The radar sees the birds, but now science can say which ones they are.
For generations, weather radar has traced the great river of bird migration across continents — detecting movement, yet remaining blind to identity. Now, researchers at Cornell Lab of Ornithology and partner universities have woven together two vast, improbable threads — billions of amateur birdwatcher observations and meteorological radar — into a method that can finally name the species behind the signal. The breakthrough speaks to something quietly profound: that the patient, collective gaze of ordinary people, accumulated over decades, can illuminate what the most sophisticated instruments alone cannot see.
- Radar has tracked bird migration for decades, but its fundamental blindness to species identity has left conservationists, public health officials, and researchers working with an incomplete picture of one of nature's most consequential movements.
- The gap matters urgently — different species face different threats from windows, turbines, and aircraft, and waterfowl carrying avian influenza cross international borders nightly, largely unidentified.
- The BMTR method fuses over 2 billion citizen science observations from eBird with radar signals and GPS-tracked birds, creating species-level migration estimates validated against 28 years of continental radar data.
- Models now cover 153 migratory species across 39 families, with plans to integrate species-specific forecasting into BirdCast — a system that could tell a Chicago building manager exactly when to dim its lights for warblers.
- The methodology points toward a future of global migration monitoring, contingent only on whether citizen science networks in Europe, Asia, and Africa can match the density of data North America has spent decades building.
For decades, scientists watched birds cross the continent on weather radar — thousands of blips moving with the seasons — but the technology could never tell them what it was seeing. A warbler or a goose? A thrush or a shorebird? The radar detected movement; it could not name the migrant. Researchers at Cornell Lab of Ornithology, working with colleagues at the University of Massachusetts and University of Illinois, have now bridged that gap by combining two unlikely sources: billions of sightings logged by amateur birdwatchers and the same radar systems meteorologists use to track storms.
The solution grew from BirdFlow, an AI-driven project modeling how bird populations move across North America. The team developed a method called BMTR — BirdFlow Migration Traffic Rate — which draws on eBird's global database of more than 2 billion citizen science observations to infer which species are responsible for the radar signals detected overhead. Because eBird reveals where different species appear at different times of year, it creates a kind of continental fingerprint. When radar shows a surge of birds moving through a region in early May, BMTR can now estimate which species are most likely responsible, even where radar coverage is sparse.
The researchers validated their approach against 28 years of data from 152 weather surveillance radars and against birds individually tracked by GPS and radio telemetry. The results confirmed strong accuracy across 153 migratory species spanning 14 orders and 39 families — a scope impossible through traditional methods alone.
The practical stakes are immediate. Knowing which species move through a region on any given night allows building managers to time light-dimming interventions precisely, helps wind energy operators reduce turbine strikes, and gives public health officials a tool to track avian influenza spread among waterfowl crossing international borders. The team is already integrating BMTR into BirdCast, an existing forecasting system that currently lacks species-level detail, and has expanded the BirdFlow model library from 4 to 60 vetted models.
If citizen science networks in Europe, Asia, and Africa accumulate sufficient data, the same methodology could enable global migration monitoring — a capability that does not yet exist. For now, the breakthrough resolves a constraint that has defined ornithology for generations: the radar has always seen the birds, but science can finally say which ones they are.
For decades, scientists have watched birds cross the continent on weather radar—thousands of blips moving north and south with the seasons—but the radar could never tell them what they were looking at. A wood thrush or a warbler? A goose or a shorebird? The technology detected movement; it could not name the migrant. Now researchers at Cornell Lab of Ornithology, working with colleagues at the University of Massachusetts and University of Illinois, have found a way to bridge that gap, combining two unlikely sources: billions of bird sightings logged by amateur birdwatchers and the same radar systems that meteorologists use to track storms.
The solution emerged from a project called BirdFlow, which uses artificial intelligence to model how bird populations move across North America. The team developed a method called BirdFlow Migration Traffic Rate, or BMTR, that takes observations from eBird—a global database where citizen scientists have submitted more than 2 billion bird sightings—and uses those records to infer which species are responsible for the radar signals scientists detect overhead. The approach works because eBird data reveals where different species are present at different times of year, creating a kind of continental fingerprint. When radar shows a surge of birds moving through a region in early May, BMTR can now estimate which species are most likely responsible for that surge, even in areas where radar coverage is spotty or nonexistent.
To test whether the method actually works, the researchers compared their species-specific migration estimates against 28 years of data collected by 152 weather surveillance radars scattered across North America. The results showed strong correlations, confirming that the new approach produces accurate predictions. They also validated the models against birds that had been individually tracked using GPS devices and radio telemetry, ensuring that the population-level movements the models predicted matched what real birds were actually doing. The team produced working models for 153 migratory bird species spanning 14 different orders and 39 families—a scope that would have been impossible using traditional tracking methods alone.
The practical applications are immediate and varied. Different bird species face different risks during migration. Some are prone to colliding with windows; others are vulnerable to strikes from wind turbines or aircraft. By knowing which species are moving through a region on any given night, conservation groups and building managers can now time their interventions more precisely. A building in Chicago might dim its lights during peak warbler migration but leave them on during other periods. The method also promises to help public health officials track the spread of avian influenza, particularly among waterfowl populations that move across international borders and can carry the virus long distances.
Adriaan Dokter, the project leader for BirdFlow at Cornell Lab, described the breakthrough as opening new directions for real-time migration monitoring and forecasting. The innovation works because it synthesizes multiple types of data—citizen observations, radar signals, GPS tracks from individual birds, and banding records—into a single coherent model. Rather than relying on any one source, which might be incomplete or biased, the method draws on the strengths of each. Citizen scientists provide coverage across the entire continent, including remote areas where radar is sparse. Radar provides continuous, automated detection of movement. GPS and banding data anchor the models to actual bird behavior.
The research team is already planning to integrate BMTR into BirdCast, an existing migration forecasting system that currently provides general predictions but lacks species-level detail. They have also expanded the BirdFlow model collection from 4 to 60 vetted models available to researchers through specialized software. The methodology could eventually extend beyond North America. If citizen science networks in other regions—Europe, Asia, Africa—accumulate sufficient data, the same approach could enable global migration monitoring, a capability that does not currently exist. For now, the breakthrough solves a problem that has constrained ornithology for generations: the radar sees the birds, but now, finally, science can say which ones they are.
Notable Quotes
BirdFlow opens up exciting new directions for monitoring and forecasting bird migration in real time. The new BMTR metrics allow us to estimate the most likely species responsible for the movements we detect with radar.— Adriaan Dokter, project leader for BirdFlow at Cornell Lab of Ornithology
By resolving these population-level movements, BirdFlow can support research and applications in migration ecology, conservation planning, disease surveillance, aviation risk assessment, and public outreach.— Yangkang Chen, Ph.D. student at University of Illinois Urbana-Champaign
The Hearth Conversation Another angle on the story
How does the system actually know which species is moving through the radar beam? The radar itself can't see that.
Right. It uses probability. If eBird data shows that wood thrushes are present in a region in early May, and radar detects birds moving through that same region at that same time, the model calculates the likelihood that those radar blips are wood thrushes. It's not guessing blindly—it's using the geographic and temporal patterns that citizen scientists have documented over years.
So it's only as good as the eBird data. What if there are blind spots in citizen science coverage?
That's a real constraint, yes. But the researchers found that even with uneven coverage, the method works well. And they validated it against 28 years of radar data and GPS-tracked birds, so they have confidence in the accuracy. The gaps matter less than you might think because the patterns are strong enough to emerge even with incomplete data.
What changes for a bird conservationist reading this?
Suddenly they can predict which species will be moving through their region on a given night. A building manager in Chicago can now say, "Warblers peak on May 10th, so we'll dim the lights that week." Before this, they were working with general migration timelines. Now it's species-specific.
And the disease tracking angle—how does that work?
Avian influenza spreads along migration routes. If you know which waterfowl species are moving where and when, you can position surveillance and response efforts more strategically. You're not just watching for sick birds; you're predicting where the virus is most likely to travel.
Does this method work everywhere, or is it limited to North America?
Right now it's North America because that's where the eBird data is densest. But the researchers see no reason it couldn't work globally if other regions built up similar citizen science networks. That's the real long-term vision—a worldwide system for understanding bird movement in real time.