Introduction

Unraveling the drivers of local species diversity is urgent and essential. Species detected on a regional scale might be absent in specific localities of such region for several reasons (Kraft et al. 2015). For instance, they could be unable to colonize a given locality due to dispersal limitations (Padial et al. 2014) or because of environmental filtering, when the abiotic conditions inhibit their persistence (Gibb et al. 2015). Even if they are able to colonize a locality, their persistence may be hindered by biotic filtering (Ellwood et al. 2016), when predators, lack of relevant interactions, and better competitors may exclude them. These factors’ interactions determine which of the regions’ species constitute local communities (Gray et al. 2018), with their importance varying in different spatial and temporal scales (Kaspari et al. 2003; da Silva and Hernández 2015). Thus, environmental, spatial, and temporal processes are simultaneously responsible for the local assembly mechanisms and community dynamics (Leibold et al. 2004; Mouquet et al. 2005), although they can have different roles in community assembly depending on the species groups and ecological systems (Kaspari et al. 2003; Graham and Fine 2008; da Silva and Hernández 2014). The effects of these processes across spatiotemporal scales have properly been underlined using the metacommunity approach (da Silva et al. 2021). A metacommunity is characterized by the connection of local communities via dispersal of multiple potentially interacting species (Leibold et al. 2004). Therefore, these multiple perspectives of processes on community patterns can shed light into the main mechanisms driving species distributions across an increasingly patchy and heterogeneous world (Mouquet et al. 2005; da Silva and Hernández 2015).

Naturally fragmented landscapes are adequate systems for evaluating species distribution patterns (Cuissi et al. 2015; Rossetti et al. 2017). However, the majority of these landscapes are oceanic archipelagos. Therefore, the theories describing island diversity were formulated with marine archipelagos as a model. A major example is the theory of island biogeography (TBI) that predicts a positive species-area relationship and a negative species-isolation relationship (MacArthur and Wilson 1967), which is now part of the theoretical framework of the metacommunity approach (Leibold et al. 2004). Since TBI’s elaboration, these predictions have also been applied to anthropogenically fragmented landscapes, where isolated patches are formed due to habitat loss (Leal et al. 2012; Rossetti et al. 2017). Various organisms in terrestrial archipelagos have shown similar patterns to those predicted by TBI (Adams et al. 2017; Palmeirim et al. 2018), but some do not follow these predictions (Haila 2002) due to the differences between the effects of habitat loss and fragmentation per se (Fahrig 2017). This leaves an unanswered and fundamental question of island biogeography: “How the patterns and processes that generate biodiversity behave when comparing marine and terrestrial archipelagos systems?” (Patiño et al. 2017). In order to fully answer this question, it is imperative to gather more information on biodiversity responses in naturally formed terrestrial archipelagos, where the landscape matrix may be permeable for only a few species, generating a metacommunity structure (Leibold et al. 2004). Besides, naturally fragmented landscapes are excellent systems for evaluating patterns and mechanisms that determine the distribution of species in environments without confounding effects of human-driven fragmentation and habitat loss (Fahrig 2017; Haddad et al. 2017).

Although crucial for the maintenance of species diversity, small and isolated patches are often neglected in conservations practices (Wintle et al. 2019). One example of those systems is the forest archipelagos immersed in grassland-dominated ecosystems on tropical mountaintops (Streher et al. 2017). In Brazil, it is common to find such systems across the mountaintops of the Espinhaço Range Biosphere Reserve (hereafter treated as forest islands) (Coelho et al. 2018b). The Espinhaço mountain range extends over 1,200 km from southeast to northeast Brazil and harbors high biodiversity and endemism (Silveira et al. 2016; Fernandes et al. 2020). The forest islands are surrounded by a campo rupestre matrix (montane grasslands), which creates a complex vegetational mosaic in the landscape, giving rise to a typical forest archipelago (Coelho et al. 2016; Itescu 2018). In this scenario, the composition and structure of the landscape (i.e., the habitat types and their configuration) can influence the colonization and establishment of species differently (i.e., a metacommunity dynamic) since different habitats provide distinct resources and conditions (Fahrig et al. 2011). Furthermore, temperature and rainfall vary throughout the year, with a marked influence on forest conditions. There is an average difference of almost 5°C and 140 mm/month between the summer and winter seasons in this forest archipelago (da Silva et al. 2019). Despite its relevance to ecology and conservation, the mountaintop forest islands have increasingly suffered from anthropogenic disturbances, such as illegal fires, grazing, and selective logging (Coelho et al. 2018b). The distribution of these forest islands is also expected to be affected by ongoing climate change, which is predicted to be more intense in mountainous areas (Mayor et al. 2017; Rumpf et al. 2018). Therefore, there is an urgency in defining clear strategies to conserve these mountaintops ecosystems (Noroozi et al. 2018) by establishing priorities and guidelines that can support informed decision-making policy towards sustainable use of the campo rupestre and its forest islands (Coelho et al. 2018b; Fernandes et al. 2020; Neves et al. 2021a).

In this tropical forest archipelago system, vagile organisms like wasps, bees, dung beetles, and fruit-feeding butterflies can benefit from a heterogeneous landscape by seeking different resources and conditions in the various landscape habitats. For instance, wasps and bees use the floral resources of the matrix while nesting in the forest islands (Perillo et al. 2020). Dung beetles and fruit-feeding butterflies also seem to use the favorable conditions of the forest islands during winter, taking shelter from the less favorable conditions of the matrix (Pereira et al. 2017; da Silva et al. 2019). On the other hand, for less vagile organisms, such as ants, the habitat diversity and the landscape configuration influence the sources of colonizing propagules. Once present in a locality, ants have a greater dependence on local resources and conditions, with little or no use of the other landscape features (Spiesman and Cumming 2008; Andersen 2019). Therefore, one might expect different responses between vagile and sessile organisms inhabiting this forest archipelago (Neves et al. 2021a).

Due to their dependence on microclimate and local resources (Andersen et al. 2002) and considering their extreme diversity and distribution (Levy 1995; Dunn et al. 2007; Parr et al. 2017), ants have been used as good model organisms for understanding the effects of habitat loss and fragmentation (Vasconcelos et al. 2006; Cuissi et al. 2015; Ahuatzin et al. 2019). Moreover, ants are sensitive to environmental changes, especially changes in vegetation structure (Andersen 2019; Corro et al. 2019) and forest cover (Ahuatzin et al. 2019). Therefore, forest islands should harbor ant species that exhibit narrow environmental tolerances, constraining them to a forested environment (forest-dependent species) and species with broader environmental tolerances, which thrive in a higher diversity of habitats (habitat generalist species). In order to find the most important processes for both species’ groups (forest-dependent and habitat generalists), we can deconstruct the metacommunity according to these tolerances (Pandit et al. 2009). Doing so, we avoid the interference of one group’s ecological responses on another’s. For instance, the habitat generalist species mask the ecological response of the forest-dependent dung beetle species to both patch and landscape features (da Silva et al. 2019). Nevertheless, we still do not know if this is the case for sessile organisms in the forest archipelago.

Here, we aimed to evaluate the spatial and temporal patterns of ant metacommunity’s structure and dynamics according to their habitat tolerance (forest-dependent and habitat generalist species) over the local (forest island) and landscape (forest archipelago) features of a montane forest archipelago. In order to do so, we have tested the following predictions: (i) the variation in species composition (β-diversity) is higher spatially than temporally due to the low mobility and longevity of ants for both groups (forest-dependent and habitat generalist species); (ii) the forest islands’ vegetation structure and landscape structure determine the ant richness and temporal species composition variation (temporal β-diversity), since larger, denser, and less isolated forest islands are richer and more temporally stable; and (iii) the effect of forest island and landscape structure differs for each species group, since forest-dependent species are more sensitive to variations in local vegetation structure (such as canopy cover and understory density) and landscape structure (such as habitat amount and isolation) than habitat generalist species.

Material and methods

Study area

This study was carried out in the forest islands from a naturally formed montane forest archipelago at the Serra do Cipó (see Coelho et al. 2018b) located in the southern portion of the Espinhaço Range Biosphere Reserve, Brazil (19°14′19″S, 43°31′35″W; Fig. 1). The Espinhaço is an important mountain range in terms of biodiversity and water resources in Brazil (Fernandes 2016). It extends over 1200 km from southern Minas Gerais to the center of Bahia States, and it is amidst three great Brazilian biomes, the Caatinga (semi-arid) to its north, the Cerrado (Brazilian savanna) to its west, and the Atlantic Forest to its east; these last two are considered hotspots of biodiversity conservation (Myers et al. 2000).

Fig. 1
figure 1

Map showing the 14 naturally fragmented forest islands (green shapes) in a campo rupestre matrix of the Serra do Cipó, southern portion of the Espinhaço Range Biosphere Reserve, Minas Gerais, Brazil, where we sampled ants

Full size image

Associated with these mountaintops occurs the campo rupestre, an ancient, climatically buffered, and diverse ecosystem that is considered the hottest Brazilian biodiversity hotspot and hosts various endangered and endemic species (Fernandes et al. 2020; Hopper et al. 2021). This ecosystem has a high edaphic and vegetational heterogeneity (Le Stradic et al. 2015; Ferrari et al. 2016) and enables the occurrence of relictual forest islands associated with specific edaphic-climatic characteristics, such as erosion valleys that lack boulders (Coelho et al. 2016). In addition, the occurrence of forest islands in the Espinhaço mountain range starts at 1200 m a.s.l. and is also frequently associated with headwaters (Coelho et al. 2018b). These forest islands’ floristic resembles semi-deciduous Atlantic Forests (Coelho et al. 2018a), and the most common plant families are Myrtaceae, Lauraceae, Melastomataceae, and Fabaceae.

The regional climate is Cwb (dry-winter subtropical highland climate) according to Köppen’s classification (Alvares et al. 2013). In the summer, the temperature was 19.22°C and humidity 88.62% (mean values of December, January, and February 2014 and 2015), while in the winter, they were 14.68°C and 92.58%, respectively (June, July, and August 2014 and 2015 mean values). In the Espinhaço range’s high elevations, the relative humidity is always high due to the nebular condensation of humid air (Coelho et al. 2016), but the temperature varies seasonally, with a mean difference of 4.54°C between seasons (data from a weather station installed at 1400 m a.s.l., near our study site: Onset HOBO® U30 data logger; Long-Term Ecological Research Program Project – PELD Campos Rupestres da Serra do Cipó; see Silveira et al. (2019)).

Sampling design

We selected 14 forest islands, varying in dimension, shape, and distances from each other and from a continuous forest (Table S1) to access ant diversity. All of them occur in the same elevation belt (ca. 1300 m a.s.l.). In the nuclear region of each forest island, we installed a permanent plot of 1000 m2 (20 × 50 m). In each plot, we installed five soil pitfall traps (9 cm deep × 15 cm diameter) (Bestelmeyer et al. 2000) and five identical arboreal pitfall traps (Ribas et al. 2003), which were left in the field for 48 h per sampling campaign. The soil traps were installed at the vertices and at the center of each plot (20 × 50 m), while arboreal traps were installed 1.5 m above the ground in the nearest tree from each soil trap. Pitfall trapping is the best method to capture ants given our aim and the conditions of our study area (Montgomery et al. 2021). We sampled these plots four times: two times in the summer season (February) and two times in the winter season (August) in 2014 and 2015.

The sampled material was taken to the Laboratório de Ecologia de Insetos (Insect Ecology Lab) at the Universidade Federal de Minas Gerais (LEI-UFMG) for processing and identification. We followed Baccaro et al. (2015) for genera identification and used LEI-UFMG Ant Collection to confirm species identification. The LEI-UFMG is an important collection of campo rupestre’s ant species that has been continuously reviewed by various ant specialists. We used a qualitative approach to separate the species into groups according to their habitat occurrence. Based on published information (Costa et al. 2015; Castro et al. 2020; Nunes et al. 2020; Neves et al. 2021a), our own field experience, and by comparing the ant species sampled in forest islands with the pre-existing ant species collection from the LEI-UFMG, we classified each species as “forest-dependent species” (species only found in forest islands) and “habitat generalist species” (species found in both forest islands and campo rupestre in local studies). A similar approach was used to classify ant species of Cerrado (Vasconcelos et al. 2018).

Forest island and forest archipelago structure

We measured forest island size and distances (from each other and from the continuous forest) using the latest satellite images of Google Earth Pro v.7.3.2 software. Each forest island was manually delimited, and its area was measured in km2. The distances from each forest island to its neighbors and from the continuous forest were measured in a straight line in km. To calculate the canopy cover and understory structure, we took four canopy hemispheric photos and 16 understory photos within each forest island in each sampling event. The canopy photos were taken from 1.5 m above ground using a fish eye’s lens attached to a Canon T5i camera (Nassar et al. 2008). These images were processed using the Gap Light Analyzer software (Frazer et al. 1999), which calculates the number of white pixels (i.e., open spaces) in each image. In order to determine the understory density, we used a method recommended by Nobis and Hunziker (2005), which consists of analyzing digital images of the local shrub and herbaceous vegetation. Understory images were taken using a 1 m2 white sheet as background, positioning the camera 3 m far from it and 1 m above the ground. Therefore, we measured the following variables as local predictive variables: (i) forest island size, (ii) distance to the continuous forest, (iii) distance to the nearest forest island, (iv) canopy cover, and (v) understory density (see also Pereira et al. 2017; Perillo et al. 2020).

To classify the landscape structure of the forest archipelago, we used images from a high-resolution multispectral satellite (RapidEye ~5m resolution) and classified these images using the R package randomForest v4.6-12. Then, we selected 10 landscape metrics from each one of the 14 forest islands, using data from buffers with a radius of 250 m considering the sampled plot center as the centroid areas (Table S2) (see also da Silva et al. 2019; Perillo et al. 2020).

Statistical analysis

We performed all the analyses using data of forest-dependent and habitat generalist species separately in the R software (R Core Team 2020). We calculated the sampling coverage as a measure of sample completeness for each forest island using the R package iNEXT (Chao and Jost 2012; Hsieh et al. 2016) using two times the reference sample size for the extrapolation curves. We calculated the accumulated richness of all sampling seasons for each forest island (n = 14) and the richness of each sampling (four events in each forest island; n = 56). Furthermore, we calculated the relative frequency of occurrence for each species as a specific indicator of the spatiotemporal distribution using the percentage of sampling unities where they are present.

To determine the variation of the ant species composition between forest islands (spatially) and between seasons (temporally), we calculated the β-diversity for forest-dependent and habitat generalist species based on the Sorensen dissimilarity coefficient using the R package betapart (Baselga and Orme 2012). We chose the Sorensen dissimilarity coefficient because it gives more importance to shared species in a sample (Baselga 2012) and indicates the proportion of shared species between sites (Anderson et al. 2011). In our case, as the study area does not occupy a vast region (less than 25 km2), the occurrence of spatiotemporally shared species is expected. The spatial β-diversity represents the composition dissimilarity among the 14 forest islands, while the temporal β-diversity represents the dissimilarity in each forest island through time (the four sampling periods). We decomposed the total β-diversity (Sorensen dissimilarity) into species replacement (or turnover) and nestedness-resultant components for forest-dependent and habitat generalist species to verify which is the main process driving spatiotemporal patterns (Baselga 2010; Baselga 2012). Both components were calculated accounting for multiple sites (spatial) or sampling periods (temporal) and accounting for pairwise dissimilarities, using the function beta.multi of the R package betapart (Baselga and Orme 2012). This function calculates the overall spatial and temporal β-diversity and its turnover and nestedness-resultant components.

Finally, to verify the effects of the vegetation structure and landscape structure on species richness (pooled values for all samplings) and temporal variation of species composition (temporal β-diversity) of forest island ants, we first tested the correlation between the forest island structure variables (local scale) and between forest archipelago structure variables (landscape scale). Those variables with ≥ 50% correlation (or p-value < 0.05) were removed from the analyses (see correlated variables in Figure S1). After that, we used the remaining six non-correlated metrics (local variables: (i) canopy cover, (ii) distance to a continuous forest; landscape variables: (iii) percentage of the forest island in the landscape, (iv) number of forest islands in the landscape, (v) mean distance to forest island neighbors, (vi) forest island’s total edge) as explanatory variables in generalized linear models (GLMs). We used species richness and temporal β-diversity of forest-dependent and habitat generalist species as response variables. We also simplified them by removing the non-significant variables one by one considering the p-value. Hence, we reached the most adequate model, in which we performed a residual analysis to verify its error distribution (Crawley 2013). The most adequate distributions were Poisson for richness models and Binomial for β-diversity models, adjusting both for overdispersion (with their quasi families).

Results

We sampled 99 ant species distributed in 40 genera and nine subfamilies across all forest islands (Table S3). Among the sampled species, 18 species were singletons (18%), 15 doubletons (15%), and seven tripletons (7%), totaling 40 rare species (40%). Only eight species (8.1%) occurred in all forest islands. From the total number of species, 65.7% were classified as forest-dependent (N = 65) and 34.3% as habitat generalist species (N = 34). The most frequent species were Pachycondyla striata Smith (100% frequency in the forest islands), followed by Gnamptogenys striatula Mayr (89%) and Pheidole jelskii Mayr (82%), all classified as habitat generalists. The most frequent forest-dependent species were Camponotus lespesii Forel (76%) and two morphospecies of Pheidole Westwood: Pheidole sp.13 (57%) and Pheidole sp.2 (55%) (Table S3). We obtained between 78.3 and 94.6% of sample completeness (average ± standard deviation = 88.8 ± 5.1%; Table S3).

Variation of the ant species composition (β-diversity)

We found that the spatial variation (spatial β-diversity) in ant species composition was greater than the temporal variation (temporal β-diversity) regardless of the species group evaluated (1.8 and 2.1 times higher for forest-dependent and habitat generalist species, respectively) (Fig. 2). Species replacement contributed more to spatial β-diversity than nestedness for both forest-dependent and habitat generalist species (95.0% and 92.3%, respectively). For temporal β-diversity, we found an increase of the nestedness-resultant component for both forest-dependent and habitat generalist species (41.6% and 50.5%, respectively). In all cases, the spatiotemporal β-diversity of habitat generalist species was lower when compared to β-diversity of forest-dependent species (Fig. 2).

Fig. 2
figure 2

Partitioning of β-diversity (Sorensen dissimilarity) of forest-dependent (FD) and habitat generalist (HG) ant’s species into its components of replacement (βrep; darker bars) and nestedness (βnes; lighter bars) spatially (among forest islands) and temporally (among seasons per forest island) in forest islands of Serra do Cipó, southern portion of the Espinhaço Range Biosphere Reserve, Minas Gerais, Brazil

Full size image

Effect of environmental variables

We found that forest-dependent species richness was not affected by any of the local or landscape variables, while canopy cover and average distance to neighbor forest islands influenced habitat generalist species richness (Table 1, Fig. 3). Surprisingly, we found more habitat generalist species in forest islands with denser canopies and also in forest islands that are closer to their neighbors (Fig. 3a, b). In addition, we found that an increase in the amount of forest or in the number of forest islands in the landscape negatively affected the temporal composition variation of forest-dependent species (Table 1, Fig. 3c, d). Moreover, the number of islands in the landscape affected in the same way the habitat generalist species (Fig. 3d), thus determining an increase of temporal stability for both groups.

Table 1 Results of generalized linear models testing the effects of forest island and landscape structure variables on species richness and temporal β-diversity (Sorensen dissimilarity) of ants sampled in the Serra do Cipó, southern portion of the Espinhaço Range Biosphere Reserve, Minas Gerais, Brazil
Full size table
Fig. 3
figure 3

Relationship between species richness (upper panels) and temporal β-diversity (Sorensen dissimilarity; lower panels) of forest-dependent and habitat generalist ant species with canopy cover (local variable), mean distance to neighbor forest islands at 250 m buffer, habitat amount represented by forest island at 250 m buffer, and the number of forest islands at 250 m buffer (landscape variables)

Full size image

Discussion

We evaluated the effects of local and landscape attributes on spatial and temporal patterns of an ant metacommunity inhabiting the montane, naturally formed forest archipelago. According to our predictions, we highlight as main findings that (i) the species composition variation across forest islands (spatial β-diversity) is higher than the species composition variation across sampling periods (temporal β-diversity); (ii) the forest island vegetation structure and the landscape structure influence the ant richness of habitat generalist species only; (iii) only the landscape structure determines the temporal β-diversity for either group; and (iv) both forest island and landscape structure metrics affected ant species’ groups differently.

Most of the β-diversity of the ant metacommunity, both spatial and temporal, is explained by species replacement. Species replacement has been described as the primary process driving spatial differences in species compositions for most organisms across many spatial scales and ecosystems (Soininen et al. 2018). The mechanisms behind this phenomenon may be related to the low dispersal capacity and the high dependence on forest habitats of the ants sampled (65.7% of forest-dependent species), which would be directly related to landscape structure, especially distance to neighbor forest islands, determining their dispersal dynamics (Soininen et al. 2007). The contribution of the nestedness-resultant component was a little higher than the replacement component only for temporal β-diversity of habitat generalist species. This implies more losses than replacement of species in the ant assemblages found when the environmental conditions are harsher and the resources scarcer (winter season). This result contrasts with those found by Nunes et al. (2020) and Neves et al. (2021b), who have reported a high temporal replacement of ant species in the campo rupestre and in a tropical dry forest, respectively. Both systems are seasonal, so there is a high temporal variation in resource availability, which implies more replacement than loss (or nested loss) of ant species over time. Habitat generalist species are also those that contribute more to nestedness when compared with habitat specialists in the Espinhaço Range, showing a spatial dynamic related to the mass effects’ metacommunity process (Neves et al. 2021a). Ants are sessile organisms, so we are aware that either a variation in species activity or a sampling artifact can contribute to the values of species replacement found. Accumulating spatiotemporal replicated data via long-term studies will help unravel these issues, in addition to contributing to the advancement of metacommunity research (Record et al. 2021).

Although the species-area relationship is well documented for ants, especially in areas with anthropogenic fragmentation (Schoereder et al. 2004; Vasconcelos and Bruna 2012; Ohyama et al. 2021), species richness of some ants’ groups may not be associated with or dependent on forest fragment area (Lozano-Zambrano et al. 2009), as we also found here. For instance, Lozano-Zambrano et al. (2009) found distinct relationships between the richness of some ant species’ groups related to habitat requirements and fragment area of a tropical dry forest in Colombia. Therefore, the ant metacommunity inhabiting the forest archipelago studied, which is formed by a high diversity of species sampled by both soil and arboreal pitfall traps, may harbor species with different niche requirements that are not affected by forest island area per se in terms of species richness. Similarly, no relationship between species richness of bees and wasps (Perillo et al. 2020), fruit-feeding butterflies (Pereira et al. 2017), dung beetles (da Silva et al. 2019), and forest island area in the same montane forest archipelago has been found. This implies that for ants, the deconstruction of the metacommunity into intra-groups within the forest-dependent species group according to habitat requirements can be an appropriate approach to investigate such relationships. In addition, habitat quality can be more important than habitat amount to determine the species habitat use in naturally fragmented landscapes (Regolin et al. 2021).

As we have shown, forest island features and landscape structure metrics affected ant species’ groups differently. Surprisingly, the species richness of forest-dependent species was not affected by any forest island and landscape variables. Besides, the canopy cover positively affected the species richness of habitat generalist species, while the mean distance to neighbor forest islands negatively affected this group. Canopy cover varies seasonally and spatially among the forest islands sampled (Pereira et al. 2017; da Silva et al. 2019). However, the seasonal variation in canopy cover was insufficient to affect forest-dependent ant species in terms of species richness and compositional changes. Our results for habitat generalist species also contrast with those found by Ahuatzin et al. (2019). On the other hand, for dung beetles, da Silva et al. (2019) found a positive relationship between habitat generalist species’ abundance and richness and canopy cover only during the cold season. These authors suggested that these species may use the forest islands as refuges during times of harsher climatic conditions and scarcer resource availability in the campo rupestre open matrix. Although we do not address this pattern in this study, the accumulated richness of habitat generalist species over the cold and hot seasons sampled may suggest an even larger flow of or use by habitat generalist species inhabiting canopy-denser forest islands and consequently more climatically buffered habitats. Another possible explanation is that the preference of habitat generalist species by forest island conditions and resources can be a general pattern. Ant species richness is positively influenced by higher forest structural complexity in tropical forests, especially when soil to canopy strata is considered (Andersen 1986; Vasconcelos and Vilhena 2006; Neves et al. 2013). Habitat generalist species richness also decreased with increasing distance between neighbor forest islands. This implies a direct relationship between forest islands’ aggregation and habitat generalist species richness, which can generate a high spatial flow that decreases the possibility of habitat generalist species entry when forest islands are more isolated. This result also suggests a forest islands’ spatial distribution effect on the species richness of these ants: the closest the forest islands are on average, the higher the overall species richness is. In terms of metacommunity patterns, we can assume this result as a dispersal-driven effect, which implies a higher probability of a locality being influenced by closer sites (Leibold et al. 2004; Soininen et al. 2007), with important effects on the temporal changes or stability of the ant metacommunity.

The amount of forest island in the landscape, however, was important in determining the species composition temporal changes (β-diversity) in the assembly of forest-dependent ant species. In our study, the amount of forest island in the landscape was positively correlated with other similar variables such as forest island size, largest patch index, total core area, and area-weighted mean core area index distribution (see Figure S1; see Table S2 for variable description), but also positively correlated with forest islands’ understory density (Pearson’s rho = 0.67). Forests with low tree density (or understory density) can favor the occurrence of specialized opportunistic ant species, which can result in high species richness (Ahuatzin et al. 2019) but can also result in decreases in resource availability used by more sensitive forest-dependent species. Seasonal changes in resource availability and climatic conditions are known to influence ant foraging activity (Calazans et al. 2020) and consequently the diversity and composition of assemblages through time (Nunes et al. 2020). Therefore, forest-dependent ant species are more temporally stable in less isolated, bigger forest islands with a denser understory, which can indicate high resource availability (Philpott and Foster 2005; Klimes et al. 2012; Nunes et al. 2020).

The negative relationship between temporal β-diversity and the number of forest islands in the 250 m buffer was the only common response of both ant species’ groups. This result can be explained by the fact that more islands in the neighborhood mean less isolation, providing higher stability in local communities over time (lower temporal β-diversity) due to dispersal dynamics, since isolation inhibits ants’ dispersion (Tonhasca Jr et al. 2003; Leibold et al. 2004; Fahrig 2013). Changes in patterns of temporal β-diversity due to some kind of isolation (e.g., distance to closest forest island, distance to continuous forest) have been documented for bees and wasps (Perillo et al. 2020) and for both forest-dependent and habitat generalist dung-beetles (da Silva et al. 2019) in the same mountainous system. Habitat generalist species, especially considering vagile organisms such as butterflies and dung beetles, also benefit from a more forested landscape since the forest patches are more stable micro-climatically and buffer the severe climate in this mountaintop during the winter season (Pereira et al. 2017; da Silva et al. 2019). We now add new information on the effects of isolation on temporal patterns of ants, a less vagile group of organisms.

In this study, we showed the importance of a naturally formed forest island archipelago for the maintenance of the mountaintop’s ant diversity. Without these relictual forest islands, a large proportion of the ant fauna would not be able to establish and persist. In addition to being suitable as a refuge during adverse winter conditions for habitat generalist species (da Silva et al. 2019), most ants found in these forest islands are directly dependent on their resources and conditions. The responses of the ant metacommunity seem to be determined jointly by the landscape structure of the forest island archipelago (i.e., isolation) and local features (e.g., forest island area, canopy cover). The landscape structure influences the dispersal capacity of ants between forest islands, and local forest structure influences their establishment capacity, which in conjunction determine the local (α) and regional diversity (γ). In summary, the landscape structure acts as a spatial filter and the forest island structure as an environmental filter. We also advocate the importance of deconstructing communities into habitat specialists and habitat generalists (da Silva et al. 2019; Neves et al. 2021a), or other habitat-related requirements (Lozano-Zambrano et al. 2009). Besides, habitat specialists are expected to be more sensitive to forest habitat loss and land-use changes — the current main causes of entomofauna loss (Sánchez-Bayo and Wyckhuys 2019). Therefore, management plans of the mountaintop areas must also consider the forest island archipelago in order to maintain the biodiversity and functioning of these montane systems.