A floristic classification of the vegetation of a forest-savanna boundary in southeastern Zimbabwe

The vegetation of Chirinda Forest boundary was classified into eight types using Two-way Indicator Species Analysis (TWINSPAN) and Detrended Correspondence Analysis (DCA). The moist forest comprises three types: Strychnos mellodora-Chrysophyllum gonmgosanum Forest on deep dolerite soils; Chrysophyllum gorungosanum-Myrianthus holstii Forest on shallow dolerite soils; and Teclea iiobilis-Ehretia cymosa Forest on drier, but deep dolerite soils. The non-forest vegetation comprises five types: Themeda triandra Grassland on shallow dolerite soils; Psidium guajava Bushland on sandstone; Bridelia micrantha-Harungana madagascariensis Mixed Woodland not restricted to any one particular soil type; Acacia karrooHeteropyxis dehniae Woodland on shallow soils derived from sandstone but sometimes on dolerite; and Julbemardia globiflora-Brachystegia spiciformis (Miombo) Woodland on sandstone.


INTRODUCTION
Forest patches occur in many tropical countries, often embedded in a matrix of non-forest vegetation types, forming unique units along their boundaries.Ranney et al. (1981) emphasized the importance of forest fringes in the structure and dynamics of forest patches.The bound ary creates microclimatic gradients that result in differ ences in environment between the forest interior and the outside, non-forest areas.In this context, the influence of soil moisture and other edaphic changes at the forest-savanna boundary is particularly important (Furley 1992;Hopkins 1992).
The ecotone between forest and non-forest areas often has high plant densities and diversity, and represents the juxtaposition of two contrasting habitats.Along the eco tone there is generally a high incidence of wind and ani mal activity, especially mammals and frugivorous birds.As a result, the predation of seed and fruit and the op portunity for their dispersal are both enhanced.
In Zimbabwe, forest patches with sharp boundaries be tween them and non-forest vegetation are common along the eastern highlands where three main centres of moist forest development can be identified, namely the Nyanga, the Vumba and the Chimanimani massifs.Many other smaller patches of forest occur elsewhere, including Chir inda, the area of this study.The factors responsible for these boundaries vary from place to place, but fire, altitude and edaphic factors are among the most important.Crook (1952) and Phipps & Goodier (1962) discussed the vege tation patterns of the Chimanimani Mountains, with em phasis on the determ inants of the forest, woodland, grassland and Ericaceous scrub.Similar situations can also be encountered in the Vumba, Nyanga and Chirinda areas.
Inventories of the vegetation of Chirinda Forest have been compiled by, among others, Goldsmith (1976) and Muller (1991).Much of this work focused mainly on the moist forest proper, with very little attention paid to the surrounding non-forest vegetation.Mapaure (1993) dis cussed the factors influencing the vegetation of the forest boundary whilst Timberlake et al. (1994) described the composition of the forest and some ecological factors af fecting the vegetation.A list of flowering plants and ferns found in and around the forest has been compiled by Drummond & Mapaure (1994).
The main objective of this study was to inventory the woody vegetation of Chirinda Forest boundary, including adjacent bushland.grassland and woodland areas and to produce a classification and description of the vegetation types.

STUDY AREA
The study was carried out in Chirinda Forest, situated near Mount Selinda Mission (20° 25'S, 32° 43'E) in south east Zimbabwe (Figure 1).The forest represents the south ern end of moist forest distribution in Zimbabwe and is the best preserved example of medium altitude moist for est (Muller 1991).It is surrounded by pine plantations, Mission settlements, Chako Business Centre, and com mercial farmlands.It is a gazetted Forest Land managed by the Forestry Commission, covering 947 ha, of which 606 ha is moist evergreen forest.The forest lies on two broad highlands rising from 1 076 m to 1 250 m in alti tude.These form two crests, essentially subdividing the forest into two, a northern and a southern section with a narrow forested saddle in between.
The geology of the area comprises red and purple phyllitic mudstones interbedded with pale fine-to mediumgrained feldspathic sandstones of the Upper Argillaceous series of the Precambrian Umkondo System (Watson 1969).Fine-grained dolerite sills have intruded into these sandstones, forming a cap over much of the higher ground.The moist forest is generally found on dolerite soils.
The area receives an average rainfall of 1 470 mm per annum (Anon. 1977).Average winter and summer tem peratures are 14°C and 20°C, respectively (Sayce 1987).The prevailing winds are southeasterly.

METHODS
An interpretation of 1:12 500, black-and-white aerial photographs of 1987 was carried out and stratification of the vegetation around the forest was done, based on the textural differences on the photographs.A map was pro duced based on both the interpretation of the aerial pho tographs, which aided in the marking of the boundaries of the vegetation types, and field sampling.
Ten belt transects (two through each of the five vege tation units apparent on the aerial photographs), each 200 m long and 20 m wide, were established (Figure 2A-J) so that half the distance was in the forest and the other half in the non-forest vegetation.The edge of the forest was identified by either an obvious noticeable change in species composition or by a difference in vegetation struc ture, or both.Each belt transect was subdivided into ten 20 x 20 m contiguous quadrats, five on either side of the forest boundary, resulting in a total of 100 quadrats.The appropriateness of this quadrat size was confirmed by a species-area curve.In each quadrat, the woody species were identified and assigned to height classes as follows: seedlings, saplings (< 0.5 m), understorey (0.5-3.0 m), subcanopy trees (3-10 m), and canopy trees (>10 m).An overall cover-abundance value for each woody species in each quadrat was assessed following the Braun-Blanquet scale (Mueller-Dombois & Ellenberg 1974).Climbers were rated sepa rately according to the number of stems encountered in the stands rather than for a cover value.
Soil depth was determined by augering six quadrats per transect, three on either side of the forest edge.The soils were considered deep if no rocks were encountered down to a depth of 80 cm.The geology was determined from any exposed bedrock and by reference to the geo logical map of the area.

Data analysis
Vegetation data were analysed using Two Way Indica tor Species Analysis (TWINSPAN) (Hill 1979) and De trended Correspondence Analysis (DCA) (Gauch 1982).TWINSPAN was applied on the full species data set, con sisting of 261 plant species belonging to 204 genera and 76 families and DCA was applied on the sixty quadrats from which soil data were collected.Rare species were downweighted and for pseudospecies cut-levels of 0, 1 ,3 and 5 were used in the application of TWINSPAN.Coverabundance values of species were used in both analyses.Minor refinements were done to the TWINSPAN output (Table 1) to improve its clarity, especially the removal, after analysis, of the species which occurred three times or less (except indicator species) and lianes and vines.

Vegetation classification
Eight vegetation types were identified (Figures 3 & 4) on the basis of the TWINSPAN and DCA analyses.The primary division of the stands by TWINSPAN separated out the moist forest stands from non-forest (woodland and grassland) stands.The second level of division separated out forest stands in the northwestern part of the forest from the rest of the forest stands.These forest stands (Type III) are situated close to the miombo stands, on the drier side of the highland.The third level of division fur ther subdivided the remaining forest stands into a group which occurred in the eastern to southeastern part of the forest (mainly transects E, F, G and H) (Type II).The other group (Type I) consists mostly of forest stands from transects A, B, C and D. The forest stands, therefore, fall into three main types.
Among the non-forest vegetation stands, the second level of division separated out grassland stands, which oc curred in the southeastern part of the forest (Type VII), from the rest of the non-forest vegetation.The remaining non-forest stands were divided irfto a further two groups at the third level of division.One type (Type IV), which consists of a wide mixture of stands from several transects but close to the apparent forest edge, was separated from the rest of the remaining non-forest vegetation.The re maining stands were further divided into two, one of which comprises two types.Type VI consisted of stands mainly from an area close to Chako Business Centre (the sandstone enclave) and stands from the northwestern side forming Types V and VIII.Further subdivision of these types was considered unnecessary because the number of stands in each group was too small to justify the subdi vision.

O rdinatio n
The first DCA axis, which accounted for 63% of the observed variation, separated the forest and adjacent non forest vegetation types, with a wide gap of up to one DCA unit between them (Figure 4).In general, stands close to the forest edge, from either side, were closer to the centre of Axis 1 than stands further from the forest edge.Nev ertheless, the stands lying adjacent to one another along the forest edge were quite separate with 0.5-2 DCA units between them.Three groups of forest stands were distin guishable, corresponding to those identified in the classi fication by TWINSPAN.These groups are not separated along the first DCA axis but are distinguishable along the second DCA axis, which accounted for 19.6% of the vari ation.A similar pattern is also apparent in the ordination of the woodland groups of stands.
M iombo woodland is clearly separated from other woodland stands along the second DCA axis.The grass land stands are, likewise, clearly separated from the wood land stands.There is considerable overlap between the stands from the sandstone enclave and those from the northwestern part of the forest (i.e. the Psidium guajava B ushland (Type VI) and Acacia karroo-Heteropyxis dehniae Woodland (Type V).
The existence of the Chrysophyllum gorungosanum-Myrianthus holstii Forest (Type II) on the southeastern to the southwestern sides adjacent to the grassland was associ ated with a higher altitude and a wetter moisture regime, regardless of the shallow soils.Psidiiim giiajava Bushland (Type VI) occurred where the soils were predominantly sand stone and relatively deep.Where the sandstone soils were steeper and shallower, miombo and sometimes Acacia kar roo-Heteropyxis dehniae Woodland (Type V) occurred.On dolerite, w ith the sam e conditions, Strychnos mellodora-Chrysophyllum gonmgosanum Forest (Type I) oc curred, whereas the Teclea nobilis-Ehretia cymosa Forest (Type ID) occurred where it was drier.The Bridelia micrantha-Harwigana madagascariensis Mixed Woodland (Type IV) occurred on predominantly dolerite soils w ith other fac tors being intermediate.This was also found in what ap peared to be geological transitional zones.

Vegetation types
The vegetation of Chirinda Forest boundary has been classified into eight types.This represents only those types occurring within 100 m on either side of the forest edge.
The TWINSPAN analysis clearly separated forest from non-forest stands.The stands close to the forest edge, however, were not clearly separated on the basis of whether they were just inside or just outside the forest, as might have been expected.The apparent lack of clear differentiation may be due to the existence of some spe cies that are transitional between forest and woodland.Such a vegetation unit, which normally consists of both fire-tolerant savanna and fire-tender forest tree species (Hopkins 1992), represents a mosaic of communities of each of the two vegetation types.The transitional vegeta tion between savanna and forest shows great variation in both structure and species composition from place to place around the forest.Some of the types described in this category are restricted in extent whereas others occur widely, though in patches.
The transitional vegetation includes forest pioneer spe cies.Most pioneer species in Chirinda Forest are Afromontane endemics or near-endemics which are absent or rare in lowland forests.These include Albizia gummifera, Anthocleista grandiflora and Maesa lanceolata.Hence, the forest has sometimes been labelled a 'transitional' for est because of the co-existence of both lowland and Afromontane forest species (White 1978).Even though the majority of the moist forest species have lowland phyto geographic affinities, several species such as Chrysophyl lum gorungosanum (which is one of the commonest), Casearia battiscombei, Drypetes gerrardii, Halleria lucida, Myrianthus holstii, Prunus africana, Strombosia scheffleri and Xymalos monospora are Afromontane in origin.Chirinda Forest is, therefore, Afromontane in origin and character but with lowland phytogeographic affinities.
The southern section of the forest supports Chrysophylliun gorungosanum-Myrianthus holstii Forest (Type II), w ith more Khaya anthotheca and Trichilia dregeana, whereas the northern section mainly supports the Strychnos mellodora-Chrysophyllum gorungosanum Forest (Type I) with more Craibia brevicaudata and Strychnos mitis.Perhaps the greatest difference between these forest types mani fests itself in the variation in species composition of the understorey.This difference seems to be determined by soil depth and slope, which also influence the moisture regime.Shallow and steep sites drain faster than areas of moderate slope, leading to differences in the undergrowth species composition.Understorey species are therefore important in defining the limits of these two types.
The forest and woodland vegetation types share only a few species between them.Greater overlap in species composition was observed between the forest and the ecotonal vegetation than between forest and, for instance, miombo woodland.Shared species appear to occur mostly as seedlings, and less so as trees, in those habitat types to which they do not characteristically belong.Some of these, such as Croton sylvaticus, Harungana madagas cariensis and Bridelia micrantha are, however, true pio neers which thrive well under gap conditions in forests.

O rd in atio n
Forest generally occurs in wetter environments, and sa vanna in drier ones (Backeus 1992).In Chirinda, the amount of precipitation received from rainfall in the area is lower than the normal requirement for forest develop ment.The extra moisture comes in the form of orographic drizzle, made possible by several factors: the high ground, the southeasterly aspect, and the tall trees which facilitate the release of this extra moisture from the low clouds.The southern section of the forest has shallower soils, but a higher average altitude and receives more moisture from the south-easterlies, advected in from the Mozambique Channel, than the northern section.In general, Chirinda receives up to 28% more moisture than the average of five surrounding stations, much of which can be attributed to orographic drizzle (Mapaure 1993).
Miombo Woodland (Type VIII) was clearly classified separately from the Acacia karroo-Heteropyxis dehniae Woodland (Type V) by DCA but not by TWINSPAN.This might have been due to the apparent importance attached to Heteropyxis dehniae by TWINSPAN in Types V and VIII, resulting in the recognition of these types as one.Also of much interest, is the distance of separation be tween the dominant forest type [Chrysophyllum gorungosanum-Myrianthus holstii Forest (Type II)] and the rest of the non-forest types on the DCA ordination diagram.The DCA distance between the stands represents the av erage standard deviation of the species turnover, where a full species turnover occurs in about four DCA units (Gauch 1982)

£
FIGURE 2.-A map showing the major vegetation types in Chirinda Forest area.The positions o f the transects (A -J) are indicated.
FIGURE 3.-Dendrogram showing the TWINSPAN results of the quadrats in Chirinda Forest (eigenvalues are shown in brackets).

Myrianthus holstii, Strombosia scheffleri and Craibia brevicaudata in
the upper canopy stratum.The area covered by this type coincides with the wetter parts of the forest.Khaya anthotheca is quite common in some places.Dracaena fragrans is the dominant shrub and ap parently locally excludes other shrubs due to its high den sities.Strychnos mellodora is largely absent from the subcanopy, which is instead dom inated by Heinsenia

pentagona, Hippocratea goetzei, H. pallens and Oncinotis tenuiloba. Teclea nobilis-Ehretia cymosa Forest
(Type III)This forest consists of a mixture of species of appar ently drier forest affinities.It is dominated by

Teclea no- bilis and Ehretia cymosa. Common associates include Chrysophyllum gorungosanum, Diospyros abyssinica and Ochna arborea. It
is found in the northwestern part of the forest adjacent to an area which seems to have been cul tivated during the 1940s.

Celtis africana, Newtonia bucha nanii and Peddiea africana form
the dominant group in the shrublayer.There is high prevalence of small trees of pioneer species such as

Newtonia buchananii and Croton sylvaticus. Bridelia micrantha-Harungana madagascariensis Mixed
Thickets of Toddalia asiatica, Asparagus falcatus and, sometimes, Smilax anceps are a common feature of the shrub layer.Harungana madagascariensis, how ever, ceases to be dominant in some localities where Pterocarpus rotundifolius trees and Albizia gummifera saplings form an important component of the woodland.Peddiea africana shrubs and Croton sylvaticus seedlings are also common.

Acacia karroo-Heteropyxis dehniae Woodland
(Type V)This woodland, which is dominated by Acacia karroo, Heteropyxis dehniae and Faurea saligna, is on the drier side of the forest and just merges into the miombo wood land (dom inated by Julbernardia

globiflora, Uapaca kirkiana, U. sansibarica and Brachystegia spiciformis).
It is mostly found between the moist forest proper and welldefined non-forest associations.It contains a wide variety of species including pioneer species such as

Bridelia mi crantha, Albizia gummifera, Croton sylvaticus and New tonia buchananii
. A form of this vegetation type also occurs in small patches elsewhere around the forest, with varying degrees of dominance among the component spe cies.Occasional emergents include Parinari

curatellifolia, Catha edulis and Primus africana. Shrubs are represented by Psidium guajava, Vangueria apiculata, Rhus longipes and R. transvaalensis. Psidium guajava Bushland
(Type VI)This bushland is dominated by the exotic Psidium gua java and by Canthium mundianum bushes and sometimes assumes the structure of a low woodland.It is found in the sandstone enclave.A few emergent trees of Parinari curatellifolia and Albizia gummifera occur.Thickets of Toddalia

asiatica, Lantana camara and Rubus rigidus are
common, particularly near the boundary with the moist forest.Peddiea africana and Pteridium aquilinum are also common.

The meda triandra and Cymbopogon caesius.
It occurs on shallow dolerite soils with occasional exposed boulders.There is local variation in grass species dominance, with Loudetia simplex apparently becoming more dominant on east-facing slopes of the southern part of the forest where the soils are deeper and mostly sandy, sometimes with Parinari curatellifolia bushes.Other woody species found within the

Julbernardia globiflora and Brachystegia spiciformis with Uapaca kirkiana, Het eropyxis dehniae and Faurea saligna as
common associ ates in the canopy layer.It is relatively open, with poorly developed shrub and herb layers.Common trees in the subcanopy layer include Heteropyxis