Food system emissions: a review of trends, drivers, and policy approaches, 1990–2018

The food system, spanning from pre-production processes to post-production stages, is responsible for about one third of global greenhouse gas emissions and requires significant mitigation efforts to prevent dangerous levels of global warming. This article summarises trends and drivers of global food system emissions from 1990 to 2018. We highlight regional diversity in patterns of food system emissions and identify the highest global emitters. While food system emissions have stabilised in some regions and countries, global emissions are increasing, with growth in certain sectors and countries outweighing the handful of cases where sustained emissions reductions have been realised. Emissions from livestock rearing account for a large portion of global emissions, and the contribution of post-production emissions is steadily increasing in all regions. We also provide an overview of food system policies at the national level, mapping them to each emissions segment. This highlights the significant shortfall in policy activity required to address the challenge of climate change mitigation in general, and the impacts of livestock and post-production emissions in particular. Our work lays the groundwork for addressing specific country-level questions on optimal policy pathways to achieve emission reductions.


Introduction
The food sector provides for the daily nutritional needs of 8 billion people, while generating income and employment for a large portion of them. It is also a major source of global greenhouse gas (GHG) emissions (Crippa et al 2021, 2022. If countries are to realise the Paris Agreement goal of limiting global warming to well below 2 o C, the food sector-like all others-must be quickly and comprehensively mitigated towards net zero emissions by the middle of the 21st century. GHG emissions occur across the entire supply chain of food production and consumption (figure 1). They are released when land and forests are cleared for agriculture (and vice-versa, are absorbed when those lands are abandoned or reforested); they occur during on-farm production such as when livestock animals are reared, or when fertilisers are applied to soils; and they are released when fossil fuels are combusted to supply energy on-farm and for processing, packaging, transporting, retailing, consuming, and disposing of food products. All major categories of greenhouse gases-CO 2 , CH 4 , N 2 O and fluorinated gases (F-gases)-are produced by the food system, and all countries produce food system emissions, to varying degrees.
In this article we examine trends and drivers of food system emissions from 1990-2018 and provide an overview of policies in place in the food and agriculture industry. We build on several important advances in the literature. Recent work based on the Emissions Database for Global Atmospheric Research , estimated that in 2015 food systems represented 34% of total GHG emissions, and that a large portion of these emissions are attributable to land use and to agricultural production (Crippa et al 2021). More recent estimates by the same authors confirm the same figures also for 2018 (Crippa et al 2022). This magnitude of food system emissions impacts is also supported by recent work by FAO (Tubiello et al 2021b). An earlier review by Vermeulen et al (2012Vermeulen et al ( ) estimated that, in 2008 and 29% of total greenhouse gases were produced by the food system, of which 80%-86% were emitted at the production stage. Other studies have contributed to a better understanding of the drivers of food system emissions. Hong et al (2021) focus specifically on land use and land use change. They highlight how an extensive growth of agricultural production and disproportionate emissions from specific products (e.g. beef and other red meat) are among the main drivers of GHG emissions. Nonetheless, there is significant heterogeneity in patterns of emissions at the production stage, both due to varying types of commodity production, as well as varying efficiencies in producing them (see Poore and Nemecek 2018 and, for a recent synthesis, see Mbow et al 2019 andRosenzweig et al 2020). Along with trends and drivers of food system emissions, we also present an overview of the policies that have been put in place in the food system space. We distinguish policies based on the stated target, to see how much attention climate change mitigation in food and agriculture receives from policymakers, and on which stages of the food chain this attention focuses. This, similarly to the approaches proposed by, e.g. Meerman et al (2016), Fanzo et al (2020)and Jiang et al (2021), aims at providing a picture of policymaking in a specific sector, in response to the need (see Marshall et al 2021a) for accurate data for evidencebased policymaking.
The purpose of our work is to both update and extend this literature, providing further insights into the drivers of global and regional food system emissions trends. The specific aims of our analysis are to: (1) report total food system emissions from 1990 to 2018, with a focus on recent (2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) trends by greenhouse gas, stage, region, and high emitting countries; (2) analyse drivers of recent regional trends in production, land use, and nonfarm emissions; (3) examine progress in the adoption of food system policies worldwide, in the context of these on-going emissions trends. Beside the analysis, a further goal of this study is to present data sources that can be jointly analysed, enabling a comparative approach to identify effective food systems policies. Transparent, unbiased, and comprehensive data are essential for evidence-based policymaking. In this study, we acknowledge the significance of these aspects and provide, along with our analysis of trends and drivers, also an overview of data sources that can be used for quantitative causal analyses, allowing for the identification of effective policy interventions.
Our analysis draws from the Emissions Database of Global Atmospheric Research food model (EDGAR-FOOD), which includes estimates of land use, land-use change and forestry emissions from FAO (2021) database and on a collection of food system related policies collected from several online public databases and harmonised. We use a recent version of EDGAR-FOOD, which includes limited methodological changes from the version presented in Crippa et al (2021) and extends the time coverage to 2018. We also use a recent database of food system related policies, described in Lowder et al (2023). Section A of the supplementary materials provides more detailed descriptions of both data sources and the methods used to build them, as well as information on their limitations and any changes from the previous versions of EDGAR-FOOD. In the following sections we structure our results in terms of global, sectoral and regional trends and drivers. We then analyse more in depth top emitters and policy progress.

Results and discussion
2.1. Global trends Total food system emissions in 2018 were 17GtCO 2 eq, an increase of 16% (+2.4GtCO 2 eq) since 1990 and of 2% since 2010 (+0.5GtCO 2 eq). Compared to non-food GHG emissions, which increased by 74% (+18.2GtCO 2 eq) since 1990 and 14% (+5.0GtCO 2 eq) since 2010, this is a relatively modest rate of growth. The share of food system GHG emissions in total GHG emissions has declined as a result, from 40% in 1990 to 31% in 2018.
Global food system emissions comprise of primarily CO 2 emissions (7.7GtCO 2 eq in 2018, 46% of total), followed by CH 4 (6.3GtCO 2 eq, 38%), N 2 O (2.3GtCO 2 eq, 13%), and F-gases (0.6GtCO 2 eq, 3%). Of the CO 2 emissions component, approximately half again (3.8GtCO 2 , 23% of total) are related to LULUCF emissions, while the rest are energy, waste and industrial production related emissions (figure 1). Emissions from these different gases are growing at different rates-the fastest from F-gases which were, in 1990, almost nonexistent, at 0.0004GtCO 2 equation However, this only includes HFCs, SF6, PFCs and NF3-not CFCs, which were very high in the 1990s but have since declined due to successful regulation under the Montreal Protocol. Albeit from a very low base, they were the fastest growing gases in the food system (+0.6GtCO 2 eq since 1990 and +0.3GtCO 2 eq or +77% since 2010), followed by CH 4 (+1.1GtCO 2 eq or +20% since 1990 and +0.4GtCO 2 eq or +7.5% since 2010), N 2 O (+0.5GtCO 2 eq or +29% since 1990 and +0.1GtCO 2 eq or +5.5% since 2010) and CO2 (+0.21GtCO 2 eq or +2.8% since 1990 and −0.4GtCO 2 eq or −4% since 2010). As a consequence of these varying growth rates, the composition of gases in food system emissions changed over time. The share of CO 2 emissions moved from 52% of the total in 1990 to 46% in 2018, while the share of CH 4 emissions increased from 36% in 1990 to 38% in 2018. N 2 O emissions, mainly driven by organic and inorganic fertiliser use in agriculture, also occupy an increasing share of total food system emissions, from 12% to 13% in the same time frame. The largest share increase can be observed in F-gases, which had an almost irrelevant contribution in 1990, and account for 3% of the total as of 2018. Table 1 provides a summary of emissions separated by gas, stage of production, global region, and top contributors.
The food system is a key sector for near term climate mitigation, due to its large share of global CH 4 (60%) and N 2 O (84%) emissions, which have higher short-term warming impacts. Taking into account a shorter time horizon of warming effects, e.g. with a Global Warming Potential conversion of 20 years instead of 100 years, the large volume of CH 4 and N 2 O emissions in the food system increases its contribution to more than 40% of total net anthropogenic emissions in 2018 (figure 2).
Emissions are growing at different rates depending on the food system stage. The LULUCF emissions data suggests an overall decrease since 1990 (−1.1 GtCO 2 eq, −22%) and 2010 (−0.6 GtCO 2 eq, −13%), however extremely high uncertainties may prevent a meaningful assessment of these changes. Production emissions have gradually increased since 1990 (+1.1 GtCO 2 eq, +18%) and 2010 (+0.4 GtCO 2 eq, +6%). All other post-production emissions stages are growing at an overall rate of +2.4 GtCO 2 eq/yr (+76%) since 1990, and +0.63 GtCO 2 eq/yr (+13%) since 2010. As of 2010-2018, the fastest growing postproduction components were retail (+0.3 GtCO 2 eq, Table 1. Food system emissions by stage of production, greenhouse gas and region. Column 1 reports the average emissions per year during the 2010-2018 period. Column 2 contains the emissions of the year 2018. Column 3 details the average growth rate in the 2018-2018 period. Column 4 indicates how much the respective row accounts on the total emissions of that category (e.g. in the first line, the share of food system CO2 emissions on total emissions; in correspondence with Africa, it is the share of food system emissions on total in Africa). Column 5 details the contribution of the row to the total food system emissions.
( +41%), end of life (+0.2 GtCO 2 eq, +13%), transport (+0.1 GtCO 2 eq, +12%), and consumption (+0.04 GtCO 2 eq, +10%), while processing and packaging had rates of +4% and +3%, respectively. As a result of this fast growth, post-production emissions now occupy a much larger fraction of total food system emissions: 32% in 2018, compared to 21% in 1990. LULUCF emissions are predominantly driven by deforestation, which comprise 74% of the total in this stage. Indeed, in accounts of total LULUCF emissions (including non-food system components), deforestation is still the single largest driver of land use change CO 2 emissions (Houghton and Nassikas 2017). Overall there is agreement in the literature that deforestation emissions are on a decreasing trend (Tubiello et al 2021a, Flammini et al 2022, even though net forest conversion continues apace, with global primary forest areas shrinking by 13% (−7.25 million Km 2 ) between 1990 and 2018, partially offset by growth in secondary forests (+22%, +5.73 million Km 2 ) (Hurtt et al 2020, Lamb et al 2021b. Since land is the main limiting resource for agricultural production, ongoing net positive deforestation reflects both an increase and a change in the global demand for food produce-with the increase driven by population growth, and the change driven by a dietary transition towards high land intensity produce such as meat (Schmidhuber and Shetty 2005, García-Dorado et al 2019, Milford et al 2019. Improvements in agricultural productivity have reduced relative pressure on land resources, but so far not to the extent that LULUCF CO 2 emissions have realised consistent annual reductions with a high certainty (Hong et al 2021).
Production emissions occur mainly on farms during livestock raising and crop cultivation. The largest subcomponent at this stage is enteric fermentation from dairy and non-dairy cattle, which accounts for close to 40% of total emissions in 2018. Other large subcomponents are managed soils and pasture (19%) and rice cultivation (14%), with the rest attributable to manure management, fertiliser application, and energy-related emissions from farming and fertiliser production. Globally, production emissions are on the rise for the same reason that deforestation continues: more food is being produced to satisfy a rising global demand (from a total calory equivalent of 1.01 × 10 10 mkcal in 1990 to 1.83 × 10 10 kcal in 2018), the impacts of which have not been significantly offset by improvements in agricultural efficiency (Hong et al 2021). The global beef industry, alongside other red meats and dairy, is notable for its outsized impact in this segment, as it requires significant land and feed resources, and produces a large portion of CH 4 emissions, despite supplying only 1% of global food calories (Gerber et al 2013, Springmann et al 2016, Hong et al 2021. The consumption of these products is highly weighted towards developed countries-and to high-income households in developing countries, where a rapid dietary 'westernization' is taking place (Caballero and Popkin 2002, Popkin 2003, 2015, Pathak et al 2010. Post production emissions take place after the farm gate and are driven by processes of economic development and the global modernization of food systems. Unlike in the case of production and LULUCF CO 2 emissions, there is a strong relationship between national gross domestic product (GDP) per capita and post-production emissions (see figure 4). As developing countries have urbanised, the development of their retail and food service sectors, which moved from the pure sale of commodities to the commercialisation of products, has been shown to drive rearrangements in their food supply chains, for example inducing more processing and packaging, and longer transportation distances (see, e.g. Reardon and Timmer 2007, DeFries et al 2010, Liu et al 2017, Reardon and Zilberman 2018, Borsellino et al 2020. In particular the retail segment stands out for its very rapid growth, emitting close to four times the GHG emissions in 2018 compared to 1990. Much of this growth is attributable to a surge in F-gas emissions, which now accounts for 64% of the retail total in 2018. This is a direct indication that contemporary food systems are highly dependent on refrigeration to preserve products along (increasingly complex) supply chains and at the point of sale (Heard andMiller 2018, Dong et al 2021).
These changes are steadily shifting the main contributors to global food system emissions since 1990, reducing the share of emissions from Developed countries (from 24% to 20%) and increasing it from developing regions. This shifting global distribution of food system emissions is consequential for strategies to limit short-term warming impacts, as non-CO 2 GHG emissions comprise a greater proportion of emissions in developing regions. Indeed, the rapid growth of CH 4 and N 2 O emissions reported in section 2.1 is mainly attributable to Africa and Asia and developing Pacific, and from the shift away from LULUCF CO 2 in Latin America towards an increase in CH 4 from agricultural production. As such, efforts to combat short-term warming by mitigating these gases would require a focus on these 3 regions, which now account for 72% of all non-CO 2 food system emissions.
LULUCF CO 2 remains a substantial component of food system emissions in Africa (48% in 2018) and Latin America and the Caribbean (40%). In the latter, LULUCF CO 2 emissions are decreasing (−43% from 1990 and −48% from 2010), as measures against deforestation in the Amazon started to bear fruit in the 2000s and 2010s. Much more recent data in the period 2018-2020 indicate that this trend may have reversed as policy measures were abandoned following the Brazil election of 2018 (Escobar 2020). On the other hand they have steadily increased in Africa (+27% from 1990 and +23% from 2010), and in Asia and the developing Pacific since 2010 (+32%). In both cases LULUCF CO 2 emissions are linked to the clearing of carbon-dense tropical forests in the equatorial regions of Central Africa and South East Asia-trends that are driven by rising regional and global demands for food (and land-intensive food types), alongside slower regional gains in agricultural productivity (notably in Africa) (Hong et al 2021).
The production stage accounts for at least 25% of food system emissions in all regions. No regional reductions in production emissions have been observed, with the exception of the Developed countries (−10% from 1990 and +2% from 2010) and Eurasia (−36% and +11%) (the latter coinciding with the dissolution of the former Soviet Union). The slowest increase occurred in Asia and the developing Pacific (+25% from 1990 and +5% from 2010), while the fastest was in Africa, which almost doubled its 1990 emissions (+95%) and increased by 18% from 2010. Production emissions arise in part due to agricultural intensification: the use of synthetic fertilisers has increased per hectare yields, but comes at the cost of increased N 2 O emissions (Fowler et al 2013). At the production stage, the increase of livestock (and particularly cattle) has also been central to the surge in emissions in Africa, with a growth of emissions from enteric fermentation of 93% from 1990, and in Latin America, with a growth of 38% in the same period. Cattle rearing releases direct emissions from enteric fermentation and manure management, whether used as fertiliser or deposited on pastures. It also releases indirect emissions through the production of feedstocks and associated land use, which are not reflected in this data (Eshel et al 2014, Mottet et al 2017). In general, a trend of increased consumption of animal products has been observed in all regions except from the developed countries, driven by increasing incomes (OECD 2021a).
The less dominant role that production emissions play in developed countries and the Middle East is also an artifact of the global trade in food commodities. The food system emissions estimates presented here are territorial-based and do not take into account imports or exports of commodities and their supply chain emissions impacts. Consumption-based analysis of food system emissions estimate that Latin America, Asia and developing Pacific, and Africa export respectively 23%-34%, 44%-49% and 9%-32% of their emissions from the land use and production stages (Pendrill et al 2019). Conversely, the US is the largest importer of livestock products (12.8%-15.8% of all traded livestock, followed by Europe (Sun et al 2020). If the deforestation embedded in animal products was allocated to final consumers, the per capita emissions in Europe could rise by one-sixth (Sun et al 2020).
Post-production emissions have been increasing in parallel to economic growth in every region since 1990, with the fastest growth in the Middle East (+252%), Asia and developing Pacific (+136%), Africa (+133%) and Latin America (+89%). Post-production emissions growth has been more moderate in middle and high income countries (+31% in Eurasia and +12% in Developed countries). The growth is substantial even considering only the 2010-2018 time frame, with +37% in the Middle East, +28% in Africa, +20% in Asia and developing Pacific, +17% in Latin America and +7% in Eurasia. Only the Developed countries had a reduction in post-production emissions since 2010 (−4%), perhaps indicating a saturation effect as countries reach a very high level of per capita income. However, while some evidence suggests the existence of a Kuznets curve (a reduction after having reached a tipping point) in meat consumption (see, e.g. Cole andMcCoskey 2013 andVranken et al 2014), no evidence exists of a similar phenomenon concerning postproduction emissions prevalence.
Large differences among regions emerge when looking at per capita emissions. Latin America produces the largest amount of GHG per inhabitant, with 4.6 tCO 2 eq yr −1 , although this value has significantly decreased over time (it was 7.2 tCO 2 eq yr −1 in 1990). This is mostly led by the heavy carbon intensity of the agricultural sector due to the importance of livestock for exports (Mammadova et al 2020, OECD 2021b) and for the high internal demand (Alexandratos and Bruinsma 2012). The Developed countries and Eurasia follow with 2.9 and 2.8 tCO 2 eq yr −1 respectively. While in both regions the intensity of agricultural production (in relation with its GDP contribution) is decreasing, it is 2.8 tCO 2 eq in the Developed countries and 4.2 tCO 2 eq yr −1 in Eurasia. Emissions per capita in the Middle East (2.1 tCO 2 eq yr −1 ) are mainly to be attributed to the post-production stages. Agricultural production is both small and not very carbon intense. Africa, with 2.1 tCO 2 eq yr −1 per inhabitant, is the second region per carbon intensity of the agricultural sector. Asia and the developing Pacific is, in absolute values, the highest contributor to global food system emissions. When accounting for the large population, however, it is the region with lowest per capita emissions (1.7 tCO 2 ). The intensity of agricultural production is equally low: it has a lower prevalence of animal products and, among them, the larger share comprises of poultry and pigs (OECD 2021b). The share of emissions from rice cultivation in especially high (30% of production stage emissions). This is partly due to the large quantity produced for both internal demand and exports (Bandumula 2018) and partly due to high emissions caused by the agricultural techniques used (Yan et al 2009, Burney et al 2010. Figure 5 provides a comparison of food system drivers by region. In the figure we also highlight the 6 countries that, according to our data, produced 51% of all food system emissions in 2018. These countries are China, Brazil, USA, India, Indonesia, and the EU. Further details on their emissions, their trends, and the relevance of food emissions on their total emissions are summarized in table 1. The stages of origin  and their changes from 1990 to 2018 are depicted in figure 6. These countries account for, combined, more than half of the emissions from Asia and developing Pacific, Developed countries, and Latin America. As a consequence, they incorporate several of the trends highlighted on the regional level. Figure 5 highlights how food systems in different countries may share broad similarities. In particular, Hong et al (2021) distinguish between poorer regions with substantial and increasing emissions from land-use change, including Latin America, sub-Saharan Africa, and Southeast Asia, modernising countries with small emissions from land-use change but increasing emissions from agriculture, including East Asia, South Asia, and the Middle East, richer regions with negative emissions from land-use change but substantial and stable emissions from agriculture, including North America, Europe, and Oceania. The potential in reducing LULUCF emissions in the first of the three categories of countries is extremely high (Brazil alone had a 0.9GtCO 2 eq lower CO 2 emission from LULUCF in 2018, compared with 2010). However, sustained emission reductions must take into account the reliance on exports for the local economies and the necessity to safeguard other objectives, such as food security and nutrition. For the remaining two groups, improved management of agricultural inputs, soil management, and reductions in food waste might be relevant paths. However, groupings based on global regions or on singular indicators might not be enough to highlight broad patterns in place and fully consider all trade-offs behind mitigation strategies. To detect the similarities of food systems and the specific constraints they face, an analysis based on multiple indicators may be an appropriate direction to inform policymakers (see e.g. Fanzo et al 2020, Marshall et al 2021b).

Policies
The Food System Policy Database comprises 15 592 policies, across 195 countries (including dependent territories, such as Hong Kong and the Cook Islands), covering all stages of the food system analysed here: from LULUCF to production and post-production. In terms of the number of policies documented, there is a wide variation between countries. For example among the high emitters, Europe has the most documented 7 (2357 as of 2018), followed by Brazil (464), India (375), China (325), Indonesia (178), and USA (125). While the database is expansive-combining all major global food policy databases-it is not necessarily representative of all food relevant policies in a given country or stage, such as in federated countries where many policies are adopted at a regional level (e.g. in Brazil and India).
The database categorises the policies according to their stated objective. The main categories are dietary health, inclusion (which includes food security, along with measures reducing inequality in food access), and environmental concerns, which also include natural resource management measures. Among these categories, dietary health appears to be the most common policy objective on the global scale (38%), followed by the environment (23%), food access and inclusion (23%), and natural resource management regulations (6%). Figure 7 shows the most common policy targets among the top emitting countries. Dietary health is of great relevance for policymakers especially in the United States (39% of total) and in Europe (34%), while it represents a smaller share (from 22% to 26%) of all policies in the other countries. Environmental concerns are almost equally prominent in all top emitters (ranging from 10% to 17%) except from Brazil, where it accounts for 5% of the total policy coverage. An opposite situation can be observed for policies focusing on inclusion, of which Brazil has a large share (47%) and which represent a smaller percent (7% to 14%) in the remaining five. Finally, natural resource management regulations, not necessarily with an environmental focus, represent a smaller fraction of the total policy coverage, ranging from 1% to 10%. This is only a quantitative measure of the laws and measures (approved or implemented) explicitly stating these objectives, independently of their stringency or efficacy. This is, in other words, a measure of attention toward certain objectives but not a measure of progress toward them.
On the global level, most policies address the LULUCF and production stages (45% of total), which are indistinguishable in the database. The relevant underlying categories for these stages in the policy database are inputs (which encompass measures influencing the use of agricultural inputs such as fertilisers, fuel, seeds, and machinery), farming (which includes measures influencing agricultural production, both directly, as in subsidies and price measures, and indirectly, as with impact assessments and training programs) and natural resource management (which includes forestry, fishing, regulations on land and water use). Within the post-production stages, there is a greater weight towards trade (17% of total) and consumption oriented policies (10% of total). The data includes very few entries for transport (2%), processing (2%), and retail (2%), suggesting these segments may be less regulated, whether only environmental impacts are considered or not, at least for what concerns specific food system-related legislation. Finally, a significant portion of the policies either encompasses multiple stages of the food chain (5%) or is composed of regulations that apply to sectors both inside and outside of the food chain (16%), e.g. health strategies or regional development plans.
These trends are mostly still present when considering top emitters specifically. When observing all policies independently of their specific target, production-related policies represent the majority in most countries, mostly focusing on inputs and on-farm activities. This trend does not apply, however, to policies with an explicit environmental motivation. In this case, the majority comprises of natural resource management regulations, with the exceptions of Indonesia and Europe, where Please note that Other goals might also refer to policies without an explicitly stated goal. A substantial number of policies mention more than one goal, therefore the total does not coincide with the total number of policies.
production-related policies are still more prominent (24% and 72%, respectively). Finally, climate change appears to be in large part addressed through broad, wide-ranging plans and strategies. Two exceptions are Brazil, in which a large amount of natural resource management regulations are motivated by climate change, and Europe, where natural resource management, production, and general frameworks are almost equally prominent. Figure 8 shows the distribution of policies by target stage for the top emitters and for the rest of the world.
What proportion of documented food system policies address climate change specifically? A more specific search for climate relevant synonyms such as 'climate' , 'emissions' and 'deforestation' (for details on the word searches used, please see appendix A.8) finds that 1053 policies (7% of the total) list these outcomes as relevant objectives, or mention them in their policy descriptions. The majority of these policies (60%) are framework policies, which act on more than one specific segment and describe a broader strategy or objective rather than one specific intervention, while most of the remaining cover the LULUCF and production stages (37%). Among the top emitters, Europe has the largest number of climate change related policies (42), followed by Brazil and India (10), Indonesia (7), China (5), and the United States (4). It is worth noting that many food system policies will have environmental outcomes regardless of whether these are explicitly stated. These outcomes, however, are not guaranteed to be positive. Indeed, approximately half of all the documented production and LULUCF stage policies are supportive measures that aim to increase agricultural production, for example via fossil fuel and fertiliser subsidies, and hence may hinder mitigation efforts. In contrast with our findings, the existing literature emphasises the importance of implementing both supply and demand-side measures and-for many countries with high consumption of animal-source foods-should include a shift to diets with higher shares of plant-based foods (see Stehfest et al 2009, Clark et al 2020 Panel on Agriculture and Food Systems for Nutrition 2020, Temme et al 2020, Latka et al 2021), including emerging food technologies such as controlled environment agriculture and cellular agriculture. We do not find evidence of similar strategies being implemented on a broad scale.
Which policies are oriented towards particularly emission-intensive food system supply chains? A further keyword search for synonyms relevant to livestock rearing identifies 944 policies (6% of total). Unsurprisingly, these policies predominantly focus on the production stage (56%), for instance as input subsidies (feed distribution being the most common). A smaller part comprises of general regulations and trade measures (9% and 26% respectively). Again, it is certainly the case that many policies not specifically targeted at the livestock sector are relevant for it; but it is nonetheless noteworthy that one of the most widely documented emissions-intensive food sectors has not received more specific attention from policymakers. A similar observation can be made for fertiliser use. While the database lists 528 policies mentioning fertilisers, only 90 state an environmental goal, and 17 consider climate change, while the remainder mostly comprises of input subsidies and trade regulations.
We now examine the specific policy levers used more often in each country. These are the specific policy tools that affect the food system. As reported in table 2, market interventions targeting agricultural production (here referred as Agricultural market measures), are prominent in all top emitters excluding the United States and the EU. While some support of agriculture in the form of subsidies is important in all countries examined, EU and United States show a focus on rules/standards and labelling that is not found in the rest of the top emitters. Focusing on mitigation efforts, the table shows that climate change is never mentioned when considering the top levers adopted, whether they are measures supporting agriculture, lower prices of food, or affect trade. In Brazil, United States, and India, climate change is mainly mentioned in specific framework policies, which represent a very small percentage of the total. The same applies to Indonesia, where, instead, long term strategies addressing climate change are much more prominent. China addresses it in a number of rules on conservation of natural resources, which include protected areas and forestry management. Finally, the climate change implications of implemented policies seem to be often considered in the EU: even policies with goals different from-or potentially in contrast with-climate change mitigation, like production support or subsidies, take often into account the expected impacts on GHG emissions. The findings support the argument in favor of a global strategy toward decarbonising the agricultural sector. As the resources available to pursue this goal vary substantially between countries, with many governments across the globe not being able to divert many resources from the fight against poverty (see, e.g. Godfray et al 2010), coordination might prove necessary.
We look at the temporal distribution of the data, to investigate if policy activity on the climate impacts of the food system might have intensified over time 8 . The result, summarised in figure 9, shows a moderately growing trend for policies specifically addressing climate change, livestock, and fertilisers reaching a peak, in absolute numbers, between the years 2012 and 2015. The subsequent decline can be ascribed to the very limited policy coverage in 2017, 2018, and 2019, with 257 entries jointly for these years (compared with, e.g. 1032 policies for the year 2020). As a Table 2. Percentage of total policies in which a specific lever appears, for each country. When a lever is used in one or more cases to address climate change, it is reported in bold. Only the 10 most commonly used policy levers are reported.

Brazil
% Indonesia % Agricultural market measures 20.5 Agricultural market measures 12.9 Support for finance and credit 12.1 Import tariff 6.7 Knowledge-agricultural training 6.5 Import quota 5.6 Land ownership, tenure and titling 3.7 Import-other import restrictions 5. consequence, no conclusions can be drawn for these three years. However, to verify if more recent policies might target the specific areas of climate change, livestock rearing, and fertiliser use, we look at their prominence in the set of policies which do not report a year of implementation. This subgroup is composed by policies which are still active and, in larger part, recently approved/implemented. Among these, 15% explicitly mentions climate change, well above the peak observed in the rest of the sample in 2016.
The share is however lower for livestock (5%) and fertilisers (1%), indicating that these aspects might indeed still be inadequately targeted. These numbers are even more concerning if considering that only 29% of policies concerning livestock rearing, and only 4% of fertilisers-related policies mention an environment-related goal. A recent report focusing on OECD countries (OECD 2022a) draws similar conclusions to what our results suggest: more policy ambition is needed. Policies that end up subsidising polluting inputs (e.g. synthetic fertilisers) should follow a decreasing path: while overall intensification had likely contributed to lower agricultural emissions (Fowler et al 2013), synthetic fertilisers tend to be overused, and significant reductions could be achieved without incurring in decreases in yield (Blandford and Hassapoyannes 2018). Moreover, targeted interventions that can reduce emissions from the most contributing sectors (e.g. methane inhibitor food supplements for livestock) should be introduced (for more details, please see the specific report chapter: OECD 2022b). Approaches to reducing GHG emissions from agriculture have the potential, however, to affect other important goals, such as food security and economic development. For example, policies aimed at reducing deforestation and promoting reforestation can limit the availability of land for agriculture and livestock production, which can have negative impacts on food security (Frank et al 2017). To address these trade-offs, it is important to develop integrated approaches that balance the goals of reducing emissions with food security and economic development.
Our data highlight how only 2% of policies focus on the transport stage. While this might reflect that the segment is largely unregulated, many policies on trade deeply affect the segment. Export-promoting and, in general, trade facilitating policies can impact the relative profitability of internal demand and exports, leading to more or less emissions from transport. The most common regulations affecting this stage have to do with the sanitary/phytosanitary measures in place, which regulate the necessary requirements for the import/export of agricultural commodities. Moreover, the policy database exclusively refers to food system policies. In most cases, food-system related transport might be regulated as a specific case in national transportation laws or might not need specific regulations altogether. Similar considerations can be made on the extremely low coverage of policies addressing waste and, in general, to all areas where a clear distinction between food system emissions and general emissions is difficult. An increase in renewable shares in energy production, for example, would affect most sectors.

Conclusion
Emissions from the food system are higher than ever and on an upward trend in most regions. They are not growing as fast as non-food system emissions but, unlike energy system emissions, they are not substantively decreasing in developing countries (see Lamb et al 2021a). Any minor reductions achieved so far have been absorbed by growth elsewhere or in other stages of the food system.
A large portion of growth is coming from non-CO 2 GHG emissions, much of this in developing regions. China, India, Brazil, USA, Europe, Indonesia, Pakistan, Russia, Mexico, Argentina, Bangladesh and Nigeria account, together, for more than 60% of total non-CO 2 emissions. Excluding Asian countries, livestock production accounts for at least one quarter of all non-CO 2 emissions in all the countries listed. Post-production is now a major source of emissions growth, outpacing all other segments as food supply chains have emerged and continue to grow in developing countries. This has also caused a substantial increase in F-gases, which, from being almost nonexistent in 1990, are now a small but relevant part of food emissions. In regions where the share of GHG emissions from the post-production stages is still limited (e.g. Africa), this process of growth might be just at the very beginning, highlighting this as an important sector for interventions to prevent a lock-in of emissions intensive activities.
We observe a discrepancy between the global climate impact of agricultural production, the insight presented in the literature, and the policy attention given to the issue. At the global level, largest share of policies targeting the food system aims at sustaining the level of agricultural outputs through subsidies to inputs and to farm activities. This may in turn negatively affect their environmental performance (Blandford and Hassapoyannes 2018), revealing a focus mostly on food security and food access. Regulations looking at the environmental impacts of the production stage are in limited number and mostly regulate the access to natural resources like forests, water and fisheries. High impacting sectors, like livestock rearing, rice production, and fertiliser do not appear to be prominent part of national policy activities. A similar picture emerges when looking at the post-production sectors. Very few policies aim at reducing the impacts of GHG emissions in these later stages. Some policy actions may have indirect effects on emissions (e.g. the implementation of guidelines for healthy diets and the procurement rules for public food service), but targeted policy action is lacking across all regions.
This paper provides a first step towards linking food system emissions and a database of policy action. Further research might address the causality and impact of policies, assessing which measures and packages have been more successful in reducing counterfactual emissions. Given the significant emissions impacts of the food system worldwide-and the multitude of other services and social functions it provides-learning about both successes and failures in food system policymaking is of primary importance. Achieving the goals of the Paris Agreement will require renewed efforts to address food system emissions, despite the difficulties that policy action in this delicate and crucial system involves.

Data availability statement
No new data were created or analysed in this study.

Acknowledgments
This work has been supported by the Food System Economics Commission, funded by the IKEA Foundation, Grant Agreement No. G-2009-01682. The opinions, findings, and conclusions or recommendations expressed in this material are those of the author/authors and do not necessarily reflect the view of the IKEA Foundation. We thank Francesca Diluiso, Franziska Gaupp, Hermann Lotze-Campen, and two anonymous reviewers for the useful comments.