Effect of stocking rate and rainfall on rangeland dynamics and cattle performance in a semi-arid savanna, South Africa

Authors


R.W.S. Fynn (fax 27 33 2605708; e-mail Fynn@agric.unp.ac.za).

Summary

1. In order to examine the emerging paradigm of non-equilibrium behaviour of plant–livestock relations in semi-arid rangeland, the effect of stocking rate, rainfall and their interaction on changes in botanical composition, primary production and live weight gain per animal and per hectare, was studied in a semi-arid African savanna. The objective was to evaluate the relative influence of rainfall and grazing on animal and vegetation dynamics in a temporally varying environment.

2. Two adjacent trials, with different starting conditions of rangeland (good vs. poor) and each of three stocking rates replicated twice, were established in 1986 and maintained for 10 years. A simple rotational grazing system using Brahman weaners was employed.

3. Although changes in botanical composition were strongly influenced by rainfall variability, with a dramatic compositional shift induced by the 1991–92 drought, stocking rate had an additional effect over time in the paddocks on sloping land, particularly on the site which started in good condition. High rainfall and light grazing promoted tufted perennial grasses (Themeda triandra, Digitaria argyrograpta, Cymbopogon excavatus, Sporobolus ioclados); heavy grazing and low rainfall promoted some annuals and weakly tufted perennial grasses (Urochloa mosambicensis, Sporobolus nitens); while other annuals (Aristida adscensionis, Enneapogon cenchroides) were favoured by heavy grazing and high rainfall. Patterns of compositional change supported a state-and-transition model.

4. Rainfall had the most marked effect on variability in herbaceous production. Long-term heavy grazing on sloping land resulted in a decline in herbaceous production in both trials. The depletion of herbaceous biomass in a paddock when grazed heavily was more pronounced if botanical composition had changed as a result of drought and grazing.

5. Long-term heavy grazing did not reduce cattle performance (gain animal−1 and gain ha−1). However, during drought cattle performance was worse at high stocking rates on poor condition rangeland than on good condition rangeland. Rainfall was a better predictor of cattle performance than herbaceous biomass and accounted for far more of the variance in gain per animal than did stocking rate. Cattle performance had a curvilinear relationship with rainfall, indicating that a rainfall year of 680 mm is optimal for cattle production in this region.

6. The notion that semi-arid African savannas are non-equilibrium systems in which rainfall overrides grazing was contradicted. Stocking rate determined the requirement of supplementary feeding and influenced gain ha−1 on poor condition rangeland during drought years. In addition, herbaceous productivity was linked to herbaceous composition, which was linked to stocking rate.

7. Key implications for management are (i) the inequality of different parts of the landscape in supporting livestock and in their sensitivity to grazing, slopes being more easily degraded than bottomland; and (ii) the pronounced changes that grazing can induce in semi-arid savanna during and subsequent to drought years. Opportunistic management is a prerequisite for sustained utilization of semi-arid African savanna.

Introduction

The theoretical underpinnings of range management, predicated upon the concept of Clementsian succession and an equilibrium theory of ecosystem functioning, have been challenged for arid and semi-arid environments in which rainfall variability appears to be a potent determinant of system change (Westoby, Walker & Noy-Meir 1989; Behnke & Scoones 1993). Traditionally, stocking rate has been considered the tool whereby range managers can adjust the successional trend of vegetation. It is now argued that grazing has minimal or no influence on the vegetation dynamics of arid environments because of pronounced interannual rainfall variability, such that these grazing systems are considered to be non-equilibrial (Ellis & Swift 1988). The notion that grazing has a minimal impact on vegetation has been extended to the pastoral systems of the semi-arid savannas of Africa (Behnke & Scoones 1993). Certainly, rainfall variability can exert a strong influence on production (Deshmukh 1984; Le Houerou, Bingham & Skerbek 1988) and species composition (O’Connor 1985), although the slight effect of grazing on species composition in one year can accrue over time to manifest as a striking impact (Milchunas & Lauenroth 1993; O’Connor & Roux 1995). Although large ungulate biomass is positively correlated with mean annual rainfall for the savannas of Africa (Fritz & Duncan 1994), implying a relationship between animal biomass and food supply, it is not a logical sequitur that interannual variation in animal biomass in a local area is regulated by density dependence, but may rather be a direct response to rainfall variability (Ellis & Swift 1988). Alternatively, drought episodes may potentiate herbivory and result in tight plant–herbivore coupling during these episodes (Illius & O’Connor 1999).

Degradation is an unambiguous indicator of whether grazing adversely affects system functioning, when degradation is defined as ‘an effectively permanent decline in the rate at which land yields livestock products under a given system of management. This definition excludes reversible vegetation changes even if these lead to temporary declines in secondary productivity. It includes effectively irreversible changes in both soils and vegetation’ (Abel & Blaikie 1989). This definition precludes reversible changes in soil resources that would impact both primary and secondary production (Abel 1993a, b; Biot 1993). This definition is demanding of research because long-term data sets are required for the detection of an irreversible decrease in primary or secondary production, although Wilson & Macleod (1991) have identified that departure from a linear form of the relation between animal production and stocking rate can constitute evidence of degradation. Even the occurrence of erosion at a point in the landscape need not be evidence of degradation because eroded material may become redistributed within the landscape (the concept of erosion cell mosaic; Pickup 1985) such that, although primary production of the eroded area is decreased, there may be no decrease of production over the landscape because of a corresponding increase in productivity in the areas receiving eroded material (Scoones 1992). Empirical support of this claim is, however, lacking (Illius & O’Connor 1999).

The non-equilibrium school has further challenged the reliance of the traditional school on botanical composition as an indicator of range condition for semi-arid African savanna (Abel 1993b) because, in some cases, transformation of a ‘climax or subclimax’ grassland (normally assumed to support the greatest animal production) to early seral stages may result in increased animal production (Harrington & Pratchett 1974), and because it is difficult to extricate the effects of animals on vegetation from those of rainfall (Abel 1993b). Compositional change in response to grazing is far more widely studied in African savannas than production or yield (O’Connor 1985), such that a key question is whether changes in composition are of any significance for primary production.

Illius & O’Connor (1999) have disagreed about the non-equilibrium behaviour of semi-arid African savannas based on theoretical and empirical reviews of density dependence in animal populations, and of the effects of grazing on species composition, primary production, water balance within the system, erosion and animal performance. Grazing management of semi-arid African savannas therefore awaits a resolution of the theoretical conflict between equilibrium and non-equilibrium schools of thought, or at least clarification of the controls of vegetation and animal dynamics for systems across a rainfall gradient.

This study was based on manipulation of stocking rate for cattle, and examined changes in vegetation composition and production, and animal production, over a period of 10 years, including extremes of rainfall. The design of the study allowed us to compare the efficacy of equilibrium and non-equilibrium concepts for this semi-arid savanna (we examine only one position along the rainfall gradient) and to explore the concept of degradation. Specifically, the aim of the study was to account for the relative contribution of (i) rainfall; (ii) stocking rate; (iii) dependence of stocking rate on rainfall; and (iv) botanical composition on changes in (a) botanical composition, (b) herbaceous biomass and (c) animal performance.

Methods

Study area

The study was conducted on the farms Llanwarne (27°35′S; 31°45′E) and Dordrecht (27°36′ S; 31°46′E), which are located in the Pongola region of the Zululand bush veld. The topography can be described as a gently undulating low-lying basin (274 m a.s.l), incised by ephemeral streams draining into the Mkuze river. The climate is typical of a subtropical semi-arid environment, with a low and variable rainfall (Fig. 1), 70% of which falls in the summer months (November–April) (mean 1977–96 = 568 mm, coefficient of variation (CV) = 30%). The months with the highest mean maximum and lowest mean minimum temperatures are in January (31·2 °C) and June (6·2 °C), respectively. The lithology at both sites consists predominantly of sandstones and shales of the Karoo Sequence. Soils are generally medium to fine textured with a blocky structure. The dominant soil type at both Llanwarne and Dordrecht is the Swartland form (Soil Classification Working Group 1991). The vegetation has been classified as Lowveld (Acocks 1953), which consists of tropical and subtropical species occupying the plains (150–600 m a.s.l) between the interior plateau and the Lebombo range. It is an Acacia-dominated savanna at both Llanwarne and Dordrecht, with broad-leafed species dominating along drainage lines. Themeda triandra is a common grass species in the open and Panicum maximum under trees. Nomenclature is according to Arnold & De Wet (1993).

Figure 1.

Seasonal rainfall data collected at Llanwarne meteorological station between 1982 and 1996.

Experimental design

The trials were established in 1986 and terminated in 1995 at Llanwarne and in 1996 at Dordrecht. In 1986, the Llanwarne and Dordrecht sites were determined to be in good and poor range condition, respectively (Turner 1988). Details of the experimental design of the trials are shown in Table 1. The first replicate of the low, medium and high stocking rate treatments will be referred to as L1, M1 and H1, respectively, and similarly as L2, M2 and H2 for replicate 2. The stocking rates chosen for the study were based on recommended norms for commercial beef production in this region, but the stocking rates of communal systems may be 2–3 times higher (McKenzie 1982).

Table 1.  Details of the stocking rate experiments at Llanwarne and Dordrecht
 LowMediumHigh
  • *

    AU = animal units; 1 AU = 450 kg steer.

  • Treatment sizes (ha) are the sum of the areas for each replication of a treatment.

Llanwarne
Stocking rate (AU ha−1)*0·1560·2380·313
Treatment size (ha)51·334·229·2
Number of cattle889
Number of replications222
Dordrecht
Stocking rate (AU ha−1)0·1640·2080·278
Treatment size (ha)36·628·625·3
Number of cattle667
Number of replications222

The cattle used in the trials were weaners of a Brahman-cross type and weighed about 250 kg on introduction, attaining weights of up to 500 kg during a year. The experimental cattle were replaced with new weaners in October of each year. A rotational grazing system at each stocking rate was used, whereby the cattle were rotated between the two replications of a treatment. The period of stay in each replication was variable, depending on the amount of forage available in that replication (Hatch 1994). The drought of the 1991–92 season resulted in critical fodder shortages such that cattle in all treatments had to be supplemented with sugarcane tops during the winter of 1992, and subsequently the cattle in the medium and heavy stocking rate treatments at Dordrecht were supplemented in the winter of 1993.

Data collection

Daily rainfall data was collected at both Llanwarne and Dordrecht for the duration of the trial. Species composition of each replication of a treatment was surveyed biennially between 1986 and 1996 using the nearest plant method, which yields relative abundance values for each species (Foran, Tainton & Booysen 1978), with 150 samples taken systematically along each of two diagonal transects across a paddock (300 samples per paddock in total). Grasses were identified to species level, sedges were grouped, and other herbaceous species were grouped as forbs. If no herbaceous plant (woody species were ignored) occurred within 20 cm of the point then bare ground was recorded.

An estimate of grass biomass was made in each treatment paddock every 3 weeks for the duration of the trial using a disc-pasture meter (Bransby & Tainton 1977), with 50 readings taken along a diagonal transect in each camp. Cattle were weighed every 3 weeks after being penned the night before weighing.

Data analysis

Species composition

In order to examine the magnitude of compositional change between 1986 and 1996, Euclidean distances (Manly 1986) were calculated for the composition data in 1986 and 1996 for each paddock, which were then compared with the 99% confidence interval about the mean Euclidean distance between paddocks for 1986. A one-tailed t-value was used to calculate confidence intervals because the objective was to determine if the paddocks had changed more between 1986 and 1996, than the differences among paddocks in 1986 (i.e. had a paddock moved outside the bounds of the original composition of the trial).

Patterns of compositional change were examined with ordination analysis using canoco (ter Braak 1988). The effect of repeated measures on a site was first removed and partial ordinations were then conducted on the residual variance using correspondence analysis (CA) and canonical correspondence analysis (CCA). A combined data set for Llanwarne and Dordrecht was used to test whether the different starting conditions of the two sites had had an effect on compositional change. Data were subjected to an unconstrained analysis (CA) for the purpose of obtaining an undistorted plot of site trajectories, and a constrained analysis (CCA) in order to test directly whether included environmental variables accounted for a significant amount of variance in species composition. A Monte Carlo permutation test (restricted form, with 99 permutations) was used to test whether the ordination and first axis were significant (ter Braak 1988).

Environmental variables included in the ordination analysis (after checking for colinearity) were: stocking rate by time (number of years); summer rainfall preceding the survey, as well as its product with the previous season's rainfall (e.g. a good yield in the previous season may result in abundant seed production, which then provides the potential for successful establishment in the following year); site, which was coded as a dummy variable because of the different starting condition of sites (Turner 1988); and an interaction between site and the grazing by time variable, in order to determine if the starting condition of the sites had an influence on the effect of grazing on compositional change. Panicum species (Panicum maximum, Panicum colouratum and Panicum deustum) were lumped because of possible inconsistencies in identification by different field workers over the years.

Seasonal peak herbaceous biomass

The effects of rainfall, species composition, grazing and their interactions on herbaceous biomass were examined with multiple regression analysis, using genstat 5.3 (genstat 1993). The Llanwarne and Dordrecht data sets were analysed separately.

Because the estimates of biomass were derived from a two-stage sampling process using a disc-pasture meter (Bransby & Tainton 1977), for which the calibration is linear (Turner 1988), we decided to use disc height rather than estimates of biomass in order to improve the stability of regression solutions. The peak of disc height for a growing season (i.e. following the first substantial rains) was used as the response variable, after loge transformation. The effect of grazing within a season, however, will probably result in an underestimate of peak herbaceous biomass. The number of grazing days ha−1 that a paddock had received during a growing season was included in the regression model to account for any underestimate of peak herbaceous biomass that may have resulted from grazing.

The following were chosen as predictor variables after checking they were not correlated.

  • 1. Rainfall, because of its obvious effect on herbaceous productivity.
  • 2. Stocking intensity of a season (Fourie, Opperman & Roberts 1985; Ralphs, Kothmann & Taylor 1990), represented as grazing days ha−1, which was calculated as the number of days an area was grazed multiplied by the stocking rate in animal units ha−1 (AU ha−1) for the period from the first rains until peak grass biomass was attained.
  • 3. The accumulated amount of grazing (sum of grazing days ha−1 for all years, from start of the trial up to and including the year in question) that a treatment had sustained over time was included to identify possible long-term influences on grass biomass.
  • 4. Species composition, using the site scores from the correspondence analysis. Because surveys of composition were conducted biennially, the score for a year without a survey was obtained by taking the mean of the scores for the previous and following year. Only the site scores of the first two axes were used because they captured most of the variance (see the Results).
  • 5. The interaction between composition (axis 1 and axis 2) and grazing days ha−1.
  • 6. The interaction between slope (upper or lower at Dordrecht and flat or sloping at Llanwarne) and the accumulated amount of grazing, because differences in topography are likely to result in differences in the availability of resources to grasses, which may affect their tolerance to heavy grazing. The main effect of slope could not be fitted in either the Llanwarne or Dordrecht models owing to a strong correlation (r = 0·75) with the accumulated grazing–slope interaction.

If high stocking rates were responsible for a loss of herbaceous productivity over time, this should then be reflected by an increasingly negative coefficient (βX) in the relationship between stocking rate and herbaceous production for each year. In order to test this the coefficient (βX) of the line between stocking rate and herbaceous production was determined for each year using regression analysis. These coefficients were then regressed against time (year 1, 2, 3, etc.) to determine if they became increasingly negative as the trial progressed.

Animal production

Multiple regression analysis using genstat 5.3 (genstat 1993) was done for seasonal gain animal−1 and for seasonal gain ha−1 (the response variable was loge transformed).

The following were the chosen set of predictor variables.

  • 1. Stocking rate (Jones & Sandland 1974).
  • 2. Accumulated grazing days ha−1 in order to examine whether the effect of heavy stocking manifests over time.
  • 3. Peak herbaceous biomass (as disc height); a separate analysis was conducted in which peak herbaceous biomass was replaced by rainfall. These two variables could not be included in the same model because one is a function of the other.
  • 4. The interaction of rainfall and stocking rate because animal production may be affected with high rather than low stocking rates at low rainfall. Also, the possible effect of stocking rate on sward structure, and hence forage intake (Stobbs 1973, 1975), may depend on the amount of rainfall.

In addition, the coefficient (βX) of the line between stocking rate and gain per animal was determined for each year using regression analysis. These coefficients were then regressed against time (year 1, 2, 3, etc.) to determine if they became increasingly negative as the trial progressed. A significant negative trend would confirm a departure from the linear form of the relation between animal production and stocking rate (i.e. declining gain per animal in the high stocking rate paddock) and constitute evidence of degradation (Wilson & Macleod 1991).

Results

Species composition

A summary of species composition at the Llanwarne and Dordrecht sites in 1986 and 1996 is presented in Table 2. It was evident that at the start of the trial the Llanwarne site was in better condition than the Dordrecht site, as seen by the lower percentage of palatable perennial grass species at Dordrecht, such as Digitaria argyrograpta, Panicum species, Themeda triandra and Heteropogon contortus, and a higher percentage of pioneer grasses, such as Sporobolus nitens, Urochloa mosambicensis and forbs (Table 2).

Table 2.  Summary of plant composition data (relative abundance) for Llanwarne and Dordrecht in 1986 and 1996
Llanwarne 1986Llanwarne 1996Dordrecht 1986Dordrecht 1996
SpeciesL1M1H1L2M2H2L1M1H1L2M2H2L1M1H1L2M2H2L1M1H1L2M2H2
Aristida adscensionis          0·70·7            
Aristida congesta0·30·3 0·31·00·32·73·31·70·73·71·30·3 1·7   2·31·02·31·05·03·0
Bothriochloa insculpta0·30·32·74·00·36·70·70·71·05·31·71·71·00·30·30·70·31·01·3  1·3 0·3
Cenchrus ciliarus1·72·30·30·30·7        0·3          
Cymbopogon excavatus             0·3          
Chloris virgata      0·70·32·01·72·01·0         1·01·30·3
Dactyloctenium australe      0·31·70·71·3            0·3 
Digitaria eriantha5·02·7   0·3    0·34·3  1·30·30·32·72·0 1·00·3  
Digitaria argyrograpta24·725·323·017·715·322·38·014·717·711·02·74·024·07·310·312·02·02·76·310·39·77·02·72·7
Enneapogon cenchroides                     0·3 0·3
Eragrostis curvula      0·30·30·3 0·3             
Eragrostis superba1·01·71·70·30·71·30·32·01·72·01·00·3  2·06·715·39·72·31·31·32·00·71·3
Eragrostis sp.1·0 0·31·0 0·70·31·0 0·7  4·71·04·30·30·30·72·00·3  0·712·3
Eustachys paspaloides2·31·01·01·72·03·3      3·3 0·31·01·71·3      
Fingerhutia africana        0·3             0·3 
Heteropogon contortus1·00·71·00·30·30·3 0·70·70·71·3   0·3   2·00·30·3   
Panicum spp.46·330·742·636·438·037·752·729·330·037·742·722·734·728·411·631·726·038·739·630·719·737·341·741·6
Melinis repens              0·3         
Sporobolus fimbriatus       0·3   0·3      0·30·7  0·30·3
Sporobolus nitens      0·7 0·7  0·3 9·735·0  5·71·33·74·71·03·31·0
Sporobolus ioclados1·08·36·09·36·010·30·38·37·72·32·74·011·06·04·022·715·30·73·73·014·74·33·72·0
Themeda triandra6·05·06·75·34·72·01·01·71·09·02·01·70·30·73·31·711·08·33·3  1·70·30·7
Tragus racemosus0·71·01·0 2·02·31·72·31·01·32·31·32·31·02·31·7 2·31·70·70·71·01·7 
Trichoneura grandiglumis                  0·30·3    
Urochloa mosambicensis1·317·78·612·05·33·714·712·710·011·721·719·012·314·39·07·35·014·713·319·37·315·36·06·3
Forbs7·33·05·010·017·78·714·319·720·313·710·035·35·725·710·013·020·711·717·324·032·021·323·326·3
Sedge      0·71·00·30·32·00·3      0·70·70·71·0  
Unallocated bare ground   1·36·0 0·3 3·00·73·01·70·35·03·71·02·0  3·74·72·04·70·3

The correspondence analysis identified clear-cut patterns of vegetation change that were clearly associated with measured variables (Table 3). There was little divergence in the site trajectories over time among the various stocking rate treatments except for the latter few years (Fig. 2a–d), which indicates site trajectories were correlated mainly with time rather than with stocking rate but that stocking rate had a greater influence in later years. The trajectories moved mainly along axis 1, which was primarily a rainfall effect with some influence of grazing (Fig. 2e). The significant effect of rainfall on compositional change was primarily a product of current and preceding season's rainfall rather than a single season's rainfall, as indicated by the longer arrow (Fig. 2e), and was the variable most strongly correlated with axis 1 (Fig. 2e; P < 0·01). Grazing by time was not significant on axis 1 (P > 0·10) but was significant on axis 2, being the most strongly correlated variable with this axis (Fig. 2e; P < 0·01). Thus axis 2 was primarily a grazing gradient, although rainfall also had a significant effect on this axis (P < 0·01). The drought of the 1991–92 season precipitated a pronounced compositional change, as seen by the large distance moved in ordination space between the paddocks in 1990 and 1993 (Fig. 2a-d), after which stocking rate had an increasing influence.

Table 3.  Summary of the overall performance of the correspondence analysis in terms of the variance accounted for by each axis and species–environment relations
Correspondence analysis
Axes
1234Total inertia
Eigenvalues0·1070·0780·0680·0610·716
Species-environment correlations0·6800·6830·7040·623 
Cumulative percentage variance
 of species data
19·633·946·357·5
 of species-environment relation30·552·873·688·1 
Sum of all unconstrained eigenvalues (after fitting covariables)0·545    
Figure 2.

Time trajectories and plot of species and environmental variable scores (CA) of the combined Llanwarne and Dordrecht data set. Replicate 1 of Llanwarne (a); replicate 2 of Llanwarne (b); replicate 1 of Dordrecht (c); replicate 2 of Dordrecht (d); species and environmental variable scores (e) (scores of environmental variables are multiplied by 10 to bring to the same scale as the species scores). Circle, low stocking rate; triangle, medium stocking rate; square, high stocking rate; AAD, Aristida adscenionis; ABA, Aristida congesta; BIN, Bothriochloa insculpta; CEX, Cymbopogon excavatus; CVI, Chloris virgata; DAU, Dactyloctenium australe; DAR, Digitaria argyrograpta; DER, Digitaria eriantha; ECE, Enneapogon cenchroides; FOR, Forbs; FAF, Fingerhuthia africana; MRE, Melinis repens; PAN, Panicum species; SIO, Sporobolus ioclados; SNI, Sporobolus nitens; TGR, Trichoneura grandiglumis; TRA, Tragus racemosus; TTR, Themeda triandra; UMO, Urochloa mosambicensis; UNA, unallocated bare ground; Rain, rainfall immediately preceding the survey; Rain2, product of the season's rainfall preceding the survey and the previous season's rainfall; Graze, stocking rate × time since start of trial; SiteXGr, interaction between site (Llanwarne and Dordrecht) and the grazing time variable.

The greater shift in ordination space by the heavy stocking treatment compared with other treatments by 1996 (Fig. 2b,c) indicated the instability of the vegetation of certain high stocking rate paddocks. This was supported by a comparison of the Euclidean distances between paddocks in 1986 and 1996 (the greater the Euclidean distance, the greater the shift in composition between 1986 and 1996) with the confidence limits of the mean Euclidean distance between paddocks in 1986 (the mean Euclidean distance between paddocks in 1986 indicated the mean range in composition between paddocks in 1986). At Llanwarne, all except the L2 paddock had changed outside the bounds of the Euclidean distances between paddocks by 1996, although the H2 paddock had changed considerably more than the other paddocks (Fig. 3a). This compositional shift involved large decreases in certain tufted perennial grasses, such as Digitaria argyrograpta, Sporobolus ioclados, Cenchrus ciliarus and Panicum species, and large increases in pioneer grasses, such as Urochloa mosambicensis and forbs, and moderate increases in other pioneer grasses, such as Aristida congesta and Chloris virgata (Table 2). At Dordrecht, the site that started in poor condition, only the H1 paddock had changed by 1996 outside the bounds of the Euclidean distances between paddocks in 1986 (Fig. 3b). The composition of the site that started in poor condition (Dordrecht) had therefore changed less and was less prone to movement outside its original bounds (Fig. 3) than composition of the site that started in good condition (Llanwarne). The direction of change, however, was not influenced by the starting conditions of the sites (direct correlation of the site by graze variable with the graze by time variable; Fig. 2e).

Figure 3.

Euclidean distances between 1986 and 1996 composition data for each treatment (i.e. a measure of compositional change in each paddock over the duration of the trial where large values of Euclidean distance indicate greater compositional change than smaller values) compared with the 99% confidence interval (horizontal line) about the mean Euclidean distance between treatments for 1986 (i.e. setting a confidence limit to the mean range in composition between paddocks at the start of the trial, in order to be able to test whether any grazing treatments resulted in a Euclidean distance value greater than the original differences between paddocks at the start of the trial). Llanwarne (a) and Dordrecht (b).

The classical antagonistic effect of high rainfall with light grazing causing relative increases in densely tufted perennials such as Themeda triandra, Cymbopogon excavatus, Digitaria argyrograpta and Sporobolus ioclados, and heavy grazing with low rainfall (drought) causing relative increases in annuals and weakly tufted perennials such as Urochloa mosambicensis, Aristida congesta, Tragus racemosus and Sporobolus nitens, was apparent (Fig. 2e). This trend did not hold for all species, however, as the annuals Aristida adscensionis and Enneapogon cenchroides, the creeping perennial Dactyloctenium australe and the weakly tufted perennial Trichoneura grandiglumis, increased most rapidly under a combination of high rainfall and heavy grazing (Fig. 2e). Thus annuals tended to increase relative to perennials with high stocking rates, but rainfall may elicit inconsistent responses among different species (i.e. some annuals increased with high rainfall and some with low rainfall), probably because of unknown species-specific requirements for germination and establishment. Certain densely tufted perennials, such as Panicum species, Bothriochloa insculpta and Cenchrus ciliarus, increased optimally at moderate levels of both rainfall and grazing. Compositional change did not appear to occur in gradual transitions, but rather in relatively rapid changes. For example, there was relatively little compositional change that could be attributed to grazing in the above average rainfall seasons prior to the 1991–92 drought (i.e. all grazing treatments remained close together in ordination space), but large and sudden changes were observed after the drought, in certain high stocking rate paddocks (i.e. the high stocking rate treatment moved much further in ordination space than other treatments; Fig. 2b).

The results of the CCA were closely congruent with those of the CA, and are therefore not discussed further in detail. The Monte Carlo test showed that the environmental variables included had a significant (P < 0·01) effect on compositional change.

Seasonal peak herbaceous biomass

Depending on the amount of rainfall in a season, herbaceous biomass of low, medium and high stocking rates, respectively, ranged from 2·4 to 3·7, 2·1 to 3·5 and 2·1 to 3·2 t ha−1 at Llanwarne, and 2·5 to 3·6, 2·2 to 3·3 and 2·0 to 3·7 t ha−1 at Dordrecht. Not surprisingly, rainfall had a strong effect on seasonal peak herbaceous biomass at both Llanwarne and Dordrecht, accounting for nearly double the variance in peak herbaceous biomass than that accounted for by grazing (Table 4). The grazing days ha−1 that a camp received had a negative effect on seasonal peak herbaceous biomass at both Llanwarne and Dordrecht (Table 4). Furthermore, grazing days ha−1 reduced peak herbaceous biomass to a greater degree on poor condition rangeland (vegetation with positive scores on axes 1 and 2; Fig. 2e). At the site which started in good condition (Llanwarne), the amount of grazing days ha−1 that a paddock received reduced seasonal peak herbaceous biomass to a greater degree in vegetation that had changed largely through drought (negative coefficient of the grazing days ha−1–axis 1 interaction; Table 4), while at the site that started in poor condition (Dordrecht) seasonal peak herbaceous biomass was reduced more in vegetation that had changed largely through long-term heavy grazing (negative coefficient of the grazing days ha−1–axis 2 interaction; Table 4). Therefore the peak herbaceous biomass of vegetation that had been degraded by drought or grazing was more rapidly reduced by heavy grazing than vegetation in good condition, but may, however, have had a higher biomass when not grazed than vegetation in good condition (positive coefficients for axes 1 and 2; Table 4). The peak herbaceous biomass of the previous year had an effect on the peak herbaceous biomass of the following year at Dordrecht, which suggests a carryover effect within this perennial grassland. Rerunning the regressions with the non-significant variables eliminated did not alter the interpretation of the results.

Table 4.  Estimates of the regression coefficients of factors hypothesized to affect peak seasonal herbaceous biomass at Llanwarne and Dordrecht (see the Methods for an explanation of the variables in the model)
 Coefficientt-prob%Var
  1. Ra2 = adjusted r2.

  2. %Var = variance accounted for by each variable.

Llanwarne
Model (Ra2 = 67·3)
Constant1·711< 0·001 
Grazing days ha−1−0·00444< 0·00117·24
Axis 10·3117< 0·0010·44
Axis 20·29470·0012·67
Accumulated grazing days ha−1−0·0001780·21310·62
Accumulated grazing–slope interaction−0·0003190·0045·61
Previous year's biomass0·01210·2901·94
Rainfall0·001142< 0·00130·98
Grazing days ha−1–axis 1 interaction−0·007550·0173·33
Grazing days ha−1–axis 2 interaction0·001590·6920·03
Dordrecht
Model (Ra2 = 71·6)
Constant1·040< 0·001 
Grazing days ha−1−0·00433< 0·0018·43
Axis 10·11010·1571·73
Axis 20·26310·0025·53
Accumulated grazing days ha−10·0001870·1782·81
Accumulated grazing–slope interaction−0·0004670·0035·37
Previous year's biomass0·0583< 0·0018·61
Rainfall0·001124< 0·00137·13
Grazing days ha−1–axis 1 interaction−0·003180·2120·00
Grazing days ha−1–axis 2 interaction−0·006730·0386·76

Importantly, the accumulated grazing–slope interaction was significant at both Llanwarne and Dordrecht (Table 4), indicating that seasonal peak herbaceous biomass had declined with heavy stocking rates over time in those paddocks on steeper slopes. This site-specific effect is well illustrated by examining the trends in the coefficient (βX) of the line of stocking rate against herbaceous biomass at Llanwarne and Dordrecht (Fig. 4). The down-slope paddocks (replication 2) at Dordrecht, although variable, had no trend of increasing negative coefficient over time (t = 0·03; d.f. = 8; P = 0·98) while the up-slope paddocks of replication 1 did (t = −3·20; d.f. = 8; P = 0·01). Likewise, replication 1 at Llanwarne, which was on flat land, had no significant trend of increasing negative coefficient over time (t = −1·76; d.f. = 8; P = 0·12), while replication 2, which was on sloping land, did (t = −2·32; d.f. = 8; P = 0·05). An increasingly negative coefficient means that the high stocking rate paddock was declining in productivity relative to the low stocking rate paddock over time. It would appear, therefore, that this rangeland was more vulnerable to degradation in landscape positions that were susceptible to loss of resources by erosion. The H1 paddock at Dordrecht and the H2 paddock at Llanwarne were also the paddocks that underwent large compositional changes (Fig. 3), which demonstrates a link between compositional change and declining primary productivity. Furthermore, these changes occurred only after the drought of 1991–92, which suggests that some form of perturbation may have been needed to precipitate the observed changes. It also suggests that this semi-arid system may undergo change in a discontinuous fashion, rather than gradually.

Figure 4.

Trends over time of the slope of the line, in the relationship between stocking rate and grass biomass at Llanwarne and Dordrecht. Triangle, replicate 1 (up-slope) of Dordrecht; circle, replicate 2 (down-slope) of Dordrecht; square, replicate 1(flat land) of Llanwarne; double triangle, replicate 2 (sloping land) of Llanwarne.

Animal production

Gain per animal

Depending on a season's rainfall, gain per animal for the low, medium and high stocking rates, respectively, ranged between 113 and 225 kg, 82 and 220 kg and 102 and 217 kg at Llanwarne, and 151 and 241 kg, 103 and 225 kg and 76 and 215 kg at Dordrecht. Animal performance (gain animal−1 and gain ha−1) over the duration of the study is summarized in Table 5. In the regression analysis the Ra2 (adjusted r2) for the rainfall and herbaceous–biomass model, respectively, was 83·8% and 64·8% for Llanwarne, and 70·7% and 33·8% for Dordrecht, indicating that rainfall is a better predictor of gain per animal than herbaceous biomass. The quadratic function for rainfall was significant at both Llanwarne and Dordrecht (Fig. 5a,b and Table 6). Differentiation of the relationship between rainfall and gain per animal showed that optimum gain per animal can be expected in a rainfall season of around 620 mm (Llanwarne) to 680 mm (Dordrecht). Rainfall (sum of linear and quadratic functions) accounted for more than five times the variance in gain per animal than stocking rate (Table 6), emphasizing its dominant effect in controlling primary and secondary productivity. Heavy stocking rates tended to reduce gain per animal (negative coefficient of stocking rate; Table 6), this being most apparent at the site which had started in poor condition (significance of stocking rate at Dordrecht but not Llanwarne; Table 6) and especially so in low rainfall years (significant rainfall–stocking rate interaction at Dordrecht but not Llanwarne; Table 6 and Fig. 5a,b). The rainfall–stocking rate interaction, however, was found to be correlated with stocking rate (r = 0·5), which may have affected the overall results of the model. The removal of the interaction term from the model resulted in stocking rate being shown to be significant at Llanwarne, but the results of the other variables were not affected by its removal. Although stocking rate was shown to affect gain per animal, especially under low rainfall conditions in poor condition rangeland, accumulated grazing days ha−1 were not significant at either Llanwarne or Dordrecht, indicating there had been no long-term decline in animal performance with heavy stocking rates. This was confirmed using the logic of Wilson & Macleod (1991), where regression analysis indicated that there was no departure from the linear form of the relation between gain per animal and stocking rate at Llanwarne (t = 0·77; d.f. = 7; P = 0·465) or Dordrecht (t = −1·39; d.f. = 7; P = 0·208).

Table 5.  A summary of cattle performance data (kg) at Llanwarne and Dordrecht for the duration of the trial
LlanwarneDordrecht
Stocking rate (AU ha−1)Gain animal−1Gain ha−1Stocking rate (AU ha−1)Gain animal−1Gain ha−1Season
0·15618228·40·16417028·01986–87
0·23817340·40·20814931·31986–87
0·31315347·30·27814840·91986–87
0·15622034·30·16421835·71987–88
0·23821249·60·20822547·21987–88
0·31320663·50·27820657·11987–88
0·15622535·00·16421134·61988–89
0·23820748·40·20820543·01988–89
0·31319961·30·27819453·61988–89
0·15617226·80·16420132·91989–90
0·23816338·00·20817035·71989–90
0·31313340·90·27815944·11989–90
0·15622134·50·16421234·71990–91
0·23822051·60·20821144·21990–91
0·31321766·80·27821559·51990–91
0·15611317·60·16415124·81991–92
0·2388219·20·20810321·71991–92
0·31310231·50·2787621·01991–92
0·15617927·80·16424139·51992–93
0·23818042·00·20815532·41992–93
0·31315246·80·27816746·01992–93
0·15619830·80·16422536·71993–94
0·23820447·60·20819440·61993–94
0·31319861·10·27817949·31993–94
0·15615825·20·16422036·01994–95
0·23816739·40·20819541·11994–95
0·31313141·50·27819052·61994–95
Figure 5.

The relationship between rainfall and gain animal−1 at Llanwarne (a) and Dordrecht (b), and rainfall and gain ha−1 at Llanwarne (c) and Dordrecht (d). Circles and solid line, low stocking rate; triangles and dotted line, medium stocking rate; squares and broken line, high stocking rate. All except the low and medium stocking rates at Dordrecht had a significant quadratic relationship (P < 0·05) between rainfall and gain animal−1 and between rainfall and gain ha−1 (lines represent best fit for each stocking rate).

Table 6.  Estimates of the regression coefficients for factors hypothesized to affect gain per animal at Llanwarne and Dordrecht (see the Methods for an explanation of the variables in the model)
 Coefficientt-prob%Var
  1. Ra2 = adjusted r2.

  2. %Var = variance accounted for by each variable.

Llanwarne
Model (Ra2 = 83·8)
Constant3·182< 0·001 
Rainfall0·007615< 0·00145·36
Rainfall squared−0·62E-05< 0·00137·66
Stocking rate−0·990·3893·88
Accumulated grazing days ha−10·00004550·6300·00
Rainfall–stocking rate interaction0·000270·8890·01
Dordrecht
Model (Ra2 = 70·7)
Constant4·807< 0·001 
Rainfall0·003680·00228·85
Rainfall squared−0·37E-05< 0·00129·27
Stocking rate−5·920·00511·49
Accumulated grazing days ha−10·0002700·0561·97
Rainfall–stocking rate interaction0·006190·0534·74

Gain ha−1

Depending on a season's rainfall, gain ha−1 for the low, medium and high stocking rates, respectively, ranged between 18 and 35 kg, 19 and 51 kg and 32 and 67 kg at Llanwarne, and 25 and 40 kg, 22 and 47 kg and 21 and 60 kg at Dordrecht. In a similar vein to the gain animal−1 analysis, rainfall rather than herbaceous biomass was a better predictor of gain ha−1 in terms of the Ra2 of each model (90·1% vs. 80·0% at Llanwarne and 75·7% vs. 40·6% at Dordrecht). The quadratic term for rainfall was significant at both Llanwarne and Dordrecht (Table 7 and Fig. 5c,d). Once again, rainfall accounted for more of the variance in gain ha−1 than stocking rate (Table 7), emphasizing its importance in controlling secondary production in semi-arid regions. As with gain per animal, heavy stocking rates reduced gain ha−1 in low rainfall years to a greater degree at the site that had started in poor condition (Dordrecht) than at the site that had started in good condition (Llanwarne) (significant rainfall–stocking rate interaction at Dordrecht but not Llanwarne; Table 7 and Fig. 5c,d). Despite this effect of heavy stocking rate in low rainfall years, there was no effect of accumulated grazing days ha−1 on gain ha−1 at either Llanwarne or Dordrecht, indicating that long-term heavy grazing had not reduced cattle production per unit area.

Table 7.  Estimates of the regression coefficients for factors hypothesized to affect gain ha−1 at Llanwarne and Dordrecht (see the Methods for explanation of the variables in the model)
 Coefficientt-prob%Var
  1. Ra2 = adjusted r2.

  2. %Var = variance accounted for by each variable.

Llanwarne
Model (Ra2 = 90·1)
Constant0·6570·078 
Rainfall0·00763< 0·00124·32
Rainfall squared−0·62E-05< 0·00120·96
Stocking rate3·370·01046·71
Accumulated grazing days ha−10·00005600·5730·004
Rainfall–stocking rate interaction0·000230·9100·004
Dordrecht
Model (Ra2 = 75·7)
Constant2·294< 0·001 
Rainfall0·003630·00225·16
Rainfall squared−0·36E-05< 0·00125·42
Stocking rate−1·430·44923·90
Accumulated grazing days ha−10·0002650·0531·64
Rainfall–stocking rate interaction0·006270·0444·28

Discussion

Compositional change: rainfall versus grazing

Compositional change of the studied savanna was closely related to rainfall (Fig. 2a–d), but stocking rate nonetheless had an effect on composition over the course of the trial, as indicated by the larger distance moved in ordination space with certain high stocking rate paddocks by 1996 (Fig. 2b,c) and the larger Euclidean distances between composition in 1986 and 1996 for these paddocks (Fig. 3). That rainfall exerts a pronounced effect on compositional change in semi-arid African savannas is almost axiomatic (O’Connor 1985; Abel 1993a), especially when a fair proportion of the sward comprises short-lived perennial or annual grasses such as in this study. The drought/grazing-induced decline in abundance of densely tufted perennials such as Themeda triandra observed in this study have been witnessed elsewhere (Moyo, Sikosana & Gambiza 1995; O’Connor 1995), involving heightened mortality after a severe drought, but with elimination under long-term heavy grazing and persistence under light grazing (O’Connor 1995). In a similar vein to this study, the weakly tufted perennial Aristida congesta (Moyo, Sikosana & Gambiza 1995) and the annual Tragus berteronianus (O’Connor 1995) have been shown to colonize after a severe drought, with greater colonization by Tragus berteronianus under heavy grazing. The grazing-induced pattern of change, which involved partial replacement of long-lived perennial grasses by short-lived perennial and annual grasses, and some herbaceous dicotyledons, is a predictable response to grazing at a global level (Milchunas & Lauenroth 1993).

Compositional change had an apparent Clementsian nature, where later successional species increased relative to other groups under high rainfall and low grazing, while early successional species increased relatively under heavy grazing (Fig. 2e,f). The discontinuous nature of this change, however, does not support traditional Clementsian theories of gradual transitions from one sere to another, but supports rather the state-and-transition model (Westoby, Walker & Noy-Meir 1989), where transitions between states are induced by perturbations such as drought and grazing or a combination thereof. While the abundance of individual species in this study varied markedly in response to rainfall variability, heavy stocking in combination with drought and landscape position resulted in discontinuous changes, although in a generally consistent direction to effect ultimately a conspicuous shift in sward composition in certain paddocks. Compositional changes were observed as a result of drought alone (Fig. 2a,d) and drought combined with heavy grazing (Fig. 2b,c). The fact that grazing-induced change was observed only after the drought of 1992 suggests that drought is required to induce a compositional shift, as suggested by Livingstone (1991). The fact that little compositional change was evident in the heavy stocking rate paddocks on flat or bottomland (depositional) sites, but that pronounced change was apparent on paddocks vulnerable to soil erosion, suggests that savanna grasslands in certain landscape positions are resistant to grazing-induced change, and that erosion processes may be implicitly associated with compositional change. Circumstantial evidence of soil loss from the heavy stocking rate paddocks on sloping land was evident in the form of partially buried fences, but this issue requires formal investigation.

It has been suggested that semi-arid grasslands with a long evolutionary history of grazing (such as this one) show greater resistance to compositional change due to grazing than grasslands with a short evolutionary history of grazing (Milchunas, Sala & Lauenroth 1988), but the trends revealed by this study contradict this model. All species recorded in this study can be considered well adapted to defoliation and the rainfall variability characteristic of semi-arid environments, yet certain long-lived perennial grasses were depleted by sustained heavy grazing in combination with drought. This study suggests that the Milchunas, Sala & Lauenroth (1988) model needs to be elaborated to include rainfall variability (drought frequency and intensity), grazing intensity, and temporal patterns of defoliation in relation to plant phenophase, if it is to achieve relevance for African grasslands (Landsberg, O’Connor & Freudenberger 1999). Greater grazing-induced compositional change at sites that started in good condition compared with sites that started in poor condition (Fig. 2b vs. Fig. 2c and Table 2) suggests that this rangeland attains a new resilient state with long-term heavy grazing, beyond which it does not easily move, in support of Walker (1980). The apparent resilience conferred on this savanna has been achieved by a replacement of long-lived perennial grasses by short-lived perennial and annual grasses that are better able to respond to constantly changing environmental conditions.

Primary production: role of rainfall, grazing and composition

While the expected effect of rainfall on above-ground primary production of herbaceous plants in a semi-arid savanna was obvious, heavy stocking ultimately had a deleterious influence on primary production in paddocks vulnerable to erosion (Table 4 and Fig. 4), in agreement with global patterns (Milchunas & Lauenroth 1993). Paddocks that experienced a decline in primary productivity with heavy stocking rates (replication 2 at Llanwarne and replication 1 at Dordrecht; Fig. 4) also exhibited the most pronounced change in species composition (Fig. 3), such that a clear relationship between primary production and composition was established. These findings suggest that Abel's (1993b) rejection of the ‘species paradigm’ for semi-arid African savannas is premature; the effect of degraded composition on primary production in this semi-arid savanna was negative, as shown for elsewhere on the globe over a range of climatic conditions (Milchunas & Lauenroth 1993).

Declines in primary production with heavy stocking were, however, only evident in paddocks of greater slope (replication 2 at Llanwarne and replication 1 at Dordrecht; Fig. 4), in accordance with the observations on compositional change. This suggests that substantial loss of moisture and nutrients through increased run-off and soil erosion may have occurred, but neither were studied. It has been suggested that degradation would occur primarily at times of poor (< 30%) vegetation cover, as found at the beginning of a growing season on heavily grazed paddocks, in particular following droughts (Livingstone 1991; Illius & O’Connor 1999). The results of this study support this suggestion; namely, heavy stocking rates induced little change in composition or productivity in good rainfall seasons before the 1991–92 drought, but resulted in large changes in both composition and productivity after the drought in certain paddocks. The fact that in 1996 the heavy stocking rate paddock in the bottomland position at Dordrecht (replication 2) exhibited the highest production of all the paddocks for all years over the duration of the trial, and maintained livestock production at a time when adjacent paddocks could not, shows that certain landscape positions may be well buffered against the effects of heavy grazing. This indicates that productivity of key resource areas is being maintained by erosion of their surrounds (Scoones 1992), but begs the question of whether the lack of degradation on bottomlands has been sufficient to off-set degradation of uplands. This cannot be answered by this study.

Animal production

Despite observations of compositional changes, and declines in grass production in some of the high stocking rate paddocks, no decline in cattle production at high stocking rates was detected. This suggests that the spatial scale of management may play an important role in determining the sustainability of secondary productivity. The rotation of cattle between the two replications of high stocking rate paddocks, according to the availability of forage, may have allowed a non-degraded paddock to offset any forage shortages in degraded paddocks during a dry period. The link between composition, primary production and animal performance should be examined in continuously grazed systems. However, the unavailability of forage on all paddocks in the dry season of 1992, and notably only the medium and heavily stocked paddocks at the poor condition site (Dordrecht) in the dry season of 1993, such that sugarcane tops had to be provided in order to avoid livestock mortality, is illustrative of a tight herbivore–resource linkage during dry periods (Illius & O’Connor 1999) that is weakened during wetter episodes. The interaction between stocking rate and rangeland condition that was observed at two locations in Australia (Ash et al. 1995) was not observed in this study. In the Australian example, animal performance was better on poor condition rangeland than good condition rangeland at low stocking rates, but with the converse being true at high stocking rates. In this study, animal performance at high stocking rates was similar at both Llanwarne and Dordrecht, probably because the H2 paddock at Dordrecht was a particularly productive paddock.

The decline in cattle performance during the highest rainfall seasons observed in this study has also been reported by ’t Mannetjie (1982), who observed a strongly quadratic relationship between gain per animal and rainfall from a Cenchrus ciliarus–Macroptilium atropurpureum pasture grazed by steers. Increased grass growth that changes sward structure (decreased leaf–stem ratios) has been found to reduce the intake of forage by cattle (Stobbs 1973, 1975) and result in dilution of plant nitrogen with carbon (Jarrell & Beverly 1981). Thus during the very high rainfall years, exceptional grass growth and resultant changes in sward structure and leaf nitrogen are expected to have a negative impact on gain per animal.

Sustainability of semi-arid grazing systems

Ellis & Swift (1988) proposed that livestock–plant systems with an interannual CV of rainfall greater than 33% were non-equilibrial in the sense that livestock would not experience density-dependent effects because drought-induced mortality would intervene before animal numbers had built up to a sufficient level; plants and animals are thus decoupled. This argument has been extended to semi-arid African savannas supporting livestock production (Behnke & Scoones 1993). The results of this study, however, suggest that density-dependent effects (consumer-resource coupling) were present but that it was mostly expressed on erodible landscapes during and following drought. Grazing clearly had an effect on plant composition and production on erodible slopes, where supplemental feeding was necessary during drought years on heavily stocked treatments and especially on poor condition rangeland. In addition, animal gain ha−1 was negatively affected by stocking rate on poor condition rangeland during drought years. These findings suggest a possible pattern of degradation in this savanna: it is dependent on drought episodes, with mainly pronounced compositional change in certain landscapes occurring during the first drought episode coincident with heavy stocking, and the effect on animal production becoming manifest only with subsequent drought episodes. The contradiction offered by this study to non-equilibrium notions of semi-arid grazing systems is tempered by theoretical effects which demonstrate that degradation apparent in field trials with fixed stocking rates, such as this one, may be absent from real systems in which animal numbers always vary (Van de Koppel, Rietkerk & Weissing 1997). In practice, it is unlikely that any system achieves a fixed stocking rate, but plant–mammalian herbivore systems vary from those with no management (e.g. some wildlife systems), through limited management (e.g. Turkana; Ellis & Swift 1988), moderate management (e.g. crop residues, buying and selling; Scoones 1993), to the carefully regulated herd sizes of commercial agriculture. The potential for degradation revealed by fixed stocking rate experiments needs to be examined for a range of similar systems in which animal numbers vary in response to management or extrinsic factors such as climate. Insights from this study are further tempered by recognition of the restricted spatial scale of the study. If more flexible movement had been possible over a greater area, impacts of grazing may well have been better buffered, as is suggested by the partial compensation of production by some paddocks in this study.

In conclusion, it has been proposed that systems with highly variable rainfall patterns are non-equilibrial, namely consumer–resource dynamics are uncoupled (Ellis & Swift 1988). Despite rainfall having an interannual coefficient of variation of 30% over the duration of this trial, there was evidence of grazing-induced changes in herbaceous composition and productivity brought about during the 1991–92 drought when the ratio of herbaceous biomass to herbivore biomass was at its lowest. Such situations are likely to result in low plant cover and elevated soil erosion (Livingstone 1991), such that changes in composition and productivity (degradation) are more likely on erodible sites. In terms of animal performance (the criterion for degradation), there was no trend of decreasing animal performance over time, yet stocking rate determined the requirement of supplementary feeding and influenced gain ha−1 on poor condition rangeland during and following drought. Considering that the heavy stocking rate applied in this trial is far less than that practised in adjacent communal livestock systems, it can be expected that grazing may exert a significant impact in those savannas where livestock numbers are allowed to vary.

Acknowledgements

John Turner established the trials and Grant Hatch continued them; we thank both for making their data available. We thank the Agricultural Research Council for funding this study; the owner Russell Anderson and the general manager Jannie Bender of Llanwarne estates, for allowing the establishment of the trials and providing accommodation; Andrew Illius for comment on an initial draft; and Craig Morris for his critical input.

Received 24 December 1998; revision received 2 February 2000

Ancillary