Rising temperatures and unpredictable weather are a challenge for plants, but some are able to adapt and survive. Warm-season perennial grasses, for example, can thrive and reproduce even in high heat and drought.

Understanding how these grasses adapt to a changing environment is critical — affecting everything from pasture quality to seasonal allergies. Now, thanks to an AI tool developed by a University of Delaware professor, scientists will be able to better understand when perennial grasses produce flowers. 

The study, published in the journal Cell, used computer vision AI to comb through tens of thousands of images of perennial grasses taken by everyday people. It shows the power of AI and citizen science in helping researchers digest large datasets while also revealing how plants adapt in diverse, native habitats. 

Why bloom time matters

When plants flower, they’re at the peak of development. Flowering time matters because it determines the environmental conditions a plant will shed pollen under and whether its seeds will mature, impacting whether a species will survive or not. Plants, such as grasses, time their blooming to avoid bad weather conditions. 

“Flowering is a critical biological milestone,” said Yin Bao, an assistant professor of digital agriculture with joint appointments in UD’s College of Agriculture and Natural Resources and College of Engineering. “Once a plant flowers, it produces seeds and then the plant can grow its future generations. It’s critical for surviving and reproducing.” 

Perennial grasses are food for livestock, so understanding their growth and reproduction cycles can help farmers fine tune what they do on their land to help these grasses grow to provide forage for livestock. Knowing the blooming trends of switchgrass can also help meteorologists forecast pollen counts, so people with seasonal allergies can be better prepared to manage their allergy symptoms.

Developing an AI tool

The research team, led by Xianran Li from the U.S. Department of Agriculture’s Agricultural Research Service (USDA-ARS), brought on Bao to develop the AI tool.

Bao and his AgCyPhER lab built a “flowering labeler for open-source research-grade images via self-supervised transformer” — aptly named FLORIST for short. The tool analyzed nearly 44,000 photos from crowdsource network iNaturalist of four North American prairie warm-season perennial grasses — switchgrass, big bluestem, Indiangrass and little bluestem — and sorted them into two groups: flowers showing and no flowers showing.

“It learned what a flower looks like from a few examples that we taught the machine to recognize,” Bao said.

FLORIST filtered out about 80% of the data, leaving around 9,000 photos for the researchers to manually analyze for flowering trends. The use of AI in this study allowed the researchers to drastically reduce the time it could have taken to carefully examine tens of thousands of photos — from weeks to a few hours. 

“Without AI, humans would have to manually inspect over 40,000 photos, looking for tiny anthers or stigmas in the pictures to decide if the species is flowering or not flowering in the picture,” Bao said. “The FLORIST is very efficient with data. It significantly reduced the manual time to inspect those photos.”

The future of farming

Bao specializes in digital agriculture, which leverages advanced technologies, such as AI, in modern farming practices. While digital agriculture isn’t new, the field is constantly evolving. New technologies emerge and farmers and agricultural companies work to adopt the latest and greatest tech. It can make farming, which has always been known as a more labor-intensive career, more efficient and more science-driven. 

“Digital agriculture is essentially giving a farm a smart nervous system,” Bao said. 

Digital agriculture bridges agronomy, engineering and data analytics to help farmers understand what is happening in their fields. AI, for example, enables farmers to predict what their crop yield will look like. Farmers can even use AI to capture tiny details about their crops, like early signs of crop stress, before they might see it with their own eyes. 

Bao’s FLORIST is just one example of what digital agriculture can do, but Bao said there are many more opportunities for technology in agriculture, especially when it comes to more labor-intensive farming.

“Ultimately, this is the future of farming,” Bao said. “We are facing a rapidly growing global population and an unpredictable, shifting climate. We have to figure out how to grow more food using less land and fewer resources.” 

The role of citizen science

Citizen science observations — analyzed by the AI tool — revealed that switchgrass flowers earlier farther north. But genetic experiments, in which plants from different regions were grown side by side in common gardens, showed the opposite trend. The discrepancy stemmed from the fact that plants in the wild grow only where they are already adapted to local conditions, while garden experiments expose them to a much wider range of environments.

Bao was not involved in the genetic experiments, but he said the study shows how citizen science and controlled experiments can work together to explain how plants adapt in their native habitats.

Using a modeling approach developed by senior authors Xianran Li of USDA-ARS and Jianming Yu of Iowa State University, the researchers reconciled the two findings. When they limited the garden analysis to plants collected near each garden site, the experimental results matched the citizen science observations.

“Citizen science observations in nature show what is currently surviving, but they lack the underlying genetic data,” Bao said. “Conversely, garden experiments reveal a plant's full genetic potential and plasticity, but they rely on artificial growth conditions.”

The findings highlight the strengths and limitations of both approaches. Together, Bao said, they provide a more complete picture of how plant populations may respond to changing climates. He said citizen science played a big part in identifying general phenological trends across native habitats. 

“Without citizen science, we wouldn’t be able to establish that trend of going north, flowering earlier,” Bao said.  

The research, Harnessing citizen science to contextualize adaptation mechanism discovery, was supported by the U.S. Department of Agriculture’s Agricultural Research Service, the USDA National Institute of Food and Agriculture, the Iowa State University Plant Sciences Institute, and the U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research, Genomic Science Program grant. The Center for Bioenergy Innovation supported by the U.S. Department of Energy, Office of Science, Biological and Environmental Research provided funding supporting production and characterization of transgenic switchgrass. The paper was published in the journal Cell, DOI 10.1016/j.cell.2026.04.039.