What do museum drawers and deep learning have in common? In our new paper published in Functional Ecology, we show that together, they can help uncover how bees have adapted to climates around the world.
The study, a collaboration between the University of California, Santa Barbara and the University of Kansas, leveraged high-resolution images of preserved bee specimens from both institutions' bee collections. By training convolutional neural networks to measure hair coverage (pilosity) and body lightness from over 600 bee species, the team was able to quantify complex morphological traits that are often too time-consuming or inconsistent to capture manually.The findings are striking: bees from hot, arid environments, especially deserts, tend to be both hairier and lighter in color. These traits likely help bees manage heat stress through increased reflectance and insulation, providing strong support for ecogeographical rules like the thermal melanism hypothesis and Gloger’s Rule. Deserts, the study finds, are hotspots for bees adapted to extreme conditions, echoing patterns seen in plants and other animals.
This work wouldn’t have been possible without digitized natural history collections. Museum specimens provide a time-stamped, spatially explicit archive of biodiversity that, when paired with modern computer vision tools, enable large-scale ecological and evolutionary insights. The collaboration with KU's Entomology Collection was essential in expanding the dataset’s taxonomic and geographic reach.
As conservationists race to understand how pollinators will respond to climate change, this study shows the power of combining historical specimens with modern AI. It’s a compelling reminder that the future of conservation may lie not only in the field—
but also in the museum.
Read the full paper: Ostwald et al. 2025, Functional Ecology