Our study was conducted in the Santa Bárbara Ecological Station (hereafter SBES; 22^°46′–22^°51′S/49^°10′–49^°16′ W, 600–680 m above sea level, Águas de Santa Bárbara municipality, São Paulo, Brazil), one of the few protected areas that preserves open savannas in the southern Brazilian Cerrado (Durigan et al., 2020). SBES is characterized by a mosaic of savannas and Atlantic Forest patches (Melo and Durigan, 2011), comprising 2,715 ha. Currently, SBES vegetation is mostly represented by native Cerrado formations, from grasslands (“campo sujo” and “campo cerrado”) and savannas (“cerrado sensu stricto”) to woodlands (“cerradão”) (Oliveira-Filho and Ratter, 2002; Araujo et al., 2010; Melo and Durigan, 2011). The SBES was under a fire suppression policy for 30 years (1985–2015; Abreu et al., 2017; Durigan et al., 2020). The grasslands that we studied here are among the few ones that had some accidental fire events more recently. The 2003 “campo limpo” (“campo sujo” in the 2017–2018 survey) patch had at least four fire events before 2008, when the last one was registered. One “campo cerrado” patch had only one fire event recorded in the last 30 years, in 2001, and the other patch in 2011, and both have been under fire suppression since then. The other plots that were sampled, burned for the last time before 1985 (Melo and Durigan, 2011; Conciani et al., 2021). Although these sparse fire events contributed to the grasslands maintenance in the area, SBES have been suffering under woody encroachment across this period (Melo and Durigan, 2011; Abreu et al., 2017, 2021). The study site also holds stands of exotic timber plantations, including Eucalyptus sp. and Pinus sp. Dry/cold (April to September) and wet/warm (October to March) seasons are strongly marked. The climate is classified as Köppen Cwa-type, with annual rainfall between 1010 and 2051 mm and an average of 1454 mm (Alvares et al., 2013; CIIAGRO, 2016). The mean temperature of the coldest months is 17°C and for the hottest months, 24°C, with a maximum of 35.2°C and a minimum of 3.4°C. These data correspond to the period from 1995–2014 and come from the weather station in the municipality of Manduri, São Paulo, Brazil, 20 km from our study area (CIIAGRO, 2016). SBES soils are characterized as deep oxisols with low nutrient and high sand content, high saturation of aluminum, and low capacity of holding water (Melo and Durigan, 2011).

We used different datasets for each analysis performed here, according to its scale and goals. The datasets are derived from the total data we collected in both surveys (2003 and 2017–2018) and are described in Table 1 and at each analysis item below.

Statistical analyzes were performed in R (version 4.1.0; R Core Team, 2021). The model selection for all linear models described below was made through hypothesis testing approach (ɑ = 0.05), comparing nested models through ANOVA function using the variance partition from the F-statistic for linear models and the residual deviance and chi-square tests for generalized linear models (Zuur et al., 2009). The diagnoses of the fitted models were made with the “DHARMa” package, we tested the model fitness for over and underdispersion, uniformity, outliers (along with Cook’s distance plot), and zero-inflation (Harting, 2021, Supplementary Table 1). The prediction graphics of the models were made with the package “ggiraphExtra,” function ggPredict (Lüdecke, 2018).

We captured 1,112 individuals of 17 native species of marsupials and rodents, in the gradient from grasslands to woodland savanna, considering both surveys (2003 and 2017–2018). Rodents and marsupials represented 67 and 33% of the total richness, respectively, (Table 2).

The 2003 survey recorded 10 species of small mammals (60% of total SBES richness), of which 70% were rodents and 30% marsupials, with a total of 124 individuals captured. In the 2017–2018 survey, 988 individuals were captured belonging to 17 species (100% of SBES total richness), of which 65% were rodents and 35% marsupials. Comparing both surveys, for 2003 we did not record the rodents Hylaeamys megacephalus, Nectomys cf. squamipes, Oxymycterus delator, and Rhipidomys cf. macrurus, and the marsupials Didelphis albiventris, Gracilinanus agilis, and Gracilinanus microtarsus in the habitats analyzed.

The species richness observed for the complete 2003 survey and for “campo sujo” and “cerradão” habitats were relatively close to the richness estimated by the Chao1 estimator (Table 3), indicating that the sampling effort was appropriate to survey this community, as was shown also by the coverage estimate (98, 97, and 83%, respectively). On the other hand, for the “campo limpo” and “cerrado sensu stricto,” the observed richness was only half of the estimated richness, which is clear through the low coverage value for “cerrado sensu stricto” (55%), but not for “campo limpo” (92%), indicating that a more extensive sampling in the “cerrado sensu stricto” would be necessary (Table 3). Despite the extremely low species richness value obtained for “cerradão,” the rarefaction-extrapolation curves for the 2003 assemblage indicate no differences in the species richness among the habitats sampled, with values estimated for extrapolated data, corroborating the results of the Chao1 richness estimator (Figure 1A).

For the 2017–2018 survey, all the observed and estimated species richness were also similar, with high coverage values (99% for total data and habitats), evidencing that the sampling effort was sufficient to estimate the community richness (Table 3). This result corroborates those obtained by the rarefaction/extrapolation curves, which also indicates that only the “campo cerrado” differed in relation to the other habitats surveyed, showing comparatively lower richness (Figure 1B).

The species with the highest abundance for the 2003 survey was Necromys lasiurus, representing 71% of the assemblage (Figure 2A), followed by other rodents such as Cerradomys scotti (6.5%) and Oligoryzomys mattogrossae (4.8%). Furthermore, the less abundant species were the rodents Akodon cf. montensis and Calomys tener, with only one individual captured each, representing, together, 1.6% of total abundance (Figure 2A). This pattern became clear in the Abundance-based Coverage Estimator (ACE) analysis, which considered only N. lasiurus as an abundant species, and the nine remaining species as rare (k < 10). For the 2017–2018 survey, the species of Oligoryzomys were the most abundant, with O. nigripes comprising 25.4% of the assemblage and O. mattogrossae 18.6%. Other abundant species in this assemblage were the rodents Calomys tener (14.8%) and Cerradomys scotti (11.2%). The rarest species were the rodents Nectomys cf. squamipes (one individual) and Oxymycterus delator (two individuals), representing, together, less than 0.4% of total abundance (Figure 2B). For 2017–2018, the ACE analysis indicates 12 abundant species (k > 10), and five rare species.

For patterns of habitat use, the non-metric multidimensional scaling for the 2017–2018 survey (Figure 3; stress = 0.09, Supplementary Figure 1) showed distinct species groups, one composed by Calomys tener, Cerradomys scotti, Oligoryzomys mattogrossae, and Cryptonanus chacoensis that seemed more related to “campo cerrado,” a grassland formation; and another formed by D. albiventris, Gracilinanus agilis, and Hylaeamys megacephalus that were located closer to “cerradão,” a forested habitat. Also related with “cerradão” are Gracilinanus microtarsus and Rhipidomys cf. macrurus, but more dispersed than the previous species. Akodon cf. montensis and Oligoryzomys nigripes were located in between “cerrado sensu stricto” (savanna) and “cerradão” (forest). The species Cryptonanus aff. chacoensis and Clyomys laticeps were a bit more dispersed, but relatively close to the “campo sujo,” the most open habitat at SBES. Monodelphis kunsi was located in a transition between “cerradão,” “cerrado sensu stricto,” and “campo sujo,” and Necromys lasiurus between “cerrado sensu stricto,” “campo cerrado,” and “campo sujo.”

The seven microhabitat variables were reduced to two principal components with eigenvalues > 1 (3.12 and 1.58, respectively, Supplementary Table 2) that represented about two-thirds of the total variance in our dataset (67.2%, Supplementary Table 2 and Supplementary Figure 2). The first principal component (hereafter PC1 variable) represented 44.6% of the total variance and is an indicator of high herb cover and low tree cover. The second principal component retained (hereafter PC2 variable) is an indicator of high shrub cover and low herb cover and represented 22.6% of the total variance (Supplementary Table 3). The gradient in relation to vegetation cover and canopy openness is visible in the PCA graph and is evidently related to the PC1 and PC2 variables (Supplementary Figure 2). There was no difference in richness considering the PC1 and PC2 variables (null model: p = 0.50), i.e., richness did not differ across the habitat gradient. Despite that, total abundance was positively affected by shrub cover and negatively by herb cover (PC2), with an average increase of 18% in abundance per unit of PC2 (p < 0.001).

For the relative abundance (for species with n > 10 captures), the rodents Calomys tener (p < 0.001) and Clyomys laticeps (p = 0.01) were positively associated with herb cover. The estimated average increase in the relative abundance of Calomys tener in areas with high herb cover was 80.7% per PC1 unit (p < 0.001). On the other hand, Clyomys laticeps showed an average decrease of 74% per unit of PC2 (p = 0.004), which indicates a high positive relation with herb cover and a negative relation with shrub cover. Cerradomys scotti (p = 0.015) and Oligoryzomys nigripes (p = 0.02) were the species positively associated with shrub cover, with an average increase of 45 and 51.8% (p = 0.02) in their relative abundance per unit of PC2, respectively. Oligoryzomys mattogrossae was associated with both herb and shrub habitats (p < 0.001), with an estimated average increase of 53% (p = 0.04) in herbaceous areas and 58% in shrub units (p < 0.001). The marsupial D. albiventris was positively related to tree and shrub cover (p < 0.001), with an average increase of 67.3% in its abundance in forest areas and 18% for shrublands. The marsupial Gracilinanus agilis and the rodent Hylaeamys megacephalus were not influenced by PC1 and PC2 in their relative abundances (p = 0.14 and 0.58, respectively).

Our linear model showed an increase of 0.1 on the EVI mean between the 2003 and 2018 survey (R² adj. = 0.26, F₁,₁₄ = 6.38, p = 0.02). This result indicates an increase in vegetation density of 27.4% in 15 years, affecting primarily the open areas (Figures 4, 5A). The linear model for EVI regarding the habitat types demonstrated a clear difference in its EVI mean values (R² adj. = 0.62, F₄,₁₁ = 7.1, p = 0.004, Figure 5B), with the open habitats (“campo limpo,” “campo sujo,” and “campo cerrado”) not different considering their EVI mean values, despite following a gradual increase on averages (0.30, 0.36, and 0.41, respectively). However, for the intermediate habitat, “cerrado sensu stricto,” we estimate an increase of 46.6% on the EVI mean value compared to the more open habitat (“campo limpo”), with an EVI mean value of 0.44. For the forest formation, “cerradão,” the model estimated a difference in the EVI mean value of 70% higher in relation to the “campo limpo” area, reaching the highest EVI mean value, 0.51 (Figure 5B).

The richness of small mammals was not affected by the increase in wood density (p = 0.39). We also tested the richness per mammal order, and obtained the same results, with no difference in richness as a function of EVI mean (Rodents: p = 0.27; Marsupials: p = 0.90). The total abundance of small mammals, on the other hand, was negatively affected by an increase in wood density (p = 0.02, Figure 6A), and for the rodents alone, a similar result was obtained (p = 0.01, Figure 6B). However, the total abundance of marsupials was not affected by changes in mean EVI (p = 0.61). We also fitted models excluding captures of Necromys lasiurus (due to its dominance in the 2003 survey), and the null models were selected for total and rodent abundances (Total community: p = 0.17; Rodents: p = 0.06).

Among the species with at least 10 individuals captured, the rodents Calomys tener and Oligoryzomys mattogrossae seemed to be unaffected by the increase in EVI mean values (p = 0.44 and 0.20, respectively). The marsupial Cryptonanus chacoensis and the rodents Cerradomys scotti and Necromys lasiurus were negatively affected by the local woody encroachment, with a decrease in their abundances related to higher mean EVI (p = 0.001, <0.001, and 0.01, respectively, Figures 6C,E,G). The marsupial D. albiventris and the rodent Oligoryzomys nigripes were positively affected by increases in EVI mean (p = 0.001 and p < 0.001, respectively, Figures 6D,F).

The Beta diversity analysis showed high dissimilarity between the 2003 and 2018 surveys (β_total = 0.84), which is explained by the replacement of species (β_repl = 0.84 and β_rich = 0). The composition in the 2018 survey presented an increase of four species compared to the 2003 survey, D. albiventris, Gracilinanus agilis, Hylaeamys megacephalus, and Rhipidomys cf. macrurus, most of them associated with the “cerradão,” a forest formation (Figure 3).

A high diversity of small mammals was found in our study area. We recorded at SBES 15% of the 113 native small mammal species from Cerrado (Mendonça et al., 2018; Fegies et al., 2021). The total richness observed (17 species) is high, given that only about 8% of the Cerrado small mammal communities are composed of more than 10 species (Mendonça et al., 2018), evidencing the completeness of the surveys in SBES. Rodents were the richest order, a recurrent pattern (Quintela et al., 2020; Abreu et al., 2021), with five sigmodontine rodents representing 74% of the total abundance in the SBES assemblage (Oligoryzomys nigripes, O. mattogrossae, Calomys tener, Cerradomys scotti, and Necromys lasiurus). The general abundance pattern recorded for SBES is also in accordance with previous Cerrado studies (see Mendonça et al., 2018 review), with Necromys lasiurus being the dominant species in the 2003 survey, and O. nigripes in the 2017–2018 survey.

Although based on different sampling designs and efforts, the two temporally spaced surveys showed high values of sampling coverage, both for the total assemblages, and for each similar habitat surveyed, allowing general comparisons between the community structure patterns found. While the 2003 survey was characterized by lower species richness, the habitats with higher observed and estimated richness were the grasslands (“campo limpo” and “campo sujo”) and the savanna (“cerrado sensu stricto”), with the forest “cerradão” being the poorest habitat. On the other hand, the higher species richness found in the 2017–2018 survey was evenly distributed among grasslands (“campo sujo”), savannas (“cerrado sensu stricto”) and “cerradão,” but with “campo cerrado,” a grassland habitat, showing lower species richness. Moreover, the seven additional species recorded in the 2017–2018 survey are mostly associated with denser covered habitats, such as savannas and forests, as is the case of D. albiventris, Gracilinanus agilis, Hylaeamys megacephalus, and Rhipidomys cf. macrurus (e.g., Santos-Filho et al., 2012; Carmignotto et al., 2014; Carmignotto, 2019; this study). Indeed, two species typical of the Atlantic Forest (G. microtarsus and Nectomys cf. squamipes), which also occur in southern and eastern Cerrado, especially using gallery forests, seasonal forests, and “cerradão” patches (Costa, 2003; Carmignotto et al., 2012), were only represented in the 2017–2018 survey.

Regarding species abundances, we observed that in the 2003 survey, the most abundant species were also represented by grassland and savanna species. Although Necromys lasiurus is considered a habitat generalist (Vieira et al., 2005; Ribeiro et al., 2019), in Cerrado, it prefers open habitats, represented by grasslands and savannas (Becker et al., 2007; Carmignotto, 2019). A similar pattern was found for C. scotti and O. mattogrossae (Henriques et al., 1997; Vieira et al., 2005; Weksler and Bonvicino, 2015). In the 2017–2018 survey, in turn, the most abundant species, the rodent O. nigripes, is more associated with savanna and forests (Weksler and Bonvicino, 2005, 2015). Moreover, some species that were more abundant in the 2003 survey, became rare in 2017–2018, such as the grassland specialist Cryptonanus aff. chacoensis (Fegies et al., 2021). Vegetation shifts may trigger shifts in small mammal abundance (Loggins et al., 2019), but other factors, such as resource availability, reproductive activity, and presence of competitors (Verberk, 2011), may also play a role. Since several of our comparisons were made between 2003 and 2017–2018 surveys controlling for the period of sampling (January–February), we believe we were able to reduce the influence of some of these confounding factors in our results.

Cerrado small mammals present high habitat association, contributing with the well-known pattern of open (grasslands and savannas) versus forest specialists found across this ecoregion (e.g., Alho, 2005; Carmignotto et al., 2012). The SBES assemblage also fits into this pattern, showing open and forest specialists as well as generalists (occupying both open and forest habitats). Considering habitat use, the 2003 survey was mostly represented by open-habitat species, with few generalists. In the 2017–2018 survey, we observed an increase in generalists and forest specialists. Despite the maintenance of grassland specialists between surveys, two species (Cryptonanus aff. chacoensis and Clyomys laticeps) were restricted to the most open habitat currently present at SBES (“campo sujo”), indicating a strong association with open grasslands (Carmignotto et al., 2014; Bezerra and Bonvicino, 2015; Bezerra et al., 2016; Fegies et al., 2021).

The microhabitat preference of species corroborates the results found at a larger, habitat scale in our study. For instance, Calomys tener and Clyomys laticeps show a higher association with high herbaceous cover (Carmignotto and Aires, 2011; Rocha et al., 2011; Bezerra et al., 2016). For Clyomys laticeps, however, our models indicate a negative relationship with high shrub cover, highlighting the dependence of this species on open grasslands (Vieira, 2003; Bezerra and Bonvicino, 2015). Other rodents (Cerradomys scotti, Oligoryzomys nigripes, and O. mattogrossae) were positively affected by shrub density. Cerradomys scotti and O. mattogrossae are open-habitat species, exhibiting a preference for grassland areas with higher shrub cover (Vieira et al., 2005). Oligoryzomys nigripes, however, is known to be associated with forest habitats, such as gallery forest in the Cerrado (Weksler and Bonvicino, 2005). Our findings are in accordance with these results since most captures were recorded in the “cerradão.” Püttker et al. (2008) also reported a higher association of this species with areas with low canopy and dense understory in the Atlantic Forest. D. albiventris, on the other hand, seems to benefit from increasing canopy cover and tree density, agreeing with previous studies where its presence was related to fallen logs and shrub (Piper sp.) density (Melo et al., 2013). Gracilinanus agilis and Hylaeamys megacephalus showed no clear response to the microhabitat variables tested, and their presence should be due to other factors not considered in our study, such as resource abundance or disturbances (Verberk, 2011; Loggins et al., 2019). These results highlight the importance of evaluating habitat use at different scales to better understand the potential reasons behind differential occurrence of small mammal species across the Cerrado, and to reinforce why some species would be favored or disfavored in a woody encroachment scenario.

We observed an increase in vegetation density in SBES throughout the 15 years between the two surveys analyzed (from 2003 to 2018), corroborating that woody encroachment may be a common phenomenon in the Cerrado (e.g., Moreira, 2000; Roitman et al., 2008; Pinheiro and Durigan, 2009; Abreu et al., 2017, 2021). The 27% increase in the mean EVI affected primarily the grasslands. Indeed, the “campo sujo” surveyed in 2018 were located at the same areas previously classified as “campo limpo” in the 2003 survey, clearly showing a shift in vegetation with the increase in density of shrubs and small trees (see also Melo and Durigan, 2011). Overall, the small mammal species richness was not affected by the woody encroachment observed in the study area, but the total abundance was negatively related to the increase in mean EVI, as observed for other plant and invertebrate groups studied at SBES (Abreu et al., 2017). Indeed, our results show a clear negative effect of woody encroachment on the density of rodents, but not for marsupials. This is expected since most marsupials are associated with savannas or forest habitats due to their scansorial and arboreal habits (Astúa, 2015). Rodents have shown a two-way relationship with vegetation thickening in other open regions. On one hand, they can contribute to seed removal and consequently decrease woody density at habitats in the initial phase of encroachment, as recorded for open habitats of Africa, Argentina, and Australia (Busch et al., 2012; Bergstrom, 2013; Teman et al., 2021). On the other hand, they can be negatively affected by the encroachment, which impacts their ecological features, such as reducing predator detection, habitat use, and local persistence (Loggins et al., 2019), similar to what our results have indicated here. Our results for total and rodent abundance may have been influenced by the high number of Necromys lasiurus individuals in the 2003 survey. When we excluded this species from the analyses, no influence of woody encroachment for small mammals and rodent abundance was detected. Thus, these results should be carefully interpreted due to this outlier influence and low sample size, and additional sampling should be made to confirm our interpretations.

Regarding the seven most abundant species, two open-habitat specialists (C. tener and O. mattogrossae) were not affected by woody encroachment. These species can occupy a wide range of open habitats, from grasslands to savannas in the Cerrado (Carmignotto et al., 2014; Salazar-Bravo, 2015; Weksler et al., 2017; Bezerra et al., 2020). This pattern indicates plasticity in their habitat use. On the other hand, the open-habitat specialists Cryptonanus chacoensis, Cerradomys scotti, and Necromys lasiurus were negatively affected by woody encroachment, indicating a dependence on grasslands formations at SBES. The negative impacts of woody encroachment on species of the genus Cryptonanus are expected, since its diversification is highly associated with the open habitats of the Cerrado (Fegies et al., 2021). In fact, SBES shelters a sympatric species, Cryptonanus aff. chacoensis, considered rare and endemic of Brazilian Cerrado, which may face even more severe impacts by the woody encroachment (Fegies et al., 2021). Cerradomys scotti showed a slight preference for shrubby habitats in our microhabitat analyzes, which is in accordance with the negative response to the tree encroachment found here, a pattern previously observed by Vieira et al. (2005), where this species was related to grass height but not to arboreal cover. In fact, there seems to be a tenuous line for species that may benefit from shrub cover due to protection against predation, and also may be negatively affected by a decrease in food resources, since several species use herbaceous sources for feeding (Vieira et al., 2005; Ribeiro et al., 2019), which, in turn, decrease with shrub encroachment, as shown for small mammals in African shrub-invaded grasslands (Loggins et al., 2019).

The rodent Necromys lasiurus also showed a decrease in abundance between 2003 and 2017–2018 potentially due to woody encroachment, since it is associated with open and grassy areas (Vieira et al., 2005; Becker et al., 2007; Rocha et al., 2011). This species was the most abundant in open areas in 2003 but among the rarest ones in 2018, extending its distribution to savanna habitats, its preferred habitat in other studies (Henriques and Alho, 1991), indicating its habitat plasticity within open formations. Our results show that Cryptonanus chacoensis and Cerradomys scotti may not persist in areas with woody encroachment if grasslands disappear. Meanwhile, forest-specialists and opportunistic species seem to benefit from woody encroachment at the SBES. D. albiventris, although considered a habitat generalist (Cáceres et al., 2012), was not captured in the 2003 survey, but in 2018 became the most abundant marsupial in the assemblage. This didelphid was also the only species that was positively associated with microhabitats with higher tree density, which corroborates its fostering by woody encroachment. The rodent Oligoryzomys nigripes was also favored by the encroachment, and its higher association with forest habitats may allow its spread in habitats with higher woody density (Weksler and Bonvicino, 2015). Mammals across savannas worldwide are differently influenced by vegetation encroachment (Stevens et al., 2016a), but they seem to be more sensitive than other vertebrates, such as birds, due to their specialized habitat preferences and foraging strategies. Turnover in species composition following woody encroachment is also recorded in African savannas, with browser mammals replacing grazers (Smit and Prins, 2015), pointing to woody encroachment as a general concern across savannas.

Our models indicate shifts in species abundance across time, with such changes being predictable to some extent. Species typical of grasslands show declines, while forest-dwellers are increasing in abundance. In the case of forest-savanna ecotone regions, climate, and land-use change, especially fire and deforestation, are leading to an invasion of savanna species into disturbed forests, shifting the forest fauna assemblages toward a “savannization” (Sales et al., 2020). Similarly, the woody encroachment is changing southern Cerrado assemblages toward a “forestization,” with the invasion of species from adjacent forest biomes and the loss of savanna specialists (Abreu et al., 2017, 2021). Our findings highlight the importance of the maintenance of the mosaic of open formations in Cerrado remnants in order to shelter a high diversity of small mammal grassland specialists. Species that showed a clear decline and high association with open vegetation structure in fact can become locally extinct, consequently altering species range, since these areas are mainly located at Cerrado boundaries. This may be particularly true for Cerrado endemics and regionally vulnerable species such as Clyomys laticeps and Cerradomys scotti (Percequillo et al., 2008; São Paulo, 2018), besides a rare and still undescribed species (Cryptonanus aff. chacoensis; Fegies et al., 2021). In fact, the SBES was created with the goal to protect the open formations of the southern Cerrado, but 30 years of fire suppression are probably the main cause of the local woody encroachment and the resulting changes in biodiversity (Abreu et al., 2017, 2021). Fire management is considered a key tool to maintain open savannas and its associated diversity, and the current fire experiments at SBES so far demonstrate no loss in small mammal diversity with prescribed fire (Durigan and Ratter, 2016; Durigan et al., 2020). Without active management of the landscape to keep open habitats, the long-term maintenance of the open-habitat specialists (with special attention to the rarest ones) will give place to an increasing replacement by forest specialists and habitat generalists in the SBES small mammal community and in other remnants in the southern portion of the Cerrado. Woody encroachment needs to be treated as a global scale problem to natural open ecosystems (Stevens et al., 2016a), and its impacts on biodiversity at local scales should continue to be investigated in order to guarantee the conservation of savanna and grassland biodiversity.