Climate TRACE underestimates vehicle CO2, study finds
A peer-reviewed study led by Kevin Gurney of Northern Arizona University, published in Environmental Research Letters, finds that the Climate TRACE emissions database underestimates vehicle CO2 emissions in U.S. cities by an average of 70% compared with the Vulcan Project's on-road inventory, based on a matched comparison of 260 U.S. urban areas. Climate TRACE disputes the finding: co-founder Gavin McCormick told Gizmodo the organization's own comparisons to official city datasets have not shown a discrepancy of that magnitude, and in later correspondence Climate TRACE said the study period coincided with a bug in its city-aggregation layer that was fixed in mid-2025 releases. Gurney's team re-examined Climate TRACE's newer county-scale data and still found a persistent average gap of about 16%, meaning the dispute remains unresolved and methodological.
For teams that ingest third-party emissions inventories into models, dashboards, or compliance workflows, this dispute is a live example of why cross-dataset validation and documented uncertainty bounds matter: a widely used ML- and satellite-based emissions database is now in an unresolved, data-driven disagreement with a government-calibrated benchmark over a factor-of-several gap in a core metric.
What happened
New research led by Kevin Gurney of Northern Arizona University and published in Environmental Research Letters compared urban on-road CO2 estimates in Climate TRACE with the Vulcan Project's on-road inventory and found that Climate TRACE reports vehicle CO2 emissions in U.S. cities that are, on average, 70% lower than Vulcan's estimates, based on a matched comparison covering 260 U.S. urban areas, according to the paper and an NAU press release. Co-author Bilal Aslam said in the NAU release that Vulcan's on-road uncertainty of about 14% is far lower than the gap found versus Climate TRACE, and co-author Pawlok Dass said individual cities such as Indianapolis and Nashville were lower by more than 90%. Gizmodo and Phys.org, reporting on the same paper, include comment from Climate TRACE co-founder Gavin McCormick, who told Gizmodo in an emailed statement that the organization has compared its data to official city datasets worldwide and has not found results consistent with the study's claims; McCormick said Climate TRACE's own U.S. road-transport total (1.5 billion tonnes CO2-equivalent for 2021) is close to the official U.S. inventory submitted to the UNFCCC (1.45 billion tonnes).
Timeline
Gizmodo publishes reporting on the NAU-led study, including comment from Climate TRACE co-founder Gavin McCormick disputing the findings.
Climate TRACE tells Let's Data Science in emailed correspondence that the study's findings reflect a bug in its city-aggregation layer, already fixed in its May 2025 (v4.3.1) and June 2025 (v4.4.0) releases.
Study lead author Kevin Gurney responds with additional technical detail, disputing that the bug fully explains the gap and reporting a persistent average discrepancy of about 16% across roughly 2,000 re-matched U.S. counties using Climate TRACE's newer county-scale data.
Technical context
The paper compares two fundamentally different approaches to allocating urban road-sector CO2. The Vulcan Project uses government-sourced traffic, energy, and infrastructure data with established emission-factor methods to produce high-resolution on-road CO2 estimates, and the study treats Vulcan as a benchmark because it is calibrated to official datasets. Climate TRACE uses machine-learning methods that incorporate satellite observations, atmospheric measurements, and other remote-sensing-derived inputs to build a global emissions inventory. Per Gurney's later correspondence with Let's Data Science, the paper traces part of the discrepancy to a single activity scale factor (0.3255) that Climate TRACE v2 applied to align onroad emissions with the country-scale EDGAR v8.0 inventory, which the paper calls the main driver of the difference; it also documents machine-learning bias in predicted vehicle activity, use of uniform national-average fleet distributions instead of location-specific ones, and fuel-economy values about 26.5% lower than measured values for U.S. passenger cars.
Industry context
Independent, high-resolution inventories such as Vulcan and remote-sensing/ML-based global systems like Climate TRACE are complementary but not interchangeable for urban on-road allocation; methodological mismatches in activity proxies, emission-factor assumptions, spatial allocation, and temporal coverage commonly produce systematic offsets between them. Offsets this large can matter for city-scale policy and verification work, particularly when one inventory is used as a baseline for mitigation tracking; the paper also references a prior comparison flagging discrepancies at power plants, which the authors combine with the new urban findings.
For practitioners
The reported 70% average disparity is large relative to Vulcan's 14% uncertainty, which raises comparability questions beyond raw accuracy, and the dispute is not settled: Climate TRACE's newer county-scale data, matched by Gurney's team across roughly 2,000 of about 3,100 U.S. counties, still showed an average difference of about 16% from Vulcan, ranging from -30% to +65%, well above Climate TRACE's own claimed 1.4% county-level gap. Teams integrating any third-party emissions inventory should expect this kind of scaling and methodology mismatch, document uncertainty bounds explicitly, and treat single-source or single-version comparisons cautiously until independently replicated.
What to watch
Independent replications using other high-resolution city inventories; disclosure of the specific Climate TRACE on-road methods and sensitivity tests used in the version compared in the paper, including a methodology document for the new county-scale output, which Gurney's team says has not yet been published; city-level audits matching local traffic and fuel-consumption records to both inventories; and whether either team issues further updates that change the magnitude or sign of the discrepancy.
Key Points
- 1A peer-reviewed NAU study found Climate TRACE reports 70% lower vehicle CO2 than the Vulcan Project across 260 U.S. cities, a gap Climate TRACE disputes.
- 2Climate TRACE says a since-fixed city-aggregation bug explains much of the gap, but the study's lead author found a persistent ~16% discrepancy in newer county-scale data.
- 3The unresolved dispute underscores why practitioners integrating third-party emissions datasets should cross-validate and document uncertainty bounds rather than trust a single source.
Scoring Rationale
The story matters to data scientists and practitioners who use emissions inventories because it challenges the comparability and reliability of a widely cited, ML-powered dataset, and the dispute remains live and evidence-based on both sides (Climate TRACE's bug explanation vs. Gurney's persistent county-scale re-analysis). It is notable but not paradigm-shifting; resolution depends on further replication and methodological transparency.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
