Materials and methods
Study sites and landscape structure
The study was conducted in 27 farms growing fruit or vegetable crops in Eastern Ontario and the Outaouais region of Québec, Canada. Farms planning to grow cucurbit crops were chosen initially for inclusion because we wished to focus on pollinator-dependent, late-season crops; however, many farms were not able to grow cucurbits due to drought conditions experienced throughout the region. To maximize independence among farm sites (i.e., to minimize the chance that an individual bee could move between farms), chosen farms were 4–211 km apart. Across all farm sites, 102 locations were sampled for bees and FR abundance (as described below), with one to six locations per farm, depending on the number of distinct land patches in which resource-providing flowers were present, and when permission was given from landowners. Sampling locations within patches of land were selected based on the estimated location of the patch’s centre, or, if the patch was over 25 m wide, was located at least 10 m from an edge. In three patches wider than 25 m, sampling locations less than 10 m from the edge were used due to a complete absence of flowers in bloom in the centre. The distance between sampling locations within a farm ranged from 3.8 m to 1040 m. Sites were visited in rotation over four time periods during one season in 2016: the first took place between May 20–June 10 (n = 38 sampling locations), the second from June 10–July 4 (n = 33), the third from July 5–August 1 (n = 37), and the fourth from August 1–September 1 (n = 39). If sampling locations contained open flowers during more than one sampling period, the same location was sampled in multiple time periods.
The composition of the landscape within 250 m, 500 m, and 750 m radii of each sampling location was quantified to estimate landscape-scale FR abundance. The 250–750 m scale has been found in previous studies to be the range at which non-Apis bees respond to landscape structure (Steffan-Dewenter et al. 2002), and 500 m was chosen as an intermediate spatial scale. Sampling locations within the same farm site (and with overlapping radii at the 750 m scale) were not treated as independent (see Statistical analysis). Within a 750 m radius around each sampling location, the boundaries between land patches were manually digitized in QGIS version 2.18.7, using both waypoints taken on-site with a Trimble Juno SD handheld GPS unit (Trimble Navigation Limited, Westminster, CO, USA), and from Google Earth and Bing Aerial satellite imagery.
Each land patch was then categorized by the type of land-use (hereafter, “land type”), through ground-truthing and raster imagery from Agriculture and Agri-Food Canada’s (AAFC) 2016 Annual Crop Inventory. Land types fell into three categories: non-resource land, resource-providing land, and unknown land (see Appendix Table A1, Supporting Information, for detailed descriptions of each land type). Non-resource land was defined as any area that did not provide FR, which included crops with exclusively wind-pollinated flowers and crops with anecdotal or no evidence of bees collecting resources from flowers. Urban and developed land was also included in non-resource land; although urban gardens or lawns may provide FR for bees, the amount is inconsistent over time and space, and the number of bee species supported is usually low (Cane 2005, Matteson et al. 2008). Resource-providing land was defined as land areas that provided FR for bees at some point during the season and was categorized into 14 different land types (Appendix Table A1). Sampling locations were located only within resource-providing land, and at least one of each resource-providing land type was sampled during each time period. Unknown land was comprised of areas where we could not determine the crop grown (2.3% of all area surrounding sampling locations); hedgerow (1.8%); or crop land where potentially resource-providing crops were grown, but FR were not measured (0.7%). The total area of each land type was then calculated within radii of 250, 500, and 750 m around each sampling location.
Bee observations
Bee observation methods were adapted from frequently used pollinator surveying designs (Memmott 1999, Alarcón et al. 2008, Gibson et al. 2011). At each sampling location, a transect was set up to survey bee activity within a 30 m × 4 m area (89 transects); a 30 m × 2 m area was surveyed when only one crop row (< 4 m wide) was present (eight transects); and 25 m × 4 m (one transect) or 24 m × 4m areas (four transects) were surveyed when crop rows were shorter than 30 m. Bee observations occurred over one minute per 4 m2 of transect intervals by slowly walking the length of the transect. The shaded and unshaded temperature, maximum wind speed, and average wind speed were recorded for at least one minute using a Kestrel 2000 Pocket Weather Meter (Nielsen-Kellerman, Boothwyn, PA, USA) held at approximately 1.5 m above ground preceding each observation period. If there was a noticeable change in conditions during the observation period, temperature and wind speed were recorded again at the end of the period and averages were recorded. All bee observations were conducted when shaded temperatures were above 11.9°C, average wind speeds were below 1.9 m/s, and maximum wind speeds were below 4 m/s.
During observation periods, all occurrences of bees visiting open flowers were recorded by two observers, standing on either side of the transect width, and recording all visits within 2 m each. A visit was counted when a bee was seen contacting sexual organs of an entomophilous flower or was probing a flower for nectar. All visited flowers were identified to genus (9 out of 77 taxa) or species (68 out of 77 taxa), and bees were identified on the wing to genus or species. When identification was not possible on the wing, the observations were paused and both observers attempted to catch the bee to take a photograph from inside a glass vial or to collect as a voucher (79 specimens total). Vouchers were then identified to species or genus and are stored in the Forrest lab’s collection at the University of Ottawa (Ottawa, ON, Canada). Overall, 82% of bees were identified to species, 17% to genus, 0.1% to family, and 1% as Anthophila.
Floral resources
Floral density was recorded at each sampling location, using three quadrats of 1.5 m × 1.5 m. Quadrats were placed in random locations within the same transect used for bee observations, immediately following the observation period. If no open flowers were present in all three quadrat locations, an additional location was randomly selected and the mean count across the four quadrats was recorded. Within a quadrat, the number of open flowers was counted for each non-graminoid species encountered; for species with many-flowered inflorescences, five individuals were haphazardly selected, and the number of flowers was counted on a randomly selected inflorescence. The mean number of flowers per inflorescence for many-flowered species was then multiplied by the number of inflorescences in a quadrat to obtain the number of flowers per quadrat. In members of the Asteraceae family, capitula were treated as single flowers (see Appendix Table A2 for descriptions of floral units used for counts of each species). For 29 out of 96 species encountered, the number of flowers per inflorescence was obtained from either literature sources or digital images of herbarium specimens due to the large number of flowers encountered in the field, and for a few species that inadvertently were not measured in the field (see Appendix Table A2 for literature values for each species).
To estimate the amount of FR (nectar and pollen) provided by a species, floral dimensions were measured on five haphazardly selected individuals of each species. The length and width of the receptacle (or capitulum in Asteraceae species) were measured at right angles to each other, as well as the height from the receptacle to the end of the longest sexual organ (stamen or pistil); in species with sexual organs completely hidden within a corolla, height was measured from the receptacle to the end of the corolla. Measurements were made using calipers and were rounded to the nearest 1 mm. Thirty-one of 96 species were not measured in the field, and floral measurements were instead obtained from literature sources or digital images of herbarium specimens (see Appendix Table A2 for measurements and literature sources for each species). Floral measurements were used to calculate both the surface area (A ) of flowers: