Researchers at Northern Arizona University have found that a major global greenhouse gas database may be substantially undercounting carbon dioxide from cars and trucks in U.S. cities. The study, published in Environmental Research Letters, reports that Climate TRACE estimates for urban vehicle CO2 were about 70% lower on average than estimates from the Vulcan on-road emissions database.
The finding matters because cities rely on emissions inventories to decide where to cut pollution, how to measure progress and how to spend public money. If vehicle emissions are underestimated, local climate plans can miss one of the largest sources of fossil fuel CO2 in urban areas.
The research focused on Climate TRACE, a global emissions tracking effort co-founded by former U.S. Vice President Al Gore. The database uses artificial intelligence-based methods to estimate greenhouse gas pollution across many sectors. Kevin Gurney, a professor in NAU’s School of Informatics, Computing and Cyber Systems, led the analysis with Bilal Aslam and Pawlok Dass.
The 70% Emissions Gap
The headline result is stark. Across 260 U.S. urban areas, the NAU team found that Climate TRACE vehicle CO2 estimates were lower than Vulcan estimates by an average of 70.4%. That comparison centered on emissions from cars and trucks, a major share of city-level fossil fuel pollution.
Bilal Aslam, a postdoctoral researcher and co-investigator at NAU, described the gap plainly. “The Climate TRACE CO2 emissions were, on average, 70% lower than those same emissions in the Vulcan onroad CO2 emissions database.”
The study examined 2021 emissions data. That year matters because it allowed the researchers to compare city-level estimates across a common time period. The team used Vulcan as the independent benchmark because it draws on official traffic records and energy consumption data.
Gurney said the new result adds to earlier concerns from a separate study of Climate TRACE power plant emissions. Taken together, the studies suggest that some AI-based estimates need more testing before they are used for detailed policy decisions.
How Researchers Checked Climate TRACE
To test the database, the NAU team compared Climate TRACE with the Vulcan emissions database, an on-road CO2 system developed by Gurney’s laboratory. Vulcan estimates fossil fuel carbon dioxide emissions at fine spatial scales across the United States.
The Vulcan system is built from multiple constraints. It incorporates official traffic information and energy use data, then calibrates those estimates against atmospheric science methods. According to the NAU researchers, Vulcan’s uncertainty for on-road emissions is about 14%.
That uncertainty is important because every emissions inventory contains some error. The scale of the mismatch in this study was much larger than Vulcan’s reported uncertainty. The authors therefore treated the gap as a meaningful signal that required close attention.
Gurney explained why the team focused on this comparison. “Given the importance of vehicle CO2 emissions in cities, we carefully examined the Climate TRACE data which relied on promising new artificial intelligence-based approaches.”
The study matched Climate TRACE urban vehicle CO2 data to Vulcan estimates across 260 cities. This gave the researchers a broad look at how the two systems behaved in many different urban settings, rather than in a single city case study.
Why Vehicle CO2 Data Matters
Vehicle emissions often make up a large part of a city’s greenhouse gas footprint. Cars, delivery trucks, buses and freight vehicles burn fossil fuels across dense road networks. That activity produces carbon dioxide close to where people live and work.
For city governments, urban vehicle emissions data can shape climate action plans. Officials may use these inventories to evaluate public transit investments, road pricing, fleet electrification, charging infrastructure and land-use decisions. A low estimate can make transportation appear less urgent than it is.
The stakes reach beyond city hall. Carbon accounting informs state and national climate goals, private-sector planning, academic models and public expectations. When a widely used database gives very different numbers from an established benchmark, researchers have to ask how those numbers were produced.
Accurate local emissions data also helps identify hotspots. A citywide total can show the big picture, but street-level or neighborhood-level estimates can reveal where targeted changes may have the greatest effect. That is one reason Gurney’s group has spent years developing detailed emissions mapping systems.
Climate TRACE was designed to provide broad, global visibility into emissions. The NAU study highlights the scientific challenge behind that goal. A tool that works across countries, sectors and data conditions must still perform well when its estimates are checked against independent datasets.
Where the Biggest Differences Appeared
The 70% average gap was only part of the story. In some cities, the difference between the two databases was even larger. Pawlok Dass, a research associate in NAU’s School of Informatics, Computing and Cyber Systems, pointed to two striking examples.
“Individual cities such as Indianapolis and Nashville were lower by more than 90%,” Dass said.
Those city-level results show why averages can hide important local effects. A national or multi-city mean helps summarize the study, but climate planning happens in specific places. A 90% difference could strongly affect how a city ranks its largest emissions sources.
The NAU researchers suggest that the undercounting may extend beyond the United States. Their direct comparison focused on U.S. urban areas because Vulcan provides a strong benchmark there. If the same modeling problems appear elsewhere, global city estimates could be affected as well.
That possibility remains a key limitation and next step. The study gives a detailed test in one country, using one independent comparison dataset. Broader international checks would require reliable local benchmarks in other regions, which can be difficult to obtain.
What Went Wrong in the Estimates
The paper points to several likely sources of the discrepancy. The researchers cite biases in Climate TRACE’s machine learning model, along with fuel economy values and fleet distribution values. Each factor can change how much CO2 a model assigns to road travel.
Fuel economy values determine how much fuel vehicles are assumed to burn over a given distance. If a model assumes vehicles are more efficient than they really are, emissions can be underestimated. Fleet distribution values also matter because cars, pickup trucks, buses and heavy-duty trucks produce different amounts of CO2.
Machine learning can identify patterns in large datasets, but the output depends on the data and assumptions used to train the system. In transportation emissions, small errors can multiply across millions of vehicles and thousands of roads. A city’s final total can drift far from reality if key inputs are biased.
The NAU team framed the issue as a scientific measurement problem. Climate inventories need transparent methods, independent checks and uncertainty estimates that policymakers can interpret. Without that structure, impressive technology can produce numbers that look precise while carrying hidden weaknesses.
The study does acknowledge that no emissions estimate is perfectly exact. The central question is whether a dataset is accurate enough for its intended use. For sub-national policy guidance and climate science applications, the NAU authors urge caution when using Climate TRACE on-road CO2 estimates.
Why AI Climate Tools Need Scrutiny
Artificial intelligence is becoming more visible in climate monitoring. It can process satellite data, identify infrastructure, fill data gaps and estimate emissions in places with limited reporting. That makes AI-based systems attractive for global climate accountability.
Still, AI-based emissions tracking needs careful validation. A model can scale quickly across the planet, but scale alone does not guarantee accuracy. Researchers need to compare estimates with ground-based records, atmospheric measurements and independent inventories whenever possible.
Gurney’s comments focused on trust as much as technology. “We will never estimate emissions with perfect accuracy, but we must ensure that the data shared with policymakers and the public is unbiased.”
That message is especially relevant for climate policy. Governments use emissions data to justify decisions that can affect transportation systems, energy markets, infrastructure spending and public health. When the data are weak, even well-intended decisions can drift away from the largest sources of pollution.
Scientific review also helps the developers of emissions databases. Independent studies can expose weak points, sharpen methods and increase confidence in tools that perform well. In that sense, the NAU study is part of a larger effort to make climate data more reliable.
Gurney’s Vulcan Emissions Work
Kevin Gurney has worked for more than two decades on detailed fossil fuel CO2 accounting. His research spans atmospheric science, ecology and public policy. At Northern Arizona University, his team has developed systems that estimate emissions from power plants, roads, neighborhoods and other sources.
The Vulcan Project is one of those systems. It produces high-resolution maps of fossil fuel carbon dioxide emissions across the United States. These maps help scientists and planners see where emissions are concentrated, which sectors dominate and how emissions patterns change over time.
Gurney is also associated with the Hestia project, which visualizes greenhouse gas emissions at neighborhood and infrastructure scales. Together, Vulcan and Hestia support a more detailed view of fossil fuel CO2 than broad national totals can provide.
The NAU announcement notes that Gurney’s emissions estimates have shown strong agreement with direct atmospheric monitoring measurements. That history is one reason Vulcan was used as the benchmark in the Climate TRACE vehicle emissions study.
The new paper adds another layer to a long-running scientific challenge. Climate policy depends on knowing where emissions come from with enough accuracy to act. For urban transportation, the NAU team’s findings suggest that greenhouse gas inventories need rigorous testing before their numbers guide local and global decisions.
