Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
A hotspot revisited – a biogeographical analysis of West African amphibians
Museum für Naturkunde, Leibniz Institute for Research on Evolution and Biodiversity at the Humboldt University Berlin, Invalidenstrasse 43, 10115 Berlin, Germany
Johannes Penner, Museum für Naturkunde, Leibniz Institute for Research on Evolution and Biodiversity at the Humboldt University Berlin, Invalidenstrasse 43, 10115 Berlin, Germany. E-mail: firstname.lastname@example.org
Johannes Penner, Museum für Naturkunde, Leibniz Institute for Research on Evolution and Biodiversity at the Humboldt University Berlin, Invalidenstrasse 43, 10115 Berlin, Germany. E-mail: email@example.com
Aim The study was aimed at testing whether West Africa can be regarded as a distinct biogeographic region based on amphibian assemblages. If so, we asked what were the relationships of these assemblages with those in Central Africa, and whether West African amphibian distributions showed biogeographic substructure. We further investigated what events or processes may explain the observed patterns.
Location Sub-Saharan Africa.
Methods Presence–absence data of amphibian assemblages derived from field surveys and the literature were statistically analysed using three different multivariate techniques (consensus clustering, Monmonier analysis and nonmetric multidimensional scaling) to emphasize consistent results.
Results We showed that West Africa has unique amphibian assemblages, which could be clearly demarcated from Central African assemblages, particularly by the geographic barrier of the Cross River. Further biogeographic subdivisions were detected to the west of this barrier. Habitat, mainly forest, was the best factor explaining our observed pattern. Overall, intra-regional similarity (e.g. within West Africa) was higher than intra-habitat similarity (e.g. within forest) across regions.
Main conclusions Our results are compared with previous works and interpreted in the light of the known evolutionary history of West and Central Africa. The observed pattern may be explained by postulated differences in river continuity through time, with West African rivers serving as more or less constant barriers in contrast to those in Central Africa. Our results demonstrate the uniqueness of West African amphibian assemblages, highlighting the need for their conservation as many are under acute anthropogenic pressure.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The delineation of biogeographical units has puzzled scientists for more than two centuries (Lomolino et al., 2004). Among the major scientific aims of biogeography remain the description and explanation of faunal and floral distribution patterns, as well as the identification of centres of high species richness and/or endemism. In Africa, as elsewhere, existing areas of high biodiversity have probably persisted during periods of extensive environmental change, e.g. glacial periods. They are, at least in part, considered to have served as Pleistocene refugia (e.g. Diamond & Hamilton, 1980; Crowe & Crowe, 1982; Mayr & O’Hara, 1986; Grubb, 1992).
Biomes with a high endemicity and/or numbers of threatened species are of special conservation interest, resulting in the identification of ‘hotspots’, places of elevated endemic and threatened biodiversity. The original hotspot concept was based mainly on data from vascular plants and selected terrestrial vertebrates (Myers, 1988, 1990; Myers et al., 2000). This global approach currently recognizes eight hotspots in Africa, of which the ‘Guinean Forests of western Africa’ is one. Similar ‘hotspot’ concepts have been applied in Africa for vascular plants (e.g. Linder, 2001; Küper et al., 2004) and selected animal groups, such as mammals (e.g. Kingdon, 1990; Kreft & Jetz, 2010), birds (e.g. Crowe & Crowe, 1982; De Klerk et al., 2002) and fishes (e.g. Roberts, 1975; Hugueny & Lévêque, 1994) but see e.g. Kareiva & Marvier (2003) for criticism of the concept. Recent collaborative studies have combined a number of different vertebrate taxa (mammals, birds, snakes and amphibians) in order to identify areas in Africa containing high biodiversity (see Hansen et al., 2009 for a comprehensive list of publications). Conservation biologists are embracing biogeographical research in response to the steadily growing human-induced pressures on biodiversity, as well as dwindling conservation resources such as financial and human capital. As a consequence, conservation efforts have to be directed and channelled necessitating hard choices concerning threatened sites and species (e.g. Brooks et al., 2006; Wilson et al., 2006, 2009; Carwardine et al., 2008; Underwood et al., 2008). Consequently, it is necessary and enlightening to relate patterns of diversity not only to biogeography, but also to phylogenies and conservation, environmental, and/or social variables.
Mostly because of the limitations of data availability, many previous studies identified particular hotspots based on a coarse geographic scale, e.g. grid cells of 1°, c. 111 × 111 km at the equator. However, in general, areas used for conservation planning are much smaller (Shriner et al., 2006), and although conservation planning, such as priority setting, can be derived from large scales (Larsen & Rahbek, 2003), a finer resolution often achieves better results (Warman et al., 2004; Hurlbert & Jetz, 2007; Jetz et al., 2007). Although most countries have a conservation network with at least some kind of legal status, gaps in space and taxa remain. It is frequently suggested that these gaps should be closed, but final selection criteria for decision makers are manifold and often include economic priorities. Many protected areas in Africa, for example, are located in less fertile regions, or have other limitations that prevent human settlement or agriculture, e.g. diseases or parasites (Ford, 1971). Whether existing protected areas effectively cover areas of high biodiversity often remains untested, and the knowledge to prioritise areas based on, for example, postglacial colonization routes, historical refugia, barriers, and/or biogeographic units is scarce.
A variety of taxa have been used to approach such questions. Among vertebrates, amphibians have experienced the highest increase in species with Red List status during the last decades (Stuart et al., 2008). Although this could be because of a hidden bias, as complete assessments for fish and reptiles are lacking. In part, their vulnerability stems from the comparatively high habitat specificity of many species and their low mobility. Currently, more than 30% of all amphibian species are threatened and included on the Red List, making amphibians one of the most threatened class of organisms world-wide (Stuart et al., 2004). The main threats are habitat destruction and alteration (e.g. Stuart et al., 2008), and this situation may become worse if forecasted climate changes are correct (Carey & Alexander, 2003; Corn, 2005; Araujo et al., 2006). The combination of habitat specificity, low mobility, and ease of sampling of the group in a standardized manner makes it an ideal biogeographical model group (Zeisset & Beebee, 2008).
For Africa, the highest regional amphibian diversities have been mapped for the Cameroon Highlands and for the ‘Eastern Afromontane hotspot’ (see Hansen et al., 2009). In these and other publications, West Africa (WA) is frequently also considered a hotspot, although often, either directly or implicitly, regarded as being a subset of the Central African (CA) bioregion (see review by Werger, 1978). For example, Poynton (1999) in a continental analysis of amphibian biogeography stated ‘… part [of West Africa] could be regarded as a subtraction margin of the fauna of Cameroon.’ In this analysis, the whole ‘west equatorial’ region was subdivided into four blocks: central, south, east and west, the latter comprising WA west of the Dahomey Gap. In another study, Schiøtz (1967) analysed the distributions of reed frogs (Hyperoliidae) and other selected amphibian taxa and identified major zoogeographical barriers in WA by comparing visually the detected barriers to distributions of mammals (mainly primates) and birds. Two main barriers were confirmed: the Dahomey Gap and the Cross River. Throughout these and other studies (cited earlier), the exact delineation of WA biogeographic units with respect to CA appears to be haphazard. This stimulated us to pose four questions that we addressed using a data set of African amphibian assemblages. First, we tested whether a distinct WA bio-region could be defined, or whether the region was a subset of the CA bio-region. Second, if WA was shown to contain unique assemblages, we asked where the boundary between the CA bio-region was geographically located. Third, we examined whether WA displayed sub-regions, comprising smaller, but distinct biogeographical units. Finally, we analysed which factors could cause the observed biogeographic pattern. The overarching goal behind these questions was to discover regions that may require specific conservation measures.
Geography and data set
Our study is confined to mainland sub-Saharan Africa. Data on 120 amphibian assemblages have been compiled through our own field surveys (65 sites) and from literature records (55 sites). Species taxonomy was harmonized (see Appendix S1 in Supporting Information). Literature records have been updated to current taxonomy, and when necessary checked for plausibility. The genus Arthroleptis has been omitted because of the unresolved taxonomic status of most WA taxa (Rödel & Bangoura, 2004). In addition to the presence of a species at a particular site, the following information was noted (after IUCN, 2010): Red List status, presence between our postulated barriers (see section on causes for delineations), irrespective of the database record and habitat preferences (see the following paragraphs for details). The final data set comprised binary data for a total of 528 species (3161 presence records). Country codes throughout the text follow ISO standards (ISO 3166-1 accessed 11th January 2010). Ideally, rarefaction curves or estimators should be used to assess the sampling efficiency of each site and to estimate its alpha diversity (Magurran, 2004). However, our data set did not allow for that, because seasons and survey designs differed considerably between sites and some literature records did not contain the necessary information (e.g. daily species lists and sampling effort). Despite these caveats, analyses can be adjusted to presence/absence (binary) data, and these analyses achieve similar result to abundance-based analyses (Furse et al., 1984; Marchant, 1990).
Delineation of WA amphibian assemblages
Our first question was whether WA has unique amphibian assemblages, i.e. if they are more similar to each other than to other assemblages on the continent. This question can be addressed if there is either a gradual change in assemblage similarity, or a clear cut boundary with respect to CA assemblages. In the latter case, we investigated where this boundary or barrier is geographically located.
Binary data of all amphibian assemblages were translated into three dissimilarity matrices using three different indices. The choice of indices is crucial as it heavily influences further analyses (e.g. Learner et al., 1983; Legendre & Legendre, 1998), and there is a multitude (over 80) of different indices plus transformations, although the use of binary data limits this choice. It is important to weigh the alternate states, presence/absence, differently (Legendre & Legendre, 1998), as absences are more difficult or even impossible to ascertain (e.g. Kéry, 2002) and thus presence data are more informative. However, there is no agreement on how to weigh the two different states, and there is no single index that is solely recommended. Consequently, we chose three indices that are well suited for binary data and vary in the weights given to the two states.
Jaccard (1908)– is the simplest index (Legendre & Legendre, 1998) and is often used for binary data and can therefore be easily compared with other studies. It is calculated as the number of shared species divided by the number of shared species, plus the number of singletons (species recorded in only one assemblage). To use it as a dissimilarity index, the formula has slightly been modified, using the Bray–Curtis dissimilarity (Oksanen, 2008). Its major disadvantage is its sensitivity to sample size (Oksanen, 2008), thus making it difficult to compare assemblages with very different species richness or different sampling intensities.
Mountford (1962)– the advantage of this index is that it is less sensitive to different sample sizes. It is derived from Fisher’s log series. There are several disadvantages; it is not commonly used, and the index is non-metric (Shepard, 1984; Oksanen, 2008) which means that there is no linear relationship within the index itself.
Raup & Crick (1979)– is also a non-metric index and a measure of the probability of observing the same species in the compared assemblages. The probability is derived from a hypergeometric distribution (Legendre & Legendre, 1998). Absent species from two compared assemblages are weighted more strongly than in the other two indices (Oksanen, 2008).
The consensus of the three indices allows for more robust conclusions, and the emphasis in our study was placed on consistent results, although different methods were applied. Using these three indices, we gained three dissimilarity matrices for the 120 assemblages. To uncover potential general geographic patterns, we used a Mantel test to test whether sites close to each other had more similar assemblages than sites at a greater distance. The test searches for correlations between geographic distances (Euclidian distances) species compositions. However, the test does not categorize assemblages according to their similarities. For that we grouped assemblages via cluster analyses, using the dissimilarity matrices as distance measures. A variety of different linkage methods are available (e.g. Leyer & Wesche, 2007; Mouchot et al., 2008) and groupings are strongly dependent on the distance measures as well as the cluster criteria used (Gordon & Vichi, 2001). Consequently, we again applied the consistency (consensus) principle. If groups were detected consistently by different indices and different methods, we considered them as being well supported (Leyer & Wesche, 2007). As we aimed at avoiding any presumptions on the number of groupings, we employed hierarchical clustering. We used an optimization approach to construct a final single cluster. As a result, 21 clusters were combined, i.e. we combined three distance measures (Jaccard, Mountford, Raup-Crick) and seven linkage methods [single linkage (nearest neighbour), complete linkage (furthest neighbour), average linkage (UPGMA), median linkage (WPGMC), centroid linkage (UPGMC), McQuitty’s method and Ward’s method; see Sneath & Sokal, 1973 for details]. The chosen linkage methods differed in their grouping properties, meaning that they either tended to build single similarly sized groups (single linkage), few large groups (complete linkage) or behaved neutral (conservative, all other linkage methods). When minimizing Euclidian dissimilarity (Hornik, 2009), a single consensus cluster was gained. Agreements between pairs of clusters were calculated as one minus the rate of inversions between associated ultrametrics (Hornik, 2009).
The above cluster analysis groups assemblages but does not take their spatial array into account. Therefore, assemblages might be grouped close to each other, although they are geographically separated by large distances. This hampers biogeographical explanations, especially if assemblages containing different amphibians are located between them. One way to include information on the geographic location of an assemblage is to use the Monmonier algorithm (Monmonier, 1973). Here, assemblages are first directly connected in a geographic space (Delauney network). Subsequently, a boundary orthogonal to the connecting line is drawn in the middle between two assemblages. This procedure is conducted from every assemblage towards every neighbouring assemblage (Voronoi tessellation; see Dupanloup et al., 2002; Manni et al., 2004; Jombart, 2008). Subsequently, the algorithm searches homogenous areas and delineates them towards other areas. Barriers are drawn on the orthogonal boundaries in order of their significance, starting with the most significant one. Hence, geographical boundaries are depicted between the most dissimilar assemblages that are geographically close.
It is important to evaluate whether most known species are listed for a specific region because generally species that remain undetected may have a negative impact on statistical analyses. Thus, species richness estimations of regions were calculated as rarefaction curves for the whole continental data set and for the speculated WA subset. Rarefaction adds numbers of species per site cumulatively. The order of sites is randomly arranged and the process iterated, leading to a smoothed average of all curves. When this smoothed curve reaches a plateau, it is likely that most species have been recorded (Sanders, 1968; Hurlbert, 1971; Simberloff, 1972). However, no extrapolation on the number of expected species can be made (Bush et al., 2004).
Causes for the delineations
Monmonier analysis, as described earlier, draws the location of the barriers always in the middle between two assemblages. This is independent of geographic features on the ground. Rivers especially are often neglected as barriers though their function is confirmed for amphibians (Li et al., 2009) and even large mammals (Anthony et al., 2007). To test specific barriers, we hypothesized nine potential barriers based on altitude, river systems, floral regions (Udvardy, 1975; White, 1983; Olson et al., 2001) and previous work on amphibians (Schiøtz, 1967, 2007; Poynton, 1999). Specifically, we tested the following barriers (from east to west): Cross River, Niger River, the Dahomey Gap, Volta River (now Lake Volta), Banadama River (which continues in the rain forest zone as a v-shaped gap, called V Baolé), Cavally River (including the Upper Guinea highlands), Mano River, Kolenté River (including the Fouta Djallon) and Géba River (see Fig. 1).
To assess which factors are responsible for the observed distribution patterns, three types of information have been added for each species: (1) a simple ecological classification, detailing whether the species occur in forest, farmbush, woodland, savanna, montane grasslands or fynbos (after IUCN 2010; compared with Fig. 2; multiple allocations were allowed); (2) nine potential biogeographical barriers dividing the species’ ranges into ten potential partitions. A third category, (3) IUCN Red List classification (Appendix S1), was also added to detect where threatened assemblages occur and how they are potentially related to biogeographical patterns. The threat status was also weighed, either linearly (NL = 2; DD = 2; LC = 1; NT = 3; VU = 4; EN = 5; CR = 6) or exponentially (NL = 2; DD = 2; LC = 1; NT = 4; VU = 8; EN = 16; CR = 32), to emphasize higher threat categories.
To test the influence of these factors, habitat, barriers and threat status, non-metric multidimensional scaling (NMDS) was used. This is an iterative optimization procedure and is preferred over similar techniques, e.g. principal component analysis, because it is flexible and has no underlying assumptions, such as linear relationships or parametric data (Kruskal & Wish, 1978; Clarke, 1993). This indirect gradient analysis results in a reduction to a few dimensions, or axes, on a real or hypothetical environmental gradient. No real environmental data of the sites is recorded for this analysis. To avoid the statistical problem of the analysis getting ‘trapped in local optima’ (McCune & Grace, 2002), the NMDS was repeated 10.000 times per run. The NMDS places the sites into an n-dimensional space, n being the number of factors included. Factors, habitat, occurrence within hypothesized barriers and red list classification were fitted as new axes and therefore as explaining vectors for the observed pattern. All analyses were conducted with the software R 2.9.0 (2009) using the packages ‘Adegenet’, ‘Clue’, ‘Mass’, ‘Stats’ and ‘Vegan 1.15-2’.
Overall, 120 amphibian assemblages have been analysed, comprising a total of 528 species (Appendix S1). Total species richness and richness for all sites is actually higher, as the genus Arthroleptis has been omitted (see Methods). Our main question was whether there are distinct amphibian faunal regions. A first Mantel test for all African assemblages confirmed that amphibian assemblages in close geographic proximity have a higher similarity than assemblages in greater distance (P < 0.001). In a second step, we examined whether this pattern remained on a regional scale.
Consensus clustering grouped amphibian assemblages close to each other according to their faunal similarity (Fig. 2). The congruence between the 21 different groupings (three similarity indices multiplied by seven linkage methods) was 68%. Within the derived single consensus tree, one single central cluster was obvious. It contained only sites west of the Cross River which is roughly the border between Nigeria and Cameroon. The right column of the graph shows this graphically (Fig. 2). The lengths of the bars indicate species richness, and colours indicate to which region the assemblage is allocated. Based on our results WA, the green group, is herein defined as the region west of the Cross River and south of the Saharan desert. Hence, Nigeria is included and Cameroon excluded. Two further distinct clusters are apparent. The first one (bottom) groups East and Southern Africa. Two assemblages from Congo (CD) were embedded within this group. The second cluster comprised all remaining CA assemblages. Habitat preferences for all recorded species are depicted in six classes (middle section of Fig. 2). Interestingly, WA savanna and rain forest assemblages were more similar to each other than rain forest assemblages of WA and CA or savanna assemblages in general. However, within the large WA cluster, rain forest and savanna assemblages grouped separately.
Of the 528 species, 172 were recorded in the WA cluster. Slightly more than 50% occurred only there (90 species). In WA, 22 species were only represented in one single assemblage. On a continental basis, three regions in Cameroon were the most species rich (Mt. Nlonako, Korup, Nkongsamba). Similar rankings were observed for genus richness (Korup, CM; Mt. Nlonako, CM, Mt. Doudou, GA). Family richness was highest in sites in Gabon and Tanzania (Mt. Doudou, GA; Mahenge, TZ; Usambara Mts., TZ; Crystal Mts., GA). Rankings remained the same when all taxa of unresolved taxonomy were excluded from the analysis. Highest numbers of such taxa were noted in CA (Mt. Doudou, GA; Mt. Manengouba, CM; Korup, CM; Tchabal, CM). Within the WA cluster, the assemblages of Mt. Nimba (CI, GN, LR), Pic de Fon (GN) and the Taï National Park (CI) were the taxa richest. The species-richest sites also contained the highest numbers of threatened species (Spearman’s rank correlation test, P < 0.001). Thus, the top ranking sites kept their status when species occurrences were weighed by threat status. Exponential weights changed the order slightly, by putting more emphasis on three sites, i.e. Obudu (NG), Ankasa and Atewa (both GH; see Appendix S1).
Rarefaction results have to be treated carefully. Comparisons between all assemblages, the WA ones and the cumulative number of species were conducted. In the latter, the sites were ranked geographically from the west to the east and to the south (Fig. 3). The boundaries between regions can be identified by a sharp increase in the cumulative number of species. The comparatively flat rarefaction curve for WA confirms that this region is better represented in our database than the whole continent. More generally, the flat curve shows that for WA, most species are present in the database; only three valid species are missing Amietophrynus danielae, Amietophrynus perreti, Phrynobatrachus brongersmai.
As the cluster analysis does not take into account the geographic relationships of the assemblages, a Monmonier analysis was conducted, searching for differences between neighbouring assemblages. The analysis partly supported the proposed barriers. In particular, it confirmed the Cross River as dividing the West from the CA amphibian assemblages. Furthermore, the analysis confirmed the Kolenté River and the Lake Volta as separating distinct groups of species assemblages within WA. Between these two rivers, we detected the assemblages with the highest number of species in WA. To reveal the finer structure within the WA data set, the Monmonier analysis and ordination were conducted for the WA assemblages only (n = 74).
The NMDS analysis also clearly separated WA from all other assemblages (stress values for the 3 dissimilarity measures used: Jaccard 20.73, Mountford 22.31, Raup-Crick 20.46). Within WA, species occurrences between the hypothesized barriers proved to be significant and therefore not randomly structured (Appendices S2 and S3). Nine geographic partitions were highly significant (P < 0.001) throughout all used dissimilarity measures. The other partitions also showed varying degrees of significance (Appendix S2). The major factor responsible for the groupings within WA is the habitat specificity of the species. Assemblages dominated by forest and farmbush species separate well from assemblages containing mostly savanna and woodland species (Appendix S2). The IUCN Red List status also differentiates assemblages (Appendix S2). This is because of a correlation between Red List status and habitat preference, threatened species being predominantly found in threatened habitats, e.g. forests and montane grasslands.
In a nutshell, our results show that WA amphibian assemblages are unique compared with other African assemblages. Several geographic partitions have been indicated and the Cross River has been confirmed as the major barrier towards CA. Explanatory variables are multifaceted, with species habitat preferences being dominant.
Two of the most important aims in conservation are to protect species and sites. Certainly, the best strategy is to preserve species in their natural habitats. Therefore, it is vitally important to know where unique areas are located, e.g. in terms of rare and unique species, and how these areas can be delineated. To reveal such delineations in Africa, several attempts with different definitions using different organisms have been put forward and discussed (see Introduction). Our study is the first comprehensive analysis of WA amphibian assemblages, showing that this region has unique species compositions compared with other African realms. This is due to the fact that a large number of species occur only in WA. Previous works on amphibians (Schiøtz, 1967; Poynton, 1999) did suggest the Cross River as important geographic barrier for species distributions, but did not weigh it against other barriers. They did also detect another clear-cut species boundary in the western part of the region, the Mano River approximately on the border between Liberia and Sierra Leone. However, our analyses did not confirm the Mano River as a very sharp boundary and other barriers within the WA region were also not prominent, e.g. by marked drops in species diversity, as previously suggested. This is probably due to two facts: first, the areas ‘outside’ the barriers, e.g. west of the Mano River, are climatically suitable for forest (Harcourt et al., 1992) and therefore provide suitable habitats for many species. Second, Guinea and Sierra Leone have not been studied formerly in detail and this is the first time that recent and detailed amphibian surveys of these countries have been included in biogeographic analyses. Several other biogeographic studies (see Introduction) focused on mammals and plants in the region. Most did not reveal the Cross River as the delineation between WA and CA bio-regions, and usually placed emphasis on the Dahomey Gap or the Niger River. However, two exceptions exist, i.e. studies on bushbucks (Moodley & Bruford, 2007) and duikers (Colyn et al., 2010), which both detect a clear difference in genetics and morphology between species on either side of the Cross River.
The factor consistently explaining the division of WA towards CA and the subdivisions within the region is habitat; ‘forest’ assemblages are especially unique. It is important to note that many WA amphibian species occur only in primary (=undisturbed) and not in disturbed forests (Ernst & Rödel, 2005; Ernst et al., 2006; Hillers et al., 2008b). Beside forests, montane grasslands likewise had a large impact on the clustering of assemblages, as this rare WA habitat is home to many specialized species.
In addition to the inter-regional and well-defined WA and CA assemblages, separated by the Cross River, an intra-regional grouping by habitat is evident in WA. Thus, within the WA region, two major groups of amphibian assemblages could be identified: forest and savanna assemblages. That intra-regional grouping by habitat is not as distinct as that by region may have several explanations. One is the way in which species habitat preferences is recorded, as some species occur in more than one habitat type. Thus, similar assemblages may comprise species of differing habitat preferences. Additionally, there is no sharp boundary between forest and savanna biomes but rather a broad transition zone (see White, 1983; Burgess et al., 2004). Assemblages located in the transition zone are more likely to contain higher percentages of species from both habitat types, compared with assemblages from very distinct habitat zones. This gradual change in biomes is partly reflected in the composition of WA amphibian assemblages. Another reason is that most forests in WA are severely fragmented and threatened (Achard et al., 2002), and within the true forest region, new habitats are anthropogenically generated. These generally favour savanna species, enabling the invasion of farmbush and savanna species into assemblages that would naturally feature a higher percentage of true forest species.
The origin of anurans, comprising the vast majority of our taxa, is dated to the beginning of the Triassic (ca. 250 Ma; e.g. Roelants et al., 2007). African anurans might have evolved later, around 100 Ma (see Zimkus et al., 2010 and references therein). Recent speciation pulses have occurred c. 15–10 Ma ago (review in Moritz et al., 2000; Wieczorek et al., 2000; Zimkus et al., 2010). However, these dates have to be treated with caution as no data on WA amphibian fossils exists to root these phylogenetic trees. Also information on the WA climate beyond the last glacial maximum (> 20.000 years) is generally meagre. Hence, comparisons between the evolution of the WA amphibian taxa and the evolution of the observed biogeographic pattern under palaeo-climate scenarios are difficult. A more interesting question is why WA forest and savanna assemblages form a cluster separate from CA forest and savanna assemblages. In general, three hypotheses may apply. Species may have evolved along an ecological gradient (Endler, 1982; Fjeldså, 1994), in riparian refugia (Colyn et al., 1991; Aide & Rivera, 1998) or in refugia of climatic stability (e.g. Diamond & Hamilton, 1980; Crowe & Crowe, 1982; Mayr & O’Hara, 1986; Grubb, 1992). If, as is commonly suggested, the evolutionary history of both regions was more or less the same, inter-regional similarity of savanna dominated assemblages should be higher as no WA–CA barrier is commonly suggested for the savanna region. This suggests that the evolutionary history differs between the regions. A possible scenario is that forest and savanna assemblages in WA have evolved together. The main driving force for this co-evolution could have been substantial expansions and contractions of forest extents between and during the ice ages (Hamilton, 1976; Maley, 1996). Thus, WA assemblages might have evolved in a mosaic landscape, with cyclical fragmentation and re-connecting of forest and savanna patches. As a result, forest and savanna assemblages evolved in close proximity and species exchanges were likely. In contrast, CA assemblages might have evolved in a situation where forest and savanna blocks retained greater connectivity. As a consequence, the habitat assemblages may have evolved more separately. However, it is often suggested that forest remnants in CA were minute (Amiet, 1987; Hamilton & Taylor, 1991; Maley, 1996). A model for forest–savanna mosaic in dry periods has been discussed for plants and primates in the Congo Basin (Colyn et al., 1991), but this remains controversial and it has not been applied to WA. Overall, the question exactly how forests were expanding or shrinking during different climatic periods remains open. The CA forests could have remained more ore less stable throughout time and without change, as has been hypothesized for the Eastern Arc Mountain forests (Finch et al., 2009; but see discussion therein).
In this respect, two other differences between WA and CA are notable, potentially explaining the uniqueness of WA assemblages. First, around 54–49 Ma ago, several ‘bizarre pollen types’ appeared in CA only (Morley, 2000). The identification of the plants associated with these pollens, and the ecological changes associated with their appearance, remains unknown. They may represent the traces of unique habitats that occurred only in CA, and were absent in WA. Later, when the species assemblages of both regions had diverged, the unique habitats in CA were replaced by similar habitats in both regions. Alternatively, hydrological features in WA have remained more or less consistent since the Miocene (23–5 Ma; John, 1986). This coincides, for example, with the diversification of Hyperolius and Phrynobatrachus species during the late Oligocene, early Miocene (Wieczorek et al., 2000; Zimkus et al., 2010), and probably other amphibians as well (see review by Moritz et al., 2000). Rivers probably broadened during wet periods (Nicolas et al., 2008), thereby increasing their effectiveness as barrier. This contrasts with the biogeography of eastern Africa, where the recent hydrological system is much younger (John, 1986). The hydrological history of CA is less clear. It is speculated that the Congo River may have been unconnected to the ocean until 30 Ma ago (review by Goudie, 2005) and therefore did not act as a barrier in western CA. In these hydrological models, the WA amphibian assemblages may have been ‘trapped’ between river barriers, resulting in regional similarity despite the presence of the major forest and savanna habitats. In contrast, continuous exchange within habitats was possible for most of this period for CA assemblages.
Areas with a high biodiversity are often equated to areas where Pleistocene refugia may have been located. Previous work (see Introduction) on the location of these refugia has yielded conflicting results. Generally, three WA refugia have been postulated. The first contains south-eastern Liberia and south-western Côte d’Ivoire, including Mt. Nimba, although the latter may have also been a separate refuge. This whole area is also often called ‘Upper Guinea’. The second refuge has been thought to be located between south-eastern Côte d’Ivoire and south-western Ghana. The third refuge belongs to CA but stretches partly into the far south-eastern tip of Nigeria and includes areas west of the Cross River (see Fig. 1; Maley, 1996). The exact delineation of these refugia remains imprecise and varies between authors, e.g. the forest block ranging from the Taï National Park in Côte d’Ivoire through to the Liberia-Sierra Leone border, may have acted as one single Pleistocene refugium (Laurent, 1973). This block is often considered a single ecoregion (Burgess et al., 2004), and the rough position, extent and entity of this potential forest refugium has been supported in various studies (e.g. Chapman, 1983; Mayr & O’Hara, 1986; Brooks et al., 2001; De Klerk et al., 2002; Küper et al., 2004). Using geomorphology derived from satellite imagery, Nichol (1999) likewise showed only one forest block as a historical refuge, but placed it slightly further to the north-west. Our results suggest that this single block can probably be divided into at least two refugical blocks based on similar amphibian assemblage. One covers the area around the Taï National Park in Côte d’Ivoire, and the other an area in south-western Sierra Leone. Although the Mano River may act as the barrier between these refugia, the scale of our analyses and the little data available for Liberia do not allow more precise positioning.
We report a clear distinction of WA amphibian assemblages from those of other African regions (central, eastern and southern Africa). Within WA, the intra-regional change in similarity of amphibian assemblages is gradual west of the Cross River. The latter is the most prominent and distinct barrier clearly delineating the hotspot ‘Guinean forests of West Africa’ from those in the east. Several other barriers of lesser importance structure the assemblage composition within WA.
Our analyses do not allow any firm conclusions on barriers between the other large African regions (east, south & central). Judging from the rarefaction results, where the cumulative number of species increases from region to region and sudden slopes indicate sharp transitions, one could speculate that a sharp boundary between east and southern Africa does not exist. This combined region is differentiated from that of CA by having an ‘arid corridor’ that ranges approximately from the Horn of Africa to the Cape of Good Hope (Balinsky, 1962; Poynton, 1995). Herpetologically, this disjunction is supported by data from hyperoliid frogs (Seymour et al., 2001) and reptile assemblages (Wagner et al., 2008).
Our demonstration of the uniqueness of WA amphibian assemblages shows that specific conservation plans are needed for WA. These should prioritise mainly the area between the rivers Volta and Kolenté. This is the most species rich and consequently outstanding area in WA. In addition, it contains a high number of threatened and endemic species. Within this area, amphibian assemblages of two sites are outstanding: Mt. Nimba and the Taï National Park. The former contains mountain grassland habitat, which is very rare in WA: the latter comprises the largest protected lowland rain forest in WA. Both sites are threatened by anthropogenic activities, i.e. mining (Hillers et al., 2008a) and logging (Hillers et al., 2008b), respectively. Their surroundings already comprise highly fragmented forest, or are now cleared of forest (Chatelain et al., 1996; Mayaux et al., 2004), which places further pressure on the fragments.
We are grateful to the German Ministry for Education and Research (BMBF) for funding the BIOTA-West project (funding number 01LC0617J). We further thank the German Science Foundation (DFG; VE 183/4-1, RO 3064/1-2), the Rapid Assessment Program and Critical Ecosystem Partnership Fund of Conservation International, the World Wide Fund for Nature and BirdLife International for funding field work in West Africa. Our gratitude also includes all the responsible governments and their ministries for issuing permits, and the numerous field assistants who provided invaluable help. We are indebted to Bill Branch who checked language and content.
Johannes Penner is interested in the macroecology of West African herps. He uses various statistical tools to assess distribution patterns, focusing mainly on amphibians. Other groups studied are monitor lizards, chelonians and selected snake species. Currently, he is with the herpetology working group of the Natural History Museum in Berlin.
Author contributions: J.P., M.W., M.S. and M.-O.R. drafted the research; J.P., A.H. and M.-O.R. gathered the data set; J.P. analysed the data; J.P. and M.-O.R. led the writing. All authors read, commented on, and approved the final manuscript.