A novel computer vision tool to quantify pilosity in bees: Are hairy bees more cold-tolerant?

Understanding the mechanistic underpinnings of thermal tolerance variation is imperative for predicting consequences of climate change for insect pollinations. For bees in particular, variation in thermal tolerance may be related to variation in pilosity (hairiness), which can act to reduce convective heat exchange. However, explorations of pilosity in bees and other insects are rare and methodologically rudimentary due to the labor-intensive nature of counting and measuring individual hairs. Manual measurements likewise fail to capture the complexity of hairiness, which varies structurally and spatially both within and among individuals. We present a novel machine learning model capable of quantifying various metrics of pilosity from images of bees. We leverage this technology to test the hypothesis that pilosity varies with thermal tolerance in males of the solitary bee, Melissodes tepida, which exhibit strong intraspecific variation in pilosity. We combine physiological, behavioral, and morphological trait data to understand how pilosity impacts activity timing in M. tepida. These results provide evidence for the role of morphological traits in mediating thermal tolerance variation, and emphasize the value of computer-automated phenotype quantification in physiological studies.