BS 55 229 A numerical re-evaluation of the sub-Saharan phytochoria of mainland Africa

H. PETER LINDER, JON LOVETT, JENS M. MUTKE, WILHELM BARTHLOTT, NORBERT JÜRGENS, TONY REBELO AND WOLFGANG KÜPER

LINDER, H.P., LOVETT, J., MUTKE, J.M., BARTHLOTT, W., JÜRGENS, N., REBELO, T. & KÜPER, W. 2005. A numerical re-evaluation of the sub-Saharan phytochoria of mainland Africa. Biol. Skr. 55: 229-252. ISSN 0366-3612. ISBN 87-7304-304-4.

The delimitation of the sub-Saharan mainland African phytochoria was investigated by cluster analysis and non-metric multidimensional scaling of the distributions of 5438 species, recorded from 1918 one-degree grid squares. The clusters obtained were in many instances very similar to the phytochoria delimited by White. The Guineo-Congolian Regional Centre of Endemism (RCE) was retrieved with almost the same borders, including the northern and southern transi- tion zones and the Lake Victoria Regional Mosaic (RM). A larger Zambesian phytochorion was found – this included the Zanzibar-Inhambane Regional Mosaic, as well as part of the Somali- Masai RCE and all of the Ethiopian and Kenyan parts of the Afromontane RCE. In southern Africa the Cape RCE, the Namib-Karoo RCE, as well as an expanded Tongaland – Pondoland RM, which included all the eastern slopes of the subcontinent were located. The central parts of the subcontinent (Kalahari-Highveld Regional Transition Zone (RTZ)) was expanded to include the Drakensberg, but divided into a south-eastern and north-western unit. None of the regional mosaics were retrieved, and the blocks of the Afromontane RCE were included in the various phytochoria in which they are embedded. Cluster analysis retrieved a Sudanian phytochorion, but ordination suggested that the delimitation between the floristic zones in West Africa is com- plex, and that there may be very broad transitions from one phytochorion to the next.

H. Peter Linder, Institute of Systematic Botany, Zollikerstrasse 107, CH-8008 Zurich, Switzerland. E-mail: [email protected]

Jon Lovett, Environment Department, University of York, York YO10 5DD, England. E-mail: [email protected]

Jens Mutke, Botanical Institute Bonn, Meckenheimer Allee 170, 53115 Bonn, Germany. E-mail: [email protected]

Wilhelm Barthlott, Botanical Institute Bonn, Meckenheimer Allee 170, 53115 Bonn, Germany. E-mail: [email protected]

Norbert Jürgens, Institut für Allgemeine Botanik der Universität Hamburg, Ohnhorststr. 18, 22609 Ham- burg, Germany. E-mail: [email protected]

Tony Rebelo, National Botanical Institute, P. Bag x 07, Claremont, South Africa. E-mail: [email protected]

Wolfgang Küper, Botanical Institute Bonn, Meckenheimer Allee 170, 53115 Bonn, Germany. E-mail: [email protected] 230 BS 55

Introduction However, because a rigorous definition of these concepts seems not to be possible, we are “If the distributions of all species are used as left with almost arbitrary decisions as to a basis for the subdivision of a Region, no whether a particular distribution pattern is a clear picture will emerge or it will be consid- phytochorion, vegetation type or historical erably obscured, since the distinctive pat- track. White did not attempt to distinguish terns of any particular ecological element between historical and ecological causes of the will be masked by those of other ecological distribution patterns, and elaborated a set of elements or swamped in the mass of statistics principles as rules by which to delimit his phy- produced by such a crude approach.” tochoria (White 1965, 1971, 1983, 1993). (White 1965, p. 652). These can be summarized as follows: 1) The classification should not be hierarchi- In a remarkable series of papers from 1965 to cal. Instead of Kingdoms, Regions and 1993 Frank White produced an integrated phy- Provinces, areas are not ranked, but encom- tochorological classification of Africa. This pass “centres of endemism”. Since they are geo- classification, which has become very widely graphically delimited, he referred to them as used (his “The vegetation of Africa” has been “regions”. cited at least 600 times to date, according to 2) The delimitation of the choria is based the Web of Science citation index, February only on species distributions. White often 2003), is remarkable in that it is based on referred to these distributions as the “facts”. clearly formulated theoretical principles, a Thus neither the vegetation types, nor higher broad empirical knowledge of the distribution taxonomic groups like genera, were taken into patterns in the African flora, and access to vast account. field knowledge. The classification synthesized 3) Three types of regions are recognised. the knowledge accumulated in the post Sec- Regional Centres of Endemism (RCE) contain ond World War botanical exploration of Africa, at least 1000 endemic or near-endemic species, and exchanged at the regular meetings of and at least 50% of its phanerogam flora AETFAT. should be endemic. Between centres are broad Phytochoria may be defined as large areas Regional Transition Zones (RTZ), which (c. 10,000 km2 or more) with largely homoge- always have fewer than 1000 endemic species, nous plant species composition, which is differ- and these make up less than 50% of the flora. ent from that of other phytochoria. The distri- Regional Mosaics (RM) constitute a mosaic of bution of species in sub-Saharan Africa is deter- vegetation types, and an intermingling of oth- mined by modern climates and soils, as well as erwise distinct floras. White (1979) discussed past climates. This is manifested by three con- the concept of a RCE in detail. Although cepts: semantically a centre has no dimensions, he 1) The phytochoria should reflect the zonal used the concept in a territorial sense, to flora, the flora found in the general area, and describe a region with a high concentration of the zonal vegetation type. species, which are largely endemic to this 2) Each phytochorion has many ecological region. The other two concepts describe the habitats, occupied by an azonal flora (azonal transitions between these RCE. vegetation types). 4) Although in theory all species should be 3) Relicts of previous climates which show taken into account, the delimitation is based disjunct distributions are labelled as tracks. largely on the dominant species or groups (the BS 55 231 zonal flora), which can be studied in detail were obtained from numerous datasources, (White 1971). In effect this means largely the including the Distributiones Plantarum Africa- trees. narum, published by Jardin Botanique The phytochorological system provides an National de Belgique (Anonymous 1969-), the efficient framework within which to summarize Flora of Jebel Marra (Wickens 1976), the Pro- information about the flora and vegetation of tea Atlas Project (Rebelo 1991), the arid flora Africa. As such it is a ready source of informa- of North Africa (Frankenberg & Klaus 1980), tion on species richness, endemism, as well as the SIG Ivoire project (Chatelain et al. 2001), peculiar features of the flora, vegetation or South African arid plant distribution data (Jür- biota of all the parts of Africa. gens 1997), and numerous taxonomic revi- White’s phytochorological classification of sions (e.g., Linder 2001a; Linder & Ellis 1990; Africa forms a coherent system, which cannot Linder & Kurzweil 1999; Polhill 1982, pers. be evaluated critically in part only. Continuing com.). Raw data in the form of maps were digi- with his research methodology, mapping tized with a digitizing tableau (see also La Ferla species or groups of species over the phytocho- et al. 2002 for futher method descriptions). ria, constitutes no test of the system: at most Further datasets (Protea Atlas data, distribu- there will be no support for it. This is illus- tion data of southern African Orchidaceae, trated by White’s (1990) analysis of a number Restionaceae and the grass genus Pentaschistis) of disjunct species. A more interesting were obtained as one quarter degree resolu- approach would be to employ precisely the tion presence / absence data for each species. methods White argued against (see the citation The data were integrated into Microsoft above), using a large, ad hoc selected set of ACCESS databases and analysed with the help species, and objective numerical methods, to of the programmes ArcView GIS 3.2 and evaluate the support for African phytochoria. WorldMap 4:20:17 (Williams 2002). Based on Here we use the largest body of sub-Saharan original data the resolution of which varies plant distribution data assembled to date, and between exact coordinates of the plant record subject it to cluster analysis as well as ordina- localities and nearest one degree square, all tion, to establish (1) whether a numerical clus- data were rescaled to a one degree gridded res- ter analysis of the present data delimits phyto- olution. This resolution is a compromise choria similar to those of White, (2) whether between the loss of biogeographical detail on the transition zones recognised by White can the one hand and the sampling inadequacy of be identified from cluster analysis and ordina- finer resolution as well as the standard software tions, and (3) whether our dataset and meth- capacities on the other hand. It also guarantees ods of analysis are sufficient to detect comparability to current zoological analyses “unusual” phytochoria such as archipelago-like (Brooks et al. 2001; Burgess et al. in prep.) and regional centres and regional mosaics. former studies on African plant diversity (Denys 1980; Linder 1998, 2001c; Lovett et al. 2000), as well as animal distribution data (e.g., Material and Methods de Klerk et al. 2002). The geographical cover- The plant dataset age of the dataset is limited to the African con- The plant dataset analysed in this study com- tinent south of 20° N and comprises 1918 one- prises 79,648 data points for 5438 taxa, c 13% degree grid squares. of the total sub-Saharan African flora (species Parts of the data are presented at the follow- as well as infra-specific taxa). Distribution data ing addresses: 232 BS 55

www.york.ac.uk/res/celp/webpages/projects/ the entire dataset. Besides, it would be worldmap/worldmap.htm (York) extremely difficult to label the almost 2000 www.nbi.ac.za/protea (Cape Town) points on one ordination. Consequently, we www.botanik.uni-bonn.de/system/biomaps/biota/ floristicdatabases.html (Bonn) used NMDS to explore regional patterns: the transition between the rainforest and the The current continental database is adminis- desert in West Africa, and the phytogeographi- trated and regularly updated as part of the Bio- cal patterns in east Africa and southern Africa. geographical Information System on African A Principal Coordinates Analysis (PCOA) Plant Diversity (URL see above), which is estab- was first performed, by extracting the eigenval- lished by the Biomaps Working Group, Bonn, ues from the double-centered Jaccard similar- as part of the BIOTA Africa Project (www.biota- ity matrix. The results from the PCOA were africa.de). then used as starting configuration for the NMDS, done for three dimensions with the Multivariate Analyses double-centered Jaccard similarity matrix. As a The data were assembled into a square table of test of fit, the stress was calculated. Each ordi- grid-cells (1918) and taxa (5438), and the simi- nation was presented as a three-dimensional larity between every pair of grid-cells calcu- plot, and the cells labelled with the phytocho- lated using the Jaccard Coefficient of similarity ria assigned on the basis of the cluster analysis. (Jaccard 1901). The Jaccard Coefficient is par- All analyses were done using NTSYS-pc (Rohlf ticular suitable for large phytogeographical 1998). analyses because it does not take shared absences into account (Jardine 1972). Calculation of regional richness and The cells were clustered using the UPGMA endemism algorithm. The cluster analysis found at least The species richness and endemism for each of one tie, consequently a set of 50 dendrograms the delimited phytochoria were calculated was calculated. However, it was not possible to from the underlying dataset in ACCESS. We build a consensus tree, since there were more used both the phytochoria as delimited by items than the consensus-building program White, as well as those determined from the could accommodate. Instead several trees were present cluster analysis. For the initial analysis visually inspected to establish whether there cells, which could not be placed, were ignored were major differences between them. Clusters (not counted for species richness, nor used in were not recognised at a consistent level of sim- the determination of endemism). Because ilarity (the “phenon-line” approach), but we most phytochoria are not geographically rather searched for large groups of cells that coherent (e.g., include geographically isolated clustered together, and investigated the sub- cells), a second analysis with “simplified” phy- clustering within these clusters. tochoria was performed. Here the phytochoro- Cluster analysis assumes a hierarchical struc- logical assignment of the cells was changed ture in the data, and consequently may “force” according to the following rules. Unassigned clustering, thus distorting the true distances cells that had at least seven of the eight neigh- between the cells (Sneath & Sokal 1973). Non- bouring cells belong to one phytochorion were metric multidimensional scaling (NMDS) was assigned to that phytochorion. Cells assigned used to present these distances, as it maintains to one phytochorion but embedded within the same order of similarity as indicated in the another phytochorion (all eight neighbouring data. It was not technically feasible to ordinate cells belonging to the same, other phyto- BS 55 233 chorion) had their assignation changed. Cells this group, but not resolved to one of the sub- assigned to a phytochorion, but not in contact groups). Recognising these subgroups, as well with any other cells of that phytochorion, but as the clusters of unplaced cells, resulted in 19 bordering on cells assigned to more than one groups or phytochoria (Fig. 1). These phyto- phytochorion, were changed to “unassigned.” choria were labelled with names somewhat dif- In a small number of cases, almost all involving ferent from the phytochoria names used by outliers of the Somalian phytochorion in the White, in order to keep the two sets of concepts Kalahari and the Sahelian phytochoria, the distinct. The cells placed to a major group, but rules were interpreted in a more relaxed fash- not to one of the subgroups, are labelled as ion to allow clusters of outliers to be trans- “undifferentiated”. ferred to the host phytochorion. This resulted in a geographically more coherent set of phyto- Richness and endemism of phytochoria choria. The species richness and endemism for the The question whether the level of endemism phytochoria as delimited by White (1983), and reported here for the phytochoria is higher calculated from our dataset, are presented in than random could not be addressed, since Table 1. We sampled an average of 24.3% there is a dramatic variation in the proportion (between 12% (Guineo-Congolian RCE) and of range-restricted species in sub-Saharan 48% (Sahel RTZ)) of the species richness of Africa (Kier & Barthlott 2001; Linder 1998, each phytochorion. We seemed to have less 2001c). Consequently the variance around any success in sampling the endemic species, statistic of the continent-wide average ranges of retrieving an average of 18% of the endemic the species would be enormous and difficult to species as predicted by White (between 0.6% interpret. (Sudanian RCE) and 68% (Sahara RTZ)). There is enormous variation in the species richness and endemism of the phytochoria Results delimited in the present study (Table 2). For Cluster Analysis the broad phytochoria richness ranges from The dendrogram for the whole sub-Saharan 410 species (Somalian) to 2612 species (south- Africa shows the major groupings only at a very ern Africa), and endemism from 1% (Sudan- low level of similarity. At the broadest level, six ian) to 85% (Southern African). For the nar- groups were delimited, as well as a number of row phytochoria (excluding the transitions, cells that were not clustered, but linked more which did not form groups) the range is from or less directly to the stem of the dendrogram. 168 species in the Sahara, to 1822 species in This lack of clustering may be the result of a the Cape. Endemism ranges from 0% for the low number of species recorded for some cells Sudanian-north to 77% for the Cape. There is (236 cells with less than 5 species). Cells with no obvious relationship between area and few species are often unplaced, probably due endemism. to sensitivity to sampling stochasticity. These The calculated results for the “simplified” unplaced cells were not mapped to any phyto- phytochoria (Table 3) show clear trends: the chorion, but shown as blank cells (with those species richness of the areas is more or less the cells for which no data were available) on the same as for the unmodified phytochoria, but phytochorological map (Fig. 1). Five of the six the endemism is generally higher. For the groups showed clear internal subgroups (as Sudanian, Somalian and Angolan phytochoria well as a number of cells that were assigned to the richness is substantially reduced, and for 234 BS 55

Fig. 1. The mapped results of the cluster analysis. Country boundaries are indicated in black, the boundaries of the White Regions are indicated in white, and the phytochoria retrieved in our analysis are colour coded.

the Ethiopian – Kenyan part of the Zambezian but that they are largely recognisable. How- phytochorion the richness increases substan- ever, it is evident that the data are not readily tially. The only reduction in endemism is seen shoe-horned into three dimensions, as indi- for the Congolian part of the Guineo-Congo- cated by the final stress values of 0.64241 for lian phytochorion. East Africa, 0.57490 for West African and 0.62141 for southern African. Values more Ordinations than 0.40 are regarded as a poor fit (Rohlf The NMDS ordinations (Fig. 2) show that the 1998). clusters overlap to a greater or lesser extent, BS 55 235

Fig. 2. NMDS ordination of (a) West Africa, (b) East Africa and (c) southern Africa. The circles are labelled with the phy- tochoria they represent, but not all cells are always included in the circles. The area codes used for all the ordinations are: A = Sudanian undifferentiated; B = Sudanian-north; C = Sudanian-south; D = Congolian undifferentiated; E = Upper Guinean; F = Congolian; G = Congolian transitions; H = Zambezian-Angolan; I = Zambezian-central; J = Zambezian- Ethiopian-Kenyan; K = Cape; L = Eastern Karoo; M = Natal; N = Namib Karoo; O = Kalahari; P = Sahelian; R = Saharan; S = Somalian; T = undifferentiated southern African; X = unplaced. 236 BS 55

Table 1. Species richness and endemism of the phytochoria as delimited by White (1983), comparing his estimates of the species richness and endemism with that obtained from our data.

White’s estimates Our data Proportion sampled Species richness endemic species Number species % endemic Species richness endemic species Number species % endemic species Endemic species

Phytochorion name

Guineo-Congolian RCE 12000 6400 53 1375 399 29 11.5 06.2 Cape RCE 08600 5870 68 1599 838 52 18.6 14.3 Zambezian RCE 08500 4590 54 1725 377 22 20.3 08.2 Karoo-Namib RCE 07000 3500 50 1036 201 19 14.8 05.7 Afromontane RCE 04000 3000 75 1564 078 05 39.1 02.6 Somali-Masai RCE 04000 1250 31 0931 103 11 23.3 08.2 Kalahari/Highveld RTZ 03000 0050 02 0583 010 02 19.4 20.0 Lake Victoria RM 03000 0050 02 0504 003 01 16.8 06.0 Tongaland-Pondoland RM 03000 1200 40 0813 084 10 27.1 07.0 Zanzibar-Inhambane RM 03000 0450 15 0576 048 08 19.2 10.7 Sudanian RCE 02750 0960 35 0684 006 01 24.9 00.6 Guinea-Congolia/Sudania RTZ 02000 0050 03 0711 005 01 35.6 10.0 Guinea-Congolia/Zambezia RTZ 02000 0050 03 0571 028 05 28.6 56.0 Sahara RTZ 01620 0050 03 0289 034 12 17.8 68.0 Sahel RTZ 01200 0050 04 0579 023 04 48.3 46.0

Discussion pretation. While White recorded 53% endemism for his region, our data indicate that Phytochoria only 29% of the species in our sample are Guineo-Congolian restricted (endemic) to this Centre (Table 1). The Guineo-Congolian Regional Centre of The Upper Guinean phytochorion is delim- Endemism (RCE) was retrieved with minor ited to the east by the Cross River, immediately variations in the delimitation from the sur- west of Mt Cameroun, with the lower Niger and rounding phytochoria. These involve the tran- its delta clearly included in Upper Guinea. The sition zones to the north (Guinea- Cross River boundary was suggested by Congolia/Sudania Regional Transition Zone Léonard (1965) and Clayton and Hepper (RTZ)), the east (Lake Victoria Regional (1974), but White could not recognize it from Mosaic (RM)) and the south (Guinea-Congo- his data. The major deviation from White’s lia/Zambezia RTZ). No East African coastal delimitation is the inclusion of the Fouta Djal- outliers of the Guineo-Congolian RCE were lon in the Upper Guinean phytochorion, while identified, thus corroborating White’s inter- White placed it in his Guinea-Congolia/Suda- BS 55 237

Table 2. Species richness and endemism in the phytochoria as delimited by our analysis, based on the data used in this paper. phytochoria Broad Species richness endemic species Number species % endemic phytochoria, Narrow Species richness endemic species Number species % endemic

Sudanian 0729 0008 01% Sudanian undifferentiated 0452 0001 00% Sudanian-north 0307 0000 00% Sudanian-south 0494 0003 01% Guineo-Congolian 1708 0739 43% Congolian undifferentiated 0750 0014 02% Congolian 1046 0200 19% Congolian + Congolian undifferentiated 1177 0460 39% Upper Guinean 0578 0061 11% Congolian transitions 1059 0099 09% Zambezian 1886 0766 41% Zambezian – Ethiopian-Kenyan 0506 0042 08% Zambezian-central 1472 0497 34% Zambezian-Angolan 0810 0090 11% Southern African 2612 2223 85% Karoo transition 0046 0000 00% Cape 1822 1396 77% Eastern Karoo 0320 0004 01% Natal 0882 0295 33% Namib-Karoo 0339 0065 19% Kalahari 0294 0017 06% Sahara-Sahelian 0474 0182 38% Sahara 0168 0001 01% Sahel 0420 0173 41% Somalian 0410 0047 11% Somalian 0410 0047 11% nia RTZ. The Upper Guinean phytochorion is The Congolian phytochorion includes both rather species poor (564 species, compared to Lower Guinea and the Congo basin. We the 1137 for the Congolian phytochorion) and included in the phytochorion both the “Con- with a lower level of endemism (13% com- golian” and “Congolian-undifferentiated” pared to 30%; Table 3). Note that the species cells. The cells of the “Congolian undifferenti- numbers, and percentages of endemism, ated” group generally have fewer species than referred to here and in the rest of the paper the “Congolian”, and this might reflect collect- are based on our sample, and are not estimates ing intensity (Küper unpublished), rather than of the actual species richness of these phyto- actual low species richness. Consequently, in choria. the discussions we combine these two clusters. 238 BS 55

Table 3. Species richness and endemism in the simplified phytochoria delimited in this paper, based on the data used in this paper. phytochoria Broad Species richness endemic species Number species % endemic phytochoria, Narrow Species richness endemic species Number species % endemic

Sudanian 637 53 8% Guineo-Congolian 1700 799 47% Congolian 1137 346 30% Upper Guinean 564 71 13% CongolianTransition 1060 102 10% Zambezian 1893 850 45% Zambezian – Ethiopian-Kenyan 804 104 13% Zambezian-central 1477 541 37% Zambezian-Angolan 479 55 11% Southern African 2615 2268 87% Eastern Karoo 320 1 0% Cape 1822 1409 77% Kalahari 298 20 7% Natal 881 297 34% Namib-Karoo 339 69 20% Sahara 467 199 43% Somalian 344 53 15%

The Congolian phytochorion is both species endemism from 19% to 39% (Table 2). Thus rich and high in endemism, with the greatest the unplaced cells can have a major impact on species richness located in Lower Guinea, from the calculation of the endemism levels. How- Mt Cameroun to the mouth of the Congo ever, if outlier cells are “simplified” into the River. In tropical Africa, this is the most surrounding phytochoria, the endemism is species-rich area with also the highest concen- again reduced to 30% (Table 3), but without tration of endemics (Barthlott et al. 1996; loss of species richness. Cheek et al. 2001; La Ferla et al. 2002; Linder The areas along the borders of the Congo- 1998, 2001c; Mutke et al. 2001). The distribu- lian phytochorion were assigned to the “Con- tion of endemics through the region was not golian transitions”. Three regions can more or investigated, but the cells assigned to “Congo- less be distinguished in the Jaccard clustering: lian undifferentiated” have only 2% endemism Bamenda – Adamaoua (part of White’s (restricted to the cells assigned to this cluster) Guinea-Congolia/Sudania RTZ), Kivu and even though 750 species are included in our Uganda (partially the Lake Victoria RM) and database for these areas. Adding these to the the areas around Kananga and Mbuji-Maji cluster of species rich cells adds only 131 (largely the Guinea-Congolia/Zambezia RTZ). species to the cluster, but increases the It is not clear what combines these three areas, BS 55 239 since they abutt onto different floras. The attributed to the Sahel RTZ and the Sudanian largest extension beyond White’s boundaries is RCE by White (1983). The second (Sudanian- found in the southern Sudan, where the Ima- south) has a mix of Sudanian RCE and the tong Mountains are included in our Congolian Guinea-Congolia/Sudania RTZ (Fig. 1). The transition, while White assigned them to the NMDS ordination (Fig. 2a) reveals that there Sudanian RCE. According to Friis (1994) the are no distinct clusters of cells in West Africa. foothills of these mountains are clothed in This suggests that although there is a large Guineo-Congolian forest, and this could con- change in the flora from the coastal rainforests tain the species on which the groupings pro- to the Sahara, there are no sharp floristic, geo- posed here are based. Surprisingly, this transi- graphical, boundaries. Thus it appears as if tional phytochorion has a 10% level of there may be no large set of species in West endemism (Table 3). Africa which have co-incident limits to their It is surprising that such consistent group- distributions. This is consistent with the distrib- ings were obtained for the Guineo-Congolian ution of forest types (van Rompaey 1996) as RCE, since it is the most poorly sampled phyto- well as individual tree species (Bongers et al. chorion, with only 12% of the species repre- 1999) in West Africa, which are strongly deter- sented in our analysis (Table 1). However, all mined by the steep rainfall gradients from previous phytochorological studies have delim- coast to desert. ited this centre (e.g., Lebrun 1947; Monod It may be possible that our results are due to 1957; Wickens 1976), indicating that it is dis- sampling artifacts. Although almost 25% of the tinct from the other phytochoria. Sudanian species have been included in the study (Table 1), more than twice as large a pro- Sudanian portion as for the Guineo-Congolian RCE, it is Our data clearly separate the Sudanian RCE possible that our sampling has missed many of and the Zambezian RCE, thus corroborating the “typical” Sudanian species. The low level the results of the careful analysis of White endemism indicated by our data (1%, com- (1965), which upset the previous assumption pared to the 35% suggested by White) suggests (Lebrun 1947, Monod 1957) of a horse-shoe such a pattern of undersampling. Further- shaped phytochorion that partially surrounded more, the large proportion of cells with few the equatorial rainforests of Congo and species may also indicate undersampling, in Guinea with its own flora. The similarity in addition to distorting the results of the analy- savanna vegetation structure is not matched by sis. Simplifying our Sudanian phytochorion by a similarity in floristics. There is as yet no sim- removing the cells that clustered with the ple explanation for this dissimilarity – whether Somalian phytochorion, including the cells it is the effect of differences in the modern cli- that were not placed, and removing isolated mates and topography, the consequence of a outlier cells of neighbouring phytochoria long period of isolation either side of an resulted in an increase in the endemism levels Atlantic to Indian Ocean rainforest, or the to 8% (Table 3). result of two independent derivations from Our data indicate that a geographically rainforest. coherent, clearly delimited Sudanian phyto- Our analyses retrieved a Sudanian phyto- chorion might not exist. This should be tested chorion broadly similar to that of White. The in more detail by focussed investigations, par- cluster analysis retrieves two subgroups. The ticularly using ordinations, of the distribution first (Sudanian-north) contains a mix of cells patterns of a much larger sample of West 240 BS 55

African plant species. Indeed, our concerns much more so than the Sahelian flora. Under about the Sudanian phytochorion were already these circumstances it might make sense to voiced by White in 1993. think in terms of a Saharo-Sindian RCE, albeit with less than 2000 species (Quézel 1978). Sahara-Sahelian The Sahara-Sahelian regions continue the pat- Somalian tern implied by the NMDS ordination for the The Horn of Africa, according to White (1983) Sudanian phytochorion, and it would have forms part of the Somali-Masai RCE, which been satisfying if they grouped together. How- includes most of Kenya, and reaches south ever, on the cluster analyses the two sets of clus- through Tanzania into the valley of the Great ters are widely separated. Curiously, the Sahel Ruaha. However, our analysis, similar to the RTZ is the only region or transition zone for earlier analysis (Linder 1998), extracts a region which the endemism suggested by White (4%) limited to Somalia, the Ogaden region of is matched by our data. For the Sahara our Ethiopia and the North Eastern Province of level of endemism is four times more than sug- Kenya. gested by White, which is explained by our only Mapping the White phytochoria onto our partial inclusion of the Sahara. Thus the cluster analysis indicates that there are paired Saharo-Sindian elements (Lebrun 1947; White clusters assigned to the Somali-Masai RCE and & Léonard 1994) may be represented in our the Afromontane RCE, which in our analysis dataset by only a part of their distribution are embedded in the Zambezian phyto- ranges, while the vast majority of the remain- chorion. This overlap is also very evident from ing species have their full distribution ranges the ordination (Fig. 2b), which shows a portion included. Quézel (1978) demonstrated that of the Somalian centre, while another portion the endemism in the Sahara is only 12%, but overlaps with the Kenyan-Ethiopian upland that a further 23% of the species could be part of the Zambezian centre. regarded as “Saharo-Arabian”, further support- Possibly this area was undersampled – our ing the substantial contribution of the Saharo- data record 410 taxa and 11% endemism. Sim- Sindian element. This leads to the spuriously plifying the Somalian phytochorion by remov- high endemism figure for this small part of the ing the outliers in the Sahelian and Kalahari Saharo-Sindian region. In addition, some of phytochoria results in a reduction in species our data are artificially truncated at 17.5º N, richness to 344 taxa and an increase in thus enhancing the floristic dissimilarity endemism to 15% (Table 3), indicating that between Saharan and Sudanian regions, and the narrower definition makes more phyto- increasing the endemism of the Saharan flora. chorological sense. If all the cells that fall The Sudanian – Saharan boundary is 1-2 within White’s circumscription of the Somali- degrees north of the boundary between Masai RCE are included the richness increases White’s Sudanian RCE and his Sahel RTZ, to 931 species, but the endemism stays at 11% about halfway across his Sahel RTZ, but (Table 1). This sampled figure is slightly over aligned more or less parallel to the White 23% of the flora suggested by White, indicating boundary. that the area is actually quite well sampled. The distinctiveness of the Saharan flora from However, it is possible that our figures are that of the Sudanian woodlands is also sup- inflated by the inclusion of Afromontane ported by the ordinations. This suggests that patches included in the cells assigned to the the Saharo-Sindian flora could be distinctive, Somali-Masai RCE. Thulin (1994) estimated BS 55 241 the endemism of the flora of Somalia at 20%, This huge area was not retrieved in an ear- thus substantially more than our figure, which lier analysis of 794 species (Linder 1998), and is for a somewhat larger area. The explanation instead three separate regions were located: for this could be that in the central portions of (1) Malawi, Tanzanian, Kenyan and Ethiopia; Kenya and Tanzania the elements of this flora (2) Mozambique, Zimbabwe and south-eastern are found in the rift valley bottoms, while the Zambia, and (3) the rest of Zambia, Angola ridges and mountains which rise to 5000 m and Shaba. Our much larger analysis also contain an Afromontane flora. Consequently located three subdivisions, but they are some- many cells contain at least some Afromontane what different: (1) Angola and Barotseland elements, as well as some Somali elements. (“Zambezian-Angolan”), (2) the rest of South- There are outliers of the Somalian phyto- central Africa to Mt Kilimanjaro, inclusive of chorion in the central Kalahari of Botswana, as Shaba and Malawi (“Zambezian-central”), and well as along the Sahel to Senegal. These (3) the Kenyan and Ethiopian uplands (“Zam- reflect the “arid corridor”, a set of disjunct dis- bezian – Ethiopian-Kenyan”). Of these three tributions across the arid parts of Africa. areas only the Zambezian-central is species- Although the arid corridor has been well docu- rich, with a high level of endemism. This may mented (Balinsky 1962; de Winter 1966, 1971; not be surprising, since this area includes sev- Ihlenfeldt 1994; Jürgens 1997; Thulin 1994; eral local centres of richness and endemism: Thulin & Johansson 1996; Verdcourt 1969), the Zambezi-Congo watershed (Linder 2001c), there have been no phylogeographical analyses Nyika Plateau (Willis et al. 2001), the Southern of any plant species that are part of the corri- Highlands of Tanzania, and the eastern arc dor, we still lack a reasonable estimation of the mountains of Tanzania (Lovett 1993). The date (or dates) of the establishment of this dis- Angolan area is at least partially undercol- tribution pattern. lected, and includes only one centre of endemism, on the Huilla Plateau (Linder Zambezian 2001c). The Zambezian phytochorion, as delimited by The grouping of the uplands of Ethiopia, our analysis, is huge. It covers the whole of Sudan and East Africa into the Zambezian – south-central Africa, from the Atlantic Ethiopia-Kenya phytochorion is consistent with seaboard of Angola to the whole of Mozam- the distributions of woody and herbaceous bique, Tanzania, and the uplands of Kenya and Ethiopian Afromontane species, where the Ethiopia. As such it includes White’s Zambez- most common distribution pattern is of species ian RCE, most of the Afromontane RCE, and restricted to Ethiopia and the mountains of the Zanzibar-Inhambane RM, as well as the East Africa, and the second most common dis- Masai parts of the Somali-Masai RCE. This area tribution pattern is of species widespread in includes 1886 species, with 41% endemism the Zambezian woodlands (Friis 1994). More (Table 2), or 1700 species and 47% endemism curious, though, is the inclusion of those cells if the phytochorion is simplified (Table 3). It with predominantly Somalian species in these cannot be directly compared with any of clusters. Also puzzling is the low species rich- Whites centres of endemism, but although the ness of the uplands (Zambezian – Ethiopian- cells falling within the limits of White’s Zam- Kenyan phytochorion), and their low bezian RCE have 1725 species (thus not much endemism. The low endemism could partially less than our much wider definition), the result from the inclusion of species from the endemism stands at only 22% (Table 1). Somalian centre that penetrate along the low- 242 BS 55 lands and rift valleys of eastern Africa. The low RCE included northern Namibia, northern species richness suggests rather that the flora Botswana and northern South Africa. This shift has been undersampled, but many of these in the boundary might be the result of orchid cells belong to the Afromontane RCE, which at distributions being truncated along the politi- 39% is the best sampled phytochorion. Thus cal border, but it does seem remarkable that the results remain somewhat puzzling. The some 50 truncated distribution ranges could analysis grouped Jebel Marra with the other have such an effect. upland areas, and this might be the result of Southern Africa has a high species richness including the distributions of all species of (2615 species) and endemism (87%; Table 3). Jebel Marra, as documented by Wickens According to Arnold and de Wet (1993) the (1976) in the dataset. region includes c 21,087 species, while sub- The east and central African NMDS (Fig. 2b) Saharan tropical Africa (the rest of Africa shows the relationships in this area very clearly. excluding the Maghreb, Libya and Egypt) con- The cells assigned to the Congolian transitions tains 26,274 species (Lebrun & Stork 1997). If lie between the Congolian and the Zambezian we accept 80% endemism to southern Africa cells. An overlap of a different nature is seen (Goldblatt 1978), then 48% of the sub-Saharan between the Somalian and Zambezian cells, flora is found in southern Africa. this is made up of the cells assigned to the Not surprisingly, the Cape Floristic Region, Kenyan-Ethiopian uplands. Possibly this as delimited by Goldblatt (1978), is clearly region, with its complex arrangements of retrieved (called here “Cape”), both in the mountains capped with Afromontane flora and cluster analysis and in the ordination. Our valleys with a Somalian flora, should be analysis included 1822 of the c 9000 species regarded as a Regional Mosaic. (Goldblatt & Manning 2002) attributed to the The remaining two areas are a reasonable flora, and the endemism of 77% (Table 3) is match for White’s Zambezian RCE. The east somewhat higher than the just below 70% cal- coast Zanzibar – Inhambane RM will be dis- culated by Goldblatt and Manning (2002). cussed below. Again the Afromontane cells are This higher endemism could be the result of included within their “matrix” flora, this will including the Grahamstown fynbos outlier in also be discussed below. The distinction our Cape phytochorion. between the Angolan and central portions of The rest of southern Africa can be sum- the Zambezian phytochorion are more diffi- marised as two trends. The first ranges from cult to determine. It could be that the Angolan the wet eastern coastline (“Natal”). West of this phytochorion reflects the dominance of zone is the semi-arid “Eastern Karoo”, which miombo woodland species, while the central includes the grassland of the Eastern Cape phytochorion contains a rich mixture of interior and the plateau of the Free State and Afromontane species, and few miombo wood- the Northern Cape. Along the Atlantic coast- land elements. line, but also including the drier western mar- gins of the Great Karoo, is the Namib-Karoo Southern African phytochorion. The north-south trend is indi- The southern African regions were all cated by the distinction of the Kalahari from retrieved as one group (Fig. 1), with the north- both the Eastern Karoo and the Namib-Karoo, ern border along the political borders: the although it is evident from the NMDS analysis Cunene River in the west and the Limpopo in that there is no clear boundary between these the east. White suggested that the Zambezian three groupings (Figure 2c). These groupings BS 55 243 are unusual, and do not readily fit the previous Karoo phytochorion in the south, and the Kala- phytochorological classifications proposed for hari phytochorion to the north. White esti- southern Africa, as summarised by Werger mated his RTZ to include some 3000 species, (1978). most marginal transgressors, with a very low The Namib-Karoo phytochorion is similar to level of endemism. Our data support this: that suggested by White, except that the coastal while the Namib-Karoo phytochorion has 20% strip following the Namib desert into southern endemism, and the Natal phytochorion 34%, Angola was not detected. This is most likely a the combined Eastern Karoo and Kalaharia sampling artifact, since this coastal strip is phytochoria muster 7% endemism (Table 3). species poor and very narrow compared to grid Curiously, the uplands of Lesotho with their size, and in total only 339 species from the subalpine vegetation are also included in this whole phytochorion were included (Table 2, semi-arid phytochorion. 3). The distinction between the Succulent Karoo and Nama Karoo, described by Ruther- East Coast phytochoria ford and Westfall (1986) was not detected, White recognised two phytochorological enti- which could be due to both undersampling of ties along the African east coast (White 1983; Succulent Karoo elements and the coarse reso- White & Moll 1978). North of Inhambane he lution of our sampling. Although there have delimited the Zanzibar-Inhambane RM, whilst been suggestions that the Succulent Karoo south of the Limpopo River, to East London, region should be included in the Cape flora, the Pondoland-Tongaland RM. Neither of and separated from the Nama Karoo region these were retrieved as groups equivalent to (Bayer 1984; Jürgens 1991), our data show that the other large groups by our analysis. These the Cape and Nama regions are widely sepa- two phytochoria do not form geographically rated in the ordination, and that the similari- coherent entities (hence the term “regional ties of the Cape region lie rather with the more mosaics”). The description of the vegetation of mesic Natal region. This could of course be dri- these regions (Moll & White 1978) indicates ven by the shared Afromontane elements, as that they are mosaics of very different floristic well as the coastal thicket floras, which have elements – forests with species with Guineo- more in common between the Cape and the Congolian (Faden 1974), or Afromontane mesic east coast, than the arid west coast. A fur- affinities (White 1981), elements endemic to ther element is that the taxa that link the Cape the more humid eastern coast (Burgess et al. and Succulent Karoo floras (Crassulaceae, 1998), Zambezian elements in disturbed or Aizoaceae, Oxalis, Iridaceae, etc.) were not somewhat drier areas (Lovett 1993), and Soma- included in the analysis. The exact delimita- lian elements in the rain-shadows behind the tion between the western margin of the Cape coastal mountains (Lovett & Friis 1996). Such a flora, and the subtropical flora of southern mixture is not likely to be resolved by the analy- Africa, will need to be investigated in a more sis of data aggregated into 1° grids, but would detailed study (Linder 2003). need a more flexible matrix. The mesic east coast was included in the Nonetheless, the cells ascribed to the Zanz- Natal phytochorion, which encompasses the ibar-Inhambane RM largely group together whole of the Pondoland-Tongoland RM, as (with some additional cells) in a cluster well as the more tropical portions of the north- embedded within the Zambezian-central phy- ern parts of South Africa. White’s Kalahari- tochorion. The RM is species-rich (Clarke Highveld RTZ is divided between the Eastern 1998; Linder 1998, 2001c), and our data set 244 BS 55 includes 19% of these species (Table 1). Clarke land to upland forest, but that would be no dif- (1998), in a recent analysis, demonstrated that ference from the transition zones between any it includes 1356 endemic species, most of other phytochoria. However, Friis (1992) was which are found in the northern part of the able to show that at c 1500 m there is a change region (Somalia to the Tanzania-Mozambique in forest composition in Ethiopia, and sug- border). On these grounds he suggests that gested that this represented the transition this northern region should be separated as a from lowland to afromontane forest. For the Regional Centre of Endemism, as it includes forests of the Eastern Cape of South Africa, more than 1000 endemic species. However, it Cawe et al. (1994) were able to demonstrate a might better be regarded as a local centre of clear distinction between coastal subtropical endemism within the Zambezian-central phyto- forests and inland afromontane forests. How- chorion. ever, many species transgressed into the The cells associated with the Pondoland- “other” forest type, and the frequency of the Tongaland RM do not aggregate separate from species in the two different forest types was an the Natal phytochorion, and the Natal phyto- important factor to consider. Simple presence chorion could be regarded as an expanded ver- – absence data, as used in this study, would not sion of the Pondoland-Tongaland RM. The detect the differences between these two types main differences are that some inland areas, of forests. attributed to the Afromontane RCE by White, The critical issue, though, is whether the are included. affinities of the Afromontane flora on average lies with the lowlands surrounding the moun- Afromontane tains, or with other mountains. Continent-wide The recognition, or otherwise, of the numerical analyses, including the present one, Afromontane phytochorion remains con- fail to detect an Afromontane phytochorion. tentious. It is, according to White, character- There could be a number of reasons for this. ized by a small, but consistent number of tree Firstly, the cell size is too large to obtain “pure” species (Chapman & White 1970; White 1978, Afromontane flora. This could apply to por- 1981). Because it does not form a geographi- tions of East Africa and South-Central Africa, cally contiguous area, White referred to it as an and might account for the mixture of cells “Archipelago-like Regional Centre of obvious from both the ordinations (Fig. 2b, c) Endemism”. From the species richness, and the cluster analysis. In East Africa, due to endemism, and internal consistency of the the steep topography, transitions between the flora he found the Afromontane RCE compa- Somali and Afromontane vegetation can be rable to the other RCE’s. The richness of this very abrupt, for example on Mt Kulal and Mt flora decreases rapidly from East Africa to the Marsabit. However, this does not apply to Cape. Ethiopia or southern Africa. In both these It is not clear to what extent the Afromon- regions there are extensive areas of Afromon- tane flora can be distinguished from the sur- tane forest and grassland. rounding lowland forest flora. Regional East Secondly, the most characteristic groups of African studies mostly fail to detect a sharp plants may not have been included in the boundary to the Afromontane, and see a gradi- analysis. However, the Afromontane flora is the ent going up the mountains (Hamilton 1975; best sampled in this study, including 1564 of Hamilton & Perrott 1981; Lovett 1993, 1998a). the estimated 4000 species (Table 1). Of the 12 Possibly there is a gradual transition from low- species listed by White (1983) as defining the BS 55 245

Afromontane region over its whole range, four spread Afromontane distribution, most species (Ilex mitis, Nuxia floribunda, N. congesta and are endemic to one of three blocks – Kenyan- Rapanea melanophloes) are included in our Ethiopian, Zimbabwian – southern Tanzanian, analysis. In addition, the understory herbs and southern African. Possibly if genera were Impatiens and Begonia were also included. It used, then the whole track would be retrieved seems therefore unlikely that our dataset might (as has been demonstrated for a number of be especially skewed against the Afromontane taxa, such as the Disinae, Erica, Protea (Linder flora. et al. 1992)), but using species only portions of A third factor is the high level of species-level this range appear. Further detailed resolution turnover between mountain blocks, especially could be obtained if the Afromontane for herbaceous species. Carbutt and Edwards endemic species are sister taxa (e.g., Griswold (2002) estimated that 20% of the species of the 1991). If, however, they are derived indepen- southern African Drakensberg are endemic to dently from the surrounding lowland flora, the region, and Lovett (1993) showed 71% of then the Afromontane is more equivalent to a the Eastern Arc forest species that extend to vegetation type or perhaps as isolated indepen- the south reach the Limpopo, but that only dent phytochoria converging to a common 44% cross the river. Friis (1994) demonstrated structural flora. that the affinities of the Ethiopian Afromon- Most likely the combination of undersam- tane species were largely with East Africa, with pling the “typical” widespread Afromontane only 8.3% of the species described as Wide species, the high degree in species-level local Afromontane taxa (reaching Cameroon and endemism, and the interdigitation between W. Africa), and a mere 3.3% of the taxa reach Afromontane and the “matrix” phytochorion southern Africa. Similarly, 42% of a sample of might have led to the “loss” of the Afromon- 331 vascular plants from the Kitulo Plateau are tane phytochorion in this analysis. Or maybe described as “Eastern Afromontane”, although the Afromontane phytochorion really does not it is not clear how widespread they are (Lovett exist. et al. 1994). It is also evident from White’s The Afroalpine Region, proposed by Hau- (1983) account of the Afromontane phyto- man (1955), was also not retrieved in this chorion that there is extensive regionalism. His analysis. However, this could be expected, “Afromontane rainforest” ranges from since this is a very small flora (Hedberg 1957), Ethiopia to Malawi, and he lists a number of restricted to very small areas. It would certainly undifferentiated forest species found only in be interesting to attempt to delimit these areas, southern Africa. Many of the most characteris- and record this flora, to investigate the geo- tic species are regionally restricted: Podocarpus graphical patterns. falcatus, Widdringtonia whytei, Ocotea usambaren- There seem to have been no complete sis and Juniperus procera. The West African floristic analyses of the phytogeographical Afromontane flora contains very few species, affinities of any set of floras of isolated and is generally nested in extensive disturbed Afromontane areas, possibly because of the grassland and forests of more Guineo-Congo- incomplete knowledge of these floras lian affinities (Thomas & Achoundong 1994), (Grimshaw 2001). Clearly, a new Africa-wide so it is not surprising that these were not analysis of the Afromontane region is needed related to the East African montane floras. to determine whether the Afromontane is bet- Although there might be a small number of ter regarded as a phytochorion delimited by species (maybe several hundred) with a wide- the common possession of a significant num- 246 BS 55 ber of species, or a biogeographical track char- not illustrate these transitions, except as less acterised by a number of closely related but robust clusters. They show up very clearly on allopatric species. the ordinations. The delimitation of geograph- ically coherent areas for transition zones is Phytochoria, tracks and azonal vegetation largely arbitrary, but field experience indicates In tropical Africa the Guineo-Congolian, Zam- that the zonal vegetation is more clearly delim- bezian, Somalian and Saharan phytochoria ited. Possibly the problem is caused by the appear readily distinct. These occupy substan- zonal flora / vegetation of one region pene- tial integral areas, have distinctive genera, have trating as an azonal flora / vegetation into the “outliers” in the other areas, and are associated neighbouring region (along rivers, or into well- with very different climates. The Guineo-Con- drained habitats). Presence – absence data, at a golian penetrates into the Zambezian and 1° gridscale, will simply show a gradual transi- Sudanian along rivers, and has outliers on the tion. African east coast (White 1979, 1990). The Similarly, regional mosaics remain difficult Somalian phytochorion could be seen as the to conceptualize. Typical is the African east hub of the “arid track” with outliers in the coast, with its huge vegetational diversity, and Sahel and Kalahari (Thulin 1994). In historical high endemism mixed with outliers from terms, it suggests that the Horn of Africa would diverse other areas. On endemism it would be be a refuge area for the arid flora. The Zam- possible to argue for a Regional Centre of bezian penetrates into the Congolian along Endemism (Clarke 1998), but there are two drier ridges, but otherwise seems to lack out- arguments against such a classification. The liers. Possibly the Saharan flora could also be first is lack of geographical coherence (similar recognised as a huge, almost empty, phyto- to the problem of the Afromontane Centre), chorion (White & Léonard 1994). the second is the absence of a typical vegeta- In southern Africa three centres emerge. tion type associated with a distinct climate. The Cape is remarkably species rich (21% of This is similar to large areas of East Africa, with the sub-Saharan African flora!) and has out- a mixture of Afromontane above 1000 m and liers north along the higher mountains (Car- Somalian floristic elements below this altitude. butt & Edwards 2002; Weimarck 1933, 1936, Possibly one of the most difficult phytogeo- 1941). Along the West Coast is the Namib- graphical elements to integrate into a phyto- Karoo phytochorion, with its unique leaf-suc- chorological classification is the Afromontane. culent flora, and outliers along the east coast, While almost everywhere distinct from the sur- in the Albany Centre (Hartmann 1991), as well rounding floras, it also has a large local as much further north in shallow-soil habitats endemism component. It seems to be a track over bedrock. The Natal phytochorion along without a regional centre on which it could be the east coast has outliers along the south coast based. of the Cape phytochorion. Most phytochoria contain distinct local cen- The transition zones are phytochorologically tres of species richness (Lovett 1998b). These difficult. This applies to the south-north transi- are separated by extensive areas low in range- tion from the rainforest on the Gulf of Guinea restricted species, but where the widespread to the Sahara. A similar transition is found in species are found over the whole area. Thus southern Africa, but this runs in an east-west the Lower Guinea centre of endemism (from direction, from the rainforest on the Natal Mt Cameroun south to Gabon) has a high con- coast to the arid Kalahari. Cluster analysis does centration of range restricted species. The BS 55 247 remaining area of the Congolian phytochorion Mosambique, or locally within regions, such as largely lacks range-restricted species, and the within the Congo basin. species found in this area are also found in the Curiously, the percentages do not simply add Lower Guinean centre. This centre is therefore up. While some 13% of the flora has been sam- interpreted as nested within the Congolian pled, if we compare our species from each of phytochorion. Similarly, in the Cape flora a the White regions, the average sampling is number of species are widespread in the flora, closer to 20%. The simplest explanation for but some parts have a high concentration of this discrepancy is that our sample was biased range restricted species (Linder 2001b). towards the more widespread species. The sec- “Endemism” has become a confusing term: ond explanation, that White underestimated technically it means that the species are the species richness and overestimated the restricted to a defined area. Species with endemism of his centres, seems more likely. restricted ranges should rather be called This could be the result of his underestimating “range-restricted species”, and centres of the distribution ranges of the species, conse- range-restricted species should not be con- quently more species occur in neighbouring fused with phytochoria. The latter are much phytochoria (maybe present as azonal ele- broader, and include both diverse vegetation ments), increasing the species richness, and types, as one or several centres of range- decreasing the endemism. restricted species. Theoretically, the actual proportion of the flora included for phytochorological work is Implications of data quality for not critical, since the delimitation of phytocho- phytochorological analyses ria is based on the shared presence of species. Data representativity However, in practice cells represented by few There are 26,274 species in sub-Saharan tropi- species tend not to join clusters, possibly cal Africa (Lebrun & Stork 1997) and 21,087 in because they lack the species that delimit those southern Africa (Arnold & de Wet 1993). With clusters. Undersampling is therefore not posi- 80% endemism in southern Africa (Goldblatt tively misleading, in that it leads to incorrect 1978), it indicates a total flora of 42,700 clustering, but rather in that it leads to a lack of species, consequently our sample includes resolution. Undersampling will correctly assign some 13% of this flora. In an ideal case, the cells for which endemics are present, but not if selected data should representatively reflect all species are widespread between clusters. the geographical distribution of the entire The correct classification of species poor areas flora as well as its taxonomic and ecologic com- (as opposed to undersampled areas) will position. However, the availability of distribu- depend on whether endemics fall out faster tion data restricted the selection of taxa. At the than common species as richness declines, and broadest level, the distribution of taxa seems this will usually be the case, except on special balanced, with 11% from southern African, substrates. Consequently marginal areas will and 12% from tropical Africa, but several geo- have a higher incidence of misclassification. graphical areas are still comparatively under- The calculations of species-richness and represented. This occurs at different spatial endemism are strongly affected by undersam- scales – be it the areas in intermediate distance pling. This could have regional biases. For to urban centres that are not as well collected example, including all the Jebel Marra species as the areas adjacent to the metropols, or in the data set means that all Jebel Marra entire regions such as Angola or southern endemics are included, but from the other 248 BS 55 areas only the widespread species (that also Data structure occur on Jebel Marra) will be sampled. The Parts of the data are characterized by peculiari- data also contain an inherent taxonomic struc- ties: while the overall majority of the distribu- ture, both at the level of their actual distribu- tion data cover the full continental areas of the tion (the distribution of one species is most species, this is not the case for the data from likely not independent of the distribution of its Frankenberg and Klaus (1980) and many of phylogenetic relatives, thus one finds genera of the exclusively south African plant distribu- rainforest, savanna or Mediterranean climate tions (Pentaschistis, Restionaceae, Orchi- species respectively), and at the level of sam- daceae). The distribution areas of 80% of the pling. Only data of taxa that have recently been species from Frankenberg and Klaus (1980) taxonomically revised can be included, and are restricted to the area north of 17.5º N, such studies are mostly organised taxonomi- which considerably reduces the floristic over- cally. Since most taxa contain a restricted range lap and artificially enlarges the “endemism” of of growthforms, this also means that the the Saharan flora. This effect will be minor in growthform sampling is biased. the case of southern Africa case, were about For phytochorological work it may be impor- 80% of the species are endemic (including all tant to gradually include more taxa “typical” of Restionaceae, and all but one species of Pen- the different vegetation types, or “ecologically taschistis), but may still have shifted the bound- important species”, as suggested by White ary of the Zambezian phytochorion to the (1968). Yet such arguments are dangerously north. circular, or can cryptically lead to the defini- Almost all of the maps have been established tion of vegetation types, rather than phytocho- without a direct biogeographical context, e.g. ria. It is obvious that all species are not phyto- by accompanying taxonomic revisions. But 206 chorologically identical, and this makes the species were obtained from the Flora of Jebel sampling issue difficult. Marra (Wickens 1976), so that this certain bio- geographical element is disproportionately Spatial resolution represented in the dataset. This should not The general implication of working with a one negatively impact on the phytochorological degree gridded dataset is a limitation of analyt- analysis, although it will have dramatic impacts ical detail. Especially in areas with steep envi- on the patterns of species richness and ronmental gradients and a subsequent high endemism. floristic turnover, the one degree grids will pool species with very different biogeographi- cal affinities and therefore blur the analysis. Conclusion This is a particular problem for the Afromon- Numerical analysis based on 5438 species dis- tane region. A related problem may be with tributions (13% of the total flora) was able to long, narrow phytochorological units, which retrieve a set of credible phytochoria. These contain a high proportion of cells with admix- largely match the existing phytochorological tures of other phytochoria. This problem classification (for the Guineo-Congolian RCE, might have occurred in and the narrow coastal Sudanian RCE, Zambezian RCE, Cape RCE, strip of the Namib desert, the two east coast Karoo-Namib RCE, Pondolan-Tongaland RM), Regional Mosaics, and the Sudanian and Sahe- and so corroborate the most important ele- lian phytochoria. ments of the existing phytochorological classi- fication of Africa. BS 55 249

However, there are a number of problems Acknowledgements with the existing system. The Sudanian RCE may be arbitrarily delimited, as there seems to The analysis is based on a combined database be no interval in the change in species compo- established by the authors in several long term sition from the Guinean coast to the Sahara. projects. Their work has been generously The Afromontane RCE is not supported. funded by several sponsoring institutions. Jon There are a number of possible reasons; the Lovett would like to thank Conservation Inter- most likely could be that it is not as coherent national. Compilation of data on Restionaceae, over the whole of Africa as thought previously. Orchidaceae and Pentaschistis was funded by The Regional Mosaics could be dubious phyto- the University of Cape Town, while the Pro- chorological entities, since it could be difficult teaceae data were obtained from the Protea to assign a geographically contiguous area to Atlas, funded by WWF-SA, Department of Envi- them. Possibly they should be regarded as ronmental Affairs and the National Botanical areas of endemism nested within a broader Institute. phytochorion. The southern African phyto- Currently, the administration of the com- chorological delimitation should be revisited, bined continental dataset as well as further dig- possible a re-interpretation of the limits of the itizing of data is done in the context of the areas would be useful. BIOTA Africa project (sub project W03, However, the danger is still that the dataset is www.biota-africa.de) funded by the German not yet adequate to refute White’s negative Federal Ministry of Education and Research comments about using large datasets and (BIOLOG programme). numerical analyses. But contrary to the situa- Further data were generously provided by tion a few years ago, it is now possible to make Cyrille Chatelain and Laurent Gautier a number of concrete suggestions: (Geneva), Peter Frankenberg (Stuttgart), 1) Afromontane and Afroalpine taxa should Roger Polhill, Henk Beentje and Justin Moat be targeted for inclusion in the dataset, to eval- (Kew), Georg Zizka (Frankfurt), Jerome uate the coherence of the Afromontane phyto- Degreef (Meise), Adjima Thiombiano (Oua- chorion. Possibly the most effective approach gadougou), and Mauricio Velayos (Madrid). would be to incorporate as far as possible all Daud Rafiqpoor (Bonn), James Taplin (York), species from a number of sample sites, similar Henning Sommer (Bonn) and Anne-Marie to what was done for Jebel Marra. Sites should Bürger (Copenhagen) helped with the integra- include the Drakensberg, Malawi or the South- tion of the plant distribution data. Paul ern Highlands of Tanzania, Ethiopia and Mt Williams is acknowledged for providing the Cameroun. WORLDMAP software. 2) A number of transects should be analysed We would also like to thank Ib Friis, Henrik for West Africa, to determine whether there Balslev and the three anonymous referees for are co-ordinated species distribution limits, many constructive comments. thus testing for the existence of the Sudanian Phytochorion. 3) The geographical scale of the floristic – Literature cited vegetational change in West and East Africa Anonymous. 1969-. Distributions Plantarum Africanarum 1-. should be evaluated, to determine whether Jardin Botanique National de Belgique. geographically coherent phytochoria can be Arnold, T.H. & de Wet, B.C. 1993. Plants of Southern delimited in the region. Africa. Mem. Bot. Surv. S. Africa 62: 1-825. 250 BS 55

Balinsky, B.I. 1962. Patterns of animal distribution on the tropical Africa based on a mathematical analysis of distri- African continent. Ann. Cape Prov. Mus. 3: 299-310. bution maps. Bull. Jard. Bot. Belg. 50: 465-504. Barthlott, W., Lauer, W. & Placke, A. 1996. Global distribu- Faden, R.B. 1974. East African coastal – West African rain tion of species diversity in vascular plants: towards a forest disjunctions. In: Lind, E.M. & Morrison, M.E.S. world map of phytodiversity. Erdkunde 50: 317-327. (eds.), East African vegetation. Longman, London. Pp. Bayer, M.B. 1984. The Cape flora and the Karoo – a winter 202-203. rainfall biome versus a fynbos biome. Veld Fl. (1975+) 70: Frankenberg, P. & Klaus, D. 1980. Atlas der Pflanzenwelt des 17-19. Nordafrikanischen Trockenraumes – Computerkarten Bongers, F., Poorter, L., van Rompaey, R. & Parren, M.P.E. wesentlicher Pflanzenarten und Pflanzenfamilien. Arbeiten 1999. Distribution of twelve moist forest canopy tree aus den Geographischen Instituten der Universität species in Liberia and Cote d’Ivoire: response curves to a Bonn, Reihe A 133, Bonn. climatic gradient. J. Veg. Sci. 10: 371-382. Friis, I. 1992. Forests and forest trees of Northeast Tropical Africa. Brooks, T., Balmford, A., Burgess, N.A., Fjeldsa, J., Hansen, Her Majesties Stationary Office, London. L.A., Moore, J., Rahbek, C. & Williams, P. 2001. Toward a Friis, I. 1994. Some general features of the Afromontane blueprint for conservation in Africa. BioScience 51: 613-624. and Afrotemperate floras of the Sudan, Ethiopia and Burgess, N.D., Clarke, G.P. & Rodgers, W.A. 1998. Coastal Somalia. In: Seyani, J.H. & Chikuni, A.C. (eds.), Proceed- forests of eastern Africa: status, endemism patterns and ings of the XIIIth plenary meeting of AETFAT, Zomba, Malawi, their potential causes. Biol. J. Linn. Soc. 64: 337-367. 2-11 April 1991. National Herbarium and Botanic Gar- Carbutt, C. & Edwards, T. 2002. Cape elements on high- dens of Malawi, Zomba. Pp. 953-968. altitude corridors and edaphic islands: historical aspects Goldblatt, P. 1978. An analysis of the flora of Southern and preliminary phytogeography. Systematics and Geogra- Africa: its characteristics, relationships, and origins. Ann. phy of Plants 71: 1033-1061. Missouri Bot. Gard. 65: 369-436. Cawe, S.G., Moll, E.J. & McKenzie, B. 1994. An evaluation Goldblatt, P. & Manning, J.C. 2002. Plant diversity of the of the phytochorological classification of the forests of Cape Region of southern Africa. Ann. Missouri Bot. Gard. Transkei. In: Seyani, J.H. & Chikuni, A.C. (eds.), Proceed- 89: 281-302. ings of the XIIIth plenary meeting of AETFAT, Zomba, Malawi, Grimshaw, J.M. 2001. What do we really know about the 2-11 April 1991. National Herbarium and Botanic Gar- Afromontate archipelago. Systematics and Geography of dens of Malawi, Zomba. Pp. 1043-1059. Plants 71: 949-957. Chapman, J.D. & White, F. 1970. The evergreen forests of Griswold, C.E. 1991. Cladistic biogeography of Afromon- Malawi. Commonwealth Forestry Institute, Oxford. tane spiders. Austral. Syst. Bot. 4: 73-89. Chatelain, C., Gautier, L. & Spichiger, R. 2001. Application Hamilton, A.C. 1975. A quantitative analysis of altitudinal du SIG IVOIRE à la distribution potentielle des espèces zonation in Ugandan forests. Vegetatio 30: 99-106. en fonction des facteurs écologiques. Systematics and Hamilton, A.C. & Perrott, R.A. 1981. A study of altitudinal Geography of Plants 71: 313-326. zonation in the montane forest belt of Mt Elgon, Cheek, M., Mackinder, B., Gosline, G., Onana, J.-M. & Kenya/Uganda. Vegetatio 45: 107-125. Achoundong, G. 2001. The phytogeography and flora of Hartmann, H.E.K. 1991. Mesembryanthema. Contr. Bolus western Cameroon and the Cross River-Sananga River Herb. 13: 75-157. interval. Systematics and Geography of Plants 71: 1097-1100. Hauman, L. 1955. La “region afroalpine” en phytogeogra- Clarke, G.P. 1998. A new regional centre of endemism in phie Centro-Africaine. Webbia 11: 467-469. Africa. In: Huxley, C.R., Lock, J.M. & Cutler, D.F. (eds.), Hedberg, O. 1957. Afro-alpine vascular plants. A taxo- Chorology, taxonomy and ecology of the floras of Africa and nomic revision. Symb. Bot. Upsal. 15: 1-411. Madagascar. Royal Botanic Gardens, Kew. Pp. 53-65. Ihlenfeldt, H.-D. 1994. Phytogeography of Pedaliaceae Clayton, W.D. & Hepper, F.N. 1974. Computer-aided R.Br. In: Seyani, J.H. & Chikuni, A.C. (eds.), Proceedings chorology of West African grasses. Kew Bull. 29: 213-234. of the XIIIth plenary meeting of AETFAT, Zomba, Malawi, 2- de Klerk, H.M., Crowe, T.M., Fjeldsa, J. & Burgess, N.D. 11 April 1991. National Herbarium and Botanic Gardens 2002. Biogeographical patterns of endemic terrestrial of Malawi, Zomba. Pp. 1063-1075. Afrotropical birds. Diversity and Distributions 8: 147-162. Jaccard, P. 1901. Distribution de la flore alpine dans le de Winter, B. 1966. Remarks on the distribution of some Bassin des Dranes et dans quelque regions voisines. Bull. desert plants in Africa. Palaeoecol. Africa 1: 188-189. Soc. Vaud. Sci. Nat. 37: 241-272. de Winter, B. 1971. Floristic relationships between the Jardine, N. 1972. Computational methods in the study of northern and southern arid areas in Africa. Mitt. Bot. plant distributions. In: Valentine, D.H. (ed.), Taxonomy, Staatssamml. München 10: 424-437. phytogeography and evolution. Academic Press, London. Denys, E. 1980. A tentatative phytogeographical division of Pp. 381-393. BS 55 251

Jürgens, N. 1991. A new approach to the Namib Region. endemism: a candidate for World Heritage Status? J. E. Vegetatio 97: 21-38. Africa Nat. Hist. Soc. Natl. Mus. 87: 359-366. Jürgens, N. 1997. Floristic biodiversity and history of Lovett, J.C. & Friis, I. 1996. Patterns of endemism in the African arid regions. Biodiversity and Conservation 6: 495- woody flora of north-east and east Africa. In: van der 514. Maesen, L.J.G., van der Burgt, X.M. & van Medenbach Kier, G. & Barthlott, W. 2001. Measuring and mapping de Rooy, J.M. (eds.), The biodiversity of African plants. endemism and species richness: a new methodological Kluwer, Dordrecht. Pp. 582-601. approach and its application on the flora of Africa. Biodi- Lovett, J.C., Gereau, R.E. & Sidwell, K.J. 1994. Vegetation versity and Conservation 10: 1513-1529. and phytogeography of the Kitulo Plateau, Southern La Ferla, B., Taplin, J., Ockwell, D. & Lovett, J.C. 2002. Tanzania. In: Seyani, J.H. & Chikuni, A.C. (eds.), Proceed- Continental scale patterns of biodiversity: can higher ings of the XIIIth Plenary Meeting of AETFAT. National taxa accurately predict African plant distributions? Bot. J. Herbarium and Botanical Gardens of Malawi, Blantyre. Linn. Soc. 138: 225-235. Pp. 1025-1041. Lebrun, J. 1947. Exploration du Parc National Albert. Fascicule Lovett, J.C., Rudd, S., Taplin, J. & Frimodt-Møller, C. 2000. 1. La Végétation de la plaine alluviale au sud du lac Édouard. Patterns of plant diversity in Africa south of the Sahara Institut des Parcs Nationaux du Congo Belge, Bruxelles. and their implications for conservation management. Lebrun, J.-P. & Stork, A.L. 1997. Enumération des plantes à Biodiversity and Conservation 9: 37-46. fleurs d’Afrique tropicale: IV Gamopétales: Ericaceae à Lami- Moll, E.J. & White, F. 1978. The Indian Ocean coastal belt. aceae. Ville de Genève, Conservatoire et Jardin In: Werger, M.J.A. (ed.), Biogeography and Ecology of south- botanique, Geneva. ern Africa. Junk, The Hague. Pp. 561-598. Léonard, J. 1965. Contribution a la subdivision phytogeo- Monod, T. 1957. Les grandes divisions chorologiques de l graphique de la region Guineo-Congolaise d’apres la Afrique. C.C.T.A./C.S.A., London. repartition geographique d’Euphorbiacees d’Afrique Mutke, J., Kier, G., Braun, G., Schultz, C. & Barthlott, W. tropicale. Webbia 19: 627-649. 2001. Patterns of African vascular plant diversity – a GIS Linder, H.P. 1998. Historical interpretation of the African based analysis. Systematics and Geography of Plants 71: phytochoria. In: Huxley, C.R., Lock, J.M. & Cutler, D.F. 1125-1136. (eds.), Chorology, taxonomy and ecology of the African and Polhill, R.M. 1982. Crotalaria in Africa and Madagascar. A.A. Madagascan floras. Royal Botanical Gardens, Kew. Pp. 67- Balkema, Rotterdam. 86. Quézel, P. 1978. Analysis of the flora of Mediterranean and Linder, H.P. 2001a. The African Restionaceae. Contribu- Saharan Africa. Ann. Missouri Bot. Gard. 65: 479-534. tions from the Bolus Herbarium, Cape Town. Rebelo, A.G. 1991. Protea Atlas Manual: Instruction booklet to Linder, H.P. 2001b. On areas of endemism, with an exam- the Protea Atlas Project. Protea Atlas, Rondebosch. ple from the African Restionaceae. Syst. Biol. 50: 892-912. Rohlf, F.J. 1998. NTSYSpc. Numerical Taxonomy and Multi- Linder, H.P. 2001c. Plant diversity and endemism in sub- variate Analysis System, 2.02i. Exeter Software, New York. Saharan tropical Africa. J. Biogeogr. 28: 169-182. Rutherford, M.C. & Westfall, R. 1986. Biomes of southern Linder, H.P. 2003. The radiation of the Cape Flora. Biol. Africa – an objective categorization. Mem. Bot. Surv. S. Rev. Cambridge Philos. Soc. 78: 597-638. Africa 54: 1-98. Linder, H.P. & Ellis, R.P. 1990. A revision of Pentaschistis Sneath, P.H.A. & Sokal, R.R. 1973. Numerical taxonomy. (Arundineae: Poaceae). Contr. Bolus Herb. 12: 1-124. Freeman, San Francisco. Linder, H.P. & Kurzweil, H. 1999. Orchids of Southern Africa. Thomas, D.W. & Achoundong, G. 1994. Montane forests of A.A. Balkema, Rotterdam. Western Africa. In: Seyani, J.H. & Chikuni, A.C. (eds.), Linder, H.P., Meadows, M.E. & Cowling, R.M. 1992. His- Proceedings of the XIIIth plenary meeting of AETFAT, Zomba, tory of the Cape flora. In: Cowling, R.M. (ed.), The ecology Malawi, 2-11 April 1991. National Herbarium and of fynbos: nutrients, fire and diversity. Oxford University Botanic Gardens of Malawi, Zomba. Pp. 1015-1024. Press, Cape Town. Pp. 113-134. Thulin, M. 1994. Aspects of disjunct distributions and Lovett, J.C. 1993. Eastern Arc moist forest flora. In: Lovett, endemism in the arid parts of the Horn of Africa, partic- J.C. & Wasser, S.K. (eds.), Biogeography and ecology of the ularly Somalia. In: Seyani, J.H. & Chikuni, A.C. (eds.), rain forests of eastern Africa. Cambridge University Press, Proceedings of the XIIIth plenary meeting of AETFAT, Zomba, Cambridge. Pp. 33-55. Malawi, 2-11 April 1991. National Herbarium and Lovett, J.C. 1998a. Continuous change in Tanzanian moist Botanic Gardens of Malawi, Zomba. Pp. 1105-1119. forest tree communities with elevation. J. Trop. Ecol. 14: Thulin, M. & Johansson, A.N.B. 1996. Taxonomy and bio- 719-722. geography of the anomalous genus Wellstedia. In: van Lovett, J.C. 1998b. Eastern tropical African centre of der Maesen, L.J.G., van der Burgt, X.M. & van Meden- 252 BS 55

bach de Rooy, J.M. (eds.), The biodiversity of African plants. White, F. 1981. The history of the Afromontane archipel- Kluwer, Dordrecht. Pp. 73-86. ago and the scientific need for its conservation. Afr. J. van Rompaey, R.S.A.R. 1996. Rain forest refugia in Liberia. Ecol. 19: 33-54. In: van der Maesen, L.J.G., van der Burgt, X.M. & van White, F. 1983. The vegetation of Africa. Unesco, Paris. Medenbach de Rooy, J.M. (eds.), The biodiversity of White, F. 1990. Ptaeroxylon obliquum (Ptaeroxylaceae), African plants. Kluwer, Dordrecht. Pp. 624-628. some other disjuncts, and the Quaternary history of Verdcourt, B. 1969. The arid corridor between the north- African vegetation. Bull. Mus. Natl. Hist. Nat., B, Adanso- east and southwest areas of Africa. Palaeoecol. Africa 4: nia 12(2): 139-185. 140-144. White, F. 1993. The AETFAT chorological classification of Weimarck, H. 1933. Die Verbreitung einiger Afrikanisch- Africa: history, methods and applications. Bull. Jard. Bot. montanen Pflanzengruppen, I-II. Svensk Bot. Tidskr. 27: Belg. 62: 225-281. 400-419. White, F. & Léonard, J. 1994. Rapports phytogeo- Weimarck, H. 1936. Die Verbreitung einiger Afrikanisch- graphiques entre l’Afrique et le sud-oest Asiatique. In: montanen Pflanzengruppen, III-IV. Svensk Bot. Tidskr. 30: Seyani, J.H. & Chikuni, A.C. (eds.), Proceedings of the XII- 36-56. Ith plenary meeting of AETFAT, Zomba, Malawi, 2-11 April Weimarck, H. 1941. Phytogeographical groups, centres and 1991. National Herbarium and Botanic Gardens of intervals within the Cape flora. Acta Univ. Lund. 37: 3-143. Malawi, Zomba. Pp. 1085-1103. Werger, M.J.A. 1978. Biogeographical division of southern White, F. & Moll, E.J. 1978. I. Phytogeography (of The Africa. In: Werger, M.J.A. (ed.), Biogeography and Ecology Indian coastal belt). In: Werger, M.J.A. (ed.), Biogeogra- of Southern Africa. Junk, The Hague. Pp. 145-170. phy and ecology of southern Africa. Junk, The Hague. Pp. White, F. 1965. The savanna woodlands of the Zambezian 563-575. and Sudanian domains. Webbia 19: 651-681. Wickens, G.E. 1976. The flora of Jebel Marra (Sudan Republic) White, F. 1968. A proposed atlas of ecologically important and its geographical affinities. Her Majesties Stationary plant species in Africa. Acta Phytogeogr. Suec. 54: 307-310. Office, London. White, F. 1971. The taxonomic and ecological basis of Williams, P. 2002. WorldMap, 4.20.17. Spots before your chorology. Mitt. Bot. Staatssamml. München 10: 91-112. eyes productions, London. White, F. 1978. The Afromontane Region. In: Werger, Willis, C.K., Burrows, J.E., Fish, L., Phiri, P.S.M., Chikuni, M.J.A. (ed.), Biogeography and Ecology of Southern Africa. A.C. & Golding, J. 2001. Developing a greater under- Junk, The Hague. Pp. 463-513. standing of the flora of the Nyika. Systematics and Geogra- White, F. 1979. The Guineo-Congolian Region and its rela- phy of Plants 71: 993-1008. tionship to other phytochoria. Bull. Jard. Bot. Belg. 49: 11- 55.