Life-cycle assessment of food

Life-cycle assessment (LCA) is the standard instrument food-systems science uses to compare the environmental impact of one food, one diet, or one production system against another. It is also the instrument most often misread. An LCA does not return a verdict; it returns a number conditional on choices about scope, boundary, allocation, functional unit, and impact category. Understanding those choices is the difference between citing an LCA and being misled by one.

What an LCA measures

An LCA follows the ISO 14040 and 14044 standards (ISO, 2006). The workflow has four phases: goal and scope definition, life-cycle inventory (LCI), life-cycle impact assessment (LCIA), and interpretation. In practice, a food LCA traces material and energy flows from cradle — seed, fertiliser, feed, fuel — through farm production, processing, packaging, distribution, retail, consumption, and waste, then converts those flows into a common set of impact indicators.

The impact categories most commonly reported for food are:

• Greenhouse-gas emissions in kilograms CO2-equivalent, typically using the IPCC’s 100-year global warming potentials.
• Land use in square-metre-years, often split into arable, pasture, and permanent crop.
• Freshwater use, divided into “blue water” (irrigation drawn from rivers, lakes, aquifers) and “green water” (rainfall on crops).
• Eutrophication, measured as kilograms of phosphate-equivalent or nitrogen-equivalent reaching freshwater or marine systems.
• Acidification, measured as sulphur-dioxide-equivalent deposition.
• Energy demand in megajoules of primary energy.

Each indicator answers a different question. Greenhouse-gas emissions speak to climate; eutrophication speaks to dead zones; land use speaks to biodiversity and opportunity cost. A food that scores well on one category does not automatically score well on another, which is why serious LCAs report a panel of indicators rather than a single “footprint.”

Boundary choices and why they decide outcomes

Boundaries are where LCAs diverge most. The two most consequential boundary decisions in food LCA are where the system starts (cradle-to-farm-gate, cradle-to-retail, cradle-to-plate, cradle-to-grave) and how co-products are allocated.

A dairy cow produces milk, veal, and eventually cull beef. A hen produces eggs and eventually spent-hen meat. A soybean crush yields oil, meal, and hulls. Allocation rules — mass, economic value, protein content, or system expansion — decide how the impacts of the shared production are split between co-products. Change the rule and the per-kilogram emissions of milk, beef, soy oil, or soy meal can move by a factor of two or more (Notarnicola et al., 2017). The same physical system, different arithmetic, different headline.

Land-use change (LUC) is the other boundary with outsized influence. Including the carbon released when forest or grassland is cleared for soy, palm, or pasture pushes the emissions of feed-linked animal products substantially upward; excluding it flatters them. Poore & Nemecek (2018) make a strong case for including LUC on a 20-year amortisation, the approach most subsequent LCAs have adopted. Studies that omit LUC can report beef emissions roughly half those of studies that include it — the physical system is identical; the accounting is not.

The functional unit — the “per what” of the comparison — matters equally. Comparing foods per kilogram rewards water-heavy items like cucumbers. Comparing per kilocalorie rewards dense foods like grains and oils. Comparing per gram of protein, or per unit of nutrient density, changes the ranking yet again. Searchinger et al. (2018) argue for a “carbon opportunity cost” metric that expresses per-unit-of-protein emissions alongside the carbon forgone by not restoring the land — a framing that widens the plant–animal gap further by accounting for what the occupied land could otherwise sequester.

The Poore and Nemecek dataset

The dominant reference for food LCAs is the meta-analysis by Poore & Nemecek (2018), published in Science. The authors synthesised 570 peer-reviewed LCAs covering 38,700 commercial farms across 119 countries and 40 products, harmonised under consistent system boundaries (cradle-to-retail, 20-year LUC amortisation, biophysical allocation where possible). The dataset is openly available and underlies much of the food-policy modelling that followed (Clark et al., 2020; Crippa et al., 2021).

Headline findings: animal products supplied roughly 18 percent of calories and 37 percent of protein while using 83 percent of farmland and generating 56–58 percent of food-related greenhouse-gas emissions. A global shift to plant-based diets would cut food’s land use by about 76 percent and food’s greenhouse-gas emissions by about 49 percent. The earlier Clune et al. (2017) systematic review of 369 food LCAs, using slightly different scope rules, reached compatible per-kilogram rankings — confidence in the central tendency comes from replication across independent meta-analyses.

Why farm-to-farm variation exceeds food-type averages — sometimes

One of the most important and least quoted findings in Poore & Nemecek (2018) is the size of the spread within each food. For beef, the highest-impact 10 percent of producers emit roughly 12 times more greenhouse gases per kilogram of protein than the lowest-impact 10 percent. Similar spreads exist for dairy, pork, and aquaculture. For some categories, farm-to-farm variation within a single food exceeds the mean difference between that food and its lower-impact alternatives.

That variation is not uniform across indicators. It is largest for GHG emissions and land use in ruminant systems, where feed quality, pasture productivity, herd management, enteric-methane intensity, and climate all compound. It is smaller for water use of field-grown crops in the same basin, and smaller still for eutrophication from standardised fertiliser regimes. McAuliffe et al. (2018) documented how even within a single pasture-based beef system, animal-level variation in growth and feed efficiency produces a roughly two-fold spread in emissions intensity.

The important caveat is that variation does not erase the gap. Poore & Nemecek found that the lowest-impact producer of beef still emitted more greenhouse gases per gram of protein than the highest-impact producer of peas, tofu, or most pulses. The distributions overlap for some impact categories and some product pairs (for example, low-impact dairy vs. high-impact nuts on water use in specific basins), but for climate and land, the central tendency dominates the tails. “Buy the best beef” is a real lever on an individual farm; it is not a substitute for dietary shift.

LCA-based findings at the system scale

Clark et al. (2020) took the LCA distributions from Poore & Nemecek, projected them forward under population and dietary scenarios, and asked whether the 1.5 and 2 °C Paris targets could be met if every non-food sector fully decarbonised. The answer was no — food-system emissions alone, on current trajectories, would exhaust the 1.5 °C budget by mid-century and make 2 °C difficult. The interventions needed were a combination of plant-rich diets, higher yields, reduced food loss, and cleaner production.

Crippa et al. (2021), using the EDGAR-FOOD global inventory rather than bottom-up LCA, estimated food-system emissions at 18 Gt CO2-equivalent in 2015 — about 34 percent of total anthropogenic emissions — with land-use change and agricultural production dominating. The two methodologies converge on the same order of magnitude and the same qualitative conclusion: food is a primary climate lever, and animal products concentrate the impact within it.

Heller & Keoleian (2015) applied LCA at national dietary scale for the United States, estimating that current US diets generate roughly 4.7 kg CO2-equivalent per person per day from food alone, with food loss adding around 0.9 kg. Shifting toward USDA-recommended patterns cut emissions modestly; shifting toward plant-based patterns cut them substantially more. LCA at the diet level is where the methodology has the clearest policy implication.

Criticisms and blind spots

LCA is powerful, and it is incomplete. Four limits deserve attention.

Biodiversity is poorly represented. Most food LCAs report land area but not the biodiversity value of that land. A hectare of degraded pasture, a hectare of soy monoculture, and a hectare of agroforestry all count the same in the standard indicator — which is why high-land-use plant foods like nuts can appear worse than dairy on “land occupation” while supporting different ecological communities. Characterisation factors for biodiversity impact exist (species-year lost per square-metre-year) but are not consistently applied.

LCA is static. A standard LCA represents a snapshot of current production. It does not capture soil-carbon dynamics under multi-year management change, rebound effects from efficiency gains, or the trajectory of a transitioning system. Regenerative-agriculture claims — soil sequestration, rotational grazing, holistic management — sit awkwardly in this framework. Some practices do accumulate soil carbon, but rates saturate, are reversible, and rarely offset the enteric methane they accompany over any meaningful horizon. LCA assessments of “carbon-neutral beef” claims typically find that the sequestration, where real, is partial and time-limited (Searchinger et al., 2018).

Allocation choices remain contested. There is no physical fact of the matter about how to divide emissions between milk and beef in a dairy system. Different standards produce different numbers, and practitioners sometimes choose the rule that suits the story. ISO 14044 permits multiple approaches and requires transparency rather than enforcing one.

Health, welfare, and social dimensions sit outside. LCA measures environmental flows, not animal welfare, not nutrient bioavailability, not worker conditions, not cultural food roles. Using an LCA to settle whether a food is “good” smuggles a value judgement past the method.

When LCA favours plant-based — and why

The pattern across the peer-reviewed food-LCA literature is consistent, not universal. Plant-based foods outperform animal-based counterparts on greenhouse-gas emissions, land use, eutrophication, and acidification in the vast majority of comparisons, across boundary choices, allocation rules, and regions (Poore & Nemecek, 2018; Clune et al., 2017; Crippa et al., 2021). They do not universally outperform on blue-water use (irrigated almonds and rice can exceed grass-fed beef in specific basins), and they do not outperform at all on biodiversity or soil when the comparison is monoculture soy versus well-managed silvopasture — a specific, small slice of production.

The reason the central tendency is so durable is thermodynamic. Feeding a crop to an animal and eating the animal wastes the majority of the original calorie, protein, and embedded input. Trophic losses of roughly an order of magnitude are not an accounting artefact; they are a property of biological conversion. LCA makes that property visible, which is why its results bend the direction they do even when every boundary choice is contested.

The correct use of LCA is neither to canonise a single number nor to dismiss the method because numbers vary. It is to report a panel of indicators with transparent scope, to read variation as information rather than noise, and to distinguish the questions the method answers well (comparative impact under defined conditions) from the ones it answers poorly (biodiversity value, soil trajectories, welfare). Done that way, food LCA is among the most useful tools in the environmental toolkit — and its cumulative message about where the pressure sits has not meaningfully changed in two decades of refinement.