Although the Old World sunbirds are generally considered to be an ecological analogue of the New World hummingbirds, it is commonly believed that in contrast to hummingbirds, sunbirds perch while feeding. Nevertheless, recent studies have shown that Old World nectarivores hover while feeding more frequently than previously thought, and some Old World plants seem to have adapted to hovering bird pollinators. To reveal the importance of sunbird foraging behavior in nectar acquisition and to test whether this behavior is determined by plant architecture and rain intensity, we focused on the specialized West African pollination system of Impatiens sakeriana and its two major pollinators, the Cameroon Sunbird (Cyanomitra oritis) and the Northern Double-collared Sunbird (Cinnyris reichenowi). C. oritis hovered more often than C. reichenowi while feeding on flowers, although both species were observed hovering regularly. For both species, hovering reduced the feeding time, but this reduction and the consequent estimated changes in energetic intake were species-specific. We found that both floral pedicel and peduncle length, as well as precipitation, negatively affected the probability of C. reichenowi hovering but did not have any significant effect on the behavior of C. oritis. Our study demonstrates that hovering behavior is common in the studied sunbird taxa and that plant architecture and environmental factors can influence sunbird foraging behavior. Nevertheless, the extent of hovering versus perching behavior, as well as the effects of biotic and abiotic factors, is species-specific.

The data were collected during August 2017 and August 2019 on Mt. Cameroon (Cameroon) in the area around Mann’s Spring. All visitors of the target I. sakeriana plants were recorded for four days using VIVOTEK IB8367-T security cameras during 12-h day periods. The type of behavior (hovering or perching) on the flower and while moving among flowers and the exact time spent on each activity was noted. The linear distances between each pair of opened flowers (i.e., apertures of their corollas) were measured using a measuring tape, the lengths of straightened peduncles and pedicels (PedPed length) were measured using Vernier caliper, both with an accuracy of 1 mm. The actual precipitation was measured in mm by two rain gauges placed in two different forest clearings within the area of selected plants; a photo of the water level in each was taken every 10 min. The average sugar consumption was estimated as the mean difference in sugar amount between non-visited and visited flowers. Nectar was collected using 5 µl microcapillary tubes. Sugar concentration was measured by a pocket refractometer (PAL‐1, Atago Co.).

To test whether the frequencies of the observed types of behavior of each sunbird species on a flower differ significantly from each other, we used the Chi-square test (performed in R; R Core Team 2019). To investigate the effect of sunbird behavior on flower visitation rate, we used nested analysis of covariance (nested ANCOVA) with plant identifier as an error variable and sum of distance between visited flowers as a covariate (performed in R; R Core Team 2019). The visitation rate was logarithmically (ln) transformed to normalize its distribution. Because of the many zero values and impossibility of improving the data distribution, we tested the effects of bird species and their behavior on the nectar amount left after the bird visit using nonparametric mixed effect PERMANOVA with the plant identifier as a random factor. PERMANOVA was performed in the PERMANOVA program, which is an extension of the software PRIMER (Anderson et al. 2008). To determine the effect of behavior and sunbird species on the time spent per flower and on energy intake in time we used a generalized linear mixed model (GLMM) with plant ID nested in year as a random factor (performed in R; R Core Team 2019). For analysis of the effects of sunbird species, PedPed length and precipitation on the type of sunbird behavior, we used a binomial GLMM (package lme4; Bates et al. 2015) with an assumed Bernoulli distribution, setting flower identifiers nested within an individual plant nested within a year as random factors. Average precipitation was logarithmically (ln+1) transformed to reduce the skewness towards larger values, as we obtained many hours of recordings and visit events during light rain but only a few visit events in heavy rain. Both analyses were performed in R (R Core Team 2019).