Our study was conducted at the Dona Benta e Seu Caboclo Natural Heritage Reserve, located in Pirambu municipality on the northern coast of Sergipe, northeastern Brazil. This area is a Private Natural Heritage Reserve (RPPN), established by Ordinance No. 71 on August 27, 2010, and protects 26.6 ha of Restinga vegetation (Fig. 1). The reserve is predominantly composed of shrubby Restinga interspersed with herbaceous fields, small patches of Restinga forest, and both permanent and temporary ponds. The regional climate is classified as As (tropical wet climate with a dry summer) according to the Köppen-Geiger classification, characterized by distinct wet and dry seasons (Alvares et al. 2013). The rainy season extends from March to August, whereas the dry season occurs from September to February, resulting in a predominance of autumn-winter rainfall. Annual precipitation ranges from 1,500 to 1,800 mm, and the mean annual temperature is approximately 26°C (dos Santos-Neto et al. 2024).

Figure 1. 

Representative landscapes of the study area at the Dona Benta e Seu Caboclo Private Natural Heritage Reserve, Pirambu municipality, Sergipe, northeastern Brazil. A. Sandy dune slope with shrubby Restinga vegetation interspersed with exposed sand patches. B. Shrub-dominated Restinga bordering a temporary freshwater pond, illustrating the interface between terrestrial vegetation and seasonally flooded habitats. C. Panoramic view of the Restinga mosaic, showing sandy dunes, shrub vegetation, and a permanent pond embedded within the coastal plain. D. Open herbaceous–shrub Restinga on sandy, nutrient-poor soils, characterized by sparse ground cover and low woody vegetation. Photos by Jean Carlos Santos.

We established 20 permanent plots of 5 m² in areas with the physiognomy of non-seasonally flooded herbaceous fields (Oliveira and Landim 2014). All species within these plots were included in this study, identified through consultation of specialized literature (Prata et al. 2013; Oliveira et al. 2015) and following the classification proposed by the Angiosperm Phylogeny Group IV (The Angiosperm Phylogeny Group 2016). The correct scientific names were checked on the Flora e Funga do Brasil online platform (https://floradobrasil.jbrj.gov.br/reflora).

The phenological data collection was based on Nordt et al. (2021), with a few modifications. Instead of 1 m² plots, we had 20 plots of 5 m² each, distributed along an environment with the same physiognomy, non-flooded herbaceous fields (Oliveira and Landim 2014), spaced 10 meters apart. Phenological monitoring was conducted in 2021, beginning on January 5 and ending on December 27, with surveys performed weekly over two sampling days per week, totalling 104 data collection campaigns, which was considered an efficient frequency (Morellato et al. 2010). Each weekly campaign included observations across all plots to capture temporal variation in flowering phenology throughout the year (Suppl. material 1).

To characterize flowering, we used the Fournier Intensity Index (Fournier 1974). Flowering intensity was estimated for each species within each plot during weekly surveys, based on the proportional abundance of reproductive structures (flowers and/or inflorescences) relative to the maximum flowering expression observed for that species in each plot. In this method, phenological activity is scored semi-quantitatively on a scale from 0 to 4, where a value of 0 indicates the total absence of flowering activity, and values from 1 and 4 represent an increase of approximately 25% in flowering activity, 2 to 26–50%, 3 to 51–75%, and 4 to 76–100% of the maximum flowering expression observed within the sampling unit. Here, 100% phenological activity refers to peak flowering intensity, when the species exhibited its greatest observed abundance of flowers or inflorescences within a plot during the study period.

To characterize pollinator assemblages, floral visitors were observed for 20 minutes each hour in plots containing open flowers, 20-minute focal observations conducted sequentially in randomized order for each plot. Observations were conducted during two daily sampling periods: morning (07:00–10:00 h) and afternoon (15:00–17:00 h), with each sampling day consisting of a full rotation of all 20 plots (approximately 6.7 hours of observation effort per day). Sampling was conducted on two days per week throughout the year of 2021, resulting in 104 sampling days and approximately 693 hours of total observation effort across the study period. All insects visiting flowers within the quadrats were recorded, photographed, and, when necessary, captured using an insect net. Specimens were identified to the lowest possible taxonomic level using specialized taxonomic literature and speciesLink (https://specieslink.net), and with the assistance of expert entomologists.

Pollination syndromes were assigned based on published information available for each species or, when species-level information was unavailable, for the corresponding genus. Classifications followed standard pollination syndrome categories reported in the literature, including melittophily (bee pollination), psychophily (butterfly pollination), ornithophily (bird pollination), and generalist entomophily. Floral traits recorded in the field, specifically flower colour and floral symmetry, were assessed visually from fresh flowers following standard descriptive botanical protocols commonly used in pollination ecology and were used only as complementary information to support syndrome assignment rather than as the primary classification criterion. These trait observations were not based on colour space models or perceptual frameworks, but on direct field-based categorization consistent with comparative ecological studies.

We applied circular statistics to characterize temporal patterns in phenological activity. For each species, we calculated mean angle and its corresponding mean week (mean date), the length of vector (r), which quantifies the concentration of activity around the mean angle, as well as the variance and circular standard deviation. Directionality and departure from a uniform temporal distribution were assessed using the Rayleigh test, and statistical significance was evaluated based on its associated p value.

We evaluated whether the community flowering phenology was temporally aggregated, random, or segregated at both the community and trait levels using null model analyses of temporal niche overlap, following the framework originally proposed by Staggemeier et al. (2010). Phenological overlap among species was quantified using the mean pairwise overlap index described by Pleasants (1990), calculated for all species combinations using Schoener (1970) overlap formula. For each pair of species i and j, overlap was estimated as:

$$Oij=Oij=1-0.5(i\sum _{k=1}^n|pik-pjk|)$$

where and represent the proportional use of time interval k (weekly resolution) by species i and j, respectively. Weekly flowering data were converted to proportional values for each species, and the observed mean overlap was calculated as the average of all pairwise overlap values at the community level.

Departures from random expectations were assessed using Monte Carlo simulations (1,000 iterations). In each simulation, flowering weeks were randomly permuted within species, thereby randomizing the onset of flowering while preserving the shape and intensity of each species’ phenological distribution, under the assumption that flowering may occur throughout the year. The observed community-level overlap was compared to the null distribution using a right-tailed test, following the criteria of Wright and Calderon (1995). Flowering was considered temporally aggregated when the observed overlap exceeded 95% of simulated values, segregated when it was lower than 5% of simulated values, and random otherwise. Floral traits were categorized prior to analysis, with floral colour assigned according to the predominant floral display (white, yellow, pink, blue, and green) and floral symmetry classified as actinomorphic (radially symmetrical) or zygomorphic (bilaterally symmetrical) following established pollination morphology terminology (Neal et al. 1998). Using the same overlap matrix, we further evaluated whether temporal overlap was associated with floral traits by calculating mean pairwise overlap among species sharing the same category of floral colour or floral symmetry. Trait-based null models were generated by randomly permuting trait labels among species (1,000 permutations) while keeping phenological data unchanged, allowing us to test whether species with similar traits overlapped in flowering time more or less than expected by chance under the same statistical framework.

To evaluate the climatic drivers of flowering time, we used a model selection approach based on multiple linear regression and Akaike’s Information Criterion (AIC). Flower richness was modelled as a function of climatic variables describing precipitation, temperature, and photoperiod, using monthly time lags to account for delayed climatic effects. Climatic predictors included accumulated precipitation and mean temperature measured during the current month and during one to four months prior to flowering. Photoperiod was included as an additional predictor representing seasonal variation in day length. All predictors were evaluated in an additive framework.

We fitted all possible candidate models combining at most one precipitation variable, at most one temperature variable, and the presence or absence of photoperiod, including an intercept-only (null) model. This procedure generated a comprehensive set of competing models representing alternative hypotheses about climatic control of the number of species flowering. Models were fitted using ordinary least squares, and model performance was evaluated using AIC (Akaike 1974). Models were ranked by AIC, and ΔAIC values were calculated relative to the best-supported model. Models with ΔAIC ≤ 2 were considered to have substantial support. Variance inflation factors (VIFs) were calculated for the best-supported model to assess multicollinearity among predictors.

To test whether temporal overlap in flowering predicts similarity in pollinator assemblages, we constructed pairwise matrices of flowering phenological overlap and pollinator niche similarity among plant species. Flowering phenology was represented as a month-by-species presence-absence matrix, from which pairwise Jaccard similarity indices were calculated to quantify the degree of temporal overlap in flowering between species. Pollinator assemblage similarity was estimated using Bray-Curtis similarity based on quantitative visitation frequencies from plant-pollinator interaction networks. We then extracted the upper triangular elements of both matrices to generate a species-pair dataset. The relationship between flowering overlap and pollinator similarity was assessed using a beta regression model with a logit link, appropriate for proportional response data bounded between 0 and 1.

To test whether floral traits influence pollinator assemblage composition, we used permutational multivariate analysis of variance (PERMANOVA) based on Bray-Curtis dissimilarities derived from plant–pollinator visitation frequencies. Pollinator assemblages were summarized as a plant × pollinator interaction matrix, and pairwise dissimilarities among plant species were calculated using Bray-Curtis distance. Floral traits included flower colour (white, pink, yellow) and floral symmetry (actinomorphic vs zygomorphic), which were included as predictor variables in a PERMANOVA model (adonis2) with 9,999 permutations and marginal effects tested for each predictor. In addition, we performed Mantel tests using Spearman correlations to assess the relationship between floral trait similarity (Gower distance) and pollinator assemblage dissimilarity, and partial Mantel tests were used to control for phylogenetic relatedness among plant species.

To test whether overlap in flowering phenology predicts similarity in pollinator assemblages, we constructed a month-by-species binary matrix (presence/absence of flowering across 12 months). Phenological dissimilarity among plant species was calculated using Jaccard distance. Pollinator assemblage composition was again summarized using a plant × pollinator visitation matrix, and Bray-Curtis dissimilarity was used to quantify differences in pollinator assemblages among plant species. Relationships between phenological dissimilarity and pollinator assemblage dissimilarity were tested using Mantel tests with Spearman correlations and 9,999 permutations, and partial Mantel tests were performed to account for phylogenetic relatedness.

We reconstructed a phylogenetic hypothesis for the sampled herbaceous plant species using the R package U.PhyloMaker v.0.1.0 (Jin and Qian 2023). Species names were standardized for genus and family to match a megatree backbone and a genus-family reference list. Phylogenies were generated under scenario 3, which places missing species at the genus or family node when necessary. The resulting tree was exported, visually inspected, and polytomies were resolved using the multi2di function from the R package ape v.5.8-1 (Paradis et al. 2004).

Phylogenetic diversity (PD) of the 15 focal herbaceous species recorded in the study community was quantified using Faith’s PD, calculated as the sum of branch lengths connecting all species in the pruned phylogeny. To evaluate whether the observed phylogenetic structure differed from random expectations, we compared the observed PD of the entire sampled community (n = 15 species) to a null distribution generated by randomly drawing 15 species from the regional species pool described by Oliveira et al. (2015). This procedure was repeated 999 times to generate standardized effect sizes (SES) and two-tailed p values. Additionally, mean pairwise distance (MPD) and mean nearest taxon distance (MNTD) were calculated using the R package picante v.1.8.2 (Kembel et al. 2010) to assess phylogenetic structure at deeper and shallower evolutionary scales, with SES values obtained from 999 null randomizations.

We quantified phylogenetic signals in both flowering phenology (mean angles) and pollinator use at the species and functional-group levels. To account for the circular nature of flowering data, mean flowering angles were first linearized following the approach of Staggemeier et al. (2010). This involved calculating pairwise angular distances between species’ mean flowering dates, extracting species ordination scores via eigenvector analysis, and retaining the first eigenvector, which captured the majority of phenological variance. The linearized scores were validated by comparing Euclidean distances of the scores to the original angular distances using Mantel tests, confirming a strong correspondence. Phylogenetic signal in flowering phenology was then assessed using Blomberg’s K (Blomberg et al. 2003) and Pagel’s λ (Pagel 1999), testing whether closely related species flowered more similarly than expected by chance. For pollinator use, which was represented as normalized interaction matrices at the species and functional-group levels, phylogenetic signal was evaluated using Mantel correlations (Sokal 1979) between the dissimilarity matrices (1 – Schoener’s D) and the phylogenetic distance matrix. This combined approach allowed us to test whether both the timing of flowering and patterns of pollinator use were influenced by evolutionary relatedness.

At the community level, flowering phenology exhibited significantly greater temporal overlap than expected under the null model. The observed mean overlap across all species pairs was 0.174, whereas null simulations yielded consistently lower mean overlap values, indicating non-random synchrony in flowering phenology (p = 0.001). Trait-based analyses revealed variation in temporal overlap among functional categories. For floral colour, species with yellow flowers showed significantly higher flowering overlap than expected by chance (observed mean overlap = 0.535; simulated mean overlap = 0.381; p = 0.013), indicating aggregation in flowering timing within this group. In contrast, species with pink flowers (observed = 0.324; simulated = 0.381; p = 0.862) and white flowers (observed = 0.251; simulated = 0.291; p = 0.853) did not differ from null expectations, consistent with random temporal overlap. While for flower symmetry, neither functional group deviated significantly from the null model. Species with actinomorphic flowers exhibited an observed mean overlap of 0.252, comparable to the simulated mean overlap (0.274; p = 0.761). Similarly, species with zygomorphic flowers showed no significant departure from null expectations (observed = 0.335; simulated = 0.292; p = 0.121), indicating random flowering overlap with respect to floral symmetry.

We produced 36 models (ranging from one to three climatic variables), and eight models were statistically indistinguishable (ΔAIC < 2, S1), with our top model including the temperature from four months prior and photoperiod, showing low multicollinearity (VIF = 2.89). Moreover, both predictors in the highest-ranked model had negative effects on the monthly richness of flowering species. Specifically, each 1°C increase in mean temperature four months before the observation was associated with a reduction of approximately 1.6 flowering species, while each 1-hour increase in photoperiod was associated with a reduction of about 8.3 species (Fig. 4). These results indicate that the community flowering peak occurs before the maximum temperature and photoperiod values, reflecting a delayed phenological response to past thermal conditions and seasonal timing (adjusted R² = 0.54, p < 0.05).

Figure 4. 

Marginal effects of photoperiod (A) and lagged temperature (4 months, °C) (B) on flowering species richness. Grey points show raw data, blue lines show model predictions, and shaded areas represent 95% confidence intervals around the predicted relationships. Both variables exhibited significant negative effects on species richness, with richness decreasing as lagged temperature and photoperiod increased. Regression coefficients (β) and associated p values are shown within each panel.

The plant-pollinator interaction network (Fig. 5) was characterized by low overall connectance (0.11), indicating that only a small fraction of all possible plant-insect interactions was realized. Despite this sparsity, the network exhibited a strongly non-random structure, with high interaction specialization (H2' = 0.82) and significant nestedness (NODF = 22.97). Both metrics differed markedly from expectations under null models: observed nestedness was significantly higher than randomized networks (mean null = 17.36; p = 0.039), and specialization was far greater than expected by chance (mean null H2' = 0.10; p < 0.001). Species-level metrics revealed a heterogeneous distribution of interactions. Among insects, Xylocopa cearensis, Synoeca cyanea, and Eurema elathea were the most connected species (all degree = 4), with weighted betweenness values of 0.46, 0.26, and 0.14, respectively, indicating a central role in linking multiple plant species. On the plant side, Mitracarpus frigidus (Willd.) K.Schum. (degree = 5, weighted betweenness = 0.097), C. flava (degree = 4, weighted betweenness = 0), and Chamaecrista ramosa (degree = 2, weighted betweenness = 0.25) interacted with the greatest number of pollinators, with Paullinia pinnata (degree = 3, weighted betweenness = 0.28) also exhibiting high betweenness, suggesting an important connector role within the network. Overall, these results indicate a highly specialized and significantly nested plant-pollinator network structured around a small set of generalist and highly connected species, embedded within a broader matrix of specialized interactions.

Figure 5. 

Bipartite interaction plot of observed plant-pollinator relationships. Upper bars (green) represent insect visitors and lower bars (yellow) represent focal plant species. Bar widths indicate the relative frequency of recorded interactions. Grey bands link insects to the plant species they visited, with thicker bands corresponding to more frequent visitation events.

The subset of species monitored for phenology was phylogenetically clustered relative to the full herbaceous community. Faith’s phylogenetic diversity was significantly lower than expected under a null model of random species draws (SES-PD = -2.51, p = 0.022). This clustering was primarily driven by deep phylogenetic structure, as indicated by a significantly negative mean pairwise distance (SES-MPD = -2.67, P = 0.020), whereas clustering among closely related taxa was weak and not statistically significant (SES-MNTD = -1.92, p = 0.068).

The Mantel correlation between the plant pollinator‐niche distance matrix and the phenological distance matrix was weak and negative (r = -0.076) and not statistically significant (p = 0.748), indicating that plant species that are more similar in their pollinator assemblages are not necessarily more similar in flowering phenology, suggesting that phenological overlap does not structure pollinator niche similarity among the studied plant species. Consistent with this pattern, analyses of phylogenetic signal revealed that flowering phenology exhibited significant phylogenetic structure, with Pagel’s λ = 1.06 (p = 0.04) and Blomberg’s K = 0.96 (p = 0.036), indicating that closely related species tend to flower at similar times. In contrast, pollinator niches showed little evidence of phylogenetic conservatism. At the species level, Schoener’s D overlap among plant species and their insect visitors yielded a Mantel correlation of r = 0.087 (p = 0.196), while functional-group-level niche overlap gave a Mantel correlation of r = -0.072 (p = 0.74).

Floral traits did not explain variation in pollinator assemblage composition among plant species. PERMANOVA showed no significant effects of flower colour (R² = 0.137, p = 0.780) or floral symmetry (R² = 0.048, p = 0.973), with most variation remaining unexplained (residual R² = 0.802). Consistent with this, Mantel tests showed no significant association between floral trait similarity and pollinator assemblage dissimilarity (r = -0.044, p = 0.654), and this pattern remained unchanged after controlling phylogenetic relatedness in partial Mantel tests (r = -0.043, p = 0.639).

Overlap in flowering phenology was also not associated with similarity in pollinator assemblages among plant species. Mantel tests revealed no significant relationship between phenological dissimilarity and pollinator assemblage dissimilarity (r = -0.090, p = 0.776), and this result remained consistent after accounting for phylogenetic relatedness (partial Mantel r = -0.110, p = 0.832).

Non-random synchrony in flowering among co-occurring species likely enhances pollinator attraction and retention, particularly in species-rich or environmentally stressful systems where pollinator availability is temporally limited (Rathcke and Lacey 1985; Morellato et al. 2010). In open and resource-limited environments, synchronized flowering can produce a “magnet effect”, increasing floral display size and visitation rates while balancing competition among co-flowering species (Mitchell et al. 2009). Phenological synchrony varies among functional groups, with aggregation in certain floral colours potentially reflecting shared pollination niches or pollinator sensory biases (Chittka et al. 1994). In contrast, traits such as floral symmetry appear less influential on temporal overlap, consistent with evidence that flowering synchrony is shaped more by pollinator availability, abiotic constraints, and shared environmental cues than by individual floral traits (Ollerton et al. 2015; CaraDonna et al. 2017). These patterns suggest that flowering synchrony emerges from interactions between community-level pressures and trait syndromes rather than uniform trait convergence.

Flowering in herbs of Restinga ecosystems and other open plant communities often contrasts with the positive correlation between daylength and temperature observed in many tree species (Cesário and Gaglianone 2008; Ramírez 2009; Alves et al. 2021). In stressful herb-dominated environments, reproductive activity tends to precede the annual maxima of temperature and daylength, likely reflecting constraints imposed by sandy, nutrient-poor soils with low water retention (Scarano 2002; Ramírez 2004). Early-season heat can exacerbate water limitation, affecting carbon allocation to floral development and reducing overall reproductive output (Scarano 2002). These lagged phenological responses underscore the role of historical climatic conditions interacting with edaphic stress in shaping reproductive strategies (Borchert et al. 2005, 2015; Pau et al. 2011; Diez et al. 2012; dos Santos-Neto et al. 2026).

Environmental filtering strongly shapes plant–pollinator networks in Restingas, producing sparse yet structured interactions where generalist pollinators link specialized plants, buffering communities against temporal fluctuations in floral resources (Bascompte et al. 2006; Olesen et al. 2007; Freitas et al. 2009). In herbaceous and subshrub communities, patchy and short-lived flowering favours flexible pollinators, reducing competition-induced exclusion and promoting coexistence among species with overlapping requirements (Fonseca et al. 2015). Nestedness and high interaction specialization likely enhance network resilience by concentrating interactions among generalist pollinators while permitting specialized species to exploit subsets of this interaction core (Bascompte et al. 2006, Thébault and Fontaine 2010). Such structures facilitate temporal and functional niche partitioning, supporting the stability of pollination services and species coexistence in dynamic coastal ecosystems. Within this framework, our results suggest that even when species flower synchronously, pollinator sharing is not deterministic and instead reflects flexible redistribution of pollinator visits across co-flowering plants, consistent with a dynamic balance between competition and facilitation processes in the interaction network.

Phylogenetic structure reflects how evolutionary history influences species coexistence in Restingas. Deep phylogenetic clustering suggests that conserved traits, such as floral morphology, reward type, or drought tolerance, act as filters determining which lineages persist under harsh environmental conditions (Webb et al. 2002; Cavender-Bares et al. 2004). Divergence among close relatives in functional traits or interaction niches reduces direct competition and allows coexistence within lineages (Vamosi et al. 2014). The combination of deep phylogenetic constraints and shallow-scale divergence helps explain the persistence of multiple species with overlapping ecological requirements, highlighting the role of evolutionary history in structuring plant communities under environmental stress.

Temporal overlap in flowering does not appear to structure pollinator assemblages, indicating that simultaneous flowering does not necessarily lead to shared pollinator interactions (Elzinga et al. 2007). Pollinator assemblages are context-dependent, influenced by local conditions, availability, and floral traits rather than flowering time alone (Ollerton et al. 2006; CaraDonna et al. 2017). Flowering phenology, however, exhibits strong phylogenetic conservatism (Pagel’s λ, Blomberg’s K), suggesting that reproductive timing reflects shared evolutionary history interacting with environmental filters (Staggemeier et al. 2010; Silva et al. 2011; Brito et al. 2017; Santos de Oliveira et al. 2021). However, the relative importance of phylogenetic constraints and environmental drivers appears to vary among systems. For example, reproductive phenology of closely related Myrcia species in a Brazilian coastal forest was primarily explained by environmental variation rather than phylogenetic relatedness (dos Santos-Neto et al. 2026), indicating that environmental filtering can sometimes override evolutionary constraints on flowering schedules. In contrast, pollinator interactions are evolutionarily labile, with closely related species often engaging with distinct assemblages depending on local conditions (Rezende et al. 2020). This contrast reinforces that different ecological dimensions are subject to varying degrees of evolutionary constraint. Future studies with broader taxonomic and geographic sampling are necessary to disentangle the relative roles of phylogenetic constraints and ecological filtering in shaping reproductive phenology and pollinator assemblages.

Importantly, our additional analyses showed that neither floral traits nor flowering phenology overlap explained variation in pollinator assemblage composition, reinforcing the idea that pollinator niche similarity is largely independent of both intrinsic floral characteristics and temporal co-flowering structure.

Despite the lack of a direct relationship between flowering overlap and pollinator assemblage similarity, our results suggest an important decoupling between temporal co-flowering and realized pollinator niches. In systems where multiple species flower simultaneously, pollinator sharing may be structured not by phenological similarity per se, but by a balance between competition and facilitation among co-flowering species. On one hand, synchronous flowering can enhance pollinator attraction through increased floral display and density-dependent visitation (i.e. facilitation or “magnet effects”), potentially increasing the pool of available pollinators at the community level (Rathcke 1983; Ghazoul 2006). On the other hand, such overlap may intensify interspecific competition for pollinators, promoting niche partitioning through differences in floral traits, spatial structure, or pollinator behaviour (Mitchell et al. 2009).

Our finding that floral traits did not significantly predict pollinator assemblage composition further suggests that such partitioning is not strongly driven by the measured traits (flower colour and symmetry) but may instead emerge from unmeasured floral attributes such as reward quantity or scent, or from flexible foraging by generalist pollinators (Chittka and Thomson 2001; Ollerton et al. 2006). These patterns support a scenario in which co-flowering species share temporal windows but diverge in realized pollinator interactions, maintaining coexistence through a combination of weak facilitation at the community level and fine-scale niche differentiation in biotic interactions.