Fig 1.
Visual representation of the plant-pollinator networks tested (a) and the hypothesized response of each network metric against honey bee abundance (b). a) Plant-pollinator interaction data from each of 19 sampling transects (left), four datasets were created (right): i) a full season all taxa dataset; ii) a mid-season all taxa dataset; iii) a full season bees-only dataset, where only bee interactions, including honey bee interactions, were kept from the full season network; and iv) a without-honey bees dataset, where only honey bee interactions were removed from the full season network. b) Predictions of the effects of increasing honey bee abundance on each network metric, for six metrics of resource overlap and six metrics associated with network stability. Yellow trend lines show predictions for metrics of total network structure, whereas blue trend lines show predictions for metrics calculated from just wild pollinator-plant interactions, excluding honey bee interactions. The rationale and references to support each prediction are in the Supplementary Information. Insect images are from SVG SILH [30].
Fig 2.
The research location, Mattheis Ranch, in southern Alberta, Canada.
Sampling transects are shown as circles and honey bee hives as diamonds. Each apiary contained a cluster of hives (northernmost: 48 hives, central: 32 hives, southernmost: 16 hives). The transect G5000 indicates the new location established when sampling was discontinued at F5000 after July 9th, 2019. Land outside the Mattheis Ranch is managed by the Eastern Irrigation District. Map shapefiles from Natural Resources Canada under an Open Government license [43]. Map created in QGIS [44].
Fig 3.
Plant-pollinator meta-networks from the full season all taxa dataset.
Data from transects 100 m (top), 500 m (middle) and 5000 m (bottom) from honey bee hives are pooled across the season. The upper row depicts the pollinator species, coloured by their given taxonomic “group” or Order (see S6 Table for the list of pollinator species). “Moths” are Lepidoptera excluding butterflies (Papilionoidea). “Wasps” are Aculeata excluding ants (Formicoidea) and bees (Anthophila). The bottom row (in black) depicts the plant species (see S7 Table for the list of plant species). The width of each upper and lower bar represents the relative frequency of interactions observed for that species. The interactions between plants and wild pollinators are represented by blue lines; width indicates the frequency of the interaction. Honey bees and their interactions are indicated in red.
Fig 4.
Predictions vs. results of network metric analyses from with-honey bees networks.
(a) and without-honey bees networks (b). a) Relationships between honey bee abundance and the network metrics: (i) pollinator functional complementarity, (ii) plant functional complementarity, and (iii) interaction evenness that were significantly related to honey bee abundance across three datasets (full season all taxa, mid-season all taxa, and full season bees-only). Solid regression lines indicate significant relationships with Bonferroni-Holm correction, whereas dotted lines indicate insignificant relationships. b) Relationships between honey bee abundance and the network metrics: (i) pollinator niche overlap, (ii) pollinator functional complementarity, (iii) plant functional complementarity, (iv) weighted connectance, and (v) interaction evenness in the without-honey bees dataset. Dashed regression lines indicate non-significant relationships with Bonferroni-Holm correction. Although (i) and (iv) appear to contain trends, these cannot be unambiguously attributed to honey bee abundance, as they can be explained by flower community variables and collection effort (Table 2D) that were somewhat correlated with honey bee abundance (S4 Table).
Table 1.
Results of the simple linear regression (SLR) model with the lowest AICc values for each response variable, where the full model for each response variable contained honey bee abundance as the only predictor variable.
Table 2.
Results of the multiple regression (MR) model with the lowest AICc value for each response variable.