Vegetation of the Sileza Nature Reserve and neighbouring areas , South Africa , and its importance in conserving the woody grasslands of the Maputaland Centre of Endemism

An analysis o f the plant communities of the Sileza Nature Reserve and surrounding areas (± 4 124 ha) is presented. The study area falls within the Maputaland Centre o f Endemism, which is part ot the Maputaland-Pondoland Region, a centre of plant diversity rich in endemic plants and animals. A TWINSPAN classification, refined by Braun-Blanquet procedures, revealed 12 distinct, mainly grassland plant communities. A hierarchical classification, description and ecological interpre­ tation ot these communities are presented. The level o f the water table, either directly, or indirectly through its role in soil formation, is the deciding factor in defining plant communities on the geologically young sandy substrate. Fire is an essen­ tial factor, particularly in maintaining the woody grasslands, a rare vegetation type rich in geoxylic suffrutices. and unique to the Maputaland Centre. A comparison between the endemic complement in the subtropical coastal grasslands of Maputaland and the high-altitude Afromontane grasslands of the Wolkberg Centre of Endemism shows marked differences in grow th form and vegetation type partitioning between the two centres. This can probably be ascribed to the relative youth (Quaternary) of the Maputaland coastal plain and its associated plant communities. Notable for their richness in Maputaland Centre endemic/near-endemic taxa. the conservation of sand forest and woody grasslands should receive high priority. Afforestation with alien trees is the most serious threat to the biodiversity of the Maputaland coastal grasslands, not only because of habitat destruction, but also through its expected negative effect on the hydrology of the region


INTRODUCTION
M aputaland, previously know n as Tongaland, co m p rises the n o rth ern m o st secto r o f the T ongaland-Pondoland Regional M osaic, one of the m ain A frican p hytochoria recognised by W hite (1983).M ost o f this phytochorion falls w ithin the M aputaland-Pondoland R egion, a centre o f plant endem ism (Van Wyk 1994).S ituated at the southern end o f the tropics in A frica, em phasis on M aputaland as a region o f biotic transition has largely obscured its status as a regional centre o f endem ism in its own right (Figure 1).Recent formal recognition o f the M aputaland Centre [MC] (Van Wyk 1994. 1996) acknow ledges the high levels o f endem ism and rem arkable biodiversity o f this region.
The vegetation o f the MC is diverse, with at least fif teen broad vegetation types described for the KwaZulu-Natal portion o f the region (Tinlev 1976;M oll 1977. 1980).H ow ever, w ith the exception o f M yre's (1964. 1971) pioneering studies on the vegetation o f southern M ozam bique, very little quantitative w ork has been done on the grasslands o f the centre.The present study on the vegetation o f the Sileza N ature Reserve [SN R], an area dom inated by grasslands, presents m ore specific infor m ation on the phytosociology o f one o f the M C 's co m paratively little-know n vegetation types.
With a large, impoverished and rapidly increasing human population in M aputaland.there is an urgent need to improve people's living standards by utilising the region's readily available natural resources on a sustainable basis.The SNR. an area under the control o f the KwaZulu D epartment o f Nature Conservation, is currently being developed as part of a com m unity-driven ecotourism ini tiative aimed at local com munity upliftment and develop ment.Also included in the project are sections of the sur rounding Endlondlweni Com munal Area, which com prises both undeveloped land and a section previously used for com mercial plantations by the KwaZulu D epartment o f Forestry.The bordering Nguni Cattle Area, which is being used for the breeding o f pedigree Nguni cattle, will even tually be linked to form a single conservation unit that w ill be jointly managed by the Kw aZulu Department o f Nature Conservation and local com munities.K now ledge o f the vegetation o f a region form s the basis for understanding and m anaging its plant resources and other aspects o f the environm ent.One o f the main problem s facing the plant conservationist is the lack o f sound inform ation on w hich to base conservation strate gies.The present study aim s to provide ecological and floristic data for the w oody grasslands o f M aputaland by identifying, characterising and interpreting the m ajor vegetation units and their variations that occur in the SNR and surrounding areas.Floristically rich (high b io diversity/endem ism ) and interesting habitats w ill also be highlighted together with com m ent on som e o f the unusual structural features of the M aputaland grasslands.

G eology
G eologically the study area is very sim ilar to the rest o f the M aputaland and southern M ozam bican coastal plain.The principal stratigraphic units are unconsolidat ed Q uaternary grey aeolian sands and Q uaternary y el lowish redistributed sands, underlain by a m ulti-layered sedim entary sequence o f Q uaternary, more clayey co n solidated sands (Port D urnford Form ation) and Tertiary white sandy lim estone w ith basal conglom erate, all rest ing u n co n fo rm ab ly on an undulating im perm eable C retaceous siltstone floor w hich slopes dow nw ards to the east (H obday 1979; M aud 1980; South A frican C om m ittee for Stratigraphy 1980).

Topography and clim ate
The study area is characterised by undulating sand ridges (linear north-south-trending dunes) interspersed with depressions w hich may form pans or sw am ps due to the poor drainage and high w ater table o f the region.The highest dune in the study area reaches 98 m asl.The low est point lies at ± 76 m asl-a difference o f only 22 m.O f the few pans to be found in the study area, Gondetem be (± 50 ha2) is the largest.
M aputaland has a w arm to hot, hum id, subtropical cli m ate (Schultze 1982).Average annual hum idity is high, even in the drier inland parts o f the region.W inters are drier than the sum m ers, although rain is received throughout the year.Rainfall data are given in Table 1. with Phelendaba being the w eather station nearest to the study area (± 2.5 km ).T his station is com pared to S ihangw ana and N dum o (both from drier inland areas) as well as the m uch w etter Kosi Bay on the coast.O w ing to the undeveloped nature o f m uch o f M aputaland, there is a shortage o f certain environm ental data, such as tem perature.

H ydrology
A shallow w ater table (ground level to a depth o f 7 -1 5 m) exists w ithin the aeolian sands and its level varies according to ground w ater m ovem ents, topography and rainfall.The exact ground w ater levels (m ore specifical ly, the deeper ones), are difficult to determ ine and fluc tuate greatly over tim e.Values deeper than 2 m are.therefore, approxim ate.G round w ater is alm ost exclu sively replenished by rainw ater.T his shallow w ater table feeds all the m arshes and pans in the study area.A fter rains, quick fluctuations in local w ater table levels can be experienced, with an estim ated vertical seepage rate in the order o f 0 . 1 m /day (K ruger 1986).Fluctuations unre lated to the seaso n 's rains are ascribed to w ater m ove m ents through the sands, w ith K ruger (1986) reporting an average transm issivity o f 20 n r/d a y and a co-efficient of storativity of 1 x 10 '.The general flow direction is from w est to east, away from the w atershed w hich lies m ore centrally in M aputaland.

Soils
Soils are developed from the relatively hom ogeneous, grey, siliceous, aeolian sands.Soil types are lim ited, but soil patterns are intricate, though predictable, as a result o f the relationship betw een topography and w ater table levels (areas subjected to periods o f inundation).Soil nom enclature follow s the FAO soil classification system (FA O -U N ESC O 1974) andSoil C lassification W orking G roup (1991).M ost of the soils show signs of high lev els o f leaching (dystrophic) as well as being relatively acidic [approxim ate pH (w ater) 6.1].An unusual feature is the presence o f above average levels o f the m icronu trient Boron (G.M ann pers.com m .).
Three main soil types (form s) are present in the study area, nam ely dystric regosols (SA -N am ib), histosols (SA -C ham pagne) and hum ic gleysols.D ystric regosols are m oderate to w ell-drained acidic sands found in ele vated places such as dune crests and slopes.H istosols are sour organic soils with an organic rich A horizon thicker than 400 mm and are found in sw am py areas and pans.Humic gleysols are wet acidic sands with an abnorm al accum ulation o f organic m atter and are found in depres sions (areas with a high w ater table).The regosols cover m ost o f M aputaland and are Q uaternary sand deposits o f generally low fertility (W atkeys et al. 1993).

METHODS
Fifty 100 m : sam ple plots (10 x 10 m) were d istrib uted in a stratified m anner throughout the study area.As far as was possible, plots were equally distributed in the d ifferent p h y sio g rap h ical-p h y sio g n o m ically h o m o g e neous units, distinguished on the basis o f physical en v i ronm ent, physiognom y, as w ell as d o m in an t plant species com position and abundance.Sam pling was car ried out from January to M ay 1993.Scientific nam es o f plant taxa follow A rnold & De Wet (1993).The follow ing inform ation was recorded for each plot: 1. total floristic com position and cover-abundance value for each species, according to the B raun-B lanquet coverabundance scale as described by M ueller  M a putaland C entre endem ics and near-endem ics m entioned in the text are based m ainly on the work of Van Wyk (1996).The term 'end em ic' refers to a taxon lim ited in its range to a restricted geographical area, or a particular substrate; w hen a taxon is also m arginally pre sent elsew here (som etim es in the form of distant satellite p o p u latio n s), it is referred to as a 'near-en d em ic' (M atthew s et al. 1993).

Classification o f vegetation
A classification o f the vegetation is presented in a phytosociological table (Table 2).Five m ajor plant co m m unities were identified.These com m unities closely reflect differences in w ater table depth (soil m oisture) w hich is directly correlated with the regional topography (Figure 2).Soil type also plays an im portant role, partic ularly in the case of hygrophilous com m unities.The num ber of species per plot (= species richness) ranges from a m inim um o f seven to a m axim um o f 60, with an average o f 25.H ierarchical classification o f the vegeta tion reinforces the correlation betw een habitat and co m m unities (Figures 2 & 3).The distribution am ong plant com m unities o f M C endem ic/near-endem ic taxa, is list ed in Table 3.A sum m ary of selected com m unity attrib utes is supplied in Table 5

I. A rtabotrys m onteiroae-D ialium schlechteri forest of deep w ater table areas (relatively dry sands)
Bothalia 29,1 (1999) Found on dune crests and slopes, this com m unity is associated w ith grey dystric regosols in places w ith the deepest w ater table.In four o f the five investigated plots the w ater table could not be reached w ith the soil auger (>2 m deep).B iom ass m easurem ents were not undertak en in this com m unity as the field layer in m ost instances was m ainly short shrubs w ith very few grasses and forbs.
Characterised by species group A (Table 2) with a high constancy in this com m unity, is a character taxon o f species group C. Species group C links the p re sent com m unity with the w oodland com m unity (2.1).On average 40 species per plot were recorded.N o less than 12 MC endem ic/near-endem ic species were found in these forests, thus m aking this one o f the richest co m m unities for such taxa (Table 3).

Syzygium cordatum (suffruticose form) Myrtaceae
X X T his com m unity is structurally a forest, the canopy varying from as low as 5 m up to about 20 m.The species com position is unique and other authors have referred to this com m unity as Sand Forest (e.g.M oll 1977, 1980; M oll & W hite 1978; W ard 1981), typical elem ents o f w hich include Dialium schlechteri, Pteleopsis m yrtifolia, Ptaeroxylon obliquum.Croton pseudopu lchellus and Ochna barbosae.Pendulous lichens ( Usnea spp.) are particularly abundant in the protruding crow ns o f m any o f the trees.Sand forests in the study area lie at the eastern lim its o f their distribution and are floristically notably im poverished com pared w ith their inland counterparts (unpublished data).B ecause of their proxim ity to the coast, they also contain species m ore characteristic o f coastal dune forest, such as M im usops caffra, A cacia kraussiana and the understorey herb lsoglossa woodii.

T hem edeto-S alacietu m M. M vre (1964); w oody grasslands o f relatively dry sandy areas
An association restricted to grey dystric regosols on dune crests, slopes and relative high-lying level plains.Average w ater table depth was ± 1.8 m. w hereas in many o f the other sam pling sites it could not be reached by the soil auger (>2 m).A verage biom ass was 2091.1 k g .h a 1.
T his asso cia tio n is subdivided into tw o su b a sso cia tions on the basis o f structure and floristic co m p o si tion, thus in d icatin g a strong correlation betw een the tall w ood lan d s and the w oody (suffru tex -rich ) g rass lands.

Catunaregam spin osa-A cacia burkei w oodland and w oodland areas o f forest edges
A subassociation o f grey dystric regosols on crests and slopes o f dunes.A verage w ater table depth was ± 1.8 m, w ith tw o sam ples deeper than the soil auger length (>2 m).A verage biom ass w as 2291.5 k g .h a 1, a value slightly higher than those of related grassland com m uni ties and attributable to a slightly higher woody com po nent.The last four species belong to group E, w hich represents the species shared betw een this and the w oodland com m u n i ty (2.1).T his again underscores the relationship betw een the w oodlands and the w oody grasslands.On average 36 species per plot were recorded, with 12 MC endem ic or near-endem ic species (Table 3).

Urelytrum agropyroides-Trichoneura grandiglumis w oody grassland o f relative shallow er w ater table areas
A variant associated with grey dystric regosols on dune crests, slopes and sandy plains.Average w ater table depth was ± 1.8 m, w ith only one sam ple extending deeper than the soil auger length (>2 m).Average b io mass for this subcom m unity was 1985.8k g .h a 1.
A lthough lacking a characteristic species group, this subcom m unity is characterised by the absence o f species group D (Table 2).C onsistent species include the grass es Urelytrum agropyroides and Trichoneura grandiglu mis.O ther species w hich show a strong affinity to this com m unity are the grasses Themeda triandra and A mtida stipitata, sedge Abildgaardia hygrophila, geoxylic rhizom atous suffrutex Eugenia albanensis, palm H \phaene coriacea and the forb Cham aecrista plum osa.This forb com es from species group H. a group repre senting the species shared betw een the woody grassland com m unities (2.2) and the hygrophilous grasslands (4).T his connection signifies a floristic relationship between the 'w e tte r' w oody grasslands on dystric regosols (this variant) and the truly h y g ro p h ilo u s grasslan d s on C ham pagne soils.On average 32 species per plot were recorded, w ith 13 M C endem ic or near-endem ic species (Table 3), the highest num ber for all the investigated plant com m unities.

E ragrostis lappu la-H elich rysopsis septentrionale hy
grophilous grassland on hum ic gleysols (inter-dune depressions) T his plant com m unity is found in interdune depres sions and other low -lying areas, and is associated with grey to dark grey sands (hum ic gleysols).These soils show signs o f slightly higher organic m atter accum ula tion than those o f the preceding com m unities.Average w ater table depth was ± 1.2 m.Average biom ass was 1758.3 k g .ha'1, the low est value o f all the plant com m u nities and quite obvious to the naked eye in the field.
C haracterised by species group I (Table 2), diagnostic species include the forbs H elichrysopsis septentrionale, Wahlenbergia abyssin ica, Vahlia capensis, Hypericum lalandii and the hem i-parasite Striga junodii.Species show ing a strong affinity to this com m unity are the sedges Cyperus tenax and Cyperus obtusiflorus and the forb D esm odium dregeanum.The grass Eragrostis lappula attains high cover values.On average 19 species per plot were recorded, w ith six MC endem ic or near-endem ic species (Table 3).This com m unity shows relationships with drier grass lands (species group H) as well as with w etter (hygro philous) com m unities (species group J) on histosols.

Ischaemum fasciculatum -E ragrostis mamoena hygro philous grasslands on C ham pagne soils
Found on wet, seasonally w aterlogged, bottom lands and associated w ith dark grey to black histosols, with high levels o f organic m atter com position com pared to the substrates ol the preceding com m unities.Average w ater table depth was ± 0.65 m.Average biom ass was 3454.7 k g .h a 1.
Characterised by species group M (Table 2).diagnostic species include the grasses Ischaemum fasciculatum and Eragrostis inamoena.The form er species attains some of the highest cover values in places and is very prom inent.On average, 13 species per plot were recorded.
This com m unity show s relationships w ith the hygro philous grassland com m unity on N am ib soils (species group N).It is subdivided into three subcom m unities on the basis o f the prom inence o f Ischaemum fasciculatum .a species reflecting periods o f inundation.

Monocymbium ceresiiforme-Eragrostis lappula grass land o f areas not being as regularly inundated by w ater as the Ischaemum fa scic u la tu m -C yp e ru s sphaerospermus grassland
A plant subcom m unity o f bottom lands, associated with dark grey to black soils (mixture of histosol and humic gleysols very rich in organic matter) and usually bordering on com m unities 4.2 and 4.3 (Figure 3).Average water table depth was ± 0.7 m. with three sam ples lying at a depth of 1 m.Average biomass was 2419.5 k g .h a 1.
Characterised by species group K (Table 2), the grass M onocymbium ceresiiform e is the m ost d iagnostic species.A nother diagnostic grass (>50% constancy) is the small-growing Panicum genuflexion.O ther prom inent species are the sedges Cyperus obtusiflorus and C. sphaero sperm us and the grasses S porobolus su btilis and Eragrostis lappula.The last two species are from group J and indicate a strong relationship between this subcom munity and the hygrophilous grasslands on humic gleysols (3).On average 14 species per plot were recorded, with two M C endem ic or near-endem ic species (Table 3).

Ischaemum fasciculatum -C yperus sphaerosperm us grassland o f depressions scattered throughout the grassland areas or directly bordering on m arshes/ pans
A plant subcom m unity of bottom lands, associated with dark grey to black histosols rich in organic m atter and usu ally bordering on pans and marshes.Average w ater table depth was ± 0.5 m. with only one sample reaching a depth of 1 m.Average biom ass was 2986.8 k g .h a '.C haracterised by species group L (Table 2).the m ost prom inent diagnostic species is the tall grass Sorghastrum stipoides.O ther diagnostic species (>50% co n stancy) include the sedge Fuirena pu bescen s, forbs Lobelia fla ccid a and Polygala capillaris and the grass Andropogon eucomus and.although o f lesser constancy, the grasses Ischaemum fasciculatum and E ragrostis inamoena.A lso characteristic are the sedge Cyperus sphaerospermus and the prostrate forbs Centella a siatica and Desmodium dregeanum.The latter three species are from group N, thus show ing a relationship betw een this subcom m unity and other hygrophilous grasslands (3 & 4.1).On average 15 species per plot w ere recorded, w ith no MC endem ic/near-endem ic species (Table 3).

lschaemumfasciculatum grassland representing low species diversity but relatively high biomass, dominated by lschaemum fasciculatum
Found in bottom lands on dark grey to black organicrich soils (histosols).Average water table depth was ± 0.7 m (with one sample reaching 1 m), and is thus slight ly deeper than that of the preceding two subcommunities.Average biomass was 4957.7 kg.ha*1, the highest value of all plant communities in the study area, which is remark able considering that none of the grasses making up the major proportion o f the biomass are particularly tall plants.The dominant grass, lschaemum fasciculatum, only reaches a height of ± 400 mm. 2), the most prominent taxon is the grass lschaemum fasciculatum , a species which contributes significantly to the high average biomass.There are few other prominent species, the most consistent perhaps being Cyperus natalensis, a sedge from group P, which is common to all communities.Other species showing >50% constancy include the forb Desmodium dregeanum and the grasses Hemarthria altissima and Acroceras mac rum.The last two species are diagnostic for commu nity 5, the very wet one.On average nine species per plot were recorded, the lowest number among the stud ied plant communities, with only one MC endemic/nearendemic (Table 3).

Leersia hexandra-Hemarthria altissima grassland o f marsh/pan areas (water table on surface during most years)
A plant community of bottom lands and associated with dark grey to black histosols very rich in organic matter.Average water table depth was ± 0.1 m with only two of the samples not having the water table present on the surface.Average biomass was 4584.4 kg.ha ', making this the community with the second highest biomass.Not all plots were sampled for biomass because of the pres ence of surface water.2), diagnos tic species include the grasses Leersia hexandra, Hemarthria altissima and Acroceras macrum, the latter attaining constantly relative high cover values.Other diagnostic members (>50% constancy) are the sedges Pycreus polystachyos, Fuirena obcordata, Eleocharis dulcis and Cyperus natalensis, the latter being common to all communities.Diagnostic forbs include Oldenlandia cephalotes and Hydrocotyle bonariensis.Many of these species are also sporadically present in other hygrophilous grasslands, notably in localised patches of wetter conditions.On average 13 species per plot were recorded, with only one MC endemic or near-endemic species (Table 3).

Ordination
Distribution of the sample plots (relevés) along the first and second axes of ordination is given in the form of a scatter diagram (Figure 4).A third axis of ordination contributes little to the interpretation of the communities and will therefore not be considered further.

Figure 4 illustrates a water table depth (moisture) gra dient along the horizontal axis, with the wetter habitats to the right and the drier habitats to the left. There is only a slight discontinuity in the distribution of the sample plots representing the deep (community 1) and the shallow water table areas (community 2)
. However, all plant communities are more or less restricted to specific areas of the diagram.

A distinct discontinuity exists between the communi ties of the dystric regosols (Namib soil) and the histosols (Champagne soil), with the intermediate communities being placed centrally in the diagram. Subcommunities of the areas with a deep (1) and shallow (2) water table
show no separation, thus supporting their treatment as closely related subcommunities.This pattern of one community merging with another along a water table depth gradient can be clearly seen in the field.There can be no doubt that the level of the water table, either direct ly, or indirectly through its role in soil formation, plays the deciding role in defining plant communities on the coastal plain of Maputaland.

Plant communities
Twelve ecologically interpretable plant communities have been distinguished and described.This information can be used in reserve and area management, land-use planning, extrapolation to other parts of Maputaland as well as in further ecological and floristic studies.The results of the ordination not only confirm the classifica tion, but also give an indication of floristic and associated habitat gradients.Dynamics of water in the landscape, in this case the water table, clearly control the structure of the community at the first physiognomic level.This is a common phenomenon in savanna areas (Solbrig 1993).
All the communities are easily distinguishable in the field on the basis of growth form, general species com position and character species, despite the gradual environ mental gradients.

Ochna natalitia
X X x = rare; x x = average; x x x = abundant; blank = absent.
Themedo-Salacietum Parinarietosum, two syntaxa which are easily recognisable throughout the MC.We have, how ever, considered it prem ature to formally describe the other syntaxa.M ore phytosociological studies in other areas o f the M aputaland C oastal plain are needed to correlate results.

G eoxylic suffrutices
An outstanding feature of some o f the Maputaland grasslands (e.g. the Themedeto-Salacietum woody grass lands o f the present study) is the abundance of geoxylic (often rhizom atous) suffrutices (e.g.Henkel et al. 1936;M yre 1964, 1971;Moll & W hite 1978).These dw arf woody plants can be com pared with extremely stunted trees, a fact which led W hite (1976) to refer to them as the 'underground forests o f A frica'.Furthermore, in this type o f grassland the phytom ass o f the suffrutices greatly exceeds that of the grasses (W hite 1976; unpublished data).
The geoxylic suffruticose habit is characterised by annual or short-lived w oody shoots sprouting from m as sive or extensive woody, perennial, underground axes.T his rather uncom m on growth form appears to be best developed in A frica, with the greatest concentration in the Z am bezian Region (W hite 1976. 1983).Despite W hite's (1976) statem ent that relatively few suffruticose sp ecies arc co n fin ed to the T o n galand-P ond o lan d Regional M osaic, plants with this growth form are abun dant in the study area w here they are alm ost exclusively confined to the Themedeto-Salacietum woody grass lands (Table 4).
W hite (1976) suggests that in Africa the geoxylic suf frutex probably originated as a response to unfavourable edaphic conditions, notably in strongly oligotrophic, sea sonally waterlogged sandy soils in regions o f extremely low relief.He argues that this habit is not primarily an adaptation to fire or frost, as has been supposed by Burtt Davy (1922), whilst acknow ledging that for some species, at least occasionally, fire is necessary for vigorous growth.
The present study confirm s the observation that geoxylic suffrutices arc prevalent in areas o f sandy soil and relatively low relief (Table 5).H ow ever, doubt is cast on the validity o f W h ite's perhaps too sim plistic claim that seasonally w aterlogged soil is the prim ary determ inant for the presence o f this growth form.In the study area the distribution o f geoxylic suffrutices rather reflects the depth o f the w ater table, with these plants being m ost abundant in sites where it is high, yet still below 1.8 m.These sites com prise m ainly those relative ly high-lying, w ell-drained areas such as dune crests and slopes, the surface soils o f w hich are never w aterlogged.In fact, geoxylic suffrutices are noticeably absent from inter-dune depressions, the only areas w hich are clearly seasonally w aterlogged.
Fire m ight have played a m ore significant role in the evolution o f the suffruticose habit than suggested by W hite (1976).In the absence o f fire, the above-ground shoots of the suffrutices in frost-free areas (such as the study area) becom e less floriferous, m oribund, and may even die back.Flow ering and the sprouting o f new shoots in suffrutices are considerably enhanced by the frequent annual burning o f the M aputaland grasslands.

G rassland o r savanna ?
Physiognom ically, typical grassland is characterised by strong dom inance o f hem icryptophytes of the Poaceae.Savanna, on the other hand, has been defined as a vegeta tion type com prising an herbaceous, usually gram inoid, layer with an upper layer o f woody plants of which the canopy cover does not exceed 75% (Edw ards 1983).A non-quantitative, m ore functional savanna definition is that it is a tropical vegetation type in w hich ecological processes, such as prim ary production, hydrology and nutrient cycling, are strongly influenced by both woody plants and grasses, and only weakly influenced by plants o f other growth form s (Scholes & W alker 1993).
Due to the abundance o f geoxylic suffrutices, it is som ew hat o f a m isnom er to refer to these com m unities m erely as 'grassland*.We suggest 'woody grassland' as a m ore appropriate descriptive term.In a sense these woody grasslands imitate a savanna in which the tree stra tum has been reduced to alm ost the sam e level as that o f the gram inoids and with m any o f the dw arf trees actually com parable to hem icryptophytes, particularly if fire is given its due recognition as a natural factor.Rutherford & W estfall's (1994) inclusion o f the M aputaland grasslands under their Savanna B iom e m ight therefore be more appropriate than would appear at first glance.Detailed com parative studies on the ecology o f these unusual woody grasslands, in relation to conventional grassland and savanna, would be m ost instructive.

Fire
G row th in the coastal grasslands o f M aputaland is not distinctly seasonal and herbage production is high, despite the infertile soils.R egular fires are a natural phe nom enon in these grasslands, although today m ost are caused by hum ans.The sam e patch of grassland may be burned up to three tim es a year, w ith at least one fire a year being the norm.T here can be no doubt that fire is an essential factor in m aintaining the w oody grasslands o f coastal M aputaland.M argins o f sand forest exposed to fire often acquire w oodland elem ents and w oodland character species.This, how ever, appears to be a tem po rary stage tow ards sand forest recovery.
An increase in burning frequency ow ing to an expand ing hum an population over the last few decades has led to the com m onsense assum ption that tree-dom inated vege tation types in the study area must be decreasing.However, a com parison o f aerial photographs of the area taken in 1942, 1975 and 1991 indicates the contrary.In most cases the patches o f w oodland and sand forest have either increased in size (albeit slightly) or becam e more densely w ooded.T hese forests and w oodlands also turned out to he very stable features-all being strictly confined to the sam e sites over at least the past 50 years, although size and Holistic com position m ight have changed.
It is hypothesised that the observed increase in size and the thickening-up o f the forest/w oodland patches could have been caused by changes in w ater table level A drop in w ater table over the long term w ould be co n ducive tow ards the creation o f habitats more suitable for the developm ent of w oodland, thicket and eventually forest.The continued afforestation o f the w oody grass lands with pines, eucalypts, cashew nuts and other alien trees will, alm ost certainly, bring about such a drop in the level of the w ater table.T his could result in a dram atic increase in the woody com ponent of the vegetation o f the region.W oody grasslands might, in future, have to be increasingly m aintained by fire in order to keep then in their current 'underground savanna/forest' state.

Primary' or secondary grassland?
Are the extensive coastal grasslands o f M aputaland prim ary or secondary?Rutherford & W estfall (1994) do not include them in their G rassland Biome.but consider them part o f the Savanna Biome.They differentiate the true G rassland Biom e clim atically from the S avanna Biome in term s of m inim um w inter tem perature in co n junction with m oisture levels, thus restricting it m ainly to the grasslands o f the high central plateau and eastern parts o f the G reat E scarpm ent of South Africa.
Moll & W hite (1978) and W hite (1983) distinguish tw o broad types o f g rasslan d in the T o ngaland-Pondoland Regional M osaic, nam ely edaphically co n trolled grassland associated w ith scattered palm s on badly drained sandy soils and secondary fire-m aintained grassland that has replaced anthropogenically destroyed coastal forest.On deeper soils along the coast, grassland has been considered a phase in the prim ary succession to coastal dune forest, a fire-subclim ax.In the absence o f fire, succession quickly proceeds from grassland to dune scrub and forest (W eisser 1978).T his trend tow ards for est developm ent dim inishes inland and in the study area, w hich falls within the edaphic grasslands o f W hite (1983).grassland appears to be a more stable feature.
A lthough the frequency of fire in the coastal g rass lands of M aputaland has obviously increased due to hum an activities, floristic and m orphological evidence clearly indicate that they have been edaphically co n trolled and/or fire-m aintained for a very long tim e.The high incidence o f endem ics, several o f w hich developed an obligate geoxylic suffruticose habit (Tables 3 & 4), signifies a long evolutionary history for this particular vegetation type in the region.
As in the case o f the A fro m o n tan e g rasslan d s (M atthew s et al. 1993).the presence o f coastal g rass lands in M aputaland is not the result of the relatively recent anthropogenic destruction o f savanna or forest, although the presence o f hum ans may have led to their expansion or contraction in certain parts.We contend rather that these grasslands are essentially prim ary in nature and not secondary in the sense o f being ■unnatu rally' degraded forest or savanna.In fact, the association betw een f ire and these grasslands m ust be as old as the grasslands them selves, thus m aking fire a regular feature o f the environm ent.E xtensive afforestation w ith exotic trees, so-called conservation m easures to curtail fire, w o o d cu ttin g , grazing and shifting cu ltiv atio n have already resulted in the all but total disappearance o f grassland in m any parts.B iodiversity has dim inished accordingly, particularly in the southern coastal region o f M aputaland (W eisser 1978).

Endemics
A bout 2 500 species (but probably more) o f vascular plants occur in the MC.O f these at least 230 species or infraspecific taxa and three genera are endem ic or near endem ic to the region (Van Wyk 1996; unpublished data).T hirty one o f these M C endem ics were recorded in the study area (Table 3), but the actual num ber present is expected to be slightly higher due to incom plete sam pling.M ost endem ics w ere associated w ith only two of the five m ajor plant com m unities.By far the m ajority of plant endem ics are confined to non-hygrophilous grass land, a vegetation type hitherto usually considered sec ondary (anthropogenic) in origin (e.g.Henkel et al. 1936; W eisser 1978; W hite 1983).The observed low species diversity and paucity o f regional endem ics in hygrophilous grassland com m unities is a com m on phe nom enon throughout southern Africa.M C endem ics recorded in the study area (and even more so for the centre as a w hole) represent a wide spec trum o f grow th form s, including trees, shrubs, suffrutices, lianas, forbs, geophytes and annual herbs (Van Wyk 1996).M ost noticeable am ong the more than 130 W olkberg C entre endem ics is the com plete lack o f annu als and large trees and the fact that nearly all these species are confined to grassland (M atthew s et al. 1993; unpublished data).There is a conspicuous lack o f local endem ism in the associated patches o f A from ontane for est.In contrast, M C endem ics are well represented in both grassland and sand forest, with the latter, on a regional basis, being perhaps the single richest com m u nity in M C plant endem ics.It is hypothesised that one of the reasons for this m arked difference in growth form and for the vegetation-type partitioning o f endem ics betw een the two centres of endem ism is the relative youthful age o f the sandy M aputaland coastal plain (Q uaternary) and its associated plant com m unities.

Threats and conservation
Conservation efforts in the M C have hitherto centred mainly around areas containing species o f large gam e.The vegetation o f nearly all existing nature reserves is dom inated by various types o f savanna.A notable ex cep tion is the Tembe Elephant Reserve w hich contains wellpreserved stands o f endem ic-rich sand forest.The grass lands in the region are particularly poorly conserved and managed.
A fforestation is currently the m ost serious threat to biodiversity on the coastal plan of the MC.U ncontrolled cattle grazing and random , uncontrolled fire certainly effects species com position and diversity, but probably do not change the structure of the grasslands.The plant ing o f alien trees over large tracts o f grassland, on the other hand, not only destroys the grassland habitat but is also expected to affect the hydrology o f the region nega tively.Further afforestation by the tim ber industry and private individuals is expected.The establishm ent of other extensive m onocultures, for exam ple cashew nuts, sugar cane and coconut palm s, also threatens the grass lands o f this relatively unspoiled part o f A frica.W ith so little grassland form ally conserved, the SNR.despite its small size, fulfils a crucial role in ensuring the preserva tion o f the endem ic-rich flora and fauna o f the MC, one of the w orld's m ost unique centres o f endem ism .

A C K N O W L ED G E M EN T S
Our thanks to M artie D ednam , H.G.W.J. Schw eikerdt H erbarium , U niversity o f Pretoria for the processing of plant specim ens and to the C urator and staff o f the Natal Herbarium .Durban, for assistance with plant identifica tion.We thank Peter G oodm an, form erly from M kuze G am e Reserve, for valuable discussions and constructive criticism .Special thanks to Ian Felton, G raham M ann.Jam es M itchell and M iranda D cutschlánder for their help and support.Jerem y H ollm an critically read and im proved the m anuscript.This research was partially funded by the K waZulu D epartm ent of N ature C on serv a tion and the U niversity o f Pretoria.
FIGURE 1.-Map showing the Maputaland Centre o f Endemism and the location o f the study area.
FIGURE 3.-Schematic representation of the distribution of the different plant communities in the Sileza Nature Reserve and surrounding com munity areas.Dotted lines indicate overlapping distributions. 1 .Artabotrys monteiroae-Dialium schlechteri forest in deep water table areas (relatively dry sands).2.1.Catunaregam spinosa-Acacia burkei woodland and woodland areas o f forest edges.2.2.Themedo-Salacietum Parinarietosum; woody grasslands in deep water table areas (relatively dry sands), e.g.dune crests and slopes.3. Eragrostis lappula-Helichrysopsis septentrionale hygrophilous grasslands on humic gleysols (inter-dune depressions).4. Ischaemumfasciculatum-Eragrostis inamoena hygrophilous grasslands o f Champagne soils. 5. Leersia hexandra-Hemarthria altissima grassland of marsh/pan areas (water Four M C endem ic/near-endem ic bird species, Neergaard 's sunbird (Nectarinia neergaardi), R udd's apalis (Apalis ruddi), pinkthroated tw inspot (Hypargos margaritatus) and W oodw ards' batis (Batisfratum) are found m ostly in sand forest.A subspecies o f the pinkthroated longclaw (Macronyx ameliae ameliae), a rare M C e n dem ic bird, is m ainly associated with the Ischaemum fasciculatum-Eragrostis inamoena hygrophilous grass lands.The present study enables com parison between the coastal plain grasslands o f the M C and the high-altitude A from ontane grasslands o f the W olkberg Centre, a botanical centre o f endem ism along the northeastern Transvaal E scarpm ent (M atthew s et al. 1993).E nviron m ental factors associated with the m ontane plant co m m unities are m diverse and include often com plex interactions betw een lithology, soil type and depth, topography, precipitation (rain and mist), altitude, rocki ness, slope and fire regim e (M atthew s et al. 1992a, b; 1994).A m uch sim pler situation exists in the M C with the key determ inants being the interconnected effects of w ater table, soil type and topography, upon w hich fire has been superim posed.The study on the W olkberg Centre, how ever, covered a considerably greater area.Casual observations and com parison with other studies (notably M yre 1964, 1971) have nevertheless show n that the plant com m unities (and thus associated environm en tal factors) o f the present study extend over m ost o f the M aputaland coastal plain.
K. 1979.G eo lo g ica l evolution and geom orphology o f the Zululand coastal plain.In B.R. A llanson.Lake Sibaya: 1-20.M onographiae Biologicae 36 Junk, The Hague KRUGER, G .P 1986.A report on the results o f a hydrogeological and geom orphological survey.In An investigation into the feasibili ty and economy o f rice cultivation on the Ngwavuma coastal plain, Northern Natal.U np u b lish ed report.

Table 2
), prom i nent diagnostic species include the shrubs Catunaregam spinosa and Vangueria infausta, shrubs/trees Acacia burkei and Sapium integerrim um , grass Panicum m axi mum and the forb Corchorus junodii.Termirmlia sericea is consistent as well as achieving its highest abundance.O ther prom inent species (>50% constancy) are the sh ru b s/trees A

Variation in water table level, the over riding environmental factor, is much less easily distin guished and is best reflected by changes in plant com munity and the presence of different soil types. Water table levels fluctuate seasonally in accordance with rain fall as well as unseasonally because of non-rainfall related ground water movements (Kruger 1986). It is possible that water table levels rarely falls below 15 m. With the exception of the excellent broad-scale classi fication of the coastal grasslands of southern Mozam bique (Myre 1964, 1971), the present study is the first attempt at a more detailed, larger scale phytosociological classification of some of the Maputaland coastal plain grasslands. A correlation between the plant communities of the present study and the various syntaxa described by Myre (1964, 1971) has been attempted. Here we recog nise his formally described Themedeto-Salacietum and D eep w a ter table (D ry) < > S h allow w a ter table (W et) < B R egosols soils D H istosols & H um ic G leysols soils . --------. : g 19 f 24 is ! ; *
FIGURE 4.-Ordination o f the vegetation o f the Sileza Nature Reserve and surrounding community areas.A, Artabotrys numteiroae-Dialium schlechteri forest o f deep watertable areas (relatively dry sands) B, Catunaregam spirutsa-Acacia burkei woodland and woodland areas on forest edges.C, Themedo-Salacietum Parinarietosum; woody grasslands of deep water table areas (relatively dry sands), e.g.dune crests and slopes.D, Monocymbium ceresiiforme-Eragmstis lappula grassland o f areas not as regularly inundated by water as F E, Eragmstis lappula-Helichrysopsis septentrionale hygrophilous grasslands on humic gleysols (inter-dune depressions F, Ischaemum fasciculatum-Cyperus sphaerospermus grassland o f depressions scattered throughout the grassland areas or directly bordering on marshes/pans or/as well as the Ischaenum fasciculatum grassland subcommunity representing low species diversity but relatively high bio mass.G, Leersia hexandra-Hemarthria altissima grassland of marsh/pan areas (water table on surface during most years).

T A BLE 5 .-Environm ental factors associated with the different plant com m unities. Other selected attributes are also presented Plant com m unity number
* = estim ates only: ** = not determ ined