Study design and sample
We used purposive sampling to include the US and a diverse range of countries that would shed light on the variation in potential impacts of FBDG. We relied on the United Nations Food and Agriculture Organization’s (FAO) online repository of FBDG [1] to examine a list of such countries. Only FBDG available in English or Spanish were considered for inclusion in this study, as were those with a list of clearly defined food groups (e.g. grains) and daily quantitative recommendations for each food group, including defined weight- or volume-based serving sizes, which are needed to model greenhouse gas emissions. We selected two European countries, Germany and the Netherlands, for which earlier research had shown that diet shifts towards their guidelines would lower GHGE [26, 27, 38]. To seek maximum variation in GHGE, we chose India [39] and the US vegetarian FDBG [40], which do not include meat in their FBDG, since animal proteins are a major dietary contributor to this impact. Other countries, including Oman [41], Thailand [42], and Uruguay [43] were chosen for diverse geographic representation. Additional countries, including Australia, Belize, Malta, and Malaysia were considered, but they did not have either specific quantitative recommendations for all food groups (Malta) or defined weight- or volume-based serving sizes for all food groups (Australia, Belize, Malaysia). Finally, we also included the EAT-Lancet dietary recommendations, since these are global guidelines explicitly designed with environmental sustainability in mind [44].
Food-based dietary guidelines (FBDG)
In addition to the FAO repository, we used government websites to obtain additional details on country-level FBDG. Quantitative recommendations for discrete food groups were included in the analysis. “Discretionary calories” were included in the analysis based on each country’s recommendations. For example, Germany, India, and Uruguay specified additional intakes of sugar, so we included those quantities in our analysis. The US and the Netherlands included discretionary calorie amounts, but no specificity on the composition of those calories, so we increased GHGE to cover the discretionary calories assuming the same composition as the rest of the recommended diet. We did not include non-caloric items, such as coffee or tea, since there were no specific recommendations for them. Additionally, qualitative recommendations (e.g. “eat a variety of food to ensure a balanced diet”) were not included in this analysis. For consistency, the main dietary pattern equal or closest to 2000 kcal was selected from each country’s FBDG, regardless of the targeted age-sex group. If the main pattern was higher, as in the EAT-Lancet recommendation of 2500 kcal, or the Netherlands’ recommendation of 2053, or Uruguay’s recommendation of 2200 kcal, quantities were scaled to 2000 kcal. If recommended quantities of food groups were only listed in ranges, we used the mid-point of those ranges for our analysis, otherwise we used the specified amounts.
FAO food balance sheet data
Although the FBDG used in this study provide daily quantitative recommendations for clearly defined food groups, the guidelines do not provide recommendations for the specific foods within each group. For example, Thailand recommends 135 g/d of protein foods, but does not make further recommendations on how much beef, pork, or chicken should be consumed. There is tremendous variation in the global warming impacts of these individual foods [45], so item specificity within a food group is needed. To address this, we used national consumption patterns of specific foods within each food group. We relied on FAO food balance sheets to obtain these country-level patterns, specifically relying on the annual per capita food supply to humans. This is ‘apparent consumption,’ since it is obtained by summing the contributions to total domestic food availability (production, imports, and stock changes) and subtracting exports and other foods utilized for non-human purposes (feed, seed, processing, waste, exports, and other uses). The 2013 food supply quantity data (kilograms/capita/year) for 58 commodities in the protein foods, grains, fruit, vegetables, and oils/fats food groups in seven countries were used in this study [46]. The FAO food balance sheets did not provide sufficient data on dairy products for this analysis, so we used the 2013 Organization for Economic Co-operation and Development (OECD)-FAO Agricultural Outlook dairy data [47] in the same manner as the FAO food balance sheets for four types of dairy products.
Apparent consumption patterns were described by a series of proportions, that is, the food supply quantity of an individual food item (e.g. poultry) divided by the sum of quantities of all items in a food group (e.g. protein foods). These consumption patterns were different for each country and not all countries consumed all items in a group. Prior to calculating these apparent consumption patterns, commodity weights were converted to cooked edible weights. This required a series of factors to convert individual protein foods (i.e. meats, poultry, and fish) from carcass weight to raw, boneless weight [45] and then again to cooked, boneless weight using an average of dry and wet cooking method conversion factors [48, 49]. Conversion factors were also applied to remove the shells from groundnuts, nuts and products, eggs, molluscs, and crustaceans [45, 48] and also to convert legumes and grains from raw weight to cooked weight [50].
Environmental impact data: dataFIELD
Data on the greenhouse gas emissions (GHGE) for the production of different foods come from the database of Food Impacts on the Environment for Linking to Diets (dataFIELD), which is based on an extensive review of the life cycle assessment literature [45]. Articles and reports available in the public domain and published in English from 2005 to 2016 were included in this review, with cradle-to-farm gate impact factors used for the vast majority of the 332 commodities in this database. DataFIELD includes studies throughout the world, and although differences in production practices affect GHGE values, there is not sufficient research to identify GHGE values of foods disaggregated to the country level. Thus, we used the same mean GHGE values, expressed as carbon dioxide equivalents per kilogram of food (CO2-eq/kg), for all countries studied here, and applied them to 58 FAO commodities and four OECD-FAO dairy products. Aggregate values were created for certain commodities that included multiple foods (e.g. “nuts and products”, “freshwater fish”). GHGE data were not available for five other commodities and were excluded from our analysis. These commodities – cephalopods, aquatic animals other than fish or seafood, sorghum, palm kernel oil, and rice bran oil – contributed a trivial amount to consumption patterns, accounting for less than 10 kcal per capita per day, on average.
Carbon footprint calculations
To calculate the GHGE for each country’s FBDG recommendation for each food group ‘g’, we calculated a weighted average of the GHGE for the specific commodities in that group and multiplied it by the recommended daily amount of that group using the following formula:
$$ {GHGE}_g={\sum}_{c=1}^C{REC}_g\times {Proportion}_c\times {CF}_c\times {GHGE}_c $$
where RECg is each country’s FBDG recommended daily amount of group g, for example, 2 cups of fruit. Proportionc is the quantity share of each commodity c consumed out of the total food group quantity, for example, the proportion of oranges out of total fruits consumed, which comes from food balance sheet data for each country. CFc is a conversion factor for adjusting the recommendation to kilograms of commodity c, and GHGEc is the greenhouse gas emissions (kg CO2-eq) to produce a kilogram of this commodity. For the EAT-Lancet guidelines, the same procedure was followed using the apparent consumption pattern of the US. Finally, the GHGE of all food groups within a country were summed for the total GHGE that could be attributed to eating a diet as recommended by the FBDG of a particular country.
Conversion factors were used in several ways in the final calculations. For the FBDG that provided recommendations in units of weight other than grams (e.g. ounces), conversions were made using the standard conversion factor for ounces to grams. For recommendations given by volume (e.g. cups), we used food composition tables [50] that provided the necessary conversions to grams of an equivalent amount of food in calories. For example, a cup of black beans has 662 kcal and 100 g of black beans has 341 cal, so a cup of black beans would be equivalent to 194 g. For Thailand’s recommendations, which define one recommended serving of fruit as a “piece” of fruit, one piece of fruit was estimated to be equivalent to one cup of fruit. If the commodity group was aggregated, an average equal calorie conversion was calculated based on the individual foods included in that commodity group. Some commodities, including meat, fish, grains and legumes, were converted from cooked weights (used in the recommendations) to raw, edible weights (used in the global warming impact factors). Grains and legumes were converted using equal calorie conversion factors, and meat and fish and seafood were converted using an average of dry and wet cooking method conversion factors [49].
Determining differences in FDBG ‘controlling’ for national dietary patterns
Since country recommendations do not specify food items within a group, we used FAO food balance sheet data for each country to calculate these specific food consumption patterns, as described above. This means that differences in the carbon footprints between countries are partly due to each country’s current consumption pattern, as reflected in these FAO data. To control for this effect, we recalculated each country’s FBDG carbon footprint using the same food consumption pattern across all countries, that of the United States. At times, this required different aggregations of foods. For example, the protein foods group in India’s FBDG does not include meats, so we only considered plant protein foods – beans, peas, and legumes – when applying the US consumption pattern to the India FBDG.