Determining the climate impact of food for use in a climate tax—design of a consistent and transparent model

In comparison with the transport and energy sectors where emissions of GHGs are dominated by fossil fuel related CO₂, the emissions generated during the life cycle of foods include substantial amounts of methane (CH₄) and nitrous oxide (N₂O). Emissions of CH₄ mainly arise from feed digestion in ruminants, while the majority of the N₂O emitted originates from fertilised soils. Another important source of food-related GHGs is emissions of CO₂ due to soil carbon changes resulting from changes in land management and land use change (LUC), i.e. transformation of land from one use to another (Goglio et al. 2015). Additional emissions arise in the production of inputs to agriculture and in processing, refrigeration, packaging, storage and transportation of food commodities. In addition, use of refrigerants, e.g. for storing wild-caught fish on vessels, causes emissions of the hydrochlorofluorocarbon R22 (HCFC-22) (Ziegler et al. 2013).

Carbon taxes or emission trading systems (ETS) covering 13% of global GHG emissions have already been implemented in 40 countries, including Sweden (Författningssamling 1994; World Bank et al. 2016). These carbon-pricing mechanisms cover some of the CO₂ emissions associated with food production, namely those from the use of electricity and fuels. However, emissions of CH₄ and N₂O are currently untaxed except for N₂O emissions from mineral fertiliser production in the European Union (EU), which are included in the EU ETS. The compound HCFC-22 is not subject to any tax, but its use is now banned under an EU directive (European Parliament and Cote 2014) and it is thus currently being phased out. To the best of our knowledge, specific climate taxes on food have not yet been levied in any country.

To obtain an ideal cost-efficient tax on GHG emissions, all emissions, irrespective of source, should be priced at the same level, as mitigation options will then be implemented in sectors where emissions can be reduced at the lowest cost (e.g. Baumol and Oates 1988). Therefore, taxes should ideally as closely as possible reflect the varying emissions generated when producing different products using different technologies. In the case of food, this would include differentiating between, e.g. vegetables grown in energy-demanding heated greenhouses and those grown in open fields and between different energy sources (e.g. fossil or renewable energy), in order to correctly reflect the true emissions from production. In addition, double taxation should be avoided by considering existing carbon taxes from the use of electricity and fuels in food production when establishing a tax on different food groups (Gren et al. 2019).

However, due to the high number of different food products available on the globalised food market of industrialised countries and the rapid development of new products, it is unlikely that a climate tax on food in practice could be based on detailed LCAs on each and every food product on the market. The cost of calculating and verifying the climate impact of all individual products would be unreasonably high (e.g. The Guardian 2012). For the same reasons, it is unlikely that differentiated charges for different technologies used during production can be implemented. Although it is possible to use national statistics to establish national averages for different foods from different countries, differentiating taxes based on country of origin of the product would not be a viable option, as that would likely violate the ‘most-favoured nation’ principle of the World Trade Organization (WTO). This principle states that members of the WTO must apply the same trading rules to all other WTO members, so differentiated tax rates for different countries would risk being discriminatory (Bähr 2015). Based on this, it would probably be necessary to base a tax on broader, more aggregated food groups.

It is important to also note that excise duties seldom are set so that they reflect the true externalities. Rather, they are a result of political negotiations including other considerations of the tax, such as being administratively simple (Government of Sweden 2009).

The proposed method uses a top-down approach to account for the GHG emissions associated with production of food from different countries available on the Swedish market. These are then aggregated based on import statistics to establish an average value for food sold on the Swedish market (Fig. 1). This approach enables the use of primary site-specific data for the most influential parameters in the calculations, such as yield levels and slaughter statistics, which are easily available in the form of datasets on national level for many countries (e.g. in national official statistics, National Inventory Reporting or national guidelines and reports from advisory services) (Fig. 2). This means that input data on the most influential parameters can easily be updated on a regular basis and that the results reflect the average carbon footprint of that food group on the Swedish market.

While primary site-specific data are often readily available for Sweden and many European countries, there are also cases when such data is not available for certain countries or food groups, e.g. fruits imported from South America. Data is then collected by the following order of prioritisation based on availability: data from the World Food LCA Database (Nemecek et al. 2015), peer-reviewed LCA studies and LCA reports. The World Food LCA Database is considered suitable, as it is available through the Ecoinvent database version 3.5 (Ecoinvent Centre 2018) and covers a variety of food products on the global market and offers access to the input data. Note however that using these data sources is a fallback solution that is used for food groups where primary data is not available. With future improvements in data availability, this can easily be changed.

For emission sources that are of less importance to the final results (e.g. packaging and electricity use in processing), standard values based on both primary and secondary literature data sources (carefully chosen to be representative for foods on the Swedish market) are used, which can simplify maintenance of the datasets. To further simplify maintenance, processes making a minor contribution to the final results (seed and seedling production, pesticide and mineral phosphorus and potassium fertiliser production, energy use for storage at wholesaler and retailer) are excluded. All input data to the model and further justifications for data choices are presented in the Electronic Supplementary Material.

To align the method as far as possible with existing efforts to make LCA calculations consistent in the LCA community, thereby maximising acceptability and trust, the method follows the ISO 14067 carbon footprint standard (ISO 2013) for everything except the exclusion of biogenic carbon uptake, as this carbon is released to the atmosphere when the food is consumed.

There have been several initiatives to standardise calculation of the climate impact of products and foods in particular (e.g. BSI 2008, 2012; Food SCP RT 2013; European Commission 2016). However, none of the available standards are sufficiently specific and detailed to be used in the present application without modification. For example, most existing standards are based on the use of product category rules (PCR), which are sets of specific rules applied to different products (e.g. PCRs for beer, olive oil and meat have been developed within the EU’s product environmental footprint (PEF) initiative (European Commission 2016)). For taxation, a method is needed that is consistent across food groups, not just within food groups.

Figure 5 shows how the choice of system boundary affects the climate impact of an illustrative set of foods, using the system boundaries in Fig. 3 (for all food groups, see Electronic Supplementary Material). Limiting a climate tax on food to agricultural emissions of CO₂, CH₄ and N₂O would exclude on average 46% of emissions for pork and 50% for chicken but only 18% for beef and 32% for milk (see Electronic Supplementary Material). Emissions from agriculture also dominate the climate impact of rice, due to high emissions of CH₄ from flooded rice paddy fields. For plant-based foods grown in greenhouses on the other hand, e.g. tomatoes, emissions from energy use are a major contributor in the life cycle. For tomatoes, taxing only agricultural emissions or full cradle-to-retail emissions would result in a very different tax.

As mentioned, taxing only agricultural emissions would probably be the easiest option with respect to administration. In addition, it would target emissions that are currently untaxed and avoid double taxation of products. However, it would risk excluding many of the emissions from e.g. plant-based foods, such as packaging, processing and transportation, which might lead to less understanding and acceptance of a tax as these processes are often perceived by consumers to have great influence on the overall climate impact (Shi et al. 2018). Thus, choosing to tax either up to farm gate or retail gate might gain more understanding and acceptance of a tax. Using set values for processes after farm gate, as suggested here (‘General model choices and input data used in the modelling’ section), would ease the administration and implementation of a tax from cradle to retailer.

In conclusion, the choice of system boundary for a tax involves a trade-off between ease of administration and accuracy of emission levels, where the latter is important for cost efficiency and acceptance.

The ICBM model used here to assess the ‘soil carbon change potential’ (section 3.3.5) requires input of regional data, which is lacking for many food production systems outside Sweden. The Tier 1 approach of the IPCC NIR guidelines for assessing soil carbon changes, on the other hand, is based on fixed factors for different levels of land management, tillage intensity and inputs of organic material (IPCC 2006). As the IPCC method is very simple and could be used consistently for foods from different world regions, we tested if this would be a suitable method for accounting for soil carbon changes. As in the ICBM approach, the ‘soil carbon change potential’, annualised over 100 years, was assessed from cultivation of 1 kg of cereals, legumes, oilseed rape, root vegetables and grass-clover ley. Figure 6 shows the results of the two different approaches for an illustrative set of food groups.

For ruminant products, the climate impact was expected to decrease due to the large proportion of grass in ruminant diets. The results from the ICBM modelling followed this assumption, with a decrease in the overall impact of beef by 5% and milk by 3% (Fig. 6). When using the IPCC Tier 1 approach however, the climate impact of both products instead increased (Fig. 6), as the emission rates for annual crops used in ruminant diets exceeded the sequestration rate of ley (Electronic Supplementary Material). Hence, the IPCC approach failed to capture the sequestration potential of Swedish grass-clover leys that is well described in the literature (Poeplau et al. 2015), while the ICBM approach gave results in line with empirical data for the Nordic countries (Kätterer et al. 2013). Hence, using the ICBM approach is a better choice in this case, although it might not correctly reflect the soil carbon changes associated with imported products. Due to the uncertainties in accounting for soil carbon changes, it could be argued that these emissions should be excluded in a climate tax on food. However, as accounting for soil carbon changes may substantially affect the carbon footprint of foods, we argue that including the emissions or sequestration in an uncertain way with average values is more accurate than excluding them entirely. As LCA studies neglecting to account for soil carbon have been criticised repeatedly by researchers (e.g. Stanley et al. 2018) and by proponents of grass-fed ruminant production (e.g. Rundgren and Bremen 2017), including soil carbon changes can be especially important for acceptance of a climate tax on food, as it decreases the difference between ruminant and monogastric meat.

A general limitation of both models is that it is far from certain that the emissions or sequestration of carbon attributed to different crops will actually take place (compared with CH₄ emissions from enteric fermentation, which are known to take place), as the methods are based on average regional conditions rather than site-specific conditions. For example, much ruminant production takes place on soils that are already high in carbon, as cropping systems on these farms have been dominated by ley for a long time, and these soils therefore have limited potential for sequestering additional carbon. Conversely, many annual crops are cultivated on soils less rich in carbon and these soils would not emit as much carbon during cropping as ‘average’ carbon-rich soils.

As previously mentioned, excise taxes should in theory be designed according to the marginal damage and associated cost of the climate impact. The marginal damage cost can be set on separate GHGs (one price for CO₂ emissions, one for CH₄ emissions and so on) or on CO₂e based on e.g. GWP₁₀₀ (Marten et al. 2015). Figure 7 shows the marginal damage cost using different approaches for weighting different GHGs: GWP₁₀₀, the Global Temperature Potential for 100 years (GTP₁₀₀) (Myhre et al. 2013), i.e. based on the temperature change caused by the different gases in 100 years, and GTP factors based on Persson et al. (2015). Persson et al. (2015) argue that using a 100-year time horizon for GTP is not in line with the Paris agreement target to limit warming to 2 °C, as this warming is expected to be reached within 100 years (at some point between 2050 and 2100). They therefore suggest using GTP corresponding to times when the target is expected to be met and to account for uncertainty in these timings. For GWP₁₀₀ and GTP₁₀₀, the cost both with and without inclusion of climate-carbon feedbacks for CH₄ and N₂O is shown. Further, the marginal damage cost of separate GHGs based on the cost estimates by Marten et al. (2015) is applied.

Differences in how the GHG emissions were weighted were most pronounced for beef, cheese and rice, whereas for other products such as potato, the different weighting approaches had minor effects. This is due to the large emissions of CH₄ that arise from the production of these products and to the difference between the characterisation factors used for CH₄. With the exception of rice, non-ruminant products generally cause more emissions of CO₂ and N₂O, for which weightings according to different characterisation factors were quite similar (Electronic Supplementary Material). Including climate-carbon feedback increased the climate impact for all products, but the differences were more pronounced for products causing large emissions of CH₄ (Fig. 7).

Implementing a tax based on the marginal damage cost of each separate GHG is argued to be most cost-efficient as an optimal climate tax should be differentiated between GHGs due to their different climate impact (Gren et al. 2019). Current climate reporting is however based on GWP₁₀₀, why using GWP₁₀₀ would be an advantage for acceptance and ease of administration. And as the difference between using GWP₁₀₀ in taxation and taxing GHGs separately is small in most cases, as shown in Fig. 7, we choose to use GWP₁₀₀ in the suggested method.

Both GWP and GTP rely on an arbitrary choice of time horizon, often fixed to 100 years, which can affect climate mitigation priorities. Using GTP factors as suggested by Persson et al. (2015) would be a more accurate option in order for a tax to move towards an actual climate goal. However, using GWP₁₀₀ for a climate tax would be more consistent with how climate policy is developed to date. Including climate-carbon feedbacks for all GHGs and not only CO₂ in GWP factors would introduce a trade-off between consistency in the methodology and accuracy, due to the uncertainty of the calculations. As recommended by UNEP/SETAC Life Cycle Initiative (2016), our proposed method includes climate-carbon feedbacks, for consistency in calculations.

The method suggested here to calculate the climate impact values has several limitations. It could be critiqued both for being overly complicated, including too many (unimportant) processes and requiring too much input data, e.g. for packaging and transports, and for being overly simplified, e.g. not distinguishing between different processing or packaging alternatives. The method is the result of a balancing act between completeness in order to be acceptable and ‘correct’ and simplicity in calculations and maintenance. Ultimately, how taxes are set is the result of political negotiations and compromises, the final result not always reflecting what is ‘optimal’ or ‘correct’. The transparent method presented here can however provide valuable input to such discussions.

As for data availability, this is also a challenge for certain production countries, food groups and production systems, especially as regards agricultural statistics and data on regions outside Europe. For food groups such as fish and seafood, we were unable to find any official data on key parameters such as fuel use and landed catch of fishing vessels. Hence, for some products, input data for the present study had to be taken from the inventories in earlier LCA studies. However, ongoing initiatives in compiling databases for fish and seafood may facilitate future assessment (e.g. Parker et al. 2018). Tracing food trade is also a challenge; trade statistics may hide the true country of origin, which hinders country-specific assessments for some food products and assumptions regarding the export country have to be made.

Regarding the maintenance of the dataset, this can also be a challenge considering the amount of data that is required. However, input data on the most influential parameters can easily be updated on a regular basis, as these come from reports that are compiled by authorities for other purposes. For emission sources where there is limited data availability or where set values are used, updates can be made less frequently. Detailed suggestions for maintenance of datasets are presented in the Electronic Supplementary Material.