DENTON (UNT), Texas — Researchers at the University of North Texas are part of a multi-institutional team using artificial intelligence to help ecologists conduct field studies at a larger scale and at more locations than before plus help economists better understand the value and impacts of pollination.

Collecting ecological data often requires researchers to travel to a location and survey the land in person. But that process comes with challenges, said project member Elinor Lichtenberg.

“On-the-ground ecological data collection is very time intensive,” Lichtenberg said. “Even then, the data is limited to a small geographic area and typically one crop.”

Lichtenberg is an assistant professor in the Department of Biological Sciences in the College of Science. She and computer science and engineering professor Xiaohui Yuan are working on the project alongside faculty from Michigan State University, UC Santa Barbara and the University of Maryland. The transdisciplinary team includes ecologists, geographers, entomologists, computer scientists and economists. The project is funded by a $3.6 million grant from the National Science Foundation with $1.4 million coming to UNT.

“We see this as extending to a broad range of ecological processes that are integral to human welfare,” Lichtenberg said. “We're leveraging the intersection of AI-enhanced remote sensing, ecology and economics and applying it to a broader range of settings that involve quantifying ecosystem services.”

The team will focus on the valuing pollination services provided to crop fields by wild and managed bees. Using satellite imagery, the AI tools will help researchers track the locations of bee boxes, estimate floral availability and monitor crop growth across seasons. Members of Lichtenberg’s lab will verify results in the field.

Yuan and his students are developing and refining the AI models with assistance from the other team members to identify bee boxes and blooms of crops using high-resolution satellite image sequences. The new AI technologies and the data products will enable timely surveys of crop pollination to allow biological and economic discoveries.

“We detect a bee box on the ground, which gives us a rough estimation of the number of bees in that area. Then we also track how long it’s in that area plus crop growth around it,” Yuan said. “Then the crop yield will let economists determine things like how pollinator proximity affected growth.”

Development on the model began in 2024 with a seed grant Lichtenberg and Yuan received through the College of Engineering and the College of Science. The initial work focused on training the model to detect bee boxes in satellite imagery, and the findings have been published in Expert Systems With Applications in 2026.

“Manual image annotation is very tedious work. You have to draw a square or rectangle around every box yourself, but sometimes it turns out the objects are water tanks or haystacks,” Yuan said. “We had many sessions with Elinor and another collaborator who would share their expertise on these objects.” Developing AI tools to detect such small objects across vast geographic areas will transform how scientific research is conducted.

During the same period, Lichtenberg worked on a separate project with collaborators from UC Santa Barbara and the University of Maryland. Funded by a U.S. Department of Agriculture grant, that project examined priorities for managing pollinator habitats.

“When farmers need to pollinate their crops, they have two options: one is buying or renting managed bees. The other is relying on wild bees, but that still isn’t free,” Lichtenberg said. “You have to set aside land for those pollinators to live in and manage that land in a certain way.”

The earlier project examined when farmers would benefit most from using managed bees versus relying on wild bees to pollinate their crops. The new grant will build on that research by considering how collective land management decisions affect pollination and related economic outcomes.

“There are two AI tools that will come out of the project. One is a model that will estimate floral availability at any given point in time, which can be an indicator of habitat quality,” Lichtenberg said. “The second is a small object detection model that we trained with honeybee boxes but could be adapted for any sort of small object detection in the future.”

Yuan said the small-object detection model could have applications beyond ecological research, including autonomous driving and defense.

“AI relies on details like texture when detecting objects, which is harder to do at a distance,” Yuan said. “For example, pieces of tire on the road. If an autonomous car can correctly detect it faster, it can respond faster before it becomes a major issue.”

Lichtenberg said she looks forward to seeing how the tools developed through the project can support future research and how collaboration across disciplines can lead to new approaches.

“Working with people in other disciplines, we realized we needed to change how we think about things,” Lichtenberg said. “It changes how we see an issue and leads us to developing new approaches that we would not have considered before.”