Online Resource 2. Comparison of insect visitation to tarweed flowers: video footage versus direct observation
We used video cameras (GoPro™ Hero 3) to record insect visitation to Deinandra fasciculata at control plots in the Apis-removal experiment. During observation periods in each control plot, we set up video cameras on an angled 45-cm tripod at the edge of the plot, such that the camera lens was no more than 70 cm away from the farthest capitula in a 50 cm x 50 cm plot. Although video footage appeared to be of high quality, we explicitly tested the ability of cameras to record insect visitation to flowers by comparing direct observations of floral visitation with video footage taken simultaneously at the same plots. We performed such comparisons using 60 min of observation in each of ten 50 x 50 cm plots containing approximately 20 Deinandra fasciculata individuals. The same observer (AJN) conducted the field observations and also transcribed the video recordings. Video transcription was conducted blind and approximately one month after observational data were collected in the field.
Comparisons of visitation between records based on direct observation and those based on video footage revealed that videos provided a reliable record of Apis visitation. Direct observations recorded 15.3 ± 3.6 (mean ± SE) Apis per 60-min trial, whereas video recordings recorded 14.8 ± 3.3 Apis per 60-min trial (paired t-test: t = 1.25, df = 9, P = 0.24). Based on the consistency of these measurements, we use video footage as a record of the rate of Apis visitation to control plots as well as the number of capitula contacted per Apis visit.
Video footage was a less useful means to quantify visitation by non-Apis insects; many of these insects are smaller and move faster compared to honey bees. Direct observations recorded 1.8 ± 0.5 (mean ± SE) non-Apis visitors per 60-min trial, whereas video recordings recorded 1.3 ± 0.4 non-Apis visitors per 60-min trial (paired t-test: t = 1.63, df = 9, P = 0.14). Given that videos captured only 72% of non-Apis visits that were recorded by direct observation, we do not use non-Apis visitation data from control plots in any of our main analyses.
Although video recordings of visitation at control plots did not capture all non-Apis visitors, the removal method itself did not appear to decrease visitation by other insects. Raw mean counts of non-Apis visitors to removal plots (3.99 ± 1.13 visits per hour) may have been slightly higher compared to raw counts of non-Apis visitors on control plots (2.26 ± 0.75 visits per hour) (Wilcoxon signed- rank test: W = 79.5, P = 0.07). When we accounted for non-Apis visitors that might have been missed in video recordings of control plots by multiplying non-Apis visitation in all control replicates by the estimated amount of visitation missed by the videos, non-Apis visitation in control plots (3.14 ± 1.05 visits per hour) did not differ from that in removal plots (W ilcoxon signed-rank test: W = 102, P = 0.34).