The role of fire disturbance on habitat structure and bird communities in South Brazilian Highland Grasslands

Grassland ecosystems are evolutionarily linked to disturbances such as grazing and fire. These disturbances define grassland plant communities and habitat heterogeneity, which influence animal communities. We evaluated the influence of fire disturbance on plant and bird communities and on habitat structure by sampling grassland fragments with different time elapsed since the last fire event. Habitat structure was sampled using plant life forms and abiotic variables and birds were sampled through point counts. We recorded 862 bird individuals from 70 species. Intermediately-burnt sites harbor higher habitat heterogeneity and plant species richness in comparison with recently or long-burnt sites. Bird abundance and taxonomic diversity decreased linearly as time since fire increased. Finally, time since fire influenced the relative distribution of plant life forms and bird food guilds. Our results indicate that fire management should be included in the framework for conservation and sustainable use of grasslands, because it promotes habitat heterogeneity and diversity. To maintain habitat heterogeneity and the related habitat-specific bird species and functional groups, conservation efforts should maintain grassland patches under different management intensities and frequencies on a landscape level. However, studies focused on determining the periodicity with which fire management should be used are still lacking.


Materials and methods
Sampling sites. The study sites were located in South Brazilian Highland Grasslands (in the state of Rio Grande do Sul, Brazil 29º05′S, 50º05′W), inserted in the Atlantic Forest biome, where there is a physiognomy of grassland-forest mosaics ( Fig. 1; see 75 and 78 for a review on plant and bird communities from these ecological systems, respectively). We selected seven sites (A-G; Fig. 1) with natural grassland vegetation that comprised a chronosequence of different fire disturbance histories. This chronosequence describes the time elapsed since the last fire event in each site ('time since fire'), i.e., the amount of time that plant and bird communities in each site had to reassemble after the reset promoted by fire. This terminology was already applied in previous studies of fire disturbances in South American grasslands 14,17,26 . The time since fire varied from 3 to 180 months (see Fig. 1), and was either monitored by the researchers or reported by landowners and park managers. Sites A and C are in the immediate vicinity of Tainhas State Park, in private farms, while site B is inside of this protected area. Site D is inside of Serra Geral National Park and sites E-G are inside Aparados da Serra National Park (Fig. 1). All sites have been under cattle grazing for many decades. A recent regional classification of South Brazilian grasslands indicated 'highland grasslands' as a discrete unit that share not only soil features and plant community structure, but similar management as well 79 . Natural grasslands in the region are historically used Scientific Reports | (2020) 10:19708 | https://doi.org/10.1038/s41598-020-76758-z www.nature.com/scientificreports/ as the primary forage source for free-range cattle ranching. Grazing pressure is usually very low, with stocking rates of less than 1.0 animal unit (450 kg of live weight) per hectare 75 . These low stocking rates are based on the low carrying capacity of the grasslands during winter, but are maintained throughout the year. This leads to the accumulation of large quantities of dead biomass and, ultimately, to use of fire to 'reset' the system with new (and more palatable) plant biomass 26,75,80 . Because part of the land of the protected areas are still under private management (i.e., federal or state authorities are still in the process of paying and reclaiming the ownership of the land from original landowners), grazing pressure is similar in farms and protected areas. In addition, even in protected areas that are officially owned and managed by federal or state authorities, cattle grazing by animals from neighboring private properties is not uncommon. We did not directly estimate grazing pressure in the sampling sites, because we assumed that the low and similar stocking rates would have similar effects on biological descriptors in all sites. In addition, we assumed that the effect of time since fire would be stronger and more detectable in comparison with the effect of eventual small variances in grazing pressure across sites.
Bird sampling. We sampled birds during the peak of breeding season, between December 2017 and February 2018, using point counts of 10 min and 80 m radius 81 . Point counts were carried out under appropriate weather conditions and by the same observer in the mornings (sunrise until 10:00 a.m.). The number of points was proportional to the size of the grassland fragment at each site, and points were allocated at least 300 m apart. They were allocated systematically in sampling sites, the sampling effort in these different habitats was proportionate to their spatial representativeness in each site and distributed to cover as much of the grassland-covered area as possible.  (40). These quadrats were allocated from the bird point count 10 m to the north, 25 m to the west, 50 m to the south, and 75 m to the east. This approach was used to maximize the coverage of the habitat heterogeneity that may exist at each point count 85,86 . In addition, we counted all individuals of Eryngium and shrubs in 10 m buffers around each quadrat, so that the sampling method encompassed larger specimens and covered a larger portion of the habitat used by birds sampled in point counts.
Data analysis. The first step in our analytical framework was to summarize plant and bird data, as well as time since fire (log-transformed for all analyses), in matrices with pooled values for each sampling site ( Fig. 2A).
In this paper, all analyses were carried out at the site level, and patterns related to within-site variations will be addressed elsewhere. Plant community and habitat descriptors were pooled in matrix E of seven sampling sites described by mean values of each variable: plant species richness (total and separate by graminoids, forbs, and shrubs); abundance (forbs, shrubs, and Eryngium); percentage (cover estimation in percentage) of vegetation, water, rock, and bare soil; and mean vegetation height. Abundance of shrubs and Eryngium was sampled in 1 m 2 plots and in 10 m buffers. This matrix was used as a descriptor of habitat structure. Since the variables that comprised matrix E could show varying levels of collinearity, we used the Variance Inflation Factor (VIF) to detect multicollinearity 87 , using package 'olsrr' 88 . We calculated VIF for all variables in matrix E, which resulted in values lower than 5 for all variables (which indicates insignificant inflation due to multicollinearity 89 ). Bird community data were summarized in matrix W, with seven sites described by mean values of species abundance ( Fig. 2A). This matrix was used as a descriptor of bird community taxonomic structure. We divided bird species in six food guilds (treated as separate binary traits in all analyses): (Ne) nectarivores, (Gr) granivores, (Fr) frugivores, (In) insectivores, (Ca) carnivores, and (Om) omnivores. We consider a species as omnivore when it could be simultaneously included in three or more guilds. The sorting of species into guilds was based on the 'Handbook of the Birds of the World Alive' 90 , and on personal field experience. Bird food guild information was summarized in matrix B of bird species described by six food guilds as binary functional traits ( Fig. 2A). Trait information was scaled-up to the community level by matrix multiplication B'W, resulting in matrix T with community-weighted mean (CWM) trait values, which in our case represent the relative contribution of each food guild at each sampling site 91 (Fig. 2B). This matrix was used as a descriptor of bird community functional structure.
We performed Principal Component Analyses (PCA) to extract the scores of the first ordination axis of matrices E, W, and T (Fig. 2C). With this, we obtained vectors describing habitat structure and bird taxonomic and functional structure, respectively. Additionally, we calculated bird taxonomic diversity using Simpson's 1-D 92 from matrix W, and bird functional diversity using Rao's entropy based on Gower dissimilarities 93 2C). In this sense, the heterogeneity index can be interpreted similarly to the species diversity index, theoretically ranging from zero (complete dominance of one life form) to one (relative contribution of life forms evenly distributed).
All analyses were carried out in the R environment 94 , and we assumed the probabilistic threshold of 95% for statistical significance. To test most of our hypotheses, we used a statistical framework based on Linear Mixed-Effect Models 95 ("LME"; Fig. 2C), with package 'nlme' 96 . We built separate models with time since fire (logtransformed) as independent variable and individual variables and ordination axis from matrices E, W, and T as dependent variables, to test the hypotheses related to the influence of disturbance frequency on plant/habitat and bird community taxonomic and functional descriptors, respectively. Additionally, we tested the relationship between habitat structure (first ordination axis of Matrix E) on bird taxonomic and functional descriptors. Regression models that involve the response of CWM traits to environmental variables may render highly biased results, such as the inflation of type-I error [96][97][98][99][100] . To avoid this potential problem, we used the multilevel method (MLM3) proposed by ter Braak 99 to test these specific relationships. In this method, models are corrected by weighting with Hill's effective number (N 2 ) of species' occurrence and N2-diversity of sites, which are then combined in a sequential permutation-based test ('max test'). The test returns two separate p-values for the model, one for sites and other for species that generated the CWM traits, and we report both in the results. Since our sampling sites were distributed in two spatially segregated areas ( Fig. 1), we included the area (Tainhas and Serra Geral/Aparados area) as a random variable in all models to control for spatial autocorrelation. For each tested relationship, we built two concurrent models, considering a linear and a quadratic relationship between dependent and independent variables. We then compared each pair of concurrent models using analyses of variance with permutation and, when comparisons resulted in significant differences, we chose the best model based on the Akaike Information Criterion (AICc; 95 ). When models were equivalent, we used the linear relationship. We estimated the explanatory power of all models by calculating pseudo-R 2 (Maximum Likelihood) and p-values through comparisons with null models 95 using package 'rcompanion' 101 . We chose to compare between concurrent models because we wanted to evaluate if plant and bird community descriptors response to time since fire was either linear or peaked (i.e., with highest values in intermediate disturbance frequencies, as described by a quadratic model). www.nature.com/scientificreports/ Bird community descriptors were separated in two dimensions for statistical analysis: taxonomic descriptors-abundance, species richness, diversity, and composition (first ordination axis of matrix W), and functional descriptors-functional diversity, individual CWM traits, and guild composition (first ordination axis of matrix T). After, to address the hypotheses related to the influence of habitat structure on bird taxonomic and functional composition, we used a Canonical Correspondence Analyses (CCA; 102,103 ; Fig. 2C). The CCA was performed with matrices W and T, using matrix E to constrain the ordinations. To avoid overfitting, we reduced the original matrix E by removing all variables that showed significant collinearity. The remaining variables were mean vegetation height and abundance of forbs, shrubs, and Eryngium in a 10 m radius. We tested the statistical significance of the CCA models using Analysis of Variance with permutations.

Results
We recorded 862 individuals from 70 bird species, belonging to six food guilds. Insectivores and omnivores were the most representative food guilds, corresponding to 43% and 33% of the species, respectively, followed by granivores (11%). Carnivores, nectarivores, and frugivores together comprised 13% of the bird species. Four species are globally threatened (Anthus nattereri, Scytalopus iraiensis, Xanthopsar flavus, and Xolmis dominicanus), and three are near-threatened (Cinclodes pabsti, Limnoctites rectirostris, and Sporophila melanogaster) 104 . Three of these species with conservation concern (A. nattereri, C. pabsti, and X. flavus) were restricted to recently-burned sites, whereas S. melanogaster was recorded in all sites but showed higher abundance in sites with longer time since fire. (See Supplementary Table S1 with matrices W and T containing complete taxonomic and functional bird data, as well as species common names and IUCN threatened species categories).
The relationship between plant community descriptors and time since fire fitted a quadratic relationship in all cases, with the exception of mean vegetation height, which increased linearly with time since fire (Table 1, Fig. 3). However, linear and quadratic models were equivalent regarding the response of bird community descriptors to time since fire and plant/habitat descriptors, with the exception of bird taxonomic diversity explained by habitat structure (Table 1, Fig. 4).
Recently-burnt sites presented lower vegetation height in comparison with sites with longer time since fire (Fig. 3A). Sites with intermediate time since fire presented the higher values of habitat heterogeneity (Fig. 2B), plant species richness (Fig. 3C), graminoid species richness (Fig. 3D), and forb species richness (Table 1, Fig. 3E).  Fig. 3F). Habitat structure (the first ordination axis of matrix E) was also explained by time since fire (Table 1). However, total plant abundance and Eryngium abundance were not predicted by time since fire (LME p = 0.088; p = 0.297, respectively). See Supplementary Table S2 for the complete list of habitat variables (Matrix E).
The relationship between bird descriptors and time since fire fitted a linear relationship in all cases. Bird taxonomic diversity and abundance were higher in recently-burnt sites ( Table 1, Fig. 4A,B), but time since fire did not influence neither bird species richness (LME p = 0.051) nor bird functional diversity (LME p = 0.698). In addition, bird species composition (first ordination axis of matrix W) was predicted by time since fire (Table 1, Fig. 4C) and by habitat structure (first ordination axis of matrix E; Table 1). Habitat structure also predicted bird taxonomic diversity and abundance (Table 1). Bird community functional composition described by food guilds (CWM traits, first ordination axis of matrix T) was also predicted by time since fire (Table 1, Fig. 4D). The relative contribution of granivores was higher in sites with longer time since fire (MLM3 p SITES = 0.012, p SPECIES = 0.036, Fig. 4E).
Finally, Constrained Correspondence Analyses also revealed a significant relationship between habitat variables (mean vegetation height and abundance of forbs, shrubs, and Eryngium in a 10 m radius) and bird communities, considering both taxonomic (Matrix W; ANOVA χ 2 = 1.048, F = 2.2189, P = 0.038; Fig. 5) and functional (Matrix T; χ 2 = 0.5922, F = 13.787, P = 0.003, Fig. 6) dimensions. Considering the taxonomic dimention, bird species like Emberizoides ypiranganus and Sporophila melanogaster were related to taller vegetation and Eryngium in sites with longer time since fire. On the other hand, species such as Cinclodes pabsti and Anthus nattereri

Discussion
Here we evaluated the effect of fire disturbance, described by the time elapsed since the last fire event, on habitat and bird community structure descriptors, in Highland Grasslands from South Brazil. We corroborated the general hypothesis that grassland habitat structure and heterogeneity vary according to time since fire. Grasslands with long periods without disturbances usually present an increase in biomass, with homogenization of vegetation structure 39 , which leads to the exclusion of species 14,80 and to overall diversity losses 14 . Accordingly, we found that sites with longer time since fire presented lower plant species richness for all groups, as well as lower habitat heterogeneity (i.e., high dominance of one or few plant life forms), and overall taller vegetation (which is an indication of increased biomass, although we did not measure this variable directly). These findings corroborate other studies from South American grasslands, providing further evidence that grasslands from this region are a discrete ecological unit that responds similarly to fire disturbance: by reducing the dominance of a few species and life forms that outcompete other taxa in the absence of disturbance 14,15,17 . In grasslands from Argentina, burned sites presented fast recover of plant cover and species richness 15 . In Uruguayan grasslands, fire removed above-ground biomass of the dominant species (Saccharum angustifolium), causing a release of resources, such as space and light, consequently promoting habitat heterogeneity 17 . Interestingly, plant community responses to time since fire were peaked (or hump-shaped), i.e., plant descriptors showed the highest values in areas with intermediate time since fire (with the exception of the linear response of vegetation height). Peaked diversity-disturbance relationships (DDR) are a common pattern in nature, although not the dominant one: a relatively recent study reviewed almost 200 papers on the subject, and found peaked DDR in ca. 20% of them, including responses of richness, diversity, and evenness 49 . In this study, peaked DDR were more commonly found when the disturbance could be considered 'natural' rather than anthropogenic, and intrinsic to the study system. Researchers that find peaked DDR in their study systems often rely on the intermediate disturbance hypothesis (IDH; [50][51][52] to explain their patterns, although the mechanisms by which the IDH explain peaked DDR have been severely questioned, both on theoretical and empirical grounds 53 . Nevertheless, we found peaked DDR for most plant community descriptors, providing further empirical evidence that this is a recurrent response pattern of grassland plant communities to disturbance, even though the mechanisms behind this pattern may be different from those postulated in the IDH. Furthermore, the responses of species richness and habitat heterogeneity to fire management relaxation (i.e., lower fire frequency) that we have reported here are similar to what has been reported for grazing relaxation in the same region 41 . This is the expected response of grassland plant diversity to grazing intensity/frequency 54,55 , and it seems that fire disturbance can promote similar response patterns. Differently from other plant descriptors, vegetation height and shrub abundance increased with time since fire (Fig. 3A,D). Sites with longer time since fire are usually dominated by tussock grasses and Sampling sites indicated by letters (as in Fig. 1) followed by time since fire. Bird species shown in red (see Supplementary Table S1 for complete species list with species common names and IUCN threatened species categories). Selected representative species (due to differential habitat use; see discussion) shown in photographs and in bold red.  www.nature.com/scientificreports/ shrubs ('shrub encroachment' 43 ). Current climatic conditions allow for forest development over grasslands in the study region. In absence of disturbances regimes, shrub encroachment may take place 39,43 , which would ultimately lead to losses of characteristic grassland species 75 . Furthermore, large areas that stay long periods without fire disturbance accumulate woody flammable biomass, and may be subject to a future fire event of greater intensity (often catastrophic), even reaching wetlands and forests 105,106 . However, the structurally homogeneous and species-poor vegetation that derives from management exclusion seems to provide resource for some specialized birds. Vegetation height has been reported as an important predictor of grassland bird composition 10,107 . Accordingly, we observed some specialized bird species associated with tall vegetation (e.g., Emberizoides ypiranganus, Sporophila melanogaster, and Phacellodomus striaticollis), whereas other species were observed only in shorter vegetation (e.g., Vanellus chilensis, Cinclodes pabsti, Anthus spp.). The differences in species bird composition in distinct vegetation types are due to the biology of each species, which requires different structures of the vegetation to forage, nest, and take refuge. These findings reinforce the importance of habitat heterogeneity, here mediated by fire, for birds with different habitat requirements to colonize 10 .
We partly corroborated our hypotheses relating time since fire to bird community descriptors. Areas with longer time since fire presented lower bird taxonomic diversity and abundance as expected, but no such pattern was found for bird functional diversity. Previous works showed that bird taxonomic diversity and relative abundance decline as disturbance-excluded grasslands become denser with both plant individuals and litter, which ultimately hampers foraging for ground-feeding birds 64 . The overall bird functional composition was predicted by time since fire, but not by habitat structure. The CCA, with fewer selected habitat variables, revealed a significant pattern (Fig. 6). Frugivore birds were associated with forbs and Eryngium (Figs. 5, 6), which plays a critical role in the ecology of many bird species (e.g., nesting or foraging 66,85 ). We observed the occurrence of Long-tailed Cinclodes (Cinclodes pabsti) and Ochre-breasted Pipit (Anthus nattereri) in recently burned sites or with an intermediate time since fire. The Ochre-breasted Pipit has been associated with natural grasslands under fire influence 108,109 . In addition, nectarivores and carnivores were also abundant in these sites, probably due to the high abundance of forbs and Eryngium, which provide more resources in these shorter grassland sites in comparison with areas with higher and homogeneous vegetation (Fig. 6). Raptors (Accipitridae and Falconidae) are attracted to newly-burned areas for the availability of prey 66 . A study carried in the Highland grasslands observed that the extinction-threatened Saffron-Cowled Blackbird (Xanthopsar flavus) is a specialist species restricted to areas under disturbance regimes, with shorter vegetation 110 . Similarly, in Argentina and Uruguay, X. flavus selects habitats with short grasses and bare soils that result from livestock grazing 111 . Accordingly, our results indicated that this species was related to recently-burnt areas with shorter vegetation. These findings reinforce that the inclusion of management (i.e., prescribed disturbances) in grassland conservation policies and practices is paramount, since threatened and near-threatened bird species seem to depend on specific habitat configurations that are absent from management-excluded areas.
In an opposite way, granivores were associated with tall vegetation and longer time since fire. This could be related to the higher availability of graminoid seeds for these birds, since sites with longer time since fire had more time to accumulate biomass and produce more seeds. However, we did not sample graminoid species abundance or seed production, therefore this tentative explanation should be considered with caution, and explicitly tested in future research, taking into account the potential negative effects of self-shading processes on seed production 112 . In these sites we found species like the Lesser Grass-Finch (Emberizoides ypiranganus) and the Black-bellied Seedeater (Sporophila melanogaster), which is a near-threatened species. These species are specialized (and therefore dependent) on tall vegetation structure to nest, forage and seek refuge 71,[112][113][114][115] . Besides the Black-bellied Seedeater, other granivore species (e.g., Saffron Finch Sicalis flaveola, Grassland Yellow-Finch Sicalis luteola, and Long-tailed Reed Finch Donacospiza albifrons, although the last species also includes insects in its diet) were more abundant in these sites as well, which could also be related to higher seed availability.
Habitat structure was a good predictor of bird community structure, corroborating our last hypothesis. Bird composition, as well as bird taxonomic diversity and abundance, were predicted by habitat variables. Therefore, we corroborate the idea that the response of the bird community to fire disturbance was related to shifts in the relationships of abundance/dominance between plant species in each fire history, an example of community structuring through bottom-up effects 116,117 . Considering the differences in plant descriptors among different times since fire, it seems that the heterogeneity of vegetation structure, and the differential resource availability that follows it, may play an important role in defining patterns of bird communities, as reported before for other ecosystems (e.g., 9,10,[30][31][32]. The responses of all bird community descriptors to disturbance were linear, which contrasts with the peaked responses found for plant community descriptors. Although areas with longer time since fire showed overall lower bird taxonomic diversity and abundance, they also seemed to benefit granivores birds, a guild that includes important grassland species. Therefore, to maximize conservation results regarding bird species, it is necessary to have grassland patches under different disturbance regimes (i.e., with different fire frequencies), which result in different habitat types for the associated bird species 30,72 . However, seasonal timing and frequency in prescribed fires in grassland management are key issues, especially considering bird species. A fire occurring during the nesting season, a vulnerable time in the life cycle of a bird, may result in loss of active nests or mortality of fledglings 70,72 . Taken together, the responses of plant and bird communities to time since fire we reported here reinforce the well-established close relationship between grassland ecosystems and fire disturbance 1,5,13,14,22,26 . Scientific Reports | (2020) 10:19708 | https://doi.org/10.1038/s41598-020-76758-z www.nature.com/scientificreports/

Conclusions
Our results indicated that habitat variables related to the grassland vegetation respond to time since fire, and that these habitat variables shape bird community patterns. Plant communities and habitat heterogeneity responded to time since fire in a peaked relationship, whereas bird community descriptors responded linearly. Some bird descriptors responded to time since fire but not to habitat descriptors, suggesting that fire may influence bird communities through direct mechanisms we were not able to detect here. Our results corroborate and expand the bulk of empirical evidence on how South American Grasslands bird communities respond to fire disturbance 71,108,109,[112][113][114][115] , indicating that a mosaic of areas under different times since fire, which results in a mosaic of diverse grassland habitat structures, seems like the optimal conservation strategy. This conclusion reinforces the idea that the conservation of grassland plant and bird communities must include oriented management, such as prescribed patch burning. However, we emphasize that the conclusions and (especially) management suggestions drawn from our results should be taken with caution, given our relatively small number of spatial replicates. We are confident that the general patterns we presented can be extrapolated to the South Brazilian Highland Grasslands as a unit, but similar studies distributed across the different grassland systems in the region are needed to build more general and widely applicable fire management policies. Nevertheless, periodic prescribed burning has great potential to be used as a tool in the management of natural grassland areas, since it maintains not only the structure and scenic beauty of the grassland landscape, but also the diversity of both plants and birds. We emphasize the importance of habitat heterogeneity, that is, the habitat complexity that results from patches under different fire frequencies, as well as forests when considering the landscape-level mosaic, for the long-term conservation of grasslands in Southern Brazil. Finally, long-term studies are fundamental to determine the periodicity with which fire management should be used for conservation, and the effect of different intensities of grazing combined with fire should be considered in future studies to aggregate financial sustainability to biodiversity conservation.