Environment & Energy

LETTER • THE FOLLOWING ARTICLE IS OPEN ACCESS
Assessing the accuracy of the Climate Trace global vehicular CO₂ emissions
Kevin R Gurney*, Bilal Aslam and Pawlok Dass
Published 5 May 2026 • © 2026 The Author(s). Published by IOP Publishing Ltd
Environmental Research Letters, Volume 21, Number 9
Citation Kevin R Gurney et al 2026 Environ. Res. Lett. 21 094018
DOI 10.1088/1748-9326/ae6355

…

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.

Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 license. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

1. Introduction
The emission of greenhouse gases (GHGs) into the earth’s atmosphere is the primary driver of current and projected climate change [1]. Accurate estimation of GHG emissions is not only essential for planning and deploying emissions mitigation activities, but remains fundamental to questions in carbon cycle science and climate change projections [2, 3]. Estimation of these fluxes has a long history and reflects efforts at multiple space and time scales, using different estimation approaches, and aimed at different questions across the climate science and policy landscape [4–6].

As GHG emissions mitigation activity has increasingly focused on sub-national spatial scales, with particular emphasis on urban, provincial, and facility/corporate activities, the need for a mixture of high-resolution estimation that is also globally comprehensive has grown [7]. In response, a variety of datasets have been produced mostly from the research community using rigorous science-based approaches [8–14]. Most of these are expressed in a gridded format ranging in resolution from approximately 0.01 × 0.01 to 1 × 1° spatial resolution. This is often done with the expressed aim to interface with atmospheric modelling research. Scientific studies have attempted to quantify emissions at the scale of emitting infrastructure representing emissions for buildings, roadways, or powerplants [15–18]. These efforts have mostly been limited to domains representing a single country or smaller (e.g. province, city) due to the limited data and techniques able to resolve information at these very fine space/time scales.

The first attempt to mix asset-scale GHG emissions estimation with global coverage is the results provided by the Climate Trace (CT) coalition (https://climatetrace.org/). Established in 2019 and co-founded by former Vice President Al Gore, CT estimates emissions by ‘…training AI algorithms to fuse data across multiple wavelengths and from more than 300 satellites; 11 100 air-, land-, and sea-based sensors; and other data streams to identify all of the largest point sources and track them over time’ [19]. The granularity was further emphasized in a recent data release, ‘This expanded data set also includes inventories for more than 72 000 state and local regions, providing comprehensive GHG data for over 90% of the world’s cities’ [20]. And further claiming ‘unprecedented granularity that pinpoints nearly every major source of GHG emissions around the world and provides independently produced estimates of how much each emits’[21]. These impressive claims are attracting important companies such as Boeing, Tesla, and General Motors to use CT data [22].

A recent study examined the CT emissions estimation for powerplant fossil fuel carbon dioxide (FFCO₂) emissions in the U.S. [23]. In 2022, power production accounted for 36% of global FFCO₂ emissions and 34.3% of the same emissions in the U.S., emphasizing the importance of this sector when considering emissions management [14, 24, 25]. When compared to Vulcan, a well-calibrated, high-resolution estimate of FFCO₂ emissions in the U.S., the study found a mean relative difference (MRD) of 50% across 1726 paired power plants. More surprisingly, the study concluded that only 3.7% of facility emissions estimation used AI techniques. A simple and error-prone approach was used for the remaining 96.7% of the facilities analysed. Of these facilities, only 8.7% agreed to within ±20% of the Vulcan estimates.

The publicity and promotional efforts by the CT coalition, the critical need for this type of data, combined with the large and persistent differences found in the previously completed power plant CO₂ emission assessment, warrants further examination of the CT GHG emissions data.

In this study, we focus on analysis of the on road FFCO₂ emissions from recent CT coalition data releases. On road FFCO₂ emissions account for 30% of the U.S. total FFCO₂ emissions and hence, are the second largest single emitting sector in the U.S. behind power production [25]. Within cities the on road emissions share is typically larger. For example, on road FFCO₂ emission accounted for 43% of the total FFCO₂ emissions in the Los Angeles megacity (California, USA) domain [16]. In Salt Lake city (Utah, USA), the on road FFCO₂ share was estimated at 37.6% and in Indianapolis (Indiana, USA), the share was estimated at 42% [15, 26].

As with the earlier study comparing power production facilities, the comparison provided here is made to the Vulcan Project version 4.0 on road FFCO₂ emissions [25, 27]. We provide a succinct description of the CT and Vulcan methods used to estimate on road FFCO₂ emissions and the comparison metrics relies upon. The differences are reported and an attempt is made to distinguish what elements of the estimation procedure account for the differences found. We end by recommending some best practices for improving the CT coalition estimates.

…