Tropical savannah ecosystems are characterized by extensive grasslands with more or less sparse trees and thickets, and are threatened globally by anthropogenic forces. These grassland habitats house a rich and diversified fauna assemblage, with some of its conspicuous elements (for instance, small mammals) that have not been sufficiently investigated so far. In this paper, we meta-analyze the literature data available on the community structure and diversity patterns of shrews and rodents in West African savannahs. Overall, 10,197 small mammal individuals belonging to 111 species of Rodentia and 55 species of Soricomorpha were found in the various studies carried out in the countries covered by the present study. Studies using a combination of methods (e.g., live trapping, pitfalls, cover boards, visual encounter) detected more species in both Soricomorpha and Rodentia, and there was a positive survey (= trap ⁄ night) effort effect on the species richness in rodents. GLM models showed (i) that there was also no effect of trapping design (transect versus grid) on species richness per site, (ii) in both rodents and soricomorphs, the number of savannah species by country depended on the total species richness of that given country, but there was no effect of the relative surface covered by savannahs in that country. The number of sympatric species per site was 2.73±1.7 (range = 1-7) in Soricomorpha and 6.33±3.8 (range = 1-15) in Rodentia. Dominance index was significantly different among countries, with Nigeria having lower values than all other countries and Ghana, Benin and Sierra Leone had significantly higher values. The conservation implications of the observed patterns are discussed.

Tropical savannah ecosystems are characterized by extensive grasslands with more or less sparse trees and thickets, and are threatened globally by anthropogenic forces producing broad-shifts in woody vegetation that tend to homogenize their structure (McCleery et al. 2018). These environments house a remarkable diversity of small mammal species (e.g., Happold and Happold 1991; Decher and Bahian 1999), but concerning the Afrotropical regions there is no comprehensive study meta-analyzing the diversity patterns, the number of sympatric species by site, and more in general the community structure of savannah rodents and shrews. Conversely, several local studies are available (e.g., Happold and Happold 1991; Decher and Bahian 1999; Papillon et al. 2006; Olayemi and Akinpelu 2014; Decher et al. 2021), but is still unclear whether the locally observed patterns may be generalized to broad savannah regions in the African continent.

The case of the savannahs of Western Africa is especially interesting, given that the species diversity of various animal groups has been shown to be remarkably different in West Africa compared to the rest of the savannah environments in the African continent. In this paper, we provide a meta-analysis of the available data on small mammal (shrews and rodents) communities from the West African savannahs, with a focus on (i) the species richness patterns (diversity, dominance, evenness) per site, (ii) the species’ detectability by surveying methods, and (iii) the geographic variations, if any, of the observed patterns.

We obtained data from the literature by searching across ISI Web of Knowledge and Google Scholar (key words: “Small mammals west African savannah” (31,900 results); “rodent west African savannah” (13,000 results); “shrews west African savannah” (3,359 results) from 20 to 25 October 2021. We selected for further analysis only articles with the following characteristics: (i) they present raw data of numbers of individuals captured in the various study areas; (ii) they used only standard methodology to capture animals, with only data from carefully explained trapping and not, for instance, opportunistic sightings or general surveys being reported; and (iii) they report explicit mentioning of the study place-names. Some studies, albeit interesting, did not provide this type of data: for instance, they presented figures without raw data. These latter types of articles were not considered in the present paper. The various selected articles, although presenting fully-analyzable data, presented details of bait and trap type (live traps, pitfalls, etc.), trapping design (transect versus grid), and survey effort (expressed as trap ⁄ nights) in a heterogeneous way across studies. However, details of some of these aspects were not explained in some of the reviewed studies and hence were not taken into account in this review. For the present study, we considered the following categories of habitat: (i) Guinea savannah, (ii) Sudanese savannah, (iii) Sahel savannah. Some studies reported additional habitats than just the ones mentioned above (for instance, suburbs, forest, etc.). In these cases, we did not consider the data relative to these additional habitats for our analyses. Overall, we re-analysed datasets for 118 populations of rodents (in 10 countries) and 116 of soricomorphs (from 11 countries) (Table 1).

Species richness by country was compiled using data from Amori et al. (2012). Each species was attributed to the savannah category by checking the “habitats” search filter in IUCN (2019).

For each site, and only for rodents due to a sufficiently high species richness per site, we calculated various diversity indices used to analyze the community: species richness (i.e. the number of species that were captured in each site), Shannon and Weaver (1949) diversity index, Piélou (1966) evenness index, Dominance index and Simpson’s diversity index (Piélou 1969, Pearson and Rosenberg 1978). Piélou’s evenness index allows to evaluate whether the individuals are equitably distributed among the species of the target site and varies between 0 and 1. It tends towards 0 when almost the totality of the captured individuals is concentrated on one species and towards 1, when all species have the same abundance within the given sample. We also used the Chao-1 bias corrected index. This is an estimate of total species richness:

where F1 is the number of singleton species and F2 the number of doubleton species.

Bootstrap analysis was used in order to generate upper and lower confidence intervals for the various indices. For the bootstraps, we generated 9,999 random samples, each with the same total number of individuals as in each original sample being generated (Harper 1999).

Differences in the frequencies of savannah species by country, between rodents and soricomorphs, were analyzed by χ2 test. A General Linear Model (GLM) was performed in order to analyze the effects of total number of species by country and the relative coverage of savannah habitat by country (introduced as independent variables in the model) on the number of savannah species of both Rodentia and Soricomorpha (dependent variable) (Hosmer and Lemeshow 2000). The identity of the link function and a normal distribution of error were used (McCullagh and Nelder 1989). Mann-Whitney U test was used to assess the differences between Soricomorpha and Rodentia in terms of mean number of sympatric species per site. Kruskal-Wallis ANOVA was used to evaluate the among-country differences in the mean number of species per site. Spearman’s rank correlation coefficient was used to evaluate the correlation between number of employed methodologies and species richness at each site. Effects of trap ⁄ nights effort on the number of rodent species detected were assessed by Pearson’s correlation coefficient. This analysis was not performed on Soricomorpha because their species richness per site was low (see below for details). Homoscedasticity and normality were verified by Levene’s test and Kolmogorov–Smirnov test prior applying parametric statistics. In the text, means are followed by ± 1 Standard Deviation.

The number of species of rodents and soricomorphs by country, with their relative percentage of savannah species, is given in Table 2. Overall, a total of 10,197 small mammal individuals belonging to 111 species of Rodentia and 55 species of Soricomorpha are found in the various countries covered by the present study. The synopsis of the species mentioned in the various studies is given in Appendix 1. The number of species mentioned in the various meta-analyzed studies was 74 (66.6%) for Rodentia and 27 (49.1%) for Soricomorpha. The mean percentage of savannah species by country was 83.3±15.8% in Rodentia and 76.0±22.3% in Soricomorpha. Although the frequencies of savannah species varied considerably from country to country (Table 2), there was no significant difference between rodents and soricomorphs (χ2=20.38, df=15, p=0.158).

We found that studies using a combination of methods (e.g., live trapping, pitfalls, cover boards, visual encounter) detected more species in both Soricomorpha and Rodentia (correlation between number of methods employed per site and number of species: rs>0.4, p<0.05). In addition, there was a positive survey (= trap ⁄ night) effort effect on the species richness in rodents (r=0.432, p<0.01). A GLM model showed that there was also no effect of trapping design (transect versus grid), including its interaction with survey effort, on species richness per site (F6,1=0.73, p=0.715).

The frequencies of studied populations of both Soricomorpha and Rodentia (Figure 1) did not differ significantly among countries (χ2=12.53, df=10, p=0.251), but Niger was more studied than any other country in terms of number of examined populations.

A GLM model revealed that, in both rodents and soricomorphs, the number of savannah species by country depended on the total species richness of that given country, but there was no effect of the relative surface covered by savannahs in that country (Table 3). In addition, there were no significant differences among countries in median soricomorph and rodent species richness per site (Kruskal–Wallis ANOVA: v2=7.00, df=6, p=0.214).

The number of sympatric species per site was 2.73±1.7 (range = 1-7) in Soricomorpha and 6.33±3.8 (range = 1-15) in Rodentia. The differences between the two groups in terms of mean number of sympatric species was statistically significant (Mann-Whitney U test, z=-4.79, U=361, p<0.0001).

For rodents, all but two sites (one in Cote d’Ivoire and one in Ghana) were adequately sampled, as the plateau in the new species discovered by individuals captured was clearly reached (Figure 2). Therefore, the diversity metrics were calculated only on the adequately sampled sites. The synopsis of the rodent community diversity metrics per site, including the bootstrapped confidence intervals, are given in the Table 4.

Dominance: this index was significantly different among countries (F7,118=4.534, p<0.0001); Tukey HSD post-hoc test revealed that Nigeria had significantly lower values than all other countries and Ghana, Benin and Sierra Leone had significantly higher values (in all cases, p<0.0001). Exactly the same intercountry pattern, but with opposite effects, were observed in all other indices (p<0.0001 in all cases), with Nigeria on the one extreme side of the continuum, and Ghana, Benin and Sierra Leone on the other extreme side of the continuum. However, in terms of Evenness index of the rodent communities (F7,118=3.342, p<0.01), the Tukey HSD post hoc test revealed that Ghana did not differ from Nigeria (p=0.114), whereas Benin and Sierra Leone were still very different (at least p<0.05).

Values of diversity indices for rodent communities are reported in Appendix 2.

As expected, given that we examined countries with savannahs being the most widespread vegetation zone, we observed that the mean percentage of savannah species was higher than that of forest species in each country, despite the forests are well known to house a higher species richness and a higher diversity of sympatric species than any other Afrotropical habitat (e.g. Kasangaki et al. 2003). Obviously, there are several small mammal species that are known to inhabit both forests and savannahs, so our reasoning applies only to those taxa that inhabit only one of these two biomes.

Our study revealed no effect of using line transects or grids on the probability of increasing the number of detected species by site, so, both can be used indifferently for biodiversity studies focusing on small mammals in savannahs. This fact contrasts with previous studies showing that line transects are more efficient than grids in terms of both number of detected species and number of captured individuals in small mammal surveys (Pearson and Ruggiero 2003). However, expectably our study revealed that a multimethodological approach (using live trapping, pitfalls, cover boards, visual encounter surveys) was better in finding higher numbers of sympatric species per site. Additional methods, such as examining indirect signs of presence for the various species, were however not analyzed in any of the reviewed studies, although being potentially useful. Thus, for defining a given methodology to survey the species composition of small mammals in Afrotropical savannahs, it should be preferred to apply a multi-methodological approach instead of focusing only on live trapping. There is a plethora of reasons that may explain the observed pattern. For instance, traps can be selective for body size categories, as some species are simply too large to fit into the commercially-made live traps (Longworth; Sherman) (Laurance 1992). A second reason may be that the area covered by a trapping grid or transect is generally smaller compared to the territory that can be surveyed with other methodologies. In addition, live trapping is usually time-consuming and not at all cost-effective, thus typically resulting in short-term studies especially in Afrotropical areas where the economic costs are an important constraint to the field research. Obviously, this is a minor problem in Europe or North America, where several long-term live trapping surveys have been made (e.g., Pucek et al. 1993; Krebs 2009; Kataev 2012; Amori et al. 2015; Henttonen et al. 2017; Casula et al. 2019). It is also important to remind that some species are behaviorally less likely to be trapped than others (e.g., Allan 2020). For instance, gerbils tend to avoid traps whereas murids are easily trapped (our unpublished observations).

Our study also revealed that there were no significant differences among countries in median soricomorph and rodent species richness per site, with the number of sympatric species per site being about 3 (range = 1-7) in Soricomorpha and about 6 (range = 1-15) in Rodentia. Also recent studies from Ghana and Benin revealed a number of species (in Ghana 3 for shrews and 12 for rodents; in Benin 2 for shews and 10 for rodents) that is consistent with our analyses (Nicolas et al. 2020; Decher et al. 2021), and the same is true for rodent communities in Malawi (9 sympatric species; Happold and Happold 1991). Interestingly, a relatively stable number of sympatric species per site was also seen in turtles from West African savannahs (Gbewaa et al. 2021), thus suggesting that the ecological conditions of the various West African savannah sites are functionally similar and support a inter-site similar number of sympatric species although not necessarily the same taxa.

Future studies should explore whether the history of the West African savannahs (largely a human-derived vegetation zone generated from deforestation during historical times, especially in the so-called Dahomey Gap area) has shaped the current diversity and community structure of small mammals, thus explaining the homogeneity in the observed patterns.