A product-level Trade-LCA framework for embodied carbon accounting in global trade

Every time a product crosses a border, it carries a hidden weight. This is not the physical weight of the cargo, but the total amount of greenhouse gas released to make that item, from the moment raw materials are pulled from the earth until the product leaves the factory gate. Scientists call this "embodied carbon." For decades, governments have tracked carbon emissions by counting what happens within their own borders, a method that works well for local pollution but fails to capture the full picture of global trade. When a country imports a car, it avoids the emissions from building it, but those emissions still happened somewhere else. Conversely, when a country exports goods, it takes on the responsibility for the carbon used to create them, even if those goods are never used there. This gap between where emissions occur and who ultimately consumes the goods is massive, accounting for more than a quarter of the carbon footprint for major economies. As the world moves toward policies like carbon taxes on imports, officials need to know exactly how much carbon is hidden inside specific products, not just broad industrial categories.

For a long time, the best tools available to answer this question were like looking at the world through a blurry lens. The dominant method, known as input-output analysis, groups thousands of different products into broad sectors, such as "machinery" or "chemicals." It then assigns a single average carbon value to every item in that group. This approach is useful for seeing the big picture of an entire economy, but it is terrible at distinguishing between a highly efficient, low-carbon machine and a wasteful, high-carbon one if they fall under the same label. In reality, the carbon intensity of individual products can vary wildly, sometimes by more than ten times, even within the same factory or industry. Without a way to see these differences, policies risk penalizing clean products while letting dirty ones slip through the cracks.

A new study by Florian Pothin, a researcher at the University of Rennes, builds a bridge between the detailed science of product manufacturing and the massive, complex data of global trade. The goal was to create a system that could track the carbon footprint of individual goods as they move between countries, linking specific customs codes to real-world factory data. The researcher started with a massive dataset of global trade flows, covering nearly 260 million transactions of goods moving between 120 countries over twenty-nine years. To this, they added detailed life-cycle data from two major scientific databases, which contain measurements of the carbon cost for thousands of specific industrial processes, from mining ore to refining steel.

The challenge was matching the two. Customs records describe goods using technical codes and brief descriptions, while the scientific databases describe manufacturing processes in different terms. To solve this, the researcher used a large language model, a type of artificial intelligence, to act as a translator. However, the model was not allowed to invent numbers or guess at the carbon values. Its only job was to suggest which scientific process might match a specific trade code. Every suggestion was then put through a strict set of rules to check if the match made sense physically. For example, the system ensured that a raw material was not matched with a finished product, and that the units of measurement aligned. This process was designed to be transparent and auditable, meaning every step could be checked by a human.

To ensure the system was working correctly, the researcher invited an independent expert to review a random sample of the matches without seeing the computer's suggestions. In the first round of testing, the expert found that 17 out of 100 matches were wrong. The errors were often subtle, such as matching a fresh apple with a chemical process meant for a completely different type of material. The researcher then refined the rules, adding a "ladder" that checked the stage of production to ensure raw ores were not confused with refined metals. After these repairs, a second blind review found only 8 errors out of 100. This rigorous testing proved that while the computer could help sort through the data, human oversight and strict rules were essential to prevent mistakes.

Using this repaired system, the study calculated the total embodied carbon in global trade for the year 2023. The result was approximately 14,368 million metric tons of carbon dioxide equivalent. This figure is significantly higher than previous estimates that relied on unverified matches, because the new system correctly identified that many low-quality data points had been dragging the numbers down. The study found that the most carbon-intensive items were not always the most expensive. For instance, fertilizers accounted for a tiny fraction of the total value of global trade but carried a disproportionately large share of the carbon emissions. Similarly, raw materials like iron, steel, and aluminum were major contributors, while many complex manufactured goods had lower carbon footprints than their price tags might suggest.

One of the most striking findings was how much the carbon intensity of products varied within the same industrial sector. Before the study's corrections, the data suggested that products in the same sector could differ by thirty-five times in their carbon footprint. After fixing the matching errors, the variation was still huge, but more realistic: about thirteen times. This means that even within a single category like "machinery," some products are vastly cleaner than others. A system that uses a simple average for the whole sector would be wrong for nearly half of all traded goods, misestimating their carbon impact by more than a factor of two. This error is not just a statistical curiosity; it means that policies based on broad averages could fail to target the most polluting products effectively.

The study also looked at how different countries fit into this global picture. For some nations, the new data confirmed what was already suspected: they are net importers of carbon, meaning they consume goods made elsewhere. For others, the picture changed dramatically. In the case of China, earlier estimates suggested the country was a net importer of carbon, but the new, more accurate accounting showed it is actually a net exporter. This shift happened because the old data had incorrectly assigned very low carbon values to China's major exports, such as aluminum and steel. The new system corrected these values, revealing that the carbon carried by these exports was much higher than previously thought.

The researchers compared their new product-level numbers against other major global carbon accounts and found they matched up well, capturing about 90% of the expected carbon footprint when weighted by emissions. This suggests the new method is reliable and does not systematically over- or under-count. However, the study also highlighted where the two approaches diverge. In some cases, the difference pointed to real variations in production technology, showing that some countries produce goods with higher carbon intensity than the global average. In other cases, the differences were due to the way services and re-exports are handled, areas where the product-level data is still developing.

Ultimately, this work provides a much sharper tool for understanding the climate impact of global trade. It moves beyond the blurry averages of the past to show the specific carbon cost of individual goods. By proving that a significant portion of previous estimates was based on matching errors, the study warns that earlier numbers may have been misleading. The new framework offers a way to screen products for their carbon content with a level of detail that was previously impossible. This precision is crucial for designing fair and effective climate policies, ensuring that the burden of reducing emissions falls on the right products and the right processes, rather than on broad, inaccurate categories. The study concludes that while the data is not perfect, the transparency of the method and the rigorous testing of its accuracy provide a solid foundation for the next generation of global carbon accounting.