Carbon and trace gas dynamics in coupled land-atmosphere systems of Southern Africa

African savannas can be abstracted to a few-component system: trees, grasses, grazers, browsers and fires; whose interaction is mediated by climate, soil and human use. Observational evidence suggests that savanna systems show two primary bifurcations: a long-term one leading to "nutrient poor" savannas or "nutrient rich" savannas, and a relatively rapid one leading to sparsely-treed savannas or densely-treed savannas. Although transitions occur, the potential range of savanna structure and function within the multi-dimensional environmental space is generally not realized because stable states are favored by a combination of interactions and feed-back mechanisms. In our conceptual and numerical scheme, the nitrogen and phosphorus cycles are closely coupled and determine vegetation type; the interaction of nitrogen and carbon biogeochemistry, herbivory and fire then determines vegetation structure.

1) To develop a conceptual and model-based understanding of savanna ecosystems that will allow us to predict, with quantified uncertainty, the emergent properties of savannas that result from complex interaction and feedback between climate, soil, herbivory and fire. This will include the change in stable states under varying conditions, the resilience of the different forms to change, and the thresholds between stable states.

2) To develop our understanding of the complexity of biological systems from an analytical/systems perspective, using savannas as model ecosystems.

3) To enhance future understanding of complex biological systems through graduate and undergraduate training that profits from the research undertaken in this project. To promote graduate and undergraduate teaching of biocomplexity and a broader public awareness of the complexity and inter-relatedness of environmental systems.

This proposal brings together an interdisciplinary team with complementary expertise, including savanna ecologists with interests in the carbon and water cycle, nitrogen biogeochemistry and fire ecology, and process based ecosystem modelers teamed with a mathematician specializing in the study of dynamic systems from an analytical perspective. The project will test emerging theory regarding savanna systems with data from long-term manipulative and ‘natural experiments’. The synthesis of existing information, for instance on fire exclusion experiments in Africa, is the first step. New data collection will take place on existing long-term experiments, and ‘natural experiments’ such as nutrient-enhanced patches resulting from human settlement centuries ago. A range of modeling approaches will be used, from analytical to highly detailed dynamic ecosystem models, applied at a range of spatial and temporal scales, from the patch/hour to the continent/millenium.

· a medium number of distinctly different components, with high connectivity (systems with a large number of similar components can be summarized using relatively simple statistical laws, and small number systems can be solved deterministically)

Virtually all biological systems have these attributes to some degree. However, to make progress in understanding complex ecological phenomena, our strategy is to study a system which demonstrates phenomena resulting from complexity, but which is not so complex or poorly understood as to be intractable. African savannas offer such a system.

Savannas are here defined as ecosystems in which the net primary production is shared, to a meaningful degree, by woody plants and grasses. They are thus, at minimum, two-component systems. Of course, typical savanna communities contain many different species of woody plant (henceforth referred to as ‘trees’, although they may include woody shrubs and suffrutices) and many species of grasses and forbs. Trees and herbaceous species have dramatically different properties (Table 1), and their coexistence in savannas remains a problem with many proposed solutions but no definitive answers.

Attribute	Trees	Grasses
Individual Mass at maturity	Large (typically >10 kg)	Small (< 1 kg)
Height at maturity	5-15 m	0.5-2.0 m
Area at maturity	>10 m²	<1 m²
Longevity	100 years	1-2 year (in perennial grasses, 6 months in annual grasses)
Sexual reproduction	Relatively few, large seeds, dispersed by dehiscence or animals	Very many, tiny seeds, dispersed by wind or clinging to animals
Vegetative reproduction	Epicormic or root budding following damage is common	Tillering is the dominant method of space occupation
Position of buds	Tips of branches	Base of crown
Interseasonal carbohydrate storage	Enough to support full initial leaf expansion	Only sufficient to create a small leaf area
Leaf nitrogen content	Up to several percent, sustained	Can be initially high, but rapidly declines to below 1%

Ecologists from Southern, East and West Africa often diverge in their understanding of savanna processes and the relative importance attributed to different driving variables. This is in part because the suite of climatic, edaphic and biotic factors varies widely between these regions. For example, in the savanna grasslands of West Africa the density of wild ungulates is generally too low to impact tree recruitment. The role of domestic animals, the inverse relationship between soil nutrients and rainfall/production and the division between low productivity-high palatability-low fire regions to the north and high productivity-low palatability-high fire regions to the south is invoked to explain patterns in vegetation type, structure and transhumance (Valenza, 1981; Menaut and Cesar, 1982; Breman and De Wit, 1983; Le Houérou, 1989; Hanan et al., 1991 ). In East Africa the role of wild ungulates and their migratory patterns are considered crucial elements of savanna form, function and heterogeneity (McNaugton, 1983; 1988; McNaughton and Banyikwa, 1995 ). However, soil chemical and physical limitations on tree establishment is often responsible for the low tree cover, so somewhat less emphasis has been placed on herbivore-tree interactions (Belsky, 1985; 1992; Sinclair, 1995 ), with some notable exceptions, including elephant effects on tree cover in Masai-Mara (Dublin, 1995 ). Research in East Africa has also highlighted the interaction of soil chemistry (availability of micronutrients such as Na, K, Ca, etc.) and herbivore behavior (McNaughton, 1988; 1990 ). Southern African research has emphasized fire, and the interaction of the carbon and nitrogen cycles (Scholes and Walker 1993). An important early activity in this research will be a synthesis of results from the three main regions, and a meta-analysis of the available data, designed to identify common features (or outliers) that will assist in the refinement of the conceptual and analytical models we describe below.

While perceptions differ, our working hypothesis in this proposal is that those perceptions are not mutually exclusive and that they in fact form end-points in the gradients of climate, soil, fire and herbivory across Africa. The ‘classical model’ of savanna structure and function proposes an equilibrium between woody and herbaceous species mediated by differential access to soil water (e.g. Walter, 1971; Walker et al., 1981 ). We propose a disequilibrium model in which the linkages between the carbon, nitrogen and phosphorus cycles (all of which are impacted by water and soil) determine the frequency and intensity of grazing, browsing and fire, which in turn determines tree cover. In this model, savanna ecosystems are more complex than in the classical model, including also soil type and nutrient status, herbivory, fire and humans as essential internal components, and the climate as the main external driver. In addition, our conceptual model of savanna systems subdivides the trees into several classes, based primarily on their nutrient physiology (but with many correlates), and adds a nitrogen-fixing component (typically a forb). Herbivory is by mammals or by insects, which are either grazers (grass eaters) or browsers (tree eaters).

The regional to continental distribution of savanna types in southern Africa can be classified in terms of soil type and nutrient status. As a general rule, soils derived from the old and highly weathered African Shield parent materials have few available nutrients, low soil organic matter (SOM) binding capacity and low phosphorus availability. By contrast soils on younger, sometimes volcanic, surfaces to the south of the region, along the Rift Valley, and in the lower rainfall zones to the north and east of the region, tend to be less weathered, with higher available nutrients and higher SOM-binding capacity. The typical soil types of the African and post-African erosion surfaces give rise to the "fertile" and "infertile" savannas (Figure 1).

The fertile and infertile savannas contain distinct assemblages of tree species with contrasting physiognomy. The infertile savannas are typically dominated by trees with relatively large leaves, low leaf nitrogen content (high C:N ratio) and high tannin content. The fertile savannas are dominated by fine-leaf species (mostly bipinnate species in the family Mimosaceae), spinescence, and relatively high leaf nitrogen (low C:N). The difference in C:N ratio means that the infertile savannas sustain relatively low mammalian and insect herbivore populations compared to the fertile savannas (Scholes and Archer, 1997 ). At the continental scale this pattern has apparently been stable for millenia, with adaptation of tree species to particular soil types and coevolution of herbivores. However, it is possible to transform a broadleaf savanna into a stable fineleaf savanna by addition of nutrients (Blackmore et al., 1990 ), and the reverse transformation may be possible (although more difficult to achieve experimentally).

At regional and landscape scales the fertile/infertile division of soils and associated tree species is also apparent, driven by geomorphology and soil differences related to catenal position, slope or differential erosion that result in local nutrient depletion or enrichment. Thus the soils and associated savanna vegetation on the tops of the catena in many regions are nutrient depleted relative to the lower slopes, where influx of organic material, silt and clay from upslope locations increases the overall nutrient status (Scholes et al., 2001 ).

Savanna systems can exhibit several alternate states. Firstly, the tree-grass system is apparently unstable in the general case, and in the absence of disturbance (fire, browsing and grazing) savanna vegetation typically converges on a closed woodland endpoint, where further tree growth is limited by tree-on-tree competition, and grasses are reduced to a minor component (Scholes and Archer, 1999). However, in the presence of frequent fires or browsing/grazing, and especially when fires and browsing are combined (Dublin, 1995, Edroma, 1984; Skarpe, 1990 ), savanna vegetation structure converges on a grassy state with few, scattered large trees (and typically many suppressed trees within the grass layer).

To explain the persistence of open savanna vegetation in Africa we propose a conceptual model that builds on the initial distinction between the fertile and infertile savannas at continental-to-landscape scale and adds disturbance factors that suppress tree cover (Figure 2). In this scheme, savanna structure is controlled by disturbance (rather than competitive equilibrium) but the relative importance of fire, grazing and browsing in suppression of tree cover depends on soil nutrient status and primary productivity. Furthermore, we hypothesize that the stability of the fertile and infertile forms is related to positive feedback mechanisms that promote conservation and loss of N in the fertile and infertile savannas, respectively.

In Figure 2, parent material and weathering history of soils affect phosphorus availability (P), which limits the occurrence and relative importance of biological nitrogen fixation (BNF) by ecto- and endo-mycorrhizal associations. Soil nitrogen availability (N) affects the species composition of trees and grasses and the carbon-nitrogen ratio (C:N) of leaf material. The C:N of plant material affects palatability and digestibility to herbivores, and thus limits the population density of grazers (G) and browsers (B). Grazing intensity, in combination with seasonal and interannual variations in net production of grasses (NPP), determines available fuel load during the dry season (L) and thus the frequency and intensity of fires (F). Grazers, browsers and fires combine to determine vegetation structure by suppressing (or not) the survival of tree seedling and recruitment of seedlings into the woody canopy layer. Higher grazing and browsing intensity in fertile savannas promotes recycling of nitrogen in urine and feces, conserves N and thus promotes the fertile condition. Low herbivory in the infertile savannas reduces N recycling, increases fuel-load and thus fire frequency and intensity which increases volatilization and loss of N to the atmosphere, reduces buildup of organic matter in the soil, and promotes the infertile condition. Those savannas on the ‘export path’ of the southern African regional circulation are particularly prone to nitrogen loss to the ocean, via the atmosphere.

Figure 2. Savanna complexity affecting species composition and structure at continental-to-landscape scales. Box and arrow sizes indicate relative importance of pool or process. Solid arrows indicate primary direction of control and dotted arrows indicate positive feed-back processes that maintain the "fertile" and "infertile" stable states.

Key: P = phosphorus availability, BNF = biological nitrogen fixation, N = soil nitrogen availability, C:N = carbon-nitrogen ratio of leaves, G = grazing intensity, B = browsing intensity, Rain = available soil water on a seasonal-annual basis, NPP = herb layer net primary production, L = standing fuel load, and F = frequency and intensity of fires.

Thus overall, nutrients and water control the system at a lower level, affecting in particular the species composition, productivity and palatability of savannas, but the “structure” (primarily tree cover in this context) is dependant on the complex interactions of herbivory and fire that affect survival of tree seedlings and recruitment into the adult tree population.

In addition to regional-scale differences in savanna fertility related to soil parent material, and landscape-scale differences related to geomorphology, fine-scale heterogeneity within savannas, in particular the presence of localized nutrient-rich patches, is ubiquitous and often considered a central feature of savanna ecosystem function (McNaughton, 1984; Blackmore et al., 1990 ). These fertile patches are usually the result of very different processes than those described at the broader landscape-to-continent scale. However, we hypothesize that they are similar in respect of the processes that determine vegetation structure and the positive feed-backs that sustain elevated fertility and make the patches relatively stable features of the landscape (Figure 3).

Figure 3. Savanna complexity affecting savanna structure at landscape-to-patch scales. Box and arrow sizes indicate relative importance of pool or process. Solid arrows indicate primary direction of control and dotted arrows indicate positive feed-back processes that maintain the "fertile" and "infertile" stable states. (See Fig. 2 for key).

Figure 3 is similar to Figure 2 except that the differences in nutrient status are related to local processes rather than broad-scale geological or geomorphological factors. Local scale factors that might concentrate nutrients in a particular area include grazing lawns where herbivores preferentially graze or rest, with resulting elevated N-inputs in urine and dung, old village sites, termitaria and canopy-intercanopy differences in the vicinity of larger trees. The initial conditions establishing these patches are hard to predict (e.g. a favored grazing patch may appear following a fire in an area not previously favored). However, once established the processes of nutrient recycling, fire suppression, and the import of nutrients from adjacent areas by herbivores, can result in considerable stability.

The complex of interacting factors that control savanna ecosystem structure and function at patch to continental scales (i.e. Figures 2-3) do not, in practice, result in a multitude of savanna types distributed across the various gradients. Rather, despite the complex environmental and "driver" space (in particular the potential, if not realized, gradients in herbivory and fire frequency), the range of savanna types actually expressed in Africa is relatively condensed (Figure 4).

Figure 4 implies that African savannas exist within a two-dimensional environmental "space". However, both nutrient and recruitment axes are complex variables with many interacting contributing variables as indicated in Figures 2-3. While climate has clear affects on species composition and productivity, the fertile/infertile and open/closed schematic is valid across the range of African savanna climates. Furthermore, the definition of "closed" in this scheme (i.e. the percent canopy cover of trees) is often a function of the nutrient status of the soil and climate controls on net primary productivity. Note also that the closed form we envisage is still "savanna" as defined, although herbaceous production is relatively low. Thus African savannas appear to gravitate to one of four primary "attractors" that are relatively stable and self-maintaining. The reasons for this have already been discussed in part: they include the relatively clear dichotomy in soil nutrients (particularly P) between old and young soils, the feed-backs between herbivory, fire and nutrient status, and the self-sustaining influence of herbivory and fire on savanna structure.

In this section we discuss the ways in which savannas may be induced to transition from one stable state to another, the resilience of the stable states, and the likely thresholds between states (Table 2). We expect that in most situations, "resilience" and "threshold" will be interrelated and dependent on time (i.e. duration of effort). Some transitions are frequently observed or can be easily induced experimentally. Others are rare in practice, may be difficult to induce experimentally, and, because of high effort requirement, may be more hypothetical than actual. It is the identification and quantification of the transitions, resilience and thresholds in savanna ecosystems that forms the heart of this Biocomplexity in the Environment proposal.

There are numerous examples of savanna community state transitions. For example, the savanna vegetation of the northern Serengeti plains apparently changed from open grassland to closed shrubland around the turn of the 20^th century, in large part because rinderpest reduced wildebeest and human populations (with decreased fire frequency). However, in the 1950’s the woodlands and thickets rapidly declined because of increased browse and fire and returned to the open form now widespread across the Serengeti (Sinclair, 1995 ). The savannas of South Africa, Namibia and Botswana have, in many documented cases, changed from an open to closed state following the widespread adoption of commercial cattle ranching (eg van Vegten, 1983 ).

However, many questions and unknowns remain. For example, why have the relatively stable fertile patches within the infertile zone not coalesced during the millennia? We hypothesize that it is in part because the overall N of the landscape is limited, thus the fertile patches are “harvesting” N from surrounding areas and areal coverage is limited by supply. In addition, the rate of formation of fertile patches may be balanced by the rate of loss since, when abandoned by the factors increasing fertility, the fertile patches have little autonomous ability to sustain higher N-status and thus may gradually relax back towards the infertile state.

Similarly, we know that the closed form is the usual end-point in savanna systems where herbivory and fire are suppressed and that, once established, the closed form is rather stable. Why, then, have not small patches of closed savanna persisted and gradually increased in coverage over the millennia (as rare or random events allow the formation of new closed forest patches)? Is it that the factors favoring the open form (fire and herbivory) can still operate even when a system has migrated substantially toward the closed form? For example, ‘bush encroached’ areas in Namibia are currently undergoing a fungal disease attack which may lead to widespread tree die-back. Thus, at a landscape scale, factors favoring formation of closed savanna are in equilibrium with factors that can transform a closed savanna back to the open form. The research proposed here will allow us to examine these questions and test hypotheses explaining the observed patterns.

Figure 4. Stable states and transitions in African savannas. Soil N availability is related to parent material and weathering rates that influence biological nitrogen fixation (P-limitation), import and export of N (herbivores, humans, fire). Tree recruitment is related to browsing and grazing intensity, seasonal weather variations and fire frequency/intensity.

Transition	Description	Example scenarios at continental to landscape scales (a) and at landscape to patch scales (b)
1	Fertile to infertile	a) Decline in soil fertility as soil weathers (P declines) during millennial time-scales b) Imposed high fire frequency with low grazing intensity reduces N at decade to century time scales
2	Infertile to fertile	a) Gradual build-up in N status when N inputs (e.g. industrial N deposition) exceed losses in fire and trace gas emissions. b) Nutrient redistribution/concentration in the landscape, e.g. herbivore-favored grazing lawns, old village sites, termitaria, etc.
3	Open to closed	a) and b) Replacement of wild herbivores by high and sustained populations of domestic cattle, especially with concurrent fire suppression
4	Closed to open	a) and b) Severe disturbance of tree population caused by, for example, fire after drought or pest/disease outbreak

Table 2. Transitions between stable states in African savanna. Some transitions may occur under current conditions or can be induced experimentally, others are less easily observable/are hypothetical.

From past studies in savanna ecology we know many of the aspects of savanna function and dynamics and can paint a conceptual picture of the interacting factors that create particular savanna types and structures, and we can hypothesize the factors that lead to relative stability in these states. However, we understand little of the complex interactions that operate between stable states, or the location and nature of the thresholds and feedbacks that divide the multi-dimensional system. In this study we seek to explore and, where possible, quantify the complex interactions that result in multiple stable states and which define the intervening environmental "space", and thus the combination of factors, the strength and thresholds of those factors, that can lead to state-transitions.

Objective 1: To develop a conceptual and model-based understanding of savanna ecosystems that will allow us to predict, with quantified uncertainty, the emergent properties of savannas (functional type, structure) that result from complex interaction and feedback between climate, soil biogeochemistry, herbivory and fire.

Accepting that processes in savanna ecosystems can be greatly affected by rare, random or chaotic events as well as quasi-continuous processes, we will develop the simulation model to incorporate statistical approaches where necessary to predict stable states, thresholds and likely trajectories of savanna ecosystems under historical conditions and varying alternative scenarios.

Objective 2: To develop our understanding of the complexity of biological systems from an analytical/systems perspective, using savannas as model ecosystems.

Use savanna ecosystem models of varying complexity to examine the nature of system behavior with respect to climate, soil biogeochemistry, herbivory and fire through analysis of the combination of factors that can cause state transitions and the non-linear interaction of biotic and abiotic factors.

Objective 3: To enhance future understanding of complex biological systems through graduate and undergraduate training that profits from the research undertaken in this project. To promote graduate and undergraduate teaching of biocomplexity and a broader public awareness of the complexity and inter-relatedness of environmental systems.

This component of the project will include undergraduate curriculum development as well as graduate student and post-doctoral training in the field of biocomplexity. We will use savanna systems as our primary model, drawing also on research in other fields of biocomplexity

Successful achievement of these objectives will contribute to the theoretical understanding of savanna ecosystems in general and African savannas in particular. It will contribute to our understanding of biocomplexity in ecosystems, the ways in which complexity can be described and analyzed and the ways in which even complex systems can be predicted and managed.