Introduction

Although up to 5 million km2 of the African savanna (the world’s largest savanna) still harbours low density human populations of under 25 people km−2, it is undergoing rapid transformation (Riggio et al. 2013). The population of Africa is expected to increase fourfold by 2100 (Gerland et al. 2014), and as the continent with the fastest population growth, agriculture is intensifying and extending (Haberl et al. 2005) across 25 countries from Senegal to South Africa. From an original extent of 13.5 million km2, 22% had been lost by 2000, and since then another 50% has been transformed to new cropland and villages (GlobCover 2010; Riggio et al. 2013). Despite the rapidity of change, and proof of negative impacts on both functional (Flynn et al. 2009) and overall diversity, effects of landuse change on biodiversity and ecological processes remain poorly quantified and little understood (Foley et al. 2011). This knowledge gap is worrying, given the view, widely-held by many international bodies that change in landuse from rangeland to cropland is not only inevitable, but desirable (e.g. the Food and Agriculture Organisation; http://www.fao.org/news/story/en/item/20964/icode/). To meet the human consumption needs of 2030, an additional 20% increase in crop production will require conversion of an area similar to that of France, Germany and the United Kingdom combined, and most of this land will be in Sub-Saharan Africa (McIntyre et al. 2009). Studies show that such changes in landuse can impact at a range of scales, resulting not only in biodiversity loss, but also loss in ecosystem function and ecosystem services (Carvalho et al. 2009; Foley et al. 2011).

Arthropods are amongst the most abundant of organisms (Otoshi et al. 2015) and their importance to ecological function and ecosystem services is increasingly appreciated. Ants, amongst the most diverse and abundant of arthropods (Hölldobler and Wilson 1994), perform a range of ecosystem services, including pollination (Landis et al. 2000; Gómez 2000; Carvalheiro et al. 2008; Wielgoss et al. 2014), soil mixing, nutrient transport, soil aeration (Graham et al. 2009), decomposition, carbon flux, provision of habitat (Del Toro et al. 2012), seed dispersal (Christianini and Oliveira 2010), and pest control (Henri et al. 2015). Species-rich ant assemblages have even been associated with higher agricultural yields (Wielgoss et al. 2014). They engineer and modify ecosystems (Folgarait 1998; Andersen and Majer 2004), influencing soil formation (Lobry De Bruyn 1999; Del Toro et al. 2012), and by modifying and regulating the distribution of resources (Jones et al. 1994), can alter the composition of other invertebrate assemblages (Andersen 1995; Munyai and Foord 2012). Just as ants can influence soils, the relationship also works in reverse, and several past studies have shown that plant (Picker and Samways 1996), and soil attributes (Wagner et al. 2004; Bestelmeyer and Wiens 2001; Johnson 2000).

Various studies have investigated the effects of grazing (reviewed in Hoffmann and James 2011, and more recently Arcoverde et al. 2017), fire (Parr et al. 2004) and conversion to plantations (de Queiroz et al. 2017) on ants in various habitats. In savannas, however, there appear to be no studies in the peer-reviewed literature on the effects of changing savanna into cropland or villages. Changes associated with cropland have been studied in forest and woodland habitats. Various studies (e.g. Perfecto and Snelling 1995; Yates and Andrew 2011; Salas-Lopez et al. 2017) have found that epigaeic ant diversity decreases with declining vegetation diversity, as a consequence of transformation to cropland or other habitat simplification. These findings suggest we should expect to see loss in species diversity in savanna systems in response to transformation to cropland and villages.

Given the speed and extent of transformation of the African savanna from rangeland into villages and agricultural cropland, we investigate for the first time how this transformation impacts ants, known to play a disproportionate role in ecosystem function. We anticipated that transformation of rangelands into cropland or villages would decrease ant species diversity, and alter assemblages, with potential loss of important functional groups. To determine the impacts of landuse (rangelands vs. croplands vs. villages), we asked (1) How does ant species richness vary with landuse, season, vegetation and soil variables? (2) Does ant assemblage composition vary with landuse, season, geographic location, vegetation or soil variables? (3) Are there indicator species associated with the different landuse types and do these species change with season? (4) How does the representation of foraging strategies change with landuse and season?

Methods

Study area

We sampled ants using pitfall traps at 42 sites (14 sites each within rangeland, cropland and village), with five pitfall traps per site. Study sites were located at two main locations (Ndengeza and Vyeboom villages) in the eastern part of the 30,000 km2 Vhembe Biosphere Reserve (S23.15°, E30.39°–S23. 31°, E30.40°), northeastern South Africa. South Africa’s largest national park (Kruger National Park) borders these rural communities. The landscape features open savanna characterised by Acacia nigrescens, Dichrostachys cinerea, Grewia bicolor alternating with savanna woodlands characterised by Terminalia sericea, Combretum zeyheri and Combretum apiculatum. Granite gneiss and migmatite form the major basement geology of the region. Archaean (the earliest geological age) granite and gneiss erode to form sandy soils at higher elevations, and clay soils in lowlands. The region experiences summer rainfall with dry winters and mean annual precipitation of about 450 mm (Mucina and Rutherford 2006). The landuse history of the study sites has been well documented. Until recently, there has been little human activity; historically the entire area was subjected to browsing and grazing by a variety of wild herbivores and occasional grazing from low-density Nguni cattle. Over the past four decades the area has been mainly under traditional management, with rangeland for cattle giving way to emergent villages and subsistence cropland. Large herbivore stocking rates in all rangeland sites fell within limits predicted by the relationship between mean annual rainfall and large herbivore biomass in African savannas (Coe et al. 1976), and cattle within the study area were present at densities of under 1.5 animals 10 ha−1, a density which when converted to metabolic biomass is well within the range of acceptable stocking rates. Using a combination of modelling and satellite photographic surveys across African savannas, human densities of over 25 people per km−2 have been associated with shifts from rangeland use to formation of cropland and villages (Riggio et al. 2013).

There were seven replicates for each land use type at each of the two main locations (Ndengeza and Vyeboom). These 14 replicates per landuse type were positioned in either (1) cattle rangeland (human density ≤ 25 km−2 within rangelands; cattle density within the aforementioned limits in rangelands); (2) newly established cropland (extant for under two decades), or (3) new villages (Ndengeza and Vyeboom, extant for circa five decades, with rapid growth over the past two decades (resulting in formation of new cropland), harbouring populations of under 6000 (occupancy per homestead ranged from 5 to 10 persons, with a mean of 4 children and/or grandchildren, 2 parents, and one other adult), with population densities of between 700 and 1000 people km−2. Sites were thus matched for village size, human population density, cattle density, and key environmental variables including MAP (Mean Annual Precipitation), altitude and soil type. Housing in the villages is usually at low densities of < 1 dph (dwellings per hectare). Stands are generally swept and are devoid of paving and tarred roads. They also include a mixture of orchards (mainly mango) and livestock, mostly chickens and pigs. Surrounding agricultural land mainly focuses on rain-fed maize cropping, and includes both active and abandoned crop lands. Rangelands currently are used for cattle grazing and harvesting of a wide range natural resources.

Ant sampling

At each of the 14 sites in each landuse type, we sampled epigaeic ants using five randomly-positioned pitfall traps in an area of 300 m2 at each site. The 14 sites were at least 3 km apart, and selected to reflect the environmental heterogeneity of the area. Within these sites, pitfall traps were 15 m apart. Traps were sampled once each in the wet and dry seasons. The pitfall traps (ø 62 mm) contained 50 ml propylene glycol solution, which does not attract nor repel ants (Munyai and Foord 2012). Pitfalls were left to settle for 5 days, reducing the ‘digging-in’ effect (Greenslade 1973). Pitfall traps were left open for 5 days in the dry season (April–June 2014) and for 5 days in the wet season (January 2015), to ensure adequate sampling (Munyai and Foord 2015). Specimens were sorted in the laboratory, identified to at least genus level and morpho-species defined for those that could not be identified to species level by comparison with reference collections housed in the Zoology Department, University of Venda in South Africa (Munyai and Foord 2015). Fisher and Bolton (2016) was used to update and confirm valid ant genus names. Species abundance was represented by the total amount of specimens in each pitfall. These abundances were then used as a measure of ant activity (Lasmar et al. 2017). We assessed whether sampling was sufficient by whether the incidence coverage estimators (ICEs) converged, for dry and wet seasons. We did not use Abundance Coverage Estimators, because the eusocial nature of ants means that there can be bias introduced into the data by traps placed near nests or on an ant trail.

Choice of habitat variables

We measured % active soil carbon, % soil mineralized nitrogen (N, as it is a measure of what will be available to plants during the growing season), % clay, % silt, % sand, % rock cover, % bare ground, as well as % vegetation cover and % leaf litter. Soil samples were collected in June 2014 using a soil auger of 80 mm (diameter) × 200 mm (length). A total of four soil sub samples were taken in each of the 14 replicates. Soil samples were sent to ZZ2 Laboratories (Mooketsi, South Africa) where they were analysed for active carbon, Phosphate, Magnesium and Nitrogen (PMN) (µgN/gDM/week), clay (%), Silt (%) and Sand (%). Habitat cover was quantified by placing a 1 m2 grid over each pitfall trap; this was then photographed, and images where analysed to calculate the percentage cover of bare ground, vegetation, leaf litter and rock.

Data analysis

Ant species richness

In order to assess how ant species richness varied with landuse, season and environmental variables, we used Generalised Linear Mixed Effects Models (GLMM). Log-transformed ant species richness was set as the dependent variable, landuse, season, % active soil carbon, % mineralizable nitrogen (N), % clay, % silt, % rock, % sand, vegetation cover, bare ground and leaf litter as fixed variables, with sampling site as random intercept (Gaussian error distribution, identity link). Log-transformed ant species yielded a better model fit than untransformed ant species richness. Models were run using the nlme library in R (Pinheiro and Bates 2016).

We tested for spatial autocorrelation of ant species richness using Moran’s I. Spatial autocorrelation was found, so we included various spatial autocorrelation structures, with and without nuggets. We chose the best model as that with the lowest Akaike Information Criterion (AIC) value by at least 2 units. We also compared the AIC of the best model with that of the null model (i.e., using only a constant). We calculated marginal and conditional R2 values to assess how much of the variance was explained by the models (Nakagawa and Schielzeth 2013).

Variation in assemblage composition with landuse, season, geographic location, vegetation and soil variables

To assess whether ant assemblages changed with landuse, season and geographical location, we used a PERMANOVA using 999 iterations in the vegan package of R (adonis function; Oksanen et al. 2016). We first checked for homogeneity of group dispersions, using a multivariate analogue of Levene’s test for homogeneity of variances. We also graphically represented the sites using an ordination (metaMDS, non-metric multidimensional scaling), ensuring the stress value was sufficiently low (i.e., < 0.2). The ordination was performed using a dissimilarity matrix calculated using a Bray–Curtis index on ant presence-absence data. We also ran this analysis with abundance data, while acknowledging that eusocial species may show bias in numbers owing to high capture rates near next sites and on trails, that changes in the relative abundances of species might also be expected, to help identify trends in the data. A simper analysis was used to assess which species were responsible for the differences in assemblages between the three landuse types.

To ascertain whether villages and cropland excluded species characteristic of the surrounding rangeland, we identified indicator species for each landuse using multilevel pattern analysis (‘IndVal.g’ function in indicspecies; de Cáceres and Legendre 2009). This considers a perfect indicator species to be always present in, and exclusive to a particular habitat. Scores range from 0 (no indicator value) to 1. The indicator value is calculated as the product of two quantities, A and B, where A is a probability of a species as an indicator of a landuse (or combination of landuses, e.g. rangeland and cropland), and B is a measure of how frequently the species is found at the sites of the landuse(s) being considered (de Cáceres et al. 2010). “Indicator values” are assigned to each species in each habitat, and the null hypothesis of no difference between habitats is tested using a Monte Carlo simulation, with 1000 randomizations. We used abundance data for this analysis, as the multilevel pattern analysis uses both incidence and abundance data.

Environmental variables explaining variation in ant assemblage among the three types of landuse

To assess which environmental variables best explained ant assemblage composition in the three landuse types, we used a bioenv analysis (package vegan). A bioenv analysis identifies which particular aspects (environmental variables) of the habitat provided in each landuse best correlate with the observed ant assemblages. Environmental variables included landuse, season, % active soil carbon, % mineralized N, % clay, % silt, % rock, % sand, % vegetation cover, % bare ground and leaf litter. We used a spearman correlation method with a Bray–Curtis similarity index on the assemblage presence-absence data. A Gower index was used for the environmental variables because the data set included factors (season and landuse). We then tested whether the results yielded by the analysis were significant, using a mantel test, with 999 permutations.

Representation of foraging strategies under different types of landuse and different seasons

We calculated the functional composition of the different assemblages as measured by the community-level weighted means (CWM) of trait values. CWM is defined as “the mean of values present in the community (assemblage), weighted by the relative abundance of taxa bearing each value” (Lavorel et al. 2008). First, we excluded species that were not very common, by removing species with abundances of ≤ 0.1% of the total. As many as 90 species fell into this category, so that only 36 species remained of the original 126, representing 98.6% of all ants trapped. Species were assigned a “1” or zero for whether they included various items in their diets. The categories were “predators”, “scavengers”, “nectarivores”, “honeydew”, “granivores (i.e., seed-collectors)” and “plant matter” (i.e., leaves, stalks and petioles) (Samways 1983; Löhr 1992; Hölldobler and Wilson 1994; Holm and Scholtz 2008). Foraging habits were determined through a combination of inspection of mandibular morphology and knowledge of foraging strategy for the different genera. These were used along with the matrix of ant species and their relative activity at the different sites to calculate CWMs. CWMs were then included in mixed models to assess how they changed with landuse and season and interactions between the two. Site was included as a random variable.

Results

Ant species richness

There were a total of 126 ant species caught in our traps, 114 and 111 in the wet and dry seasons, respectively. The ICEs for each season converged, so sampling was considered sufficient. Ant species richness was spatially-autocorrelated, i.e., sites with higher species richness were close to other sites with high species richness, regardless of treatment (Moran’s I observed = 0.42, expected = 0.01, p < 0.0001). Correcting for spatial correlation did not improve model fit, but we report the results of the spatially-corrected model here. Percentage of bare ground was collinear with percentage vegetation cover. Percentage vegetation cover gave a better fit than % bare ground for all models, so percentage vegetation cover was included in our models and % bare ground omitted. The best-fitting model (spherical spatial autocorrelation; no nugget) indicated that rangeland and village had significantly higher species richness than cropland. Furthermore, species richness was greater in the wet season, and decreased with soil clay content (Supplementary material, Fig. 1). There was an interactive effect with season, however. In the wet season, ant species richness was significantly lower in the village, and significantly greater in rangeland than in cropland, but not significantly different between the three landuses in the dry season (Table 1; Figs. 1, 2). Fixed effects in the best model explained 64.4% of the variation, random effects (site group and year) explained a further 10% of the variation (i.e. the model explained over 74% in total).

Table 1 Variation in ant species richness with landuse, season and environmental variables
Full size table
Fig. 1
figure 1

Species richness relative to landuse and season. The solid bar denotes median values

Full size image
Fig. 2
figure 2

Mean values for environmental variables across types of landuse (p < 0.001 indicated by ***p < 0.01 indicated by **p < 0.05 indicated by *)

Full size image

Variation in assemblage composition with landuse, season, geographic location, vegetation and soil variables

There were clear differences in assemblage composition associated with landuse and geographic location (Fig. 3). Rangeland was separate from cropland and village, as were the two main locations (i.e., Ndengeza and Vyeboom; Fig. 3). There were some slight differences in ant species composition between the two seasons, but this pattern was not as marked as for geographic location and landuse (Fig. 3). When presence-absence data were used, landuse explained 21.1% of the variance in ant assemblage differences (F = 16.8, p < 0.001, n = 28), and season a further 5.0% (F = 98.1, p < 0.001, n = 28), but geographic location had the greatest influence, explaining 24.3% of the variation (F = 38.7, p < 0.001, n = 28; adonis, package vegan). When abundance data were used, landuse explained most of the community differences with 12.2% of the variance (F = 6.7, p < 0.01, n = 28), geographic location explained another 8.9% (F = 7.0, p < 0.01, n = 28) and season a further 6.4% (F = 9.8, p < 0.01, n = 28).

Fig. 3
figure 3

Ordination produced by multidimensional scaling denoting location, landuse and season (stress value = 0.14). The two locations are represented by ovals around the points; Vyeboom is indicated by a solid line, Ndengeza by a dashed line. Cr cropland, Ra rangeland, and Vi village. “Dry” and “Wet” represent the two seasons

Full size image

Cumulative contributions of most influential species

When considering the difference between rangeland and cropland, 42% is explained by Anoplolepis custodiens (almost 50% more abundance in cropland), and 8% by Monomorium custodiens, and for differences between rangeland and village, 51% is explained by A. custodiens (Table 2).

Table 2 Results from SIMPER analysis, showing species accounting for 70% of the dissimilarities in ant assemblages with different landuse types
Full size table

Indicator species for the three landuses

Of the 126 species in the study, 43 emerged as indicators for a specific landuse; 20 species (46%) characterised rangeland, 12 (28%) cropland, and 11 (26%) villages. (Table 3). The greatest species overlap of indicator species was between village and cropland.

Table 3 Indicator values for ant species in the 3 types of landuse
Full size table

Environmental variables explaining variation in ant assemblage composition found in the three types of landuse

Percent soil mineralised N, % clay in soil, vegetation cover and leaf litter best explained differences in ant species composition, with a significant correlation of 0.35 (Mantel statistic r = 0.251, p = 0.001). Leaflitter had the greatest explanatory value of the four variables, with a correlation of 0.24; when combined with vegetation cover, the correlation was 0.31.

Representation of foraging strategies under different types of landuse and different seasons

The functional composition of the different assemblages as measured by the community-level weighted means (CWM) of trait values differed per type of landuse and season. There were significantly more scavengers in rangeland (0.94 ± 0.06; p = 0.008) and villages (0.96 ± 0.05; p = 0.0006) than in cropland (0.90 ± 0.11), and significantly more in the wet season in all landuse types (wet: 0.96 ± 0.04; dry: 0.09 ± 0.10; p = 0.0002). Granivores were significantly more active in rangeland, but this only held for the dry season, indicating a seasonal effect crop dry: 0.41 ± 0.23; range dry: 0.72 ± 0.11; village dry: 0.48 ± 0.48; crop wet: 0.46 ± 0.311; range wet: 0.38 ± 0.23; village wet: 0.42 ± 0.33; p < 0.05). There were significantly more ants that feed on plant matter in rangeland (0.72 ± 0.31; p < 0.01) than both cropland (0.53 ± 0.30) and village (0.50 ± 0.39), across seasons. There were significantly more honey-dew collectors (aphidocolous species) in villages (0.76 ± 0.22; p < 0.01) than in the other landuse types (crop: 0.63 ± 0.22; range: 0.71 ± 0.15), regardless of season. There was weak support for more (mean ± SD: crop: 0.70 ± 0.20; range: 0.70 ± 0.16; village: 0.79 ± 0.22; p < 0.06) predators in villages, but no difference in season or between rangeland and cropland. For nectar collectors, there were no significant differences between landuse and season (Fig. 4).

Fig. 4
figure 4

The functional composition of the different ant assemblages as measured by the community-level weighted means of trait values for foraging strategy for predator, scavengers, nectar-collectors, honeydew-collectors, seed-collectors and plant matter-collectors

Full size image

Discussion

This study shows for the first time that the ongoing transformation of African savannas from rangelands to cropland and villages can lead to a decrease in ant species diversity and loss of functional traits, and that ant assemblages associated with cropland and villages lack the complex combination of species that characterise rangelands. Species richness was best explained by landuse, season and % clay in the soil. Species composition also varied with % clay, but also with leaf litter, vegetation cover and % mineralised N. Given the key roles played by ants in ecosystem function and in provision of ecosystem services, and the potential ramifications of changes in ant species composition at a range of scales, the details of our findings are addressed sequentially.

Rangeland harboured the most species, in both dry and wet seasons (Fig. 1, Table 1), but the effect was particularly marked in the wet season. Species richness was significantly lower in the village in the wet season. Species richness was also greatest in all land uses in the wet season, which might be expected in situations where the difference in rainfall between wet and dry seasons is large (Delsinne et al. 2010). Heterogeneous and more complex habitats have been shown to support greater species richness (Pacheco and Vasconcelos 2012), and relative to rangeland, cropland harboured lower levels of vegetation cover, lower rock cover and less leaf litter. Thus our results are consistent with those observed in other systems, where reduced complexity has been associated with reduced ant diversity (e.g. Perfecto and Snelling 1995; Yates and Andrew 2011).

Ant activity declined with increased clay content in the soil. This effect could arise indirectly, through clay’s influence on vegetation structure, which in turn affects ant activity and species richness (Johnson 2000; Boulton et al. 2005; but see Munyai and Foord 2015), or directly, as soils with high clay content tend to be associated with limited nest availability. Reduced nest availability would be expected to be associated with lower ant activity as well as reduced relative numerical dominance of species, an effect that has been recorded for Anoplolepis custodiens in ant assemblages in nearby Kruger National Park (Parr 2008).

In this study, geographic location was the most important variable explaining the presence or absence of species, but landuse was also associated with distinct assemblages (Fig. 3). Location might be important because of local factors like soils and rockiness, topography, aspect, or microclimatic variation, as well as opportunities for colonisation. Season only explained a small amount of variance, which could be expected as ants likely do not disperse widely, although species may vary in their degree of activity in the different seasons, which would affect the likelihood and abundance of these species in traps. Where abundance data were used, landuse emerged as the most important predictor of species assemblage, followed by location; season remained relatively unimportant compared to these two. Although we acknowledge that abundance data is not always accurate for use in studies such as this for eusocial species, some patterns in abundance should be attributable to actual changes in abundance in each site, so it is possible that landuse affects relative abundances more than geographic location.

The environmental variables associated with different ant assemblage composition were % mineralized N, % clay in soils, vegetation cover and leaf litter. Leaf litter on the ground provides microhabitats for ants above and beneath litter, particularly in savannas where leaf litter is relatively rare (Munyai and Foord 2012; Pacheco and Vasconcelos 2012). This suggests that the addition of leaf litter to croplands could ameliorate changes to ant assemblage composition. As noted, clay-rich soils influence ant distribution by influencing plant distribution and through moisture retention (Johnson 2000; Boulton et al. 2005). It has long been known that even epigaic ant assemblages exhibit changes in species composition in response to changes in vegetation cover (e.g. Greenslade and Greenslade 1977; Pacheco and Vasconcelos 2012). Altered levels of mineralized N brought about by various management practices, like tilling, fertilizer addition or removal of mammalian herbivores in the different landuse types can affect different nutrient cycles, soil microorganism and vegetation patterns, all of which impact ant assemblages (Wang et al. 2001; Picker and Samways 1996; Bestelmeyer and Wiens 2001).

Of species accounting for dissimilarity between landuse types, Anoplolepis custodiens, a predatory generalist, explained about half of the dissimilarity between sites across all landuse types. A. custodiens, the pugnacious ant, is noted for its aggressive and predatory behaviour (Löhr 1992), and often dominates other ant species, impacting richness and assemblage composition, in both agricultural lands and natural vegetation across southern Africa (Samways 1990; Addison and Samways 2000; Parr 2008). Furthermore, it can be a pest to small livestock, attacking small mammals and domestic poultry (Prins et al. 1990). Activity for this species was 42% higher in cropland, and 5.2 times higher in village plots. Unlike Anoplolepis gracilipes, a species invasive to tropical Asia and Australia, which can cause marked changes to other faunal communities (Wetterer 2005), A. custodiens is native to the region. It is likely responding to disturbance by becoming dominant, and this tendency of A. custodiens to increase in response to disturbance has been observed elsewhere in the region (Addison and Samways 2000). In a comparison of rangeland with cropland, no other species explained over 10% of dissimilarity between sites. Comparisons between village and both cropland and rangeland showed the importance of Pheidole sp. 03 (megacephala gp.), which explained a little over 10% in both comparisons.

Thus, this study highlights how certain species benefit from open and disturbed ecosystems (in this study cropland and villages). Increased activity of A. custodiens in cropland and in villages could have far-reaching ecological impacts that are difficult to predict. For example, increased predation on other invertebrates has been shown to impact overall arthropod assemblages, and the honeydew-producing hemipteran mutualism that arises from its aphidicolous tendencies can alter both plant fitness and hemipteran survivorship (Holm and Scholtz 2008; Styrsky and Eubanks 2010). In a study conducted in this region, the mutualism between A. custodiens and the mealybug Planococcus citri promoted the populations of both organisms (Samways 1983).

That cropland and village shared almost half of their indicator species (Table 4), and rangeland and cropland had the lowest overlap, suggests that as vegetation is removed and transformed, the ant fauna tends towards homogenization, and a preponderance for species that are broad generalists (e.g. Monomorium and Tetramorium, both of which are native to Old World Tropics, but which now have global distributions). These two landuses (cropland and village) both have far greater intensities of human disturbance than rangeland. Indicator species in rangelands not only represented a wider diversity at both the level of species and genus; but also covered a broader range of ecological functions than those found in villages and cropland.

Table 4 Overlap of indicator species between treatment types
Full size table

Landuse and season impact foraging strategy

Although changes in ant species richness can occur with changes in landuse or vegetation, ant assemblages sometimes show remarkable resilience and maintain functional groups (see e.g. Yates and Andrew 2011). Here, however, we found quite marked differences in foraging strategies between the different landuses. This may be unsurprising, given the differences in environmental variables as a consequence of landuse type. Rangelands harboured the broadest suite of foraging strategies, and included significantly more scavengers, granivores, and plant-matter feeders than cropland. Representation of foraging strategy also changed with season, with the dry season favouring the activity of granivores, and the wet favouring scavengers. In excess of 70,000 A. custodiens were sampled in village sites, compared with < 15,000 in rangeland, and < 20,000 in cropland. This increase in predators and honey-dew collectors in villages accentuates that this already highly transformed environment (relative to rangeland) can be expected to face further changes to food web dynamics through sparing of honeydew-producing hemipterans (which can impact negatively on plants), and altering of arthropod assemblages through combinations of predation and defence that as yet remain unclear (Styrsky and Eubanks 2007, 2010).

Implications for African savannas

This study suggests that the hastening transformation of African rangeland to cropland and villages will have markedly deleterious impacts on ants across the continent, through lowering of species richness, emergence of less complex assemblages that likely do not provide the same quantity and quality of ecological and ecosystem services, and the emergence of assemblages with significant reduction in the representation of functional traits relative to rangelands. Even as this study, and those from other savanna systems show how change in landuse to cropland and villages diminishes biodiversity, ecosystem function and ecosystem services (Carvalho et al. 2009; Foley et al. 2011), the Food and Agricultural Organisation of the United Nations (FAO) continues to earmark the Guinea savanna to feed both Africa and the world (http://www.fao.org/news/story/en/item/20964/icode/). Left unchallenged, these new forms of landuse threaten to change the entire savanna system. Given the important roles played by ants in soil formation, plant and seedling recruitment, and impacts on other arthropods that in turn influence entire food webs, advocating landuse change to benefit humans without an understanding of the potential loss of ecosystem functions and services can ultimately prejudice not only the wellbeing of natural systems, but also the wellbeing of humans.