Ordination and classification of vegetation of Songimvelo Game Reserve in the Barberton Mountainland , South Africa for the assess ment of wildlife habitat distribution and quality

A vegetation survey was undertaken of the 49 000 ha Songimvelo Game Reserve in the Barberton Mountainland of Mpumalanga. South Africa with the aim to identify constituent plant communities and to assess their relative value to wild herbivores. The vegetation is highly diverse with representation of three biomes: Savanna. Grassland and Forest A total of 428 plots were sampled by means of a semi-quantitative technique. Data were subjected to ordination (CANOCO) and clas­ sification (PATN). The composition of the 19 distinct communities is determined through an intricate combination of environmental factors as evident from the ordination results. Firstly drainage line' position is critical, followed by land use history and further by the interplay between elevation and geology. These findings are in line with results obtained from other studies along the eastern Escarpment. Alluvium, mafic and ultramafic lavas support mixed veld, whereas felsic lavas, sandstones and quartzites support sour veld which has a very low forage value in the dry’ season. Each community, through its specific species assemblage, structure and location, forms a distinctly different habitat in terms of its value to the vari­ ous species of herbivores in the SGR


INTRODUCTION
The 49 (XX) ha Songimvelo Game Reserve (SGR) is located within an area of great conservation and biogeo graphic value (Fourie et al. 1988;Matthews et al. 1993), of internationally renowned geological interest (Lowe & Byerly 1999) and with aesthetically striking landscape attributes (Anon. 1986).The vegetation is highly diverse with representation of three biomes; Savanna.Grassland and Forest.
These conservation and scientific values are of limit ed consequence to the impoverished rural communities living alongside the Reserve.Tangible economic benefits through tourism are essential to obtain local community acceptance and support for the use of the land for con servation.Successful ecotourism development in this area depends largely on the introduction and mainte nance of a large wild herbivore component (Anon. 1998).
A description of the vegetation and an understanding of the underlying causal factors are required in order to assess habitat suitability for wild herbivores and in order to formulate appropriate management guidelines.Prior to the present study, no comprehensive vegetation survey had been undertaken of this area and consequently little information was available on the vegetation-herbivore interrelationships.
Vegetation composition has direct bearing on the quality and seasonality of available feed (Barnes et al. 1984;Eckhardt et al. 1993;Fabricius & Mentis 1990).
Vegetation structure (height and density) largely controls its availability to herbivores (Fabricius & Mentis 1992).
The aims of this study were firstly to classify the veg etation of the SGR into identifiable plant communities based on composition and structure, secondly to identify the main environmental factors responsible for this veg etation pattern, and lastly to assess the relative value of the identified plant communities to the wild herbivores.

STUDY AREA
The SGR is located in the southeastern part of Mpu malanga on the South African-Swaziland border at lati tude 25° 45'-26c 5' S and longitude 30c 46-31° 16' E (Figure 1).

Geology, soils, topography and drainage
The SGR is situated in the Barberton Mountainland which forms part of the African erosion surface (Partridge & Maud 1987).Elevation ranges from 6(X) to 1 9(X) m above sea level.The SGR is drained by numer ous perennial rivers and streams of which the Komati is the most important.The Komati Valley represents some 10 (XX) ha of relatively gentle topography, whereas the remainder of the Reserve is more rugged.
The Barberton Mountainland represents an early Precambrian greenstone belt (Viljoen & Viljoen 1971).The entire succession of supracrustal rocks constituting the greenstone belt is known as the Swaziland Sequence and has been divided into three groups.The Onverwacht Group represents the initial volcanic phase of the belt.For purposes of the study, this group was divided in two units.The Tjakastad unit combines the Komati and Theespruit formations consisting of basaltic and peridotitic komatiite and tholeite with various mafic and ultramafic schists.Alluvial deposits in the lower Komati Valley were lumped with this group.The Geluk unit combines the Zwartkoppie, Kromberg and Hooggenoeg formations and includes mostly mafic and felsic volcanic rocks, agglom erates, breccia, chert and shales.The Onverwacht Group is overlain by an argillaceous group (Figtree) and an arena ceous group (Moodies).These include sandstones, quartzites, shales, agglomerate and conglomerate.Acid igneous intrusives occur in the form of biotite trondhjemite gneisses in the southwestern corner and granodiorite-adamellite along the southern edge of the SGR.
Geology and elevation are not independent of each other.The different groups stretch in a wedge shape, from the broad low-lying Komati Valley in the southwest to the narrow band of mountains in the northeast.Ultramafic and mafic substrates generally occur at lower elevation, whereas felsic lavas, quartzites and sandstones are found at higher elevations.Average elevation increases from 919 m for the Tjakastad unit, 1 098 m for Geluk to 1 219 m for Figtree and Moodies.A corresponding increase in average annual rainfall contributes to more leached and acid soils derived from the Geluk and Figtree units with lowest pH values of 4.4 as compared to the highest pH value of 7.1 in the lower Komati Valley (Anon. 1986).

Climate
Rainfall, which is concentrated between November and March, varies from less than 800 mm per year in the lowlying southwestern area to over 1 600 mm in the highlying northeastern parts (Gamble 1988).Mean minimum and maximum monthly temperatures are 5.4°C and 7.9°C in July and 22°C and 34°C in January for the highland and lowland areas respectively (Anon. 1986).Frost is common during winter months in the Komati Valley.

Archaeology and land use history
The SGR is characterised by a long and diverse histo ry of land use.Witt (1983) reports middle Stone Age artefacts dating back 30 000 to 50 000 years.Stone walled sites of the Later Iron Age (the last 1 000 years) are common (Anon. 1986).The SGR area was settled by the bakaNgwane (people of Swaziland) during the reign of King Mswati II (1840-1868) (Van der Merwe & Retief 1995).During the last 100 years gold and asbestos mining took place.The area was used for winter grazing of sheep from the turn of the century.In 1985, some 350 households were living within the future SGR.The total area under dryland cultivation was ± 850 ha.Since then the number of residents has declined with only 40 fami lies remaining in 1997.By 1985, only limited numbers of small game still occurred.Since 1986 a total of 20 species of large herbivores totalling more than 2 000 ani mals have been re-introduced.

Vegetation
The vegetation of the higher-lying regions of the SGR belongs to the Grassland Biome.The lower-lying Komati Valley falls within the Savanna Biome (Rutherford & Westfall 1986).Three of Acocks's (1975) veld types occur.Veld types 8 (North-eastern Mountain Sourveld) and 63 (Piet Retief Sourveld) correspond to the North eastern Mountain Grassland of the Grassland Biome as defined by Low & Rebelo (1996).Veld type 9 (Lowveld Sour Bushveld) corresponds to the Sour Lowveld Bushveld of the Savanna Biome.The Forest Biome is represented by numerous isolated patches of forest, mostly at higher elevation and along drainage lines.

Sampling approach
Edwards' (1983) structural classes were used to describe the overall structural properties of the sampled plots.Overall cover was estimated for the woody, grass, forb and geophyte component respectively, using the semi-quantitative measures of the Braun-Blanquet approach (Mueller-Dombois & Ellenberg 1974).Cover and height classes were recorded for individual woody and grass species.Individual geophytes and forbs were omitted from the list of species for three reasons.Firstly, grasses and woody species are of most importance to the herbivores.Secondly, the great diversity of forbs would have significantly increased sampling time and would therefore have decreased the possible number of sample plots.Thirdly, because of logistical constraints, sampling was spread across seasons which would have resulted in a differential presence of geophytes depending on the sampling date.Records of environmental data included elevation, by means of an altimeter (± 20 m) and the 1:50 000 topocadastral maps; geology, according to 1:250 000 geological survey maps (Geological Survey 1986); and locally at a finer scale through personal obser vations; landscape position (Land Type Survey Staff 1989), aspect (whether predominantly N, E, S or W), slope steepness (class estimate), soil texture (using the sausage method (National Working Group for Vegetation Ecology 1986)) and rockiness (estimated as a percentage of the ground cover).
It was deemed important to achieve a high number and sufficient spread of sampling plots for two reasons.Firstly, Austin & Heyligers (1989) argue that where there is no existing information on vegetation (as was largely the case in this instance), surveys should sample various combinations of environmental variables as a means of obtaining a representative sample.Secondly, they contend that sampling the full range of environments ensures that predictive models derived from survey data can be used for interpolation rather than extrapolation.Furthermore, in resource surveys in which a major objective is the detection of as much diversity as possible, randomisation is largely irrelevant (Gillison & Brewer 1985).In this instance the authors were not concerned with a statistical estimation of the proportions of the survey area covered by different vegetation communities.Provided there has been sufficient ground coverage to ensure correct inter pretation, these areas can be measured, for example, from aerial photographs or satellite imagery with sufficient accuracy.
Different combinations of environmental variables and different vegetation communities were sampled in 428 plots of 30 x 30 m which were subjectively located.Although the whole Reserve was covered, samples tend to be clustered as dictated by terrain accessibility (Figure 2).The question with regard to this approach is whether it adequately cap tures the floristic diversity and the large range of abiotic conditions which might determine community composition.This was investigated in two ways.Firstly, the coverage by sample plots of different combinations of environmental factors was determined.The combinations of environmen tal factors such as elevation and geology were obtained through the use of the IDRISI Geographic Information System (Eastman 1992).The geographic position of the samples was determined in the field by means of a Global Positioning System (GPS) and incorporated into the GIS.Secondly, the adequacy in capturing the floristic diversity was assessed by direct comparison of the sample data with the known woody and grass diversity of the SGR.

Ordination
The CANOCO computer package (Ter Braak 1992) was selected to analyse relationships between the data set of 428 plots by 346 species and the underlying environ mental factors.CANOCO allows for canonical ordina tion which is an intermediate technique which combines aspects of regular ordination with aspects of regression (Jongman et al. 1987).The resulting ordination diagram expresses not only the pattern of variation in species composition but also the main features of species distri butions along the gradient of environmental variables (Ter Braak 1986).A step-wise approach was followed in which groups of plots, determined by specific environ mental conditions identified in the previous ordination run, were removed from the remaining data set for the next ordination.Both Principal Components Analysis (and its canonical equivalent Redundancy Analysishenceforth RDA) and Correspondence Analysis (and its canonical equivalent Canonical Correspondence Analysis-henceforth CCA) were used, based on the type of response model exhibited by the local species.As a practical guideline.Ter Braak & Prentice (1988) sug gested that most species are behaving monotonically if the gradient length is less than 1.5 sd [standard deviation or unit of ordination-length as defined by Hill & Gauch (1980)1.Gradients longer than 3 sd can be analysed using CA and related techniques.The range 1.5-3 sd for the first axis represents a 'window' over which both PCA/RDA and CA/DCA/CCA can be used to good effect.In practice, the data set was subjected to a DEC-ORANA run (Hill 1979a) which led to the identification of a gradient length of 5.347 sd along the first axis.

Classification
Classification or cluster analysis was used to identify groups and to impose structure on the raw data (Jongman et al. 1987).The PATN software package (Belbin 1988) was used.PATN incorporates a host of classification techniques including TWINSPAN (Hill 1979b).In par ticular, the ALOC module was used.
ALOC stands for 'allocation' and implements a sim ple non-hierarchical clustering strategy.In a first phase the first sample plot in the data set is used as a 'seed' to create a starting configuration that ALOC can iterate on.Each object in the data set is compared to each seed.If the resulting association is greater than an allocation radius between zero and one, the object itself becomes a seed and the number of groups increases with one.The process continues until all objects have been compared with all seeds.The second phase begins with object 1 and sequentially allocates each object to the nearest seed as generated in the first phase.The third phase dismisses the seeds and calculates the group centroids based on the objects which were assigned to each group in phase 2. The fourth phase is iteration and re-location.Each itera tion sequentially extracts each object from the group it currently belongs to and allocates it to the group with the nearest centroid.Groups may disintegrate down to a sin gle object, by the process of re-allocating them to other groups.
The PATN outcome was evaluated subjectively by testing the obtained entities against photographs of each sample plot and field knowledge.The criteria used were the homogeneity of the units and their identifiability in the field.

Landscape diversity
Based on initial field observations and a study in a similar environment (Deall & Theron 1990), elevation and geology were identified as being of major impor tance.Using six elevation classes, starting at 600 m asl and each spanning 200 m, and the 12 geological sub strates, a total of 51 elevation-geology combinations are present in the SGR.Marked differences could be observed in the field across a 200 m elevation difference, in particular when comparing different aspects.Thirteen elevation-geology combinations were not sampled.
However, these 13 combinations cover only 1.9% of the SGR with the largest combination standing at only 0.4% of the total surface area.Another seven combinations representing 1.2% of the surface area, only have one sample plot each.Thus, only 3.1% of the surface area in terms of elevation-geology combinations was not sam pled at all or was sampled without replication.Sixteen combinations covering 75.4% of the Reserve are repre sented by more than 10 sample plots per combination.The 428 plots are thus considered to be adequately spread across the SGR.

Floristic diversity
A total of 348 species was recorded for the 428 plots; 247 woody and 101 grass species.A total of 288 woody species is known for the SGR based on independent sur veys and ongoing collecting by several taxonomists.The sampled species thus represent 85.8% of this total.Species which were not encountered include inter alia Calodendrum capense, Encephalartos paucidentatus, Kiggelaria africana, Nuxia floribunda, Piper capense, Rhoicissus digitata, Rhus discolor, R. gerrardii, Scutia niyrtina, Syzygium guinense and Warburgia salutaris.They are mostly found in forests which cover less than 10% of the Reserve and where they do not constitute a dominant component.The known number of grasses based on ongoing collecting and previous research pro jects is 136 species (after lumping both in the total list and in the sample plots the subspecies of Aristida congesta, Bothriochloa spp., Cymbopogon spp., Festuca spp.and the different varieties of Setaria sphacelata as a single species each).It was necessary to lump some of these species and subspecies as sampling conditions cov ered a wide range of conditions in terms of the length and age of the sward making identification problematic in instances where heavy grazing was experienced or where a fire had recently occurred.The sampled species repre sent 74.3% of the known total.Species which were not encountered during the sampling are mostly Eragrostis spp., Sporobolus spp.and Digitaria spp., none of which were important in the field.

Ordination results
The first CCA ordination led to a dense cloud of plots representing the major part of the sample, with plots along drainage lines falling outside of this cloud.Eigenvalues of the first and second axes are respectively 0.57 and 0.56.Plots in 'drainage' positions are covered by forests and thickets, as well as more open wetland or riverine vegetation (Figure 3).Drainage positions have a higher moisture availability and are sheltered from fires.Thickets, which are more prominent at low elevation and on northerly aspects are characterised by species such as Grewia occidentalis, Olea europaea subsp.africana and Ruttya ovata.In contrast, forests are much more inesic, occurring at higher elevation and on more southerly aspects, and are characterised by species such as Combretum kraussii, Hallena lucida, Keetia gueinzii and Schefflera umbellifera.The wetland and forest commu nities which these plots represent are clearly defined The 60 forest/thicket/riverine/wetland plots identified through the first CCA were removed from the original data set.The remaining 368 plots were again subjected to CCA ordination.Eigenvalues of the first and second axes are 0.71 and 0.41 respectively.A dense cloud of 'undis turbed' samples spread along an elevation gradient was plotted against a group of samples on old lands and old settlement sites (Figure 4).Within the ordination diagram a clear spatial definition is obvious between the plant communities on old lands at higher elevation (Lippia javanica-Hyparrhenia spp.Community) as compared to the lower elevation old lands which are covered by the Cynodon dactylon-Melinis repens Community.

Syzygium
The 65 old land and settlement sites identified in the second CCA were subjected to a separate RDA ordina tion.Resulting eigenvalues were very low at 0.07 and 0.05 for the first two axes respectively.Old lands and old settlements are not so much different in terms of compo sition as they are structurally distinct.Both sets of plots share species such as Acacia nilotica, Cynodon dactylon and Heteropogon contortus.A. nilotica occurs as scat tered low shrubs on old lands but as dense mature thick ets on settlement sites.The greater woody cover on old settlement sites probably reflects a longer time span from abandonment and/or the result of better fire protection afforded by old walls.Other typical species are Hippobromus pauciflorus and Pappea capensis.
After removal of the 65 old land and settlement sites, the remaining 303 plots were again subjected to a CCA ordination.The eigenvalues were relatively high; 0.58 for the first axis and 0.36 for the second axis.Both ele vation and geological substratum are clearly important factors but their respective influence could not immedi ately be ascertained.
Their relative importance was evaluated by ordinating a subset formed by all the plots falling between 900-1 100 m and 1 500-1 700 m in the Geluk and Figtree/Moodies units.The choice of these two elevation belts and substrata was made subjectively in order to obtain enough samples (respectively 33, 19, 13 and 14 plots) while having a large enough difference in eleva tion (400 m minimum) to ensure the likelihood of elevation-induced vegetation differences.The resulting diagram clearly shows a split between low and high elevation plots within each geological unit and a split between geological units within each elevation belt (Figure 5).The CCA ordination yielded eigenvalues of 0.47 and 0.26 for the first two axes.High t-values and inter-set correlations of -0.84 and 0.63 for elevation on the first and geology on the second axis indicate the impor tance of these two factors.A total of 74.7% of these sam ple plots are classified into three communities which are spatially distinctly defined in the ordination diagram.

C E N T R O ID /B IP LO T OF EN VIR O N M E N TA L VARIABLE
Based on the above, the subsets formed by the three major geological units were ordinated separately.The 45 plots occurring on alluvium and the Tjakastad subgroup were ordinated using CCA.The eigenvalues for the first axis was 0.45 and for the second 0.27.They generally represent the lowest-lying area of the Reserve as well as the least broken part.Elevation is an important factor, but landscape position is also critical.The plains and footslopes are characterised by grass species such as Aristida congesta, Eragrostis chloromelas and Heteropogon contortus and woody species such as Acacia caffra, A. nilotica and Combretum inereroense.Upper slopes are typi cally covered by Combretum apiculatum while with ele vation an increase in the more sour grass species such as Loudetia simplex, Panicum natalense, Trachypogon spicatus and Tristachya leucothrix is apparent.
The CCA ordination diagram for the 153 plots on the Geluk subgroup again yielded elevation as being the most important environmental factor.The eigenvalue for the first axis was 0.49 and 0.21 for the second.At high elevation, Erica drakensbergensis, Koeleria capensis and Protea roupelliae are found.Warmer, northerly aspects are often characterised by Combretum apiculatum, Gymnosporia buxifolia and Pterocarpus angolensis.On cool er and moister southerly aspects sour species such as Alloteropsis semialata subsp.eckloniana are prominent.Aristida canescens/transvaalensis and Xerophyta spp.are associated with steep westerly and northerly slopes.
Elevation was also the most important factor accord ing to the CCA ordination diagram for the 81 plots on Figtree and Moodies substrata.Eigenvalues were 0.34 and 0.24 for the first two axes respectively.Alloteropsis semialata subsp.eckloniana, Koeleria capensis, Protea caffra and Rendlia altera are representative of a large proportion of the vegetation.

Classification results
The ALOC classification results are presented by means of a dendrogram.The y-axis represents decreasing association (or increasing 'distance') between groups of sample plots.A total of 24 classification groups were produced (Figure 6).
The three main splits represent respectively, mixed, 'forest/thicket' and sour communities.Sour refers to vege tation of which the forage quality declines sharply towards the dry and cold winter making it less acceptable to herbi vores.Mixed communities are intermediate between sour and sweet communities.The latter retain their forage qual ity during the dry season (Ellery et al. 1995).
Based on the frequency distribution of the cover class es, each main split exhibits a particular combination of woody, grass and forb cover (Figure 7).Mixed commu nities are generally more wooded and have a less dense grass layer.Forests and thickets have per definition a closed to virtually closed woody layer and a low grass cover.The sour communities are less wooded and have a very dense grass layer.These three main splits reflect the major physiognomic characteristics of the three biomes that occur in the Reserve, namely Savanna, Forest and Grassland (Rutherford & Westfall 1986).
Grass species were classified in terms of their p a y a bility, particularly late into the growth season and into the dry season (Ellery et al. 1995;Van Oudtshoom 1991).Based on the frequency distribution across differ ent cover classes, a cover value was approximated for each species in the mixed and sour communities.The unpalatable or sour species make up more than 60% of total cover in the sour split, whereas the palatable sweet grasses contribute less than 10% (Figure 8).This neatly fits the definition for sour grasslands derived by Ellery et al. (1995).The mixed split similarly fits the definition by being intermediate between the sour and sweet grassland with the latter having more than 20% sweet grasses and a cover of less than 30% by sour grasses.
These sour communities occur at elevations above 1 000 m where the high rainfall results in leached soils on mostly Geluk, Figtree and granitic substrates which are dystrophic (Anon. 1986).Within the sour split, groups 5, 17, 19 and 24 represent the typical sour grass lands.Groups 7 and 14 represent mixed communities with a higher cover of woody species.
Below this first split, the next branch defines the mixed communities and the forests and thickets.Clas sification groups 6, 18, 22, 9, 11 and 21 make up the forests and thickets.The mixed communities occur gen erally at lower elevation and are more wooded and are generally situated on nutrient rich alluvium and Tjakastad geology.
Within the mixed split, the riverine and wetland com munities (groups 4, 16, 8, 13, 20 and 10) are isolated.The remainder of the mixed split comprises the drier commu nities with Acacia nilotica as a visually striking element (groups 1, 2, 23, 3 and 15).The further division (group 1) comprises old lands.

Description of plant communities
Of the 24 classification groups, it is difficult to differ entiate in the field between groups 2 and 23, groups 4 and 16, groups 17, 19 and 24, as well as groups 7 and 14.After lumping these groups, 19 units or communities remain.The community concept is applied in its broad sense and reflects a recurring assemblage of grass and woody species of characteristic composition and struc-  Gabriel & Talbot (1984)).These communities occur both on a micro-(1 m2 to 106 m2) and meso-scale (106 m2 to 10,() m2) (Delcourt & Delcourt 1988).Community names were chosen subjectively so as to have practical value in the field through the use of two species which are visual ly and/or diagnostically important.Structural informa tion with regard to vegetation height and openness fol lows Edwards (1983).A synoptic table with constancy values for all species in the 19 plant communities is pro vided (Appendix).

Cynodon dactylon-Melinis repens Short Shrubland (classification group 1, n = 35 sample plots)
This community is mostly found in the lower-lying Komati Valley below 1 000 m.It occupies footslopes and river terraces which have been cultivated or settled in the past.Soils are clayey.More than half of the sample plots had a woody cover of less than 5%.The most frequently occurring shrub is Acacia nilotica (Table 1).A further 31 woody species were recorded, typically including Dichrostachys cinerea, Euclea crispa, Rhus pentheri, Sclerocarya birrea, and Ziziphus mucronata.The grass component is generally less than 0.5 m tall with a charac teristic combination of two or more of the following species: Bothriochloa spp., Cynodon dactylon and Heteropogon contortus (Table 1).In total 41 species of grass es was recorded.More than half of the sample plots were subject to moderate or heavy grazing pressure.The forb component as observed in the field comprises alien specie^ such as Conyza spp., Acanthospermum australe and Schkuhria pinnata.Community) are embedded.This community itself has been exposed for a long time to wood extraction and heavy grazing by livestock, but its soil surface has not been directly modified.This is probably because of the often extremely rocky character of the habitat.Only 5 out of 82 sample plots occupy sites of old lands or settlement.

Acacia nilotica-Heteropogon contortus
The woody component is generally open to closed.A total of 74 woody species was recorded with Acacia nilotica as the most frequently occurring tree in associa tion with Euclea crispa shrubs (Table 1).The dominant species in the grass layer is Heteropogon contortus.A total of 55 grass species was encountered.

Acacia nilotica-Euclea crispa Low Woodland (classi fication group 3, n = 19)
This community has many species in common with the previous one, but is generally a closed formation occurring as discrete patches on mostly old settlement sites, but also on dolerite dykes.These sites are mostly situated in the Komati Valley.Acacia nilotica is the dom inant woody element, often forming a closed, even-aged canopy.Of particular importance is the occurrence of valuable browse species such as Berchemia zeyheri.In total, 61 woody species was observed.Only 29 grass species were recorded of which Heteropogon contortus was dominant.Cynodon is also important.The forb com ponent as observed in the field harbours the alien Zinnia peruviana.Omithogalum saundersiae is a conspicuous geophyte which is found along old settlement walls of packed rock.

Combretum apiculatum-Xerophyta retinervis Low
Woodland (classification group 15, n = 3) This community is confined to moderately steep to very steep rocky upper slopes overlooking the Komati River on an ultramafic substrate.It is physiognomically very distinct with the tall fibrous perennial Xerophyta retinervis under a canopy of Combretum apiculatum.The     This community is physiognomically dominated by tall 'thatch' grass species such as Cymbopogon spp.. Hyparrhenia spp.and Hyperthelia dissoluta (Table 1).Other common tall grass species are Bothriochloa spp..

Eragrostis curvula, Sporobolus africana and S. pyramidalis. Total grass diversity is 25 species. It is essentially the counterpart of the Cy nodon dacty lon-Melinis repens
Community on old lands but occurs at elevations above 1 000 m as small patches of old lands or in small saddles in the landscape.The old lands are not much grazed, but the saddle positions are much favoured by cattle for graz ing and resting.This is probably because of the gentler topography, sheltered microclimate and increased nutri ent cycling established through the more concentrated activities of animals on these particular sites (Blackmore et al. 1990).Only eight woody species were recorded of which only Lippia javanica is fairly ubiquitous.

Acacia mearnsii-Eragrostis curvula Short Woodland (classification group 20. n = 1)
This alien-dominated community is found north of the Msoli River along some drainage lines and in areas of previous disturbance such as mining, roadsides and set tlement sites.The most conspicuous feature is the domi nance of Acacia meanisii (black wattle) coupled with a low diversity of other species.Grass cover is sparse with Eragrostis curvula being the most prominent species.Only seven grass species were recorded.9. Solanum aculeastrum-Eragrostis curvula High Shrub land (classification group 10, n = 1) This community was only observed in a few' small patches.It occurs as a closed shrubland of Solanum aculeastrum on disturbed sites at elevations above 1 400 m in moist areas in the north and east of the SGR.It is a species-poor community containing certain forest ele ments such as Celtis africana.Halleria lucida and the grass Oplismenus sp.under its cover and protection from fire.Only four grass species were recorded on this sam ple plot.10.Dalbergia annata-Keetia gueinzii Tall Forest (classi fication group 6. n = 9) This community is found towards the eastern side of the SGR at relatively low elevations of 800 to 1 200 m along drainage lines.It is particularly prominent along the lower reaches of the Msoli River and through the cen tral valleys of the northeastern panhandle extension of the Reserve.It is a closed dry forest with an important thorny and spiny element (Acacia ataxacantha and D. armata) (Table 1).A total of 91 different woody species was recorded which is more than in any other communi ty.The grass component of nine species is more diverse than any of the other forest communities, probably because the lower canopy and frequent gaps allow more light to reach the forest floor.The most common species however is Oplismenus hirtellus which is a typical forest-floor species (Table 1 This community occurs at low elevation in the Komati and Msoli Valleys where it is found on steep, mostly north-facing slopes overlooking the river.Due to its low elevation, steep slope with shallow soils and northern aspect, this is probably one of the most xeric habitats in the SGR.This includes some extremely broken dolerite ridges.Soils are mostly sandy loam with 15% to 20% clay.The woody component is relatively diverse with 64 species recorded.Ptaeroxylon obliquum and Ruttya ovata are the most frequently occurring woody species.Common species also found in closed communities at low elevation are Olea europea subsp.africana.Pappea capensis and Zunthoxylum capense.This is the only community in which Spirostachys africana was encoun tered.Croton gratissimus was observed on dolerite ridges.Panicum maximum was recorded in each of the six sample plots belonging to this community (Table 1).Only 14 other grass species were recorded of which Heteropogon contortus is the most prominent.Forb cover is generally below 5%.A succulent component mostly consisting of Aloe spp., Euphorbia ingens and E. evansii (only recorded in this community) was encoun tered in every plot.

Acacia caffra-Dombeya rotundifolia Short Thicket
Community (classification group 11, n = 5) This closed community occurs exclusively on the granodiorite-adamellite outcrops in the southern foothills of the Kromberg.Sample plots were located on south to southwest aspects.Soils consisted of loamy sand (10-15% clay).The community is found on steep slopes but also in some old.deeply incised (3-5 m) erosion gullies.Acacia caffra, Cussonia spicata, Diospyros whxteana and Dombeya rotundifolia have a 100% constancy value (Table 1).Another 50 woody species were identified.Forest species such as Rhus chirindensis occur within the shelter offered by the erosion gullies.Only 13 grass species were recorded with limited cover values because of the dense woody canopy.As a consequence, the grass component contains several species with forest affinities, such as Panicum deustum and Setaria megaphylla.This closed community is found at the transition of the Komati Valley to the mountains in the north and the Kromberg in the south.There are many similarities with the previous community both in terms of structure and composition.As in the previous community Acacia caffra, Cussonia spicata.Diospyros wliyteana and Dombeya ro tundifolia are prominent.Hippobromus pauciflorus occurs much more frequently and with higher cover values.However, the greatest difference lies in the occurrence of Berchemia zeyheri, Grewia occidentalis and Olea europea subsp.africana.These important browsing species w'ere not encountered in the previous community.The limited grass cover of only 12 species consists mostly of shade lov ing grasses.Forb cover is mostly in the 5-25% class.This community is characteristically found on chert ridges often in an exposed summit position or on steep north-facing upper slopes of the Kromberg and the northern mountains.The sample sites were invariably extremely rocky.This combination of factors results in a xeric com munity despite its generally high elevation.Woodv cover is mostly below 25%.A total ol 48 woody species was record ed in this sparse to open shrubland.Xerophyta retinen is is a conspicuous element.The grass layer is fairly open for this type of sour grassland.Total grass diversity was 46 species.This community is subject to higher grazing pressures than Community 18. Thirty percent of the sample plots had a medium to high grazing pressure as compared to only 3% in the Loudetia simplex-Themeda triandra Community.Aristida canescens, Diheteropogon amplectens, Loudetia simplex and Themeda triandra occur in more than 75% of the plots and achieve cover values of 25% or more.Ol inter est is the relatively high trequency ot Heteropogon contor tus.whereas this species is generally not prevalent at eleva tions above 1 2(X) m.This reflects the relatively sweet nature of this community.Almost 50% of the sample plots had a small succulent component as compared to only 15% in Community 18.Low Grassland (classification group 17,19 & 24,n = 134) This community is found across all physiographic units including isolated higher hills within the Komati Valley (Figure 9).It is the dominant community in all the higher parts of the SGR in a variety of landscape posi tions and aspects.It is a typical sour grassland in which the woody component is absent in 28% of the samples.A total of 55 woody species was recorded.The presence of several species of Protea (P.caffra, P. gaguedii, P. roupelliae and the Barberton endemic P. comptonii) is char acteristic.The grass layer is generally very dense and shorter than 0.5 m.A total of 53 species was recorded of which 10 with frequencies exceeding 60% consistently make up the bulk of the community in terms of cover and phytomass.These include Loudetia simplex, Panicum natalense, Themeda triandra, Trachypogon spicatus and Tristachya leucothrix (Table 1 This community is also widespread outside of the Komati Valley.Its peak distribution is at lower elevation than the typical sour grassland of Community 18.It is also found less on the sandstones and quartzites of the Figtree and Moodies subgroups.A total of 65% of the sample plots have soils heavier than loamy sand (10-15% clay) compared to 48% in Community 18.It is generally a more mixed community with more woody elements (both in terms of species and cover).A total of 55 woody species was recorded in both communities but the previous community had twice as many sample plots.Only 12% of the sample plots had no woody species pre sent compared with 28% in Community 18.No single woody species characterises this community.A combina tion of some of the following species is often found: Acacia caffra, Euclea crispa, Faurea speciosa, Lippia javanica and Rhoicissus tridentata.If the habitat factors and woody species are taken into account, this commu nity represents a broad transition from woodlands in the Komati Valley to the sour grasslands in the higher moun tains.The grass layer is taller than in Community 18 at 0.5 to 1 m.It is also more diverse with a total of 60 species having been recorded.The very dense grass layer has as most important species Themeda triandra and Cxmbopogon spp.(mostly C. excavatus).Forb cover is between 5% and 25% in half of the sample plots.

Vegetation of the SGR in relation to similar environments
Broad environmental determinants of grass and woody structure and composition have been identified for the SGR including land use history, elevation and geology.These are the same as those identified for simi lar montane habitats along the eastern Escarpment (Matthews et al. 1994).
At the SGR.disturbance, particularly in the form of cropping and settlement, results in a very specific vege tation overriding other factors.This is similar to findings for the Legalameetse Nature Reserve, 220 km to the north of the SGR (Stalmans 1990) and the high-altitude grasslands of northern KwaZulu-Natal, 200.km to the south (Eckhardt et al. 1996).
The mixed and sour communities found in the SGR are respectively linked to the alluvial and mafic/ultramafic volcanic substrata on the one hand and the dys trophic felsic lavas, and sedimentary sandstones/quartzites on the other hand.Similarly, the first split in the classification of the vegetation of the Suikerbosrand Nature Reserve, 250 km west of the SGR.could be ascribed to the occurrence of two very different geologi cal substrates, namely one of volcanic origin and one consisting of sedimentary dystrophic quartzite (Bredenkamp & Theron 1976).
Within individual geological substrata, elevation plays an important role in governing water availability (Ferrar & Scheepers 1988) both through increased rain fall and reduced evapotranspiration (Scheepers 1978).In the SGR.this is reflected in the occurrence of sour grass es at high elevation and on cooler and moister southern aspects of sour grasses, whereas mixed and sweet species occur at lower elevation and on warmer, northerly aspects (Figure 9).The same pattern has been observed in the Natal Drakensberg (Walker 1988) and the Bewaarkloof Mountains.250 km to the northwest of the SGR (Stalmans & De Klerk 1992).
Individual communities could not be exactly matched to each other across different studies as only grass and woody species were used in the SGR survey as compared to the full species cover assessment in most other studies.Scaling issues remain a problem, particularly as the scale of results is often determined by the study objectives.Notwithstanding the differences in approach, certain mixed, sour and forest communities of the SGR are mir rored in other areas.
With regard to mixed communities, the Cynodon dactylon-Melinis repens Short Shrubland in the SGR is similar to the Sporobolus africanus-Eragrostis cunula Young Secondary Grassland on abandoned fields in the Mlilwane Wildlife Sanctuary.50 km south of the SGR (Coetzee & Nel 1978) With regard to forest communities, the Syzygium gerrardii-Xymalos monospora Tall Forest in the SGR has great affinities with other mesophytic forests along the escarpment (Von Breitenbach 1990).Cassipourea malosana, Cola greenwayi, Englerophytum natalense, Garcinia gerrardii, Micrococca capensis and Ocotea kenyensis are exclusive to this moist forest.During a survey of Swaziland forests (Masson 1991), these species were only recorded on the Swaziland side of the Mlembe Mountain which straddles the SGR/Swaziland border.The presence of O. kenyensis was seen as a strong link to the Transvaal Drakensberg escarpment forests to the north, whilst G. gerrardii provides a link with the Indian Ocean coastal belt forests to the south (Masson 1991).Morgenthal & Cilliers (1997) similarly list C. greenwayi, O. kenyensis and G. gerrardii as species of phytogeographical importance in the 85 ha Pedlar's Bush forest, which is situated less than 3 km northwest of the SGR.They consider this as an area where an unique overlap of species common to the Transvaal and Natal forest has occurred.

Plant community suitability for herbivores
The value to herbivores of the communities described above needs to be evaluated at different scales following the ecological hierarchies identified by Senft et al. (1987).
Out of the 428 sample plots, a total of 342 are acces sible to the herbivores.Their location is not far enough away from a perennial source of water as to markedly influence herbivore distribution.The grazing impact on these plots was rated from none to light (factor 1), medi um (factor 2) and heavy (factor 3).Based on the fre quency distribution of the grazing classes, a composite value was determined for each community.
In terms of their ranked grazing values (Table 2), the first six communities are all found on the mixed and for est/thicket side of the primary split in the dendrogram (Figure 6).This supports the premise that this cluster rep- resents the more eutrophic and palatable part of the SGR.These communities occur on a basic geology at lower elevation, mostly in the Komati Valley.Forests within this cluster, however, are not much utilised.
The presence of more palatable species such as Berchemia zeyheri in the Diospyros whyteana-Hippobromus pauciflorus Community as opposed to the other wise very similar Acacia caffra-Dombeya rotundifolia Community is reflected in their respective rankings.Similarly, the transitional nature of the Themeda triandra-Cymbopogon spp.Community which falls between the mixed communities in the Komati Valley and the typ ical sour grasslands of the Loudetia simplex-Themeda triandra Community on the higher mountain slopes, is correctly reflected in its ranking.The low ranking of the Combretum apiculatum-Xerophyta retiner\}is and Ptaeroxylon obliquum-Panicum maximum Communities re flects their occurrence on extremely steep slopes or low cliffs which makes access to herbivores extremely diffi cult.
The ranked utilisation value of the communities thus reflects the theoretical delineation into mixed and sour communities based on palatability values obtained from the literature.This is also an indication that the delin eation of individual communities based on their floristic make-up is relevant in terms of their significance to her bivore utilisation.Therefore this delineation becomes useful from a game management perspective.
FIGURE 1.-Location of the study area in Mpumalanga, South Africa (information source Mpumalanga Parks Board).

FIGURE 2 -
FIGURE 2 -Distribution o f428 semiquantitative sample plots (de picted as • ) across the Songimvelo Game Reserve FIGURE 3.-Ordination diagram of the CCA o f 428 plots with elevation, drainage position and land use as environmen tal variables.Forests, A. Thicket, A; Riverine/wet land, □ : others.■ FIGURE 8.-Relative contribution of palatable, intermediate and unpalatable grass species to the mixed and sour split in the ALOC dendrogram.

15.
Diospyros whyteana-Hippobromus pauciflorus Short Thicket (classification group 21, n = 11) . Mlilwane is underlain by granites and the Combretum molle-Dombeya rotundifolia Subhumid Mountain Bushveld corresponds to the Acacia caffra-Dombexa rotundifolia Short Thicket Community occurring on the granodiorite-adamellite outcrops of the SGR.With regard to sour communities, the Loudetia simplex-Themeda triandra Low Grassland found in the SGR has many affinities to communities such as the Helichrysum wilmsii-Andropogon schirensis Low Closed Grassland of the Subhumid Mistbelt and several communities of the Humid Mistbelt grasslands of the Plateau Escarpment as defined by Deall et al. (1989) for the Sabie area.150 km to the north of the SGR.This SGR community also corresponds to the Loudetio simplicis-Alloteropsidetea semi-alatae class (as defined by Matthews et al. 1994) of the relatively low altitude (below 1 600 m asl) areas of the North-eastern Mountain Sourveld in the Sabie area.

OLD SETTLEMENT A Acacia nilotica -Euclea crispa community VERY OLD SETTLEMENT -2 Cynodon dactylon - Melims repens community Uppia javanica -Hyparrhema spp
community -1 FIGURE 4 -Ordination diagram of the CCA with elevation and land use history as environ mental variables o f 368 plots remaining after removal of forest and thicket plots Old settlement A; old lands.□ ; others.■

Cussonia spicata, Rapanea melanophloeos and Trimeria grandifolia
It occurs mostly in areas with annual rainfall above 1 100 mm on upper slopes in the landscape in the north and northeastern part of the Reserve.The canopy is char acterised by a combination of (Table1).Elements from drier communities occur, namely

Aloe arborescens, Buddleja saligna and Chionanthus foveolatus make
Short Forest (clas sification group 12. n = 1) This short forest is only known from an isolated patch on a steep, rocky, south-facing upper slope of the Kromberg.It is a very distinct community in terms of composition.up most of the cover.B. saligna was only observed in this sample plot.No grass es were observed.The forb cover of 5-25% consisted mostly of Acanthaceae.Fern cover was 1-5% with Asplenium sp. and Elaphoglossum sp.being observed.