Congruence among datasets
We found a large degree of congruence in the African species-level biogeographical patterns of plants, mammals, birds, amphibians and reptiles. This congruence is evident from a visual comparison of the clade-based maps, from the consensus classification (which is almost perfect if a congruence of three of five taxa is used, and where the core biogeographical regions are retrieved by all five taxa), as well as from the confusion matrix analysis. Globally, patterns of richness and endemism are positively and significantly correlated among mammals, amphibians, birds and reptiles (Lamoreux et al., 2006). However, congruence in the biogeographical regions has not been reported previously. Udvardy (1975) found that, at a global scale, there was agreement amongst zones developed by zoologists, except for areas of a transitional nature, but that the botanical zones were more different. At a continental scale, Rueda et al. (2010) observed substantial incongruence in the regionalization of trees, butterflies, reptiles, amphibians, birds and mammals in Europe. Consequently, it is unclear whether the congruence demonstrated in Africa is unusual, or whether this a global pattern, at least in tropical regions.
The reasons for congruence or incongruence would be very interesting to understand. The nature of the underlying data might affect the results, but a biological explanation seems more likely. Congruence could be the consequence of the vertebrate distributions being influenced by the vegetation and flora (predicted by Rueda et al., 2010), by common responses to the same climatic parameters or by a common underlying history. Interpreting the common response as being driven by the vegetation is consistent with the idea that the vegetation functions as a giant ecosystem engineer that creates the habitat for animals (Linder et al., in press), e.g. hot pyrophytic savanna (Bond et al., 2005; Beerling & Osborne, 2006) or shady cool forests (Pinto et al., 2010) can be found under the same climate. This is consistent with the very different structure in the dendrograms between the plants and the vertebrates. The poor structure in the plant dendrogram may reflect the individual response of each species to local environmental variations, while the highly resolved vertebrate dendrograms, with long branches separating the biogeographical regions (Fig. 1), could be the response to the spatially broader integrated vegetation structure, which is itself a spatially more coherent expression of the floristic patterns. The incongruent patterns in East Africa could be the result of the much smaller-scale variation between vegetation types in the region (Lind & Morrison, 1974). The importance of a common climatic response is shown by the absence of amphibians from much of the Sahara (Fig. 1d), and the different patterns of the reptiles in southern Africa. Such biological differences might also account for the different placements of the boundaries between savanna and forest, both to the north and the south of the Congolian Region. The recognition of a separate Ethiopian Region by birds, reptiles and amphibians, but not by mammals, might also be the result of different climatic responses. Historical explanations have been invoked in the differentiation of the forest regions in Africa (Lovett & Friis, 1996; Fjeldså & Lovett, 1997; Couvreur et al., 2008). These could result in the floristic fragmentation of these regions and so reduce the capacity to retrieve them as biogeographical regions. This could possibly account for the inconsistency in the retrieval of a separate Guinean Region. A detailed exploration of the explanations for congruent biogeographical regions in Africa could be very interesting, but is beyond the scope of this paper.
The combined evidence and consensus methods for obtaining a biogeographical regionalization of Africa are highly complementary. The combined evidence approach places every grid cell, and results in a well-resolved classification in which all seven main regions can be recognized, and subregions delimited in most of them. The consensus approach is clearly more conservative and less well resolved, but it separates the well-supported core biogeographical regions from the areas not supported by most groups of organisms. This offers support for the recognition of the seven (or six) core regions in Africa. Conducting both analyses offers both a detailed result and gives an indication of the support of the placement of each grid cell in this detailed result.
The Congolian, Sudanian, Zambezian, Somalian, Ethiopian, Southern African and Saharan regions are identified by most clades, and are also evident in the combined evidence and consensus analyses (except the Ethiopian Region). These common regions are very similar to the phytochoria proposed by White (1983), and the biomes map developed by the WWF for their ecoregions project (Burgess et al., 2004, 2006). However, there are also a number of differences between our combined evidence regionalization and the chorological classification of White (1983). First, White identified two regional mosaics (RMs) which were not recognized in our regionalization scheme, namely the Lake Victoria RM and the Zanzibar–Inhambane RM. It is not surprising that we did not detect the RMs, because these mostly lack endemics and so are not delimited as distinct areas. Instead, their cells are grouped within those regions with which they share most biota, which are generally geographically adjacent. Regional mosaics might have been expected to conform to some of our zones of high turnover. However, this does not seem to be the case as, for example, the Zanzibar–Inhambane regional mosaic of White is not recognized in any of our regionalization or turnover maps. Second, our classification does not conform to that proposed by White in southern Africa, the reasons for which are discussed below. Finally, White recognized two altitudinally driven and fragmented Afromontane and Afroalpine phytochoria. We found no support for the recognition of these regions, although such patterns could possibly emerge by analysing distributions of lineages (at an appropriate taxonomic level) instead of species.
The Congolian Region, as defined by the combined evidence analysis, is a close match to the Guineo–Congolian region delimited by White (1979, 1983) and Kreft & Jetz (2010). However, this belies a complex underlying set of patterns, revealed by the separate taxon datasets, and consequently evident on the consensus analysis.
African tropical rain forests are traditionally separated into three regions. Along the East African coast several small relicts are found, these are usually distinct from the main, Central and West African, forest block (White, 1979). The Kakamega Forest in western Kenya constitutes a Congolian outlier (Wagner et al., 2008). Neither of these is retrieved in our analyses. Presumably the forest effect is diluted by the surrounding vegetation, due to our larger grid cell sizes. The Congolian and Guinean forests also form two regions; these are retrieved only by the combined evidence dataset, with the boundary placed at the Sanaga River in Cameroon to the east of Mount Cameroon. There is some support for this boundary from vascular plants (Lawson, 1996), primates (e.g. Anthony et al., 2007) and the mammals analysed by Kreft & Jetz (2010). Analyses of mammals, reptiles and birds recognize the extension of the Congolian Region along the West African coast. Analyses of these taxa also reveal the ‘Dahomey Gap’ in Benin. This biogeographical interval has also been identified by analyses of plants (Brenan, 1978; White, 1983), amphibians and mammals (Jenkins, 1992) and birds (de Klerk et al., 2002a). It is evident in the vegetation maps as a region where the Sudanian savanna reaches the ocean (White, 1983), thus splitting the Guinean forest into two blocks. Recent phylogeographical research has started to unravel some of the complex patterns within this region, revealing refugia and centres of endemism (Sosef, 1996; Anthony et al., 2007; Couvreur et al., 2008; Marks, 2010). Recent studies using DNA to discriminate geographical areas have shown that many of the species of rain forest birds that are currently regarded as ‘shared’ between the Guinea and Congo areas should be recognized as different species separated since the late Miocene (Fjeldså & Bowie, 2008).
The northern Congolian forest zone boundary is strongly defined in all groups of animals, and slightly less well defined in plants, as a zone with a high turnover in species composition (Fig. 4). This contrasts particularly with an almost complete absence of spatial turnover in species composition from the central part of the Congo Basin, which contains a widespread fauna (Williams et al., 2003). West Africa (the Guinean coast and its hinterland) reflects the transition between the less seasonal (coastal) rain forest and the more seasonal (inland) Sudanian savannas. This region is mostly topographically subdued, with a relatively even north-to-south rainfall gradient; consequently, the idiosyncratic nature of species climatic tolerance limits results in the absence of congruent biochorological boundaries. The southern Congolian boundary is generally weakly defined, with few species replacements or species drop-outs, except at the Angola scarp (Figs 4 & 5). This region is characterized by a broad transitory zone with interdigitation of north–south-trending savanna ridges and forested valleys, leading towards the well-drained uplands in the southern parts of the Congo. Further south, in Zambia, Brachystegia savanna is the norm, with evergreen forest restricted to wetlands. At the 1° scale this results in a gradual transition and no clear biome boundary. Not surprisingly, there is no congruent classification for this region; in the consensus analysis it is an unplaced area, but in the combined evidence analyses it is separated as the Shaba subregion. White (1983) recognized this complex situation by labelling the region as transitional.
The savanna regions and East Africa
Our data corroborate the early detailed analysis by White (1965) showing that the northern and southern savannas are different, and show that this difference is recognized by all biotic groups investigated here. Indeed, only the mammal dataset groups the two savanna regions as sister regions (Fig. 1a), supporting the results obtained by Kreft & Jetz (2010).
The extent of the Sudanian Region in the north closely fits that proposed by White (1965, 1983), but there is incongruence among the taxon datasets as to the placement of the eastern border, with the mammal analysis including the uplands of Ethiopia in the Sudanian Region.
The Zambezian Region in the south, defined by the combined analysis, extends further east and north than White’s delimitation, with the inclusion of the Zanzibar–Inhambane RM, the Lake Victoria RM, the Ethiopian uplands (Afromontane region in White), and much of the Somalia–Masai area. Neither the separate analyses nor the more detailed combined evidence analysis retrieved East Africa as a region.
Almost all of East Africa falls into a transition zone between the Somalian Region on the arid northern edge, the Zambezian Region on the seasonally dry western and southern edge and the Congolian Region on the wet aseasonal western edge. There is also a chain of mountains running down the middle of the East African area, and a zone of forest/savanna on the coastal margins – with the mountains and the lowlands containing mixed biogeographical signals from the Congolian and Zambezian regions, as well as large numbers of local endemics (Emberton et al., 1997; Burgess et al., 1998b, 2007b). All of these biotas intermix, depending on local topography and rainfall patterns. The high rates of turnover and strong gradients of richness are evident in all groups in our data (Fig. 4a–e). Usually, biogeographical regions consist of a central area of low species replacement surrounded by a border of high species replacement (as seen in the Congolian, Sudanian and Southern African regions). We suggest that East Africa presents a very complex biogeographical mixture. The different biologies of the taxa generate idiosyncratic responses to complex climatic and topographical patterns, resulting in incongruent biogeographical signals. Assigning the area to any region (for example the Zambezian Region) will obscure much of this complexity. This is exemplified by the area along the Albertine Rift in Central Africa: in the combined evidence analysis, it is assigned to the Congolian Region, in the reptile analysis to the Ethiopian Region, in the mammal analysis it is partitioned between the Congolian and the Sudanian regions, in the bird analysis it is assigned to the Zambezian Region and in the amphibian analysis it is a separate region. Not surprisingly, in the consensus analysis (Fig. 3) it is unplaced. Maybe the best solution is to recognize the whole of East Africa (from the Kivu province of Congo, Uganda, Kenya and north-central Tanzania) as a large RM. This is equivalent to a much enlarged Lake Victoria RM.
A relatively large Somalian Region is recognized for all groups, and is somewhat smaller than the Somalia–Masai region of White. The region has low species numbers in all groups and turnover rates are also low. The levels of endemism in the region are high for plants (Thulin, 1993) and reptiles (Burgess et al., 2004). Thulin (1994) argued for this region to have been a refugium for arid-adapted plants, from which they have colonized Socotra, Arabia, East Africa and even southern Africa. The low turnover rates contrast with the high local turnover in the mountainous regions to the west and south-west.
Four out of five datasets recognize an Ethiopian Region, which could be part of the Afromontane Region of White (1983), as it is centred on the Ethiopian uplands. The Ethiopian Region extends southwards to the Albertine Rift for reptiles, the Kenyan uplands for amphibians and to the Eastern Arc for plants. However, neither combined nor consensus maps show any southward extension of the Ethiopian Region.
The boundary between the Saharan and Sudanian regions seems to be almost entirely due to species drop-outs (compare Fig. 5c, which shows a change in diversity, and Fig. 4c, which shows no change in species composition). Furthermore, there are relatively few species that occur widely in the Sahara. The Saharan Region is the only one that extends far beyond the African continent, as the Saharo–Sindian Region, which reaches to the arid western part of the Indian subcontinent (Wickens, 1976; Brenan, 1978; Kreft & Jetz, 2010), and is generally species-poor. Part of the poor geographical definition of the region could be due to mixed affinities: Leonard (2000) showed for Jebel Uweinat on the Libyan–Egyptian–Sudanian border that the lower elevation flora is Sudanian, and the upper elevation flora Saharo–Sindian, and that the level of endemism to the Saharo–Sindian Region is low. Furthermore, the amphibians are, for obvious reasons, absent from much of the region.
Contrary to all previous treatments (e.g. Werger, 1978; White, 1983), but consistent with Kreft & Jetz (2010), we recognize a single Southern African Region. White recognized five regions: the Cape, Karoo–Namib and Afromontane regional centre, one transitional zone (Kalahari–Highveld) and one RM (Tongaland–Pondoland). These segregations are recognized by the combined evidence analysis (as five areas), but not by any of the taxon-based analyses.
The absence of a Cape Region from any of the cluster diagrams is surprising, especially for the plants, as this region is recognized as a separate plant kingdom at the same level of difference as the rest of the Palaeotropics (Good, 1974; Takhtajan, 1986; but see Cox, 2001), and its distinction has long been recognized (Marloth, 1908; Goldblatt, 1978), although there are arguments that it should be combined with the neighbouring Succulent Karoo Region (Born et al., 2007). There are several factors that might have led to our analyses not recognizing the Cape, or Greater Cape, regions. First, the high level of plant species turnover within the Cape Region (Fig. 4e; Born et al., 2007) means that, although the same genera are found throughout the region, there are proportionally very few common species across this region. Therefore, an analysis based on shared species will not retrieve a Cape Region. Second, the interdigitation of arid, north-western elements in the drier inter-montane valleys with mesic south-eastern elements found along the coastal flanks of the mountains dilutes the fynbos elements. This interdigitation is particularly strong along the south-eastern coastline, where Afromontane, coastal thicket, fynbos and karoo elements co-occur in a very small area (Cowling, 1983). The non-fynbos species often have a widespread distribution (especially the forest elements), and so provide strong evidence to link these grids to the rest of the South African grids. Finally, the level of endemism among vertebrates is not exceptional in the Cape Region, and no Cape Region was detected by for mammals by Kreft & Jetz (2010) or birds by de Klerk et al. (2002a).
The grouping together of the other southern African zones is not surprising, except possibly the arid Namib biota. This is, however, very species-poor and its lack of recognition could be a result of this lack of evidence. Compared with the regions recognized in tropical Africa, the recognition of a single Southern African Region (albeit with subregions) seems a reasonable solution.
Afromontane Region and centres of endemism
The Afromontane is not retrieved as a separate region by any of the datasets, thus confirming the results of Linder et al. (2005), but contrary to White (1978, 1983) and Wickens (1976). The African montane flora and at least some of the fauna is clearly differentiated from the surrounding lowland biota, although there is often no abrupt transition in the floristic composition (Hamilton & Perrott, 1981; Lovett, 1998). The evolutionary processes also appear to be different, with diversification happening at different times (e.g. Roy, 1997). Further evidence of this in our data is the high species turnover (as measured by neighbourhood segregation in Fig. 4a–e) in the region of the Albertine Rift, the Eastern Arc Mountains, the Southern Rift and the Ethiopian Highlands. Furthermore, the dominant trees are common from Ethiopia to Cameroon and Cape Town (Chapman & White, 1970; White, 1978, 1983).
There are two explanations for the failure of objective analytical methods to retrieve the Afromontane Region. The most likely explanation is that a high intermontane turnover at species level obscures the biotic commonality among the mountains, similar to the explanation for the failure to retrieve a Cape flora region. Consequently, each montane region is included in its surrounding region, except the large Ethiopian Region.
The second possible explanation is that the grid sizes are too large, and consequently also include the biota of the surrounding lowland habitat matrix. The coarse scale of the analysis grid means that smaller-scale features, such as mountains, might be lost within the broader groups, and the mixture of lowland and montane elements in the montane grids then dilutes the effect of the narrow endemics (see also de Klerk et al., 2002b; Linder et al., 2005). Using a finer grid scale will result in numerous spurious absences, which makes the classification of grid cells into broad zones problematic. One way to address the grid scale challenge might be to use natural features as the units, rather than a rigid grid system. This was done very successfully in the Cape flora by Moline & Linder (2006), using previously defined broad habitat units (Cowling & Heijnis, 2001), and in Australia (Mackey et al., 2008). Such natural units would have to be pre-defined, however, and the success of the approach would thus depend on the availability of easily defined area delimitations.
The recognition of an Ethiopian Region is consistent with both explanations. This region represents the northern extreme of the Afromontane Region. This area is also the largest spatially coherent area, and includes several complete grid cells. This removes the dilution effect. The inability to link the southern Afromontane grid cells to this core is consistent with a species-level replacement removing much of the biogeographical signal.
The Afroalpine Region is also not recognized. This archipelago-like region is found only above 3500 m in Ethiopia, Kenya, Uganda, Congo, Rwanda and Tanzania, and shares no species with the lower slopes (Hedberg, 1955, 1957, 1986; White, 1983; Gehrke & Linder, 2009). However, it always occupies only a small part of a grid cell, and could be regarded as an extreme zonal environment, rather than a biogeographical region, rather like saline wetlands or coastal mangroves.