This chapter clarifies the concept of reliable and widespread carbon footprints as used in this report. It discusses the "cradle-to-gate" logic around which the report is organised, and the challenges of reliability. From these concepts, it is possible to derive the eight building blocks necessary to achieve the ambitious goal of reliable and widespread carbon footprints in food systems.
Measuring Carbon Footprints of Agri‑Food Products
Eight Building Blocks
Abstract
This report asks what it would take to achieve reliable and widespread carbon footprints in food systems. To clarify this concept, it is useful to go over each of the terms separately, before deriving the necessary building blocks to achieve this outcome.
The focus in this report is mainly on carbon footprints as they occur along the food supply chain up to the point of purchase by consumers (e.g. in shops, restaurants), taking into account the full life cycle of the product up to that point – including land use change and the production of inputs. However, as the term ‘food supply chain’ might be interpreted by some to mean only food processing, distribution, and retail, or starting at the farm rather than taking into account land use change and the production of inputs, the broader term ‘food systems’ will be used here. Most of the discussion will focus on land-based production, although many of the ideas apply to fisheries and aquaculture and novel foods such as meat protein alternatives as well.1
As is common in the literature, carbon footprints here refer not just to carbon dioxide (CO2) emissions but to all GHG emissions (which will typically be expressed in CO2-equivalents). This is particularly important in the case of food systems as a large share of emissions consist of methane (CH4) and nitrous oxide (N2O).
The term ‘carbon footprint’ can refer to different reporting levels, such as countries, sectors, entities (firms, organisations), or products (Deconinck, Jansen and Barisone, 2023[1]).
In this report, the focus is on product carbon footprints. Measuring product carbon footprints requires defining a denominator (e.g. emissions per kg of product). The choice of unit will be discussed in more detail in Chapter 4.
One reason for the focus on product carbon footprints is that quantifying product carbon footprints would also indirectly provide information about carbon footprints at other levels of analysis. Quantifying product carbon footprints requires clarity on how to quantify farm level or firm-level emissions as well, as these are inputs in the calculation. In turn, product carbon footprint information from suppliers can be used to quantify upstream supply chain emissions (which is part of a firm’s Scope 3 emissions).
Product carbon footprints can be defined on a ‘cradle to grave’ basis, covering all stages of the product life cycle including use and waste disposal (Figure 3.1). Other approaches are possible too, such as ‘cradle to farm gate’ or ‘cradle to purchase’.2 However, the emphasis here is on a ‘cradle to gate’ approach, where each actor in the supply chain focuses on calculating product carbon footprints of the product life cycle up to the point where the product leaves its premises.
Note: Simplified representation of the stages of a typical product life cycle for food products. (Transport is not explicitly shown here as it occurs in between each stage but is also in scope).
Source: Adapted from IDF (2022[2]).
A cradle-to-gate approach makes it easier to achieve widespread carbon footprints by “decentralising” the task of calculating carbon footprints. As pointed out by the Partnership for Carbon Transparency (PACT, 2023[3]), if product carbon footprint information on a cradle-to-gate basis is widely available from suppliers, then each actor in the supply chain can focus on calculating its own emissions, adding the carbon footprints of its inputs (provided by suppliers), and allocating the total across its outputs. The resulting product carbon footprint can then be shared with customers. In this way, the carbon footprint of a product can be built up step by step throughout a supply chain, allowing the use of primary data to the maximum extent possible (Figure 3.2).
Note: Simplified representation of product carbon footprints using a cradle-to-gate principle. A firm receives information from its suppliers on the carbon footprint of its purchased inputs, using a cradle-to-gate principle (i.e. including all upstream emissions). The firm adds its own emissions and shares the resulting cradle-to-gate product carbon footprint with its customers.
Source: OECD analysis.
How would such an approach look like in food supply chains? Starting at the input stage, suppliers of agricultural inputs (such as fertilisers) calculate the carbon footprint of their products using primary data. They in turn provide this information to farmers, either by sharing data directly with the farmer or by making their data publicly available. Farmers then use farm level calculation tools to estimate their on-farm emissions and add this to the emissions embedded in their purchased inputs. They allocate the total emissions across their different outputs (for example, a dairy farmer would need to allocate emissions across milk and meat). The resulting product carbon footprint information is then shared with processors. Processors add their own emissions calculated using primary data (e.g. on energy use, transport), allocate the result across their different outputs (e.g. a dairy processor would need to allocate emissions across cheese, milk powder, fluid milk) and share the resulting product carbon footprint with the next stage in the supply chain (e.g. food manufacturers, traders/wholesalers, retailers).
Each subsequent stage in the supply chain thus takes the ‘cradle to gate’ information received from its suppliers, adds their own emissions, allocates the result across their different products, and shares it with the next stage. A similar “modular” approach to emissions accounting in supply chains has been proposed by White et al. (2021[4]) and Reeve and Aisbett (2022[5]). Where information is not available for a supplier, firms may need to rely on secondary data, as is currently often the case.
The availability of product carbon footprint information would also help emissions reporting at other scales. For example, as noted in Box 1.1 (Chapter 1) firms are increasingly asked to report not just the total emissions from their own operations (Scope 1 emissions) and from the energy they purchase (Scope 2), but also Scope 3 emissions, which include upstream and downstream supply chain emissions. If product carbon footprint information is widely available on a cradle-to-gate basis, then calculating the upstream supply chain emissions becomes straightforward. This is another motivation for this report’s focus on product carbon footprints using a cradle-to-gate basis.3
Widespread carbon footprint information ideally means that information is available for all food products, for all producers, at all stages of the supply chain, so that stakeholders can easily take the information into account in their decision making.
Carbon footprints are an application of the life-cycle assessment (LCA) methodology to the specific issue of climate change. Historically, LCAs were conducted as highly customised one-time projects. An expert in LCA would work with a client to map the life cycle of a product and would use a variety of research methods to quantify the various flows. The resulting assessment would be used to identify hotspots (priority areas to be tackled) or to help re-design products and would often remain proprietary information of the client and/or the expert. Thus, originally, an LCA was best thought of as an individual study. Over time, as more life-cycle assessments were conducted, results were increasingly brought together in databases. These made it possible to draw comparisons between different products and processes (as in the synthesis by Poore and Nemecek (2018[6]) mentioned earlier), and to use the data to fill in gaps in LCAs where primary data is unavailable. However, not all products and geographies have been equally well studied (Deconinck and Toyama, 2022[7]).
The concept of reliable and widespread carbon footprints studied in this report can be seen as the logical next step. Databases provide valuable information, and further refinements can make them even more useful. But average data as found in a database can hide a considerable degree of heterogeneity and is static. For example, the database might contain information on the average carbon footprint of milk in Switzerland, at farm gate. But the database cannot reflect the efforts an individual farmer has made to reduce emissions, or the changing sourcing decisions made by a processor or retailer in its supply chain. In terms of the “three levers” identified in Chapter 2, databases can help shift purchasing decisions from product categories with higher average carbon footprints to product categories with lower average carbon footprints (the first lever), but they cannot capture individual heterogeneity (the second lever) and cannot identify and incentivise the different actions a producer could take to reduce their footprint (the third lever). Individual studies can do so, but are time consuming and costly, and where practitioners use different methodological choices, results may be hard to compare.
What is needed, therefore, is an approach which captures individual heterogeneity and mitigation efforts as in an individual study, while making data comparable and as easily available as in a database. This is the reasoning behind the proposal by the Partnership for Carbon Transparency (PACT, 2023[3]), as described earlier. However, this logic only works if the available product carbon footprint information is reliable.
The reliability of an estimate or measurement has two components. The first is that it should not be systematically over or under the true value – a concept known as “unbiasedness” in statistics, or “trueness” in the ISO 5725-1 terminology. The second is that the non-systematic (random) error should be small.4 For example, if firms do not include some sources of emissions in their estimates, this would lead to a systematic understatement of the carbon footprint. By contrast, if firms use industry averages for the carbon footprint of an input rather than supplier-specific information, the result will be random error, as the true carbon footprint of its suppliers might be higher or lower than the industry average – unless its suppliers strategically chose not to disclose their carbon footprint because it is above average, in which case the result would be a systematic understatement of emissions.
There is a strong case for using primary data as much as possible in calculating product carbon footprints, rather than secondary data.5 One reason is the large heterogeneity of carbon footprints even among producers in the same region, which means that averages could lead to significant random error, even if there is no systematic over- or underestimation. A second reason is that if a producer adopts mitigation techniques to reduce emissions, this should ideally be reflected in carbon footprint calculations, to provide proper incentives to the producer and other supply chain actors. These arguments apply not just to food systems, but to other sectors as well (PACT, 2023[3]). To be reliable, carbon footprints should therefore be timely and granular (OECD, 2024[8]).
However, estimates based on primary data may come with their own measurement errors. If primary data from suppliers is used as an input in calculating carbon footprints downstream, any upstream measurement error will affect downstream results. Systematic errors upstream will lead to systematic errors throughout the supply chain. Random errors, by contrast, might end up being ‘averaged out’: for example, if a dairy processor has thousands of farmers supplying milk, a random error leading to an understatement in the carbon footprint estimate of an individual supplier would probably be offset by a random error leading to an overstatement for another supplier. However, reducing random error is still important for several reasons.
First, even if random errors of individual suppliers may be ‘averaged out’ in a supply chain, they still lead to uncertainty if the number of suppliers is small. For example, if a processor has only three suppliers, it is possible that all three random errors happen to be positive (leading to an overstatement of the carbon footprints) or negative (leading to an understatement). The smaller the number of suppliers, the higher the chance of such situations occurring, creating uncertainty.6
Second, if the goal is comparability of carbon footprint information (across products, producers, countries, etc.), even random error needs to be avoided or minimised as much as possible, as comparisons might otherwise lead to wrong conclusions. For example, if the dairy processor is selecting its suppliers based on their estimated carbon footprint, random error could mean that farmers are unfairly excluded.
Third, random error, like systematic error, would send the wrong signals to individual actors about where to focus their mitigation efforts. If a farmer’s carbon footprint estimate contains measurement error (whether random or systematic), it becomes harder for the farmer to choose cost-effective mitigation measures.
Both systematic and random error can be reduced by insisting on completeness (all relevant emissions sources and sinks should be included) and consistency (assumptions, methods, and data should always be used in the same way), as well as on using the most up-to-date science-based methods. As science progresses, it seems likely that calculation methods will become more precise, reducing measurement error. In addition, a form of quality assurance such as third-party verification can also help improve reliability. Where supplier-specific information is used, data sharing tools can help avoid human error and can provide an ‘audit trail’ for quality assurance.
The requirements that product carbon footprints should be reliable and widespread are closely connected. Since generic averages could be misleading, there is a strong case for incorporating supplier-specific primary data – in other words, widespread product carbon footprints could help with reliability. In turn, achieving widespread carbon footprints is useless if data is of poor quality. However, there may also be trade-offs: increasing the reliability of carbon footprint estimates can increase the cost of calculations, which would make it harder to scale up carbon footprint calculations.
As the preceding discussion shows, the concept of reliable and widespread carbon footprints in food systems is ambitious and demanding. But it also creates clarity about the necessary building blocks and can create a common vision for how these building blocks should be further developed or adjusted. It seems likely that these efforts would in turn have positive effects in creating a better data infrastructure even if they do not achieve a near-universal system of carbon footprints.
Based on the key findings about food systems emissions and the conceptual discussion above, at least eight distinct building blocks can be distinguished for reliable and widespread carbon footprint measurement in food systems.7 They are:
Reporting standards and guidelines for carbon footprint measurement, to create a shared understanding of what to include in carbon footprint calculations.
Science-based methods for measuring or estimating emissions.
Farm level calculation tools, which allow different actors along the supply chain to use primary data on their activities and management practices as inputs to calculate their carbon footprint, in line with up-to-date science-based methods.
Databases with secondary data, to be used where primary data is not (yet) available.
A way of communicating carbon footprint data along the supply chain, so that detailed calculations by producers at one stage of the supply chain can be used as input at the next stage.
A way to ensure the integrity and quality of the data and calculations, for example through third-party verification.
A way to scale up carbon footprint calculations while keeping costs low, to ensure widespread adoption by actors with relatively limited administrative capacity, notably farmers, small and medium-sized enterprises (SMEs), and producers in developing countries.
A way to update these elements as new scientific insights and techniques become available.
Again, a detailed discussion of international trade implications is beyond the scope of this report, but it is worth noting some connections between the building blocks identified here and rules designed to avoid trade barriers. The World Trade Organization’s Agreement on Technical Barriers to Trade (the TBT Agreement) incentivises WTO members to align standards and regulations on common international standards and encourages members to accept the results of conformity assessments (verification) performed by other members. The TBT Agreement also recognises the special needs of producers in developing countries and the potential role of technical assistance in helping them meet standards. These principles (on coherence in measurement and standards, on robust verification, and on inclusiveness) are highly relevant to the question of quantifying carbon footprints in an international context (WTO, 2022[9]). Communicating carbon footprint data along supply chains also connects to issues such as trade facilitation (OECD, 2018[10]; Sorescu and Bollig, 2022[11]) and data localisation measures (Del Giovane, Ferencz and López González, 2023[12]).
The following chapters cover each of these building blocks in more detail, assessing what is already in place and which further actions would be needed.
References
[14] Bjørn, A. and M. Hauschild (2017), “Cradle to Cradle and LCA”, in Life Cycle Assessment, Springer International Publishing, Cham, https://doi.org/10.1007/978-3-319-56475-3_25.
[1] Deconinck, K., M. Jansen and C. Barisone (2023), “Fast and furious: the rise of environmental impact reporting in food systems”, European Review of Agricultural Economics, Vol. 50/4, pp. 1310-1337, https://doi.org/10.1093/erae/jbad018.
[7] Deconinck, K. and L. Toyama (2022), “Environmental impacts along food supply chains: Methods, findings, and evidence gaps”, OECD Food, Agriculture and Fisheries Papers, No. 185, OECD Publishing, Paris, https://doi.org/10.1787/48232173-en.
[12] Del Giovane, C., J. Ferencz and J. López González (2023), “The Nature, Evolution and Potential Implications of Data Localisation Measures”, OECD Trade Policy Papers, No. 278, OECD Publishing, Paris, https://doi.org/10.1787/179f718a-en.
[13] Frezal, C., C. Nenert and H. Gay (2022), “Meat protein alternatives: Opportunities and challenges for food systems’ transformation”, OECD Food, Agriculture and Fisheries Papers, No. 182, OECD Publishing, Paris, https://doi.org/10.1787/387d30cf-en.
[2] IDF (2022), The IDF Global Carbon Footprint Standard for the Dairy Sector, International Dairy Federation, https://shop.fil-idf.org/products/the-idf-global-carbon-footprint-standard-for-the-dairy-sector.
[8] OECD (2024), “Towards more accurate, timely, and granular product-level carbon intensity metrics: A scoping note”, Inclusive Forum on Carbon Mitigation Approaches Papers, No. 1, OECD Publishing, Paris, https://doi.org/10.1787/4de3422f-en.
[10] OECD (2018), Trade Facilitation and the Global Economy, OECD Publishing, Paris, https://doi.org/10.1787/9789264277571-en.
[3] PACT (2023), PACT Pathfinder Framework: Guidance for the Accounting and Exchange of Product Life Cycle Emissions, Version 2.0, https://www.carbon-transparency.com/media/b13h4b25/pathfinder-framework_report_final.pdf.
[6] Poore, J. and T. Nemecek (2018), “Reducing food’s environmental impacts through producers and consumers”, Science, Vol. 360/6392, pp. 987-992, https://doi.org/10.1126/science.aaq0216.
[5] Reeve, A. and E. Aisbett (2022), “National accounting systems as a foundation for embedded emissions accounting in trade-related climate policies”, Journal of Cleaner Production, Vol. 371, p. 133678, https://doi.org/10.1016/j.jclepro.2022.133678.
[11] Sorescu, S. and C. Bollig (2022), “Trade facilitation reforms worldwide: State of play in 2022”, OECD Trade Policy Papers, No. 263, OECD Publishing, Paris, https://doi.org/10.1787/ce7af2ce-en.
[4] White, L. et al. (2021), “Towards emissions certification systems for international trade in hydrogen: The policy challenge of defining boundaries for emissions accounting”, Energy, Vol. 215, p. 119139, https://doi.org/10.1016/j.energy.2020.119139.
[9] WTO (2022), What yardstick for net zero? Trade and Climate Change Information Brief n° 6, World Trade Organization, https://www.wto.org/english/news_e/news21_e/clim_03nov21-6_e.pdf.
← 1. On meat protein alternatives, see Frezal et al. (2022[13]).
← 2. Yet another possibility, not shown in the figure, is a “cradle-to-cradle” approach. This approach replaces the final step of waste disposal by a reusing or recycling step, so that the ‘waste’ product effectively becomes the input in another production process, creating a more circular model. See Bjorn and Hauschild (2017[14]) for a discussion.
← 3. One shortcoming of a cradle-to-gate approach is that it focuses on activities taking place within companies. This leaves out the activities by households (e.g. emissions from cooking) and waste disposal. In principle, it might be possible to add an estimate for these emissions to the carbon footprint calculation at the retail stage. However, this would necessarily need to involve average data rather than primary data.
← 4. The terms “accuracy” and “precision” are often used in this context, but the terms can be confusing. For example, in metrology, the term “accuracy” refers to the systematic error, while “precision” refers to the random error; however, in the ISO 5725-1 standard, “accuracy” describes a combination of low systematic error (high trueness) and low random error (high precision).
← 5. Intuitively, the difference between primary and secondary data is that secondary data was collected in other contexts or for different purposes and is used as an approximation instead of collecting primary data on the specific product, firm, or farm being studied. In reality, the distinction is more of a continuum. For example, on a farm, direct measurement of emissions (e.g. using sensors) is often difficult and costly. In practice, primary activity data (e.g. on the number of animals, manure management practices, feed rations, use of cover crops) is fed into a model to estimate emissions. While this is one step removed from direct observation of emissions, it still leads to a more specific estimate than using average data (e.g. based on estimates obtained on other farms). In what follows, estimates based on primary activity data will therefore also be referred to as primary data.
← 6. This can be seen more formally from the formula for the standard deviation of a sample mean, which is where is the standard deviation in the population (which in this context can be thought of as the standard deviation of the random measurement error) and is the number of observations (in this context, the number of suppliers). For large numbers of suppliers (high ), this expression becomes small, as random errors are more likely to ‘cancel out’. For a small number of suppliers, this is not the case, making it more important to reduce the random error (i.e. a lower ) to reduce the overall uncertainty.
← 7. Recent work by OECD and the International Trade Centre has developed a typology of sustainability initiatives (OECD report) to help establish a common understanding of the characteristics of different sustainability initiatives, and their similarities and differences. The typology looks at features related to an initiative’s objective, scope, operations, and governance, each broken down into differentiators, for which potential attributes are defined. The typology is sufficiently flexible that it can be used to organise the various building blocks covered here. For example, a carbon footprint standard and a farm level calculation tool would both fall under “Scope – sustainability – environmental” and “Scope – performance – outcomes”, but would differ on the objective: where the standard would have “Objective – facilitation – guidance/framework”, the farm level calculation tool would have “Objective – facilitation – tool”. Other building blocks can similarly be classified.