Vegetation and vegetation-environment relationships at Grootbos Nature Reserve, Western Cape, South Africa

The private Grootbos Nature Reserve is located at the Western edge of the Agulhas Plain in the Cape Floristic Region of South Africa, an area characterized by high habitat and floristic diversity. The Reserve is covered in near-natural fynbos shrublands with a few patches of forest and wetland. The main objective o f this study was to classify the vegetation into discrete units and relate them to the prevailing environmental conditions. The vegetation was analysed by numerical means (TWINSPAN, DC A, CCA) and mapped on GIS. At the vegetation type level. Forest & Thicket and Fynbos formed distinctive clusters, whereas the wetland releves were intermixed, but without relationships to one of these units. Fire incidence served as the major determinant of the forest-fynbos boundary. The Forest & Thicket grouping was separated into Thicket (as transitional to fynbos), Afromontane Forest and Milkwood Scrub Forest. Two broad complexes were distinguished within the Fynbos grouping, the Alkaline Sand Fynbos Complex corresponding to Coastal Fynbos. and the Acid Sand Fynbos Complex corresponding to Mountain Fynbos. They discriminated along gradients of pH. soil depth and rock cover. The complexes were further subdivided into formations by using one or a few subjectively chosen dominant species as indicators. The transitions between these formations were rather continuous than discrete. The vegetation type and complex levels correspond well to existing fynbos-wide classifications. Comparing the formations to the results of other vegetation studies is problematic even on the scale of the Agulhas Plain, due to the high regional plant diversity in the Fynbos Biome.


INTRODUCTION
An important, and well-established aspect of the species-rich Cape flora is the difficulty in resolving regional vegetation associations into meaningful, easily identifi able floristic groupings (Campbell 1986).Various authors have tried to classify the habitat types and the plant com munities of the Fynbos Biome in the previous decades, some looking at broad descriptions of major vegetation types (Taylor 1978;Moll et al. 1984;Campbell 1985;Cowling & Heijnis 2001) and others at detailed descrip tions of communities (McKenzie et al. 1977;Boucher 1978;Taylor 1983;Richards et al. 1995;Cowling et al. 1996;McDonald et al. 1996;Taylor 1996).The Agulhas Plain, covering an area of 270 000 hectares of semi-arid lowland fynbos and renosterveld.has been prioritized as an area of high irreplaceability and high vulnerabil ity, with exceptionally rich coastal lowland ecosystems (Cowling et al. 1999).Cowling et al. (1988) applied the mountain vegetation classification approach of Campbell (1985) to the Agulhas Plain.This classifica tion was produced using the Braun-Blanquet method of table sorting and was based on structural and higher taxon characteristics of the vegetation.Nine zonal com munities, at various hierarchical levels, were recognized and mapped.This phytosociological approach provided a qualitative description of community environmental characteristics on the Agulhas Plain.The only study on the Agulhas Plain to provide a quantitative assessment of the importance of environmental factors was Richards et al. (1995) who explored vegetation-environment rela tionships in a 30 ha study area, the Soetanysberg, ± 15 km west of Cape Agulhas (34 45' S; 19 50' E).However, no comparable studies have been undertaken on the west side of the Agulhas Plain.
The major objective of this study was to fill this gap through a detailed mapping and numerical analysis of the vegetation of Grootbos Nature Reserve (GNR), which is located at the western edge of the Agulhas Plain, between the villages of Stanford and Gansbaai (Figure 1 A).The geographic coordinates are 34°32'30" S and 19°24'50" E. The privately owned reserve consists of seven formerly separate farms covering an area of 1 700 ha.It is run as an upmarket ecotourism lodge.The land scape is sloping, with a maximum elevation of ± 475 m along the slopes of the Swartkransberg (Figure 1B).The underlying rock type of the more elevated hills is quartzitic Table Mountain Sandstone.On the lower slopes, deposits of sandy aeolian material overlay the bedrock.Outcrops of the Bredasdorp Limestone are exposed in some places.
The climate, which can be characterized as maritime mediterranean, shows a strong seasonality in precipita tion.May to August are the wettest months (most pre cipitation is carried by northwesterly winds), whereas the southeasterly winds prevailing in the summer months are dry.There has been no long-term precipitation record at GNR.The mean annual rainfall from 1996-1998 was 730 mm.The mean maximum daily temperatures fluc tuate between 25°C in February and 14°C in July.The GNR is frost-free.
Since 1995 the property has been managed for con servation and ecotourism.Over the last decade a team of botanists has been actively sampling and collect ing plant specimens over the entire reserve.At the lat est count, 660 species of indigenous plants have been recorded on the property, with still more being found each year (Privett & Lutzeyer in press).Fifty-one spe cies are included in the Red Data List of threatened species (Hilton-Taylor 1996), and three species (Erica magnisylvae, Cliffortia anthospermoides, Lachenalia lutzeyerii) were recently discovered and are considered endemic to the Reserve.The vegetation of GNR.largely fynbos shrubland with some patches of forest and wet land, is amongst the best surveyed in the Cape Floristic Region and lends itself to a detailed quantitative analysis of vegetation-environment relationships.

Data collection
Vegetation sampling at GNR was carried out in w inter 1997 and in spring 2004.Seventy-one 50 m2 (10 x 5 m) rectangular releves were analysed for their floristic com position and their major environmental features.They were subjectively chosen in order to represent homoge neous patches of vegetation and divided into 10 cells, 2.0 x 2.5 m each.The floristic composition was investi gated for each cell, including all identifiable species.All taxonomy followed Goldblatt & Manning (2000).The cover was recorded as a percentage.Soil samples were taken at random localities within each releve, in depths of 5-30 cm depending on soil depth.Environmental data were recorded for each releve: topographic parameters (slope, elevation, aspect), vegetation age, soil depth, rock cover, pH, resistance and nutrient levels of the soil.Flow accumulation (wetness index), solar radiation and the exposure to the fire-bearing southeasterly winds were computed from a digital elevation model (DEM).All releves were used for analysis, including those located in transitional vegetation.The GPS coordinates of each releve were stored and the southwestern comer was marked with a concrete Until or an iron peg to enable further monitoring.
In addition to the regular releves (core dataset), 127 sites were analysed to a lesser extent (these and the core dataset together are referred to as the extended data set).This was done without setting up a formal plot and only the dominant and easily identifiable species were recorded.The coordinates were located with a GPS and stored for mapping.No environmental data were taken directly for these additional releves, but the parameters extracted from the DEM were stored.

Numerical vegetation analysis
A classification and certain ordinations were applied to the sampled dataset.The classification was performed using TWINSPAN (Hill 1994).For the classification of the core dataset, the pseudospecies cut levels were set to 0, 5, 10, 20 and 50, for the extended dataset to 0, 2, 5, 10, 20, 50.The first pseudospecies of the extended data set was excluded from the classification.Splitting was allowed down to two species (default: five) for both data sets.An indirect ordination (Detrended Correspondence Analysis, DCA) was performed for the extended dataset.A direct ordination (Canonical Correspondence Analysis, CCA) was applied to the core dataset (due to the avail ability of environmental data).Both ordinations were run with the whole dataset and then repeated only including releves classified as Fynbos.The default parameters sug gested by the programs (DECORANA for the DCA and MVSP for the CCA) were applied to the analyses.

Vegetation mapping
A set of colour orthophotos (spatial resolution: 0.75 m) as well as a GIS dataset representing preliminary vegeta tion units mapped in 1997 were available.Attempts to extract vegetation units from the imagery by numerical means failed as the different fynbos types showed very similar reflection properties, while the reflection proper ties varied very much within each type.Consequently, GNR was explored during several excursions and assessed according to the subjective impression of the authors, supported by the releve information.In addition, a large number of localities were recorded by GPS in order to locate the transitions between vegetation units.All the point data were transfered to a GIS and mapped onto the composite orthophoto.The information provided by the orthophotos was combined with the point data, the results of the numerical vegetation analysis and the existing GIS dataset to generate a comprehensive map of meaningful vegetation units. {

Numerical vegetation analysis
The TWINSPAN classification clearly supported the presence of two major vegetation types on GNR: one fynbos (Cape Fynbos Shrublands) and the other nonfynbos (Forest & Thicket).The wetland releves were not clearly assigned to any of these groups, and as a result were considered as an independent vegetation type.The Fynbos grouping was further split into an Acid Sand Fynbos Complex and Alkaline Sand Fynbos Complex, the Forest & Thicket grouping in Milkwood Scrub Forest, Afromontane Forest and Thicket.No further sig nificant splitting was possible in the wetland grouping.
The level of vegetation complexes was the maximum of detail supported by the classification.However, for the fynbos, some smaller units were clearly recognizable in the field, mostly dominated by a few or even only one species and with the bulk of the species shared among all the units of the corresponding complex.These detailed units were conceptualized as formations, which are sum marized in Table 1.
The ordination results corresponded well to the clas sification results.In the DC A for the extended data set, Afromontane Forest, Milkwood Scrub Forest and Fynbos were clearly separated along the first axis (eigenvalue = 0.91), with the Thicket and the Wetland releves intermediate between Milkwood Scrub Forest and Fynbos (Figure 2A).The two fynbos complexes were discriminated along the second axis (eigenvalue = 0.65).The DCA only including the releves classified as fynbos (Figure 2B; eigenvalues: 0.68 for the first axis and 0.62 for the second axis) showed a clear cluster ing of some formations established in the classification ( Thamnochortus fraternus Restioid Fynbos, Erica sessiliflora Ericaceous Fynbos) and the post-fire releves of Protea repens Proteoid Fynbos.The transitions between the remaining formations appeared to be highly continu ous.
The CCA (Figure 3A) illustrated the clear and discrete discrimination of Fynbos and Forest depending on veg etation age, which was almost coincident with the first axis (for eigenvalues and canonical coefficients com pare Table 2).The forest complexes were discriminated by elevation, slope, aspect and certain soil characteris tics.The fynbos releves were aligned continuously along the second axis, discriminated by gradients of elevation, slope, pH, soil depth and rock cover.The Alkaline Sand Fynbos Complex and the Acid Sand Fynbos Complex were clearly recognizable, but closely together.The exclusion of all non-fynbos releves from the CCA (Figure 3B) did not lead to a clearer distinction of these clusters.No clear structure was recognizable within the two subclusters either.Only the Thamnochortus frater nus Restioid Fynbos and the Protea repens Proteoid Fynbos formed ± proper clusters.A number of environ mental variables showed similar explanatory' value.In addition to the site scores, the scores of selected species were plotted (Figure 4).The species were grouped in the same manner as the releves they dominate.The red data species appear to be associated with rather extreme con ditions as they occupy the edges of the plot.

Afromontane Forest
Five patches of Afromontane Forest are present at the GNR, all of them situated in the forest valley (Figure 5A).The CCA suggested an association with shal low, neutral to slightly acid soils rich in K and Mg.The relief is characterized by steep slopes protected from the southeasterly winds.Afromontane Forests form tall canopies with heights of more than 10 m.The canopy is frequently dominated by Rapanea melanophloeos.The well-developed subcanopy is 3-10 m high, compris ing Celtis africana, Chionanthus foveolatus, Diospyros whyteana, Kiggelaria africana and Sideroxylon inerme.Most of these species have the potential to grow to can opy height.Due to the dark interior of the forest, the ground layer is usually sparse with Asplenium adiantum-nigrum, Droguetia iners and Ehrharta erecta as the most common species.Climbers are present (Asparagus aethiopicus, A. scandens and Cynanchum obtusifolium).The species diversity is low at the 50 m2 level, averaging 13.2 identifiable species per releve.
The range of temperate forest ecosystems referred to as Afromontane Forest includes the mountains of the southern Cape but stretches far into tropical Africa where they occur at higher altitudes (Midgley et al. 1997).The Afromontane Forests of the Agulhas Plain are ecologi cally similar but floristically distinct from those consid ered by Campbell (1985).Instead they show affinities to the dune forests of the Tongaland-Pondoland Forest (Moll & White 1978).These forests, and also the Afromontane Forest along the south coast (Knysna.Tsitsikamma).have significantly higher species diversities than the forests on the GNR.They host taxa not present on GNR.such as Podocarpaceae and Cyatheaceae.One reason for the declining diversity towards the southwest may be that more and more species disappear as the climatic condi tions become harsher and the forest patches smaller.

Milkwood Scrub Forest
Four patches of Milkwood Scrub Forest are present at low elevations, associated with deep, slightly acid to alkaline, sandy, colluvial soils with high contents of Ca and P. In contrast to the Afromontane Forest, the CCA did not indicate a negative spatial coincidence with the southeasterly winds.The high levels of mineral com ponents appear to be a distinctive feature of this type of ecosystem.Ca in particular is considerably richer in the forest than anywhere else on the GNR.In addition, the soils are very fertile due to plant-induced organic enrich ment (Thwaites & Cowling 1988), which makes them suitable for agriculture and thus susceptible to anthropo genic disturbance.
A single tree layer, usually dominated by Sideroxylon inerme, attains a height of 6 m and a very dense cover.Euclea racemosa does form part of the canopy in some places, whereas Chionanthus foveolatus.Gymnosporia huxifolia and the winter-deciduous Celtis africana occur as single trees or small groups.The soil is covered by a 50-300 mm high, sparse to dense (where sufficient light is available) herb layer dominated by Droguetia iners and Ehrharta erecta.The only shrub species in the full Bothalia 38,1 (2008) Vegetation unit Diagnostic spp.

Thicket
The fynbos-forest boundary is rarely sharply defined, but is often made up of a thicket of varying height and density.This vegetation unit does not necessarily corre spond to the Subtropical Thicket described by Midgley et al. (1997) fire frequency or intensity have been reduced over a cer tain period, but not sufficiently to support forest: in the buffer zones between fynbos and the Milkwood Scrub Forest, in protected ravines as a successional stage from fynbos to Afromontane Forest and on some south-fac ing slopes of the forest valley.The latter (adjacent to the Afromontane Forest) is dominated by the fern Pteris dentata, which attains a very high cover in these places.These ecosystems are therefore considered as separate formations {Pteris dentata Shrubland, compare Figure 8 ).Forest edge thickets and valley thickets, in contrast, are floristically similar so that a separation into two for mations is not supported.The broad-leaved subtropical shrub species (Cowling et al. 1997)

Wetlands
The catchment areas supplying the GNR are not large enough to support permanent streams under the prevail ing precipitation regime.In winter, some springs can develop and small rivers may persist until the end of October.As a result, the distribution of true wetlands is extremely limited at the GNR and is confined to a few suitable habitats.Nevertheless the diversity of differ ent wetland habitats is considerable and it is difficult to point out one type of wetland characteristic of the GNR.As the different types are located adjacently along sev eral environmental gradients, they will be treated as one entity.Wetlands are related to soils with a high content of organic matter, indicated by the dark colour.The ordinations placed the wetland releves in between for- est and fynbos without clear relationships to any of the measured variables.The pH ranges from 6 .1-7.9 but the releve with the highest value is situated below a slope that may have an alkaline character.The largest patch is located at the bottom of the forest valley.South-facing slopes are covered in a dense, 1.5 m high canopy of Pteris dentata, (recognized as formation Pteris dentata Femland), whereas slightly north-facing slopes support mats of Cliffortia ferruginea and dispersed stands of Artemisia afra, Leonotis leonurus and some thicket species.Gunnera perpensa and Senecio halimifolius are dominant directly in the temporary stream.Species such as Mariscus thunbergii, Hippia frutescens and Zantedeschia aethiopica are present in all types.Transitions to Thicket are common.One characteristic feature of the wetlands is their low alpha-diversity with an average of only 9.3 species per releve.

Alkaline Sand Fynbos Complex
The Alkaline Sand Fynbos Complex covers most of the lower parts of the GNR.Two hundred species were recorded in 42 releves, averaging at 24.9 species per releve.The substrate is mainly wind-blown, colluvial sand of varying depth, pH and nutrient levels.Dune Asteraceous Fynbos, Neutral Sand Proteoid Fynbos and Protea repens Proteoid Fynbos are associated with these habitats.Some limestone outcrops show a dif ferent vegetation but they are too small to house many of the rare limestone endemics found a few kilometres to the east on larger patches.Three different limestone formations, which are frequently intermixed, have been recorded: Protea obtusifolia Proteoid Fynbos, Erica coc cinea Ericaceous Fynbos and Thamnochortus fratem us Restioid Fynbos.
Dune Asteraceous Fynbos covers the entire western part of the GNR.It forms an extensive matrix o f low ericoid/restioid fynbos, higher ericoid fynbos and thicket dominated by broad-leafed shrubs (Figure 6 A).The for mation is characterized by low elevation, alkaline, deep soils (> 100 cm) and low rock cover.It corresponds well with the Dune Asteraceous Fynbos described by Cowling et al. (1988) for the Agulhas Plain and with the Eastern Type of Coastal Fynbos (Kruger 1979) The formation shows some association with the Protea susannae-Leucadendron coniferum Proteoid Fynbos described by Cowling et al. (1988) and Richards et al. (1995), and the Neutral Sand Proteoid Fynbos described by Mustart et al. (2003).These units prefer rather deep, colluvial neutral sands at least partly derived from Bredasdorp Limestone.Protea susannae.however, does not occur on the GNR.Instead, Leucadendron coniferum and Leucospermum patersonii are dominant with shift ing composition.Relatively old, almost monospecific stands of L. coniferum resemble low forests with canopy heights of up to 5 m (Figure 6 B), but with an extremely high density of thin stems, making them almost impen etrable.In young stands (up to 10 years after fire), the two proteoids are approximately equally abundant.An ericoid layer with variable density is present.As in the Dune Asteraceous Fynbos, Thamnochortus erectus is the most conspicious restioid element, whereas low restioids occur in the ground layer, together with various species of the sedge Ficinia.North-and east-exposed slopes of the forest valley are covered in a different vegetation unit.Owing to the high frequency of Leucadendron coni ferum and Leucospermum patersonii and the scarcity of species indicating a different unit, it has been assigned to the Neutral Sand Proteoid Fynbos, but as a different unit in Figure 7. Protea obtusifolia Proteoid Fynbos grows on lime stone outcrops and is associated with shallow to mod erately deep, alkaline soils and with low rock cover.The pH ranges from 6.9-8.1.The formation is charac terized by 2 m high, sparse to medium-dense stands of P. obtusifolia often intermixed with Leucadendron coni ferum and sometimes with Leucospermum patersonii.The bulk of the species occurs throughout the Alkaline Sand Fynbos.The formation is frequently intermixed with other formations of the Alkaline and even the Acid Sand Fynbos, sometimes on a very fine scale.Despite floristic differences, it shows affinities to the Protea obtusifolia-Leucadendron meridianum Proteoid Fynbos of the Agulhas Plain (Cowling et al. 1988) and to the Leucadendron meridianum-Protea obtusifolia Proteoid Fynbos of the Soetanysberg (Richards et al. 1995).

Protea repens
Erica coccinea Ericaceous Fynbos is associated with alkaline soils rich in Na and Ca.It is not related to the Ericaceous Fynbos of Campbell (1985) and Cowling (1988)  as well.The formation is associated to Kruger's (1979) limestone fynbos.

Acid Sand Fynbos Complex
The Acid Sand Fynbos Complex is indicated by the presence of one or more Proteaceae species associated with shallow, acidic soils derived from Table Mountain Sandstone (Mimetes cuccullatus and Leucadendron salignum are the most frequent).The higher hills of the GNR are entirely covered in this complex (Figure 8 B).Acid Sand Fynbos has the highest species diversity among the vegetation complexes of Grootbos, averaging at 35 species per releve with a maximum of 48 species.One hundred and forty-nine species were recorded in the 12 releves of the core dataset altogether.Four formations of Acid Sand Fynbos were separated in this study.Part of the complex was in an early post-fire stage at the time of the survey and was dominated by Aspalathus ciliaris, Pseudopentameris macrantha, Thesium strictum and Othonna quinquedentata.
Acid Sand Proteoid Fynbos grows predominantly on hilltops and steeper slopes, with its largest patch on the Swartkransberg.It is associated with high eleva tion, shallow, rocky and acidic (pH 5.2-5.6 ) soils with low levels of all nutrients except K, and north-facing slopes.The formation shows environmental and struc tural affinities to both the Leucadendron xanthoconus-Leucospermum cordifolium Ericaceous Fynbos and the Aulax umbellata-Protea compacta Proteoid Fynbos of the Soetanysberg (Richards et al. 1995).It is floristically and structurally characterized by the absence of species that have their centre of distribution in the Alkaline Sand Fynbos Complex and by a low to medium-dense proteoid layer usually less than 1.5 m tall, including several acidophilous Proteaceae with a changing composition and without clear dominance of one species.Mimetes cucul latus and Leucadendron salignum are the most common, joined by Aulax umbel lata, Leucadendron tine turn, L. spissifolium, L. xanthoconus, Protea acaulos, P. cynaroides, P. longifolia and P. speciosa.Several smaller spe cies of Erica and further ericoid shrubs form a sparse to medium, low-height ground layer.The restioid compo nent is variable but Elegia juncea in particular is abun dant in some places, as is the large "graminoid' Iridaceae Bobartia indica.Some of the non-sprouting Proteaceae, in particular Aulax umbel lata and Leucadendron xan thoconus, may gain dominance in some places as well.
According to Campbell (1986)  Erica sessiliflora Ericaceous Fynbos grows on damp, south-facing slopes and in valleys.The Acid Sand Fynbos elements are joined by several moisture indica tors, such as Berzelia lanuginosa, Cliffortia ferruginea, Drosera capensis and Psoralea arborea.Structurally dominant is Erica sessiliflora, which can be up to 1.5 m high and can attain cover values of more than 7 5 %.Mimetes cucullatus, Leucadendron coniferum and E. glabella are common.The largest patch occurs on the southern slope of a mountain on the Steynsbos prop erty, but extended canopies o f E. sessiliflora also occur on the northern slope of the Swartkransberg, indicating that the species can also cope with less moist conditions.Transitions to other fynbos formations are manifold.The formation corresponds well to the Wet Ericaceous Fynbos described by Campbell (1986) and Cowling et al. (1988).
Elegia thyrsifera Restioid Fynbos occupies three patches o f fynbos on GNR.It is structurally domi-  Campbell's (1986) Restioid Fynbos although it is diffi cult to be assigned to one of the subseries.The sites on the GNR may represent a gradient from Mesic Restioid Fynbos with a higher share of ericaceous Ericaceae (mainly E. sessiliflora) to Dry Restioid Fynbos.Floristically it does not correspond to the Dry Restioid Fynbos of the Agulhas Plain (Cowling et al. 1988).
Transitional Proteoid Fynbos constitutes a mixture o f species centred in the Alkaline Sand Fynbos.and spe cies centred in the Acid Sand Fynbos.The transitional character is well indicated by the CCA.The pH, how ever, does not exceed 6 .Various subtypes of this forma tion are present on GNR.Some patches are structurally similar to the Neutral Sand Fynbos but acidophilous proteoids, usually Mimetes cucullatus, indicate the more acidic conditions.Damp ravines on the southern slope of the Swartkransberg support a thicket-like vegetation with a high cover of Olea capensis, but still with fynbos character.Another transition zone exists between Protea obtusifolia Proteoid Fynbos and Acid Sand Fynbos, leading to very complex situations with P. obtusifolia and Aulax umbel lata growing almost together, but rep resenting completely different soil nutrient regimes.The major difference to the transition described above is that it is not based on a gradual decrease of soil depth and pH-value, but on a fine-grained mosaic of young wind blown, shallow, calcareous soils, older limestone ridges and underlying Table Mountain Sandstone.Extremely complex situations prevail at the western slope of a mountain at Steynsbos, where Leucadendron coniferum, Leucospermum patersonii and P. obtusifolia coexist with some of the acid sand proteoids.

Vegetation mapping
The vegetation units illustrated in the map (Figure 7) largely correspond to the units established in the analy sis.Only a few changes were made (e.g.patches of fyn bos in early post-fire stages).An additional level of detail was introduced in some cases.Most of GNR is cov ered in Fynbos (1 620 ha or 95.3%).Forest & Thicket account for 78 ha or 4.6% (Afromontane Forest 4.1 ha or 0.2%, Milkwood Scrub Forest for 43 ha or 2.5% and Thicket for 30.9 ha or 1.8%), whereas only 1.4 ha (0.1%) are covered in Wetland.Dune Asteraceous Fynbos cov ers more than half of GNR (914 ha or 53.8%), and the whole Alkaline Sand Fynbos Complex occupies 1 385 ha or 81.5%.Even the second largest formation of this complex, the Neutral Sand Proteoid Fynbos (191 ha or 11.3%), approaches the same amount of cover as the whole Acid Sand Fynbos Complex (235 ha or 13.8%).
Figure 7 also gives an idea of the fragmentation of the vegetation units.The interpretation of this information has to be approached with caution because the fragmen tation may be caused by different factors, including the shape of GNR.Dune Asteraceous Fynbos and Neutral Sand Proteoid Fynbos cover continuous, hardly frag mented areas with average patch sizes of 305 and 96 ha, respectively.Protea repens Proteoid Fynbos and Erica sessiliflora Ericaceous Fynbos also exceed average patch sizes of 25 ha.In contrast, the distribution of Limestone Fynbos is rather patchy, with 27 patches ranging from tens of hectares of Protea obtusifolia Proteoid Fynbos to tiny limestone outcrops covered with Thamnochortus fratem us Restioid Fynbos.The average patch size ( 6 ha) is not representative in this case.In the Forest the patch size of the highly fragmented Afromontane Forests remains below 1 ha (0.81), whereas the patches of Milkwood Scrub Forest occupy slightly more than 10 ha on average.DISCUSSION Three levels of vegetation units were established on the GNR: the vegetation type level and the vegetation complex level were based on environmental conditions and species groupings (TWINSPAN), and the formation level on dominant species.
The units of the vegetation type level largely cor respond to the biome level of the Broad Habitat Units established by Cowling & Heijnis (2001), based on environmental variables.The Fynbos and Forest Biomes occur on GNR, the Thicket Complex identified in this study does not correspond to the Thicket Biome of Cowling & Heijnis.The complexes of this study corre spond partly to the primary units.Campbell (1985) con sidered Forest and Thicket as one group, as in this study.Wetlands are neither considered by Campbell (1985) nor by Cowling & Heijnis (2001).Cowling et al. (1988) classified them as azonal vegetation.
The separation of the fynbos on GNR into an Alkaline and an Acid Sand Fynbos Complex seems sufficiently supported by the classification and the ordinations.The two complexes correspond to the Coastal Fynbos (or Lowland Fynbos) and Mountain Fynbos, respectively (Acocks 1953;Taylor 1978;Kruger 1979;Moll et al. 1984).According to Cowling et al. (1988), a separation of fynbos in this way has to be rejected because none of the stuctural units (Campbell 1985) are restricted to one of those groups, and because the floristic changes within the Mountain Fynbos are as significant as between Mountain Fynbos and Lowland Fynbos.However, this study shows that it appears to be highly relevant on a local scale, where the gamma-diversity does not play the role it does on a broader scale while the beta-diversity is still high.
The division of the complexes into proper commu nities presents problems.Most subcomplex vegetation units are based on one or a maximum of two dominant species which give the landscape a very characteristic appearance.Nevertheless, the classic concept of plant communites as an association of several characteris tic species which differentiates it from other commu nites should not be applied to these entities.The struc tural component in such a classification is evident, and the groups (here called formations) show a fairly strong correlation to the classification systems of Campbell (1985) and Cowling et al. (1988) where structural fea tures of the vegetation were included more systemati cally.The difficulty of purely floristic classification sys tems in the Fynbos Biome, due to high gamma-diversity, is illustrated by comparing the findings of this study to the vegetation study of Richards et al. (1995) for the Soetanysberg.Despite environmentally comparable con ditions and a distance of < 50 km, substantial floristic dif ferences, also among the dominant species, are evident.
The expected dependence of fynbos vegetation on certain environmental factors was confirmed by the study.The major explanatory variables in the CCA of Richards et al. (1995) in their vegetation study of the Soetanysberg were pH, rock cover, soil depth and soil texture.Apart from soil texture, which was investigated in more detail than in this study, the explanatory vari ables are the same as for GNR.Elevation, which plays a major role here but not in the Soetanysberg, is probably mainly a surrogate for the aeolian sediment accumula tion budget, wind speed and the distribution of different substrate types, as the correlation values (e.g.0.78 with pH) indicate.
The clustering of the releves in the DCA and the CCA was rather poor (compare Figures 2B; 3B) and the tran sitions between the fynbos units many.Only the Acid Sand Proteoid Fynbos formed a clear cluster, together with some releves of Transitional Proteoid Fynbos.This finding is partly in line with the study of Richards et al. (1995): the releves connected to low pH (sandstone) were poorly clustered in the CCA biplot, but they were clearly separated from the releves on limestone.On GNR, the transition between the limestone formations and the remaining formations was continuous with the Neutral Sand Proteoid Fynbos as intermediate forma tion, but rather with affinities to the limestone forma tions.In contrast, the Protea susannae-Leucadendron coniferum Proteoid Fynbos of Richards et al. (1995), a formation that is associated with neutral sands according to Cowling et al. (1988), did not differ substantially in pH from the sandstone formations, but was clearly dis criminated from the limestone formation (Protea obtusifolia-Leucadendron meridianum Proteoid Fynbos).
The poor clustering in the fynbos of the GNR may be explained by a variety of factors.The location of the releves (some of them were placed in transitional zones) and the high level of detail in the study (leading to a considerable amount of noise) may serve as one expla nation, the topographic and geological patterns on GNR as another; alkaline sands cover the entire western part of the GNR, and limestone ridges of varying shape and size are widely dispersed.This leads to wind-blown alkaline sands of variable depth over large parts of acid substrate and to an extremely fine-scaled pattern of different physi cal and chemical substrate properties.This is especially true for the foothills and the lower slopes of the higher hills of GNR, leading to very complex vegetation pat terns; in some places Aulax umbel lata.a strong indica tor of acidic conditions, and Protea obtusifolia, a strong indicator of alkaline conditions, grow immediately adja cent to each other.Only the highest parts of the GNR. the Acid Sand Fynbos Complex, remain untouched and do not contain calcareophilous species.In such a fine mosaic of different environmental conditions, the vegetation is sensitive to influences other than topographic and sub strate variables.The problems with grouping fynbos eco systems based on reseeding proteoids (Richardson & Van Wilgen 1992;Privett et al. 2001) are well established, as these organisms are susceptible to local extinction ('drifting clouds of species abundance').Therefore the structural dominants of the vegetation of a certain place may shift from fire interval to fire interval, as Privett et al. (2001) have shown for the Cape Peninsula.However, this phenomenon only occurs at a subcomplex level.The boundaries of the fynbos complexes are not affected as steep environmental gradients prevent mixing of the spe cies pools (high beta-diversity).
Figure 9 represents the floristic diversity of the veg etation of GNR at the complex level.Fynbos and Forest constitute completely different species pools.Less than ten taxa occur in both vegetation types, whereas Acid Sand Fynbos and Afromontane Forest share no taxa at all.In contrast the complexes within each vegetation type share a considerable number of their taxa.Wetland shares only a few taxa with the other groupings, but the low diversity of the wetland may contribute to this phenomenon.Although the Alkaline Sand Fynbos hosts more taxa (200) than the Acid Sand Fynbos (149), the diversity of the latter is more than four times higher if normalized to the area (0.63 and 0.14 taxa per hectare respectively).Even though this ratio may be of limited value, it confirms the general patterns obtained from the releve data.
As a conclusion, it can be stated that the vegetation patterns of GNR were investigated in great detail, lead ing to a differentiated picture of the spatial distribution of the vegetation units and the vegetation-environment relationships.The major implications for further man agement is the potential impact of fire on vegetation structure, in particular the vegetation units dominated by reseeding proteoids.Careful planning to mimic natural fire frequencies and conditions will be necessary in order to prevent local extinction.In order to protect Red Data species, particular care has to be taken when undertaking controlled bums of extreme habitats, such as steep lime stone ridges and sandstone slopes.

SHORT NOTE
In February' 2006, after the submission of this paper, the fynbos ecosystems of the entire GNR were burnt by a fire.Since then, the number of species recorded has increased from 660 to 732 (Privett & Lutzeyer in press) and two further new species (Capnophyllum sp.nov.and Pterygodium sp.nov.) were recorded.

FIGURE 1 .
FIGURE 1.-A, study area located between Cape Town and Cape Agulhas, 5 km east of Walker Bay seashore; B, western part o f Grootbos Nature Reserve gently sloping towards Walker Bay, eastern part hilly with maximum elevation slightly below 500 m.

FIGURE
FIGURE 7.-Vegetation map of Grootbos Nature Reserve, based on orthophotos and field studies.
, although it may share certain characteris tics with it.Thickets are characterized by a mixture of forest and fynbos elements.They occur at sites where FIGURE 3.-A, results of Canonical Correspondence Analysis (CCA) for entire core dataset indicate clear separation between Fynbos complexes, Afromontane Forest and Milkwood Scrub Forest, whereas Wetland releves are intermixed; B, results of CCA for fynbos releves o f core dataset show continuous transition between formations.
, mainly Euclea race mosa, Olea capensis, O. exasperata and various species of Searsia (Moffett 2007) frequently dominate thickets on GNR together with Salvia africana-lutea.Among the fynbos elements, large individuals of Leucadendron con iferum and Thamnochortus erectus are most common.Certain forest elements, most commonly Sideroxylon inerme, may join the thicket but some thicket species may also occur as trees in the Milkwood Scrub Forest.
. Large ericoids (Metalasia muricata and Passerina vulgaris) and nonproteoid broad-leafed shrubs (Chrysanthemoides monilifera) are the structural dominants, in some places joined or replaced by fabaceous shrubs, Otholobium bracteolatum, Aspalathus forbesii.Erica irregularis and certain non-ericaceous ericoids (Anthospermum aethiopicum, Phylica ericoides) are widespread.Thamnochortus erectus is the most conspicuous restioid but the lower Ischyrolepis eleocharis has a very high cover in some sites.Lower individuals of the subtropical shrub species are very common all over the Dune Asteraceous Fynbos (Euclea racemosa, Olea capensis subsp.capensis, O. exasperata and Searsia laevigata).One characteristic feature of the Dune Asteraceous Fynbos is the lack of proteoids.If they occur, especially Leucadendron coni ferum and Protea obtusifolia, they indicate a transition to other formations.
Proteoid Fynbos occupies the lower slopes of the Swartkransberg heading northwards towards the broad Steynsbos Valley, as well as large parts of the Steynsbos Valley itself.It forms 2.5 m high, mediumdense to dense stands of Proteoid Fynbos, associated with extraordinarily high values for electrical resistance is frequently intermixed.The ericoids Cliffortia ilicifo lia and Passerina vulgaris are very frequent, growing up to more than 1.5 m, as does the restioid Thamnochortus erectus.A sparse to medium-denfee undergrowth attains heights of some tens of centimetres.Parts of the forma tion were in an early post-fire stage at the time of the survey, showing a high cover of Aspalathus microphylla and lacking visible individuals of Protea repens.
and has its centre of distribution on rather steep limestone outcrops.It is characterized by a high cover of E. coccinea (yellow-flowered variant), which is frequently joined by Indigofera brachystachya and Cullumia squarmsa.Transitions to other formations of the Alkaline Sand Fynbos are common.The usually sparse ground cover is made up of shrubs such as Indigofera brachystachya or Cullumia squarrosa, and also Erica coccinea (yellow-flowered variant), the latter indicating a transition to the E. coc cinea Ericaceous Fynbos.Protea obtusifolia and small individuals o f Leucadendron coniferum may appear and Cow ling et al. (1988), part of the Acid Sand Proteoid Fynbos distinguished in this study should instead be recognized as Asteraceous Fynbos as those authors only recognize a community as Proteoid Fynbos if it contains more than 10% cover of reseeding proteoids.The resprouting Mimetes cucullatus and Leucadendron salignum.in contrast, do not fulfil this requirement.
FIGURE 9.-Species diversity o f vegetation complexes o f Grootbos Nature Reserve.Size of each box and number inside indicate total species number o f complex, whereas width o f lines and adjacent numbers indicate number o f species shared by two complexes.AcSF.Acid Sand Fynbos Complex; A1SF.Alkaline Sand Fynbos Complex; MSF.Milkwood Scrub Forest; AF.Afromontane Forest; WL.Wetland.