After quantifying the national-level hidden costs of global agrifood systems, the next step of the TCA approach, as proposed in The State of Food and Agriculture 2023, is to engage with stakeholders at country level. This is an opportunity to validate the quantified hidden costs, acknowledge and potentially fill data gaps, and contextualize the challenges to address the hidden costs and the possible solutions to do so based on national priorities and commitments. This engagement with stakeholders is crucial if targeted assessments are to succeed in guiding policy actions based on TCA. A case study commissioned as a background paper for this report showcases the usefulness of combining scenarios based on stakeholder consultations with TCA to assess the implications for national hidden costs.
The study was conducted by the Food, Agriculture, Biodiversity, Land-Use and Energy (FABLE) Consortium, a global network of national research organizations developing national-scale food and land-use strategies aligned with national and global goals. For this study, the FABLE Consortium validated the findings of The State of Food and Agriculture 2023 and held consultations with stakeholders to identify nationally relevant desired outcomes to increase the sustainability of their agrifood systems. A set of future scenarios estimated the effectiveness of each desired outcome in addressing the hidden costs in the following countries: Australia, Brazil, Colombia, Ethiopia, India and the United Kingdom of Great Britain and Northern Ireland.
The desired outcomes identified included improvements in crop and livestock productivity, lower stocking rates (ruminant density) on pasture, and reduced post-harvest losses in all countries. In most countries, preventing deforestation beyond 2030 and increasing afforestation to meet official commitments to the Bonn Challenge (Brazil, Colombia, Ethiopia and India) or other national targets (Australia, India and the United Kingdom of Great Britain and Northern Ireland) are included in the national commitments and global sustainability scenarios. Dietary changes for healthier consumption patterns are also seen as key – except in Ethiopia, which is as expected, because health hidden costs account for only a small share (13 percent) of the country’s total hidden costs, which are dominated by social hidden costs (46 percent).54 Only a few countries include outcomes such as increased use of agroecological practices and irrigation, and changes in trade, biofuel demand, and population growth.
Three scenarios were assessed in each country: i) the current trends scenario is a low-ambition vision of feasible actions towards environmental sustainability, strongly dependent on historical trends and current policies; ii) the national commitments scenario reflects the actions needed to meet existing national commitments and targets; and iii) the global sustainability scenario corresponds to efforts compatible with achieving global sustainability targets.i Because of the large number of desired outcomes included in each scenario, the FABLE Consortium undertook a separate assessment of each to identify which would be most influential in reducing the hidden costs of agrifood systems.
The dietary indicators generated by these scenarios are expressed in terms of changes in the availability of food groups, which must be transformed into dietary intakes to be linked to the dietary NCD risk factors costed as health hidden costs. This link is non-trivial, as the way food groups are consumed (namely, fresh, processed or highly processed) has immense implications for dietary risks and NCD outcomes, as well as environmental impacts.55 To overcome this limitation, a machine-learning model was used to establish the link between food availability (FABLE model outcome) and dietary risks (linked to DALYs due to NCDs in the GBD data) to quantify the implications of the scenarios for health hidden costs (Box 7).
Box 7Description of the machine-learning exercise to link food availability to food intake
Most models used for scenario analyses only provide information on the quantities of different commodities produced, imported or exported every year in each country under different scenarios. However, what impacts the health of consumers is not the availability of food, but its actual intake, which can have an unclear correlation for various reasons.
Acknowledging this issue, The State of Food and Agriculture 2024 has estimated the health outcomes associated with the results of the FABLE simulations* using a machine-learning model. After extensive validation by The State of Food and Agriculture team and background paper authors, the model architecture selected was a mixed model using XGBoost, a method based on decision trees with good empirical performance in many fields, and a linear model.
The machine-learning model was used for dietary risks not easily associated with a specific food category in food availability statistics in FAOSTAT.** For example, it is difficult to link the excessive consumption of sodium with any major food group. Therefore, the links between food availability and intake were estimated using the machine-learning model for such food and nutrient groups. The model was trained on data on the availability of food from FAOSTAT and food-intake data from the Global Burden of Disease database, so it could learn the historical relationship patterns between the two quantities. Other controlling indicators that mediate the relationship between food availability and consumption were also used (for instance, ultraprocessed food and beverage sales by country, which proxy the way in which available food is processed).
For food groups whose supply (adjusted for trade and food loss and waste) could be directly matched to intake, the linear model was used. Specifically, the changes in available supply of fruits, vegetables, red meat, milk, legumes, vegetable oils, nuts and seeds were assumed to be proportional to the changes in intake used for their (disability-adjusted life year) DALY predictions. For example, an increase of 5 percent in the supply of vegetables (after adjusting for trade and food loss and waste) was assumed to result in a 5 percent increase in the intake of vegetables.
Although the machine-learning model provides an important missing link to facilitate scenario analysis on the impacts of changing diets, its use is limited in cases where the historical data used to train the model (based on past trends) and the context for which it needs to provide predictions (a future scenario that breaks the historical patterns) diverge significantly. Historically, countries have followed strong trends (for example, as they develop, consumption increases, not only of fruits and vegetables, but also of highly processed foods). When the targeted policy scenarios depart significantly from historical trends in the relationship between food production and intake, it is important to acknowledge that simply altering the food production mix is not sufficient to achieve transformation. Incorporating other levers that target food environments and behaviours is necessary, as discussed in Chapter 4.
SOURCE: Authors' own elaboration.
The results show significant variation from country to country in terms of which of the modelled outcomes is the most effective in reducing the quantified hidden costs of agrifood systems (Table 1). Drawing on the agrifood systems typology, however, an interesting pattern can be observed. For most of the agrifood systems studied in the industrial and formalizing categories, changing dietary patterns is not only the main means of decreasing the hidden costs due to the burden of disease, but it is also a very effective way of reducing the environmental hidden costs (due to GHG and nitrogen emissions and land-use change). Of the 11 hidden cost subcategories reported in Table 1, dietary change is the most influential outcome in Brazil and the United Kingdom of Great Britain and Northern Ireland in six subcategories. In Australia, it is most influential in four subcategories (in addition to calorie intake), including methane and nitrogen emissions and pasture use. Dietary change is found to increase the hidden costs of blue water use, highlighting the importance of combining it with improvements in crop productivity and reductions in food waste considered in the global sustainability scenario.
TABLE 1 Desired outcomes that are most effective in decreasing the hidden cost subcategories by country, 2050
SOURCE: FABLE. 2024. How to reduce agrifood systems' future hidden costs? A multi-country case study – Background paper for The State of Food and Agriculture 2024. Paris, Sustainable Development Solutions Network.
In Colombia, although improving diets was a desired outcome included in the scenarios, it is most influential only for reducing the hidden costs of nitrous oxide emissions (in addition to calorie intake). Improving crop productivity through the sustainable intensification of production emerges as most influential for five subcategories of hidden costs, including reductions in carbon dioxide and nitrogen emissions and land-use change.
Dietary change was also included in India, particularly a transition to the EAT-Lancet diet along with increased calorie intake to eradicate underweight by 2050. It was the most influential in decreasing four subcategories of hidden costs in the country, including through reductions in methane emissions (from livestock and rice), pasture, and blue water use. Curbing nitrogen runoff on croplands and managing land-use change emerged as other pivotal desired outcomes for hidden cost reductions in India.
Ethiopia, the only country where dietary change was not identified as an outcome to be modelled in stakeholder consultations, stands to benefit most from improved livestock and crop productivity, afforestation and limiting agricultural expansion into forested land to decrease environmental hidden costs. The potential actions to address social hidden costs due to poverty – the largest hidden cost in Ethiopia – were not well represented in the models used in this case study.
Overall, with the exception of Ethiopia, countries’ hidden costs under the national commitments scenario are not distinguishable from those under the current trends scenario when uncertainty is taken into account, although the former does reveal small reductions, on average. This suggests that countries should be more ambitious, striving to achieve reductions in the potential economic impacts of their agrifood systems, including levers for dietary change, which provide the clearest link to reducing overall hidden costs by freeing up land and reducing and sequestering GHGs and nitrogen, in addition to improving the accessibility of sustainable and nutritious diets for all.
The innovative machine-learning model applied to the simulations was helpful in breaking down the dietary risks associated with decreasing hidden costs so as to guide policy. The results highlight salient differences by agrifood systems category. The health hidden cost reductions between global sustainability and current trends scenarios in industrial countries are significant (−60 percent in Australia and −42 percent in the United Kingdom of Great Britain and Northern Ireland). In Australia, this is driven by an increase in consumption of nuts and seeds, fruits, legumes and vegetables and marked decreases in demand away from processed and red meat and SSBs. In the United Kingdom, this is mainly driven by lower processed meat consumption and higher legume consumption. The differences between these two scenarios are relatively lower in formalizing countries (Brazil and Colombia), with decreases in processed and red meat and SSB consumption explaining most of the reduction in hidden costs in Brazil, while decreases in processed meat and SSB consumption and an increase in legume consumption drive the reduction of hidden costs in Colombia. In traditional agrifood systems of India, healthier diets and avoiding a Western-style diet trajectory of overconsumption of sugars, salt and processed foods account for roughly two-thirds of the avoided health and environmental hidden costs. In Ethiopia, classified in the protracted crisis category, changes in health hidden costs are dwarfed by reductions in environmental hidden costs from improved production practices. Increasing the consumption of fruits, vegetables and whole grains should be envisaged – compared to the current diets modelled here – to further reduce the health hidden costs due to NCDs.
The role of stakeholder consultations in identifying nationally relevant sets of desired outcomes to be included in this case study is critical to the effectiveness of tailored assessments to guide decision-making. The overall recommendations of this case study further include the use of national datasets on land-use change and GHG emissions for tailored TCA assessments. Using country-specific thresholds for poverty and calorie consumption needs would also increase the relevance of the hidden costs to the national context. The consultations raised awareness among stakeholders and identified important data gaps, underscoring the need to invest in data collection, for instance, on nitrogen application and the value of ecosystem services in different locations. Lastly, using subnational statistics where such data are available was also highlighted as important for targeted TCA assessments to further facilitate effective policy design, especially in large countries with different agroecological zones and those with high in-country inequalities across relevant outcome indicators (Box 8). However, a limitation of this case study is that the scenarios focused on desired outcomes do not detail how these will be achieved.
Box 8THE NEED TO GO TO SUBNATIONAL LEVEL FOR TAILORED COUNTRY-LEVEL TRUE COST ACCOUNTING ASSESSMENTS
Biophysical characteristics and the spatial organization of a territory define the actions needed to transform agrifood systems for greater sustainability. Country-level results based on national average values are likely to over- or underestimate the magnitude of the impacts on hidden costs. For example, expanding the cultivated area of a certain crop would need to happen under much poorer agronomic potential, or targeting a specific area for ecosystem restoration could lead to greater-than-average benefits. Sometimes, a problem can even become invisible at national level, as it can be offset by other regions in the country. Therefore, depending on data and resource availability, national-level data should be complemented by spatial analyses to enable the assessment of heterogeneity in the main impacts and drivers of agrifood systems.
An example of a policy with targets that vary across territories is the Forest Code in Brazil. The code is one of the most important policies in place to regulate future land-use change and, consequently, whether large amounts of carbon dioxide are emitted or sequestered. The rules govern how credits can be traded between farms, offsetting deforestation above allowed levels with permitted deforestation avoided elsewhere, but they need to account for similarity in forest type and biodiversity, among other things.
Distinguishing between agricultural production systems, for example, based on farm size or intensification level would better capture the heterogeneity across food production systems at subnational scale. This might be particularly pertinent to countries such as Ethiopia, where small-scale farmers constitute 75 percent of the population and the diverse agroecological zones (from highland areas to very arid areas) offer differing potential to reduce hidden costs.
When inequalities within a country (for example, in incomes, access to healthy food, dietary patterns and infrastructure) are significant, subnational assessments are even more necessary. For example, in remote areas of Australia, food baskets cost 39 percent more than in major supermarkets in capital cities.56 Higher commodity prices can have a greater effect on populations that rely on extensive cattle farming or subsistence fisheries in remote areas.57 In India, underweight prevalence among children (under five years) varies greatly across states – from 40 percent in Bihar to 12 percent in Mizoram.58 Because inequality is not costed separately in the true cost accounting (TCA) approach used in this study, such national-level TCA assessments can mask key inequalities at subnational (population subgroup) level, which needs to be properly incorporated into policy design through consultations with civil society at the national and subnational level for inclusive transformation.
