Science Journals

SUSTAINABILITY

Reducing food’s environmental

impacts through producers

and consumers

J. Poore

1,2

* and T. Nemecek

F ood’s environmental impacts are cr eated by millions of diverse pr oducers. To identify solutions

that are effectiv e under this heterogeneity, we consolidated data cov ering five environmental

indicators; 38,700 farms; and 1600 processors, packaging types, and retailers. Impact can vary

50-fold among producers of the same product, creating substantial mitigation opportunities.

Howev er, mitigation is complicated by trade-offs, multiple ways for producers to achieve low

impacts, and interactions throughout the supply chain. Producers have limits on how far they can

reduce impacts. Most strikingly , impacts of the lowest-impac t animal products typically ex ce ed

those of veg etable substitutes, providing new evidence for the importance of dietary change.

Cumulatively , our findings support an approach where producers monitor their own impacts,

flexibly meet environmental targets by choosing from multiple practices, and communicate their

impacts to consumers.

ith current diets and production prac-

tices, feeding 7.6 billion people is degrad-

ing terrestrial and aquatic ecosystems,

depleting water resources, and driving

climate change (1, 2). It is particularly

challenging to find solutions that are effective

across the large and diverse range of producers

that characterize the agricultural sector. More

than 570 million farms produce in almost all the

world’sclimatesandsoils(3), each using vastly

different agronomic methods; average farm sizes

vary from 0.5 ha in Bangladesh to 3000 ha in

Austral ia (3); average mineral fertilizer use ranges

from 1 kg of nitrogen per ha in Uganda to 300 kg

in China (4); and although four crops provide half

of the world’s food calories (4), more than 2 million

di stinct varieties are recorded in seed vaults (5).

Further , products range from minimally to heavily

processed and packaged, with 17 of every 100 kg of

food produced transported internationally, increas-

ingto50kgfornutsand56kgforoils(4).

Previous studies have assessed aspects of this

heterogeneity by using geospatial data sets (6–8),

but global assessments using the inputs, outputs,

and practices of actual producers have been lim-

ited by data. The recent rapid expansion of the

life cycle assessment (LCA) literature is providing

this information by surveying producers around

the world. LCA then uses models to translate pro-

ducer data into environmental impacts with suf-

ficient accuracy for most decision-making (9–11).

To date, efforts to consolidate these data or build

new large-scale data sets have covered greenhouse

gas (GHG) emissions only (8, 12, 13), agriculture

only (13–16), small numbers of products (8, 14–16),

and predominantly Western European producers

(12–16) and have not corrected for important meth-

odological differences between LCAs (12–16). Here,

we present a globally reconciled and methodolog-

ically harmonized database on the variation in food’s

m ultiple impacts. Our results show the need for

far-reaching changes in how food’senvironmental

impacts are managed and communicated.

Building the multi-indicator

global database

We derived data from a comprehensive meta-

analysis, identifying 1530 studies for potential

inclusion, which were supplemented with addi-

tional data received from 139 authors. Studies

were assessed against 11 criteria designed to

standardize methodology, resulting in 570 suit-

able studies with a median reference year of

2010 (17).

The data set covers ~38,700 commer-

cially

viable farms in 119 countries (fig. S2) and

40 products representing ~90% of global pro-

tein and calorie consumption. It covers five im-

portant environmental impact indicators (18):

land use; freshwater withdrawals weighted by

local water scarcity; and GHG, acidifying, and

eutrophying emissions. For crops, yield repre-

sents output for a single harvest. Land use in-

cludes multicropping (up to four harvests per

year), fallow phases (uncultivated periods be-

tween crops), and economic allocation to crop

coproducts such as straw. This makes it a stron-

ger indicator of both farm productivity and

food security than yield.

The system we assess begins with inputs (the

initial effect of producer choice) and ends at re-

tail (the point of consumer choice) (fig. S1). For

each study, we recorded the inventory of out-

puts and inputs (including fertilizer quantity

and type, irrigation use, soil, and climatic con-

ditions). Where data were not reported, for ex-

ample, on climate, we used study coordinates

and spatial data sets to fill gaps. We recorded

environmental impacts at each stage of the sup-

ply chain. For GHG emissions, we further disag-

gregated the farm stage into 20 emission sources.

We then used the inventory to recalculate all

missing emissions. For nitrate leaching and

aquaculture, we developed new models for this

study (17).

Studies included provided ~1050 estimates

of postfarm processes. To fill gaps in process-

ing, packaging, or retail, we used additional

meta-analyses of 153 studies providing 550 ob-

servations. Transport and losses were included

from global data sets. Each observation was

weighted by the share of national production it

represents, and each country by its share of

global production. We then used randomiza-

tion to capture variance a t all stages of the

supply chain (17).

We validated the global representativeness of

our sample by comparing average and 90th-

percentile yields to Food and Agriculture Or-

ganization (FAO) data (4), which reconcile to

within ±10% for most crops. Using FAO food

balance sheets (4), we scaled up our sample data.

Total arable land and freshwater withdrawals

reconcile to FAO estimates. Emissions from de-

forestation and agricultural methane fall within

ranges of independent models (17).

Environmental impacts of the entire

food supply chain

Today’s food supply chain creates ~13.7 billion

metric tons of carbon dioxide equivalents (CO

eq),

26% of anthropogenic GHG emissions. A further

2.8 billion metric tons of CO

eq (5%) are caused

by nonfood agriculture and other drivers of de-

forestation (17). Food production creates ~32%

of global terrestrial acidification and ~78% of

eutrophication. These emissions can fundamen-

tally alter the species composition of natural

ecosystems, reducing biodiversity and ecological

resilience (19). The farm stage dominates, rep-

resenting 61% of food’sGHGemissions(81%

including deforestation), 79% of acidification,

and 95% of eutrophication (table S17).

Today’s agricultural system is also incredibly

resource intensive, covering ~43% of the world’s

ice- and desert-free land. Of this land, ~87% is

for food and 13% is for biofuels and textile crops

or is allocated to nonfood uses such as wool and

leather. We estimate that two-thirds of freshwater

withdrawals are for irrigation. However, irriga-

tion returns less water to rivers and groundwater

than industrial and municipal uses and pre-

dominates in water-scarce areas and times of

the year, driving 90 to 95% of global scarcity-

weighted water use (17).

Highly variable and skewed

environmental impacts

We now group products by their primary dietary

role and express impacts per unit of primary

nutritional benefit (Fig. 1 and fig. S3). Immedi-

ately apparent in our results is the high variation

in impact among both products and producers.

Ninetieth-percentile GHG emissions of beef

are 105 kg of CO

eq per 100 g of protein, and

RESEARCH

Poore et al., Science 360, 987–992 (2018) 1 June 2018 1of6

Department of Zoology, University of Oxford, New Radcliffe

House, Oxford OX2 6GG, UK.

School of Geography and the

Environment, University of Oxford, South Parks Road, Oxford

OX1 3QY, UK.

Agroscope, Agroecology and Environment Research

Division, LCA Research Group, CH-8046 Zürich, Switzerland.

*Corresponding author. Email: [email protected]

Erratum 22 February 2019. See Erratum.

on February 26, 2019 http://science.sciencemag.org/Downloaded from

land use (area multiplied by years occupied) is

370 m

·year. These values are 12 and 50 times

greater than 10th-percentile dairy beef impacts

(which we report separately given that its pro-

duction is tied to milk demand). Tenth-percentile

GHG emissions and land use of dairy beef are

then 36 and 6 times greater than those of peas.

High variation within and between protein-rich

products is also manifest in acidification, eutro-

phication, and water use.

Within the major crops wheat, maize, and rice,

90th-percentileimpactsaremorethanthreetimes

greater than 10th-percentile impacts on all five

indicators. Within major growing areas for these

crops (the Australian wheat belt, the U.S. corn

belt, and the Yangtze river basin), land use be-

comes less variable, but we observe the same

high levels of variation in all other indicators.

This variability, even among producers in similar

geographic regions, implies substantial potential

to reduce environmental impacts and enhance

productivity in the food system.

For many products, impacts are skewed by

producers with particularly high impacts. This

creates opportunities for targeted mitigation,

making an immense problem more manageable.

For example, for beef originating from beef herds,

the highest-impact 25% of producers represent

56% of the beef herd’sGHGemissionsand61%of

the land use (an estimated 1.3 billion metric tons

of CO

eq and 950 million ha of land, primarily

pasture). Across all products, 25% of producers

contribute on average 53% of each product’senvi-

ronmentalimpact(fig.S3).Forscarcity-weighted

freshwater withdrawals, the skew is particular-

ly pronounced: Producing just 5% of the world’s

food calories creates ~40% of the environmental

burden. We will now explore how to access these

mitigation opportunities through heterogenous

producers.

Mitigation through producers

Enable producers to monitor

multiple impacts

The first step in mitigation is estimating pro-

ducer impacts. Prior research [e.g., (7, 8, 14)] has

suggested that readily measurable proxies pre-

dict farm-stage impacts, avoiding the need for

detailed assessment. From our larger data set,

which includes more practices and geographies

than prior studies, we assess the predictive power

of common proxies, including crop yield, nitro-

gen use efficiency, milk yield per cow, liveweight

gain, pasture area, and feed conversion ratios.

Although most proxi es signi fican tly covary with

impa ct , they make poor predictors when used

alone, explaining little of the variation among

farms (coefficient of determination R

=0to27%

in 47 of 48 proxy-impact combinations assessed)

(fig. S4).

Prior research has also suggested using one

impact indicator to predict others (20). We find

weakly positive and sometimes negative relation-

ships between indicators. For similar products

globally, correlations between indicators are low

= 0 to 30% in 26 of 32 impact-impact com-

binations assessed) (fig. S4). Pork, poultry meat,

Poore et al., Science 360, 987–992 (2018) 1 June 2018 2of6

100g protein

Beef (beef herd)

724

20 50 42 164

Lamb & Mutton

757

12 20 30 185

Beef (dairy herd)

490

9.1 17 7.3 22

Crustaceans (farmed)

1.0k

5.4 18 0.4 2.0

Cheese

1.9k

5.1 11 4.4 41

Pig Meat

116

4.6 7.6 4.8 11

Fish (farmed)

612

2.56.0 0.43.7

Poultry Meat

326

2.45.7 3.87.1

Eggs

100

2.64.2 4.05.7

Tofu

354

1.02.0 1.12.2

Groundnuts

100

0.61.2 1.83.5

Other Pulses

115

0.50.8 4.67.3

Peas

438

0.30.4 1.23.4

Nuts

199

-2.2 0.3 2.7 7.9

Grains

23k

1.02.7 1.74.6

1liter

Milk

1.8k

1.73.2 1.18.9

Soymil

354

0.61.0 0.30.7

1000 kcal

Cassava

288

0.41.4 0.81.9

Rice (flooded)

7.8k

0.41.2 0.30.8

Oatmeal

139

0.30.9 1.12.9

Potatoes

604

0.20.6 0.61.2

Wheat & Rye (Bread)

8.8k

0.30.6 0.41.4

Maize (Meal)

6.2k

0.20.4 0.30.7

1liter

Palm Oil

220

3.67.3 1.72.4

Soybean Oil

497

2.4 6.3 5.3 11

Olive Oil

411

2.9 5.4 7.9 26

Rapeseed Oil

1.8k

2.5 3.8 5.2 11

Sunflower Oil

519

2.5 3.6 8.4 18

1kg

Tomatoes

855

0.42.1 0.10.8

Brassicas

0.20.5 0.20.6

Onions & Leeks

0.30.5 0.10.4

Root Vegetables

0.20.4 0.20.3

1kg

Berries

183

0.81.5 0.32.4

Bananas

246

0.60.9 0.31.9

Apples

125

0.30.4 0.30.6

Citrus

377

0.10.4 0.40.9

1kg

Cane Suga

116

0.93.2 1.22.0

Beet Suga

209

1.21.8 1.21.8

1 unit

Beer (5% ABV)

695

0.14 0.24 0.05 0.22

Wine (12.5% ABV)

154

0.07 0.14 0.07 0.14

1 serving

Dark Chocolate (50g)

162

-0.01 2.3 1.7 3.4

Coffee (15g, 1 cup)

346

0.08 0.4 0.13 0.3

0 255075

0 5 10 15

0246

Land Use

year)

GHG Emissions

(kg CO

eq)

0123

0246

00.20.40.6

0369

0123

pctl.

ruminant

meat

051015

percentile10

percentile

Median

Mean

0 100 200 300

0369

0246

00.511.5

01234

0 0.2 0.4 0.6

0246

0 102030

01530

0510

01020

02040

012

01020

03060

075150

01020

0510

01020

02040

00.51

01020

03060

0369 051001020

Acid.

(g SO

eq)

Eutroph.

(g PO

3–

eq)

050100

075150075150050100

Scty. Water

(kL eq)

050100

02040

01020

02040

00.10.2

03060

0100200

00.10.2

0751500 5 10 15

Mean

Fig. 1. Estimated global variation in GHG emissions, land use, terrestrial acidification,

eutrophication, and scarcity-weighted freshwater withdrawals, within and between

40 major foods. (A) Protein-rich products. Grains are also s hown here given th at they

contribute 41% of global protein intake, despite lower protein content. (B)Milks.

beverages (1 unit = 10 ml of alco hol; ABV, alcohol by volume). (I)Stimulants.n =farm

or regional inventories. Pc and pctl., percentile; scty., scarcity.

RESEARCH | RESEARCH ARTICLE

Erratum 22 February 2019. See Erratum.

on February 26, 2019 http://science.sciencemag.org/Downloaded from

and milk show higher correlations between acid-

ification and eutrophication (R

≤ 54%), explained

by the dominant role of manure in these impacts,

but this does not generalize to other products or

indicators. The same conclusion holds for farms

in similar geographies or systems (fig. S5).

Monitoring multiple impacts and avoiding

proxies supports far better decisions and helps

prevent harmful, unintended consequences. How-

ever, two recent studies suggest that data on

practices and geography, required to quantify

impacts, must come directly from producers

(11, 21), that quantifying impacts with the use

of satellite or census data misses much of the

variation among farms.

Set and incentivize mitigation targets

When land use or emissions are low, we find

trade-offs between indicators for many crops

(fig. S5). This reflects diminishing marginal yield

with increasing inputs as crops tend toward their

maximum yields (22). For example, for already

low-emission Northern European barley farms,

halving land use can increase GHG emissions per

kilogram of grain by 2.5 times and acidification

by 3.7 times. To explore trade-offs further , we pair

observations from the same study, location, and

year that assess a practice change (fig. S6). Of the

nine changes assessed, only two (changing from

monoculture to diversified cropping and improv-

ing degraded pasture) deliver statistically significant

r e du ctions in both land use and GHG emissions.

Geography influences these trade-offs. For ex-

ample, in the Australian wheat belt, where farmers

practice low-rainfall, low-input farming, we find

that both output per hectare and GHG emissions

are in the bottom 15% globally. The environmental

and social importance of different impacts also

varies locally, given land scarcity, endemic bio-

diversity, and water quality, among other factors.

Setting regional and sector-specific targets will

help producers navigate trade-offs and make

choices that align with local and global priorities.

Meet targets by choosing from multiple

practice changes

To meet these targets, policy might encourage wide-

spread adoption of certain practices. However, the

environmental outcomes of many practices, such as

conservation agriculture (23), organic farming (fig.

S6), and even integrated systems of best practice

(24),arehighlyvariable.Usingourdataset,wecan

generalize these findings. To do this, we disaggre-

gate each environmental indicator into its sources

or drivers. We consider practice change as a pack-

age of measures that targets one or more of these

sources. If producers have different impact sources,

the effects of practice change will be variable.

We find that sources of impact vary consider-

ably among farms producing the same product

(Fig. 2 and figs. S7 to S9). Priority areas for

reducing impact for one farm may be immaterial

for another . For example, measures to reduce di-

rect nitrous oxide emissions from synthetic and

organic fertilizer , such as biochar application, are

included in many mitigation estimates (25). How-

ever , for a third of global crop calorie production,

these emissions represent less than 5% of farm-

stage GHGs. It may be the case that low-impact

farms have similar impact drivers. We again find

variable sources of impact, even for low-impact

farms (Fig. 2, C and D). Reducing impacts means

focusing on different areas for different producers

and, by implication, adopting different practices.

Toexplorethisfurther,weusesensitivityanalysis

(26) to decompose the variance in each product’s

impact into its sources. Numerous sources con-

tribute to variance (fig. S10). Most notably, for all

crop calorie production globally, differences in fal-

low duration and multiple cropping drive 40% of

thevarianceinlanduse.Thisisimportantasmost

strategies to increase productivity are focused on

increasing single crop yields (27). But for many

producers, increasing cropping intensity through

the use of early-maturing varieties, intercropping,

ca tch crops, and enhanced irrigation can provide

more economically viable and trade-off–free ways

to boost productivity and reduce impacts (27).

Geography plays a major role in this variation

and affects the economic and environmental

desirability of different practices (28). However,

attheheartofagricultureischangingsitecon-

ditions to enhance productivity (such as liming,

terracing, or installing drainage), meaning that

statements on the importance of geography have

limitations. Nevertheless, some impact sources

stand out. We find that freshwater aquaculture

ponds

create

0 to 450 g of methane per kg of

liveweight (for context, enteric fermentation

in dairy cows creates ~30 to 400 g per kg of live-

weight). Of this variation, a third is explained by

temperature (17), which accelerates methano-

genesis and net primary production. Improving

aeration and limiting addition of surplus feed

to ponds can abate these emissions, particularly

important in warm countries. Further , for every

kilogram of nitrogen applied to crops, between

60 and 400 g is lost in reactive forms. Of this

wide range, ~40% is explained by site conditions,

including soil pH, temperature, and drainage

(17). Prior research has also found that the po-

tential of soil to store carbon varies significantly

with soil properties, slope, and prior practice (29).

Providing producers with multiple ways to

reduce their environmental impacts recognizes

the variability in sources and drivers of impact

but requires a step change in thinking: that prac-

tices such as conservation agriculture or organic

farming are not environmental solutions in them-

selves but options that producers choose from

to achieve environmental targets.

However, some practice changes can be pur-

sued across all producers. Methane from flooded

rice, enteric methane from ruminants, and con-

centrate feed for pigs and poultry are sizeable

globally, representing 30% of food’s GHG emis-

sions; are material for all producers, contributing

at least 17% of farm-stage emissions (Fig. 2B and

fig. S7); and can be mitigated with relatively trade-

off–free approaches such as shorter and shallower

rice flooding (30), improving degraded pasture

Poore et al., Science 360, 987–992 (2018) 1 June 2018 3of6

Seed

Fertilizer & Pesticide

Manufacture

Equipment

Electricity & Fuel

Direct

Indirect

Organic Fertilizer (N

Crop Residue

Urea (CO

)

Lime (CO

)

Residue Burning

Drying

Cyprus,

irrigated

Germany,

organic

Nepal,

raised beds

Synthetic

Fertilizer

25%

50%

75%

100%

Contribution of each source to farm-stage GHG emissions

Contribution of each source to farm-

stage GHG emissions

Australia,

grass-fed

Brazil,

improved

pasture

Denmark,

dairy, 9mo

at slaughter

25%

50%

75%

100%

Contribution of each source

Concentrate Feed (incl.

land use change)

Pasture Management

Housing

Enteric Fermentation (CH

)

Manure Management (CH

)

Direct

Indirect

Manure

Management

Contribution of each source to farm-

stage GHG emissions

0% 25% 50% 75% 100%

Distribution

Percentiles

10th 90th

Wheat farms Below median GHG emissions wheat farms

Beef farms Below median GHG emissions beef farms

Australia,

conventional

Australia,

no-till,

residue burnt

China,

confinement

dairy

Canada,

feedlot

with implants

Fig. 2. Contributions of emission sources to total farm-stage GHG emissions. (A and B) Gray

bars show 10th- and 90th-percentile contributions. Shaded bars represent the distribution. For

example, the 90th-percentile contribution of organic fertilizer N

O to farm-stage emissions is

16%, but for most wheat producers the contribution is near 0%. Density is estimated using a

Gaussian kernel with bandwidth selection performed with biased cross-validation. (C and D)

Contributions of emission sources for example producers with below-median GHG emissions.

RESEARCH | RESEARCH ARTICLE

Erratum 22 February 2019. See Erratum.

on February 26, 2019 http://science.sciencemag.org/Downloaded from

(fig. S6), and improving lifetime animal produc-

tivity (8). Further, emissions from deforestation

and cultivated organic soils drive on average

42% of the variance in each product’s agricul-

tural GHG emissions (fig. S10) and dominate

the highest-impact producers’ emissions (fig. S11),

further justifying ongoing efforts to curb forest

loss and limit cultivation on peatlands.

Communicate impacts up the

supply chain

Processors, distributors, and retailers can substan-

tially reduce their own impacts. For any product,

90th -percentil e postfarm emissions are 2 to 140

times lar ger tha n 10th- p e rcentile emissions, indi-

cating large mitigation potential (fig. S12). For

example, returnable stainless steel kegs create just

20 g of CO

eq per liter of beer , but recycled glass

bottles create 300 to 750 g of CO

eq, and bottles

sent to landfills create 450 to 2500 g of CO

eq.

Processing, more durable packaging , and

greater usage of coproducts can also reduce food

waste. For example, wastage of processed fruit

and vegetables is ~14% lower than that of fresh

fruit and vegetables, and wastage of processed

fish and seafood is ~8% lower (24). Providing

processors and retailers with information about

the impacts of their providers could encourage

them to reduce waste where it matters most.

For products such as beef, distribution and retail

losses contribute 12 to 15% of emissions (fig. S13),

whereas the sum of emissions from packaging,

transport, and retail contributes just 1 to 9%.

Here, reducing losses is a clear priority.

As a third strategy, procurement could source

from low-impact farms. Although this strategy is

important, and possible only with information

about the impacts of providers, it has clear limita-

tions. To be effective, it relies on high-impact

production not simply being purchased elsewhere

inthemarket.ThecaseoftheRoundtableon

Sustainable Palm Oil (RSPO) shows that this is

hard to achieve: despite one-fifth of 2017 palm oil

production being certified, there remains virtu-

ally no demand in China, India, and Indonesia

(31). Alternatively, this strategy would be ef-

fective if higher prices for sustainable produc-

tion incentivized low-impact producers to increase

output or high-impact producers to change prac-

tices. The case of organic food shows how passing

premiums to consumers limits total market size

and widespread practice change.

However, processors and retailers routinely

demand that products meet taste, quality, and

food safety standards. These markets are con-

centrated, with just 10 retailers representing

52% of U.S. grocery sales and 15% of global sales

(32). This sometimes means that standards

achieve market transformation (33), where vir-

tually all producers adhere to gain market access.

A fourth strategy for producers is setting en-

vironmental standards. These are particularly

important: Although many environmental issues

can be monitored and mitigated in a flexible

way, issues such as harmful pesticide usage and

deforestation require strict controls, and issues

such as on-farm biodiversity are hard to quantify

(28). Procurement, farming organizations, and in-

ternational policy-makers must come together to

implement a safety net for global agriculture—

comprehensive standards to manage the worst

and hardest-to-quantify environmental issues,

extending the successes of existing schemes and

enabling a flexible mitigation approach to op-

erate effectively.

Producer mitigation limits and the role

of consumers

Though producers are a vital part of the solu-

tion, their ability to reduce environmental impacts

is limited. These limits can mean that a product

has higher impacts than another nutritionally

equivalent product, however it is produced.

In particular , the impacts of animal products

can markedly exceed those of vegetable substitutes

(Fig. 1), to such a degree that meat, aquaculture,

eggs, and dairy use ~83% of the world’s farmland

and contribute 56 to 58% of food’sdifferentemis-

sions, despite providing only 37% of our protein and

18% of our calories. Can animal products be pro-

duced with sufficiently low impacts to redress this

vast imbalance? Or will reducing animal product

consumption deliver greater environmental benefits?

We find that the impacts of the lowest-impact

animal products exceed average impacts of sub-

stitute vegetable proteins across GHG emissions,

eutrophication, acidification (excluding nuts), and

frequently land use (Fig. 1 and data S2). These

stark differences are not apparent in any product

groups except protein-rich products and milk.

Although tree crops can temporarily sequester

carbon and reduce nutrient leaching, the impact

of nuts is dominated by low-yielding cashews

and water-, fertilizer-, and pesticide-intensive

almonds. Production of nuts doubled between

2000 and 2015 (4), and more work is required to

improve their resource use efficiency. Although

aquaculture can have low land requirements, in

part by converting by-products into edible pro-

tein, the lowest-impact aquaculture systems still

exceed emissions of vegetable proteins. This chal-

lenges recommendations to expand aquaculture

(1) without major innovation in production prac-

tices first. Further, though ruminants convert

~2.7 billio n metric tons o f grass dry matter, of

which 65% grows on land unsuitable for crops

(34), into human-edible protein each year, the

environmental imp acts of this conversion are

immense under any production method prac-

ticed today.

Using GHG emissions (Fig. 3), we identified five

primarily biophysical reasons for these results.

These reasons suggest that the differences between

Poore et al., Science 360, 987–992 (2018) 1 June 2018 4of6

0.3

0.6

0.2

2.1

1.3

0.5

2.1

0.6

1.0

0.5

0.1

0.2

Lamb &

Mutton

Cheese Pig Meat Fish

(farmed)

Poultry

Meat

Eggs

1.4

1.6

1.8

2.7

1.1

2.7

0.8

1.4

2.0

1.4

2.1

1.4

1.6

0.5

1.8

0.6

0.8

1.7

9.0

6.0

1.5

3.5

0.5

1.2

6.0

5.0

1.9

3.7

1.0

0.9

0.5

0.4

GHG Emissions (kg CO

eq 100g protein

–1

)

Avg. veg.

proteins

pctl.

veg. proteins

0.21

Avg. veg. proteins

(excluding nuts)

0.2

0.1

1.5

1.0

GHG Emiss.

(kg CO

eq L

–1

)

Milk

Land Use Change

Crop Production & Feed Transport

Livestock & Aquaculture

-0.23

-0.76

0.20

0.02

0.77

0.54

0.11

0.09

12.0

Mean (light shaded box) 10

Percentile (dark shaded box)

Beef

(beef herd)

Beef

(dairy herd)

Crustaceans

(farmed)

-0.2

0.5

Fig. 3. Mean and 10th-percentile GHG emissions of protein-rich products across three major

production stages. (A to C) Red lines represent average vegetable protein emissions, and blue

l ines represent 10th-percentile emiss ions. The gr ay line represents average emissions excluding

nuts, w hich ca n temporarily sequester carbon if g rown on cropland or pasture. To calculate

10th-percentile emissions by stage, we averaged across farms that have total emissions between

the 5th and 15th percentiles, controlling for burden shifting between stages.

RESEARCH | RESEARCH ARTICLE

Erratum 22 February 2019. See Erratum.

on February 26, 2019 http://science.sciencemag.org/Downloaded from

animal and vegetable proteins will hold into the

future unless major technological changes dispro-

portionately target animal products. First, emissions

from feed production typically exceed emissions of

vegetable protein farming. This is because feed–

to–edible protein conversion ratios are greater than

2 for most animals (13, 34); because high usage of

low-impact by-products is typically offset by low

digestibility and growth; and because additional

transportisrequiredtotakefeedtolivestock.Sec-

ond, we find that deforestation for agriculture is

dominated(67%)byfeed,particularlysoy,maize,

and pasture, resulting in losses of above- and below-

gr ound carbon. Improved pasture management

can temporarily sequester carbon (25),butitre-

duces life-cycle ruminant emissions by a maximum

of 22%, with greater sequestration requiring more

land.Third,animalscreateadditionalemissions

from enteric fermentation, manure, and aquaculture

p onds. For these emissions alone, 10th-percentile

values are 0.4 to 15 kg of CO

eq per 100 g of pro-

tein. Fourth, emissions from processing, particu-

larly emissions from slaughterhouse effluent, add

afurther0.3to1.1kgofCO

eq, which is greater

than processing emissions for most other products.

Last, wastage is high for fresh animal products,

which are prone to spoilage.

Mitigation through consumers

Today, and proba bly into the future, dietary

change can deliver environmental benefits on

a scale not achievable by producers. Moving from

current diets to a diet that excludes animal pro-

ducts (table S13) (35) has transformative potential,

reducing food’s land use by 3.1 (2.8 to 3.3) billion ha

(a 76% reduction), including a 19% reduction

in arable land; food’sGHGemissionsby6.6(5.5to

7.4) billion metric tons of CO

eq (a 49% reduction);

acidification by 50% (45 to 54%); eutrophication by

49% (37 to 56%); and scarcity-weighted freshwater

withdrawals by 19% (−5 to 32%) for a 2010 refer-

ence year. The ranges are based on producing new

vegetable proteins with impacts between the 10th-

an d 90th-percentile impacts of existing produc-

tion. In addition to the reduction in food’s annual

GHG emissions, the land no longer required for

foodproductioncouldremove~8.1billionmetric

tons of CO

from the atmosphere each year over

100 years as natural vegetation reestablishes and

soil carbon re-accumulates, based on simulations

conducted in the IMAGE integrated assessment

mod e l (17). For the United States, where per capita

meat consumption is three times the global av-

erage, dietary change has the potential for a far

greater effect on food’s different emissions, reduc-

in g them by 61 to 73% [s e e supplementary text (17)

for diet compositions and sensitivity analyses

and fig. S14 for alternative scenarios].

Consumers can play another important role by

avoiding high-impact producers. We consider a

second scenario where consumption of each ani-

mal product is halved by replacing production

with above-median GHG emissions with vegeta-

ble equivalents. This achieves 71% of the previous

scenario’s GHG reduction (a reductio n of ~10.4

billion metric tons of CO

eq per year, including

atmospheric CO

removal by regrowing vege-

tation) and 67, 64, and 55% of the land use, acid-

ification, and eutrophication reductions. Further,

lowering consumption of more discretionary

products (oils, sugar, alcohol, and stimulants)

by 20% by avoiding production with the highest

land use reduces the land use of these products

by 39% on average. For emissions, the reductions

are 31 to 46%, and for scarcity-weighted fresh-

water withdrawals, 87%.

Communicating average product impacts to

consumers enables dietary change and should

be pursued. Though dietary change is realistic

for any individual, widespread behavioral change

will be hard to achieve in the narrow timeframe

remaining to limit global warming and prevent

further, irreversible biodiversity loss. Communi-

cating producer impacts allows access to the

second scenario, which multiplies the effects of

smaller consumer changes.

An integrated mitigation framework

In Fig. 4 we illustrate a potential framework im-

plied by our findings, prior research, and emerg-

ing policy (9). First, producers would monitor

their impacts using digital tools (36). Data would

be validated against known ranges for each value

(e.g., maximum yields given inputs) and validated

or certified independently. In the United States

these tools have already been integrated with ex-

isting farm software (31); in Africa and South Asia

they are in trials with 2G mobile phones (37); and

in China they have been operated by extension

services with extremely successful results (24).

Second, policy-makers would set targets on

environmental indicators and incentivize them

by providing producers with credit or tax breaks

or by reallocating agricultural subsidies that now

exceed hal f a tril lion dol lar s a year worldwide

(38). Third, the assessment tools would provide

multiple mitigation and productivity enhancement

options to producers. Ideally these tools would be-

come platforms that consolidate the vast amounts

of research conducted by scientists around the

world, while also sharing producer best practices.

In particular , practice sharing offers a very effec-

tive way to engage producers (24). Maximum

flexibility also ensures least-cost mitigation (39)

and supports producer-led innovation (24).

Finally, impacts would be communicated up the

supply chain and through to consumers. For com-

modity crops that are hard to trace (31), this may

not be feasible and mitigation efforts may have to

focus on producers. For animal products, stringent

traceability is already required in many countries

(40), suggesting that communicating impacts is

most feasible where it matters the most. Commu-

nication could occur through a combination of en-

vironmental labels, taxes or subsidies designed to

reflect environmental costs in product prices (35),

and broader education on the true cost of food.

We have consolidated information on the prac-

tices and impacts of a wide range of producers.

From this research, we have provided a unified

exposition of the environmental science for mak-

ing major changes to the food system. We hope

this stimulates progress in this crucially impor-

tant area.

REFERENCES AND NOTES

1. H. C. J. Godfray et al., Science 327, 812–818 (2010).

2. J. A. Foley et al., Nature 478, 337–342 (2011).

3. FAO, “The state of food and agriculture” (FAO, 2014).

4. FAOSTAT; www.fao.org/faostat.

5. FAO, “The second report on the state of the world’s plant

genetic resources for food and agriculture” (FAO, 2010).

6. K. M. Carlson et al., Nat. Clim. Change 7,63–68 (2016).

7. P. C. West et al., Science 345, 325–328 (2014).

8. P. J. Gerber, H. Steinfeld, B. Henderson, A. Mottet, C. Opio,

J. Dijkman, A. Falcucci, G. Tempio, “Tackling climate change

through livestock: A global assessment of emissions and

mitigation opportunities” (FAO, 2013).

9. European Commission, “Recommendation 2013/179/EU on the

use of common methods to measure and communicate the life

cycle environmental performance of products and

organisations” (European Commission, 2013).

10. S. Hellweg, L. Milà i Canals, Science 344, 1109–1113 (2014).

11. K. Paustian, Environ. Res. Lett. 8, 021001 (2013).

12. S. Clune, E. Crossin, K. Verghese, J. Clean. Prod. 140, 766–783

(2017).

13. D. Tilman, M. Clark, Nature 515, 518–522 (2014).

14. M. Clark, D. Tilman, Environ. Res. Lett. 12, 064016 (2017).

15. M. de Vries, I. J. M. de Boer, Livest. Sci. 128,1–11 (2010).

16. D. Nijdam, T. Rood, H. Westhoek, Food Policy 37, 760–770

(2012).

17.

See the supplementary materials.

18. W. Steffen et al., Science 347, 1259855 (2015).

19. A. F. Bouwman, D. P. Van Vuuren, R. G. Derwent, M. Posch,

Water Air Soil Pollut. 141, 349–382 (2002).

20. E. Röös, C. Sundberg, P. Tidåker, I. Strid, P.-A. Hansson, Ecol.

Indic. 24, 573–581 (2013).

Poore et al., Science 360, 987–992 (2018) 1 June 2018 5of6

Farms Processors & Retailers

Monitor multiple

impacts

Validate and

communicate

impacts

Meet targets by

choosing from

multiple practice

changes

Policy

Set and incentivize mitigation targets

Monitor multiple

impacts using

supply-chain data

Validate and

communicate

impacts

Define and regulate sustainability standards

Require

sustainability

standards

Meet targets

Consumers

Incentivize

sustainable

consumption

Researchers

Provide multiple mitigation options

Fig. 4. Graphical representation of the mitigation framework.

RESEARCH | RESEARCH ARTICLE

Erratum 22 February 2019. See Erratum.

on February 26, 2019 http://science.sciencemag.org/Downloaded from

21. E. Beza, J. V. Silva, L. Kooistra, P. Reidsma, . Eur. J. Agron. 82,

206–222 (2017).

22. Z. Cui et al., Biogeosciences 11, 2287–2294 (2014).

23. J. K. Ladha et al., Global Change Biol. 22,1054–1074

(2016).

24. Z. Cui et al., Nature 555, 363–366 (2018).

25. P. Smith et al., in Climate Change 2014: Mitigation of Climate

Change (Cambridge Univ. Press, 2015), pp. 811–922.

26. E. Song, B. L. Nelson, J. Staum, SIAM/ASA J. Uncertain.

Quantif. 4, 1060–1083 (2016).

27. Q. Yu et al., Agric. Syst. 153, 212–220 (2017).

28. R. N. German, C. E. Thompson, T. G. Benton, Biol. Rev.

Cambridge Philos. Soc. 92, 716–738 (2017).

29. R. Lal, Global Chang. Biol. 10.1111/gcb.14054 (2018).

30. P. Smit h et al., Agric. Ecosyst. Enviro n. 118,6–28

(2007).

31. K. B. Waldman, J. M. Kerr, Annu. Rev. Resour. Econ. 6, 429–449

(2014).

32. Euromonitor; www.euromonitor.com

33. D. C. Nepstad, W. Boyd, C. M. Stickler, T. Bezerra,

A. A. Azevedo, Philos. Trans. R. Soc. London Ser. B 368,

20120167 (2013).

34. A. Mottet et al., Global Food Sec. 14,1–8 (2017).

35. M. Springmann et al., Nat. Clim. Change 7,69–74

(2016).

36. K. Denef, K. Paustian, S. Archibeque, S. Biggar, D. Pape,

“Report of greenhouse gas accounting tools for agriculture and

forestry sectors” (Interim report to U.S. Department of

Agriculture under contract no. GS23F8182H, ICF International,

2012).

37. GSM Association (GSMA), “Creating scalable, engaging mobile

solutions for agriculture” (GSMA, 2017).

38. Organisation for Economic Co-operation and Development

(OECD), “Agriculture policy monitoring and evaluation 2017”

(OECD, 2017).

39. K. Segerson, Annu. Rev. Resour. Economics 5, 161–180

(2013).

40. European Parliament and Council, “Establishing a common

organization of the markets in agricultural products”

[Regulation (EU) 1308/2013, European Union, 2013].

ACK NOW LE DGM EN TS

We thank the many researchers who provided us with additional

data, acknowledged in data S1. We are grateful to R. Grenyer,

P. Smith, E. J. Milner-Gulland, C. Godfray, G. Gaillard, L. de Baan,

Y. Malhi, D. Thomas, K. Javanaud, and K. Afemikhe for comments

on the manuscript and Tyana for illustrations. Funding: This work

was unfunded. Author contributions: J.P. conducted the analysis

and wrote the manuscript. J.P. and T.N. contributed to the study

design and data interpretation and reviewed the manuscript.

Competing interests: The authors declare no competing interests.

Data and materials availability: A Microsoft Excel file allowing full

replication of this analysis, containing all original and recalculated

data, has been deposited in the Oxford University Research Archive

(doi.org/10.5287/bodleian:0z9MYbMyZ).

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/360/6392/987/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S14

Tables S1 to S17

References (41–151)

Data S1 and S2

5 October 2017; resubmitted 8 December 2017

Accepted 17 April 2018

10.1126/science.aaq0216

Poore et al., Science 360, 987–992 (2018) 1 June 2018 6of6

RESEARCH | RESEARCH ARTICLE

Erratum 22 February 2019. See Erratum.

on February 26, 2019 http://science.sciencemag.org/Downloaded from

Reducing food's environmental impacts through producers and consumers

J. Poore and T. Nemecek

originally published online May 31, 2018DOI: 10.1126/science.aaq0216

(6392), 987-992.360Science

, this issue p. 987Science

opportunities to target the small numbers of producers that have the most impact.

environmental cost of producing the same goods can be highly variable. However, this heterogeneity creates

Thedifferent agricultural goods around the world in a meta-analysis comparing various types of food production systems.

38,000 farms producing 40∼impacts? Poore and Nemecek consolidated data on the multiple environmental impacts of

environmental costs. Given the heterogeneity of producers, what is the best way to reduce food's environmental

Food is produced and processed by millions of farmers and intermediaries globally, with substantial associated

The global impacts of food production

ARTICLE TOOLS

http://science.sciencemag.org/content/360/6392/987

MATERIALS

SUPPLEMENTARY

http://science.sciencemag.org/content/suppl/2018/05/30/360.6392.987.DC1

CONTENT

http://science.sciencemag.org/content/sci/363/6429/eaaw9908.full

file:/content

REFERENCES

http://science.sciencemag.org/content/360/6392/987#BIBL

This article cites 84 articles, 11 of which you can access for free

PERMISSIONS

http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.Science

licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. The title

(print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

on February 26, 2019 http://science.sciencemag.org/Downloaded from