What 38,700 farms show about food’s environmental footprint

Two farmers grow the same crop. One of them puts out fifty times the environmental impact of the other.

Not fifty percent more. Fifty times. That finding comes from one of the largest consolidations of farm-level food data ever assembled, and it is the most useful fact in this subject, because it says something uncomfortable: the name of a food is a weak predictor of its footprint, and the package cannot tell you which farm you are holding.

This article reports what the measurements say about food and the environment. It recommends no diet. Several neighboring articles own pieces of this: ours on local food covers food miles, ours on seasonal eating covers timing, ours on food waste covers what gets thrown away, ours on organic farming covers certification, and ours on packaging covers containers.

The study that reset the question

In 2018, Joseph Poore of the University of Oxford and Thomas Nemecek of Agroscope in Zurich published a consolidation of food production data in Science. Its scale is why it still anchors the field. The authors pulled together data on 38,700 farms measured across five environmental indicators, plus 1,600 processors, types of packaging and retailers.

Three findings came out of it.

Variation within a product is enormous. Impacts can differ by a factor of 50 among producers of the same food. That is the number quoted at the top of this article, and it comes out of the consolidated production data rather than from a scenario about the future.

Producers cannot fix it alone. The authors report that producers have limits on how far they can cut impacts, and that mitigation runs into trade-offs, multiple routes to a low-impact result, and interactions along the supply chain. Lowering one indicator can raise another.

The comparison between categories is larger than the comparison within them. The paper’s own summary, published in Science, states that “impacts of the lowest-impact animal products typically exceed those of vegetable substitutes.”

That third sentence is the one people argue about, so read exactly what it says. It compares the best performers in one group with substitutes in another, on the indicators the study measured. It is a statement about measured impacts. It is not a statement about nutrition, cost, culture or what anyone ought to have for dinner, and the authors did not present it as one.

What a 50-fold spread does to consumer decisions

Here is the practical consequence, and it cuts against most advice you will read.

If two producers of the same food differ by that much, then a label naming the food tells you very little, and almost nothing a shopper can see distinguishes the high-impact version from the low-impact one. Poore and Nemecek’s own recommendation follows from that. They argue for producers monitoring their own impacts, meeting targets flexibly through whatever practices work for them, and communicating results onward.

In other words, the information a consumer would need to act precisely does not currently exist at the point of sale. That is a measurement gap, not a reason for despair, and it explains why so much household advice in this area is weakly supported. Advice about what category of food to buy rests on large measured differences between categories. Advice about which version to buy rests on information nobody gives you.

Ranking the household decisions by measured effect

Put the household decisions in rough order of how much the measurements say they move, largest first.

What the diet is made of. The category differences in Poore and Nemecek’s data are the largest effect in the dataset that a household controls.

How much food gets thrown away. Waste multiplies every upstream impact by the fraction discarded. Our article on food waste handles this properly, including the share that comes from households.

How it was produced. Real and large, by a factor of 50, and mostly invisible to you.

Where it came from. Transport is a smaller share of most foods’ footprint than people assume, which is why our article on local food treats food miles as a weak proxy rather than a strategy.

When it was grown and how it was packed. Smaller again, and both carry genuine exceptions. Our articles on seasonal eating and packaging cover them.

Dietary composition dominates sourcing. That sentence is the summary of the hierarchy, and it is a claim about measurement rather than morality.

Now the caveats that belong next to it, because leaving them out would be reporting one side.

Averages hide the spread. A category average is the midpoint of a distribution 50 times wide, so a specific farm can sit far outside its category’s reputation in either direction.

Not all agricultural land is interchangeable. Mesfin Mekonnen and colleagues, analyzing US livestock from 1960 to 2016 in Environment International in 2019, note that ruminant animals consume a large share of feed that humans cannot eat. Land that grows that feed is not automatically land that could grow human food instead.

Efficiency changes over time. The same analysis found that water productivity for every US livestock product is far better now than it was in 1960, driven by gains in animal productivity, feed conversion and crop yields. Numbers from one decade do not describe the next.

Water and land are separate questions, and they rank differently

This is where single-number thinking breaks down fastest.

Start with water, and with a unit problem. The U.S. Geological Survey reports that about 110,904 million gallons a day went to crop irrigation across the lower 48 states in 2020, down from a peak near 138,938 million gallons a day in 2000. Roughly 63.5 percent came from groundwater. The number that matters more is this one: about 72 percent of irrigation water is consumptive use, which USGS defines as water that does not return to local water bodies or groundwater because it evaporates, is consumed, or ends up inside a crop. Withdrawal and consumption are different measures. Irrigation, thermoelectric power and public supply together account for roughly 90 percent of daily water use in the lower 48 states, and ranking those three on withdrawals alone tells you nothing about which gallons left the basin.

USDA’s Economic Research Service puts irrigation at 47 percent of US freshwater withdrawals across 2010 to 2020, with 54.9 million irrigated acres reported in the 2022 Census of Agriculture. One more figure from ERS reframes the whole subject: irrigated land made up less than 17 percent of harvested cropland but farms with some irrigation produced more than half the total value of US crop sales. Irrigation is concentrated, and so is the exposure if a basin runs short.

Now watch the ranking move when you change the denominator. Mekonnen and colleagues measured water productivity, which is product output per unit of water. Per unit of food energy, milk came out highest, followed by chicken and pork. Per gram of protein, poultry products came out highest, then cattle milk, then pork, with beef lowest. Same animals, same data, two different orderings, because calories and protein are not the same thing to divide by.

Land tells a third story. ERS reports a US land base of nearly 2.26 billion acres in 2022, with about 51 percent in agricultural use counting cropping, grazing and farmsteads. Total cropland fell to 377 million acres, the lowest since the series began in 1945.

Against that backdrop, Alon Shepon and colleagues published an analysis in the Proceedings of the National Academy of Sciences in 2018 using a construct they called opportunity food loss: the food forgone by using cropland for a resource-intensive item instead of a nutritionally comparable plant alternative. Conventional retail-to-consumer losses run around 30 percent for both plant and animal products. On their accounting, opportunity losses came to 96 percent for beef, 90 percent for pork, 75 percent for dairy, 50 percent for poultry and 40 percent for eggs, because plant replacements can yield up to twentyfold more nutritionally similar food per unit of cropland than beef.

Read that for what it is. It is not a measurement of waste. It is an accounting device that treats a different possible use of land as a loss, and it assumes those replacements are nutritionally comparable, which is an assumption doing a lot of work. The arithmetic is transparent and the premise is arguable, which is exactly the kind of result worth reporting with its premise visible.

EAT-Lancet, and the criticism it drew

In 2019 The Lancet published the report of the EAT-Lancet Commission, written by Walter Willett, Johan Rockström and 35 co-authors, proposing a global reference diet intended to be healthy and environmentally sustainable at the same time. On the EAT Forum’s own description of the diet, plant foods make up a large proportion of what is eaten, with fish, dairy and meat in moderate or small amounts, and the diet is presented as a global reference meant to accommodate cultural and regional variation.

It drew serious, published criticism, and both the report and the criticism belong in any honest account.

The mortality claim. Francisco Zagmutt, Jane Pouzou and Solenne Costard, at the consultancy EpiX Analytics, published a replication attempt in the Journal of Nutrition in 2020. They state that the report “did not meet standards for transparency and replicability” and did not fully account for statistical uncertainty. Reworking the US mortality calculation, they reported finding errors, and concluded that once uncertainty was carried through properly, the diet’s mortality benefit in the United States was no greater than what would come from energy-intake changes preventing underweight, overweight and obesity on their own. They called for independent validation before the figures informed dietary guidance.

The affordability claim, or rather its absence. Kalle Hirvonen and colleagues at the International Food Policy Research Institute and Tufts University pointed out in Lancet Global Health in 2020 that the Commission never assessed cost. Using 2011 retail prices for 744 foods across 159 countries, they found the cheapest way to meet the targets cost a global median of $2.84 a day, with fruits and vegetables the largest single share at 31.2 percent. The cost exceeded household per capita income for at least 1.58 billion people, and it ran on average 1.60 times the minimum cost of simply meeting essential nutrient requirements. That is the hardest criticism to argue with, because it is arithmetic on price data rather than a model of future health.

The counterweight. Marco Springmann and colleagues modeled 85 countries’ national food-based dietary guidelines in The BMJ in 2020. Adopting national guidelines was associated with a 15 percent average reduction in premature mortality and a 13 percent average cut in greenhouse gas emissions, but 83 of the 85 were incompatible with at least one global health or environmental target. Adopting the EAT-Lancet recommendations instead was associated with 34 percent greater mortality reduction and more than three times the emissions reduction. The authors flagged their own weak point: the health results rest on observational data and an assumed causal relationship, with evidence certainty in existing meta-analyses mostly graded as moderate.

Sort those by evidence tier and the picture is clearer than the shouting suggests. The cost finding is the sturdiest, because prices and incomes are measured. The environmental modeling uses accepted methods and carries wide uncertainty. The mortality projections are the weakest link on both sides of the argument, which is precisely what the replication attempt was about.

Why two honest studies disagree about the same food

Most contradictory food footprint numbers are not evidence that someone is lying. They are evidence that two teams drew the boundary differently.

Martin Heller, Gregory Keoleian and Walter Willett reviewed 32 studies that used life cycle assessment to evaluate diets or meals, in Environmental Science and Technology in 2013. Their central finding was a methodological one: researchers had used a wide range of bases for comparison, and the choice of basis shapes the answer. They argued the field needs a more comprehensive nutritional basis for comparison, regionally specific inventory databases, and a scope that reaches past greenhouse gases.

In plain terms, two choices decide most disagreements.

The first is what you divide by. Per kilogram, per calorie, per gram of protein, per dollar, per serving, or per day of a whole diet. Those produce different rankings of the same foods, and none of them is the wrong question. The water-productivity flip earlier in this article is that effect in miniature.

The second is where you stop. A study can end at the farm gate, or follow food through processing, packaging, distribution, retail refrigeration, home cooking and the share thrown out uneaten. Land-use change can be counted or excluded. Each choice is defensible and each moves the total. Two careful studies that stopped in different places will not agree, and neither is wrong.

For a worked example of how much method choice matters, Derek Headey, Kalle Hirvonen and Harold Alderman revisited the cost-of-a-healthy-diet calculation in Food Policy in 2024. Adjusting for differences in countries’ age structures, since younger populations need fewer calories, and improving how household food budgets are derived moved the estimated number of people who cannot afford a healthy diet from 3.02 billion down to 2.13 billion. Same underlying question, nearly a billion people of difference, from two methodological corrections.

Four questions for any food footprint number

1. Per what? Kilogram, calorie, gram of protein, or a whole day’s diet?
2. Which indicator? Climate, water, land and nutrient pollution are separate answers, and they do not have to agree.
3. Where did the measurement stop? Farm gate, store shelf, or your kitchen bin?
4. Average or range? A category average sits in the middle of a distribution 50 times wide.

If a number arrives without answers to those four, it is a headline rather than a measurement.

What the data does not decide

Environmental measurement answers environmental questions. It does not rank the other inputs to a food decision, and those inputs are legitimate.

Cost is one, and the Hirvonen analysis puts a figure on it. Nutrition is another, and our article on eating well in later life handles nutritional adequacy. Culture is a third, and the EAT-Lancet report itself frames its diet as a reference meant to flex regionally rather than a single menu. Access is a fourth. Someone without a car, a full-service grocery store within reach or a working kitchen faces constraints that no footprint figure accounts for.

One practical note on health, because dietary change is where people get into trouble on their own. If you are considering a substantial change in what you eat and you are pregnant, feeding a young child, managing diabetes or kidney disease, or taking medication whose dose depends on diet, that is a conversation with your clinician or a registered dietitian before you start, not after. Restricted diets can run short on specific nutrients, and the sensible move is to have adequacy checked rather than assumed. Unintended weight loss, persistent fatigue or new digestive symptoms after a dietary change are reasons to see a clinician rather than to search harder online.

One sentence from 38,700 farms

That dataset has one lesson above the others. The spread inside a food category is wider than anything a shopper can see from the outside of a package, and the composition of a diet is measured far more reliably than the sourcing of it.

That is a statement about what is currently measurable, not about what anyone should eat. The measurements are useful, they are incomplete, and the decision they inform has at least four other inputs the measurements never touch.

Sources

1. Poore J, Nemecek T.. “Reducing food's environmental impacts through producers and consumers. Science, 2018;360(6392):987-992. PubMed. DOI: 10.1126/science.aaq0216.” doi.org
2. Willett W, Rockstrom J, Loken B, Springmann M, Lang T, et al.. “Food in the Anthropocene: the EAT-Lancet Commission on healthy diets from sustainable food systems. The Lancet, 2019;393(10170):447-492. PubMed. DOI: 10.1016/S0140-6736(18)31788-4.” doi.org
3. Zagmutt FJ, Pouzou JG, Costard S.. “The EAT-Lancet Commission's Dietary Composition May Not Prevent Noncommunicable Disease Mortality. The Journal of Nutrition, 2020;150(5):985-988. PubMed. DOI: 10.1093/jn/nxaa020.” doi.org
4. Hirvonen K, Bai Y, Headey D, Masters WA.. “Affordability of the EAT-Lancet reference diet: a global analysis. The Lancet Global Health, 2020;8(1):e59-e66. PubMed. DOI: 10.1016/S2214-109X(19)30447-4.” doi.org
5. Headey D, Hirvonen K, Alderman H.. “Estimating the cost and affordability of healthy diets: How much do methods matter? Food Policy, 2024;126:102654. PubMed. DOI: 10.1016/j.foodpol.2024.102654.” doi.org
6. Springmann M, Spajic L, Clark MA, Poore J, Herforth A, Webb P, Rayner M, Scarborough P.. “The healthiness and sustainability of national and global food based dietary guidelines: modelling study. The BMJ, 2020;370:m2322. PubMed. DOI: 10.1136/bmj.m2322.” doi.org
7. Heller MC, Keoleian GA, Willett WC.. “Toward a life cycle-based, diet-level framework for food environmental impact and nutritional quality assessment: a critical review. Environmental Science and Technology, 2013;47(22):12632-12647. PubMed. DOI: 10.1021/es4025113.” doi.org
8. Mekonnen MM, Neale CMU, Ray C, Erickson GE, Hoekstra AY.. “Water productivity in meat and milk production in the US from 1960 to 2016. Environment International, 2019;132:105084. PubMed. DOI: 10.1016/j.envint.2019.105084.” doi.org
9. Shepon A, Eshel G, Noor E, Milo R.. “The opportunity cost of animal based diets exceeds all food losses. Proceedings of the National Academy of Sciences, 2018;115(15):3804-3809. PubMed. DOI: 10.1073/pnas.1713820115.” doi.org
10. USDA Economic Research Service. “Major Land Uses.” ers.usda.gov
11. USDA Economic Research Service. “Irrigation and Water Use.” ers.usda.gov
12. U.S. Geological Survey. “Irrigation Water Use.” usgs.gov
13. U.S. Geological Survey. “Water Use in the United States.” usgs.gov
14. EAT Forum. “The Planetary Health Diet and You.” eatforum.org