Leafy spurge is a scourge of a weed. Accidentally introduced into the United States in the early 1800s, it contains a toxic substance that has caused deaths in cattle and sheep. If that isn’t enough, it is an aggressive invader, capable of overwhelming native plant communities and destroying wildlife habitats. Between 1960 and 2000, it caused an estimated $1 billion in economic losses for farmers and ranchers. Left unchecked, the invasive plant threatens to spread farther across the United States.
“When leafy spurge flowers, it’s like someone took a yellow highlighter to the landscape. It’s so bright yellow, satellites can see it too,” said Thomas Lake, a University of Minnesota Ph.D. graduate who is now a postdoctoral researcher at the Center for Geospatial Analytics at North Carolina State University.
Lake is part of a research team using artificial intelligence to understand how leafy spurge spreads, with the hope of predicting future invasions and stopping them before they take hold. The research, conducted through the University of Minnesota’s Department of Plant and Microbial Biology, received funding from the U.S. Department of Agriculture’s National Institute of Food and Agriculture and the university’s Minnesota Invasive Terrestrial Plants and Pests Center.
The approach combines satellite imagery, AI and observations collected by ordinary people to identify where leafy spurge has established itself and where it might appear next. Rather than relying exclusively on researchers walking through fields and pastures, scientists can now examine vast stretches of land from hundreds of miles above Earth.
The findings, published Nov. 4 in the scientific journal Ecological Applications, suggest that AI could give farmers, ranchers and land managers a substantial advantage in their decadeslong struggle against one of the country’s most persistent invasive plants.
Using 21 years of satellite imagery, researchers estimated that the area of Minnesota where leafy spurge was detected increased by approximately 570%, from 412 square miles in 2000-02 to 2,763 square miles in 2018-20. The expansion was especially pronounced around the Minneapolis-St. Paul and Fargo-Moorhead metropolitan areas.
Those findings matter because leafy spurge is extraordinarily difficult to eliminate once established. AI cannot eliminate leafy spurge, but it can reveal where it has spread and identify areas most vulnerable to invasion, and that could help land managers intervene before infestations become established.
According to the University of Nebraska-Lincoln’s Extension program, the perennial can grow as tall as 3 feet, with roots extending 15 feet or more underground. Its extensive root system allows it to survive mowing and other control measures, while seeds can remain viable in the soil for eight years or longer.
Each stem produces an average of 140 seeds, some of which can be propelled as far as 15 feet from the parent plant. Birds, animals, water, people and agricultural equipment can carry those seeds farther, creating new infestations well beyond the original plants.
The weed’s milky sap also contains toxic compounds that can irritate livestock and discourage grazing. Although sheep and goats can consume leafy spurge and are sometimes used to control infestations, cattle and horses generally avoid it, reducing the amount of usable pastureland.
For researchers, the challenge has been identifying infestations early enough to intervene. Traditional field surveys are labor-intensive and cannot consistently cover enormous geographic areas. They also depend on where observers happen to look, leaving some infestations undocumented.
AI offers a way to overcome those limitations.
The Minnesota researchers trained a deep-learning model using observations of leafy spurge, including photographs and location records submitted to the community science platform iNaturalist. They combined those observations with historical Landsat satellite images supplied by the U.S. Geological Survey through a program operated with NASA.
The model learned to recognize patterns associated with leafy spurge in satellite observations collected over time. By analyzing changes across successive growing seasons, it could reconstruct the plant’s expansion and improve predictions about where new infestations might emerge.
Researchers discovered that severe drought made leafy spurge more difficult to detect because the plant produced fewer flowers, reducing one of its most recognizable features. Still, the ability to examine decades of satellite imagery provides a perspective that would be nearly impossible to obtain through ground surveys alone.
“Tracking invasive species from satellites vastly improves our capacity to dynamically monitor invasive spread and prioritize limited resources for management,” said David Moeller, a professor in the University of Minnesota’s College of Biological Sciences. “Remote sensing significantly reduces bias by allowing detection in areas where observers often miss.”
One particularly revealing finding was the weed’s tendency to appear near roadways. That association suggests highway maintenance and roadside vegetation management could play important roles in limiting its spread. The research does not establish that roads are responsible for new infestations.
The scientists now plan to expand their mapping efforts across seven states, while examining how climate change, insects and the plant’s own evolution might influence future invasions.
“Having a fuller picture of how plants invaded and where they actually occur will help us to make better predictions about where it might eventually end-up,” said Ryan Runquist, a senior research associate in the College of Biological Sciences.
Across agriculture, researchers are developing AI systems that can recognize threats before they become costly problems.
At the University of Florida, scientists are combining drones, multispectral cameras and machine learning to detect early blight in celery fields, potentially before symptoms become visible to farmers. The project aims to help growers target fungicide applications more precisely, reducing costs and unnecessary chemical use.
Researchers at Texas A&M University have demonstrated how AI can analyze camera footage of cattle to monitor feeding behavior, identify individual animals and measure feed availability. The technology could eventually help ranchers identify health problems earlier and manage livestock more efficiently.

