMs is the contrasting ways in which trees resprout following fire. In Africa, stems are typically topkilled by fire and trees resprout from the bases of stems; in Australia, savannas are dominated by eucalypts that have an exceptional ability to resprout epicormically, high above the ground (Lawes et al. 2011). DGVMs incorporating such biogeographic effects could be used to evaluate the importance of resprouting ability in limiting woody biomass in different savanna landscapes. DGVMs could also help generate hypotheses that can be tested with further field observation and experimentation. Hence, the use of DGVMs would not obviate the need for ecological field data and experimentation, but rather provide a framework within which to explore and rank the ‘constellation of weak forces’ (Mills et al. 2006) affecting forest and savanna distribution.

Given strong evidence that the development and maintenance of forest vs. savanna is likely to be strongly controlled by demographic processes, it is critical that DGVMs used to this end should realistically incorporate demographic processes at the individual level, including the impact of fire (e.g. Scheiter & Higgins 2009), rather than using simple biomass ‘pools’. Similarly, landscape-scale feedbacks between vegetation and fire are likely to be critical components of the forest–savanna system, so fire must be realistically simulated at appropriate scales. The advent of DGVMs that spatially explicitly account for fire activity (e.g. LPJ-SPITFIRE: Thonicke et al. 2010) make the prospect of understanding interactions between fire activity and biome distributions all the more plausible.

Conclusions

The factors that control the distribution of tropical forest and savanna have puzzled ecologists for over a century, and yet there is still only limited consensus on the relative importance of ‘bottom-up’ (resource-dependent) and ‘top-down’ (disturbance-dependent) controls, and the exact mechanisms by which these factors operate. Such understanding is critical if we are to predict the future of these vastly different biomes. The traditional view is that climate, specifically rainfall, is the overwhelming driver of forest distribution, and this generalisation holds at large spatial scales, albeit with substantial differences between continents. However, at some landscape and regional scales this relationship breaks down and edaphic and topographic factors are clearly important. For example, there is clear evidence that forest tends to be associated with nutrient-rich soils. However, fire is also accepted as a driver of tropical tree cover. High frequencies of fire in grassy savannas is thought to limit recruitment of forest trees that tend to be easily killed by repeated fires.

The recent appreciation of the interplay between fire and tree cover, especially within the savanna biome, has led us to propose the ‘tree growth-fire interaction’ model that has two key components. First, forest and savanna exist as alternative stable states, primarily held in check by a strong negative feedback between the forest canopy and fire activity. Strong evidence for this is provided by fire-exclusion experiments that have led to biome switches from savanna to forest within a few decades. Second, the interaction between tree growth rates and fire frequency determines the likelihood of a forest canopy forming to displace flammable savanna. Any factor that promotes tree growth, such as water or nutrient availability, will increase the likelihood of forest trees recruiting, maturing and forming a closed canopy in the interval between destructive savanna fires, as will any factor that increases the fire interval (such as topographic fire protection, low rainfall seasonality). Hence, the ‘tree growth–fire interaction’ model is consistent with the wide range of environmental variables observed to be correlated with forest distribution. It is also consistent with the recent global trend of forest expansion into savannas, albeit with reduced rates of expansion on resource-limited sites. It is becoming clear that elevated atmospheric CO2 concentrations are driving this expansion, in line with the model of Bond & Midgley (2000). Their model suggests that trees growing in open environments such as savannas will benefit most from elevated CO2, primarily in the form of increased growth rates.

Resolving the drivers of forest distribution has moved beyond simple correlative studies that are unable to establish ultimate causation. We propose that Dynamic Global Vegetation Models, that incorporate the numerous feedbacks associated with tropical biomes, provide a potentially important tool for exploring the relative importance of various controls of these systems and the impacts of global environmental change. It is clear that forest–savanna boundary dynamics are a powerful integrator of the effects of global environmental change. Given that they can be easily monitored using remote sensing techniques, they have the potential to provide powerful insights into the processes controlling the distributions of these two important biomes, and how their distributions will change in the future.

Acknowledgements

We thank Franziska Schrodt, Simon Lewis, Jon Lloyd, Clay Trauernicht and Jeremy Russell-Smith for commenting on the manuscript and Mark Cochrane for helpful discussions. Reviews by Timothy Paine, Colin Prentice, Bruno Herault and Christopher Baraloto greatly improved an earlier version of the manuscript. This work was supported by the Australian Research Council (Discovery Project DP0878177) and the UK’s Natural Environment Research Council (NERC), including a NERC international fellowship that enabled David Bowman to work on the manuscript.

Authorship

B.M. and D.B. developed the concept and wrote the paper; B.M. conceived and performed the analyses.

Ancillary