Animal-based foods have high social and climate costs

Despite the importance of animal-based agricultural greenhouse gas emissions as drivers of climate change, the climate costs of such emissions have not yet been quantified in an integrated way. Using a macroeconomic–climate framework, we coupled global agricultural and industrial economies to estimate these costs at a regional level. To be consistent with end-of-century temperature increases of 1.5–3 °C, we found that every 10-percentage-point increase in agricultural emissions required a compensating 1.5-percentage-point reduction in industrial emissions—the ‘emissions opportunity cost’ of animal-based foods. Alternatively, if agricultural emissions were not offset in the industrial sector, diets high in animal protein contributed US$72 per person per year in additional climate damage—approximately half of the annual climate damage produced by the average passenger vehicle in the United States. Our analysis revealed geographic heterogeneity in climate costs by diet and food type, suggesting opportunities for mitigation policies while recognizing food insecurity risks. Increasing greenhouse gas emissions in agriculture must be compensated by emissions reduction in other sectors if global emissions are to be capped. Using macroeconomic–climate modelling, this study quantifies such emission compensation efforts under different dietary choices.

G lobal food supply chains sustain a population of more than 7.8 billion people but produce over 26% of global greenhouse gas (GHG) emissions 1 . Nearly 80% of these emissions are attributable to livestock production 2,3 , and projections suggest that this sector's contribution to climate change will increase as populations grow and low-and middle-income countries continue shifting towards Western diets high in meat-based proteins 4,5 . For instance, in Southeast Asia, a developing region that has experienced substantial income and population growth, total agricultural GHG emissions increased by 20% during 2000-10 as meat consumption rose by over 60% (ref. 6 ). Efforts to reduce climate impacts from the global food system will thus probably entail large-scale dietary changes as well as reductions in the emission intensity of livestock farming 7-12 . Recent analyses have detailed the contribution of animal agriculture to climate change. A large body of work focusing on dietary shifts has found that reducing global consumption of animal products could reduce mid-century annual agricultural GHG emissions by up to 70% (refs. [13][14][15] ). Other work has used life cycle assessment or observational analysis methods 16 to demonstrate the potential role of technological mitigation options, including changes to livestock feed, manure management, land use for sequestering soil carbon and animal genetics and health. These studies suggest that while such innovations represent promising options to reduce emissions in this sector, they will probably not achieve net-zero emissions 1,5,17,18 .
Past work, however, has largely been silent on quantifying the trade-offs and costs associated with livestock-related emissions using integrated economic-climate frameworks. Work in this general area focuses on the climate benefits produced by a small set of policy scenarios-namely, emissions reductions arising from changes in animal product consumption or land use resulting from researcher-specified emission taxes 11,[19][20][21][22][23] . While interesting and useful, this work stops short of computing generalized costs that can be used in external cost-benefit analyses or supplemented with information about global animal agriculture's additional environmental impacts to support true-cost pricing 24 .
Furthermore, GHG emissions from livestock supply chains comprise a mix of carbon dioxide (CO 2 ), nitrous oxide (N 2 O) and methane (CH 4 ). Past efforts to assign a dollar value to (or 'monetize') the damages caused by such emissions-generally referred to as 'climate costs'-have relied on converting these non-CO 2 emissions into CO 2 -equivalent measures by way of global warming potentials (GWPs). CO 2 -equivalent measures are subject to arbitrary decisions about the time horizon of warming equivalence and ignore differences in warming dynamics known to be important for economic policymaking 25,26 . Estimating climate costs from animal agriculture emissions therefore requires explicit representation of non-CO 2 GHGs and their distinct impacts on the climate system.
Here we represent emissions from animal agriculture within a leading macroeconomic integrated assessment model (IAM)-the Dynamic Integrated Climate-Economy model (DICE 27 ; Fig. 1)and estimate a suite of climate costs associated with the global production of animal-based foods. These estimates draw on regional heterogeneity in diets and production methods, as well as distinct concepts of costs from the economics literature. Within DICE, we replace the climate model with the Finite Amplitude Impulse Response (FAIR 28 ) model, heeding calls by the National Academies of the Sciences to incorporate more realistic representations of climate system dynamics in IAMs 25 . In contrast to past work that relies on GWP conversions of non-CO 2 gases, FAIR explicitly models the gas cycles and warming effects from CH 4 and N 2 O. Additionally, we incorporate a representation of emissions from animal agriculture in DICE's economic model, calibrating emission intensities on the basis of the United Nations Food and Agricultural Organization's (FAO) Global Livestock Environmental Assessment Model (GLEAM 29 ). Our model is not the first to incorporate agriculture in a macroeconomic IAM; however, past work has focused on enriching the representation of climate damages through the effect of warming on food production 30-33 rather than measuring the climate costs of current emissions.
Using this model, hereafter DICE-FARM ( Fig. 1), we focus on two distinct concepts of the costs of dietary emissions. The first set of costs, related to cumulative emissions budgets and peak global

Animal-based foods have high social and climate costs Frank Errickson 1 , Kevin Kuruc 2 ✉ and Jonathan McFadden 2
Despite the importance of animal-based agricultural greenhouse gas emissions as drivers of climate change, the climate costs of such emissions have not yet been quantified in an integrated way. Using a macroeconomic-climate framework, we coupled global agricultural and industrial economies to estimate these costs at a regional level. To be consistent with end-of-century temperature increases of 1.5-3 °C, we found that every 10-percentage-point increase in agricultural emissions required a compensating 1.5-percentage-point reduction in industrial emissions-the 'emissions opportunity cost' of animal-based foods. Alternatively, if agricultural emissions were not offset in the industrial sector, diets high in animal protein contributed US$72 per person per year in additional climate damage-approximately half of the annual climate damage produced by the average passenger vehicle in the United States. Our analysis revealed geographic heterogeneity in climate costs by diet and food type, suggesting opportunities for mitigation policies while recognizing food insecurity risks.
temperature targets 34,35 , are inferred from model-generated combinations of industrial and animal agricultural emissions reductions necessary for limiting total warming to prespecified levels. When total warming is fixed, there is necessarily a trade-off between agricultural emissions and industrial emissions; this trade-off represents the emissions opportunity cost of animal agriculture (that is, the cost of agricultural emissions in terms of forgone industrial emissions). The second set, building on work measuring the economic impacts of CO 2 emissions, assumes that small increases in animal agricultural emissions are not offset in the industrial sector and instead result in additional warming. The social cost of animal products (SCAP) measures the discounted, monetized climate   damage from the resulting increase in global temperatures. We use this terminology to highlight the conceptual similarity between our estimates and the social cost of carbon (SCC), though we recognize that animal agriculture produces substantial non-climate costs that we do not account for in this analysis (discussed below). DICE-FARM allows us to calculate the SCAP at the global level as well as the individual diet and individual food product levels. After analysing representative diets and production methods, we reperform the analysis at the country level to catalogue variability in costs that can inform mitigation prioritization and/or internationally coordinated climate policies.

results
Under a business-as-usual (BAU) scenario, DICE-FARM projects that animal agriculture will account for 0.4 °C of the approximately 3 °C of warming at the end of the century (Fig. 2, left). The BAU scenario takes industrial CO 2 emission pathways and population growth from DICE's default scenario, assumes fixed per capita meat consumption (an assumption later relaxed) and uses Representative Concentration Pathway (RCP) 6.0 for GHGs and short-lived climate forcers other than CO 2 , CH 4 and N 2 O (Methods) 27,36 . The temperature gap between the BAU scenario and a hypothetical reference scenario in which emissions from animal agriculture are immediately set to zero ('Vegan') opens immediately and grows modestly throughout the twenty-first century (Fig. 2, left). This difference is consistent with the 12-18% livestock-related GHG emissions cited in past studies 2,23,37 and indirectly illustrates the challenge of staying under a 2 °C temperature target without global dietary change.
These temperature differences are largely driven by changes in CH 4 emissions, the primary GHG produced by animal agriculture (Methods). Compared with CO 2 , CH 4 is known to have a shorter atmospheric lifetime but produces much stronger near-term warming. To highlight the dynamics of a dietary emissions bundle, as well as compare the total warming effects of animal consumption with those of other common household activities, we depict how temperatures in our model respond to the emissions produced by a representative US individual's diet, household energy consumption and gasoline burned from one passenger vehicle 38 (Fig. 2, right). These temperature response functions are computed in DICE-FARM by adding the annual GHG emissions associated with each of the three activities for one person in the year 2020. The high CH 4 levels in dietary emissions lead to shorter-lived but more intense warming, concentrated over the next 30 years. Under standard economic discounting assumptions, this has ramifications for assessing the costs of these emissions and makes explicit the problem of using GWPs that consider only total warming. As implied by our analysis, dietary emissions (unlike industrial emissions) primarily harm the generation producing them.
The emissions opportunity cost of animal products. We define the emissions opportunity cost of animal-based foods (hereafter 'animal products') using the joint reductions in industrial and agricultural emissions required to achieve a prespecified peak temperature. Under a societal objective to keep the peak change in global surface temperature below a fixed threshold, increasing animal agricultural emissions requires additional restrictions of industrial emissions. The emissions opportunity cost is the amount of industrial activity crowded out by increases in animal production at current emissions intensities. This trade-off is measured by the slope of the combinations of industrial output and animal production consistent with staying below various global surface temperatures (Fig. 3), which we calculate for objectives of 1.5, 2, 2.5 and 3 °C to illustrate both the Paris Agreement targets and the general pattern that occurs as less stringent reductions are undertaken. Our analysis suggests that (1) a 10-percentage-point reduction in agricultural emissions allows for an additional 1.5 percentage points of industrial emissions and (2) this trade-off appears constant at these various levels of warming (Fig. 3).
These trade-offs highlight the challenge of achieving the temperature goals in the Paris Agreement without serious efforts to curb dietary emissions-results consistent with complementary work 1,5,10,12,13,39 . Limiting total warming to 2 °C, absent reductions in livestock-related emissions, requires a more than 80% reduction in industrial emissions relative to BAU (Fig. 3). These stringent industrial requirements are relaxed to reductions of about 65% under a hypothetical, immediate transition to veganism. The same dietary shift under a more modest 3 °C warming entails industrial emissions reductions of only 25%. In related work, Stehfest et al. find that energy sector emissions reductions needed for a 50% probability of meeting a 2 °C warming target are lower under scenarios in which animal products are reduced or eliminated 30 . The social cost of animal products. When animal agricultural emissions are not offset by reductions in the industrial sector, they increase total projected warming, causing losses in future economic well-being (Fig. 1). We value these losses as the discounted sum of climate damages from diets and animal products, collectively referring to our estimates as the SCAP (Methods  Table 6). Eliminating meat but retaining dairy and eggs reduces these costs to $24 (Table 1, column 3); that is, meat consumption accounts for about two-thirds of the total climate costs of an average American diet.
The climate costs of consuming an additional serving of individual animal products (Table 1, columns 4-9) are a relevant metric for policies seeking to address incremental dietary changes. They highlight which products drive these effects and approximate the optimal product-level tax to correct for the neglected climate costs. For beef, dairy and sheep/goat meat, our SCAP estimates are 35%, 18% and 23% of each product's total production costs (private plus social cost) based on averages of producer price data (Supplementary Section D). These ratios and costs per serving can guide climate-related policies such as dietary taxes recently considered in Denmark 40 , Germany 41 and Sweden 42 . Important caveats to the results above are that they rely on global averages for emissions intensities from animal agriculture and a representative Western diet, which masks geographic differences. Performing the same analysis of annual diet-level climate costs at a disaggregated level (and assuming no trade; Methods) yields estimates that vary by an order of magnitude across countries, ranging from less than $10 in parts of sub-Saharan Africa and South Asia to over $100 in parts of West Asia and Latin America (Fig. 4a). This variation comes from both an emissions intensity effect (livestock sectors, like industrial output, are more emission-intense in low-income regions; Supplementary Table 7) and a composition effect (different cultures consume different diets; Supplementary Table 8). The emissions intensity effect is apparent from the wide range of climate costs per serving that we calculate for different regions (Fig. 4b).
discussion An unresolved issue in the climate economics literature concerns the proper method for valuing damages that occur in different periods. A key benefit of using DICE-FARM rather than GWP conversions (Supplementary Section A and Supplementary Table 2) is that it allows for accurately capturing in which years climate damages occur. With the full path of damages, it is straightforward to modify the discounting approach, an exercise known to greatly impact valuations of the SCC given the long atmospheric lifetime of CO 2 (ref. 27 ).
Unlike industrial emissions, however, the animal-related temperature effects that we estimate are largely driven by short-lived CH 4 emissions and are therefore concentrated over the next few decades (Fig. 2, right). While this suggests that our SCAP estimates should be less sensitive to discounting assumptions than the SCC, we still find policy-relevant variation in the SCAP. Our baseline results follow the discounting framework used in the original DICE model (a 1.5% pure rate of time preference (ρ) and a value of 1.45 for the elasticity of marginal utility (η); Methods). US government SCC estimates in recent years instead apply constant consumption discount rates of 2.5%, 3% and 5% (ref. 43 ). Under this discounting approach, annual standard dietary costs range between $54 and $179 (Table 1). If we instead follow past work 44 that argues for a much lower pure rate of time preference on normative grounds ('Stern' , ρ = 0.1%, η = 1.45), the same dietary cost is $227, a threefold increase relative to our baseline estimate. Past literature producing comparable monetized values is scant; however, our range of system-wide estimates ( Table 1, column 1) are on the same order of magnitude as the $570 billion per year projected for the year 2050 in Springmann et al. 14 . An extensive discussion on the debate surrounding discounting is beyond the scope of this paper; nonetheless, the range exhibited in this table highlights that discounting is a key uncertainty in these valuations, and it allows policymakers to focus on the set of estimates that are closest to their government's choice of discount rate.
Other sources of uncertainty could likewise influence our SCAP results. Social cost estimates are generally sensitive to the relationship between temperature increases and economic damages 25 . By applying the DICE model, we do not consider the full range of available climate damage functions; in particular, we neglect recent empirical work documenting larger reductions in economic growth with increasing temperatures 45 . Likewise, our results do not account for structural and parametric uncertainty in DICE-FARM's underlying climate model 25 , nor do we model the evolution of GHG emissions resulting from plausible global adjustments towards plant-based diets. We also focus exclusively on climate costs, given the acute scientific and social interest in climate change and the readily available tools of climate economics for valuing such costs. However, raising livestock pressures natural systems through intensified land use, water and air pollution, biodiversity loss and the drawdown of freshwater sources [46][47][48][49][50] . Such impacts, many of which are challenging to reliably monetize at the global level, are important to recognize in agriculture and climate policy design and would increase our social cost estimates (Supplementary Section E). In this way, our estimates serve as a lower bound on the true, all-things-considered cost of animal products that would incorporate such external factors. Main results of the social cost analysis along with output from alternative discounting assumptions. Column 1 represents the annual social cost of the global system through the monetization of a one-year elimination of all emissions from animal agriculture. Columns 2 and 3 depict the climate costs of the animal products within entire diets (column 2, the SAD (Supplementary Table 6); column 3, the SAD with meat removed). The remaining columns report costs associated with an additional serving of different animal products (measured as 20 g of protein).  The results are analogous to the $72 in Table 1, row 1, column 2. Latin America, Oceania and parts of western Asia have the highest dietary costs. These results rely on regional differences in emissions intensities (Supplementary Table 7) and country-level differences in diets (Supplementary Table 8). b, regional differences in cost per item (analogous to Table 1, row 1, columns 4-9) account for some of this difference, but they also highlight where production methods are the most emission-intense.
Our emissions opportunity costs (Fig. 3) are also sensitive to various uncertainties and modelling assumptions about the evolution of technology and diets. If increasing global meat consumption leads to more agricultural emissions than projected in our BAU scenario, additional opportunities for reducing these emissions would further alleviate pressure on industrial emissions. Because our BAU scenario is conservative in this dimension-we assume that animal product consumption does not increase with income when in fact a positive relationship has been documented 4,51-53 -our emissions opportunity cost estimates are also conservative. We show that the slopes of our temperature target curves imply an increase in the emissions opportunity cost of 50% in a sensitivity analysis where animal product consumption increases with income ( Supplementary Fig. 4). For similar reasons, our results could be biased if technological innovation leads to more rapid reductions in livestock-related emissions intensities.
At the country and regional levels, our disaggregated results (Fig. 4) detail competing policy considerations. If dietary guidance (that is, promoting vegetarianism) is of the greatest interest to policymakers, then our analysis would suggest catalysing such changes in countries with the highest per capita diet-level costs (Fig. 4a). However, a serving of beef in regions such as Africa, Asia and Latin America leads to four to five times more climate damages (as measured in 2010 US dollars per 20 g protein) than in less emissions-intense regions such as Russia and western Europe (Fig. 4b), and hence equivalent reductions in these regions would confer greater climate benefits.
The mitigation burden of such policies can be contextualized by comparing our SCAP results with food expenditures. Across countries, the $72 baseline estimate of annual dietary climate costs represents approximately 8% of annual per capita food expenditures. Across regions, this fraction remains small even in low-income areas because both climate costs and food expenditures are reduced (Supplementary Table 4) 54 . So, while large in aggregate, our results suggest that the climate costs of animal products, and hence a dollar amount near the annual tax burden that an individual would face under optimal policy, would not be a large fraction of food expenditures for most of the global population. Nonetheless, equity criteria necessitate recognizing that populations in low-income countries have disproportionately higher rates of extreme poverty and food insecurity while generating far fewer per capita emissions given their low levels of animal product consumption 55 . Such a holistic policy analysis is beyond the scope of this paper. One compelling option, however, would be to expand the use of international climate funds to provide technical and financial assistance for reducing climate costs per animal-product serving in low-income, high-emissions-intensity countries. Biophysical-based evidence indicates substantial technical mitigation potential in some of those regions, especially Latin America, South Asia and eastern Africa 17,37 .

Conclusions
Livestock-related emissions are an important driver of climate change. Our findings reinforce biophysical findings that it will be challenging to achieve the goals of the Paris Agreement without substantial changes in this sector (Fig. 3) 1,5,[13][14][15] . Indeed, we estimate the climate and social costs arising from the current combination of consumption choices and production methods to be large, suggesting similarly large benefits from a multipronged approach to reducing these emissions. Regardless of the specific mitigation policies that societies adopt, efficient design requires a careful consideration of the relevant costs, benefits and trade-offs from major emission sources. The model and results in this paper allow future researchers and decision makers to better understand and act on such considerations.

Methods
Here we detail the key equations of DICE-2016R, our modifications and extensions necessary to develop DICE-FARM (for more details on a recent DICE model, see the manual made available by W. Nordhaus 56 ) and the method for computing the SCAP. DICE-2016R IAM. DICE-2016R can be summarized as a system of four conceptual modules: current economic activity impacts the current CO 2 emissions (economic/emissions module), current emissions impact the current and future stock of GHGs in the atmosphere (carbon cycle module), the stock of GHGs determines the temperature in future periods (warming module) and future temperatures impact the economic well-being of future people (damage module). Although this basic structure underlies our integrated assessment modelling, we make modifications to the original model. DICE endogenizes only the emissions, propagation and radiative forcing of CO 2 , so the climate system (the carbon cycle and warming modules) must be replaced in this paper to endogenize the life cycle and radiative forcings of CH 4 and N 2 O.
The modules of DICE-2016R take the following structure. Gross output (in time t), Y G t , is produced in a Cobb-Douglas production function: where A t , L t and K t represent the total factor productivity (or economic efficiency), the total labour force and the accumulated capital stock, respectively; α governs the contribution of labour, while (1 − α) governs the contribution of capital. Industrial emissions, E Ind t , arise linearly from gross output according to a time-varying parameter, σ Ind t : This emissions intensity declines through time as a representation of private sector improvements in energy efficiency that would arise with or without climate policy. Emissions can, in principle, be reduced through costly mitigation, but none of the exercises in this paper explore this option (namely, we do not study optimal industrial emissions). We retain DICE-2016R's projected path for population and unmitigated industrial emissions in our BAU scenario.
To incorporate GHG emissions from animal agriculture, we insert an agricultural sector into the DICE-2016R model in parallel with the industrial sector. Emissions are modelled analogously to equation (2), as we explain in the next subsection.
Gross output, prior to division between consumption and savings, is reduced by climate damages according to total warming in that year: Note that D t is the fraction of output lost to climate damages and is quadratic in temperature, T t (which we explain the computation of below). It is through losses in output that climate costs will be computed. Note that a 1 and a 2 are fixed coefficients.
Post-damage net output, Y N t , is split between consumption, c t , and savings, s t , which determine the trade-off between current welfare and future economic production. In the original model, these savings rates are optimized according to standard consumption-savings methods in economics. In principle, savings rates can adjust when climate damages change. However, we choose to fix savings rates as they are in the DICE-2016R BAU scenario, because the climate damages we consider are unlikely to meaningfully change savings behaviour. Welfare then follows from a global utility function over per capita consumption, c t : In our social cost calculations, the total welfare from W above is used to assess the monetary equivalent of the losses from additional warming. Note that ρ is the pure rate of time preference, and η is the elasticity of the marginal utility of consumption.
Development of the DICE-FARM model. The farm sector that we introduce sits alongside the traditional economic sector in DICE-2016R. Animal products are produced with linear emission intensities for the three distinct agriculture-related gases that we study. Emissions from the farm sector are thus an 18-equation module (6 products by 3 gases) taking the following form: where FE j,g t are the farm emissions of gas g from product j in time t, q j,t is the output (in kg of protein) of each product in that period and σ j,g is the emission intensity of each product-gas. We calibrate σ j,g and q j,t to the GLEAM database from the FAO 29 . GLEAM simulates global GHG emissions from the production of meat, milk and eggs among the world's predominant livestock supply chains using a life cycle assessment approach. Importantly, GLEAM accounts for direct, indirect and some embedded emissions from land use change for feed crops, fodder production, animal management, energy use and post-farm transport, processing and packaging (Supplementary Table 5 and Supplementary Fig. 5). Indirect and embedded sources account for energy used for irrigation; emissions from volatilization, leaching and runoff of nitrogenous fertilizers and manure; and the energy embedded in the construction and maintenance of farm buildings, farm equipment, tractors and irrigation systems.
To compute the evolution of temperature used to determine damages (equation (4)), we replace the simple climate module in DICE-2016R with the FAIR climate model detailed in Smith et al. 57 . The three primary benefits of using FAIR are that (1) it explicitly models a more comprehensive set of GHGs relevant to the animal agriculture sector, (2) the impulse response function to a pulse of emissions is calibrated to be consistent with the behaviours of more sophisticated Earth system models and (3) it accounts for carbon cycle temperature feedbacks previously neglected in DICE-2016R. At its core, FAIR has a carbon cycle that takes place in a four-reservoir system; the stocks of other GHGs follow simplified single-equation decay models. Temperature dynamics are governed by a two-equation system that performs better over short horizons than the native DICE-2016R climate module; this is especially important for accurately capturing the dynamics of dietary emissions. The complete set of FAIR equations are fully described elsewhere 57 .
We maintain FAIR's default parameter settings from the current model version (v.1.3), except the equilibrium climate sensitivity (ECS) and transient climate response (TCR). For these parameters, we set the ECS to the default value in DICE-2016R, 3.1 °C, and select a corresponding TCR value (1.69 °C) following FAIR's model calibration framework that accounts for the correlation between the TCR and ECS 57 . This parametric adjustment brings our social cost numbers closer to the numbers one would get from the original DICE-2016R model while retaining the integrity of FAIR, though these parametric choices do not meaningfully affect the main results (Supplementary Table 3).
Because FAIR takes as inputs the emissions of GHGs other than CO 2 , we must supply these from outside of the model (DICE-2016R treats all non-CO 2 forcings as a single exogenous time series). We do so using RCP 6.0 for all non-CO 2 gases. Of the multiple RCP scenarios, these emissions imply total warming most comparable to the DICE-2016R BAU scenario. After computing emissions from the farm sector, we remove these from the exogenous RCP emissions to ensure that animal agricultural emissions are not double-counted. The baseline model with agricultural emissions and augmented RCP emissions then-by constructionreplicates the model output using RCP 6.0 without any farm emissions. Note that forcings from solar irradiance are cyclical in the RCP scenario, driving the slight visible cyclicality in projected temperature time paths to 2100 and beyond.
One technical change we must implement to merge FAIR into DICE-2016R is to annualize the latter. DICE-2016R is traditionally run in five-year time steps and is widely available in this format; our implementation requires modifying parameters and model equations to reflect a yearly time-step setting. For completeness, we verify that our annualized implementation produces output identical to the original DICE-2016R model ( Supplementary Fig. 1). For our annualized model to match the five-year DICE-2016R model, the implied yearly depreciation rate must be slightly changed. It is originally assumed to be 10% per year over five years, but the depreciation occurs at the end of every five-year horizon by the nature of the discrete time steps. Because depreciation in our model is eroding capital every year in a compounding way, 10% per year implies a capital stock that is lower than it otherwise would be. To ensure that our model environment replicates the economic trajectory of DICE-2016R, we solve for the depreciation rate (8.2%) that minimizes the squared errors between the capital stock in our annualized model at the five-year intervals available in the five-year DICE-2016R. SCAP computation. The SCAP is computed by pricing the emissions associated with a specific product, or an entire diet, as governed by equation (7). Notice that this formulation implicitly assumes that the plant proteins that replace animal foods produce zero emissions; the relative magnitude of their emissions makes this is a reasonable approximation 1 . As we note in the manuscript, per capita meat consumption is calibrated to current levels 6 and assumed to be fixed in our main BAU calibration, with that assumption relaxed in Supplementary Section B.
The benefits and costs of dietary choices are computed as the total economic well-being gains (according to the social welfare function above) to all generations avoiding the warming dynamics associated with those choices (such as the dynamics depicted in the projected temperature time paths). We then compute the amount of economic consumption (measured in 2010 US dollars) that would need to be sacrificed today to match the cumulative well-being lost via the climate damages. The SCAP is thus where ΔAP(t) is the change in animal product consumption considered in time period t, C(t) is the consumption level (in US dollars) in that period and W is the total economic well-being across all periods.
Regional heterogeneity in our social cost estimates is computed in the same manner but employs disaggregated data to compute emissions changes. Dietary data are available at the country level, most recently in 2013 (Supplementary Table  8) 6 . For simplicity, we assume that economies do not trade these commodities, such that this consumption is produced domestically. Estimates of aggregate production from GLEAM, necessary for calibrating emissions intensities, are available only at a regional level that is less disaggregated than countries 29 . All available regions are indicated in the notes to Supplementary Tables 4 and 7. Since the model measures the global environmental costs of an emissions bundle, we can modify the emissions pulse to reflect the diet-emissions combinations from different world regions without the need for a fully specified regional model.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

data availability
The datasets generated and/or analysed during the current study are archived 58 and publicly available on K.K. 's Github website: https://github.com/kevinkuruc/ ClimateCostsofAnimalFoods_NatureFood2021.

Code availability
The code used to generate all results is archived 58  Last updated by author(s): Mar 17, 2021 Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted

Software and code
Policy information about availability of computer code Data collection No software was used to collect data.

Data analysis
Data was used to calibrate model, but not independently analyzed. Code used for analysis is archived (see references) as well as available on GitHub.
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A list of figures that have associated raw data -A description of any restrictions on data availability The datasets generated during and/or analyzed during the current study are archived and publicly available on the author's Github website: https://github.com/ kevinkuruc/DICEFARM.