Climate TRACE is a global nonprofit coalition that provides independent, open-access greenhouse gas (GHG) emissions data using satellites, AI/machine learning, remote sensing, and other datasets.

Climate TRACE combines:

• Satellite imagery and remote sensing.
• AI/ML models (e.g., for detecting activity like traffic or facility operations).
• Public and commercial datasets.
• Bottom-up and top-down methods.

Its goal is greater transparency and timeliness than traditional self-reported national inventories submitted to the UN.

It was co-founded/promoted with involvement from former Vice President Al Gore.

Climate TRACE (CT) significantly underestimates urban on-road (vehicular) CO₂ emissions in the U.S. by an average of ~70% compared to a high-quality, bottom-up reference dataset.

Recent research by climate scientist Kevin Gurney of Northern Arizona University (NAU) has identified significant underestimations in the Climate TRACE global greenhouse gas emissions database, particularly for urban vehicle CO₂ emissions.

Urban vehicle emissions: Climate TRACE underestimates CO₂ emissions from cars and trucks in U.S. cities by an average of 70% compared to the well-established Vulcan onroad database (calibrated to official traffic and energy data). Some cities show discrepancies over 90%.

Power plants (prior study): A 2024 NAU analysis found Climate TRACE underestimated emissions at power plants by an average of 50% in the U.S. Only ~4% of facilities used AI-based methods; most relied on approximate approaches.

These results come from peer-reviewed work published in Environmental Research Letters. Gurney and colleagues compared Climate TRACE data against high-resolution, bottom-up inventories like Vulcan, which integrate detailed activity data (e.g., traffic counts, fuel sales) with lower uncertainty (~15% for power sector comparisons).

The critiques highlight that for on-road vehicles and many power plants, its methods (often relying on proxies or approximations rather than direct high-resolution activity data) lead to systematic underestimation of fossil fuel CO₂.

Accurate emissions data is essential for:

• Tracking progress toward Paris Agreement goals.
• Designing effective mitigation policies.
• Carbon markets and accountability mechanisms.
• Scientific modeling of climate impacts.

Underestimation could imply that true emissions (and thus needed reductions) are higher than reported in some sectors/locations. Gurney emphasizes that reliable data is “a critical ingredient for society’s response to climate change.”

This is a specific methodological critique from within the climate science community — not a challenge to the broader reality of anthropogenic climate change or the need for emissions reductions. Similar discrepancies have been noted in other inventories before; science advances through such validation and improvement.For details, see the NAU announcement or the Phys.org coverage of the recent paper. The full study is available in Environmental Research Letters.

_____________________________________________________________________________________

Assessing the accuracy of the Climate Trace global vehicular CO2 emissions

Climate TRACE (CT) significantly underestimates urban on-road (vehicular) CO₂ emissions in the U.S. by an average of ~70% compared to a high-quality, bottom-up reference dataset.

This assessment is based on a peer-reviewed study by Kevin R. Gurney and colleagues (Northern Arizona University), published May 5, 2026, in Environmental Research Letters: “Assessing the accuracy of the Climate Trace global vehicular CO₂ emissions.”

Reference dataset: Vulcan Project (version 4.0) — an atmospherically calibrated, multi-constraint inventory using detailed U.S. traffic counts, fuel sales, vehicle registration data, and other activity data. Uncertainty is relatively low (~14% for on-road emissions).

Scope: 260 U.S. urban areas in 2021 (with analysis also for 2022 and different CT data versions).

Mean relative difference (MRD): ~70% (CT lower than Vulcan). Individual cities like Indianapolis and Nashville showed discrepancies >90%.

Systematic bias: High correlation (Pearson ~0.99) between datasets indicates consistent underestimation rather than random error. Most urban areas fell outside Vulcan’s 95% confidence intervals.

Different CT data versions (v1 and v2) showed large differences from Vulcan, sometimes flipping from over- to under-estimation due to methodological tweaks (e.g., scaling factors), but the magnitude of error remained similar.

Main Sources of Error in Climate TRACE

Machine learning model biases — CT uses convolutional and graph neural networks trained on U.S. Sentinel-2 satellite imagery to predict annual average daily traffic (AADT) on OpenStreetMap road segments. These predictions appear biased when scaled to emissions.

Fuel economy and fleet distribution — Reliance on U.S. national averages (from sources like CURB and FHWA) that do not adequately capture local variations in vehicle mix, driving conditions, or efficiency.

Scaling adjustments — Version 2 applied a global scaling factor to align with EDGAR v8.0 at the country level, which introduced further offsets at urban scales.

Vulcan integrates more direct activity data (e.g., traffic counts, fuel consumption) and has been validated against atmospheric measurements.

Context and Implications

This is a methodological critique, not a denial of climate science. Vehicle emissions are a major and growing share of urban GHG totals (often 30–40%+ in U.S. cities). Underestimation affects policy targeting, carbon accounting, and progress tracking toward mitigation goals.

Global relevance: The study focuses on U.S. urban areas (where validation data are strongest), but the authors suspect similar issues may apply elsewhere due to the global methods used. CT aims for worldwide coverage using satellite + AI approaches.

Comparison to prior work: A 2024 Gurney study found CT underestimated U.S. power plant CO₂ by ~50% on average (with non-AI methods performing poorly).

Caveats: No dataset is perfect. Vulcan has its own uncertainties, and CT’s satellite/AI approach offers valuable global, near-real-time granularity that traditional inventories lack. Independent validation like this drives improvements.

The full open-access paper is available in Environmental Research Letters. NAU’s press release and coverage on Phys.org provide clear summaries. Climate TRACE continues to update its methods, and such comparisons help refine global datasets.

Published:  Environmental Research Letters

DOI: 10.1088/1748-9326/ae6355

Provided: Northern Arizona University peer-reviewed

Authors: Kevin R Gurney, Bilal Aslam and Pawlok Dass

Abstract

Accurate estimation of greenhouse gas (GHG) emissions at the infrastructure scale remains essential to climate science and policy applications. Vehicle emissions often dominate GHG emissions in urban areas and are rapidly increasing globally. Climate Trace (CT), co-founded by former U.S. Vice President Al Gore, is a new AI-based effort to estimate roadway-scale GHG emissions. However, limited independent peer-reviewed assessment has been made of this dataset. Here, we compare CT on road CO₂ emissions in U.S. urban areas to atmospherically calibrated, multi-constraint estimates of on road CO₂ emissions from the Vulcan Project. Across 260 urban areas in 2021, we find a mean relative difference (MRD) of 70.4%. These large differences are driven by biases in CT’s machine learning model, fuel economy values, and fleet distribution values. We conclude that sub-national policy guidance or climate science applications using the on road CO₂ emissions estimates made by CT should be done so with caution.