Cape Hangklip area . I . The application of association-analysis . homogeneity functions and Braun-Blanquet techniques in the description of south-western Cape vegetation

Releve data were collected in two phases from, respectively 150 and 100 sampling points distributed by stratified random means through almost 24 000 ha of vegetation. Association-analysis, Braun-Blanquet and homogeneity function methods were used to treat the data. Only the "norm al” association-analysis method was applied. Three sorting techniques for tabulating the data were tested and were compared with a fourth method. Homogeneity functions were used to construct a dendrogram and to determine the degree of similarity between individual releves and groups of releves. After compaiison of the methods, it was concluded that the Braun-Blanquet method is consistently more efficient and more exact, even in the floristically rich vegetation of the south-western Cape Province of South Africa.


IN TRO D U C TIO N
Little descriptive work has been done on the vege tation of the south-western Cape Province of South Africa.Acocks (1953) has remarked that the Cape Fynbos vegetation " is a complex vegetation, and to divide it simply into Macchia and False Macchia is like dividing the tropical vegetation into grassveld and bushveld . . .Statistical approaches have been used by Rycroft (1951), Grobler (1964), Taylor (1969) and Hall (1970) as aids in the description of this vegetation.The largest area of vegetation described and mapped in detail is the 7 680 ha of the Cape of Good Hope Nature Reserve (Taylor, 1969).This latter study was the first attempt at applying association-analysis (Williams & Lambert, 1959. 1960, 1961;Lambert and Williams, 1962) and Zurich-Montpellier methods (Braun-Blanquet, 1932;Becking, 1957;Kuchler, 1967) to the Fynbos element (Taylor, 1972) of the South Western Cape vegetation.The Zurich-Montpellier principles, expressed by Braun-Blanquet and further developed theoretically by many followers (Werger, 1973b), will here be termed the Braun-Blanquet method.The Braun-Blanquet method has subsequent ly been used to describe some 373 ha of Fynbos vegetation in the Jonkershoek State Forest near Stellenbosch (Werger, Kruger & Taylor, 1972).
A method was required in the south-western Cape that would be most suitable for the evaluation, description and classification of large tracts of Fynbos vegetation.The Cape Hangklip area was floristically rich and included many variations of habitat.It would, therefore, be an exacting test of the suitability of any method.This served as the stimulus to apply homo * Botanical Research Unit, P.O.Box 471, Stellenbosch.
The study area consists of about 24 000 ha of coast and mountain vegetation.The boundaries were taken as those of the 1:50 000 Topographical Survey Sheet 3418BD Hangklip (Trigonometrical Survey, 1968) and the portions of the Kogelberg State Forest which occur outside this sheet.

The sampling method
Releve data were obtained from 150 and 100 sites scattered respectively over 11 506 and 12 354 ha.Sites were distributed randomly within physiographicphysiognomic units delimited on aerial photographs.In the first phase of 150 releves, each of the 10x5 m releves were strictly laid out with the longest axis on a north-south magnetic bearing.In the second phase each of the 100 releves was so positioned that a maxi mally homogenous sample was recorded.
For comparative purposes, only permanently recognizable species were listed, others being listed merely for record purposes.An additional list of species, occurring in the surrounds of the releve within the community being sampled, was also made.
Listed species were given cover-abundance and sociability values (Becking, 1957).A modified scale of abundance, similar to that described by Hanson (Brown, 1954), was used to assist in describing each community.
As required by association-analysis and homoge neity functions, a standard releve size of 10 X 5 m was used throughout.This size is identical to that used by Taylor (1969) for conformity in the event of compari sons being made.Comparisons of sample size versus information recorded are made in Table 1.The data used in these comparisons were obtained from visibly different communities.The following additional information was recorded at each sampling site: soil type, soil moisture, degree and type of stoniness, local climatic data, geology, geomorphology, species dominance, species height, vegetation age, disturbances, etc.

The association-analysis method
The various forms of association-analysis, namely, "normal" , "inverse" and " nodal" analyses, have been fully described by the developers of the method and by numerous other workers.South African workers who have applied this method include Van der Walt (1962), Grunow (1965), Downing (1966), Roberts (1966), Woods & Moll (1967), Miller & Booysen (1968), Scheepers (1969) and Taylor (1969).Scheepers (1969) found that the inverse analysis produced a stepwise arrangement of species grouping (also known as "chaining") which was difficult to interpret.Morris (pers.comm.)considered this step wise arrangement to be a fairly regular feature of the method when large numbers of species were involved, as in this case.It was considered advisable to restrict treatment of the data to the normal type of analysis.

1^-r
were available.Termination of subdivision N took place when there were less than eight releves left in the group or when the highest single jc2 equalled 3,841 or less.

The Braun-Blanquet method
The Braun-Blanquet method, commonly used in Europe and elsewhere, has been little used in South Africa.The only English source of information on this method was Fuller and Conard's (Braun-Blanquet, 1932) authorized translation of Braun-Blanquet's first edition of Pflanzensoziologie (Braun-Blanquet, 1928).More recent English descriptions of the method were published by Poore (1955), Becking (1957) and Kiichler (1967), amongst others.Taylor (1969) was the first to use the method in South Africa.He prepared a synthesis table from data collected in systematically distributed releves and obtained associations which were recognizable in the field.Werger, Kruger & Taylor (1972) then applied the phytosociological technique as further described by Ellenberg (1956) and Braun-Blanquet (1964), to test its usefulness in the floristically rich Fynbos vegetation in a portion of the Jonkershoek State Forest.A practical classification into communities based on floristic criteria was obtained.This method has subsequently been used by, amongst others, Coetzee (1972) in the Jack Scott Nature Reserve, by Werger (1973a & b)

M ethod using homogeneity functions
Homogeneity functions for identifying groups in a matrix of vegetation d ta have been developed by Hall (1967aHall ( & b, 1969aHall ( , b & c and 1970)).A function, given as H qm, is written, for the subset of t = \ . . .k sample plots and all the j = 1 . . .p species, as follows: k where sajk and shjk are the standard deviations of the subset's actual data row for the yth species, and a dummy maximally heterogeneous row, respectively; ajt is the value for the y'th species in sample plot t.These methods have only been tested on small vegeta tion data matrices from the Bains Kloof area of the Cape Province.A simplified form of this homogeneity function determines the similarity between the average members of each major group, subgroup, siblinggroup (Hall, 1969a) or core and each releve, thereby indicating the "goodness of fit" of each releve in each vegetation group delimited.
The similarity between two items or average mem bers t and k can be given by this simplified version of the homogeneity function.By similar notation, Here, the modulating factor M j is calculated exactly as before.The homogeneity expression that follows, uses abundance values scaled to a range with a maxi mum of one, a'jt and a y*.The scaling of the jth species is based on its largest operational value (Hall, 1970).

Association-analysis
For practical reasons collection and treatment of the data were divided into two separate phases.(1972).In the first phase, 150 releves were ordered into 32 final groups in the association-analysis hierarchy (Fig. 1).The hierarchy was constructed following the the conventional procedure of listing positively defined releves, at each subdivision, on the left-hand limbs and the negatively defined releves on the righthand limbs.To present the information in a more practical fashion, a logarithmic, instead of a linear scale, w'as used for values of highest single x 2. The linear representations are inset in Figs 1 and 2.

F i g . I.-F i r s t -p h a s e a s s o c i a t i o n -a n a l y s i s h ie r a r c h y from B o u c h e r
The hierarchy showed relatively uniform divisions with little tendency to chaining, except in the final negatively associated portion, where the more distinc tive communities such as the coastal dune vegetation and wet seepage communities, occurred.A reversal, or increase, in level of subdivision of a subsequent group, following the removal of a more homogeneous group, occurred after group 12 had been delimited.In certain instances, such as in the subdivision of groups six and seven, two species could equally well have been used to effect the subdivision, both resulting in the smallest total of residual significant associations in the two resulting subclasses.When such an ambi guity occurred, the computer was instructed to sub divide on the species with the lowest coding number.
In most cases communities proved to be under sampled rather than oversampled.In very few instances w'ould the recombination of adjacent final groups have resulted in ecologically valid larger groups.
A major difficulty, probably more commonly found in monothetic divisive techniques was the occurrence of apparent misclassifications.These were found especially during the first major subdivisions when large numbers of releves were involved.Here the chance absence of a species within a particular releve, although it occurred within the community being sampled, could result in the misclassification of the releve.This contingency was largely overcome by listing species not found inside the releve but occurring in its immediate vicinity, within the same community.The reclassification of any releve could be undertaken on these grounds.In a number of instances the cause was found to be the misidentification of the dividing species.This was attributed to the drought conditions prevailing at the time and to the problem of identifying vegetative specimens.The collection of 114 species of Ericaceae, characterized by ericoid leaves, underlines the reality of this problem.More than 1 400 different species were collected during this survey.
Eleven releves were regrouped after the relevant dividing species were found in the surround lists, while five were found to be wrongly grouped because of misidentifications.Two were transitional.The final groups in the hierarchy were found to vary in the degree of their floristic and ecological homogeneity.Insufficient sampling could be a possible reason.
The description and mapping of the vegetation of a portion of the study area was based on these final groups because they showed highest correlation of communities with habitat.A hierarchical arrangement has the advantage of providing a dichotomous key for the identification of communities.
In the second phase of the study, 97 of the 100 releves were divided into 25 final groups (Fig. 2).(Data from 3 additional releves were collected at a later stage to strengthen some of the Braun-Blanquet groups.)In contrast to the previous and most other normal analysis hierarchies, a total chaining of groups occurred.There was virtually no correlation of groups with habitat factors, but groups 1, 2 and 4 showed some correlation with Braun-Blanquet groups.The remaining groups did not appear to be correlated floristically or with habitat factors and were, therefore, not used in the description of the vegetation.

Braun-Blanquet
The conventional technique of manually rewriting the table while sorting the data was used initially.(No other facilities were then available.)A few major groups were distinguished which were further com pared using homogeneity functions.Further refining of the data was attempted once the TABSORT program me became available.A few more communities were extracted and the complex interrelationships between the communities were better shown (Boucher, Part II, in preparation).
The Ceskar & Roemer programme for identifying species-releve groups was tested on the same data.The criteria applied for grouping were 50% and 66% species occurrence within a group, each species having a maximum of either 10% or 20% occurrence outside that group.The groups delimited conformed to the major groups that were readily identifiable using con ventional sorting methods.Considerable further refine ment of the table using another method would there fore appear to be necessary after using this programme.This suggests that it could be a useful tool only for the initial sorting of the data.
In the second phase, where the collection of releve data conformed more to the principles laid down by Braun-Blanquet, the data could be refined to a fairly detailed level with the aid of the TABSORT program me (Boucher, Part II, in preparation).The subjective location of sampling sites resulted in their most effi cient distribution within the many variations occurring in the vegetation.An understanding of the vegetation, gained during the first phase, resulted in fewer transitional areas being sampled.The Braun-Blanquet groups obtained in the second phase were therefore clearer than those obtained in the first phase.
The TABSORT programme was used to study the meaning of the association-analysis groups in terms of the Braun-Blanquet community concept.The species rows were rearranged to determine whether any pattern of distribution would support the associationanalysis groups.Releves were also rearranged within each of these groups but not between the groups.On the basis of Braun-Blanquet differential species, some of the association-analysis groups were heterogeneous.This is possible w'hen the grouping is based on total species complement in each releve.A relationship of a different type, such as fire-age, might be indicated.This detracts from the acceptance of the associationanalysis, in comparison to the Braun-Blanquet, groups for community classification.

Homogeneity functions
The initial subdivision of the data before applying homogeneity functions was done using the Braun-Blanquet method.The easily extractable, obvious groups were accepted prior to being further analysed using homogeneity functions.The residual data matrix consisted of 108 releves having 185 perma nently recognizable species occurring in more than two releves.This matrix was still too large for analysis by this method with the available computer facilities.It was, therefore, further reduced by excluding the species of less frequent occurrence until 86 species remained.
A chaining effect was obtained in the homogeneity function dendrogram (Fig. 3) with virtually no distinct clustering.A broad moisture gradient could be discerned in the arrangement with drier sites linking onto the matrix at the higher homogeneity levels and wetter sites linking at the lower homogeneity levels.This was similar to the association-analysis grouping where the wet seepage communities linked at the lower highest single x 2 values.The data matrix for comparison using the homo geneity functions consisted of the less distinct com munities.The absence of the species of rarer occur rence probably reduced the sensitivity of the method to an excessive degree.The end product was, therefore, unsuitable for the description or the mapping of the vegetation.
The distinctness of the recognized groups, and of the members in each group, to their average member was determined using the simplified form of the homo geneity function.With this method a member, when compared to itself, would be 100% similar.All the groups recognized were compared to each other, whether they were those delimited by the Braun-Blanquet method or were indistinct groups or arbitrary cores extracted from the homogeneity function dendro gram.The distinctness of the Braun-Blanquet groups was confirmed {vide the example in Fig. 4, where the length of the bar indicates the degree of similarity).The degree of similarity between the releves in the complex data matrix was also readily shown diagrammatically by the length of the bars.
Affinities between groups and the best location of transitionary releves are easily shown by this method.

C ON CLU SIO NS
The releve size (10 m x 5 m) was found to result in an adequate sample of each community, at the scale of study involved.The releve shape could possibly have been more flexible, for instance in sampling the riparian communities which form narrow bands along streams.The less rigid sampling arrangement of the second phase, where the releve was not placed in a fixed direction, but in a position to ensure maximum homogeneity, resulted in fewer transitions being sampled and thereby proved to be more satisfactory.
Although the mechanical table sorter was not used to re-arrange the data during this study, experience in its use indicates that it has certain advantages and disadvantages over the TABSORT programme.It is quicker to compare releves with one other and with developing groups with the mechanical table sorter, because direct comparisons are possible.This is most easily simulated with the TABSORT programme by strip-cutting the data, although this can be tedious and liable to error.Copies of new arrangements must frequently be made.The TABSORT programme has the advantage of providing an immediate neatly typed copy of the table of any required stage.This could be useful for comparison between stages of refinement, particularly for teaching purposes.The mechanical table sorting method is more liable to human trans cript errors.The size of the data matrix (150x363) was not limited by the computer capacity during this study but rather by the number of releves and species which could be efficiently dealt with in the actual arranging.The mechanical sorter data matrix size is limited by practical design and ease of handling (124 x 130 in the prototype).
Taylor found that association-analysis revealed groups which were ecologically meaningful, but that most of the groups represented ..." such small isolated fragments of natural units that they do not give a harmonious picture of the vegetation" (Taylor, 1969).In the first phase of the Hangklip survey, the ecologically meaningful groups required little sub jective ordering to form a more harmonious picture of natural units.The groups in the second phase were found, in contrast, to represent isolated fragments which would require considerable subjective rear rangement.This was in agreement with Taylor's findings.The latter result was unexpected, because care had been taken to ensure the most efficient sampling.
The Braun-Blanquet method was found to be more consistent, primarily because the communities are better defined in that they do not depend on single species presence or absence for final group forming.In addition the relationship between the communities is readily demonstrated.Donselaar who used the Braun-Blanquet method in the savannas of Northern Surinam is mentioned by Werger et al. (1972) as stating that the number of species must be moderate for this method to be successful.The latter workers, in con trast, found the method to be practical in floristically rich fynbos vegetation.During the present study more than 1 400 different species were collected.In the first phase releves containing 365 species (species occurring in fewer than three or less releves were not included in the analysis) were satisfactorily ordered in a table.This method, therefore, proved satisfactory with a fairly large data matrix.The data matrix, in contrast, proved too large for analysis using homogeneity func tions.Initial Braun-Blanquet groups had to be defined prior to further analysis.The homogeneity function tests on these groups showed them to be reasonable.Little further subdivision of these groups using homo geneity functions was effected although the degree of similarity between individual members and groups was readily demonstrated.
The Ceska & Roemer programme for identifiying species-releve groups only resulted in the delimitation of the major groups which were readily delimitable using less sophisticated and cheaper facilities.

A CK NO W LED GEM EN TS
The research undertaken during the first phase was submitted to the University of Cape Town for a M.Sc.degree under Dr A. V. Hall.
The Secretary of the Department of Forestry gave permission for data to be collected in a State Forest.
Considerable assistance was received from staff members of the Botanical Research Institute, the University of Stellenbosch, the Department of Forestry at Jonkershoek and the University of Cape Town.   A. V., 1969(a).Giving ranks and names to subsidiary groups.Taxon 18,4: 375-377.H a l l , A. V., 1969(b).Autom atic grouping programs: The treatm ent of certain kinds of properties.Biol.J. Linn.Soc.1: 321-325.H a l l , A. V., 1969(c)  Biomass in stands of about two years old ranged from about 2 200 kg per ha to about 7 500 kg per ha.Mature stands comprised about 11 000 to 15 000 kg per ha in heaths and 15 000 to 26 000 kg per ha in sclerophyllous scrub.The data indicate a maximum annual growth rate of 1 000 to 4 000 kg per ha early in the develop ment of a stand, but growth rates appear to decline rapidly as communities age.

R EFER EN C ES
Young stands are dominated by hemicryptophytes, which comprise about 2 000 to 6 000 kg per ha, or about 60 to 75 per cent of the biomass in stands of about four years old.Shrubs become prom inent later, but the hemicryptophytes persist.
The data indicate that the biomass, growth rates and the shape of the growth curves of fynbos communities are on the whole similar to those of analogous vegetation in other zones of mediterranean type climate.However, there are important structural differences in that analogues of the northern hemisphere (garrigue, chaparral) do not have a significant component of persistent hemicrytophytes.Although A ustralian heath communities do have this feature, the hemicryptophytes are not as prominent as in fynbos.

R E S U M E R A P P O R T P R E L IM IN A IR E S U R A B IO M A S S E D E L A V E G E T A T IO N
On a obtenu un echantillonnage de la bio masse de la vegetation aerienne en en recolt ant a divers endroits du maquis (fy n b o s) du sud-ouest de la Province du Cap.
Les donnees indiquent que la biomasse, les taux de croissance et la fo rm e de la courbe de croissance de ces maquis sont au total semblables a celles des associations vegetales analogues dans d'autres zones du type climatique mediterraneen.Neanmoins, il y a d'im portantes differences structurelles en ceci que les analogues de Vhemisphere N ord (garrigue, chaparral) n'ont p a s une constituante significative de sem i-cryptophytes persistants.Bien que les associations de bruyeres australiennes possedent cette caracteristique, les sem i-cryptophytes n 'y sont p a s aussi importants que dans le fynbos.

IN TRO D U C TIO N
The ecology of natural communities of Mediterranean-type ecosystems has recently received con siderable attention, particularly from the point of view of ecosystem convergence, and much more information on plant communities has become avail able (Specht, Rayson & Jackman, 1958, and previous papers;Specht, 1969aSpecht, , 1969b;;Jones et al., 1969;Mooney et al., 1970;di Castri & Mooney, 1973).However, few data on the Cape fynbosf have reached the press.
In this paper data have been collated on the biomass of fynbos communities, which have become available during the course of ecological studies from 1967 to 1974.The studies were not aimed at measuring com munity production nor are the data such that they may be used as direct measures of productivity.Nevertheless, they represent an index of productivity and contain other useful information.

STUDY AREAS
Biomass surveys were conducted on various sites in three research areas, each described below.1. Jonkershoek Forest Research Station (sites 1.1-1.4).The research area at Jonkershoek is situated at about 33°57'S and 18°55'E.The ecosystem has been * Jonkershoek Forest Research Station, Stellenbosch.+ Also known as sclerophyll bush (Adamson, 1938), and including the types described by Acocks (1953) as Macchia, False Macchia and Coastal Macchia.
described by Wicht et al. (1969).The communities sampled are situated on the slopes and near the bottom of a steep-walled valley (Fig. 1) and occur on soils derived from Cape granites.Soils are about one metre deep with a brown structureless loam A-horizon on a yellow-brown apedal B. They are acid, with pH ranging from about 4,50 to 5,00.Extractable phos phorus (citric acid extract) amounts to about 12 to 40 p.p.m. and total nitrogen and organic carbon con tent amount to 0,1 to 0,2 per cent and three to eight per cent respectively, in the A-horizon (Joubert, 1965).2. Zachariashoek Research Catchment (sites 2.1-2.3).This catchment research area is situated at 34°49'S and 19°02'E and has been described by van der Zel (1974).The communities studied are situated in the Kasteelkloof subcatchment (Fig. 2).Soils are derived from sedimentary orthoquartzites and shales of the paleozoic Table Mountain Group.The soils here have not been studied, but would resemble those at Jakkalsrivier rather than those at Jonkershoek.Site 2.2 is phreatic and the soil has an organic A-horizon.
Few climatic data are available.Rainfall at the top of the catchment amounts to about 1 300 mm per annum, and at the bottom, 1 100 mm per annum (six-year records at 701 m and 274 m a.s.l., respect ively).3. Jakkalsrivier Research Catchment (sites 3.1-3.10).Plathe & van der Zel (1969) and Kruger (1974) have described the Jakkalsrivier area in some detail (Fig.

A
c o c k s , J. P. H., 1953.Veld types of South Africa.M em .Bot. Surv.S .A fr.28.B e c k i n g , R. W., 1957.The Zurich-M ontpellier School of phytosociology.B ot. Rev. 23: 411-488.B o u c h e r , C., 1972.The vegetation o f the Cape H angklip area.U npublished M.Sc.thesis, University of Cape Town.B r a u n -B l a n q u e t , J., 1928.Pflanzensoziologie.Berlin: Springer.B r a u n -B l a n q u e t , J., 1932.(Transl.by G .D. Fuller and H. S. Conard).Plant sociology.New Y ork-London: M cGraw Hill.B r a u n -B l a n q u e t , J., 1964.Pflanzensoziologie.3rd ed.W ien-New Y ork: Springer.ig .4.-G roup similarity analysis from Boucher (1972).
has been sampled by harvesting on several sites in fynbos communities of the south western Cape Province.

TABLE 1
D., 1954.M ethods o f surveying and measuring vegeta tion.Bulletin 42, Comm onwealth Bureau of Pastures and Field Crops.Hurley-Reading: Bradley and Son.C e s k a , A. & R o e m e r , H ., 1971.A computer program for identifying species-releve groups in vegetation studies.Vegetatio 23: 255-277.C o e t z e e , B. J., 1972.'// Plantsosiologiese studie van die Jack Scott-N atuurreservaat.Unpublished M.Sc.thesis, Univer sity of Pretoria.D o w n i n g , B. H., 1966.The plant ecology o f Tabamhlope V'lei, Natal.Unpublished M.Sc.thesis, University of Natal.