The floristics of Sand Forest in northern KwaZulu-Natal, South Africa

We use multivariate analyses (ordination and classification) to assess both the floristic uniqueness of the woody vege­ tation of Sand Forest in relation to a range of other forest types in the region, and the range of variation within Sand Forest. Two broad Sand Forest subtypes and related Ecotonal Forests are described and grouped under the term Tropical Dry Forest, distinct from all evergreen forests in South Africa. Sand Forest, a dry semi-deciduous type in northeastern KwaZulu-Natal is defined by the presence of the canopy dominant Cleistanthus schlechteri as well as Hymenocardia ulmoides, Psydraxfra- grantissima, Croton pseudopulchellus and Drypetes arguta. Sand Forests form a cohesive group in both DCA and TWINSPAN analyses, with similar composition of canopy dominants at sampled sites. This implies that ecological func­ tioning is similar across the geographical range in northeastern KwaZulu-Natal. However, turnover of subdominant species between recognisable Sand Forest types emphasizes the need to conserve the full range of extant forests.


INTRODUCTION
Maputaland, the northeastern tip of KwaZulu-Natal (Figure 1), forms the narrow southern portion of a large coastal plain extending up the east coast of Africa as far north as Somalia (Watkeys et al. 1993). Maputaland is bordered by Mozambique in the north, the Indian Ocean to the east and the Lebombo Mountains to the west (Moll 1978;Bruton & Cooper 1980). The southern boundary can be drawn from the southern end of the Lebombo Range to the mouth of the St Lucia Estuary (Watkeys et al. 1993). The vegetation is a complex mosaic of forest, thicket, savanna and grassland, with a high proportion of endemics (perhaps 40% of woody species) and abrupt local changes in response to soils and climate (Moll 1980). Despite this botanical importance, the area has only recently been scientifically explored. Bayer, in a 1938 study encompassing the coastbelt and midlands of Zululand, stated : '...there is no doubt that throughout the coastbelt proper, evergreen sub-tropical forest... is a true climatic climax.' However, moist, evergreen forest is certainly not the only forest type in the region. References to dry forest with a unique complement of species first appear in the literature in the mid-1960mid- 's (e.g. Vahrmeijer 1966Tinley 1967). Moll (1968Moll ( , 1978Moll ( , 1980 and Moll & White (1978) used the local term 'Sand Forest' to describe this dry deciduous or semideciduous forest occurring on sandy soils. They list a vari ety of common and widespread tree species, including Newtonia hildebrandtii, Cleistanthus schlechteri, Hyme nocardia ulmoides, Balanites maughamii, Ptaeroxylon obliquum and others. The term Sand Forest has since passed into general (e.g. Goodman 1990;Midgley et al. 1997) and popular use (e.g. Pooley 1993;Craib 1995). In South Africa, it refers to dense forests with numerous trees and shrubs (De Moor et al. 1977;Moll & White 1978;Moll 1978Moll , 1980 with a relatively short canopy White 1978;Moll 1978Moll , 1980Ward 1981). These forests tend to be patchily distributed in characteristic northsouth oriented strips (Vahrmeijer 1966;Moll 1978Moll , 1980Moll & White 1978;Ward 1981).
Although this forest type is a conspicuous feature in Maputaland (Moll 1978(Moll , 1980pers. obs.), descriptions have been cursory (e.g. Vahrmeijer 1966;Tinley 1967;. Moll 1978Moll , 1980Moll & White 1978) or very local (e.g. De Moor et al. 1977;Ward 1981;Goodman 1990). There is confusion regarding the definition of Sand Forest and which plant species are representative and characteristic (e.g. Moll & White 1978cf. McKenzie 1996. The published literature describing Sand Forest is confined to a few paragraphs. Species lists associated with this forest type appear to be derived largely from casual observation, and feature conspicuous canopy emergents such as Newtonia hildebratidtii and Balanites maughamii which may be associated with other vegeta tion types (pers. obs.). Only one study, based on a regional data set and using quantitative methods, defines Sand Forest types in terms of characteristic tree species (MacDevette et al. 1989). However, that work is pre sented only as a preliminary classification of the KwaZulu-Natal indigenous forests, and is based on species checklists of varying reliability, with whole forests as the basic sample unit.
It is also unclear how closely Sand Forest is related to other forests in the region. Moll (1978Moll ( , 1980 simply treats Sand Forest as one of thirteen or fifteen vegetation types in Maputaland. Moll & White (1978), in a descrip tion of the Indian Ocean Coastal Belt (a floristic zone stretching from just south of Somalia to the Cape), include Sand Forest in the Tongaland-Pondoland Regional Mosaic with four other forest types. Midgley et al. (1997) follow this approach and that of White (1983) which emphasizes the separation of the South African forest flora into the Afromontane types and the Tongaland-Pondoland Regional Mosaic which includes Sand Forest. MacDevette et al. (1989), in a TWINSPAN classification of the KwaZulu-Natal indigenous forests, group an Eastern and Western Sand Forest type under the title Tropical Dry Forest with four other Coastal Forest types. Their main division separates Coastal Forests from Interior Forests in a similar manner to the treatments described above. Tinley (1967Tinley ( , 1977, how ever, regards Tropical Dry Semideciduous Forest in South Africa as part of a much larger Southern Tropical Sand Forest Domain, completely separate from all moist evergreen forests (including evergreen coastal forests). Moll & White (1978) list a variety of Sand Forest species linking the Pondoland-Tongaland Regional Mosaic with Dry Forests in the Zanzibar-Inhambane Regional Mosaic, which would seem to support Tinley's approach.
Despite the lack of any comprehensive study of Sand Forest, its conservation in South Africa is considered important for a number of reasons. Sand Forest covers a small total area  and is heavily impact ed outside of reserves (Moll 1978;Geldenhuys & MacDevette 1989;pers. obs.). It is rich in woody species (Moll & White 1978) and the habitat of a number of unusual or rare animals, such as the Suni, 294 Neotragus moschatus (Lawson 1986), the African broadbill, Smithornis capensis, and Neergaard's sunbird, Nectarinia neergaardi, which is largely confined to the Sand Forest (Harrison et al. 1997). It is a drawcard for tourists (Macfarlane 1993;Craib 1995) and an important local resource, providing a range of building materials, traditional medicines and some food plants (Cun ningham 1985).
At present, there is little information available for conservation planning, or for the delimitation of sensible ecological units for management and research into the dynamics of this little known forest type. Human pres sure on the natural environment is increasingly severe as improved infrastructure leads to a rapidly increasing rural population. This description is important in ascer taining the conservation worthiness of the forest type as a whole and how well the range of floristic variation is presently conserved. In addition it serves as a basis for further ecological work.
Our aims are: to determine how similar Sand Forest in northern KwaZulu-Natal is to other South African forests, especially the moist evergreen Coastal Forests of KwaZulu-Natal; to define Sand Forest in terms of its woody species composition; and to describe any varia tion within the Sand Forest type. Bothalia 29,2 (1999) STUDY SITE Maputaland in South Africa is a low coastal plain cov ering approximately 5 700 km2 (Watkeys et al. 1993). The climate is moist subtropical along the coast where rainfall is over 1 000 mm per annum becoming dry sub tropical inland with less than 600 mm per annum. Rainfall increases again to over 800 mm per annum along the crest of the Lebombo Mountains (Maud 1980;Watkeys et al. 1993). The highest monthly precipitation falls between September and April resulting in hot, humid summers and cool, dry winters.
The soils of Maputaland are complex, although most of the area is covered by infertile, sandy Tertiary and Quaternary deposits (Watkeys et al. 1993). Marine transgressions and regressions since the end of the Cretaceous have formed these deposits into dune ridges oriented in a north-south direction (Goodman 1990), parallel to the present-day coastline. These dune cor dons decrease in age from west to east, and the oldest, most westerly dune cordon may date from the Plio-Pleistocene (Davies 1976, cited by Goodman 1990). These oldest palaeo-dunes are not well preserved and are deep red in colour due to advanced mineral diagen esis. The soils on younger, more easterly dunes are gen erally poorly developed, yellow to orange arenosols. Sand Forest occurs on the full range of these inland dunes. The tall coastal dune cordon is composed of dys trophic pallid sands, with steep slopes stabilised by dune forest and scrub. Between the dune ridges, the coastal plain is flat to gently undulating, and may be covered with loose dystrophic sands (Goodman 1990;Watkeys et al. 1993).

Data collection
Analyses were based on two data sets, a 'Sand Forest' data subset, sampled specifically for this study, was com bined with a regional Northeastern KwaZulu-Natal data set, to allow comparisons among a range of forest types. Sampling was confined to woody vegetation for a num ber of reasons. Non-woody understorey vegetation is temporally and spatially variable in these seasonally dry forests, and our short-term sampling program could not adequately assess this component. From a practical point of view, we are attempting to delimit ecological units, and it is the trees that most affect the forest environment. In Sand Forest the non-woody component is particulary sparse and contributes little to total biomass. Also, trees, shrubs and lianes are easier to find and identify than grasses and herbs, and we hope this study will be accessable to nonspecialists.

'Sand Forest' Data Set
Forests growing on or near sandy soils in Maputaland (excluding dune forests) were sampled to represent all the so-called Sand or Tropical Dry Forest types mentioned by De Moor et al. (1977). Moll (1978Moll ( , 1980, Moll & White (1978), Ward (1981), MacDevette et al. (1989) and Goodman (1990). Sampled sites include the KwaZulu-Natal Nature Conservation Services reserves; False Bay Park, Mkuzi Game Reserve, Tembe Elephant Park, Ndumu Game Reserve and Sileza Forest Reserve, the pri vately owned Phinda Resource Reserve (hereafter referred to as False Bay, Mkuzi. Tembe, Ndumu. Sileza and Phinda respectively) and relatively undisturbed near by areas which will be referred to by the same locality names. Protected areas were preferred for this survey as unprotected forests are usually heavily disturbed, compli cating the recognition and definition of sand forest types.
At each locality, areas of forest (closed canopy, woody communities, > 5 m, MacDevette et al. 1989;Midgley et al. 1997) were chosen subjectively to repre sent the range of variation in structure and species com position. Within these areas, quadrats were randomly located, with the proviso that they be al least 50 m from any previous quadrats and the forest edge. The number of quadrats located at each locality was subjectively deter mined. Sampling was halted when the variation in species composition was adequately represented. Due to the naturally patchy and discontinuous nature of Sand Forest, which is further fragmented outside of protected areas, a more structured approach was considered impractical.
Samples consisted of circular 400 n r quadrats. This is a suitable size for short forest communities (Kent & Coker 1992), representing the approximate point of inflection ot a species/area curve within a homogenous area of sand forest (Goodman 1990; D. Kirkwood unpubl. data) and is compatible with samples collected by other workers in the region. Species abundance values in each quadrat are total diameter at breast height (DBH), calcu lated from the sum of area at breast height of all individ uals. DBH of all woody individuals taller than 2 m, root ed within the quadrat was measured. Height of trees shorter than 2 m was measured, and converted to an esti mate of diameter from a linear regression of height vs DBH for all trees between 2 and 3 m high at a site. Shrubs (largest individuals usually < 2 m), woody lianes and creepers were assigned an arbitrary total DBH of 2 cm. Tree species names follow .

Northeastern KwaZulu-Natal Data Set
Data from circular 400 n r quadrats sampled in a range of northeastern KwaZulu-Natal forests (see Figure 1 These quadrats from coastal and inland types were combined with the 135 quadrats sampled for this study from as wide a range as possible of 'Sand Forest' and related types in Maputaland. This includes 16 quadrats sampled in Hell's Gate, just south of False Bay Park (D. Kirkwood unpubl. data), an area administered by the S.A. National Defence Force. The Hell's Gate quadrats are only included in the regional data set as the forest was substantially different floristically from other Sand Forest/Tropical Dry Forest types. In this regional data set. abundance values were sim plified to presence/absence of species, and small shrub, liane and creeper species were excluded, in order to overcome compatibility problems. While this reduces the information content of the data, it allows the robustness of results from analysis of the 'Sand Forest' data subset to be assessed.

Multivariate analysis
Our classification and definition of forest types, as well as the ordinations of samples are derived from indi rect gradient analyses, utilising only floristic data. The most meaningful results were obtained using two wellknown, robust and complementary techniques (Gauch 1982;Kent & Coker 1992): Two-Way Indicator Species Analysis (TWINSPAN. Hill 1979), a polythetic, divisive program and the Detrended Correspondence Analysis (IX'A) option in the package CANOCO (Ter Braak 1991). These quantitative multivariate techniques were used to analyse the two data sets.
For all TWINSPAN analyses the following defaults were used: 10 indicator species per division and a mini mum group size for division of five quadrats. Three lev els of division were adequate to separate the regional data set into groups consisting of quadrats largely from one locality. Four levels of division resulted in meaning ful final groups used for the detailed classification of the Sand Forest data subset, which has a higher information content. Pseudospecies cut levels for the Sand Forest analysis were set at 0, 2.5, 10, 20 and 40 cm total DBH.
Analyses of the regional data set utilised all quadrats and only presence/absence abundance values were used. For the Sand Forest Data, two outlier plots signif icantly changed the relationships amongst the remain ing plots. These were found to be quadrats in woodland clumps on the margin of sand forest patches in Tembe and Mkuzi and were thus eliminated from the final analysis. Our final classification of Sand Forest and related types is based on TWINSPAN analyses of both the regional and Sand Forest data sets. TWINSPAN, which is based on a reciprocal averaging algorithm, successively divides groups of samples utilising the dif ferential presence or absence of species (Gauch 1982;Kent & Coker 1992). The program identifies indicator species for each division; 'characteristic species' are those which consistently occur in only one of the two groups of samples under consideration; 'preferential species' occur in a greater proportion of samples in one group than in the other.
All quadrats were used in the DCA of the regional data set. Similarly, although the related Hell's Gate Forests were not incorporated, all other Sand Forest/Tropical Dry Forest type quadrats were used in the DCA analyses of the 'Sand Forest' data set. Two sep arate DCA analyses of the 'Sand Forest' data set were performed. The first uses species abundance values of total DBH. The subsequent analysis uses only pres ence/absence values to reduce the influence of dominant species.

Dominance
In order to evaluate the relative influence of dominant species on the grouping of Sand Forest quadrats by the DCA, dominance diversity curves were constructed for the six sites sampled. Importance values for these curves were calculated as the average of a species' relative dom inance value (species total DBH/locality total DBH) and its relative density (no. plants/total no. plants) over all quadrats sampled in an area, excluding those shown to be outliers in the DCA ordinations. Quadrats sampled from the same area are grouped together by all analyses, with very few exceptions, indicating that forests are locally uniform, and sites can be used as natural units.

Affinities
The first two ordination axes of the detrended corre spondence analysis of the northeastern KwaZulu-Natal forests (Figure 2)   The TWINSPAN classification of the same data set (Figure 3) emphasizes the separation of samples of 'Sand Forest' and it's allies from those of the moist evergreen Coastal and Interior Forests in the first division (eigen value: 0.749).

Classification
In classifying the Sand Forest types we follow the ter minology of MacDevette et al. (1989), whose group names are sensible and easily remembered. However, it is important to note that the groups are substantially modified and many of our characteristic and representa tive species differ.  TWINSPAN analyses of the regional data set ( Figure  3) and the Sand Forest data set ( Figure 4) produce com pletely congruous classifications. Characteristic and preferential species based on divisions 1 and 2 of the regional analysis and divisions 2, 3 and 4 of the Sand Forest analysis are presented in our classification.

Sand
We suggest that the term 'Tropical Dry Forest' be used in South Africa to encompass both Sand Forest and allied dry semideciduous forests. Although these forests are strictly subtropical, it seems likely that they are floristically allied and ecologically similar to other Tropical Dry Forests in Africa.

This forest type is defined by the presence of the char acteristic tree species Hymenocardia ulmoides, Wrightia natalensis, Pteleopsis myrtifolia, Cleistanthus schlechteri, Newtonia hildebrandtii and Drypetes arguta. Preferential tree species are Cola greenwayi, Hyperacanthus amoenus, Boscia foetida, Brachylaena huillensis, Combretum mkuzense, Dialium schlechteri, Grewia niicrothyrsa, Haplocoelum gallense, Monodora junodii, Psydrax fragrantissima, Ptaeroxylon obliquum, Strychnos henningsii, Toddaliopsis bremekampii and Tricalysia lanceolata.
Although we have comprehensively sampled Tropical Dry Forest, only a small range of KwaZulu-Natal forests is used for comparison here. While the tree species char acterising Tropical Dry Forest at this level of classification are representative, they may not be definitive when used in comparison with forests not included in this analysis.
Tropical Dry Forest is divided into two subtypes: Sand Forest and Ecotonal Forest (eigenvalue: 0.326).

Ecotonal Forest
These samples, while clearly allied to Sand Forest, are floristically diverse and occur on a variety ot soils. Quadrats sampled in False Bay Park on duplex soils and soils with a high clay content fall into this group, as do quadrats on grey regie sands from the Hell's Gate area. Due to the variable species composition of quadrats in this group, we will not define subtypes within this group.

Sand Forest
This type includes the majority of Tropical Dry Forest samples (103 of 143) and forms a cohesive group in the 1 Forest growth form, known locally as umHlalakolotshe : Combretum c f celastroides, an unidentified but common climber in these forests. * abundant, total DBH > 20 cm ** very abundant, total DBH > 40 cm DC A of the regional data set (Figure 2 For further divisions, the results of the TWINSPAN analysis of the Sand Forest data subset are presented. Shrubs and lianes, as well as abundance values for all woody species are recorded in this subset of quadrats. Where a species name is marked with an asterisk or dou ble asterisk, this denotes that a characteristic or preferen tial species of a group is abundant-total diameter at breast height (DBH) > 20 cm; or very abundant-total DBH > 40 cm. Despite the increased information contenl of this data set, the TWINSPAN classification of Sand Forest samples (Figure 4) corresponds exactly with the divisions produced using the regional data set (Figure 3). The third level of division in the TWINSPAN classi fication of the Sand Forest data set (Figure 4) essential ly subdivides both the Western (eigenvalue: 0.207) and the Eastern (eigenvalue: 0.260) Sand Forest types into groups of samples that reflect species turnover between geographically separate sites. Only quadrats from Phinda and those on sandy soil at False Bay are grouped together. In the fourth and final level of division (not illustrated) it is notable that both the Mkuzi quadrats (Western Sand Forest), and the Phinda/False Bay quadrats (Eastern Sand Forest) are subdivided into groups characterised by the presence or absence of Newtonia hildebrandtii. Localised stands dominated by this tree, a large, spreading canopy emergent, are a con spicuous feature of these forests (pers. obs.; Moll & White 1978).

Species turnover and dominance in Sand Forest samples
In a DCA ordination of the same Sand Forest data subset ( Figure 5), with species abundance values of total DBH, quadrats are not separated into the groups described above. Samples from False Bay, Mkuzi, Phinda. Tembe and Ndumu are grouped together, with a high degree of overlap within a range of two half changes (each DCA unit or average standard deviation of species turnover is approximately equivalent to a 50% change in species composition of samples, Gauch 1982). Only quadrats from Ecotonal Forest on duplex soils at False Bay and the Sileza samples are clearly separated from the main group. However, when species abundance values are reduced to presence/absence values, the resulting DCA ordination plot (Figure 6), groups quadrats in an identical manner to the final level of the TWINSPAN classification illustrated in Figure 4. This result confirms the validity of the classification. In addi tion, ground-truthing using this classification outside of sampled areas indicated that we have adequately covered the range of variation of Sand Forest in Maputaland.
Separation of Sand Forest subtypes in the DCA ordi nation presented in Figure 6 results from reducing the influence of dominant species. This implies that within the Sand Forest type, forests have similar dominant species, but there is significant turnover of the less com mon species between groups. Dominance-diversity curves at each locality (Figure 7) indicate that this is the case. Cleistanthus schlechteri and Newtonia hilde brandtii rank consistently high. Note the disparity between the importance values of the one or two most dominant species and the other species at most sites.
Although the division between Western and Eastern Sand Forest is justified and convenient, it seems that the most natural grouping of forests is into three groups: the Mkuzi/Ndumu Group, separated from all other samples on the first and third DCA axes in Figure 6. and the Phinda/False Bay Group separated from the Tembe/Sileza Group on the second axis. These groups may represent a soil gradient, from the oldest red sands of Western Mkuzi/Ndumu Group, through orange to yellow sands of Sand Forests at Phinda and False Bay Park, to the pre dominantly yellow to white sands of Tembe and Sileza. The samples at Sileza. an isolated forest occurring on dys trophic white sands and surrounded by Hxphaene natal ensis Palm Veld, although included in Eastern Sand   Figure 5).
Forest, are separated from other Sand Forests by all DC As and should not be considered typical.
A sample by species table of the Sand Forest data set is presented in Table 1 with a complete species list for all the localities sampled. Importance value classes in the table matrix represent total stem diameter at breast height for a species: 1 =0-2.5 cm DBH, 2 = 2.5-10 cm DBH, 3= 10-20 cm DBH, 4 = 2(M 0 cm DBH, 5 = > 40 cm DBH.  In KwaZulu-Natal, Tropical Dry Forest, including Sand Forest and Ecotonal types, is clearly distinct from both coastal forests and an interior forest. This would tend to support Tinley's (1967Tinley's ( , 1977  The similarity in terms of dominant tree species across the range ot Sand Forests indicates that these forests can be treated as functionally uniform. This is important, as it allows us to extrapolate the results of ecological research and apply similar management prac tices throughout. Significant turnover of plant species does occur however, and since many species appear to be confined to these forests (pcrs. obs.), conservation of the range of variation is crucial. Fortunately, although Sand Forest covers a smaller area than any other vegetation type in South Africa, it is well conserved (Low & Rebelo 1996). Our surv ey of woody plants indicates that the full range of variation in South Africa is represented in con served areas. Forests in Mkuzi and Ndumu Game Reserves represent the Western Sand Forest, while forests in False Bay Park. Phinda Resource Reserve, Tembe Elephant Park and Sileza Forest Reserve ade quately represent the more variable Eastern Sand Forest. The non-woody understorey component of these forests deserves further study. Turnover of herbaceous plants and grasses between sites appears to be high and these plants may include a high proportion of endemics. The naturally patchy nature of these forests suggests that the fragmentation associated with conservation in non-contiguous reserves is unlikely to be important.