Effects of season of fire on bee‐flower interaction diversity in a fire‐maintained pine savanna

Abstract Whereas the Coastal Plain of the southeastern United States historically experienced fire primarily during the mid‐summer lightning season, managers today typically apply prescribed fire during the late winter or early spring months. The ecological implications of this discrepancy remain poorly understood, especially with regard to pollinators and their interactions with flowers. In a replicated field experiment, we compared the abundance and richness of bees and bee–flower interactions among pine savanna plots in Florida that were burned either during the winter, spring, summer, or fall. We netted 92 bee species from 77 species of flowers, representing 435 unique bee–flower interactions in total. When analyzing the results from each month separately, we detected significant short‐term reductions in the number of bees and bee–flower interactions following fires regardless of season. Although bee abundance and richness did not differ over the entire season, bee–flower interaction richness was significantly higher overall in spring and summer plots than in fall plots and the composition of both bees and bee–flower interactions differed significantly among treatments. Several bee–flower interactions were significantly associated with one or more of the treatments. Some of these associations could be attributed to differences in flowering phenology among treatments. Taken together, our findings suggest that season of fire has modest but potentially important implications for interactions between bees and flowers in southeastern pine ecosystems. Because most flowering plants within our study region are pollinated by a variety of bees and other insects, and most bees endemic to the region are polylectic, season of fire may not be very important to either group overall. However, the timing of fire may be more important to particular species including certain flower specialists and fire‐sensitive taxa such as butterflies. Future research targeting such species would be of interest.

studies, to our knowledge, have investigated how pollinators are affected by the timing of prescribed fire.
In a longleaf pine savanna in Georgia, Hiers et al. (2000) compared the effects of prescribed fire during the winter/early spring vs. the summer on the flowering phenology and fruit initiation (i.e., a surrogate for pollination rates) of legumes. They found season of fire to significantly alter flowering phenology but found no differences in fruit initiation between treatments. This latter finding, combined with the presence of bees capable of pollinating legumes during all flowering periods, suggests that pollinators and pollination may be largely unaffected by fire-mediated differences in flowering phenology. Most recently, in Florida, Adedoja et al. (2022) found no significant differences in the number of pollinator visits observed on flowers among plots that had been burned during the "winter-dry" (late January to mid-March), "spring" (late May to mid-June), or "summer-wet" (August) periods. This was true both the year after the plots were burned and during the year of the burns.
However, season of fire did affect flower density in that study, and these effects varied depending on time since fire. For example, summer-wet burns reduced flower density during the year of the fire but increased the density of flowers the following year when no burns took place.
Both of these previous studies suggest that season of fire does not strongly affect pollinator abundance and visitation despite altering the phenology and density of flowers. However, little is known about species-specific responses to season of fire. It is possible that some taxa are more sensitive to season of fire than others due to differences in activity periods and resource requirements among species. Most bee species native to the southeastern United States are polylectic (Folkerts et al., 1993), meaning they visit a wide range of host plants. However, some taxa are highly host specific and these species may be more sensitive to fire during particular times of year, especially if it results in a phenological mismatch between their activity and host-flower availability. Given limited knowledge of how season of fire affects the timing of flowering, and how this may differ among plant species (Brown et al., 2017), such phenological mismatches are difficult to predict. The current study sought to more fully explore such questions by investigating the effects of season of fire (winter, spring, summer, and fall) on specific bee-flower interactions in a Florida pine savanna.

| Study area and experimental design
This study took place at Tall Timbers Research Station in Leon County, Florida ( Figure 1). The site had been fire-excluded for some time prior to being acquired by Tall Timbers in 1990, after which broadleaf hardwood trees were cleared, leaving widely spaced shortleaf pine (Pinus echinata Mill.) in the overstory. Shortly after acquisition, eight 0.4 ha (one-acre) plots, ultimately the experimental blocks used in the current study, were treated with various herbicides as a site-preparation experiment and planted with longleaf pine. Today, the herbaceous plant community is indistinguishable from the surrounding vegetation, such that the herbicide had no noticeable long-term effects (KR, personal observation). The study site has no history of agriculture and is dominated by little bluestem (Schizachyrium scoparium (Michx.) Nash) and resprouting native upland tree species top-killed by fire, including mockernut hickory (Carya tomentosa (Lam. ex Poir.)), southern red oak (Quercus falcata Michx.), post oak (Q. stellata Wangenh.), white oak (Q. alba L.), and black oak (Q. velutina Lam.). Since acquisition by Tall Timbers, and prior to 2021 when the current treatments were initiated, the study site was treated with biennial prescribed burns in March-April. The three northern-most and five southern-most blocks used in this study were last burned this way in 2020 and 2019, respectively.

| Bee sampling
Two of us (MDU and SH) sampled bees in each plot once a month from February to November 2022, with the first sampling occurring several weeks after the winter burn treatment (Figure 2). We used nets to sample bees from flowers, collecting only specimens judged to be actively foraging for pollen or nectar. Because we were specifically interested in bee-flower associations, bees flying near flowers or sitting on the leaves or petals of flowering plants were not collected. We identified flowers on site or from photographs with the help of botanists familiar with the local flora (KR and CD). The process for sampling bees followed a standardized protocol. Collectors 1 and 2 started in the southeast and northwest plot within each block, respectively, before moving to the next plot in the clockwise direction. Consequently, the two collectors never sampled the same plot at the same time. Within each plot, collectors 1 and 2 walked a diagonal transect from the F I G U R E 1 Map of study plots on Tall Timbers Research Station, Florida, USA. Light and dark green areas represent fire-maintained open pine forests and unburned hardwood-dominated forests, respectively.

F I G U R E 2
Timeline showing when bee collections and fires (above and below the line, respectively) took place in 2022. Note that multiple vertical lines reflect when collections or burns took place over multiple days. See text for specific dates, details about sampling effort by month, and for information on fires from previous years.
southwest to northeast corners and from the northwest to southeast corners, respectively. All flowers present within a net length (~1.5 m) of either side of the transect were inspected for foraging bees. After completing the transect, the collectors walked the perimeter of each plot and then, as time allowed, searched for flowers elsewhere within the plot. Sampling took place only during sunny or partly cloudy weather between the hours of 9:30 am to 5:30 pm. We ensured that each collector sampled all plots within each block consecutively to minimize the effects of weather conditions and time of day on bee activity. Each collector spent 20 min (Feb, Mar, Aug, and Nov) or 25 min (Apr, May, Jun, Sep) sampling bees in each plot. In July, due to stormy weather, the plots within the four northern-most blocks were each sampled for a total of 40 min while the others were sampled for only 30 min.
Voucher specimens have been deposited in the first author's research collection.

| Analysis
Unless otherwise stated, all analyses were conducted in R 4.2.1 (R Core Team, 2022). We pooled data by plot and month prior to analysis. Because only three of the summer plots had been burned prior to the June sampling, we included only the five unburned plots in the analysis for that month. We calculated the total bee-flower interaction richness, bee richness, and bee abundance for each plot by month. We calculated the same metrics after combining data across months. These monthly and combined responses were compared among treatments using generalized linear mixed effects models with treatment as the fixed effect and block as the random term. We used Poisson models for bee-flower interaction richness and bee richness and negative binomial models for bee abundance to mitigate overdispersion (identified using the dispersion_glmer function). For multiple comparisons of season of fire treatments, we used the ghlt function of the multcomp package (Hothorn et al., 2016). Because the winter plots yielded no bees in February, the Kruskal-Wallis rank sum test was used to compare treatments for that month. Pairwise comparisons were made using the Benjamini-Hochberg method to adjust p-values.
To test whether bees and bee-flower interactions differed compositionally among the season of fire treatments, we conducted non-metric multidimensional scaling followed by PERMANOVA in PC-ORD (McCune & Mefford, 2011). We pooled data across sampling periods when constructing the matrices. Then, to test which bee species or bee-flower interactions were strongly associated with one or more season of fire treatments, we conducted indicator species analysis using the function multipatt (multilevel pattern analysis) in the package indicspecies (Cáceres & Legendre, 2009) to produce indicator values ranging from 0 (no association) to 1 (complete association).
Finally, to estimate the total bee richness at our study site, we calculated the Chao1 estimator using the rareNMtests package (Cayuela & Gotelli, 2014) which gives a richness estimate with 95% confidence intervals based on the list of collected species and their abundances.
After combining data across sampling periods, we found no significant differences in the richness or abundance of bees among season of fire treatments (Figure 4). However, there were, on average, about 10-11 fewer bee-flower interactions detected in fall plots than in spring or summer plots, a significant difference ( Figure 4).
When we analyzed data from each month separately, we detected several significant differences between treatments. In February, following winter burns, for instance, we found lower bee-flower interaction richness, bee richness, and bee abundance in the fall and especially winter plots compared to the spring and summer plots ( Figure 5). However, no such differences were observed 1 month later. In April, after the spring burns, both bee-flower interaction and bee richness were significantly lower in spring than summer plots but there were no significant differences between the spring and fall or winter plots. In May, we found bee abundance to be significantly higher in the spring plots than in the winter plots. Otherwise, there were no other significant differences among treatments until October, following the fall burns, when bee-flower interaction richness, bee richness, and bee abundance were all significantly lower in the fall plots than in the other treatments. The same pattern was detected in November.
NMDS based on bee data yielded a one-dimensional solution (not shown). PERMANOVA revealed significant compositional differences in bee communities among the season of fire treatments  Table 2). A total of 10 bee-flower interactions were found to be significantly associated with one or more treatments based on indicspecies analysis ( Table 3). Six of these associations involved spring or summer plots while four involved fall or winter plots (Table 3).

| DISCUSS ION
The purpose of this study was to better understand how season of fire affects bees and bee-flower interactions in pine savannas on the southeastern US Coastal Plain. We found no difference in the total richness or abundance of bees among treatments across the entire season. However, bee-flower interaction richness was significantly lower in fall plots than in spring or summer plots and we detected significant differences in the composition of both bees and beeflower interactions among treatments.
We observed stronger effects of season of fire when we analyzed data from each month separately. Results from February, April, October, and November all confirm that fire greatly reduces or eliminates floral resource availability and associated pollinator activity. However, TA B L E 1 Bee taxa with the widest observed host range in this study ranked by the number of flower records (left) and flower taxa ranked by the number of collected bee species (right). these effects appear to be quite ephemeral, disappearing within 1-2 months due to the rapid recovery of plants following a fire. We even found evidence that fire can trigger flowering by some plant species with resultant increases in bee numbers relative to less-recently burned plots. In May, for example, just 5 weeks after the spring burns, we found bee abundance to be significantly higher in spring plots than in winter plots. This was driven by a fire-induced flush of Tephrosia virginiana from which we collected large numbers of bees. does not appear to be very ecologically significant in our system.

Rank
Most bee species within our study area are polylectic (Folkerts et al., 1993), meaning they collect pollen from a wide variety of unrelated plants. Similarly, many of the flower species we sampled were visited by numerous species of bees in this study. Thus, fire-driven phenological mismatches between bees and flowers are probably largely inconsequential for both parties in this region. Not one of the species pairs found to be significantly associated with one or more of the seasons of fire in this study consists of species belonging to specialist bee-flower relationships. However, as discussed below, we should not be too hasty in concluding that season of fire is unimportant to pollinators in southeastern US pine savannas.
It should be stressed that because the spring, summer, and fall burns did not all cover 100% of the plots, flowers were often available in these plots even immediately after burns. We would likely have detected stronger effects of season of fire if these plots had over multiple decades and involve mostly changes in species relative abundance rather than species composition (Glitzenstein, 2003;Smith et al., 2013;Towne & Owensby, 1984;Van Wyk, 1971). Finally, it is possible that season of fire has a stronger effect on taxa not considered here. Butterflies, for example, are generally more exposed to fire during their immature stages because of their use of plant hosts and are more sensitive to fire as a result (Carbone et al., 2019).
Season of fire has important implications for some of these species (Jue et al., 2022).
A particularly unique aspect of this study is the inclusion of fall fires, which are barely represented in the fire ecology literature pertaining to southeastern US pine savannas (Lewis, 1964;Platt et al., 1988). Although the early October burns were within the growing season, we observed little to no regrowth of perennial vegetation in burned areas until the following growing season, consistent with observations of growing season fall burning in a tallgrass prairie (Engle et al., 2000).  savannas (Lewis, 1964;Platt et al., 1988).

This work was funded by the USDA Southern Research Station and
Tall Timbers Research Station. We are grateful to Jim Cane for providing advice on standardized bee sampling. We also thank the Associate Editor and two reviewers for comments that greatly improved the manuscript.

CO N FLI C T O F I NTER E S T S TATEM ENT
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The dataset produced by this study is available at https://doi.

A PPEN D I X A
List of bee species collected in this study. With the exception of Apis mellifera, numbers reflect the total abundance of each species by month (left columns) and treatment (right columns). Data for A. mellifera represent the sum of flower records across plots as it was not possible to fully capture the abundance of this species at certain times of the year.

A PPE N D I X B
List of flower species from which bees were collected in this study. Data reflect the total number of bees collected from each species by month.