Background and Aims Ornamental flowering plant species are often used in managed greenspaces to attract and support pollinator populations. In natural systems, selection by pollinators is hypothesized to result in convergent multimodal floral phenotypes that are more attractive to specific pollinator taxa. In contrast, ornamental cultivars are bred via artificial selection by humans, and exhibit diverse and distinct phenotypes. Despite their prevalence in managed habitats, the influence of cultivar phenotypic variation on plant attractiveness to pollinator taxa is not well resolved.

Methods We used a combination of field and behavioural assays to evaluate how variation in floral visual, chemical and nutritional traits impacted overall attractiveness and visitation by pollinator taxonomic groups and bee species to 25 cultivars of five herbaceous perennial ornamental plant genera.

Key results Despite significant phenotypic variation, cultivars tended to attract a broad range of pollinator species. Nonetheless, at the level of insect order (bee, fly, butterfly, beetle), attraction was generally modulated by traits consistent with the pollination syndrome hypothesis. At the level of bee species, the relative influence of traits on visitation varied across plant genera, with some floral phenotypes leading to a broadening of the visitor community, and others leading to exclusion of visitation by certain bee species.

Conclusions Our results demonstrate how pollinator choice is mediated by complex multimodal floral signals. Importantly, the traits that had the greatest and most consistent effect on regulating pollinator attraction were those that are commonly selected for in cultivar development. Though variation among cultivars in floral traits may limit the pollinator community by excluding certain species, it may also encourage interactions with generalist taxa to support pollinator diversity in managed landscapes.

This dataset contains data on floral traits collections and pollinator visitation in both the laboratory and the field. 

All data was analyzed in R using a combination of mixed effects models and NMDS ordination. For full details of data collection and analysis, please see the associated manuscript. 

All data was processed in R. Data sheets are stored in CSV files.