Previous studies of city roads have tended to underestimate traffic emissions, according to a new study that leverages traffic camera images and mobile phone data to piece together a picture that is at once fine-grained and large-scale.
The methodology fills an urgent need for cost-effective strategies to evaluate the effectiveness of decarbonization policies such as congestion pricing.
“Traffic emissions are far more uneven than we tend to think,” says study team member Songhua Hu, a transportation sustainability researcher at the City University of Hong Kong, who conducted the work as a postdoctoral researcher at the Massachusetts Institute of Technology. “They vary block by block and hour by hour, and those differences matter for both climate policy and public exposure.”
Hu and his colleagues assembled information from a variety of sources for their study of traffic in New York City’s borough of Manhattan, including images from 331 traffic cameras and anonymized location data from more than 1.75 million cell phones. They combined these data with a model of individual vehicle emissions to paint a hyperlocal picture of emissions in near-real-time.
The result is more detailed than citywide emissions estimates and more practical and scalable than analyses based on attaching tailpipe monitors to individual vehicles, the researchers say.
Overall, the team’s estimates of carbon monoxide, carbon dioxide, nitrogen oxides, and small particulate matter emissions known as PM2.5 range from 23 to 49% higher than those from the U.S. Environmental Protection Agency. (Meanwhile, the new estimates correlate well with actual data from air quality sensors sparsely placed throughout the city.)
The researchers used traffic camera images to build a picture of variations in speed and traffic flow as lights cycle from green to red – a key determinant of emissions that has largely been ignored in the past.
“What stood out most was how large the errors can be when we simplify the system,” Hu says. “For example, if you ignore traffic lights, you can underestimate some pollutants by up to about 50 percent, because stop and go driving plays a major role in emissions.”
They also tracked the impact of congestion pricing, which was implemented south of 60th Street in Manhattan in January 2025, the first such scheme in the United States. Congestion pricing has been in place in some European cities for years, but the policy’s impact on actual emissions has remained unclear.
The new study highlights the big-picture effectiveness of the strategy. “Another striking result was how quickly the effects of congestion pricing became visible in the data,” Hu says. Eight weeks after the new policy’s introduction, traffic volumes had fallen by about 10% and emissions by 16-22%, depending on the pollutant.
But the devil, as always, is in the details. “We were able to detect not just an overall drop in emissions, but also clear spatial differences, with larger reductions on higher-class roads and near major entry points, while some areas outside the congestion zone saw even modest increases,” says Hu.
The results thus “raise an important follow-up question,” Hu says: “not just whether emissions fall overall, but who benefits most, and whether some communities may see fewer gains or even unintended spillover effects.” The answers will be key to designing city decarbonization policies that are equitable as well as effective.
Source: Hu S. et al. “Ubiquitous data-driven framework for traffic emission estimation and policy evaluation.” Nature Sustainability 2026.
Image: Pierre Vivant via Wikimedia Commons.
