• Open Access

Biome transitions as centres of diversity: habitat heterogeneity and diversity patterns of West African bat assemblages across spatial scales


  • Jakob Fahr,

  • Elisabeth K. V. Kalko

J. Fahr (jakob.fahr@gmail.com), Inst. of Experimental Ecology, Ulm Univ., DE-89069 Ulm, Germany. – E. K. V. Kalko, Inst. of Experimental Ecology, Ulm Univ., DE-89069 Ulm, Germany, and Smithsonian Tropical Research Inst., Apartado Postal 0843-03092, Balboa, Panama.


It is widely accepted that species diversity is contingent upon the spatial scale used to analyze patterns and processes. Recent studies using coarse sampling grains over large extents have contributed much to our understanding of factors driving global diversity patterns. This advance is largely unmatched on the level of local to landscape scales despite being critical for our understanding of functional relationships across spatial scales. In our study on West African bat assemblages we employed a spatially explicit and nested design covering local to regional scales. Specifically, we analyzed diversity patterns in two contrasting, largely undisturbed landscapes, comprising a rainforest area and a forest-savanna mosaic in Ivory Coast, West Africa. We employed additive partitioning, rarefaction, and species richness estimation to show that bat diversity increased significantly with habitat heterogeneity on the landscape scale through the effects of beta diversity. Within the extent of our study areas, habitat type rather than geographic distance explained assemblage composition across spatial scales. Null models showed structure of functional groups to be partly filtered on local scales through the effects of vegetation density while on the landscape scale both assemblages represented random draws from regional species pools. We present a mixture model that combines the effects of habitat heterogeneity and complexity on species richness along a biome transect, predicting a unimodal rather than a monotonic relationship with environmental variables related to water. The bat assemblages of our study by far exceed previous figures of species richness in Africa, and refute the notion of low species richness of Afrotropical bat assemblages, which appears to be based largely on sampling biases. Biome transitions should receive increased attention in conservation strategies aiming at the maintenance of ecological and evolutionary processes.

Quantifying and explaining the spatial distribution of life on Earth is a central focus of contemporary ecological research. In most taxa, species richness increases from the poles towards the equator (Hillebrand 2004). Since standardized data collection has been rarely achieved over broad spatial extents, many studies analyzed drivers of species richness using large sampling units such as gridded range maps or point records generalized to larger areas (Lyons and Willig 1999, 2002, Ceballos and Ehrlich 2006, Orme et al. 2006, Davies et al. 2007). Accordingly, these studies focused on the regional scale as their underlying data do not account for lacunarity or range porosity, that is an increasing loss of species with increasing spatial resolution (Hurlbert and White 2005, Hurlbert and Jetz 2007). The causative mechanisms driving species richness are still hotly debated, and some of the conflicting results might be explained by the scale-dependency of species richness (Rahbek 2005).

A major conceptual advancement has been the recognition that local and regional processes act in concert to result in a community or, more neutrally defined, a point estimate of overlapping regional species distributions (Ricklefs 2004). At regional scales, speciation, extinction, and immigration create, over evolutionary time, regional species pools. At local scales, habitat selection and species interactions as well as stochastic processes may be important. To predict species richness in relation to environmental conditions requires an understanding of the relative contribution of these processes along a spatially nested hierarchy (Ricklefs 1987, 2004, Cornell and Lawton 1992, Loreau 2000). The landscape scale connects local and regional scales and thus is of immense interest for studying patterns and causes of species richness (Böhning-Gaese 1997, Whittaker et al. 2001).

Within ecological time, regional diversity sets the limit for species richness at the local scale. To identify processes that determine local diversity, we need to ask how beta diversity, or species turnover, links regional and local scales. In Whittaker's (1960) multiplicative approach, regional diversity (in his terms gamma diversity) is the product of beta diversity and local (or alpha) diversity. However, this approach does not allow direct comparison of the relative contribution of these factors because regional and local diversities are measured as the number of species (or related units that incorporate the abundance of species), while beta diversity is dimensionless. Alternatively, diversities can be partitioned additively where regional diversity=local diversity+beta diversity (Lande 1996, Loreau 2000, Veech et al. 2002). This additive approach defines beta diversity as species turnover and therefore is well suited to analyze the relative contributions of diversity components across spatial scales (Wagner et al. 2000, Crist et al. 2003, Summerville et al. 2003, Freestone and Inouye 2006, Veech and Crist 2007).

Habitat heterogeneity is considered an important mechanistic factor driving species richness: only few species are found in all habitats, hence an increase in habitat types should lead to more species when sampled across a landscape (Rosenzweig 1995, Kerr et al. 2001, Tews et al. 2004). Several studies assessed this relationship across large spatial extents and employed variables such as altitudinal range or number of land cover classes per grid cell as proxies for habitat heterogeneity (Kerr and Packer 1997, Rahbek and Graves 2001, Van Rensburg et al. 2002, Ruggiero and Kitzberger 2004). Since these studies employed relatively coarse grain, habitat heterogeneity might have been missed as an explanatory variable of species diversity because ecologically relevant heterogeneity is likely to be perceived by organisms at finer grain depending on factors such as body size and dispersal ability.

We assessed diversity and assemblage structure of bats (Chiroptera) in two largely undisturbed areas in Ivory Coast, West Africa, and asked which factors drive species diversity from local to regional scales. We employed a spatially explicit and nested design that ranged from local to regional scales to account for the influence of spatial grain and extent (Wiens 1989, Whittaker et al. 2001, Ricklefs 2004, Rahbek 2005). We analyzed constant sample units (plots) that were hierarchically grouped within landscapes, hence keeping the sample grain invariant while changing the sample “focus” or area of inference (Scheiner et al. 2000). The extent of the landscape scale was chosen to match the dispersal abilities of our study group. As bats show a broad suite of habitat-related adaptations, most notably in their sensory systems (echolocation, passive listening, vision, and smell) and morphology (wing shape) (Norberg and Rayner 1987, Neuweiler 1989, Schnitzler and Kalko 2001, Safi and Dechmann 2005), we hypothesized that species richness of bats should be positively related to environmental heterogeneity as heterogeneous habitats should offer more niches than uniform ones. We differentiated between habitat complexity and habitat heterogeneity (August 1983), where heterogeneity is defined as the horizontal variability or patchiness of a habitat and complexity refers to the development of vertical strata within a habitat. In our approach, heterogeneity of vegetation types is taken as the most relevant habitat parameter for the majority of bat species as well as the most commonly used variable in previous studies (Tews et al. 2004).

Our study was conducted in two contrasting landscapes along the steep climatic gradient of West Africa, which is characterized by the staggered arrangement of biomes that stretch from the rainforest zone in the south through various savanna types up to the Sahara Desert in the north. Variables such as annual precipitation, actual evapotranspiration, and net primary productivity decrease along this S-N-gradient while seasonality increases (Tateishi and Ahn 1996, Imhoff et al. 2004). If water-related variables best explain broad-scale patterns of species richness of animals in the (sub)tropics (Hawkins et al. 2003), species richness of bats should monotonically increase from deserts to forests. If habitat heterogeneity drives species richness, one would expect a unimodal gradient with a peak at intermediate latitudes corresponding to the structurally most heterogeneous region along the biome transition (“Guinea Zone”) between forests and savannas (Goetze et al. 2006).

We hypothesized first that species diversity increases with habitat heterogeneity through the effects of beta diversity. Second, we postulated a positive relation between habitat complexity and species diversity. Third, habitat type rather than geographic distance should explain diversity patterns across spatial scales. Fourth, we expected that the structure of functional groups within a habitat type is not a random draw from the combined landscape assemblage but a selectively filtered subset. Fifth, we hypothesized that the reputed impoverishment of Afrotropical bat assemblages (Findley 1993) is largely due to sampling biases.

Material and methods

Study sites

We assessed bat assemblages in two areas in Ivory Coast, West Africa: Taï National Park (TNP) and Comoé National Park (CNP), which are ca 500 km apart. TNP (4550 km2) is located in southwestern Ivory Coast and constitutes the largest protected rainforest in West Africa in conjunction with the adjacent “Réserve de faune du N′Zo” (790 km2). The study was carried out in the vicinity of the “Centre de Recherche en Ecologie” station (CRE; 5°50′N, 7°21′W). The rolling landscape (ca 200 m a.s.l.) consists of a mosaic of drier and wetter parts. The climate is subequatorial seasonal, with an annual precipitation of 1813±268 mm in the study area (1978–1982, 1988–2002; Taï Monkey Project unpubl.) and two dry seasons: a minor one in July–August and a major one from December to February. Floristically, TNP belongs to the “Guineo-Congolian regional centre of endemism” (White 1983) and the “Western Guinean lowland forests” ecoregion (Olson et al. 2001). Our study area was composed of a mosaic of evergreen forest on the lower slopes with patches of deciduous trees on hill tops (Van Rompaey 1993). Apart from the clearing around the research station, treefall gaps and a few sparsely vegetated inselbergs, the study area is covered by a closed canopy.

In TNP, we differentiated between two major forest types according to their physical structure: hill forest (“forêt sèche”) on slopes and hill tops vs swamp forest (“bas-fond”) on the bottom of seasonally flooded valleys. Hill forest was characterized by high stature of mature trees and a comparatively open understorey (shrub and herb layer). Swamp forest had a higher density of smaller trees and a denser understorey, mainly composed of Raphia palms and Marantaceae. We established six plots arranged in three pairs, where one plot of each pair represented hill forest and the other swamp forest. Distances between plots were 0.2–2.8 km (median: 2.1 km), with distances between paired plots of 0.2–0.3 km, and distances between plot pairs of 1.0–2.6 km. Despite the short distance between paired plots, mark-recapture data showed that very few bats crossed from one plot to its neighbouring pair (19 out of 844 marked bats [2.3%], and 17.6% of all recaptures [108 individuals]), while recapture rate within plots was high (83 recaptured bats [9.8%], and 76.9% of all recaptures), thus justifying to treat each of the paired plots as an independent sample.

CNP (11 493 km2) is located in northeastern Ivory Coast and represents the largest protected area in the savanna zone of West Africa. The study was conducted around the former research station of the Univ. of Würzburg (Lola Camp: 8°45′N, 3°49′W, ca 200 m a.s.l.). The landscape generally is flat but occasionally broken by inselbergs or low rocky outcrops. The climate is of humid Sudanian type, with an annual precipitation of 1003±173 mm in the study area (1993–2002; Univ. of Würzburg unpubl.), a single dry season from November to March, and a wet season from April to October. Floristically, the southern portion of CNP belongs to the transition zone between the “Sudanian woodland with abundant Isoberlinia” and the “Mosaic of lowland rainforest and secondary grassland” (White 1983), which is part of the “Guinean forest-savanna mosaic” ecoregion (Olson et al. 2001). The study region is characterized by a matrix of bush-tree savanna with embedded patches of semi-deciduous forest islands ranging in size from >1 ha to several km2. Extensive gallery forests with evergreen elements occur along the main rivers Comoé, Iringou, and Kongo. The wider stretches of gallery forest and some larger forest islands structurally resemble rainforest and show floristic affinities to Guineo-Congolian lowland forests (Hovestadt et al. 1999). The three main habitat types (savannas: covering 84.2% of the area; forest islands: 8.4%; gallery forests: 2.3%) result in an overall mosaic-like landscape structure with clearly defined edges between vegetation types (Hovestadt et al. 1999, Hennenberg et al. 2005, Goetze et al. 2006).

In CNP, we sampled bat assemblages in three vegetation formations: open bush-tree savanna (“savanes boisée” and “savanes arbustive”), forest islands, and gallery forest. Initially, we established two plots in each of these three habitat types. A third savanna plot was added during the second half of the study period, resulting in a total of seven plots. Distances between plots ranged between 1.4 and 13.9 km (median: 5.4 km). The large distances compared to TNP result from one plot in gallery forest that was situated rather far from the other plots.

Sampling design

Each plot comprised 12 mist nets arranged in a standardized configuration along a 200×100 m-rectangle (2 ha), with equidistant (50 m) centres of the nets. The nets were oriented in an alternating fashion perpendicular to one another. These “understorey nets” (UN) were set on poles near ground level or slightly elevated, with the lower net edge level with the surrounding soil or herb layer. In addition, we set up one elevated net system in each plot, which consisted of a pulley and rope structure to hoist four stacked nets usually reaching a height of ca 25 m. These “canopy nets” (CN) were established within, or close to, the rectangle formed by the understorey nets, in TNP within natural treefall gaps, in CNP either within gaps (forest plots) or between emergent trees (savanna plots). All mist nets measured 12×2.8 m (16 mm mesh; 70 denier/2-ply netting) with four or five shelves. Furthermore, one two-bank harp trap (4.2 m2 capture area; Faunatech) was set in each plot.

Each plot was typically sampled for two consecutive nights per field season. Capturing lasted from dusk (ca 18:30) until dawn (ca 06:30). Mist nets and the harp trap were checked every 30–60 min throughout the entire night. We did not capture during nights around full moon phases, and in rare cases interrupted sampling because of heavy rain.

Bats were measured (forearm, body mass) and their sex, age, and reproductive status assessed. Most individuals were identified to species in the field and subsequently released. A few bats were sacrificed to check identifications. These synoptic collections are deposited in the Forschungsinstitut Senckenberg, Frankfurt/M., and in the research collection of JF. All adult bats except for insectivores with <10 g body mass and males of Epomophorus gambianus, Epomops buettikoferi, and Hypsignathus monstrosus were individually marked with a stainless steel ball-chain necklace and a serially numbered aluminium band. Males of the three species of fruit bats were not marked as they inflate their throats during courtship calls.

The study comprised eight field seasons in TNP between March 1999 and February 2004 (first part of the study [1999–2000]: J. Fahr; second part [2001–2004]: Stefan Pettersson, Göteborg Univ.). In CNP, we sampled bats during seven field seasons between April 1999 and June 2002 (first part of the study [1999–2000]: J. Fahr; second part [2001–2002]: Njikoha Ebigbo, Ulm Univ.). Capture seasons were selected to match similar conditions in phenology and climate, i.e. at the end of the dry season/start of wet season (TNP: Feb/Mar; CNP: Apr/May) and at the end of the wet season/start of dry season (TNP: Aug/Sep; CNP: Oct/Nov).

We also captured bats outside plots in an opportunistic fashion with mist nets set in understorey and canopy as well as with harp traps. Such opportunistic sampling (OS) targeted particular habitat types and situations (e.g. small creeks, clearings, and rocky outcrops), which were deemed to yield species that might have been missed in the standardized plots. Additionally, we included data from preliminary surveys in CNP during 1993 and 1995. Total capture effort for CNP and TNP combined was 1765.0 mist net nights and 102.6 harp trap nights (Table 1).

Table 1.  Capture effort expressed as mist net nights (UN: understorey nets, CN: canopy nets) and harp trap nights (HT: harp traps). 1 net night: one 12 m-net opened for 12 h, 1 trap night: one trap set for 12 h. Opportunistic sampling in CNP includes UN-data from 1993 and 1995.
 Standardized plotsOpportunistic sampling

Data analysis

Recaptures of marked bats were excluded from analysis if they were caught during the same sampling period in the same plot. Estimated species richness (Sest) was calculated with the programs EstimateS 7.5 (Colwell 2005). We followed Brose and Martinez (2004) for the choice of the least biased and most precise estimator to extrapolate estimated species richness, Sest. In a first step, we calculated Sest of a given sample with a suite of non-parametric and parametric estimators of species richness (Abundance-based Coverage Estimator [ACE], Incidence-based Coverage Estimator [ICE], First-order Jackknife [Jack1], Second-order Jackknife [Jack2], and Michaelis-Menten [M-M]). We then calculated the range of sample coverage (observed species richness [Sobs]/estimated species richness [Sest]) and its mean. Variation in estimated sample coverage was generally rather low (8–29%). In a second step, we chose the estimator recommended by Brose and Martinez (2004) as the final estimate of species richness for a given sample.

Interpolated species accumulation curves (sample-based rarefaction) of plot data were calculated with the “Mao Tau”-function in EstimateS (Colwell et al. 2004, Colwell 2005). The graphs were rescaled by individuals, resulting in individual-based rarefaction curves sensu Gotelli and Colwell (2001). Rescaling by individuals allows direct comparison of species richness as opposed to rarefaction curves scaled by samples, which represent species density (Gotelli and Colwell 2001). We used 95% confidence intervals to test for significant differences in species richness.

We followed Hill (1973) and Jost (2006) in the use of effective number of species when reporting diversity measures other than species richness. In short, the effective number of species equals species richness if all species of a sample have the same frequency and decreases with declining evenness of a sample. Shannon diversity, which is equivalent to Hill's (1973) N1 diversity index, was calculated as eH, with


and Simpson diversity was calculated as 1/D, with


where pi=the proportion of individuals in the ith species. We also employed the nonparametric estimator of Shannon entropy implemented in Spade 3.1 (Chao and Shen 2006), which accounts for unseen species in a sample, thus resulting in Shannon diversities (eĤ [est]) that are unbiased by sample size. We stress that frequency data derived from captures represent relative abundances of individuals, which in turn are affected by sampling bias of capture techniques. Since our sampling protocol was standardized, comparisons within our study system are valid since data are affected by the same bias.

Species richness was used to assess the total number of species in a sample (“diversity of order 0” sensu Jost 2006), Shannon diversity was employed as a diversity measure that weighs species directly proportional to their frequencies (“diversity of order 1”), and Simpson diversity was used as a complement to focus on the most frequent species in a sample (“diversity of order 2”). Evenness was calculated as E=eH/S, where S=number of species in a sample (Buzas and Hayek 1996). Evenness equals 1 if all species in a sample have the same frequency and decreases as samples are increasingly dominated by a few species, hence reducing the effective number of species (eH). We calculated observed evenness from observed Shannon diversity and Sobs as well as estimated evenness (Ê) derived from eĤ and Sest. For the latter, estimators were chosen based on sample coverage, see above.

Spatial variation in beta diversity

To analyse whether variation in assemblage composition among sites within a region (variation in beta diversity, Tuomisto and Ruokolainen 2006) is explained by geographical location (spatial autocorrelation), we ran independent Mantel tests for each study region. Plot data (relative species abundances per plot) were transformed as dissimilarity matrices based on the quantitative Sørensen (Bray-Curtis) index. Geographical locations of plots had been measured with a hand-held GPS (Garmin GPS II plus) and the Euclidean distances between the midpoints of each plot were calculated with ArcView 3.2a. We compared both distance matrices of each study region, using PC-Ord 5 (McCune and Mefford 2006) to run 10 000 Monte Carlo randomizations.

Additive partitioning

The program PARTITION (Veech and Crist 2009) was employed to assess additive partitioning of species richness for each study region (CNP, TNP). We tested the null hypothesis that the observed components of diversity at increasingly higher levels (α1, β1, β2,…, βi) could have been obtained by the random placement of individuals among samples at all hierarchical levels (Crist et al. 2003, Crist and Veech 2006). For this approach, the observed numbers of individuals of each species are randomly placed among samples at the lowest hierarchical level, and these samples are then grouped into progressively larger samples at each higher level. Under this null model, each species is not constrained to a particular sample (in our case representing a specific habitat type) but reshuffled among samples, thereby effectively removing the influence of species-specific associations with a particular habitat type (see also Veech and Crist 2007). The program PARTITION places individuals randomly in samples while preserving the original species-abundance and sample-size distribution. For randomizations (10 000), we arranged our samples (plots) in replicates corresponding to their spatial location in the landscape, thus matching a nested design. For TNP, this resulted in three replicates each containing two samples representing distinct habitats (hill forest, swamp forest). For CNP, this resulted in two replicates each containing three samples representing distinct habitat types (forest islands, gallery forests, savannas). Samples were weighted by the relative number of individuals in each sample, i.e. each sample received a weight equal to the individuals in this sample divided by total number of individuals.

Our analytical design spanned three hierarchical levels from the local to the landscape scale (Fig. 1). Additive partitioning allows the expression of the proportional contributions of diversity at each level in this hierarchy. Since diversities are calculated as an average of the samples at a given level regardless of how they are nested within the next higher level, this approach is robust to unbalanced sampling designs (Summerville et al. 2003). Although the number of structural vegetation types, and therefore our sampling design, differed between CNP and TNP, the relative contributions of diversity components across spatial scales can be compared between both study regions.

Figure 1.

Schematic representation of the hierarchical levels studied in CNP. For TNP, the arrangement differed in that there were two habitat types (rather than three) and three replicates (rather than two). The right-hand circles illustrate how each lower level adds to the next hierarchical level (γ=α112).

Functional group composition

Bats were classified into five broad functional groups following Schnitzler and Kalko (2001) based on diet (frugi- and nectarivorous [F] vs animalivorous [A]), foraging mode (gleaning [g] vs aerial [a]) and habitat (degree of structural clutter: narrow space [NS; foraging within dense vegetation], edge and gap [EG; foraging close to, but not within dense vegetation], open space [OS; foraging distant from vegetation]) (Appendix).

We assessed whether habitat type structured functional group composition of assemblages by testing whether the observed composition in a given habitat type conformed to a random sub-sample of each study area (TNP, CNP) or if the proportional composition of functional groups shows habitat-specific patterns. We employed the program Resampling Stats (Resampling Stats 2006) to create 10 000 random assemblages for each habitat type. These were constrained by drawing the observed number of species without replacement from the species pool of each study region (TNP: 40 species, CNP: 57 species). Statistical significance was calculated as the proportion of null values greater than (or less than) the observed values. This proportion is a p-value that indicates the probability of obtaining a value as great as (or as small as) the observed value by chance.


Landscape diversity and sample coverage

We captured a total of 75 species, 22 of which were shared between the two study areas (Appendix). We recorded 40 species in TNP, with 32 species caught in plots (P) and eight species captured opportunistically (OS). The total for CNP was 57 species, with 51 species recorded in plots and six species that were found off-plot.

Standardized plot data revealed significantly higher species richness for CNP than for TNP when rarefied to the assemblage with the lower number of individuals; thus, at 1569 individuals, TNP had 39 species (upper 95% CI: 42 species) whereas CNP had 50 species (lower 95% CI: 45 species; Fig. 2). The completeness of sampling was similar for both areas as indicated by high sample coverage (Sobs/Sest– CNP: 80.3%, TNP: 78.5%) despite the very unequal number of individuals captured in each study area (Table 2). Including data from opportunistic sampling, estimated sample coverage increased to 88.1–91.5% for CNP and 87.7–95.5% for TNP, respectively. At this level of sample coverage, the magnitude of the higher species richness of CNP compared to TNP was much more pronounced (TNP: 39 species, upper 95% CI: 42 species; CNP: 55 species, lower 95% CI: 51 species).

Figure 2.

Estimated number of species (extrapolated species accumulation: mean Jackknife 1±1 SD) for CNP (black diamonds) and TNP (grey circles), and sample-based rarefaction curves (interpolated species accumulation: Sobs) with 95% confidence intervals rescaled by individuals.

Table 2.  Species richness of CNP and TNP broken down to approach (P: plots, OS: opportunistic sampling, pooled: P+OS) and method (UN: understorey net, CN: canopy net, HT: harp trap, combined: UN+CN+HT). Species numbers in brackets refer to those captured with a single method. Sobs: observed species richness, Sest: estimated species richness, eH: observed Shannon diversity, eĤ: estimated Shannon diversity, Eobs: observed evenness, 1/D: Simpson diversity. Sest– a: Michaelis-Menten, b: ICE, c: Jackknife 1, d: Jackknife 2. In CNP, one additional species (Nycteris gambiensis) was found in its day roost, another additional species was found in its day roost and recorded by its echolocation calls (Rhinolophus landeri); in TNP, one additional species (Myotis bocagii) was captured when flying into station building.
 Comoé NPTaï NP
 CombinedUN pooledCN pooledHT pooledCombinedUN pooledCN pooledHT pooled
 pooledPOS   pooledPOS   
Pteropodidae99999188778 (1)2
Emballonuridae1111 (1)111 (1)
Nycteridae5525 (4)15434 (3)11 (1)
Hipposideridae6556 (2)1366454 (1)4
Vespertilionidae20171313 (5)12 (4)5 (3)111067 (1)59 (3)
Molossidae13134213 (11)7275 (1)6 (2)
Sobs55513536 (11)38 (16)10 (3)39322829 (5)25 (5)17 (4)
eH12.112.2 9.811.
eĤ12.312.4 13.86.411.4
Eobs (eH/Sobs)0.220.24 0.430.250.57

Although species richness (Sobs) was much greater using understory and canopy nets than with harp traps, each method (UN, CN, HT) yielded species not documented with other methods (Table 2). In CNP, observed Shannon diversity (eH) was highest for the combined sampling methods (both P and P+OS), decreased slightly for mist net samples (both understorey and canopy level), and was lowest for harp trap samples. In TNP, Shannon diversity was highest for mist net samples at understorey level, intermediate for harp trap samples and combined methods (P and P+OS), and lowest for canopy samples. For most samples in both study areas, estimated Shannon diversity (eĤ) was only marginally higher compared to observed Shannon diversity, showing that the influence of undetected species on Shannon diversity was negligible due to high sample coverage. Only in TNP, understorey net samples (12.4 vs 13.8) and harp trap samples (9.7 vs 11.4) had noticeably higher estimated than observed values.

Observed evenness generally was very similar among methods and between samples except for harp trap samples (both CNP and TNP) and understorey net samples in TNP, which had a much higher evenness than the remaining samples (Table 2). Finally, Simpson diversity (1/D), which emphasizes the most frequent species, was highly similar among CNP samples except for HT, which had a lower diversity. Simpson diversity in TNP was high for UN- and HT-samples, intermediate for combined methods (P and P+OS), and low for CN-samples.

The estimated total species richness (Sest) of CNP when based on a single method (UN, CN, HT) was lower than the observed number of species (Sobs) and much lower than the estimate derived from all methods combined. In TNP, samples from understorey and canopy nets estimated a species total close to the observed as well as estimated richness of all methods combined while harp trap samples predicted roughly two-thirds of the total species richness.

Spatial variation in beta diversity and additive partitioning of species richness

Variation in community composition among plots was uncorrelated with geographic distance in CNP (Mantel test: r=0.372, p=0.2146) in contrast to TNP (Mantel test: r=0.520, p=0.0067). However, when substituting the average community dissimilarity value of TNP-plots for the three plot pairs (P1–P2, P3–P4, P5–P6) in close proximity to each other, geographic distance no longer had a significant effect on community composition (Mantel test: r=0.183, p=0.2171). We obtained similar results (not shown) for all tests when applying Jaccard or relative Sørensen distance indices.

At the local scale (plots: α1), additive partitioning revealed significantly lower species richness both in TNP and CNP compared to the individual-based randomizations, that is, a lower proportion of species richness was found on this level than expected from a random distribution of species between plots in both study regions (Table 3, Fig. 3). CNP had significantly higher beta diversity on both levels (β1: among plots representing different habitat types, β2: between replicates) compared to the null model of random placement of species. Both beta components of species richness in TNP did not differ from the null model. In TNP, 41% of total species richness was apportioned on the beta levels while in CNP the combined contribution of beta diversity to total species richness was 50%. The study areas also differed in the proportional partitioning of beta diversity on the two hierarchical levels of our analysis: in CNP, a larger fraction of observed species richness was found on the first level (β1), corresponding to high species turnover between plots representing different habitat types, whereas in TNP a larger fraction of species richness was found on the second level (β2), representing a higher species turnover between replicates (paired plots).

Table 3.  Additive partitioning of species richness for standardized plots in CNP and TNP. Observed values compared to mean (min-max) values of individual-based randomizations (10 000) and the proportion (p) of randomized values with a diversity estimate greater than the observed. Arrangement of samples conform to the spatially nested design (CNP: three plots each representing different habitat types on the first level, which are aggregated into the second level [two replicates]; TNP: two plots each representing different habitat types on the first level, which are aggregated into the second level [three replicates]). Note that diversity partitions add up to total Sobs in plots (α112; CNP: 51, TNP: 32).
 Comoé NPTaï NP
 Observed diversity% of total diversityExpected diversitypObserved diversity% of total diversityExpected diversityp
Within plots (α1)25.650.3%33.1 (30.4–35.5)>0.999918.959.0%20.4 (18.5–21.8)0.9996
Among plots (β1)15.630.5%11.7 (8.5–14.8)<0.00014.714.6%3.9 (1.9–5.8)0.0715
Between replicates (β2)9.819.2%6.3 (3.8–9.4)0.00018.526.4%7.7 (5.9–10.1)0.0866
Figure 3.

Additive partitioning of diversity components (α1, β1, β2) for CNP (left) and TNP (right) as the percentage of total species richness. Expected values are derived from the means of 10 000 individual-based randomizations (Table 3).

Local (alpha) diversity

In CNP, estimated species richness was highest in gallery forests and one savanna plot (CP5), while estimated Shannon diversity was highest in gallery forests, medium in forest islands, and lowest in savanna plots (Table 4). In pair-wise comparisons (95% CI), Shannon diversity in gallery forest plots (CP3, CP4) was significantly higher than in forest island and savanna plots except for one forest island plot (CP2), which did not differ significantly from CP3. Estimated evenness was very similar among plots except for CP5, which had a much lower evenness. Despite the high species richness in savanna plot CP5 (both Sobs and Sest), Shannon and Simpson diversities were very low as a result of the disproportionately high dominance of two fruit bat species (Micropteropus pusillus and Nanonycteris veldkampii), which is also reflected in the much lower evenness of this plot. In TNP, estimated species richness was very similar among plots except for one located in swamp forest (TP5), which was predicted to have much higher species richness. Estimated Shannon diversity and evenness was variable and not consistently related to habitat type (in pair-wise comparisons [95% CI], only TP4 was significantly more diverse than TP2). Plot TP5 was distinguished by the exceptional dominance of one fruit bat species (Eidolon helvum), which might have been attracted to food resources nearby. The same plot also had a very high number of singletons (7), which resulted in a high Sest.

Table 4.  Species diversities of plots in CNP and TNP. Indiv.: individuals, Sobs: observed species richness,% of Sobs (total): percent of total Sobs (CNP: 51, TNP: 32), eH: observed Shannon diversity, Eobs: observed evenness (eHobs/Sobs), 1/D: Simpson diversity, Sest: estimated species richness, Est.: estimator used for Sest (Jack1, Jack2, Jack3),% of Sest (total): percent of total Sest (CNP: 63, TNP: 41), eĤ [est]: estimated Shannon diversity, eĤ [est] SE: standard error of estimated Shannon diversity, Eest: estimated evenness (eĤest/Sest).
  Indiv.Sobs% of Sobs (total)eHEobs1/DSestEst.% of Sest (total)eĤ [est]eĤ [est] SEEest
CNP P1forest island3012345.1%9.10.406.332.7Jack251.9%
CNP P2forest island2452345.1%9.90.436.429.0Jack146.0%
CNP P3gallery forest1792549.0%13.10.529.639.0Jack361.9%
CNP P4gallery forest2092956.9%14.30.4910.044.2Jack270.1%
CNP P5savanna4773058.8%
CNP P6savanna1851937.3%8.00.425.828.2Jack244.7%
TNP P1swamp forest1591856.3%6.10.343.021.3Jack152.0%6.71.310.31
TNP P4swamp forest1892062.5%9.60.486.324.3Jack159.2%
TNP P5swamp forest3262165.6%
TNP P2hill forest2821856.3%
TNP P3hill forest1051650.0%8.40.535.923.8Jack257.9%
TNP P6hill forest2461856.3%8.80.496.022.3Jack154.4%

Overall, species richness of plots was weakly and positively correlated with Shannon diversity (Fig. 4; linear regression for observed and estimated values, respectively: eH=2.230+0.295 Sobs, R2=0.207, p=0.137; eĤ=2.503+0.230 Sest, R2=0.293, p=0.069). Species richness and evenness were independent from each other (linear regression for observed and estimated values, respectively: E=0.488–0.004 Sobs, R2=0.026, p=0.617; Ê=0.412–0.003 Sest, R2=0.075, p=0.390). Both observed and estimated mean species richness of CNP-plots was significantly higher than TNP (t-test, Sobs: mean(CNP)=24.8, mean(TNP)=18.5, t=3.463, 10 DF, p=0.006; Sest: mean(CNP)=35.4, mean(TNP)=24.7, t=3.245, 10 DF, p=0.009).

Figure 4.

Linear regression between species richness and Shannon diversity of plots. Above: observed; below: estimated. (TNP-plots: TPx, CNP-plots: CPx, Table 4).

There were neither significant differences between TNP and CNP with respect to Shannon (both observed and estimated) and Simpson diversities nor with regard to evenness of plots. However, potentially significant differences on the level of Shannon and Simpson diversities might have gone undetected as a result of low sample size (n=6 for TNP and CNP, respectively). On the level of observed species richness, plots in CNP harboured a significantly lower proportion of total species richness compared to TNP (t-test,% of Sobs: mean(CNP)=48.7%, mean(TNP)=57.8%, t=−2.286, 10 DF, p=0.045). However, there was no significant difference between CNP and TNP when based on estimated species richness (t-test,% of Sest: mean(CNP)=56.2%, mean(TNP)=60.3%, t=−0.642, 10 DF, p=0.535).

In CNP, forest islands had significantly lower species richness than gallery forests and did not differ from savannas when Sobs was rarefied to the smallest sample (GF, Table 5). Although the corresponding values of Sobs differed widely between gallery forests and savannas, the significance level was just missed due to wide confidence intervals for both habitats. Estimated species richness (Sest) of gallery forest approached the total estimated for CNP and was much higher than in savannas, which in turn had higher Sest than forest islands. Shannon diversity (eH) was significantly higher in gallery forests compared to both forest islands and savannas while no significant difference was found for Simpson diversity (1/D; note very wide confidence intervals in the latter). Observed evenness decreased from gallery forests to forest islands and savannas, whereas estimated evenness was similar among forest habitats but lower in savannas. Higher evenness in forest habitat, particularly in gallery forest, was caused by much lower capture frequencies of the two dominant fruit bats (Micropteropus pusillus, Nanonycteris veldkampii).

Table 5.  Bat diversity in relation to habitat type (CNP – GF: gallery forest, FI: forest island, SA: savanna; TNP – SF: swamp forest, HF: hill forest). CI: 95% confidence interval; Sest– a: Jackknife 2, b: ICE, c: Michaelis-Menten, d: Jackknife 1; rarefied: comparison restricted to the sample with the lowest number of individuals.
HabitatIndiv.SobsCI (Sobs)SesteHeĤ [est]CI (eĤ)1/DCI (1/D)EobsEest
 GF [P3-4]3883931.8–46.262.1a18.019.715.4––20.20.460.32
 FI [P1-2]5462826.5–29.530.5b–35.6c10.210.68.0––11.50.360.30–0.35
 FI – rarefied39026.324.6–28.1        
 SA [P5-6]6623429.4–38.643.6d7.57.95.1––
 SA – rarefied38828.123.8–32.4        
 SF [P1,4,5]6742820.1–35.936.5d8.18.46.1––
 SF – rarefied63927.519.8–35.3        
 HF [P2,3,6]6332220.9–23.123.1b–24.4c8.38.45.3––8.00.370.34–0.36

TNP-plots grouped by habitat type did not differ in any of the diversity measures (Sobs when rarefied to the smaller sample [HF], Shannon, Simpson). However, hill forests had lower species richness (Sest) but higher observed and estimated evenness than swamp forests.

Structure of functional groups in relation to habitat type

In CNP, the composition of functional groups showed several significant departures from the null model of a random draw of species into local assemblages (Table 6). The bat assemblage of forest islands was characterized by a significantly higher richness of frugi- and nectarivorous gleaning narrow space bats (FgNS) and by a significantly lower richness of animalivorous aerial open space bats (AaOS). The assemblage of gallery forests constituted an almost perfect random draw from the species pool. The savanna assemblage showed the greatest departure from the null model: richness of animalivorous aerial narrow space bats (AaNS) was significantly lower, but higher for AaOS bats; FgNS bats tended to be more species-rich than expected (p=0.0517) while the functional group of animalivorous gleaning narrow space bats (AgNS) tended to be impoverished (p=0.0854).

Table 6.  Functional group composition of bat assemblages in relation to habitat type: observed compared to expected species richness (10 000 randomizations: mean and 95% CI in parentheses; significant differences [proportion of observed values ≤ or ≥ randomizations] shown in italics). Functional group classification modified from Schnitzler and Kalko (2001; Appendix): F – frugi- and nectarivorous, A – animalivorous, gNS – gleaning narrow space, aNS – aerial narrow space, aEG – aerial edge and gap, aOS – aerial open space; Landscape: pooled data from plots and opportunistic sampling, SF: swamp forests; HS: hill forests, FI: forest islands, GF: gallery forests, SA: savannas.
SampleFunctional groupTotal
 FIobs9 (p=0.0008)4672 (p=0.0029)28
 FIexp4.4 (2–7)4.4 (2–7)3.9 (1–6)8.3 (5–12)6.9 (4–10) 
 GFexp6.2 (4–8)6.1 (4–8)5.5 (3–8)11.6 (8–15)9.6 (7–12) 
 SAobs832 (p=0.0368)912 (p=0.0214)34
 SAexp5.4 (3–8)5.4 (3–8)4.8 (2–7)10.1 (7–13)8.3 (5–11) 
 SFobs8 (p=0.0443)4772 (p=0.0048)28
 SFexp5.6 (3–8)4.9 (3–7)4.9 (3–7)7.0 (4–9)5.6 (3–8) 
 HFobs8 (p=0.0036)3641 (p=0.0092)22
 HFexp4.4 (2–7)3.9 (2–6)3.9 (2–6)5.5 (3–8)4.4 (2–7) 

In TNP, the AaOS group exhibited lower species richness both in hill and swamp forest while the FgNS group had higher species richness than expected by the null model in both habitat types. Most other functional groups showed proportional sampling close to that expected from the species pool although AaNS bats tended to be more species-rich in both habitats compared to the null model, albeit not significantly so (p=0.0612 and 0.0827 for SF and HF, respectively).

At the landscape level, the functional group composition of both assemblages (TNP, CNP) constituted random draws from the regional species pool recorded for Ivory Coast (data not shown). We obtained the same result when comparing the observed composition of each study area (TNP: 40 species, CNP: 57 species) with a random draw from the combined species total of TNP and CNP (75 species) or with a random draw from the entire species pool of Ivory Coast (87 species; Fahr unpubl.).


We discuss our results in a spatially hierarchical order. First, we assess patterns of local diversity in relation to habitat type. Second, we evaluate how composition of functional groups is a function of differential recruitment from the species pools as mediated by habitat type. Third, we show how patterns on the local level scale up to the landscape level through species turnover (beta diversity), which is linked to habitat heterogeneity, and introduce a conceptual model that aims to dissect the contribution of habitat complexity and heterogeneity along biome gradients. Finally, we discuss our findings in the light of previous studies on tropical bat assemblages.

Relating local (alpha) diversity to habitat type

We used the contrasting landscape configuration and habitat types of CNP and TNP to assess the influence of habitat complexity on local diversity and found that habitat type had a pronounced influence on local diversity in CNP but not in TNP. With the exception of one savanna plot (CP5), estimated species richness decreased from gallery forests through forest islands to savannas. Shannon diversity revealed a similar pattern except for one pairwise comparison between forest island and gallery forest (CP2-CP3). These results suggest that local bat diversity is positively related to complexity along the vertical axis. It remains unclear, however, why gallery forests supported higher local diversity than forest islands in CNP. Possibly, the linear structure and narrow width of gallery forests might have fostered higher permeability for savanna species along the edge.

At the level of species richness, mean single-plot diversity was significantly higher in CNP than in TNP (i.e. across habitat types). Considering that plots size and sampling were standardized, local diversity in CNP was possibly increased through a “spillover” of species between different habitat types. This could have been partly caused by our sampling design, which deliberately included edge habitat, and may constitute a mass effect (sensu Shmida and Wilson 1985) through species occasionally extending their core habitat into adjacent habitat types. Compared to closed forests, heterogeneous landscapes such as the forest-savanna mosaic of CNP comprise much more edge habitat (Hennenberg et al. 2005), thus offering more opportunities for species foraging in this situation. When broadly viewed on the plot level, none of the estimates of species richness approached the respective values on the landscape scale, indicating that spatial patterns of species aggregation and turnover found on the local scale precluded accurate estimation of species diversity on the landscape scale.

Our results contradict those of Rautenbach et al. (1996), who studied the effects of structural complexity on bat species richness along a latitudinal transect across Kruger National Park, South Africa. They found no significant differences in diversity patterns between paired plots in gallery forests and woodlands. While their study of bat assemblages is one of the few with a largely standardized sampling protocol, sampling bias has probably seriously impacted their results. First, it is likely that many species were missed since neither elevated nets nor harp traps were used. Second, edge habitat but not the forest itself was sampled with large mist nets employed perpendicular to gallery forest, which probably explains the surprisingly low number of species in gallery forests adapted to dense vegetation (Rautenbach et al. 1996).

To summarize, we propose that the generality of species diversity-habitat complexity relationships is critically dependent on how much a given group utilizes the habitat along the vertical axis. While some studies of non-volant mammals revealed links between species diversity and habitat complexity, others did not (August 1983, Williams et al. 2002, and references therein). The classic studies by MacArthur (MacArthur and MacArthur 1961, MacArthur 1964), however, demonstrated a striking relationship between species diversity of birds and structural complexity, which has been recently corroborated with high-resolution remote sensing data (Goetz et al. 2007).

Structure of functional groups in relation to habitat type

Earlier studies demonstrated a close link between ecomorphological and -physiological characters of bats and their habitat-specific foraging patterns, which were in turn related to the structure of species assemblages (McKenzie and Rolfe 1986, Aldridge and Rautenbach 1987, Crome and Richards 1988). We expected that the proportional composition of functional groups would mirror the different habitat types when compared to the composition on the landscape scale. We predicted that three functional groups foraging in dense habitat (FgNS, AgNS, AaNS) should be overrepresented in TNP-plots as well as in forest habitat in CNP (FI, GF) while the group foraging in open space (AaOS) should be underrepresented. We further expected the opposite of this pattern in savanna plots of CNP. For the group foraging in edge habitat (AaEG) we anticipated proportional sampling in both CNP and TNP. These predictions were only partially confirmed (Table 6). In CNP, functional groups showed partly idiosyncratic responses. For example, the functional group composition of gallery forests constituted a random draw compared to the species pool of the landscape scale whereas two functional groups (FgNS and AaOS) showed significant deviations in forest islands in accordance with our prediction. Two of four functional groups (AaNS and AaOS) deviated from proportional sampling in the savanna assemblage of CNP according to our expectation and another group (FgNS) showed a similar trend. Both forest types of TNP showed the closest agreement with our predictions, with two functional groups (FgNS and AaOS) revealing a significant departure and a third (AaNS) evincing a trend. The proportionally lower presence of the AaOS group should be treated with caution, however, since this group is notoriously difficult to sample with mist nets and harp traps, especially in forest habitat such as TNP where this group was only within reach of our elevated mist nets in large canopy gaps.

We expected that habitat type would lead to a more pronounced shift in proportional composition of functional groups. However, since the null model compared observed and expected values based on incidence data, possible shifts in the relative abundance structure of assemblages remained undetected. Moreover, the occasional sampling of species usually foraging in a different habitat type might have blurred our analysis – especially for the habitat mosaic of CNP. Most interestingly, the group composition of both landscape assemblages (CNP and TNP) constituted almost perfect random draws from the regional species pool. We anticipated this result for CNP, where the habitat mosaic should lead to proportional sampling when analyzed on the landscape scale (i.e. pooling group compositions from distinct habitat types). This result was surprising for TNP, where we expected a preponderance of functional groups adapted to dense vegetation (FgNS, AgNS, AaNS) as well as a lower proportion of species foraging in open habitat (AaOS), and opens intriguing questions to which extent species richness on local to landscape scales is governed by the availability of suitable habitat (bottom up) or if overlapping distribution ranges determine the regional species pool in a way leading to proportional sampling on landscape to local scales (top down). Alternatively, our classification of functional groups might have been too coarse to detect a clear signal when analyzing the pooled communities on the landscape level.

Habitat heterogeneity and partitioning of diversity components

Beta diversity as measured on the landscape scale is a function of two non-exclusive mechanisms. First, if species are characterized by high habitat specificity, increase in habitat heterogeneity should result in increased beta diversity. Second, dispersal limitation can lead to higher beta diversity, particularly in the case of spatially discontinuous habitat patches (Mouquet and Loreau 2003, Freestone and Inouye 2006). Kadmon and Allouche (2007) modelled complex relationships between habitat heterogeneity and species richness depending on area effects as well as dispersal and reproductive rates, and habitat heterogeneity only had a monotonically positive effect on species richness when dispersal rates were high and habitat patches were large.

In the habitat mosaic of CNP, heterogeneity had a strong and positive effect on beta diversity, which was significantly higher among habitat types (β1) as well as between replicates (β2) than expected (additive partitioning). On the contrary, the rather uniform TNP did not show elevated patterns of beta diversity; i.e. landscape richness was neither significantly increased by a high turnover of species between habitat types (β1) nor on the second level of our analysis (among replicates: β2). Since CNP had a significantly higher mean plot diversity than TNP at the local scale (both Sobs and Sest), we asked whether the higher landscape diversity of CNP was driven mainly by local (alpha) or beta diversity. The additive partitioning approach revealed that both factors contributed about equally to landscape diversity in CNP, but local diversity was significantly lower and beta diversities were significantly higher than expected by the null model. In TNP, the local scale contributed much more to the landscape scale (59.1%) than in CNP, although local diversity was likewise reduced compared to expected values, while beta diversities did not deviate from the null expectations.

Further support for a causal relationship between habitat heterogeneity and landscape richness is shown by the spatial partitioning of beta diversities on the two hierarchical scales in CNP. The nested plots representing different habitat types contributed almost two-thirds to landscape diversity although plots on this level were spatially much closer to each other than on the next hierarchical level. In a habitat mosaic such as CNP, community similarity should vary from high to low depending on whether similar or dissimilar habitat types are being compared. Beta diversity should correlate with geographic distance only if dispersal ability is low, thus imposing a filter on community similarity with increasing distance (Nekola and White 1999, Freestone and Inouye 2006). However, this was rejected for CNP by the Mantel test, indicating that dispersal limitations between habitat types were negligible at the distances covered by our study and due to the high mobility of bats. This matches results by Veech and Crist (2007), who found no decay of community similarity for bird assemblages within North American ecoregions but a positive relation between habitat heterogeneity and bird diversity at the landscape scale.

Furthermore, we employed a grain size that was smaller than the patch size of distinct habitat types, thereby allowing for habitat-specific responses in assemblage composition. Larger grain sizes would have averaged beta diversity over different habitat types (Nekola and White 1999), such that changes in species turnover could have been detected only over larger distances and linear gradients. The lack of decay in community similarity with geographical distance in CNP matched the spatial configuration of the study region. Here, plots were located in distinct habitat types that were defined by sharp and stable boundaries (Hennenberg et al. 2005, Goetze et al. 2006). In contrast, structurally defined habitat types in the rainforest region of TNP lacked pronounced boundaries. Parallel to the gradual rather than abrupt changes in floristic composition with local topography in TNP (Van Rompaey 1993), turnover in the composition of bat assemblages was unrelated to structurally defined habitat types. Mantel tests showed that the significant variation of beta diversity with distance was the result of very similar assemblage structures in neighbouring plots (paired sampling design) although they represented different forest types. This pattern broke down when we used average similarity values for the three neighbouring plots pairs, indicating that habitat differences had a minor influence on assemblage patterns at this spatial scale. Interestingly, community composition of leaf-litter anurans in the very same plots in TNP was explained by geographic distance rather than by environmental variables (Ernst and Rödel 2005). Here, low mobility of leaf-litter anurans seems to result in assemblages that are predicted by geographic distance rather than environmental factors.

High mobility or dispersal rates do not necessarily lead to homogenization of assemblages in distinct habitat patches if the latter are characterized by contrasting environments. In CNP, forest and savannas are structurally very different, hence species adapted to dense habitats should have disadvantages when foraging in open savannas with strongly reduced complexity, while species adapted to open habitats might be almost completely excluded from entering dense habitat because of ecomorphological and -physiological constraints (Aldridge and Rautenbach 1987, Schnitzler and Kalko 2001). In our view, the structure of bat metacommunities on the landscape scale might be largely regulated by the filter properties imposed by physical habitat parameters (e.g. vegetation density), rather than by species interactions within a given habitat. Thus, we suggest differentiating between strict dispersal limitation on the one hand (i.e. organisms with low mobility, or those that rarely cross unsuitable habitat) and on the other hand high dispersal but low recruitment due to specific habitat preferences (i.e. organisms that regularly cross or encounter unsuitable habitat without establishment of permanent populations). We suggest that variation in local habitat conditions lead to intraspecific aggregation of bats species to the extent that these conditions meet their habitat requirements. Intraspecific aggregation may reduce local diversity (He and Legendre 2002, Veech et al. 2003), in agreement with our findings of reduced diversity at the local scale (plots) both in TNP and CNP, albeit with a much higher effect in CNP. The latter finding is in accordance with the pronounced contrasts between savanna and forest habitats in CNP, which should result in higher intraspecific aggregation in relation to habitat type (i.e. within distinct patches), and a stronger reduction of local diversity when compared to expectations of random placement.

Our study is one of few to assess the influence of habitat heterogeneity across spatial scales in natural habitat mosaics; most previous research has been conducted in anthropogenically fragmented landscapes (Tews et al. 2004). Recent anthropogenic fragmentation might lead to qualitatively and quantitatively different patterns as the regional species pool from which local assemblages are recruited might be fundamentally different from a natural biome transition such as CNP, where habitat types as well as distribution ranges of species interdigitate. In Paraguay, Stevens et al. (2004) documented much higher diversity of bats in a forest-savanna mosaic compared to a fairly uniform forest region and linked this pattern to the effects of habitat heterogeneity. Similarly, the importance of forest habitats in the southern part of CNP, which cover a mere 10.7% of this region (Hovestadt et al. 1999), is illustrated by the high species overlap between CNP and TNP where more than half of the species recorded in TNP also occur in CNP. In this context, gallery forests along larger rivers such as the Comoé might be critical for linking forest populations in the rainforest zone with more isolated populations in the forest-savanna mosaic to the north. Habitat heterogeneity also predicts diversity patterns of non-volant mammals in Australia (Williams et al. 2002). The latter study demonstrated scale-dependency, where heterogeneity became an excellent predictor of species richness at larger spatial scales, showing that the effects of heterogeneity depend on the perceived grain of the study organism. The lack of heterogeneity-diversity relationships reported by Cramer and Willig (2005) might be explained by the microscale at which they studied rodents.

Mixture effects of habitat complexity and heterogeneity on species richness across biomes

Since a vast majority of studies evinced a positive relationship between habitat heterogeneity and species diversity (reviewed by Tews et al. 2004), we sought to conceptualize and integrate the effects of complexity and heterogeneity along a highly simplified gradient of biomes from forests to savannas and steppes, building on a virtual transect across West Africa (Fig. 5). As one moves from rainforests in the south to steppe in the north, we posit that habitat complexity decreases monotonically as vegetation strata are increasingly lost with diminishing stature of woody plants, finally giving way to biomes that are dominated by herbs and low bushes. Along this gradient, we conjecture that habitat heterogeneity is rather low over a broad climatic range, specifically, within the forest zone where canopy is broken mainly by elements such as treefall gaps and water courses. Once a critical threshold is crossed, increased seasonality and/or reduced precipitation leads to a biome transition between forests and savannas (Sankaran et al. 2005). Within this ecotone, habitat heterogeneity increases sharply until forest elements such as forest islands and gallery forests are completely lost further north. For flying organisms such as bats and birds utilizing both the vertical and horizontal axis, we hypothesize that local (alpha) diversity is driven mainly by habitat complexity while beta diversity results from heterogeneity. If true, landscape diversity would result from the joint effects of complexity and heterogeneity, giving rise to a pronounced peak within the biome transition between forests and savannas. A mixture model of habitat complexity and heterogeneity has been also invoked by Roth (1976) to explain contrasting patterns of bird diversity from North American grasslands to forests.

Figure 5.

Hypothetical model of the contribution of habitat structure on species richness along a forest-savanna-steppe gradient: habitat complexity decreases along this gradient with diminishing vegetation height, habitat heterogeneity shows an intermediate peak at the biome transition between forests and savannas, and the effects of habitat complexity and heterogeneity, in their sum, lead to a diversity peak at the biome transition.

We compared patterns documented by our study for bats with patterns of species richness of birds in CNP and TNP. In CNP, 494 bird species have been recorded (Salewski 2000). When we subtract Palaearctic migrants, which are mainly occurring in northern CNP, and aquatic species, 373 species remain for the southern part of CNP. For TNP, more than 230 bird species have been recorded (Gartshore et al. 1995), and 215 remain after excluding aquatic species and Palaearctic migrants. In contrast to bats and birds, plant diversity shows the opposite pattern, with 1233 species known from TNP and 720 species from CNP (Poilecot 1991, Hovestadt et al. 1999, Denguéadhé Kolongo et al. 2006). Parallel to plant diversity, actual and potential annual evapotranspiration are higher in TNP than in CNP (Tateishi and Ahn 1996: TNP: 1397 mm AET, 1515 mm PET; CNP: 1027 mm AET, 1388 mm PET). Estimated annual net primary production is almost twice as high in TNP as in CNP (Imhoff et al. 2004: TNP: 1177.6 g C m−2, CNP: 609.9 g C m−2). Contrary to hypotheses developed on the regional scale (Hawkins et al. 2003), landscape diversity of bats and birds does not match the marked differences in energy availability, productivity, precipitation, or floristic diversity. In conclusion, we suggest that species richness of bats and birds follows a mixture model integrating habitat complexity and heterogeneity, with peak richness found in the forest-savanna biome transition. In agreement with our study, Williams et al. (1999) documented pronounced peaks of both species richness and turnover for sub-Saharan birds in the transition zone bordering equatorial forests. Moreover, Kark et al. (2007) demonstrated that bird species richness peaks along ecoregion boundaries of the New World. Bridging these studies, which employed coarse sampling grains over large geographical extents, our study supports the notion that biome transitions harbour significantly increased diversity from local to landscape scales.

Biome transitions and ecotones are increasingly seen as significant centres of evolutionary processes for the maintenance, and possibly generation, of diversity (Moritz et al. 2000, Spector 2002). Our study lends further support to the importance of heterogeneous habitats for conservation strategies aiming at safeguarding ecological and evolutionary diversification (Kark et al. 2007). As a focal point, “keystone structures” within heterogeneous landscapes deserve particular consideration (Tews et al. 2004). In the habitat mosaic of CNP, these keystone structures constitute forest islands and gallery forests as they allow for the persistence of forest-adapted species in the surrounding savanna matrix. However, forest patches are increasingly lost through land use outside the protected area of CNP (Goetze et al. 2006), with potentially far-reaching consequences for the maintenance of ecological and evolutionary processes such as gene flow between populations along corridors such as gallery forests. Thus, systematic conservation planning should specifically consider biome transitions with an evaluation of their spatial connectivity to neighbouring source areas.

Species richness of Afrotropical bats

In a first attempt to characterize regional species richness of bats globally, Findley (1993) suggested that the Afrotropical region is impoverished compared to the Australasian and Neotropical realms. He further concluded that species richness of African bats does not peak in equatorial forests but rather in the grasslands and savannas of East Africa, where, according to his data, species richness reaches 60–70 species per 250 000 km2 grid cell compared to 100–120 species in the equatorial regions of the Neotropics and south-east Asia. Although this pattern has been widely cited (Kingston et al. 2003, Willig et al. 2003, Procheş 2005), we argue that species richness of bats has been largely underestimated in the Afrotropics.

Compared to the few available data, species richness of bats in CNP and TNP by far exceeds any African sites surveyed so far on similar spatial scales. Our figures from CNP even surpass documented bat richness of vast areas such as Kruger NP, Garamaba NP, or the Ivindo Basin (Table 7). Furthermore, what Findley (1993) called “East African grasslands and savannas” in fact comprises the Albertine Rift, the Eastern Arc Mountains, and the coastal forests of East Africa, all regions distinguished by pronounced habitat heterogeneity and known to harbour both high diversity and levels of endemism (Brooks et al. 2001). Second, East Africa has been historically much better explored than most of the Central and West African regions, hence data on species richness of Africa bats should be carefully evaluated against sampling artefacts. Finally, species richness in CNP compares well with results from studies in the Neotropics and Australasia, refuting a general impoverishment of Afrotropical bat assemblages. Future comparisons should be made with great caution to account for confounding effects such as sampling methods, sampling effort, and area effects.

Table 7.  Documented (Sobs) and estimated (Sest) species richness of Afrotropical bats in relation to area (CNP and TNP – #: estimated from samples, see Table 2; §: estimated from nearby records within the same biome, see Supplementary material; other areas –1: Rautenbach et al. 1996, 2: Verschuren 1957, 3: Brosset 1966).
Study regionSobsSestArea [km2]
Comoé NP50 6
 57 170
  65–71§entire park (11 500)
Taï NP4041–44#6
  44–50§entire park (4500)
Kruger NP, South Africa141 ca 20 000
Garamba NP, D.R. Congo2ca 40 ca 5000
Ivindo Basin, Gabon334 ca 2500

Conclusions and future perspectives

We believe that the mixture model of habitat heterogeneity and complexity on species richness is likely to have general applicability for African bats on local to landscape scales, and possibly also for other taxonomic groups such as birds. It will be interesting to see whether this pattern holds true for bats in other tropical regions when studied on similar spatial scales. Idiosyncratic differences among continents will be particularly exciting to analyze both with respect to the historic legacies of these regions and potential evolutionary constraints such as niche conservatism or divergent radiations (Gavrilets and Vose 2005, Wiens and Graham 2005, Stevens 2006).

Download the Supplementary material as file E5510 from <www.oikos.ekol.lu.se/appendix>.


We greatly appreciate the contribution of Stefan Pettersson, Njikoha Ebigbo and Katja Soer to the data set analyzed here. We thank Kouakou Kouamé, Koffi Kouadio and Kouadio Kouakou (all CNP) and Georges Gbamlin (TNP) for their dedicated assistance in the field. Vital logistical support was provided by K. Eduard Linsenmair, Frauke Fischer and the employees of the “Projét Biodiversité” at CNP, and by the “Centre de Recherche en Ecologie”, the “Projet Autonome pour la Conservation du Parc National de Taï”, and the “Taï Monkey Project” at TNP. Research permits to work in CNP and TNP were kindly granted by the “Ministère de l'Agriculture et des Ressources Animales” and the “Ministère de l'Enseignement Supérieur et de la Recherche Scientifique”, République de Côte d'Ivoire. Dieter Kock, Senckenberg Museum Frankfurt, offered invaluable help with species identification. Thomas Crist, Lou Jost, Martin Pfeiffer, Erica Sampaio and Joseph Veech helped in various ways with the development and interpretation of our analyses and in shaping our ideas. We appreciate general support by Hans-Ulrich Schnitzler and Mark-Oliver Rödel. This is a contribution of the BIOTA program, funded by the German Federal Ministry of Education and Research (BMBF, project 01LC0017, 01LC0411 and 01LC0617E1). We acknowledge additional funding by the Landesgraduiertenförderung BW and the German Academic Exchange Service (DAAD). Douglas Kelt, Egbert Leigh, Christoph Meyer and three reviewers helped to improve previous versions of the manuscript.


Appendix. Capture frequency of bat species in TNP and CNP, and their classification into functional groups.


  1. Re-use of this article is permitted in accordance with the Term and Conditions set out at http://www3.interscience.wiley.com/authorresources/onlineopen.html