Influence of environmental covariates on pollinator community occupancy, detection, and richness across urban gardens in Richmond, Virginia (U.S.A.)

Pollination is an essential ecosystem service that supports reproduction and propagation of most of the world’s flowering plants. The dramatic decline in pollinators, especially insect pollinators, due to climate change and pesticide use threatens not only our food supply, but also the diversity of native plants. Urban areas, if well managed, can serve as corridors and reserves for pollinator species and benefit agricultural and natural ecosystems well beyond the urban environment. In this study, we assessed Mid-Atlantic (U.S.A.)-region urban garden plant-pollinator interactions, focusing on activity associated with two regionally-native plants: dense blazing star (Liatris spicata; Asteraceae) and clustered mountain mint (Pycnanthemum muticum; Lamiaceae). We conducted 350 visual surveys across 52 gardens and identified 14 taxa in 361 detection events, with 5 taxa dominating at 331 detections. We built multi-species occupancy models (MSOMs) in a Bayesian framework using site and survey covariates to evaluate variables that influenced species occupancy, detection, and richness. We found little influence of any variables on occupancy, and the intercept-only models resulted in species-specific occupancy that ranaged from 0.04 (Halyomorpha halys) to 0.86 (Halictidae). For detection, we found that plant species and survey start time (or the interaction between the two) influenced detection of a majority of pollinators at the community level, while Julian date and urban distance (interaction) influenced a small number of species. Comparisons between the two plant species indicated that honey bees (Apis mellifera) and wasps (Vespoidea) were more likely to be detected on P. muticum compared to L. spicata, while the reverse was true for A. campestris. All taxa became more detectable as it became later in the day. A. mellifera and Bombus spp. had higher detection earlier in the year. Halictidae detections increased closer to the urban areas, while Bombus spp. detection increased farther from urban areas. The posterior medians of the number of taxa per site ranged from 5 – 8 and showed little evidence of differences across sites, but the composition did vary. The estimated number of taxa occurring across all sampled sites was 18, indicating that ~25% of taxa present at our study sites went completely undetected. Our study demonstrates that MSOMs can be an effective tool for monitoring and investigating the pollinator community. We were able to estimate occupancy for 14 observed insect taxa, 9 of which were detected fewer than 8 times. We also estimated effects of detection covariates that impacted multiple taxa and provide insight into ways to improve future pollinator monitoring efforts. These findings further our understanding of how plant species and the urban setting may variably influence pollinator activity and highlight the importance of urban gardens in supporting divserse insect communities.

Methods
Data are the result of visual pollinator surveys conducted in the greater Richmond, VA area at 50 gardens (sites) from June 28 - July 27, 2021. At each garden, observations were conducted on Liatris spicata or Pycnanthemum muticum, or both. For a given plant type, 5 separate inflorescences on different plants (spatial replicates) were observed for 5 minutes. Pollinator visits were tallied and identified to the lowest taxonomic level possible. 20 gardens were surveyed on both plant types (but different days for each plant type), for a total of 10 spatial replicates, while 20 other gardens were surveyed only on P. muticum (5 spatial replicates) and 10 gardens were surveyed only on L. spicata (also 5 spatial replicates).

Potential explanatory variables (covariates) were measured, which can broadly be divided into two types. Site covariates did not vary over time and were consistent across all spatial replicates at a given site, whereas survey covariates could vary across spatial replicates at a given site. Survey covariates can be further distinguished between values that were only measured once per site visit (thus they would differ between plant types, but not between spatial replicates on the same plant type) vs those for which unique values were measured for every spatial replicate.

Data were analyzed in a multi-species occupancy modeling framework, as demonstrated in the included R scripts (see Code/Software section).