William Weaver was staring at a digital herbarium image of a centuries-old magnolia specimen when he realized the petiole—the slender stem connecting leaf to branch—held a secret. Hidden in that tiny stalk was a clue to how plants adapt to warming climates, and with machine learning, he could unlock it at scale. Weaver, a Schmidt AI in Science fellow at the University of Michigan, has developed a new algorithm embedded in his software LeafMachine2 that can automatically measure leaf area and petiole width across thousands of digitized plant specimens. In a study published in New Phytologist, Weaver and his team analyzed over 22,000 leaves from 1,580 species of woody angiosperms—flowering plants like oaks, maples, and magnolias—and revealed that temperature, not rainfall, is the dominant force shaping leaf thickness across the globe.
This discovery matters because leaf thickness, measured indirectly through petiole width and area, reflects how much energy a plant invests in its foliage. Thicker leaves with higher mass per area are more common in tropical climates where evergreen plants keep their leaves year-round. The new algorithm allows scientists to estimate this trait across vast digital collections, turning static herbarium archives into dynamic climate records. For the first time, researchers can analyze global patterns in leaf construction investment at a resolution previously impossible—without leaving the lab.
The implications stretch far beyond modern botany. Paleobotanists like Aly Baumgartner, one of the study’s co-authors, use these modern plant traits to decode ancient climates. Fossilized leaves rarely preserve mass or thickness, but their petioles do. "The petiole has to hold up the blade of the leaf, so the greater the mass of the leaf, the thicker the petiole needs to be to support it," Baumgartner explained. "It’s simple physics, but it’s a game changer." By calibrating petiole dimensions to leaf mass in living species, scientists can now infer the climate conditions of forests that vanished millions of years ago.
The study is part of the 2026 State of the World’s Plants and Fungi report from Royal Botanic Gardens, Kew, which highlights how digital tools are revolutionizing biodiversity science. Herbarium collections, once used mainly for taxonomy, are now climate archives. Weaver’s work—done with U-M colleagues Thais Vasconcelos, Baumgartner, Zoë Bugnaski, and James Boyko—shows how artificial intelligence can extract new knowledge from old specimens. As climate change accelerates, tools like LeafMachine2 offer a faster, smarter way to understand how plants respond to environmental stress. And as more collections go digital, the past may hold the key to predicting our ecological future.