Biodiversity responses to land-use and restoration in a global biodiversity hotspot: Ant communities in Brazilian Cerrado

Introduction

Although protected areas will always play a key role in conserving biodiversity, land-use systems such as agroforestry, exotic tree plantations, and selectively logged forests have significant biodiversity value (Bennett et al. 2006; Bhagwat et al. 2008), and such human-modified landscapes are increasingly recognized as being critical for biodiversity conservation (Gardner et al. 2009; Anand et al. 2010). Finding productive systems that provide social and economic benefits while reducing environmental impacts of land use on biodiversity and ecosystem services is, therefore, a challenge for the whole society (Foley et al. 2005). This was the motivation for the Millennium Ecosystem Assessment (2005), which was conducted to inform decision makers on how human actions have changed natural ecosystems and caused biodiversity losses, to show the consequences of those changes on ecosystem services and human well-being, and to indicate how management decisions can contribute to environmental conservation and therefore improve human well-being (Reid & Mooney 2016). There is a particular need to adopt land-use practices that maintain high levels of biodiversity (Karp et al. 2012), and to restore degraded land that previously had high biodiversity value (Harvey et al. 2008; Ferreira et al. 2012).

High biodiversity loss typically occurs when land-use systems involve marked simplification of vegetation structure (Karp et al. 2012), or substantial ecosystem inputs such as through application of fertilizers and pesticides (Perfecto et al. 1997; Tscharntke et al. 2005; Gibbs et al. 2009; Solar et al. 2016).

In these situations, restoration interventions can potentially reverse land degradation and enhance biodiversity and associated ecosystem services (Chazdon 2008; Rey-Benayas et al. 2009). However, although ecological restoration is considered a global priority for the future of biodiversity (Dobson et al. 1997; Aronson & Alexander 2013), the challenges in meshing science, practice and policy are still to be overcome (Hobbs & Harris 2001). There is a pressing need to understand the biodiversity impacts of alternative land-use systems, and the effectiveness of different restoration options (Suding 2011).

The Brazilian Cerrado is a global biodiversity hotspot (Myers et al. 2000) that has undergone rapid transformation due to agricultural expansion. The Cerrado is dominated by grassy ecosystems, and includes the world's most biodiverse savannas (Durigan & Engel 2012; Overbeck et al. 2013). In the last four decades, about half of its total area of 2 million km² has been converted to pasture, agriculture and silviculture (Beuchle et al. 2015; Overbeck et al. 2015). A further shift in land use has occurred over more recent decades, with pastures being replaced by cultivation of soybean (the largest Brazilian export commodity) and, most recently, by sugarcane for ethanol production (Lapola et al. 2014).

Current Brazilian environmental legislation requires every rural property to have at least 20% of its area as preserved or restored native vegetation, and this has driven a huge demand for Cerrado restoration. In the absence of validated techniques for this biome, forest restoration techniques based upon planting tree seedlings in high density have been widely applied in the Cerrado, although rarely documented (Bertoni 1992; Melo et al. 2004; Santilli & Durigan 2014). Innovative techniques have been tested, such as direct seeding (Pereira et al. 2013a,b; Silva et al. 2015; Cava et al. 2016), topsoil translocation (Ferreira et al. 2015; Pilon et al. 2017), hay transfer (Le Stradic et al. 2014a; Pilon et al. 2017) and transplantation of shrub and tree seedlings (Le Stradic et al. 2014b). The outcomes of Cerrado restoration, however, have not been adequately assessed. One problem is that indicators used to assess forest restoration, such as tree cover and biomass, have been uncritically applied to Cerrado (Durigan et al. 1998; Melo et al. 2009; Pereira et al. 2013a,b; Cava et al. 2016). The extent to which contemporary restoration techniques can recover the structure, biodiversity and functioning of the different Cerrado structural types, from grasslands to savanna woodlands is still to be demonstrated. Cerrado is therefore a globally significant biome for investigating both the biodiversity impacts of alternative land-use systems, and the effectiveness of different restoration strategies.

Here, we use ants as bio-indicators to examine the effects of different land uses and restoration practices on Cerrado biodiversity. Ants are an ecologically dominant faunal group in most terrestrial ecosystems (Folgarait 1998), and are especially diverse and abundant in Brazilian Cerrado (Campos et al. 2011). Ants are widely used as bio-indicators in land-use studies (Hoffmann & Andersen 2003; Andersen & Majer 2004), including in the evaluation of ecosystem restoration (Majer 1983; Andersen & Sparling 1997; Lawes et al. 2017). In our study, we examine ant functional as well as taxonomic diversity, in order to identify ant functional groups that are particularly sensitive to disturbance. We also examine the extent to which ant diversity and composition can be explained by vegetation attributes that apply across the full land management spectrum. These examinations will provide insights into broader ecological change associated with disturbance (Andersen 1995; Andersen & Majer 2004). We specifically ask three questions. Firstly, how does ant diversity and composition compare across the dominant land-uses: agriculture, pastoralism, silviculture and conservation? Secondly, how effective are different management practices in restoring ant biodiversity in lands previously used for production? Thirdly, what vegetation structural attributes are the best predictors of variation in ant communities across all management regimes?

Methods

Study sites

The study area is located in the state of São Paulo, which has the longest history of conversion of Cerrado to pastures or agriculture in Brazil. In just 30 years, the Cerrado has been reduced to <7% of its original area in the state of São Paulo (Durigan et al. 2007). The study was conducted in and around the Assis State Forest and Assis Ecological Station, between 22°33′36″ to 22°37′09″S latitude and 50°25′46″ to 50°21′56″W longitude, which has the southernmost occurrence of Cerrado.

The study area has an average altitude of 550 m above sea level and experiences a Cfa humid subtropical climate, according to Köppen's classification (Alvares et al. 2013). The average annual rainfall is about 1400 mm, concentrated in the austral summer, and severe frosts sporadically occur (Brando & Durigan 2005).

Locally, there is a gradient of native vegetation from “open” “cerrado sensu stricto” (typical savanna) to “closed” “cerradão”, and recent studies have shown extensive succession from open to closed cerrado due to decades of fire suppression (Pinheiro & Durigan 2009; Durigan & Ratter 2016). There is a toposequence of three main soil types from high to low elevation: red oxisol, yellow oxisol and yellow-red oxisol (Juhász et al. 2006). For this study, all sites were located in areas with Rhodic Haplustox soil. As is the case for Cerrado soils generally, the soils in the study site are typically deep and well-drained, have low fertility, high aluminium content and low soil water-holding capacity (Juhász et al. 2006).

We surveyed vegetation attributes and ant communities in 65 plots of 100 m² (20 m × 5 m) (Fig. 1), comprising five replicates of each of 13 land-use and restoration treatments that represented the most common land uses and restoration practices in São Paulo State, including two cerrado vegetation types as reference ecosystems. We first mapped all land-use and restoration treatments in the study area, and distributed replicates of each as widely as possible across the landscape. Although all treatments were interspersed with other treatments, we acknowledge that the five replicates of some treatments occurred in a single patch, not being ideally independent (Fig. 1).

The treatments represented four classes of ongoing land-use comprising cropping, pastoralism, silviculture, plus two physiognomies of native vegetation as conservation reference sites, and three classes of restoration, comprising passive, active mixed plantings and active monoculture, as follows:

1. Sugarcane (SC): sugarcane (Saccharum officinarum) grown for at least 10 years on land previously used as pasture. Soil has been deeply ploughed and fertilized (nutrients and lime) every 5 years, and insecticides and herbicides regularly applied.
2. Pasture with cattle (PC): grazed by cattle (1.5–2.0 head per ha from 2008, reduced to 0.5–1.0 head per ha hereafter) after two decades of silviculture.
3. Plantation Eucalyptus (PlE): 6-year old Eucalyptus hybrid plantation (E. urophylla × E. grandis) with 3 × 2 m spacing, without ploughing the soil, and without thinning (previously Pinus plantation for several decades). Herbicides used for the first 2 years. Insecticides (for leaf cutting ants) applied when necessary.
4. Plantation Pinus young (PlPY): 5-year old plantation of Pinus caribaea hondurensis with 3 × 3 m spacing, without ploughing the soil, and without thinning (previously Pinus plantation for several decades). Herbicides used for the first 2 years. Insecticides (for leaf cutting ants) applied when necessary.
5. Plantation Pinus old (PlPO): low density (200 trees per hectare) plantations 25–48 year old, featuring dense regeneration of native species in the understory. Herbicides never applied, insecticides (for leaf cutting ants) applied when necessary.
6. Agriculture abandoned (AA): passive restoration, abandoned for 2 years after 15 years of intensive agriculture (5 years of sugarcane and 10 years of grain crops, including genetically modified soya bean, Glycine max Roundup-ready). Cultivation practices include ploughing the soil, fertilizers, lime, insecticides, herbicides.
7. Pasture abandoned (PA): passive restoration; same as PC, but fenced for 7 years before survey to exclude cattle.
8. Restoration 2 (Rest2): 2-year old mixed stands of native species planted in an area cultivated for more than four decades with Pinus spp. Planted seedlings was spaced at 3 × 2 m, with fertilizers and without ploughing, and included about 30 cerrado species native to the region. Management practices included application of herbicides and insecticides for controlling invasive grasses and leaf-cutting ants, respectively.
9. Restoration 10 (Rest10): As in Rest2, but 10-year old. Fertilizers used before planting, herbicides and insecticides applied for the first 2 years.
10. Restoration monoculture, Anadenanthera falcata (RA) (18-year old; 3 × 3 m spacing): planted stands with a single native species previously used as pasture for several decades. No insecticides, fertilizers or soil ploughing.
11. Restoration monoculture, Inga laurina (RI) (14–16 year-old; 3 × 3 m spacing): planted stands with a single native species. Two plots had been planted immediately after clearing, and three plots had first been used as pasture for several decades. No insecticides, fertilizers or soil ploughing.
12. Cerrado sensu stricto (CSS): savanna physiognomy, mean basal area of 7.8 m² ha⁻¹, 48% canopy cover, average height of the tallest trees 6.6 m, and mean density of 897 trees per hectare with DBH ≥ 5 cm (Pinheiro & Durigan 2012).
13. Cerradão (C): forest physiognomy where trees form a continuous layer and grasses are sparse or absent. This vegetation has an average basal area of 21.4 m² ha⁻¹, 86.5% canopy cover, canopy height around 16 m and mean density of 1779 trees per hectare with DBH ≥ 5 cm (Pinheiro & Durigan 2012).

Results

Ant communities

Overall, 14 035 individuals from 131 ant species and 36 genera were collected (Appendix S2). The most diverse genus was Pheidole with 25 species, followed by Trachymyrmex (11), Camponotus (10), Hypoponera (8) and Solenopsis (8). The most common species were Pheidole sp. 14 (27.8% of total ant abundance), Wasmannia auropunctata (9.7%), Dorymyrmex sp. 2 (9.5%), Atta sp. 1 (7.6%) and Pheidole diligens (5%). The most common functional group was Generalized Myrmicinae, which represented 42% of all ants, followed by TCS (22%), Dominant Dolichoderinae (13%) and opportunists (11%). Hot-climate specialists were represented by just two individuals from one species, and so this group was not included in functional group analyses.

There was significant variation among land-use and restoration treatments in the number of individuals (GLM, Deviance_(N = 65) = 3281; P < 0.001), observed species richness (GLM, Deviance_(N = 65) = 95.9; P < 0.001) and estimated species richness (GLM, Deviance_(N = 65) = 129.5; P < 0.001) (Fig. 2). Several land-use and restoration treatments had comparable plot richness to that of the native reference habitats.

The GLMs indicated that, across all land-use and restoration treatments, ant abundance was best explained by a combination of grass cover, litter biomass and plant density (DBH < 5 cm), and ant species richness was best explained by the combination of basal area, grass cover, herbaceous plant richness and woody plant richness (Table 1; Fig. 3).

Table 1. The best fitted models (Generalized Linear Models – GLMs) and respective values of AIC and pseudo R² representing the relationships between observed species richness (S_obs), estimated species richness (S_est), number of individuals and number of individuals in each functional group in response to environmental variables

Response variable	Variables included in the best fitted model	GLM family	AIC	Pseudo R²
Observed richness	Herbaceous plant richness Woody plant richness Basal area Grass cover	Poisson	416.1	0.32
Estimated richness	Herbaceous plant richness Woody plant richness Basal area Grass cover	Poisson	485	0.31
Number of individuals	Plant density (DBH <5 cm), litter biomass, grass cover	Negative binomial	793.74	0.25
Cryptic Species	Herbaceous plant richness Basal area Grass cover	Negative binomial	469.5	0.16
Dominant Dolichoderinae	Herbaceous plant richness Wood plant richness Basal area Grass cover	Negative binomial	467.4	0.43
Specialist Predators	Herbaceous plant richness Woody plant richness Plant density (DBH < 5 cm) Basal area	Negative binomial	379.8	0.29
Tropical-Climate Specialists	Plant density (DBH <5 cm) Grass cover	Negative binomial	625.7	0.16
Opportunists	Plant density (DBH <5 cm) Litter biomass	Negative binomial	528.5	0.24
Subordinate Camponotini	Woody plant richness Basal area Grass cover	Negative binomial	305.1	0.21
Generalized Myrmicinae	Nonsignificant	Negative binomial	713.3	0

Ant species composition varied significantly among land-use and restoration treatments (Permanova, F_(12,65) = 3.41; P = 0.0001) (Appendix S2). There was a general gradient in habitat openness along the primary NMDS axis, in the sequence sugarcane, pastures, cerrado sensu stricto, plantations and cerradão. Permanova pair-wise tests showed significant differences in species composition between all pairs of land-use and restoration treatments, except for: pasture with cattle and cerrado sensu stricto (F_(12,65) = 1.24; P = 0.053), Eucalyptus and Pinus plantations (F_(12,65) = 1.18; P = 0.19), old Pinus plantation and cerrado sensu stricto (F_(12,65) = 1.16; P = 0.096), and old Pinus plantation and cerradão (F_(12,65) = 1.16; P = 0.065) (Appendix S3). Ant functional group composition also varied significantly among land-use and restoration treatments (Permanova: F_(12,65) = 5.12, P = 0.0001). There was a similar gradient in habitat openness, except that cerradão was not differentiated from plantations (Appendices S2 and S4).

Consistent with Permanova results, CAP showed very strong separation of land-use and restoration treatments across both axes based on the composition of both species (CAP1 = 0.97 and CAP2 = 0.94) and functional groups (CAP1 = 0.85 and CAP2 = 0.78). In both cases there was a strong separation between the very open sites (agriculture abandoned, sugarcane and restoration 2) and other sites, with significant associations with basal area, woody plant richness, plant density (DBH ≥ 5 cm) and litter biomass (Fig. 4a,c). A range of species (Fig. 4b) and functional groups (Fig. 4d), particularly Dorymyrmex sp. 2, Atta sp. 1 and Dominant Dolichoderinae, were strongly associated with open sites. In contrast, Cryptic Species, Generalized Myrmicinae and Specialist Predators were most abundant at closed sites, all positively associated with basal area (Fig. 4d; Table 1). The abundance of TCS was positively correlated with plant density, whereas this relationship was negative for the abundance of opportunists (Fig. 4d; Table 1). The abundance of Subordinate Camponotini was positively associated with the richness of woody plants, and negatively with plant density (Table 1).

Nine species were recorded only at either cerrado or cerradão plots. Almost all other land-use and restoration treatments also had species occurring nowhere else, and all these species were likewise rare, mostly occurring in a single plot. However, these were all very rare (most were recorded from a single plot), and so their “exclusive” occurrences cannot be attributed to habitat preferences.

Discussion

Our study used ants to assess the effects of different land uses and restoration practices on Cerrado biodiversity, and to examine the extent to which diversity and composition across the full land management spectrum can be explained by simple vegetation attributes. We firstly asked how ant diversity and composition compares across the four dominant land-uses: agriculture, pastoralism, silviculture and conservation. Mean ant abundance differed among treatments, but did not follow any pattern among land uses, which is consistent with previous studies showing that ant abundance is a poor indicator of ant community change (Andersen & Majer 2004; Hoffmann 2010). Mean observed and estimated richness also differed among treatments, and showed far more systematic variation in relation to treatment. Estimated richness was lowest in sugarcane and highest in cerrado sensu stricto, and in older restoration treatments and abandoned pasture was similar to that in natural reference sites.

Both the taxonomic and functional composition of ant communities were more sensitive to variation in land use compared with either total abundance or richness. This is consistent with previous studies of ant community responses to land-use change (Hoffmann & Andersen 2003; Andersen & Majer 2004; Hoffmann 2010; Solar et al. 2016), as well as of succession gradients in ecosystems undergoing restoration (Andersen et al. 2003; Ottonetti et al. 2006; Lawes et al. 2017), and of habitat change more generally in the Cerrado biome (Pacheco & Vasconcelos 2012). In contrast with other land uses, pastures with cattle and old pine plantations had similar species composition to that occurring in native cerrado. These two land uses were not subject to ploughing or fertilization, had limited use of herbicides and insecticides, and supported a considerable number of Cerrado plant species. The vegetation of cattle pasture – with sparse trees and a continuous grass layer – is structurally similar to cerrado sensu stricto. In the old pine stands, thinning operations reduced tree density by nearly 90%, making light availability intermediate between that in cerrado sensu stricto and cerradão (Abreu et al. 2011).

Different species had clear habitat preferences, especially in relation to vegetation cover. For example, Pachycondyla striata and Linepithema sp.1 were most abundant in structurally complex habitats, whereas Dorymyrmex sp. 2 and Atta sp. 1 strongly preferred very open habitats. Dorymyrmex is a highly thermophilic genus that is characteristic of open habitats throughout its range (Andersen 1997), and species of Atta are widely known to respond positively to disturbance, often becoming agricultural pests (Leal et al. 2014). Such contrasting responses of specialist forest versus open-habitat ant species is typical of forest management throughout the world (Andersen et al. 2009; Arnan et al. 2009; Dolek et al. 2009; Solar et al. 2016).

Functional group composition showed predictable patterns of variation. The highly specialized functional groups Specialist Predators, Cryptic Species and Tropical-Climate Specialists were all associated with high tree cover. These groups are dominated by taxa with forest origins (e.g. Odontomachus, Pachycondyla, Hypoponera, Pseudomyrmex, Trachymyrmex) and are known to be especially sensitive to disturbance (Andersen 2000; Leal et al. 2012; Solar et al. 2016). In contrast, Dominant Dolichoderinae and Opportunists were associated with the more highly disturbed sites, as occurs globally (Andersen 2000; Hoffmann & Andersen 2003). Dominant Dolichoderinae comprises highly thermophilic species that strongly prefer open habitats, and Opportunists often proliferate at disturbed sites because of their wide habitat tolerances (Andersen 1995; Hoffmann & Andersen 2003).

Our second question asked how effective different management practices are in restoring ant biodiversity in lands previously used for production. Ant species richness at actively restored sites was similar to that at reference sites, but species composition was markedly different, especially when compared with cerrado sensu stricto. This reflects differences in vegetation structure and composition between cerrado sensu stricto and restoration sites, where trees have been planted to produce a forest rather than savanna structure. These results are consistent with previous studies showing that although restoration interventions can improve degraded conditions, they have limited success in restoring historical ecosystems (Andersen et al. 2003; Rey-Benayas et al. 2009; Suding 2011). We found limited success with restoring functional as well as species composition of ants. It is therefore likely that the ecosystem services provided by ants, such as seed dispersal and predation, have similarly not been fully restored (Bihn et al. 2010).

Finally, we asked what vegetation structural attributes are the best predictors of variation in ant communities across all management regimes. The strongest predictors of ant species richness were tree basal area, grass cover, and the richness of both woody and herbaceous plants. These were all also key correlates of variation in ant species and functional composition. Tree basal area and grass cover are key measures of habitat structure, and drive important abiotic conditions such as the thermal environment, as well as the availability of nest sites and foraging environments for arboreal ants (Fagundes et al. 2015). Trees can also promote the diversity of ground-nesting species in Cerrado, even up to 30 m away (Frizzo & Vasconcelos 2013). It is likely that higher plant species promotes ant richness through the provision of a greater diversity of nest sites and food resources. There is a well-documented positive relationship between richness of arboreal ants and tree richness in the Cerrado (Ribas et al. 2003). Changes in tree and grass cover mediated by land management can also have important indirect effects on ant communities through changes in the competitive influence of dominant species (Gibb 2011; Moranz et al. 2013).

Over recent decades Brazilian Cerrado has undergone dramatic transformation due to agricultural and silvicultural development. Our results reinforce the notion that highest biodiversity loss occurs when land-use systems involve transformation of vegetation structure. Cattle grazing, a land use that retains the general structure of savanna, maintains ant-species composition similar to native cerrado, as observed in previous studies (Frizzo & Vasconcelos 2013; Carvalho 2014). Conservation policies based on fire suppression in remaining native habitat have caused widespread woody encroachment in the Cerrado, resulting in the homogenization of the structural mosaic of the Cerrado and loss of the exceptionally diverse savanna flora (Pinheiro & Durigan 2009; Durigan & Ratter 2016; Abreu et al. 2017; Stevens et al. 2017). The policy of fire suppression is based on a complete mis-understanding of tropical grassy biomes, which require frequent fire to maintain their high levels of diversity (Parr et al. 2014). Our results show that the forest structure of the cerradão similarly does not support open-habitat ant species, and so the woody encroachment following fire exclusion leads to loss of Cerrado ant diversity, as demonstrated by Abreu et al. (2017).

Brazil's legislative requirement that at least 20% of each rural property must be set aside to preserve or to restore the native vegetation can only be met in the Cerrado biome through extensive restoration. Current restoration techniques have focussed on planting trees in high density even in savanna regions, failing to restore native grasses, forbs and shrubs (Pilon & Durigan 2013; Correa et al. 2015). It is therefore unsurprising that the species composition of Cerrado ant communities is not restored. A restoration philosophy focussed on high tree cover is clearly inappropriate for Cerrado.

Our finding of major compositional differences among land-use types has important implications for the ant diversity of agricultural landscapes, and therefore for the challenge of developing production systems that reduce impacts on biodiversity and ecosystem services (Foley et al. 2005). Compared to monocultures, mosaics of different agricultural uses markedly enhance biodiversity at the landscape scale (Bennett et al. 2006). Recognizing the emergent properties of landscape mosaics and the negative consequences of agricultural intensification and homogenization at large scales is crucial for conservation planning (Vrdoljak & Samways 2014; Baiamonte et al. 2015). Policy frameworks and management solutions are needed for preventing and reversing intensive homogenization in order to sustain biodiversity in agricultural systems (Benton et al. 2003; Bennett et al. 2006).