Which of these continents is not like the other? Comparisons of tropical savanna systems: key questions and challenges


(Author for correspondence:
tel +61 8 8946 7762; fax +61 8 8946 7720; email caroline.lehmann@cdu.edu.au)

Continental comparisons of tropical savanna systems workshop, Working Group 49, ARC-NZ Research Network for Vegetation Function, Darwin and Kakadu National Park, Australia, September 2008

The evolution and dynamics of mixed tree–C4 grass systems has long been fertile ground for debate. These systems cover a large extent of the earth's surface, predominantly in seasonal tropical and subtropical latitudes. One-fifth of the human population and most livestock and wild herbivore biomass are found in the world's savanna. At the global scale we still know little about current and future responses of this biome to climate change and the multiscale feedbacks among climate, fire, herbivory and vegetation (Beerling & Osborne, 2006). Furthermore, the vegetation dynamics of these systems, which exhibit enormous spatio-temporal variability in woody and herbaceous biomass, structure and plant functional forms, is poorly understood. The complexity of the biome across population, landscape, continental and global scales have prohibited generalizations. Tropical savannas are currently undergoing rapid and radical human-induced transformation as a result of land-use change, climate change and changing herbivore and predator populations (Grace et al., 2006). Fire is highly prevalent and a major agent of change. For example, rapid biome shifts from forest to savanna are introduced in the Amazon region via fire and land-use change (Barlow & Peres, 2008), and the regional and global climate implications of such a rapid forest–savanna biome switch are extensive (Bond et al., 2005; Bonan, 2008). Less dramatic, but no less important, are the widely reported occurrences of bush encroachment across the savannas of Africa and Australia (Roques et al., 2001; Fensham et al., 2005). Woody encroachment alters savanna function and their potential as global carbon sinks. Given these changes, a global understanding of the evolution and dynamic of tropical savannas, the mechanics of woody encroachment, and the dynamics of grassland–savanna and savanna–forest boundaries is urgent.

‘... no single factor explains the presence or absence of grasslands, savanna or closed forests across the wet/dry tropics’

Recently, a Savanna Structure and Variation Working Group (Working Group 49; http://www.vegfunction.net/wg/49/49_Savanna_Structure.htm) of the Australian Research Council-funded Vegetation Function Network met in Darwin to undertake continental comparisons of tropical savanna ecosystems. This brought together 15 savanna ecologists with expertise from three continents to assess deficiencies in knowledge and propose research initiatives to scale up our understanding of this biome at landscape to continental scales. We asked the following questions: (1) are patterns of woody and grass biomass across environmental gradients observed in one continent comparable with those in another; (2) which quantitative and analytical models can enable a common framework within which global patterns of savanna structure can be meaningfully interpreted; (3) how and why does the extent of tropical savannas vary across the three major continental regions (Africa, Australia and South America); and (4) are savannas governed by the same factors in Australia, Africa and South America? If not, what may we infer about the generalized representations of this biome in dynamic vegetation models that predict global-scale responses to climate change and, finally, what sort of demographic, structural and compositional data do we need to capture this heterogenous biome accurately in global vegetation models?

During this first meeting, this synthesis was progressed by collating data (both field-based and remotely sensed) on tree basal area, woody cover, grass biomass, species traits and climatic variables for Africa, Australia and South America savanna systems. The immediate goals are twofold: first, these data will be used for empirical comparisons of savanna vegetation traits across continents; and, second, we will use these data to parameterize analytical and simulation models to predict potential responses of this biome to climate change and shifts in disturbance regimes.

Based on current knowledge, it is clear that no single model or paradigm explains savanna vegetation dynamics (Sankaran et al., 2004). It is equally clear that no single factor explains the presence or absence of grasslands, savanna or closed forests across the wet/dry tropics. It is, however, clear that the relative importance of water, nutrients, climate stochasticity, fire and herbivory vary substantially across these regions in determining climatic limits to standing biomass and its variance (Bond, 2008). Understanding the potential limits to standing biomass and the variability in biomass invoke vastly different mechanisms and require different ways of thinking, for example Fig. 1, discussed below.

Figure 1.

(a) Woody maxima occur in savanna systems, with potential maxima increasing with rainfall. This bound on woody cover is seen as a fundamental descriptor of a savanna region and is primarily expected to reflect limits on woody biomass set by climatic variables. However, the rate at which maximum woodiness increases with rainfall, as well as the point where canopy closure occurs, can differ in different savanna regions across the globe and may be closely linked to differences in the nature of the rainfall regime as well as differences in growth rates, architectural forms and physiological adaptations of dominant woody species in different regions. (b) Another defining aspect of savanna biomes may be the observed mean and variation in woodiness across an environmental gradient. Understanding these aspects of woodiness in savannas is a fundamentally different question. This invokes a complex interplay of drivers of savanna dynamics such as fire and grazing regimes, effects of soil types on water and nutrient availability, woody species demography as it relates to fire, nutrient and grazing regimes, and the relative prevalence of deterministic vs stochastic processes.

Climatic limits to woody biomass

Work presented by Mahesh Sankaran (University of Leeds, UK), Jayashree Ratnam (University of Leeds, UK) and Niall Hanan (Colorado State University, USA) reports a bound on woody cover in arid and semi-arid regions of Africa (Sankaran et al., 2005, Fig. 1), with potential maximum woody cover increasing linearly with mean annual rainfall, until potential canopy closure. This bound on woody cover is primarily limited by water, and may be interpreted as a fundamental descriptor of a savanna region. After examination of data from Australia and South America presented by Richard Williams (CSIRO Sustainable Ecosystems, Australia), Caroline Lehmann (Charles Darwin University, Darwin, Australia) and Jeanine Felfili (University of Brasilia, Brazil), it is evident that the rate at which maximum woodiness increases with rainfall, as well as the points of potential canopy closure, differ across savanna regions of the world (Fig. 1a).

Spatio-temporal patterns of woody biomass

Another defining aspect of savannas explored by the working group is the observed mean and variation in woodiness across environmental gradients (Fig. 1b). With quantified differences between continents in fire, rainfall, soil, herbivore communities, cyclonic disturbances and historical contingencies, we expect major differences in savanna structure across continents, but how can we interpret this in a common framework? Deciphering differences in tree allometries, architecture and demographics of the savanna regions is at the heart of understanding the real structural variability in the biome. Once such nuanced variability is accounted for we can undertake a standardized assessment of the spatio-temporal trends in tree biomass. Furthermore, we can then usefully delve into questions regarding the role of phylogenetic history in determining the characteristics of each region and ask: has the co-evolution of tree and grass species of each continent informed the unique signature of a savanna region?

To achieve this, the working group began using several statistical and simulation modelling approaches, each addressing a different aspect of the cross-continental comparison. Michael Andersen (University of Gröningen, the Netherlands) discussed structural equation modelling, a technique borrowed from the social sciences that is useful for comparing large multivariate data sets with different causal pathways, interactions and feedbacks between driver and response variables (Grace, 2006). Adam Liedloff (CSIRO Sustainable Ecosystems, Australia) parametized the FLAMES simulation model of tree demographics developed for a mesic Australian savanna (Liedloff & Cook, 2007) with data for a southern African savanna. This exercise generated questions about variance in water-use and phenology of tree species between Africa and Australia. Simon Scheiter and Steve Higgins (both at the Goethe University Frankfurt, Germany) presented an exciting dynamic global vegetation model (DGVM) based on African savannas that operates novel mechanistic submodels of fire, tree allometry and phenology (Scheiter & Higgins, in press). This DGVM was parametized for the Australian savanna systems in cooperation with Lindsay Hutley (Charles Darwin University, Darwin, Australia) and we anticipate beginning simulations of Australian systems in the coming months.

The global extent of the biome

There has been surprisingly sparse quantitative analysis of the global extent of the savanna biome. In order to characterize the dynamics of this system at a global scale, there must be an accurate appreciation of extent and boundary conditions. Hence, we classified expertly derived vegetation maps of Africa, Australia and Brazil as to whether or not there was a dominant C4 understorey. Overlaying this information against indicators of productivity, burnt area and grazing pressures presented us with clear envelopes of savannas for Africa, Australia and South America. Our attempts to explain these differences led to an exciting conceptual development. William Hoffmann (North Carolina State University, USA) proposed that tropical savannas potentially lie between two points of intersection where the time to canopy closure intersects the potential minimum fire-return time. We suggest that this may be an interesting new way to explore how the savanna biome transitions between savanna and nonsavanna at arid and wet ends of its distribution. Sally Archibald (CSIR, South Africa) and Caroline Lehmann discussed the likelihood of changes in plant architecture along these environmental gradients being related to the strategy most likely to avoid top-kill by fire (i.e. growing tall vs wide). We suggest that critical insights will come from paired comparisons of woody species traits relating to fire-responses, growth rates and seedling colonization of different grass mixtures under the boundary conditions with analysis for each savanna continent.

A system maintained via demography

William Bond (University of Cape Town, South Africa) noted that the savanna biome is the only biome on earth to be maintained by demography and not by climate (Bond, 2008). The stochastic processes of fire, herbivory, interannual climate variability and cyclones present major challenges for small trees trying to escape a period of vulnerability before becoming disturbance tolerant and reproductive. In small trees (that are not necessarily young), key to establishment success are architecture, growth rates and resprouting ability. A comparison of species traits of congeneric pairs of savanna and forest tree species by William Hoffmann demonstrated that minimum fire-return times interact with differences in growth rates, and resprouting ability potentially generated sharp forest–savanna boundaries in the Brazilian cerrado. Lynda Prior (Charles Darwin University, Darwin, Australia) presented data on growth rates and recruitment of sapling trees in northern Australia, showing that these traits are affected by the fire-return intervals, fire intensity, herbivory and climate stochasticity, whereas large trees remained relatively unaffected.

In the future, combining data on tree demographics, top-kill responses and allometry will enable us to continue to explore fundamental questions about resource use and disturbance responses of woody species of different savanna regions, and how species-level traits scale-up to feed into savanna community structure. How much wood must a tree in South America, Africa and Australia produce before becoming disturbance tolerant? And how much does it cost to produce this biomass given differences in soil water and nutrient availabilities? How fast do different woody species grow between characteristic fire-return intervals? Can we predict which tree species and/or age classes will escape a fire-determined demographic bottleneck? Answering these questions will ultimately enable us to disentangle the directions through which a changing climate and altered atmospheric CO2 concentrations will drive this biome using proximate and measurable responses of tree demographic stages.

The Future – anomalies and challenges

Over the course of our discussions, it became clear there are exciting areas for future explorations. Savanna regions across the globe are characterized by different suites of plant functional groups, but explanations for these patterns are far from obvious (Bowman & Prior, 2005). Why are mesic savannas in Australia characterized by evergreen woody species and annual grasses? Why are the arid and mesic savannas of east and southern Africa dominated by deciduous species and perennial grasses, whereas deciduous trees and annual grass species dominate in arid western Africa? How does the evolution of these traits relate to differences in herbivory and fire regimes? Does the dominant form of herbivory (i.e. large mammal, small mammal or insect) have fundamentally different impacts on the woody–herbaceous dynamic of these ecosystems, and can we develop generalizations on these? Where are the savannas of the seasonal tropics of Asia? Most literature on this region refers to deciduous or thorn forests, but these regions are climatically similar to the savanna regions of other continents. Are savannas actually infrequent in tropical Asia or is this the result of different semantic traditions in vegetation classification? Finally, how is the savanna biome placed in the context of the current larger ecological debate on biodiversity and ecosystem function?

As a group, we are excited about the future and plan to continue with a second working group meeting in 2009. With large continental data sets collated, several well-developed models becoming available and international groups working in collaboration (e.g. TROBIT, AMMA, SAFARI and Working Group 49), savanna ecology is poised for exciting ideas and insights in the coming decade that are applicable to ecology at the broadest scales.


This working group is sponsored by the ARC-NZ Vegetation Function Network. C. Lehmann would like to thank the participants of the Working Group 49, all of whom are mentioned in the body of this report, for their energy and enthusiasm in exploring these interesting questions as a collective.