Most insect pollinator species use floral resources exploited by many conspecifics and heterospecifics. In systems where floral resources limit pollinator populations and give rise to competition, understanding whether competition varies across the growing season is a fundamental question in explaining how species coexist and how communities are structured. Recent work on nectar depletion rates in plant-pollinator communities has demonstrated the value of measuring nectar depletion rates as a proxy for competition among pollinators. One hypothesis on pollinator competition, hereafter, “equilibrium hypothesis”, predicts that nectar depletion rates should be constant across the growing season within a given community. In contrast, an alternative hypothesis, hereafter, “dynamic hypothesis” suggests the opposite: that resource depletion rates vary across a growing season, with floral resource supply and demand being asynchronous through time. To test these hypotheses, we sampled nectar depletion and floral abundance across a complete plant community over the bulk of a growing season in central Kentucky, USA within an early successional old field community in spring/summer, 2023. Initially in the growing season, nectar depletion rates within this early successional vegetation community were moderate (e.g., 50-70%) but steadily climbed to 85% as the growing season advanced. Plant species varied widely in depletion rate: invasive purple deadnettle Lamium purpureum being the least depleted (48% depleted) and native common milkweed Asclepias syriaca the most (95% depleted). After correcting for species-specific abundance, Vicia sativa and Lonicera japonica produced the greatest quantity of nectar over the first seven weeks, after which time Asclepias syriaca and Monarda fistulosa yielded a 2-3x increase in available nectar for pollinators. These data add to a growing body of literature suggesting that the dynamic hypothesis explains patterns of competition in temperate pollinator populations.

Study area and sampling design. We monitored plant and insect communities within the Inner Bluegrass region of Central Kentucky, USA. This area falls within the Interior Plateau ecoregion of the southeastern United States (Drummond, 2016; Sayler et al., 2016). The Interior Plateau ecoregion is dominated by a mix of rolling hills and flat plains, being substantially flatter than the Appalachian region to the east but still punctuated by occasional dissected uplands and deeply cut valleys/rivers (Drummond, 2016; Karathanasis, 2018). The soils are largely alkaline, due to extensive underlying limestone (Karathanasis, 2018). Within the Inner Bluegrass region, we studied an early successional old field community at the University of Kentucky’s 24-ha Ecological Research and Education Center (EREC) in Lexington, Kentucky (38.081°N, -84.477°W, WGS84; Figure 1A). This community is kept in a state of early succession via periodic mowing and brush-hogging to inhibit woody vegetation from advancing in size toward a forested state. The EREC preserve occurs within a strongly suburban context with human development (primarily residential homes and businesses) occurring on all sides (Fig. 1A).

            Across the EREC preserve, we plotted a 50-m x 50-m grid of points in R version 4.2.1 (R Core Team 2022) with the rgdal and raster packages (Bivand et al., 2022; Hijmans, 2022). We then masked this grid using a shapefile of the EREC boundary which yielded 103 potential sampling locations. Finally, we assessed each location in the field to ensure feasibility of vegetation sampling (e.g., some points were in standing water or parking lots). The final sample was 90 points, systematically distributed across the EREC. At each of these 90 locations, we marked a transect that was 50m in length (25 m north and south) and 2 m in width. At these 90 locations, we surveyed 45 each week to conduct flower sampling which gave us a biweekly assessment of floral abundance across the study area. At each 50-m x 2-m transect, we conducted a visual count of every flower species and tallied the number of open flowers by species (Mathis et al., 2022). This produced an absolute density of each species at each site, per each visit.

Floral and nectar sampling. Since we sampled points on most days of the sampling period, we attempted to extract nectar from all blooming species in the study area. We made attempts to extract nectar from bagged individuals of each species on five occasions on different days. If nectar could never be extracted, it was dismissed from future nectar consideration. For species where nectar was present and could be sampled, we sampled individuals at random locations each day. For each day of nectar sampling, we sampled 1-5 pairs of bagged and unbagged flowers on each focal species. Bags attached were made of fine mesh bags that excluded all insect pollinators but did not restrict airflow or sunlight (Sponsler et al., 2024a). The bagged individual within each pair was assumed to have the “full” amount of nectar (no depletion) while the un-bagged individual within each pair was left open for pollinator consumption. Bags were deployed in the morning (before 0900) and collected after approximately 24 hours. Nectar was extracted using microcapillary tubes (0.25 and 5.0 microliter). The length of the nectar column was then measured with digital calipers to calculate nectar volume extracted using known volume of the cylindrical microcapillary tubes. 

Statistical analysis. Following Sponsler et al. (2024a, b), we quantified nectar depletion as the proportional difference between the mean nectar volume of bagged and open flowers. Prior to modeling, we normalized our measurements by dividing each volume reading by the mean volume of the bagged flowers within each species and sampling date. This step served to simplify prior selection (see below) and dampen variation extraneous to the calculation of depletion rate.

To obtain unbiased estimates of mean nectar depletion while respecting the structure of our sampling design, we used a hierarchical Bayesian Gamma regression model with treatment specified as a constant effect and day, species, and plant specified as varying slope and intercept effects (following the nomenclature of Gelman 2005). Proportion of nectar available (i.e. inverse of depletion) was used as the response variable for modeling efficiency. To accommodate the fact that nectar volume measurements are continuously positive but open-bounded at zero, we included a binomial (i.e. hurdle) submodel with the same specifications as the Gamma model to handle zeros. Priors were selected to be regularizing and weakly informative (Wesner and Pomeranz 2021) following the workflow of Gabry et al. (2019). The model was implemented with Stan (Stan Development Team, 2022), accessed via the package brms (Bürkner, 2017) in R v. 4.3.2 (R Core Team, 2023). Four chains of 5,000 iterations were fit via Hamiltonian Monte Carlo (No U-Turn Sampler) and included an initial 2,500 warmup for each chain, yielding 10,000 posterior samples. After assessing model convergence using trace plots and Gelman-Rubin statistics (R-hat) and evaluating model fit using a graphical posterior prediction check to compare empirical and modeled density functions, we used the posterior distributions of nectar availability to calculate the posterior distribution of the mean depletion rate (depletion rate = 1 - availability rate).

All other data operations were done in R. Data handling was performed with the tidyverse package suite (Wickham et al., 2019). Model predictions were extracted using tidybayes (Kay, 2021b), and models were visualized using ggplot2 (Wickham, 2016), see (Lüdecke et al., 2021), and ggdist (Kay, 2021a). For further description of statistical methods, see Sponsler et al. (2024a ,b).