Preliminary floristic analysis of the major biomes in southern Africa

Over 24 000 plant taxa are known to occur in the southern African flora, which is extraordinarily rich on a species/area basis. Lists of species and infraspecific taxa recorded for the six major biomes, Fynbos, Savanna, Grassland, Nama-Karoo, Succulent Karoo and Desert, were obtained from the PRECIS specimen database. These lists were analysed by numbers of unique and shared species and infraspecific taxa. by differential occur­ rence and life forms of large genera, and by differential occurrence of families. Each biome is floristically distinct except Nama-Karoo. The biomes form two main groupings, those with winter rainfall and those with summer rainfall. Succulent Karoo is most similar to Fynbos and Nama-Karoo is most similar to Savanna.


INTRODUCTION
The southern African flora is extrem ely speciesrich in term s of species/area ratios, with 0,0081 species/km 2 overall ( Figure 1). This value is higher than those recorded for humid tropical floras such as Brazil (0,0044) and Asia (0,0041) (Gibbs Russell 1985b). The winter rainfall Cape Floral Kingdom is well known to be extrem ely species-rich (G oldblatt 1978). H owever, even when the Cape flora is ex cluded from calculation, the species/area ratio for the rest of the southern African flora (0,0061) is still considerably higher than that of the humid tropics, and nearly twice that of Australia (0,0032), which also includes both tropical and tem perate areas.
These species/area ratios indicate in a superficial way that the rem arkable species richness of the southern African flora is not restricted to the Cape Floral Kingdom. The aim of this study is to investi gate the floristic richness of the m ajor biomes and to explore floristic relationships between these biomes using distribution data for families, genera and species.

M ETHODS
This study is based on checklists compiled from PRECIS for quarter degree latitude and longitude grids representing the m ajor biomes for southern Africa. The biomes adopted were determ ined by superimposing five recent treatm ents of southern African vegetation using floristic, structural and en vironm ental criteria (W erger 1978;Scheepers 1982based on Acocks 1975W hite 1983;Huntley 1984;R utherford & Westfall 1986). The resulting compos- ite map showed six m ajor regions that were recog nized as entities, even though none of the studies agreed on exact boundaries. Elimination of all areas of disagreem ent, and of areas smaller than a quarter degree, yielded the regions accepted as the core bi omes for this investigation ( Figure 2). Im portant en vironm ental characteristics of the core biomes are shown in Table 1.
The lack of agreem ent between the treatm ents occurred at three levels; 1, exact boundaries at quar  Besides the six core biomes adopted for this study, all the treatm ents recognized the high altitude vege tation of the D rakensberg, and the forests of the southern C ape. H ow ever, these small irregularly shaped areas were not accessible to com puter search at the scale of q uarter degree grid reference, and could therefore not be included.
The PR EC IS specimen database records label in form ation for ± 610 000 specimens in the National H erbarium (P R E ). The overall operation and imple m entation of PR EC IS have been reported several times (G ibbs Russell & Gonsalves 1984; Gibbs Rus sell 1985a). M ore recently, new program ming has allowed com pilation from the database of checklists of plant species and infraspecific taxa from any com bination of q u arter degree grids. Several special pro gramm es were w ritten to com pare the checklists by providing lists of unique taxa, lists of shared taxa, and a m atrix of all taxa with the biomes from which they were recorded.
The total num bers of unique and shared taxa, ob tained by employing these program m es, were used to calculate Sorenson's (1948) coefficients of similar ity, and percentages of unique taxa and taxa shared betw een biom es. The ranking of families for each biom e, the identification of widespread taxa, and the determ ination of centres of diversity for ia r g e ' gen era with 10 or m ore species and infraspecific taxa were obtained by manual searches of printout. A biom e was considered to be a centre of diversity for a genus if it contained 50% or more of the taxa re ported for the genus. In a few cases, slightly less than half (to 45% ) was accepted in biomes of low collect ing intensity. Life forms follow the definitions of R aunkiaer (1934) as stated by R utherford & W est fall (1986), but with the inclusion of 'Succulent', and were determ ined from Dyer (1975Dyer ( , 1976 and h er barium specimens. At all stages of w ork, doubtful records encountered on PRECIS listings were checked in PRE. An inherent weakness in the m ethod used is the uneven collecting intensity for the different biomes. Gibbs Russell et al. (1984) showed that the collecting intensity represented in PRECIS for the eastern and southern mesic areas is far higher than for the west ern arid areas. Therefore, the checklists used here undoubtedly differ in com pleteness, and it must be em phasized that these results are preliminary. Table  2 illustrates the differences in collecting intensity b e tween the biomes by comparing the specim ens and the taxa per km 2 as well as the specimens per taxon recorded in PR EC IS for each biome. A lthough Fyn bos, and to a lesser degree G rassland, appear to be better collected than the other biomes on a speci m ens/km 2 or taxa/km 2 basis, Savanna in fact exhibits m ore 'rep eat' collections than either. H ow ever, it is apparent that mesic Fynbos, Savanna and Grassland are better collected than arid N am a-K aroo, Succu lent K aroo and D esert.
PR EC IS is known to have errors in about 7% of specimen identifications and quarter degree grid re ferences. Until these errors can be corrected, an ongoing process in system m anagem ent, results must be used with discretion. In this study, identifications directly from PR EC IS are used only at the level of family and genus, while at the level of species and infraspecific taxa, only total num bers, and not iden tifications, are used unless the records were checked in PR E. For the same reason, distribution is given only at biome level, and not to individual quarter degree grids.
Despite the limitations imposed by differences in collecting intensity and by the accuracy of individual PRECIS records, at the present time PRECIS is the most reliable and com plete source of information about the distribution of taxa throughout the south ern African flora. Publication of these preliminary results is therefore considered worthwhile.
T hroughout the study, the num ber of species and infraspecific taxa, rather than species alone, were used in com parisons because of taxonomic uncer tainty about the correct level of treatm ent for many of these entities, as explained in detail in Gibbs Rus sell (1985b). For the sake of brevity, the term 'taxa' in this context is used in place of the longer phrase 'species and infraspecific taxa'.

Area, taxa and specimens
The area, taxa and specimens covered in this study are summarized in Table 3. The five recent vegeta tion treatm ents used to determ ine the biomes for this study agreed on about 40% of the total area of southern Africa at a scale available for com puter search. A bout 60% of all southern African taxa re presented in PREC IS were reported from the area designated. C ertain taxa were not included in the study for the following reasons: 1, they are known only outside the areas of the core biomes; 2, they are not represented in PREC IS; or 3, they are repre sented in PREC IS, but the distribution is not re corded as a quarter degree grid. Only about 25% of the specimens in PREC IS are reported in the study. This low figure results from the uneven collecting intensity in the National H erbarium m entioned above.

Comparison o f biomes by numbers o f species and infraspecific taxa
Widely differing numbers of taxa have been re corded for the six biomes, and the differences in taxon num bers are not related to the area sampled (Table 2). Fynbos has the most taxa although it is the smallest in area. Savanna, which covers by far the largest area, has about 1 500 fewer taxa than Fynbos.
Similarly, Grassland has about 1 700 more taxa than N am a-K aroo, although Nama-Karoo covers about twice the area of Grassland. Nama-Karoo and Suc culent Karoo have similar numbers of taxa, but N am a-Karoo covers about four times the area of Succulent Karoo. The num ber of taxa recorded for Desert is extrem ely low even though its area is slightly larger than that of Fynbos.
The checklists for each biome were com pared both by Sorenson's (1948) coefficient of similarity, and by percentage comparisons within each biome. Sorenson's coefficients give com parable values for checklists of different length. Low Sorenson's coef ficients signify low similarity between lists of taxa, while higher values show a greater similarity. The percentage comparisons show the proportion of taxa within each biome that are unique and that are shared with other biomes.

Sorenson's coefficients of similarity
The Sorenson's coefficients of similarity between the six m ajor biomes are shown in Figure 3. The values for the coefficients are generally low (30 or less), indicating that each biome has its own flora which is quite distinct from that of the others. The exception is the coefficient between Savanna and G rassland, which is considerably higher than any other.
For Savanna, the highest Sorenson's coefficients occur with Grassland and with Nam a-Karoo, and the values are low (less than 20) for the other biomes. G rassland, which has the strongest similarity to Savanna, has very low Sorenson's coefficients with Desert and with Succulent Karoo, and somewhat higher values with Fynbos and Nama-Karoo. Desert has very low values with all biomes except Nama-Karoo. Fynbos has low Sorenson's coefficients with all biomes except Succulent Karoo. Succulent Karoo and N am a-K aroo show opposite relationships. Ex cluding the Sorenson's coefficient between the two 'karroid' biom es, Succulent Karoo has its highest value with Fynbos, and very low values with D esert, Savanna and G rassland, whereas Nama-Karoo has its highest value with Savanna, high values with D e sert and G rassland, and a low value with Fynbos.

Percentages of unique and shared taxa
The percentages of taxa unique to each biome and shared betw een biom es are shown in Figure 4. The biomes vary greatly in percentages of unique taxa. Fynbos has the highest percentage (which is consist ent with a value of 68% given by Bond & Goldblatt (1984)), and Savanna is also well above the others. Grassland and Succulent Karoo are similar, and D e sert and N am a-K aroo have similar and very low per centages of unique taxa. The percentage of taxa shared between the bi om es amplifies the relationships shown by the So renson's coefficients. The few apparent contradic tions result from com paring taxon lists of very dif ferent length: where one list is long and the other short, the percentage of shared taxa differs m arkedly from the Sorenson's values.
The close floristic relationships between Savanna and Grassland and Savanna and Nam a-K aroo showr by the Sorenson's coefficients are borne out by the high percentage of Grassland and N am a-K aroo taxi that is shared with Savanna. Savanna itself shares m ost taxa with G rassland, shares the same percent age of its taxa with Nam a-Karoo as with Fynbos, and shares a very low percentage of its taxa with Suc culent K aroo and D esert. Grassland shares a very high percentage of its taxa with Savanna, and is simi lar to Savanna in that its lowest percentage of shared taxa is with Succulent Karoo and D esert, but G rass land shares a considerably higher percentage of taxa with Fynbos than with Nama-Karoo. D esert, which because of its small flora shows very low Sorenson's values with all biomes except N am a-K aroo, shares about the same very high percentage of its taxa with Savanna as with Nam a-Karoo. Desert has a very low percentage of unique taxa, and shares m ore than 20% of its taxa with Grassland, with Succulent Ka roo and with Fynbos. In contrast, Fynbos, which has a high percentage of unique taxa, does not share m ore than 20% of its taxa with any other biom e. The close relationships of Succulent Karoo to Fynbos and of N am a-K aroo to Savanna, already indicated by Sorenson's coefficients, are borne out by the high percentage of Succulent Karoo taxa shared with Fynbos, and the high percentage of N am a-K aroo taxa shared with Savanna. Both of the 'karroid' bi om es share the lowest percentage of taxa with D e sert and an interm ediate percentage with G rassland.

Comparison o f biomes by important fam ilies and large genera
D ifferential occurrence of im portant families Forty-six families comprise 1% or m ore of the taxa in at least one biome. In each biom e there are betw een 22 and 28 families that com prise 1% or m ore of the taxa, and that together account for be tween 55 and 60% of the total num ber of taxa. Each of these families can be used to distinguish and/or link the biom es (Table 4).
In order to com pare the biomes in this way, these im portant families are ranked for every biom e by num ber of taxa from largest (rank of 1) to smallest (rank of 22 to 28). Ranking is necessary for com parison at family level because the biomes differ so greatly in num ber of taxa. A family well represented in a species-poor biome may in fact have fewer taxa in that biome than the same family has in a biome with a rich flora, even though the family is a negligi ble com ponent of the more species-rich biome (G ibbs Russell 1975(G ibbs Russell , 1985b. The families are ranked in three groups in the discussion: the largest families (1-3 in bold type in Table 4); the next rank (4-10 in italics in Table 4); and the lowest rank (from 11 onwards in rom an type in Table 4). The biomes are characterized by the presence, absence or differ ence in rank of certain large families, and the occur rence of some families can be linked to simple en vironm ental param eters characteristic of certain com binations of biomes.
The seven plant families that comprise 1% or m ore of the taxa in all of the biomes are shown in Table 4a. Three families, A steraceae, Poaceae and Fabaceae are the three largest in all biomes (with the exception of Poaceae in Succulent Karoo and Fyn bos), and either A steraceae or Poaceae is the largest family in all biomes. Asteraceae and not Poaceae is the largest family in Grassland.
The six biomes are briefly discussed in turn below: Fynbos (Table 4b) is distinguished by eight fami lies that are im portant in no other biome. Of these, Ericaceae is one of the three largest families, and  R estionaceae, R utaceae and Proteaceae among the ten largest families in Fynbos only. In contrast, A s clepiadaceae is not im portant, and only in Fynbos is Scrophulariaceae not one of the ten largest families. Savanna (Table 4c) is distinguished by one im port ant family, V erbenaceae, one family, Rubiaceae, that ranks am ong the ten largest in no other biome, and one family, M esem bryanthem aceae, that does not occur am ong the im portant families. Grassland (Table 4d) is distinguished by the high rank of Orchidaceae and Lam iaceae, which are among the ten largest families only in this biom e, and only here are Sterculiaceae and Aizoaceae absent from the im p ortant families. N am a-Karoo (Table 4e) is the only biome which is not distinguished from the others by differential occurrence of families. Succulent Karoo (Table 4f) is distinguished by the high rank of Iridaceae, which is one of the three largest families, and Crassulaceae and G eraniaceae, which are among the ten largest families only in this biome. Desert (Table  4g) is distinguished by the occurrence of Pedaliaceae and B urseraceae as im portant families, by the occur rence of C henopodiaceae and C apparaceae among the ten largest families, and by the absence of Iridaceae as an im portant family.
A num ber of families indicate floristic relation ships betw een biom es with different rainfall seasona lity and am ount. The four sum m er rainfall biomes (Table 4h) are variously linked by 11 families that do not occur as an im portant com ponent of the winter rainfall biom es. W inter rainfall biomes (Table 4i) are linked by the occurrence of three families, Pro teaceae, O xalidaceae and C am panulaceae, that are not im portant in sum m er rainfall areas, and one family, Poaceae, that ranks first or second in sum m er rainfall biom es, but has a lower rank in the win ter rainfall areas.
In contrast to the above groupings based on rain fall seasonality, other families link biomes with simi lar am ounts of rainfall. The arid biomes are linked by four families (Table 4j). C henopodiaceae and Zygophyllaceae are im portant, and Aizoaceae and M e sem bryanthem aceae are among the ten largest fami lies only in the arid biomes. Finally, a group of six families, all with low ranking, weakly links the sum m er rainfall biom es G rassland and Nam a-Karoo to the w inter rainfall biom es (Table 4k). Savanna is not linked to the w inter rainfall biomes at family level.

C entres of diversity of large genera
The large genera (with 10 or m ore taxa) with centres of diversity in one, two or three biomes are listed in A ppendices 1-3. The large genera with no apparent centre of diversity are listed in Appendix 4. Figure 5 sum m arizes this inform ation by showing the num bers and percentages of large genera with centres of diversity within and shared between the biomes.
Only in the case of Fynbos and Savanna are more than half the large genera centred in a single biom e, w hereas each of the other four biomes shares more than half its large genera with another biome. The highest num ber of large genera have their centre of Savanna or with Nama-Karoo. No genera have centres of diversity in both Desert and Grassland, Desert and Succulent Karoo, or Desert and Fynbos.
Life forms and centres of diversity of large genera Figure 6 shows life form spectra for large genera with centres of diversity either only in one or in more than one biome. The basis for plant classification is that floral characters are conservative at family and genus level, whereas vegetative characters can be variable between members of a higher category. R aunkiaer's life forms indicate broad basic differ ences in vegetative states, depending on the position of the perennating bud, and indicate differences in utilization of resources. The fact that a genus has many species and infraspecific taxa in a certain bi ome suggests that the adaptations displayed by the taxa are com patible with the environm ent of that biome. Thus the differences in characteristics of the genera, as illustrated by life forms, can show conver gent adaptations in a num ber of separate evolution ary lines to the conditions in the biome. However, a centre of diversity for a genus in a particular biome does not imply that speciation occurred either in that biom e, or under current environm ental conditions. The biomes are characterized by differences in the life forms reported in the large genera. In Fynbos, cham aephytes are the most commonly reported life form in the large genera. In Savanna, phanerophytes are reported more frequently than in any other bi ome. For Grassland, nearly half the life forms re ported are hemicryptophytes. Grassland differs from Savanna by having fewer phanerophytes and cham aephytes (the woody com ponent), and from Nama-K aroo by having fewer cryptophytes. Nama-Karoo is similar to Grassland, but with more cryptophytes reported among the few genera with their centre of diversity in Nama-Karoo only. Succulent Karoo is rem arkable because it has similar values for cham ae phytes, hemicryptophytes and cryptophytes. The com parative value for cryptophytes is far higher than for any other biom e, and phanerophytes are not re ported at all. The life form spectrum for Desert may be misleading because it is based on three genera only, and therefore it is not considered further.
The differences in occurrence of each of the life forms in the biomes can also be examined. Phanero phytes appear only in genera with a centre of diver sity in Fynbos or Savanna. Chamaephytes and hem i cryptophytes show a basic difference between the summ er and the winter rainfall biomes. C ham ae phytes are reported most often in winter rainfall Fynbos and Succulent Karoo. Hemicryptophytes are the most abundant life form in the sum m er rainfall Savanna, Grassland, Nama-Karoo and Desert. C ryptophytes occur in low numbers in all the bi om es, but are reported often only in genera with their centre of diversity in Succulent Karoo, and to a lesser extent, Nama-Karoo. Therophytes are re ported in all biomes, but are less frequently re ported in genera of which the centre of diversity is confined to a single biome, and are more frequently reported in genera with centres in more than one biome. Succulents are reported in all biomes, but the mesic biom es Fynbos, Savanna and G rassland, have succulents reported in genera with centres in each one, while the more arid Succulent Karoo and N am a-K aroo have succulents reported only in gen era with centres of diversity in more than one biome.

Fynbos
Fynbos has the largest num ber of taxa, the highest percent of unique taxa, the largest num ber of im portant families that do not occur in any other bi om e, and the greatest num ber of centres of diversity for large genera. A t species level, the Sorenson's coefficient of similarity and the percentage of shared taxa show that Fynbos is most closely related to Suc culent K aroo, the other winter rainfall biome. At the generic level, Fynbos shares m ore centres of diver sity for large genera with Succulent Karoo than with any o ther biom e. At the family level, Fynbos is linked only to Succulent Karoo by four im portant families. The less-m arked relationship between Fyn bos and G rassland will be discussed under Grassland below.

Savanna
Savanna is second to Fynbos in num ber of taxa, percentage of unique taxa, and in num ber of centres of diversity for large genera. However, Savanna is distinguished at family level by only three families, while it is linked to the other sum m er rainfall biomes by eight families. The closest relationship of Sav anna is to G rassland, as shown by the very high So renson's coefficient of similarity and the percentage of shared taxa, the num ber of large genera with centres of diversity in both Savanna and Grassland, and the six families that link them , three of which are im portant only in Savanna and Grassland. A w eaker relationship betw een Savanna and Nama-Karoo is shown by a high Sorenson's coefficient and the percentage of shared taxa, a considerable num ber of large genera with centres of diversity in both Savanna and N am a-K aroo, and by four families that link them .

Grassland
A m oderately large num ber of taxa is reported for G rassland, which is distinguished by four families. Its relationship with Savanna is the closest dem on strated in this study, as discussed above. Grassland shows similar m oderate Sorenson's coefficients with both N am a-K aroo and Fynbos, but other com pari sons show that G rassland is in fact more closely re lated to Fynbos than to N am a-K aroo. The percent age of G rassland taxa shared with Fynbos is far higher than the percentage shared with N am a-Ka roo, and a num ber of large genera, nearly all hem i cryptophytes, have centres of diversity in both Grassland and Fynbos, while no large genera have centres of diversity in both Grassland and Nama-Ka-roo. A t family level, Grassland is linked to Nama-K aroo only by families that also link it to Savanna (Table 4h) or to Fynbos (Table 4k), while it is linked independently to Fynbos by two families (Table 4k).
N am a-K aroo N am a-K aroo is not well defined floristically in this study. At species level, its num ber of taxa is low, particularly with respect to its large area, and the percentage of unique taxa is very low, hardly higher than that of D esert. Nam a-Karoo is the only biome for which all Sorenson's coefficients except one (to Fynbos) are greater than 20. Over half of N am a-K a roo taxa are shared with Savanna, about a third are shared with Grassland and another third with Fyn bos. A t generic level, few large genera have a centre of diversity in N am a-K aroo, and of these, m ore have a shared centre of diversity with Savanna, with Fyn bos or with Desert than are centred in N am a-K aroo alone. At family level, Nam a-Karoo is the only bi om e that cannot be defined by differential occur rence of im portant families. It is linked to all the other sum m er rainfall biomes, and also to the winter rainfall Succulent Karoo through the arid biomes.

Succulent Karoo
The num ber of taxa reported for Succulent Karoo is similar to that of Nam a-Karoo, but the area cov ered is about a quarter as large, and Succulent Ka roo has m ore unique taxa. It is distinguished from other biomes by three im portant families. Succulent K aroo is shown by Sorenson's coefficients, by p er centage of shared taxa and by centres of diversity of large genera to be related floristically both to Fynbos and N am a-K aroo. The much higher values in every case show that the relationship is strongest to Fynbos (see Fynbos above). Over half the Succulent Karoo taxa and over three quarters of the large genera are shared with Fynbos. A t family level, the strong links of Succulent Karoo to Fynbos are shown by four families that are im portant only in these two biom es, w hereas at family level Succulent Karoo is linked to N am a-K aroo only through the group of families that links the three arid biomes.

D esert
A very small num ber of taxa are reported for D e sert, and the percentage of unique taxa is lower than for any other biome. There are no large genera with a centre of diversity in D esert alone. How ever, D e sert is distinguished by four im portant families. R e lationships of the Desert flora are shown by Soren son's coefficients and by the percentage of shared taxa, to be highest with Savanna and with Nama-Karoo, and it is only with these two biomes that Desert shares centres of diversity for large genera. In ad dition, D esert is linked to Nam a-Karoo by ten fam i lies, two of which are im portant only in D esert and N am a-K aroo, and it is linked to Savanna by four families, one of which is im portant only in Desert and Savanna. D esert is also linked to the arid but winter rainfall Succulent Karoo by four families.

Relationships
The distribution of species, genera and families and the life form spectra shows that the biomes fall floristically into two groups, which correspond to the summer rainfall region (Savanna, Grassland, Nama-Karoo, Desert) and the winter rainfall region (Fyn bos, Succulent K aroo). The present analysis of 14 000 taxa therefore supports and extends the 'win ter rainfall biom e' concept first put forward on the basis of a few genera by Bayer (1984). A detailed study of grass subfamily distributions also shows a similar basic division, with Chloridoideae and Panicoideae most abundant in summer rainfall areas and A rundinoideae most abundant in winter rainfall areas (Gibbs Russell 1986).
Nam a-Karoo and Succulent Karoo, which have previously been placed together at highest level in all vegetation studies except that of R utherford & Westfall (1986), are not closely related floristically. Nam a-Karoo is more closely related to Savanna than to Succulent K aroo, and Succulent Karoo is more closely related to Fynbos than to Nama-Karoo.
W ithin the sum m er rainfall group, at species level, the strongest relationship is between Savanna and G rassland, with a w eaker relationship between Sa vanna and Nam a-Karoo. The same relationships are shown at generic level, and the distinctness of Nama-Karoo from Grassland and of Desert from Succulent Karoo is emphasized. A t family level, particular fa milies link and dem arcate the summer rainfall bi omes and the winter rainfall biomes, but another group of families complicates this simple difference by linking the arid biomes of both summer and win ter rainfall regimes.
Secondary links connect the two m ajor groups through N am a-K aroo, which lies between the other biomes geographically. Nam a-Karoo is ill-defined as an entity, and is strongly linked at species, genus and family level to Savanna and D esert; it is more weakly linked at species and family level to Succu lent Karoo and at genus level to Fynbos. Grassland, which is very strongly allied to Savanna, shows a secondary link to Fynbos, independent of Nama-Karoo, at species, genus and family level.

CONCLUSIONS
At the highest level of floristic comparison the winter rainfall biomes and the summer rainfall bi omes form two separate groups. Within these groups, each biome is floristically distinct at the level of species and infraspecific taxa, whether m easured by Sorenson's coefficient of similarity or by percent age of shared taxa, and each biome (except Desert) is rich in taxa. Each is a centre of diversity for certain large genera, and the life form spectrum for these genera is different for each biome. Each (except Nam a-K aroo) is distinguished by differences in the occurrence of im portant plant families.
The floristic distinctness of the biomes, coupled with high taxon num bers, implies that each should be studied and m anaged as a separate entity. Be cause of the high num bers of species and infraspe cific taxa, it is unlikely that conservation of limited areas in nature reserves will protect a large propor tion of the taxa in any one biome.
This study is ham pered by the dearth of specimen records from arid areas, and for this reason it may be criticized for being too preliminary. However, the trends indicated should serve as stimulus to more precise analyses. Unfortunately precision can only be achieved when primary data are available to com pile more complete and accurate checklists. This should be done through bringing together records from many herbaria and from literature, and most im portant, through rationally planned specimen col lecting designed to cover all biomes adequately.
The conclusions are based on plant distributions as they are now known, that result from interactions over a long geological, climatological and evolution ary history. It is not apparent to what extent these distributions have been influenced by present or past environm ents. However, listing and comparing the taxa in each biome is the first step in unravelling the events that have led to the formation of its character istic flora. PRECIS has given us a preliminary look that will allow the generation of hypotheses for more rigorous testing using stronger data sets and more refined techniques.