Ecological interpretation of plant communities by classification and ordination of quantitative soil characteristics

An agglomerative cluster analysis and a principal components analysis of habitat, based on 27 quantitative soil variables, are compared with a Braun-Blanquet classification of the vegetation of the Manyeleti Game Reserve in the eastern Transvaal. The results indicate that these techniques can be successfully used to obtain relatively homogeneous habitat classes, characterized by sets of environmental (soil) variables and not only single variables individually, and which are furthermore significantly correlated with the recognized plant communities of the area.


INTRODUCTION
For the establishment of efficient wildlife man agement programmes and conservation policies for any area, a sound knowledge of the ecology of the area is an essential prerequisite (Edwards, 1972).It has often been demonstrated that different eco systems of a particular area can be recognized by the delimitation of the plant communities within the area (Major, 1969;Kiichler, 1973;Bredenkamp & Theron, 1978).For this reason, and as part of a vegetation survey programme for conservation areas in South Africa, a study of the vegetation of the Manyeleti Game Reserve was undertaken.Al though great diversity and variation occur in the vegetation, there is little conspicuous variation in the topography of the slightly undulating landscape, especially within the relatively small study area.The importance of soil characteristics as a principal ecological factor determining the distribution of plant species and plant communities in the Transvaal Lowveld has been emphasized by.inter alia, Van der Schijff (1957), Gertenbach (1978) and Bredenkamp (1982).It has often been shown that the distribution of plant species and.especially, plant communities, is the result of the totality of the present environmental factors rather than of single factors (e.g.Roberts, 1971;D. Scott, 1974;J. T. Scott. 1974;Bredenkamp, 1977).This paper summarizes the results of an attempt to correlate the plant communities of the Manyeleti The slightly undulating plains of the study area are situated at an altitude of 350-450 m with a slight rise towards the west.Numerous dry drainage lines dissect the area.Archaean granite covers most of the Reserve, but portions of a large dolerite dyke are exposed in the western parts.The soils of the uplands sites on granite are coarse, sandy, acid, leached and dystrophic whereas the soils of the bottomland sites are fine textured.neutral, mesotrophic and calcareous, and sometimes brackish.The soils of doleritic origin are very clayey, alkaline, eutrophic and calcareous.
The climate of the study area is, according to the Kóppen index, a BShw climate (Schulze, 1947) where BS = arid steppe climate; h = hot and dry, with mean annual temperature exceeding 18°C; and w = dry winter.The average annual rainfall for the period 1967 to 1979 is 614,6 mm.
Temperatures recorded for Skukuza range from mean daily maximum temperatures of 32,3°C in December and January to mean daily minimum temperatures of 5.6°C in July (Weather Bureau, 1954).
A Braun-Blanquet analysis of the floristic data from the relevés was carried out independently and before cluster analysis and principal components analysis of soil habitat data (Table 1).
Quantitative data for 27 soil variables were obtained by analysis of soil samples from the 245 relevés (Table 2).The soil variables include: percentage gravel in the soil sample in the A as well as B soil horizons; percentage coarse sand, medium sand, fine sand, total sand and clay (particle sizes according to MacVicar et. al., 1977) in the A as well as B soil horizons, after the gravel has been removed; the amounts of exchangeable K+, Na+, Mg:+, Ca2+ and the S-values (mg/100 g soil) of the A and B soil horizons; the soil conductivity (u mho/cm) of the A and B soil horizons; and the soil depth (cm).
Classification and ordination of the habitat were done on standardized data (Seal, 1964) obtained from the absolute quantitative values of these variables.
Classification was done using Orloci's agglomerative cluster analysis (Orloci, 1967) in order to obtain definite and relatively homogeneous habitat classes, based on the variation within the sets of quantitative soil characteristics.This classification was compared  to and correlated with the seven plant associations of the Braun-Blanquet classification (Table 3).
Ordination by principal components analysis was done to indicate possible gradients in the soil habitat and, by locating the position of each plant assocation in the habitat gradient, it was determined whether or not the associations are restricted to certain areas within the gradient (Table 4).From the results of this ordination, the individual soil variables which could be important with regard to the distribution of the plant communities, were also determined (Table 5).

Classification o f the soil habitat by cluster analysis
The dendrogram (Fig. 2) summarizes the results of the cluster analysis.Six soil habitat classes (CA-CF) were obtained at a similarity value of 88%.This high similarity value indicates the relatively small variation in habitat within each class.The average values for each variable in the six classes are given in Table 2.
Comparison of the values of the different soil habitat variables within and between the classes enabled the following soil habitat index to be compiled (for average values and explanation of the terms, see    nigrescens Associations.The statistical chi-square method of Bailey (1974) was used to measure coincidence between the classification of the soil habitat by means of the numerical cluster analysis, based on quantitative soil habitat data and the Braun-Blanquet classification of the vegetation based on qualitative floristic composition.Despite theoretical limitations of the method, the chi-square value of 526,4 (p=0,001 at x: = 59,7, 30°f) indicates that the two classifications are indeed highly significantly correlated.
It may be concluded that the soil habitat classes CA.CE and CD and the Perotis patens -Terminalia sericea Association, the Themeda triandra -Setaria woodii Association and the Spirostachys africana -Diospyros mespiliformis Association respectively, are mutually restricted to one another, within the study area.The soil habitats of the Euclea divinorum -Acacia nigrescens, Themeda triandra -Acacia gerrardii and Euclea divinorum -Albizia harveyi Associations are fairly closely related (soil habitat classes CB and CC, Fig. 2), at least as far as the measured quantitative soil characteristics are con cerned.These three associations occur on meso trophic clayey soils mainly of granitic origin.Although typical stands of these associations can easily be recognized, transitions in vegetation and habitat are frequently found (Bredenkamp, 1982), which could hamper identification.These transitions (gradients) are also indicated in the results of the ordination which follow.

Ordination o f soil habitat by principal components analysis
The results of the principal components analysis of soil habitat data show that the first two components explain 48,9% of the variation in the soil data, and these two components are therefore considered adequate to indicate possible soil habitat gradients and also to show whether there is an associated distribution of plant associations (Table 4).
The arrangement of relevés along the first two components of the soil habitat ordination represents gradients without consistent discontinuities (Fig. 3).Gradients in soil habitat variables (Fig. 3) were established by superimposing the quantitative values of those variables with relatively high eigen values (Table 5) and which contribute greatly to the The different Braun-Blanquet plant associations (Table 1), represented by the same relevés than those used in the soil habitat ordination, were superimposed on this ordination and the results (Fig. 3) show that the different plant associations are indeed remarkably confined to certain areas within the soil habitat gradients and consequently the plant associations could easily be delimited (Fig. 3).
The grouping of relevés by principal components analysis of soil habitat data was compared to the Braun-Blanquet classification of these relevés.The results are given in Table 4. From Fig. 3 and Table 4 it is clear that the Themeda triandra -Setaria woodii Association is entirely restricted to soil group OE; the Perotis patens -Terminalia sericea Association is almost entirely restricted to soil group OA; the Spirostachys africana -Diospyros mespiliformis Association to soil group OG; and the Cardiospermum corindum -Acacia nigrescens Association to soil group OF.The soil groups OE, OA, OG and OF, furthermore, mainly contain relevés of the abovementioned four assocations respectively.These results indicate that the four groups of the soil habitat and four plant associations are mutually restricted to one another, and largely confirm the results of the cluster analysis.
The Euclea divinorum -Acacia nigrescens Association is principally confined to soil group OB (48 of the 62 relevés, that is 77,4%), but group OB also contains relevés from the Themeda triandra -Acacia gerrardii and the Euclea divinorum -Albizia harveyi Associations.The Themeda triandra -Acacia gerrardii Association is mostly limited to groups OB and OC whereas the Euclea divinorum -Albizia harveyi Association occurs in group OB and OD.These results indicate, as the results of the cluster analysis, the relationships and transitions between the habitat of the Euclea divinorum -Acacia nigrescens, the Themeda triandra -Acacia gerrardii and the Euclea divinorum -Albizia harveyi Associations.
Despite the somewhat arbitrary delimitations of the plant associations in this ordination, the classification obtained was correlated to the Braun-Blanquet classification by means of the chi-square test of Bailey (1974).The chi-square value of 824,8 (p=0,001 at x2 = 68,0 36°f) indicates a highly significant correlation between the two classifica tions.
For each plant association, the average values of the most important variables, i.e. those with relatively high eigen values in the first two components, are given in Figs 4 & 5. Variables with relatively high eigen values in the first component include total sand, coarse sand and clay contents of the A soil horizon and the S-value, magnesium and calcium contents and the pH of both A and B soil horizons (Table 5).A [

Vv169 f /
Vv I "\ 274 According to the zone in the gradients to which the different associations are confined, a suggestion of the relationships of each assocation to the most important habitat variables is made and the transitions between related associations are also indicated.The soil habitat of the Perotis patens -Terminalia sericea Association is situated at the one end of the soil gradient where it represents acid, coarse, sandy soils, especially low in calcium and magnesium content.This habitat grades mainly into the habitat of the Euclea divinorum -Acacia nigrescens and Themeda triandra -Acacia gerrardii Associations with slightly acid to neutral, clayey and nutritionally richer soils, but also, to a lesser degree, into the habitat of the Euclea divinorum -Albizia harveyi Association which represents the neutral, clayey and mesotrophic but brackish (sodium-rich) soils.The soils of the three last mentioned Associations are closely related and gradual transi tions are common.The habitats of the Themeda triandra -Acacia gerrardii Association and, especially, the Euclea divinorum -Albizia harveyi Association, also grade into that of the Themeda triandra -Setaria woodii Association, which is situated at the other end of the gradient where the soils are alkaline, very clayey and eutrophic.Soils of the Cardiospermum corindum -Acacia nigrescens Association, which is restricted to the rocky doleritic hills in the transitional areas between dolerite and granite, show relationships with those of the Themeda triandra -Acacia gerrardii and Themeda triandra -Setaria woodii Associations.The widely scattered distribution pattern of the relevés of the Spirostachys africana -Diospyros mespiliformis Association of the riverbanks (group OG, Fig. 3) indicates the great variation in soil character of this Association.These different soils show relationships with the soils of most other Associations through which the rivers flow.In spite of this variation, these soils could easily be united into a single group (group OG, Fig. 3) which indicate the relationship of these soils to each other.

N/ \
The results indicate that, although the habitat often represents a complex continuum, soil groups which largely represent the floristic Associations can be established within the gradient.
In conclusion the following can be emphasized: 1.The use of classification and ordination techniques on quantitative soil characteristics proved very successful in obtaining relatively homogeneous soil habitat classes which are (i) characterized by sets of environmental (soil) variables rather than of individual single variables and are (ii) significantly correlated with the recognized plant communities of the area.
2. The soil habitat and distribution of the plant communities were successfully interpreted and the reality of these communities as ecologically signifi cant units emphasized.
3. The transitions in vegetation can also be explained by gradients in the habitat (e.g.Breden kamp, 1982).4. Some individual environmental variables which may influence the distribution of the plant communities can also be identified from the results of the principal components analysis.

F
i g .1.-Map indicating the posi tion of the Manyeleti Game Reserve.

F
i g . 2 .-Ad e n d r o g ra m in d ic a tin g th e c lassificatio n o f the h a b ita t by cluster analysis.

F
i g .3.-The distribution of the Braun-Blanquet Associations in relation to the first and second components of the ordination (Symbols are explained in Table 1.) distribution pattern of the relevés along the first two components of ordination.
Fig .4.-Distribution of average quantitative values in each Association of the soil variables with high eigen values in the First component of the ordination.
. 5 .-D is tr ib u tio n o f average q u a n tita tiv e v alu es in each A s s o c ia tio n o f th e soil v ariab le s w ith h ig h eig en v alu es in th e second c o m p o n e n t o f the o r d in a tio n .values of the other important variables mostly decrease from left to right (Figs 3 & 4).For the second component, variables such as clay content, soil conductivity and sodium content of the B soil horizon are important.With relatively high positive eigen values, the quantitative values of these variables mostly increase from bottom to top in Figs 3 & 5. Variables with relatively high negative eigen values include total sand and medium sand contents in the B soil horizon, calcium, potassium contents and S-value in the A soil horizon and soil depth.The values of these variables mostly decrease from bottom to top (Figs 3 & 5).

TABLE 3 .
-A comparison between the distribution of 245 relevés in the seven Braun-Blanquet plant associations and the six habitat classes obtained by cluster analysis of habitat data

TABLE 4 .
-A comparison between the distribution of 245 relevés in the seven Braun-Blanquet plant associations and the seven habitat groups obtained by Principal Components Analysis of habitat data

TABLE 5 .
-Eigen values of the soil variables in the first and second components of the ordination Concerning the division of the relevés within the soil habitat classes, it is clear that soil habitat class CA is mainly associated with the Perotis patens -Terminalia sericea Association, class CD with the Spirostachys africana -Diospyros mespiliformis Association, and class CE with the Themeda triandra -Setaria woodii Association.Soil habitat class CB is mainly represented in both the Euclea divinorum -Acacia nigrescens Association and the Themeda triandra -Acacia gerrardii Association, and class CC is also represented in these two associations as well as the Euclea divinorum -Albizia harveyi Association.Although the relevés of soil class CF are present in almost all the associations, this class is best represented in the Themeda triandra -Acacia gerrardii, the Themeda triandra -Setaria woodii, and the Cardiospermum corindum -Acacia