Table of Contents
Statement of principal findings
Diet-related environmental impacts vary substantially by diet groups within this cohort of UK adults which includes a large sample of vegans, vegetarians and fish-eaters. For measures of GHG emissions, land use, water use, eutrophication and biodiversity, the level of impact is strongly associated with the amount of animal-based products that are consumed. Point estimates for vegan diets were associated with less than half of the impact of high-meat-eater (>100 g d−1) diets for all indicators, and 95% uncertainty intervals were below 50% for all outcomes except water use and biodiversity. There are also large differences in the environmental impact of diets for groups with lower (but still some) meat consumption. For GHG emissions, eutrophication and land use, the impact for low meat-eaters was at least 30% lower than for high meat-eaters. Large food-level variation in the environmental indicators due to region of origin and method of food production does not obscure differences between diet groups.
Implications of research
The UK has a legal commitment to a 78% reduction in GHG emissions by 2035 compared to 199017 and of halting biodiversity loss by 203018. The UK Committee on Climate Change has stated that if the government is to achieve its ambitious targets for carbon reductions, then rapid progress must be made across all sectors including implementing measures to encourage consumers to shift diets19. Shifts in diets towards plant-based consumption was also emphasized in the 2021 National Food Strategy, which called for a 30% reduction in meat consumption20. Previous scenario modelling work has shown that global improvements in food technology, closure of yield gaps and reductions in food waste could potentially reduce dietary GHG emissions by about 15%, primarily through adoption of more efficient technologies in low- and middle-income countries7. Our results suggest that much bigger reductions can be achieved by increasing the uptake of plant-based diets, which aligns with other results from this field7,8,11.
There are many population-level interventions that could be implemented to support transitions towards lower meat diets. The UK Health Alliance on Climate Change recommends that sustainable diets should be supported by mandatory environmental labelling on foods, regulation of promotions and taxation of high-carbon foods21. All of these are variants on policies aimed at increasing healthy diets that either have already been introduced (for example, traffic light labelling, the UK Soft Drink Industry Levy) or have been proposed in the UK Childhood Obesity Plan22. The UK Government’s dietary policy is underpinned by its food-based dietary guidelines (FBDGs), known as the Eatwell Guide23. A recent systematic review of national FBDGs found that the large majority are not compatible with the proposed downscaling of ‘planetary boundaries’ for food production—if the UK population consumed the diet recommended by the Eatwell Guide, it would not stay within boundaries for GHG emissions, water use, land use and eutrophication suggested by the paper24. Incorporating environmental sustainability into FBDGs (such as the Eatwell Guide proposed by Plant-based Health Professionals UK25) may be the first step towards implementation of population-level policies that have been shown to support shifts away from animal-based foods26.
Strengths and limitations
This paper uses one of the largest datasets available on the diets of vegans and vegetarians to compare the environmental impact of different diet groups over ten environmental measures. The analyses contribute to the literature that shows the benefit of low-meat diets for reduction of GHG emissions14, land use, water use, water pollution and biodiversity. The paper uses only empirical measures of diet, thereby verifying previous modelling work that has suggested multiple environmental benefits of low-meat diets7,8,27. By using self-identification as vegan, vegetarian and fish-eater, we ensure that our methods include all dietary patterns within those categories including those that breach some of the definitions of the groups—this means our estimates are likely to reflect real dietary practices as opposed to comparison of idealized diet groups.
A key strength of our analysis is that it incorporates the uncertainty around the environmental parameters drawn from a review of 570 LCAs covering results from over 38,000 farms in 119 countries covering five continents3—henceforth, ‘the Poore and Nemecek database’. Doing this shows that although uncertainty for any single food group is large, when this uncertainty is combined over multiple food groups to produce aggregated dietary estimates, we can still observe clear differences between diet groups. Our primary results are based on a Monte Carlo analysis where 1,000 estimates of each food’s environmental impact are produced based on varying measures due to food sourcing and production methods. In our secondary results (shown in Supplementary Tables 1 and 2 and based on regression models that take the median estimate of the environmental parameter for each food group and ignore the underlying variation), not only are the confidence intervals much tighter than in the primary analysis, but the point estimates are also lower. The discrepancy between the two sets of results is due to the computational mathematics involved with combining multiple distributions, many of which are heavily right-skewed, all of which are bounded by zero, and in which negative scalars are not possible (as negative consumption of food is not possible). Although each random draw from the food group distributions is equally likely to be either lower or higher than the median, draws that are higher than the median are, on average, further from the median than those that are lower. When summed, these random draws produce median estimates that are larger than the sum of the medians for the individual food groups. The same principle is shown by rolling two dice. For two normal 1–6 dice (which have no skew), the median score when rolling two dice is 7, which is twice the median score for rolling each dice separately (3½). However, consider rolling two ‘doubling dice’ from backgammon that are heavily right-skewed (with faces 2, 4, 8, 16, 32 and 64). Here, the median score when rolling two dice is 35, much higher than the sum of the median scores for each single dice (which is 12).
Our secondary results (shown in the Supplementary Information) show that ignoring the uncertainty around food-level parameters can result in both underestimation of the uncertainty in diet-level outcomes and bias in the results which can reduce observed differences between diet groups. For example, our primary results show a difference in water use between high meat-eaters and vegans of 480 l d−1, with high meat-eaters consuming 2.2 times as much water as vegans, whereas the secondary results show an absolute difference of 210 l d−1 and a relative difference of 1.7. The issue of food-level uncertainty affects all areas of nutritional epidemiology that rely on food diaries or FFQs to estimate dietary intake. For example, estimates of sugar consumption produced by these methods do not account for uncertainty in the sugar level of food groups, but we know that wide variability in sugar levels for similar foods exists28.
An additional contribution of our research was providing disaggregated GHG emissions and exploring multiple CO2-equivalence metrics, whereas most previous studies report only GWP100 CO2e. Reporting emissions only as aggregated GWP100 totals results in ambiguity in climate impacts29, whereas providing footprints under multiple metrics gives users insight into temporal differences where there are both short- and long-lived gases involved, as highlighted by the Life Cycle Initiative30. As food system emissions contain important amounts of CH4, a relatively short-lived gas, metric selection can have a pronounced impact on CO2e emission reporting31. Here, however, using the alternative pulse-emission metrics explored in this study did not greatly affect results, with a fairly small change in total footprints and relative performance between dietary groups. A caveat is that emissions data from the Poore and Nemecek database are not separated into different gases, and while they are categorized to broadly infer gas compositions (for example, assuming that the CO2e emissions reported for fertilizer application represented N2O, and enteric fermentation CO2e represented CH4), for other components we had to assume emissions were entirely CO2. We reiterate calls for studies on GHG emissions, particularly those relating to agriculture and food, to provide disaggregated emissions to enable the most reliable analyses31.
Our analyses are subject to the following further limitations. The data on the environmental footprint of foods are taken from a snapshot of food and drink on sale in the UK in 2019 linked to the most comprehensive publicly available dataset of LCAs of the environmental impact of foods currently available3. However, the data on dietary consumption were collected in the 1990s, and dietary preferences are likely to have changed since then. This is mitigated somewhat by the fact that the FFQ was linked to the environmental footprint of food and drink on sale in the UK in 2019, but this will not account for category-level changes in consumption since the 1990s. More recent datasets of dietary consumption in the UK are available, including datasets based on a representative sample of the UK population (for example, Kantar Fast-Moving Consumer Goods panel32 and the National Diet and Nutrition Survey33). However, the European Prospective Investigation into Cancer and Nutrition (EPIC)-Oxford dataset (used for this analysis) is the most recent data available in the UK that has a large sample of vegan and vegetarian diets, necessary for these analyses. Data collection is underway on the Feeding the Future study34, which aims to update estimates of food intake in vegans and vegetarians (and meat-eaters) in the UK. Updating our analyses using more timely data will provide evidence of whether trends in new meat and dairy alternatives have affected the environmental impact of plant-based diets.
Our database of food and drink on sale in 2019 was not adjusted for sales, so we were not able to put extra weight on more popularly consumed foods. For our analyses, we standardized daily diets to 2,000 kcal so that differences between diet groups are entirely a result of the composition of the diets—this may result in underestimates of the difference between diet groups as meat-eaters tend to consume more calories than vegans and vegetarians35. Our sensitivity analysis (Supplementary Tables 5–7 and 11–13) shows results that have not been standardized for energy content, which suggests larger differences between the diet groups, but these figures should be treated with caution as some of the difference in kilocalorie intake between groups is caused by artefact. For example, the FFQ used to estimate dietary consumption assumes fixed portion sizes for food groups, but it is likely that portion sizes of cereals, fruit and vegetables are higher in those consuming more plant-based diets.
The FFQ that we used has been validated against food records and biomarkers for estimation of the nutritional quality of the diet, but no such validation has taken place for estimating environmental outcomes. However, a previous validation study compared dietary GHG emissions estimated by a different FFQ with estimates from a 24 h diet recall and showed acceptable levels of agreement between the two36. The FFQ in our study did not measure agricultural production methods, so differences between diet groups based on (for example) differing levels of consumption of organic produce could not be assessed. While we included multiple environmental indicators in our analyses, there are other ethical aspects that vary by region and method of agricultural production that are not included here (for example, agricultural working conditions, animal welfare). Finally, as the Poore and Nemecek database is not comprehensive and our uncertainty analyses are not weighted towards more common food production practices, our uncertainty intervals do not fully incorporate all the uncertainty associated with these comparisons between diet groups. As new agricultural practices aimed at reducing the environmental impact of the food system (for example, feed additives, genetic selection, lab-grown meat) becomes more widespread and LCA data become more readily available, our analyses should be updated.
Comparison with other literature
By scaling our results to the national level, we can compare our absolute estimates of environmental impact with other estimates from the literature. To do this, we used data from the UK’s gold standard diet monitoring programme, the National Diet and Nutrition Survey33, which estimated that in 2016–2019 the average consumption of all meat (that is, processed and unprocessed meat including poultry but excluding fish) in 19–64 year olds was 99 g d−1, and 77 g d−1 in the 65+ age group. We estimated the prevalence of vegans and vegetarians using data from a recent Ipsos Mori survey37. Using these data to scale our results to the population of the UK, we estimate that the annual dietary environmental footprint of adults in the UK amounts to 120 MT of CO2e, 230,000 km2 of agricultural land, 15 km3 of agricultural water, 690 kT of phosphate equivalents (PO4e) and 0.06 terrestrial vertebrate species destined for extinction. Our estimate of 120 MT of CO2e is similar to the most recent estimate from EDGAR-FOOD (Emissions Database for Global Atmospheric Research)38, which produces globally comparable estimates using Food and Agriculture Organization of the United Nations food balance sheet data and estimates UK food systems emissions in 2015 to be 113 MT of CO2e. Our estimates for water use, eutrophication and biodiversity are similar to results for the UK published by the World Wildlife Fund39 of 19 km3 of agricultural water, 645 kT of PO4e and 0.03 species destined for extinction each year. While our estimate of total GHG emissions is similar to that from EDGAR-FOOD, the proportion of individual gases is different. For our estimates, the contribution to CO2e of N2O is about 7% for all diet groups, and for CH4 the contribution increases from 6% in vegans to 21% in high meat-eaters. Equivalent estimates from EDGAR-FOOD for the UK are 17% for N2O and 35% for CH4. This may be a result of discrepancies in how we derive separate N2O, CH4 and CO2 emissions making inferences from the Poore and Nemecek database, as noted above, and the way separate gases are handled in the Food and Agriculture Organization Statistics Division (FAOSTAT) emissions in EDGAR-FOOD, further highlighting the challenges in obtaining individual gas data.
Previous estimates of dietary GHG emissions for vegans, vegetarians, fish-eaters and meat-eaters in the EPIC-Oxford cohort have been made using a similar method based on GHG emissions data from a single study14. The estimates presented here are slightly lower for plant-based diet groups and slightly higher for meat-eating groups. Other studies have compared the environmental impacts of observed diet groups defined by exclusion of meat or dairy40,41,42, but they have not included as many environmental measures as here nor incorporated uncertainty in estimates due to region of origin and production method. Dietary GHG emissions for US vegetarians in the Adventist Health Study 2 cohort41,42, standardized to a 2,000 kcal diet, were 70.8% (70.5–71.2%) of emissions from non-vegetarian diets, similar to the difference between vegetarians and the medium meat-eaters (58.5%) observed in our sample. An analysis of 29,210 French adults in the NutriNet-Sante Study included data on 464 pesco-vegetarians (equivalent to fish-eaters in our study), 406 vegetarians and 297 vegans40. For both GHG emissions and land use, that study40 found the same relationship as shown in our paper, with lowest environmental impact for vegans, similar impact for vegetarians and fish-eaters, and highest impact for meat-eaters. They also found similar relative differences between vegans and meat-eaters, with dietary GHG emissions of vegans being 24.5% (19.2–29.8%) of the meat-eaters and 35.6% (29.9–41.3%) for land use.