The environmental costs and benefits of high-yield farming

Abstract

How we manage farming and food systems to meet rising demand is pivotal to the future of biodiversity. Extensive field data suggest that impacts on wild populations would be greatly reduced through boosting yields on existing farmland so as to spare remaining natural habitats. High-yield farming raises other concerns because expressed per unit area it can generate high levels of externalities such as greenhouse gas emissions and nutrient losses. However, such metrics underestimate the overall impacts of lower-yield systems. Here we develop a framework that instead compares externality and land costs per unit production. We apply this framework to diverse data sets that describe the externalities of four major farm sectors and reveal that, rather than involving trade-offs, the externality and land costs of alternative production systems can covary positively: per unit production, land-efficient systems often produce lower externalities. For greenhouse gas emissions, these associations become more strongly positive once forgone sequestration is included. Our conclusions are limited: remarkably few studies report externalities alongside yields; many important externalities and farming systems are inadequately measured; and realizing the environmental benefits of high-yield systems typically requires additional measures to limit farmland expansion. Nevertheless, our results suggest that trade-offs among key cost metrics are not as ubiquitous as sometimes perceived. © 2018, The Author(s), under exclusive licence to Springer Nature Limited.

Full text

Articles
https://doi.org/10.1038/s41893-018-0138-5
1Conservation Science Group, Department of Zoology, University of Cambridge, Cambridge, UK. 2Centre for the Study of Existential Risk, University
of Cambridge, Cambridge, UK. 3Environment Centre Wales, Bangor, UK. 4Rothamsted Research, Okehampton, UK. 5Department of Animal and Plant
Sciences, University of Sheffield, Sheffield, UK. 6RSPB Centre for Conservation Science, The Royal Society for the Protection of Birds, Sandy, UK. 7School
of Biosciences, University of Nottingham, Loughborough, UK. 8Institute of Biological and Environmental Sciences, University of Aberdeen, Aberdeen, UK.
9Rothamsted Research, Harpenden, UK. 10Department of Veterinary Medicine, University of Cambridge, Cambridge, UK. 11CIPAV, Centre for Research
on Sustainable Agricultural Production Systems, Cali, Colombia. 12School of Geosciences, University of Edinburgh, Edinburgh, UK. 13Global Academy of
Agriculture and Food Security, University of Edinburgh, Edinburgh, UK. 14Oxford India Centre for Sustainable Development, Somerville College, Oxford,
UK. 15Faculty of Veterinary Medicine and Zootechny, National Autonomous University of Mexico, Mexico City, Mexico. 16Commonwealth Scientific and
Industrial Research Organisation, St Lucia, Queensland, Australia. 17Department of Geography and Environment, Pontifical Catholic University of Rio de
Janeiro (PUC-Rio), Rio de Janeiro, Brazil. 18Institute of Agricultural Engineering and Informatics, Faculty of Production and Power Engineering, University of
Agriculture in Kraków, Kraków, Poland. 19Universidade Federal da Bahia, Rua Barão de Jeremoabo, Salvador, Brazil. 20British Veterinary School, University of
Bristol, Bristol, UK. 21UN Environment World Conservation Monitoring Centre, Cambridge, UK. *e-mail: a.balmford@zoo.cam.ac.uk
Agriculture already covers around 40% of Earth’s ice- and
desert-free land and is responsible for around two-thirds
of freshwater withdrawals1. Its immense scale means that it
is already the largest source of threat to other species2, so how we
cope with very marked increases in demand for farm products3,4 will
have profound consequences for the future of global biodiversity2,5.
On the demand side, cutting food waste and excessive consumption
of animal products is essential1,5–8. In terms of supply, farming at
high yields (production per unit area) has considerable potential to
restrict humanity’s impacts on biodiversity. Detailed field data from
five continents and almost 1,800 species from birds to daisies9–14
reveal that so many depend on native vegetation that for most the
impacts of agriculture on their populations would be best limited by
farming at high yields (production per unit area) alongside sparing
large tracts of intact habitat. Provided it can be coupled with set-
ting aside (or restoring) natural habitats15, lowering the land cost of
agriculture thus appears central to addressing the extinction crisis2.
However, a key counterargument against this land-spar-
ing approach is that there are many other environmental
costs of agriculture besides the biodiversity displaced by the
land it requires, such as greenhouse gas (GHG) and ammonia
emissions, soil erosion, eutrophication, dispersal of harmful
pesticides and freshwater depletion5,7,16–18. Measured per unit
area of farmland, the production of such externalities is some-
times greater in high- than lower-yield farming systems17,18,
potentially weakening the case for land sparing. However, while
expressing externalities per unit area can help identify local-scale
impacts19, it systematically underestimates the overall impact
of lower-yield systems that occupy more land for the same
level of production20. To be robust, assessments of externali-
ties also need to include the off-site effects of management
practices, such as crop production for supplementary
feeding of livestock, or off-farm grazing for manure inputs to
organic systems20–22.
The environmental costs and benefits of high-
yield farming
AndrewBalmford1*, TatsuyaAmano 1,2, HarrietBartlett 1, DaveChadwick3, AdrianCollins4,
DavidEdwards5, RobField6, PhilipGarnsworthy 7, RhysGreen1, PeteSmith8, HelenWaters 1,
AndrewWhitmore 9, DonaldM.Broom10, JulianChara11, TomFinch1,6, EmmaGarnett 1,
AlfredGathorne-Hardy12,13,14, JuanHernandez-Medrano15, MarioHerrero 16, FangyuanHua1,
AgnieszkaLatawiec17,18, TomMisselbrook4, Ben Phalan 1,19, BennoI.Simmons 1, TaroTakahashi4,20,
JamesVause21, ErasmuszuErmgassen1 and RowanEisner1
How we manage farming and food systems to meet rising demand is pivotal to the future of biodiversity. Extensive field data
suggest that impacts on wild populations would be greatly reduced through boosting yields on existing farmland so as to spare
remaining natural habitats. High-yield farming raises other concerns because expressed per unit area it can generate high
levels of externalities such as greenhouse gas emissions and nutrient losses. However, such metrics underestimate the overall
impacts of lower-yield systems. Here we develop a framework that instead compares externality and land costs per unit produc-
tion. We apply this framework to diverse data sets that describe the externalities of four major farm sectors and reveal that,
rather than involving trade-offs, the externality and land costs of alternative production systems can covary positively: per unit
production, land-efficient systems often produce lower externalities. For greenhouse gas emissions, these associations become
more strongly positive once forgone sequestration is included. Our conclusions are limited: remarkably few studies report
externalities alongside yields; many important externalities and farming systems are inadequately measured; and realizing the
environmental benefits of high-yield systems typically requires additional measures to limit farmland expansion. Nevertheless,
our results suggest that trade-offs among key cost metrics are not as ubiquitous as sometimes perceived.
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain 477

Articles NaturE SuStaiNability
A novel framework for comparing system-wide costs
Here, we argue that comparisons of the overall impacts of contrast-
ing agricultural systems should focus on the sum of externality gen-
erated per unit of production10 (paralleling measures of emissions
intensity in climate change analyses). This approach has, for the
most part, been adopted only for a relatively narrow set of agricul-
tural products8,23 and farming systems (for example, organic versus
conventional, glasshouse versus open-field20,24). Here we develop a
more general framework, and apply it to a diversity of data on some
major farm sectors, farming systems and environmental externali-
ties. Existing data are limited but nevertheless enable us to explore
the utility of this new approach, test for broad patterns and make an
informed commentary on their significance for understanding the
trade-offs and co-benefits of high- versus lower-yield systems.
Our framework involves plotting the environmental costs of pro-
ducing a given quantity of a commodity against one another, across
alternative production systems (as in Fig. 1). We focus on examining
variation in some better-known externality costs in relation to land
cost (that is, 1/yield), because of the latter’s fundamental importance
as a proxy for impacts on biodiversity. However, the approach could
be used to explore associations among any other costs for which
data are available. Comparisons must be made across production
systems that could, in principle, be substituted for one another, so
they must be measured or modelled identically and in the same
place or, if not, potential confounding effects of different meth-
ods, climate and soils must be removed statistically. If the idea that
high-yield systems impose disproportionate externalities is true, we
would expect plots of externality per unit production against land
cost to show negative associations (Fig. 1a, blue symbols). However
observed patterns may be more complex, and could reveal promis-
ing systems associated with low land cost and low externalities, or
unpromising systems with high land and externality costs (Fig. 1b,
green and red symbols, respectively).
Our team of sector and externality specialists collated data for
applying this framework to five major externalities (GHG emis-
sions, water use, nitrogen (N), phosphorus (P) and soil losses)
in four major sectors (Asian paddy rice, European wheat, Latin
American beef and European dairy; Methods). We used both lit-
erature searches and consultation with experts to find paired yield
and externality measurements for contrasting production systems
in each sector. To be included, data had to be near-complete for
a given externality—for example, most major elements of GHG
emissions or N losses had to be included, and if systems involved
inputs (such as feeds or fertilizers) generated off-site we required
data on the externality and land costs of their production. To limit
confounding effects, we narrowed our geographic scope within
each sector (Supplementary Table 1), so that differences across sys-
tems could reasonably be attributed to farm practices rather than
gross bioclimatic variation. Where co-products were generated, we
apportioned overall costs among products using economic alloca-
tion, but also investigated alternative allocation rules.
Findings for four sectors
Our first key result is that useable data are surprisingly scarce. Few
studies measured paired externality and yield information, many
reported externalities in substantially incomplete or irreconcilably
divergent ways, and we could find no suitable data at all on some
widely adopted practices. Nevertheless, we were able to obtain suf-
ficient data to consider how externalities vary with land costs for 9
out of 20 possible sector–externality combinations (Supplementary
Table 1). The type of data available differed across these combina-
tions (which we view as a useful test of the flexibility of our frame-
work). For one combination, the most extensive data we could find
was from a long-term experiment at a single location. However
because we were interested in generalities, where possible we used
information from multiple studies—either field experiments or life
cycle assessments (LCAs) conducted across several sites—and used
generalized linear mixed models (GLMMs) to correct for confound-
ing method and site effects (Methods). Last, for two sectors, we used
process-based models parameterized for a fixed set of conditions
representative of the region.
The data that we were able to obtain do not suggest that envi-
ronmental costs are generally larger for farming systems with low
land costs (that is, high-yield systems; Fig. 2). If anything, posi-
tive associations—in which high-yield, land-efficient systems also
have lower costs in other dimensions—appear more common. For
Chinese paddy rice, we found sufficient multi-site experimental
data to explore how two focal externalities vary with land cost across
contrasting systems (Methods). GHG costs (Fig. 2a) showed nega-
tive associations with land cost across monoculture and rotational
systems (assessed separately). Our GLMMs revealed that for both
system types, greater application of organic N lowered land cost but
increased emissions (probably because of feedstock effects on the
methanogenic community25; Supplementary Table 2); by contrast,
there was little or no GHG penalty from boosting yield using inor-
ganic N (arrows, Fig. 2a). A large volume of data on rice and water
use showed weakly positive covariation in costs (Fig. 2b). GLMMs
indicated that increasing application of inorganic N boosted
Externality cost of
unit of production
Externality cost of
unit of production
Land cost of unit of production
Land cost of unit of production
YieldYield
ab
Fig. 1 | Framework for exploring how different environmental costs compare across alternative production systems. a, A hypothetical plot of externality
cost versus land cost of different, potentially interchangeable production systems (blue circles) in a given farming sector. In this example, the data suggest
a trade-off between externality and land costs across different systems. b, This example reveals a more complex pattern, with additional systems (in green
and red circles) that are low or high in both costs.
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain
478

Articles
NaturE SuStaiNability
0.05 0.10 0.15
0.6
1.0
1.4
1.8
GHG cost (tonnes CO
2
e)
More inorganic N
More organic N
More inorganic N
More organic N
Continuous
flooding to
midseason
drainage
Rice−rice
Rice−cereal
0.08 0.10 0.12 0.14 0.16 0.18
200
600
1,000
1,400
Water cost (m3)
More inorganic N
Continuous flooding
to dry soil
0.10 0.15 0.20 0.25
0.10
0.14
0.18
0.22
GHG cost (tonnes CO
2
e)
More ammonium nitrate N
More urea N
0.00.2 0.40.6 0.
81
.0
2
4
6
8
10
12
N cost (kg)
51015202530
25
35
45
55
GHG cost (tonnes CO2e)
Improved pasture
0102030405
060
0
40
80
120
GHG cost (tonnes CO2e)
Pasture
Silvopasture
Feedlot-finishing
0.05 0.10 0.15 0.20
0.9
1.1
1.3
GHG cost (tonnes CO2e)
C1
C2
C3
O1
O2
0.00 0.05 0.10 0.15 0.20
2.0
3.0
4.0
N cost (kg)
C1
C2
C3
O1
O2
0.00 0.05 0.10 0.15 0.20
0.02
0.04
0.06
0.08
P cost (kg)
C1
C2
C3
O1
O2
0.00 0.05 0.10 0.15 0.20
0
5
15
25
35
Soil cost (kg)
C1
C2
C3
O1
O2
Land cost (ha-years) Land cost (ha-years)
Land cost (ha-years) Land cost (ha-years)
Land cost (ha-years) Land cost (ha-years)
Land cost (ha-years) Land cost (ha-years)
Land cost (ha-years) Land cost (ha-years)
ab
cd
ef
g
h
ji
Fig. 2 | Externality costs of alternative production systems against land cost for five externalities in four agricultural sectors. All costs are expressed per
tonne of production (so land cost, for instance, is in ha-years per tonne; that is, the inverse of yield). a–j, Costs differentiated by sector (shown by icons).
a,b, Asian paddy rice. c,d, European wheat. e,f, Latin American beef. g–j, European dairy. Different externalities are indicated by background shading
(grey, GHG emissions; blue, water use; pink, N emissions; purple, P emissions; buff, soil loss), and different sectors (Asian paddy rice, European wheat,
Latin American beef and European dairy) are shown by icons. Points on plots derived from multi-site experiments (a–c) and LCAs (e) show values for
systems adjusted for site and study effects via GLMMs of land cost and externality cost (for 95% confidence intervals, see Supplementary Fig. 1), while
arrows show management practices with statistically significant effects (whose 95% confidence intervals do not overlap zero in the GLMMs; Methods).
In d (wheat and N emissions), progressively darker circles depict increasing nitrate application rate (0, 48, 96, 144, 192, 240 and 288 kg N per ha-year).
In f (beef and GHG emissions, estimated by RUMINANT), different colours show different system types. In g–j (dairy and four externalities), circles and
squares show results for conventional and organic systems, respectively (detailed in Supplementary Table 4). Spearman’s rank correlation coefficients
(P values) are a, rice–rice: − 0.51 (0.002); rice–cereal, − 0.36 (0.06); b, 0.19 (0.26); c, − 0.34 (0.14); d, − 0.21 (0.66); e, 0.95 (0.001); f, 0.83 (< 0.001);
g, 0.90 (0.08); h, 0.70 (0.23); i, 1.00 (0.02); and j, 1.00 (0.02). Note that these correlation coefficients do not necessarily reflect nonlinear relationships
(for example, d) accurately. Credit: Icons for Asian paddy rice and European wheat: Freepik (www.flaticon.com).
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain 479

Articles NaturE SuStaiNability
yield26, and less irrigation lowered water use while incurring only
a modest yield penalty27 (Supplementary Table 2). Sensitivity tests
of the rice analyses had little impact on these patterns (Methods;
Supplementary Fig. 2).
We found two useable data sets on European wheat, both from
the UK (Methods). Our GLMMs of data from a three-site experi-
ment varying the N fertilization regime revealed a complex relation-
ship between GHG and land costs (Fig. 2c; Supplementary Table 2),
driven by divergent responses28 to adding ammonium nitrate
(which lowers land costs but increases embodied GHG emissions)
and adding urea (which lowers land costs without increasing GHG
emissions per unit production, but at the cost of increased ammonia
volatilization). A single-site experiment varying inorganic N treat-
ments showed a nonlinear relationship between land cost and N
losses (Fig. 2d), with increasing N application lowering both costs
until an apparent threshold, beyond which land cost decreased fur-
ther but at the cost of greater N leaching (see also ref. 1).
In livestock systems, all data we could find showed posi-
tive covariation between land costs and externalities. For Latin
American beef, we located coupled yield estimates only for GHG
emissions, but here two different types of data (Methods) revealed
a common pattern. Using GLMMs again to control for potentially
confounding study and site effects, we found that across multiple
LCAs, pasture systems with greater land demands also generated
greater emissions (Fig. 2e), with both land and GHG costs reduced
by pasture improvements (using N fertilization or legumes). This
pattern across contrasting pasture systems was confirmed by run-
ning RUMINANT29 (Fig. 2f), a process-based model that also iden-
tified relatively low land and GHG costs for a series of silvopasture
and feedlot-finishing systems (for which comparable LCA data
were unavailable).
For European dairy, process-based modelling of three conven-
tional and two organic systems, parameterized for the UK, enabled
us to estimate four different externalities alongside yield (Methods).
This showed that conventional systems—especially those using less
grazing and more concentrates—had substantially lower land and
also GHG costs (Fig. 2g), in part because concentrates reduce CH4
emissions from fibre digestion30. Systems with greater use of concen-
trates (which have less rumen-degradable protein than grass31) also
showed lower losses of N, P and soil per unit production (Fig. 2h–j).
These broad patterns persisted when we used protein produc-
tion rather than economic value to allocate costs to co-products
(Methods; Supplementary Fig. 2).
Incorporating land use
As a final analysis, we examined the additional externalities result-
ing from the different land requirements of contrasting systems. To
generate the same quantity of agricultural product, low-yield systems
require more land, allowing less to be retained or restored as natural
habitat. This is in turn likely to increase GHG emissions and soil
loss, and alter hydrology—although we could find only enough data
to explore the first of these effects. For each sector, we supplemented
our direct GHG figures for each system with estimates of GHG con-
sequences of their land use following Intergovernmental Panel on
Climate Change (IPCC) methods32 to calculate the sequestration
potential of a hectare not used for farming and instead allowed to
revert to climax vegetation (Methods). Results (Fig. 3) showed that
these GHG opportunity costs of agriculture were typically greater
than the emissions from farming activities themselves and, when
added to them, in every sector generated strongly positive across-
system associations between overall GHG cost and land cost. These
patterns were maintained in sensitivity tests where we halved recov-
ery rates or assumed half of the area potentially freed from farming
was retained under agriculture (Methods; Supplementary Fig. 3).
These findings thus confirm recent suggestions33,34 that high-yield
farming has the potential, provided land not needed for production
is largely used for carbon sequestration, to make a substantial con-
tribution to mitigating climate change.
Conclusions, caveats and knowledge gaps
This study was conceived as an exploration of whether high-yield
systems—central to the idea of sparing land for nature in the face
of enormous human demand for farm products—typically impose
greater negative externalities than alternative approaches. Our
results support three conclusions. First, useful data are worryingly
limited. We considered only four relatively well-studied sectors
and a narrow set of externalities—not including important impacts
such as soil health or the effects of pesticide exposure on human
health20. Even then, we found studies reporting yield-linked esti-
mates of externalities scarce, with many widely adopted or prom-
ising practices within these sectors undocumented. We were not
able to examine complex agricultural systems (such as mixed farm-
ing or agroforestry) that might have relatively low externalities.
Relevant data on many significant developing-world farm sectors
(such as cassava or dryland cereal production in Africa) also appear
very limited. Given that a multi-dimensional understanding of the
environmental effects of alternative production systems is integral
to delivering sustainable intensification, more field measurements
linking yield with a broader suite of externalities across a much
wider range of practices and sectors are urgently needed.
Second, the available data on the sector–externality combina-
tions that we considered do not suggest that negative associations
between land cost and other environmental costs of farming are
typical (in contrast to Fig. 1a). Many low-yield systems impose high
costs in other ways too and, although certain yield-improving prac-
tices have undesirable impacts (for example, organic fertilization of
paddy rice increasing CH4 emissions; see also ref. 1), other practices
appear capable of reducing several costs simultaneously (see also
refs 1,8,24,35,36). High (but not excessive) application of inorganic N,
for example, can lower land take of Chinese rice production without
incurring GHG or water-use penalties. Similarly, in Brazilian beef
production, adopting better pasture management, semi-intensive
silvopasture and feedlot-finishing can all boost yields alongside
lowering GHG emissions. It is worth noting that although most sys-
tems we examined are relatively high-yielding, other recent work
suggests that positive associations (rather than trade-offs) among
environmental and land costs may, if anything, be more likely in
lower-yielding systems1.
Third, pursuing promising high-yield systems is clearly not
the same as encouraging business-as-usual industrial agriculture.
Some high-yield practices we did not examine, such as the heavy
use of pesticides in much tropical fruit cultivation37, are likely to
increase externality costs per unit production. Of the high-yield
practices we did investigate some, such as applying fossil-fuel-
derived ammonium nitrate to UK wheat, impose disproportion-
ately high environmental costs. Others that seem favourable in
terms of our focal externalities incur other costs, such as high NH3
emissions from using urea on wheat28, and management regimes
that reduce costs in one geographic setting may not do so in oth-
ers1. Much work characterizing existing systems and designing new
ones is thus needed. We suggest that our framework can serve as
a device for identifying existing yield-enhancing systems that also
lower other environmental costs—and perhaps more importantly,
for benchmarking the environmental performance of promising
new technologies and practices.
We close by stressing that for high-yield systems to generate
any environmental benefits they must be coupled with efforts to
reduce rebound effects. Several plausible mechanisms for limiting
these by explicitly linking yield growth to improved environmen-
tal performance have been identified—including strict land-use
zoning; strategic deployment of yield-enhancing loans, expertise
or infrastructure; conditional access to markets; and restructured
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain
480

Articles
NaturE SuStaiNability
rural subsidies15. Without such linkages, systems that perform well
per unit production may nevertheless cause net environmental
harm through higher profits or lower prices stimulating land con-
version38–40, and damage human health by encouraging overcon-
sumption of cheap, calorie-rich but nutrient-deficient foods41,42,. If
promising high-yield strategies are to help solve rather than exac-
erbate society’s challenges, yield increases instead need to be com-
bined with far-reaching demand-side interventions1,6,41 and directly
linked with effective measures to constrain agricultural expansion15.
Methods
Focal sectors and externalities. We focused on 4 globally signicant farm sectors
(Asian paddy rice, European wheat, Latin American beef and European dairy,
accounting for 90%, 33%, 23% and 53% of global output of these products43)
and 5 major externalities (GHG emissions, water use, N, P and soil losses).
We chose these sector–externality combinations because preliminary work
suggested that they were characterized quantitatively relatively oen, using
diverse approaches (single-site experiments, multi-site experiments, LCAs and
process-based models), enabling us to explore the generality of our framework.
We then searched the literature and consulted experts to obtain paired yield and
externality estimates of alternative production systems in each sector, narrowing
our geographic scope so that dierences in system performance could be
reasonably attributed to management practices (rather than gross variation in
bioclimate or soils). Our analyses have rarely been attempted previously and have
complex data requirements, so we could not adopt standard procedures developed
for systematic reviews on topics where many studies have attempted to answer the
same research question.
This process generated data on ≥ 5 contrasting production systems for 9 out
of 20 possible sector–externality combinations (Supplementary Table 1): Chinese
rice–GHG emissions (from multi-site experiments); Chinese rice–water use
(multi-site experiments); UK wheat–GHG emissions (a multi-site experiment);
UK wheat–N emissions (a single-site experiment); Brazilian beef–GHG emissions
(both LCA data and process-based models); and UK dairy–GHG emissions, and
N, P and soil losses (process-based models). Water use in the wheat and most of
the beef systems examined was limited and so not explored further. We could not
find sufficient paired yield–externality estimates for the nine remaining sector–
externality combinations.
The land and externality costs of each system were then expressed as total
area used per unit production (that is, 1/yield) and total amount of externality
generated per unit production. All estimates included the area used and
externalities generated in producing externally derived inputs (such as feed
or fertilizers). For analytical tractability, as in other recent studies1,24, we treat
impacts occurring at different times and places as being additive. Occasional
gaps in estimates for a system were filled using standard values from IPCC
or other sources, or information from study authors or comparable systems
(details below). Where experiments or LCAs were conducted at multiple sites,
we built GLMMs in the package lme444 in R version 3.3.145 to identify effects of
specific management practices on land and externality cost estimates adjusted
for potentially confounding biophysical and methodological effects. To illustrate
the effects of statistically significant management variables (those whose 95%
confidence intervals did not overlap zero; shown in bold in Supplementary
Table 2), we estimated land and externality costs at the observed minimum and
maximum values (for continuous management variables) or with the reference
category and the category that showed the maximum effect size (for categorical
variables), while keeping other variables constant; we then linked these points
as arrows on our externality cost/land cost plots (Fig. 2 and Supplementary
Figs. 1 and 2, with arrows displaced horizontally and/or vertically for increased
visibility). Where systems generated significant co-products (wheat and rapeseed
from rotational rice, beef from dairy), we allocated land and externality costs to
the focal product in proportion to its relative contribution to the gross monetary
value of production per unit area of farmland (from focal and co-product
combined)46.
0.05 0.10 0.15
2.0
2.5
3.0
3.5
4.0
ab
cd
Rice−rice
Rice−cereal
0.10 0.15 0.20 0.25
1.0
1.5
2.0
2.5
3.0
3.5
4.0
51015202530
50
100
150
200
250
300
350
0.05 0.10 0.15
0.20
1.2
1.4
1.6
1.8
2.0
C1
C2
C3
O1
O2
Land cost (ha-years) Land cost (ha-years)
Land cost (ha-years) Land cost (ha-years)
GHG cost (tonnes CO2e)
GHG cost (tonnes CO2e) GHG cost (tonnes CO2e)
GHG cost (tonnes CO
2
e)
Fig. 3 | Overall GHG cost against land cost of alternative systems in each sector, including the GHG opportunity costs of land under farming. The y-axis
values are the sum of GHG emissions from farming activities (plotted in Fig. 2a,c,e,g) and the forgone sequestration potential of land maintained under
farming and thus unable to revert to natural vegetation (Methods). All costs are expressed per tonne of production. a–d, GHG costs differentiated by
sector (shown by icons). a, Asian paddy rice. b, European wheat. c, Latin American beef. d, European dairy. The notation is as in Fig. 2. Spearman’s rank
correlation coefficients (P values) are a, rice–rice: 0.40 (0.017); rice–cereal: 0.80 (< 0.001); b, 0.99 (< 0.001); c, 0.98 (< 0.001); and d, 0.80 (0.13). Credit:
Icons for Asian paddy rice and European wheat: Freepik (www.flaticon.com).
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain 481

Articles NaturE SuStaiNability
Rice and GHG emissions. Systematic searching of Scopus for experimental studies
reporting both yields and emissions of Chinese paddy rice systems identified 17
recently published studies47–63 containing 140 paired yield–emissions estimates
for different systems (after within-year replicates of a system were averaged). To
limit confounding effects, we analysed separately the data from monoculture
systems from southern provinces (2 rice crops per year; 5 studies, 60 estimates)
and rotational systems from more northerly provinces (1 rice and 1 wheat or rape
crop per year; 12 studies, 80 estimates). The studies documented the effects of
variation in tillage (yes/no), application rates of inorganic and organic N, and (only
for rotational systems) irrigation regime (continuous flooding versus episodic
midseason drainage). There were insufficient data to examine effects of seedling
density, crop variety, organic practices, biochar application, use of groundcover to
lower emissions, N fertilizer type, or K or P fertilization.
Land cost estimates were expressed in ha-years per tonne of rice grain
(that is, the inverse of annual production per hectare farmed). GHG costs were
expressed in tonnes of CO2 equivalents (CO2e) per tonne of rice grain, and
included CH4 and N2O emissions for growing and fallow seasons (with the latter
where necessary based on mean values from refs 47–49,64), and embodied emissions
from N fertilizer production (Yara emissions database; F. Brendrup, personal
communication). We were unable to include emissions from producing manure
or K or P fertilizer, or from farm machinery. For rotational systems, we adjusted
the land and GHG costs of rice production downwards by multiplying them by the
proportional contribution of rice to the gross monetary value of production per
unit area of farmland from rice and co-product combined (using mean post-2000
prices from ref. 43).
We next built GLMMs predicting variation in our estimates of land cost and
GHG cost, for the monoculture and rotational data sets in turn. Management
practices assessed as predictors were tillage regime (binary), application rates
of organic N and of inorganic N, and irrigation regime (binary; only rotational
systems). Study site was included as a random effect. For all systems, we adjusted
for biophysical and methodological differences across sites using the first two
components from a principal component analysis of site scores for 14 variables:
annual precipitation, precipitation during the driest and wettest quarters, annual
mean temperature, mean temperatures during the warmest and coldest quarters,
maximum temperature during the warmest month, mean monthly solar radiation,
latitude, longitude, soil organic carbon content, plot size, replicates per estimate,
and start year (with all climate data taken from refs 65,66). Principal components 1
and 2 together explained 82.3% and 76.2% of the variance in these variables for
monoculture and rotational systems, respectively. Soil pH and (soil pH)2 were also
assessed as additional predictors. For the monoculture models, tolerance values
were all > 0.4 (indicating an absence of multicollinearity) except for the pH terms
(both < 0.1), which we therefore removed. For the rotational models, all tolerance
values indicated an absence of multicollinearity, but (soil pH)2 was removed
because AICc (corrected Akaike Information Criterion) values indicated that
model fit was no better than using soil pH alone. Final models (Supplementary
Table 2) were then used to plot site-adjusted land and GHG costs (as points) and
statistically significant management effects (as arrows) in Fig. 2a. We also tested the
effect of allocating land and GHG costs in rotational systems based on the relative
energy content of rice and co-products67 (rather than a relative contribution to
gross monetary value; Supplementary Fig. 2).
We adopted similar although simpler approaches for the next two sector–
externality combinations, which again used data from multi-site experiments.
Rice and water use. A systematic search on Scopus yielded 15 recent
studies57,58,64,68–79 meeting our criteria containing 123 paired estimates describing the
effects of variation in inorganic N application rate and irrigation regime on land
and water costs of Chinese paddy rice. We analysed monoculture and rotational
systems together but considered water use solely for periods of rice production.
Land cost was expressed in ha-years per tonne of rice grain, and water cost in cubic
metres per tonne of rice grain (excluding rainfall). We adjusted these estimates for
site effects in GLMMs of variation in land and water costs using as predictors the
application rate of inorganic N, and irrigation regime (a six-level factor: continuous
flooding, continuous flooding with drainage, alternate wetting and drying,
controlled irrigation, mulches or plastic films, and long periods of dry soil), while
accounting for the effect of study site as a random effect. Tolerance values were all
> 0.7. Final models (Supplementary Table 2) were then used to plot site-adjusted
land and water costs (points) and significant management effects (arrows) in Fig. 2b.
Almost all sources reported data on only one rice season per year, but one study68
included separate estimates for early- and late-season rice, so we checked the
robustness of our findings by re-running the analysis without the early-season data
from this study (Supplementary Fig. 2).
Wheat and GHG emissions. The Agricultural Greenhouse Gas Inventory
Research Platform80–83 provided 96 paired measures of variation in yield and N2O
emissions in response to experimental changes in N fertilizer application rate and
type. We expanded the emissions profile to include embodied emissions from N
fertilizer production (from the Yara emissions database; F. Brendrup, personal
communication). We derived land costs in ha-years per tonne of wheat (at 85%
dry matter) and GHG costs in tonnes of CO2e per tonne of wheat. Experiments
were run in three regions, so to adjust for site effects we built GLMMs of variation
in land and GHG costs fitting study region as a random effect and using the
application rates of ammonium nitrate, urea and dicyandiamide (a nitrification
inhibitor) as predictors. Tolerance values were all > 0.7. Adjusted land and GHG
cost estimates from the final models (Supplementary Table 2) are plotted in Fig. 2c,
with arrows showing statistically significant management practices.
Wheat and N losses. We assessed this sector–externality combination using data
from Rothamsted’s long-term Broadbalk wheat experiment, which investigates
the effects of inorganic N application rates on yields of winter wheat. During the
1990s changes in field drainage enabled the measurement (alongside yield) of plot-
specific leaching losses of nitrate84. Mean land and N costs—expressed in ha-years
per tonne of wheat (at 85% dry matter) and kilograms of N leached per tonne of
wheat, respectively—were averaged across 8 seasons (thus smoothing-out rainfall
effects), for each of 7 levels of N application (from 0 to 288 kg N (as ammonium
nitrate) per ha-year; details in Fig. 2 legend). Results are plotted in Fig. 2d.
Beef and GHG emissions. Two types of data were available for this sector–
externality combination, enabling us to compare findings across assessment
techniques. First we examined all published LCAs of Brazilian beef production85–92.
Supplementing this with a bioclimatically comparable data set from tropical
Mexico (R. Olea-Perez, personal communication) yielded 33 paired yield–
emissions estimates for contrasting production systems. These varied in whether
they used improved pasture, supplementary feeding or improved breeds (which
if unreported we inferred from age at first calving, and mortality and conception
rates). There were insufficient LCA data to examine the effects of feedlots,
silvopasture or rotational grazing. Land costs were calculated in ha-years per
tonne of carcass weight (CW), incorporating land used to grow feed, and assuming
a dressing percentage of 50%93. GHG costs were derived in tonnes of CO2e per
tonne of CW, including enteric CH4 emissions, CH4 and N2O emissions from
manure, N2O emissions from managed pasture, emissions from supplementary
feed production (where necessary using values from ref. 86), and embodied GHG
emissions from N, P and K fertilizer production. There were too few data to
include CO2 emissions from lime application or farm machinery. Milk production
was not a significant co-product. To control for site effects, we built GLMMs of
variation in land and GHG costs using site as a random effect and use of improved
pasture, supplementary feeding and improved breeds (each a binary factor) as
predictors. Tolerance values were all > 0.8. Adjusted land and GHG cost estimates
from the final models (Supplementary Table 2) are plotted in Fig. 2e, with the
arrow describing a statistically significant management practice.
For comparison we derived an equivalent GHG cost versus land cost plot
(Fig. 2f) using a process-based model of beef production. RUMINANT29 is an
IPCC tier 3 digestion and metabolism model that uses stoichiometric equations
to estimate production of meat, manure N and enteric methane for any given
pasture quality, supplementary feed quantity and type, cattle breed and region.
We used plausible combinations of these settings (Supplementary Table 3) and
corresponding values of feed and forage protein, digestibility and carbohydrate
content (judged representative of the Brazilian beef sector by M.H.) to derive yield
and emissions estimates for 86 contrasting pasture systems. To extend beyond the
scope of the LCA analyses, we also modelled 50 silvopasture systems by boosting
feed quality to simulate access to Leucaena, and 8 feedlot-finishing systems by
incorporating an 83–120 day feedlot phase when animals received high-quality
mixed ration. For each system, we included the whole herd, after determining the
ratio of fattening/breeding animals using the DYNMOD demographic projection
tool94, based on system-specific reproductive performance parameters and animal
growth rates (reflecting pasture quality and management; Supplementary Table 3).
Breeding animals experienced the same conditions as fattening animals (except
that in pasture and silvopasture they received no supplementary feed). Stocking
rates were set to sustainable carrying capacity for pasture and silvopasture, and
201 animals ha−1 for feedlots (D.M.B. personal observation). Yields were converted
to land cost in ha-years per tonne of CW, including the area of feedlots and land
required to grow feed (using feed composition and yield data from refs 43,85).
RUMINANT emissions estimates were supplemented with estimates of manure
CH4, CO2 and N2O emissions from feed production, and N2O emissions from
pasture fertilization (from refs 32,85). Carbon sequestration by vegetation could not
be included, so we probably overestimate net GHG emissions from silvopasture95.
All emissions were converted to CO2e units (using conversion factors from refs 32,85
and feedlot manure distribution from ref. 96) and expressed in tonnes of CO2e per
tonne of CW.
Dairy and four externalities. We also used process-based models to investigate
how GHG emissions and N, P and soil losses varied with land cost across five
dairy systems representative of UK practices (Supplementary Table 4; Fig. 2g–j).
We modelled three conventional systems with animals accessing grazing for 270,
180 and 0 days per year, and two organic systems with grazing access for 270 and
200 days per year. Model farms were assigned rainfall and soil characteristics
based on frequency distributions of these parameters for real farms of each type,
with structural and management data (for example, ratios of livestock categories
and ages, N and P excretion rates) based on the models of refs 31,97,98. Manure
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain
482

Articles
NaturE SuStaiNability
management was based on representative variations of the ‘manure management
continuum’99 (Supplementary Table 4). Physical performance data (annual milk
yield, concentrate feed input, replacement rate and stocking rate) were obtained
from the Agriculture and Horticulture Development Board (AHDB) Dairy
database (M. Topliff, personal communication) for conventional systems and
from the Department for Environment, Food and Rural Affairs (DEFRA)100 for
organic systems.
Yields were converted to land cost in ha-years per tonne of energy-corrected
milk (ECM), including land required to grow feed (from refs 101,102, with yield
penalties for organic production from ref. 103). As 57% of global beef production
originates from the dairy sector104, we adjusted land costs downwards by
multiplying them by the proportional contribution of milk to the gross monetary
value of production per unit area of farmland from milk and beef combined (using
prices from the AHDB Dairy database (M. Topliff, personal communication)).
GHG cost estimates for each system comprised CH4 emissions from enteric
fermentation (based on ref. 31), CH4 and N2O emissions from manure management
(following refs 32,105), emissions from N fertilizer applications to pasture (from
refs 106,107), and from feed production (from ref. 108). Emissions from farm
machinery and buildings were not included. Emissions were then summed and
expressed in tonnes of CO2e per tonne of ECM. Nitrate losses of each system were
derived from the National Environment Agricultural Pollution–Nitrate (NEAP-N)
model109,110, while P and soil losses were estimated using the Phosphorus and
Sediment Yield Characterisation In Catchments (PSYCHIC) model98,111. These last
three costs were expressed in kilograms per tonne of ECM and (as with land costs)
downscaled by allocating a portion of them to beef co-products, based on milk
and beef prices. Finally, to check the effect of this allocation rule, we re-ran each
analysis instead allocating costs using the relative protein content of milk and beef
(from ref. 104; Supplementary Fig. 2).
GHG opportunity costs of land farmed. Alongside the GHG emissions generated
by agricultural activities themselves (analysed above), farming typically carries an
additional GHG cost. Wherever the carbon content of farmed land is less than that
of the natural habitat that could replace it if agriculture ceased, farming imposes an
opportunity cost of sequestration forgone112, whose magnitude increases with the
area under production (and hence with the land cost of the system). We quantified
this GHG cost using the forgone sequestration method, whereby retaining the
current land use is assumed to prevent the sequestration in soils and biomass that
would occur if the land was allowed to revert to climax vegetation (see details in
Supplementary Table 5).
For each forgone transition, values for annual biomass accrual (≤ 20 years) were
taken from Table 4.9 of ref. 32, assuming that the climax vegetation for UK wheat
and dairy was ‘temperate oceanic forest (Europe)’, for Chinese rice it was ‘tropical
moist deciduous forest (Asia, continental)’, and for Brazilian beef it was ‘tropical
moist deciduous forest (South America)’. The carbon content of all biomass was
assumed to be 47% of dry matter (ref. 32 Table 4.3).
Changes in soil carbon values were taken from the relevant mean percentage
change in soil organic carbon values for each land conversion from a global
meta-analysis113. For UK wheat and Chinese rice, we used values for conversion
of cropland to woodland; for UK dairy and Brazilian beef, we used conversion of
grassland to woodland for grazing land and conversion of cropland to woodland
for land used to grow feed. Initial soil carbon values were taken from Table 2.3 of
ref. 32. We assumed that the soils for UK wheat were ‘cold temperate, moist, high
activity soils’, for Chinese rice, they were ‘tropical, wet, low activity soils’, for UK
dairy, they were ‘cold temperate, moist, high activity soils’ for grazing land and
for producing imported feed they were ‘subtropical humid, LAC soils’ (South
America), and for Brazilian beef for both grazing and feed production they were
‘tropical, moist, low activity soils’. In each case, the relevant percentage change in
soil organic carbon was multiplied by the initial soil carbon stock to calculate an
absolute change, which, following IPCC guidelines32, we assumed took 20 years.
Total annual forgone sequestration was then estimated by adding this annual
change in soil organic carbon and the annual accrual of biomass carbon under
reversion to climax vegetation. We assumed (as in ref. 34) that each 1 ha reduction
in land cost results in 1 ha of recovering habitat. As above, our land cost estimates
included land needed to produce externally derived inputs, and (for rotational rice
and dairy) were adjusted downwards based on the value of co-products. These
GHG opportunity costs were then added to the direct GHG emissions estimates of
each system, and the summed values were plotted against land cost (Fig. 3).
As a sensitivity test of our key assumptions, we re-ran these analyses assuming
that carbon recovery rates are halved, or that (because of rebound or similar
effects38–40) half of the area potentially freed from farming is retained under
agriculture. These two changes to our assumptions have numerically identical
effects, shown in Supplementary Fig. 3. Note that our recovery-based estimates
of the GHG costs that farming imposes through land use are conservative, in that
they are roughly 30–50% of those obtained from calculating GHG emissions from
natural habitat clearance (annualized, for consistency with the recovery method,
over 20 harvests; data not shown).
Code availability. The R codes used for the analyses are available from the
corresponding author upon request.
Data availability
The data that support the findings of this study are available from the corresponding
author upon request.
Received: 3 April 2018; Accepted: 10 August 2018;
Published online: 14 September 2018
References
1. Poore, J. & Nemecek, T. Reducing food’s environmental impacts through
producers and consumers. Science 360, 987–992 (2018).
2. Green, R. E., Cornell, S. J., Scharlemann, J. P. W. & Balmford, A. Farming
and the fate of wild nature. Science 307, 550–555 (2005).
3. Tilman, D., Balzer, C., Hill, J. & Befort, B. L. Global food demand and the
sustainable intensication of agriculture. Proc. Natl Acad. Sci. USA 108,
20260–20264 (2011).
4. Hunter, M. C., Smith, R. G., Schipanski, M. E., Atwood, L. W. &
Mortensen, D. A. Agriculture in 2050: recalibrating targets for sustainable
intensication. Bioscience 67, 386–391 (2017).
5. Godfray, H. C. J. et al. Food security: the challenge of feeding 9 billion
people. Science 327, 812–818 (2010).
6. Bajželj, B. et al. Importance of food-demand management for climate
mitigation. Nat. Clim. Change 4, 924–929 (2014).
7. Foley, J. A. et al. Solutions for a cultivated planet. Nature 478,
337–342 (2011).
8. Ripple, W. J. et al. Ruminants, climate change and climate policy. Nat. Clim.
Change 4, 2–5 (2014).
9. Phalan, B., Onial, M., Balmford, A. & Green, R. E. Reconciling food
production and biodiversity conservation: land sharing and land sparing
compared. Science 333, 1289–1291 (2011).
10. Balmford, A., Green, R. & Phalan, B. Land for food & land for nature?
Daedalus 144, 57–75 (2015).
11. Hulme, M. F. et al. Conserving the birds of Uganda’s banana-coee arc:
land sparing and land sharing compared. PLoS ONE 8, e54597 (2013).
12. Kamp, J. et al. Agricultural development and the conservation of avian
biodiversity on the Eurasian steppes: a comparison of land-sparing and
land-sharing approaches. J. Appl. Ecol. 52, 1578–1587 (2015).
13. Dotta, G., Phalan, B., Silva, T. W., Green, R. & Balmford, A. Assessing
strategies to reconcile agriculture and bird conservation in the temperate
grasslands of South America: grasslands conservation and agriculture.
Conserv. Biol. 30, 618–627 (2016).
14. Williams, D. R. et al. Land‐use strategies to balance livestock production,
biodiversity conservation and carbon storage in Yucatán, Mexico. Glob.
Change Biol. 23, 5260–5272 (2017).
15. Phalan, B. et al. How can higher-yield farming help to spare nature? Science
351, 450–451 (2016).
16. Pretty, J. Agricultural sustainability: concepts, principles and evidence. Phil.
Trans. R. Soc. B 363, 447–465 (2008).
17. Matson, P. A., Parton, W. J., Power, A. G. & Swi, M. J. Agricultural
intensication and ecosystem properties. Science 277, 504–509 (1997).
18. Tilman, D., Cassman, K. G., Matson, P. A., Naylor, R. & Polasky, S.
Agricultural sustainability and intensive production practices. Nature 418,
671–677 (2002).
19. Didham, R. K. et al. Agricultural intensication exacerbates spillover eects
on soil biogeochemistry in adjacent forest remnants. PLoS ONE 10,
e0116474 (2015).
20. Seufert, V. & Ramankutty, N. Many shades of gray – the context-dependent
performance of organic agriculture. Sci. Adv. 3, e1602638 (2017).
21. Kirchmann, H., Bergström, L., Kätterer, T., Andrén, O. & Andersson, R. in
Organic Crop Production – Ambitions and Limitations (eds Kirchmann, H.
& Bergström, L.) 39–72 (Springer, Dordrecht, 2008).
22. Madhusudan, M. D. e global village: linkages between international
coee markets and grazing by livestock in a South Indian wildlife reserve.
Conserv. Biol. 19, 411–420 (2005).
23. Nijdam, D., Rood, T. & Westhoek, H. e price of protein: review of land
use and carbon footprints from life cycle assessments of animal food
products and their substitutes. Food Policy 37, 760–770 (2012).
24. Clark, M. & Tilman, D. Comparative analysis of environmental impacts of
agricultural production systems, agricultural input eciency, and food
choice. Environ. Res. Lett. 12, 64016 (2017).
25. Yan, X., Yagi, K., Akiyama, H. & Akimoto, H. Statistical analysis of the
major variables controlling methane emission from rice elds. Glob. Change
Biol. 11, 1131–1141 (2005).
26. Pittelkow, C. M., Adviento-Borbe, M. A., van Kessel, C., Hill, J. E. &
Linquist, B. A. Optimizing rice yields while minimizing yield-scaled global
warming potential. Glob. Change Biol. 20, 1382–1393 (2014).
27. Carrijo, D. R., Lundy, M. E. & Linquist, B. A. Rice yields and water use
under alternate wetting and drying irrigation: a meta-analysis. Field Crop
Res. 203, 173–180 (2017).
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain 483

Articles NaturE SuStaiNability
28. Smith, K. A. et al. e eect of N fertilizer forms on nitrous oxide
emissions from UK arable land and grassland. Nutr. Cycl. Agroecosyst. 93,
127–149 (2012).
29. Herrero, M. et al. Biomass use, production, feed eciencies, and
greenhouse gas emissions from global livestock systems. Proc. Natl Acad.
Sci. USA 110, 20888–20893 (2013).
30. Beauchemin, K., McAllister, T. A. & McGinn, S. M. Dietary mitigation of
enteric methane from cattle. CAB Rev. Perspect. Agric. Vet. Sci. Nutr. Nat.
Resour. 4, 1–18 (2009).
31. Wilkinson, J. M. & Garnsworthy, P. C. Dietary options to reduce
the environmental impact of milk production. J. Agric. Sci. 155,
334–347 (2017).
32. IPCC 2006 IPCC Guidelines for National Greenhouse Gas Inventories (eds
Eggleston, H. S. et al.) (IGES, 2006).
33. Gilroy, J. J. et al. Optimizing carbon storage and biodiversity protection in
tropical agricultural landscapes. Glob. Change Biol. 20, 2162–2172 (2014).
34. Lamb, A. et al. e potential for land sparing to oset greenhouse gas
emissions from agriculture. Nat. Clim. Change 6, 488–492 (2016).
35. Cui, Z. et al. Pursuing sustainable productivity with millions of smallholder
farmers. Nature 555, 363–366 (2018).
36. Notarnicola, B. et al. e role of life cycle assessment in supporting
sustainable agri-food systems: a review of the challenges. J. Clean. Prod.
140, 399–409 (2017).
37. Bravo, V. et al. Monitoring pesticide use and associated health hazards in
Central America. J. Int. J. Occup. Environ. Heal. 173, 1077–3525 (2011).
38. Lambin, E. F. & Meyfroidt, P. Global land use change, economic
globalization, and the looming land scarcity. Proc. Natl Acad. Sci. USA 108,
3465–3472 (2011).
39. Ewers, R. M., Scharlemann, J. P. W., Balmford, A. & Green, R. E. Do
increases in agricultural yield spare land for nature? Glob. Change Biol. 15,
1716–1726 (2009).
40. Byerlee, D., Stevenson, J. & Villoria, N. Does intensication slow crop land
expansion or encourage deforestation? Glob. Food Sec. 3, 92–98 (2014).
41. Tilman, D. & Clark, M. Global diets link environmental sustainability and
human health. Nature 515, 518–522 (2014).
42. Yang, Q. et al. Added sugar intake and cardiovascular diseases mortality
among US adults. JAMA Intern. Med. 174, 516 (2014).
43. FAOSTAT: Food and Agriculture Data (Food and Agriculture Organization
of the United Nations, 2017); http://fao.org/faostat
44. Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting linear mixed-eects
models using lme4. J. Stat. Sow. 67, 1–48 (2015).
45. R: a Language and Environment for Statistical Computing (R Foundation for
Statistical Computing, 2016); https://www.r-project.org
46. Guinée, J. B., Heijungs, R. & Huppes, G. Economic allocation: examples and
derived decision tree. Int. J. Life Cycle Assess. 9, 23–33 (2004).
47. Shang, Q. et al. Net annual global warming potential and greenhouse gas
intensity in Chinese double rice-cropping systems: a 3-year eld
measurement in long-term fertilizer experiments. Glob. Change Biol. 17,
2196–2210 (2011).
48. Liu, Y. et al. Net global warming potential and greenhouse gas intensity
from the double rice system with integrated soil–crop system management:
a three-year eld study. Atmos. Environ. 116, 92–101 (2015).
49. Chen, Z., Chen, F., Zhang, H. & Liu, S. Eects of nitrogen application rates
on net annual global warming potential and greenhouse gas intensity in
double-rice cropping systems of the Southern China. Environ. Sci. Pollut.
Res. Int. 23, 24781–24795 (2016).
50. Xue, J. F. et al. Assessment of carbon sustainability under dierent tillage
systems in a double rice cropping system in Southern China. Int. J. Life
Cycle Assess. 19, 1581–1592 (2014).
51. Shen, J. et al. Contrasting eects of straw and straw-derived biochar
amendments on greenhouse gas emissions within double rice cropping
systems. Agric. Ecosyst. Environ. 188, 264–274 (2014).
52. Ma, Y. C. et al. Net global warming potential and greenhouse gas intensity
of annual rice–wheat rotations with integrated soil–crop system
management. Agric. Ecosyst. Environ. 164, 209–219 (2013).
53. Zhang, X., Xu, X., Liu, Y., Wang, J. & Xiong, Z. Global warming potential
and greenhouse gas intensity in rice agriculture driven by high yields and
nitrogen use eciency. Biogeosciences 13, 2701–2714 (2016).
54. Yang, B. et al. Mitigating net global warming potential and greenhouse gas
intensities by substituting chemical nitrogen fertilizers with organic
fertilization strategies in rice–wheat annual rotation systems in China: a
3-year eld experiment. Ecol. Eng. 81, 289–297 (2015).
55. Zhang, Z. S., Guo, L. J., Liu, T. Q., Li, C. F. & Cao, C. G. Eects of tillage
practices and straw returning methods on greenhouse gas emissions and
net ecosystem economic budget in rice–wheat cropping systems in central
China. Atmos. Environ. 122, 636–644 (2015).
56. Xiong, Z. et al. Dierences in net global warming potential and greenhouse
gas intensity between major rice-based cropping systems in China. Sci. Rep.
5, 17774 (2015).
57. Xu, Y. et al. Improved water management to reduce greenhouse gas
emissions in no-till rapeseed–rice rotations in Central China. Agric. Ecosyst.
Environ. 221, 87–98 (2016).
58. Xu, Y. et al. Eects of water-saving irrigation practices and drought resistant
rice variety on greenhouse gas emissions from a no-till paddy in the central
lowlands of China. Sci. Total Environ. 505, 1043–1052 (2015).
59. Yao, Z. et al. Nitrous oxide and methane uxes from a rice–wheat crop
rotation under wheat residue incorporation and no-tillage practices. Atmos.
Environ. 79, 641–649 (2013).
60. Xia, L., Wang, S. & Yan, X. Eects of long-term straw incorporation on the
net global warming potential and the net economic benet in a rice–wheat
cropping system in China. Agric. Ecosyst. Environ. 197, 118–127 (2014).
61. Zhang, A. et al. Change in net global warming potential of a rice–wheat
cropping system with biochar soil amendment in a rice paddy from China.
Agric. Ecosyst. Environ. 173, 37–45 (2013).
62. Zou, J., Huang, Y., Zong, L., Zheng, X. & Wang, Y. Carbon dioxide,
methane, and nitrous oxide emissions from a rice–wheat rotation as
aected by crop residue. Adv. Atmos. Sci. 21, 691–698 (2004).
63. Zhou, M. et al. Nitrous oxide and methane emissions from a subtropical
rice–rapeseed rotation system in China: a 3-year eld case study. Agric.
Ecosyst. Environ 212, 297–309 (2015).
64. Yao, Z. et al. Improving rice production sustainability by reducing water
demand and greenhouse gas emissions with biodegradable lms. Sci. Rep. 7,
39855 (2017).
65. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A.
WorldClim – Global Climate Data: WorldClim Version 2 (2017); http://www.
worldclim.org/version2
66. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A.
WorldClim – Global Climate Data: Bioclimatic Variables (2017); http://www.
worldclim.org/bioclim
67. Heuzé, V., Tran, G. & Hassoun, P. Feedipedia: Rough Rice (Paddy Rice)
(Feedipedia, a programme by INRA, CIRAD, AFZ and FAO, 2015); https://
www.feedipedia.org/node/226
68. Liang, K. et al. Grain yield, water productivity and CH4 emission of
irrigated rice in response to water management in south China. Agric.
Water Manag. 163, 319–331 (2016).
69. Kreye, C. et al. Fluxes of methane and nitrous oxide in water-saving rice
production in north China. Nutr. Cycl. Agroecosyst. 77, 293–304 (2007).
70. Lu, W., Cheng, W., Zhang, Z., Xin, X. & Wang, X. Dierences in rice water
consumption and yield under four irrigation schedules in central Jilin
Province, China. Paddy Water Environ. 14, 473–480 (2016).
71. Jin, X. et al. Water consumption and water-saving characteristics of a
ground cover rice production system. J. Hydrol. 540, 220–231 (2016).
72. Sun, H. et al. CH4 emission in response to water-saving and drought-
resistance rice (WDR) and common rice varieties under dierent irrigation
managements. Water Air Soil Pollut. 227, 47 (2016).
73. Wang, X. et al. e positive impacts of irrigation schedules on rice yield
and water consumption: synergies in Jilin Province, Northeast China. Int. J.
Agric. Sustain. 14, 1–12 (2016).
74. Xiong, Y., Peng, S., Luo, Y., Xu, J. & Yang, S. A paddy eco-ditch and
wetland system to reduce non-point source pollution from rice-based
production system while maintaining water use eciency. Environ. Sci.
Pollut. Res. 22, 4406–4417 (2015).
75. Shao, G.-C. et al. Eects of controlled irrigation and drainage on
growth, grain yield and water use in paddy rice. Eur. J. Agron. 53,
1–9 (2014).
76. Liu, L. et al. Combination of site-specic nitrogen management and
alternate wetting and drying irrigation increases grain yield and
nitrogen and water use eciency in super rice. Field Crop Res. 154,
226–235 (2013).
77. Chen, Y., Zhang, G., Xu, Y. J. & Huang, Z. Inuence of irrigation water
discharge frequency on soil salt removal and rice yield in a semi-arid and
saline-sodic area. Water 5, 578–592 (2013).
78. Ye, Y. et al. Alternate wetting and drying irrigation and controlled-release
nitrogen fertilizer in late-season rice. Eects on dry matter
accumulation, yield, water and nitrogen use. Field Crop Res. 144,
212–224 (2013).
79. Peng, S. et al. Integrated irrigation and drainage practices to enhance water
productivity and reduce pollution in a rice production system. Irrig. Drain.
61, 285–293 (2012).
80. Bell, M. J. et al. Nitrous oxide emissions from fertilised UK arable soils:
uxes, emission factors and mitigation. Agric Ecosyst Environ 212,
134–147 (2015).
81. Bell, M. J. et al. Agricultural Greenhouse Gas Inventory Research Platform
- InveN2Ory: Fertiliser Experimental Site in East Lothian, 2011 Version: 1
[data set] (Freshwater Biological Association, 2017); https://doi.
org/10.17865/ghgno606
82. Cardenas, L. M., Webster, C. & Donovan, N. Agricultural Greenhouse Gas
Inventory Research Platform - InveN2Ory: Fertiliser Experimental Site in
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain
484

Articles
NaturE SuStaiNability
Bedfordshire, 2011 Version: 1 [data set] (Freshwater Biological Association,
2017); https://doi.org/10.17865/ghgno613
83. Williams, J. R. et al. Agricultural Greenhouse Gas Inventory Research
Inventory Research Platform - InveN2Ory: Fertiliser Experimental Site in
Herefordshire, 2011 Version: 1 [data set] (Freshwater Biological Association,
2017); https://doi.org/10.17865/ghgno675
84. Goulding, K. W. T., Poulton, P. R., Webster, C. P. & Howe, M. T. Nitrate
leaching from the Broadbalk Wheat Experiment, Rothamsted, UK, as
inuenced by fertilizer and manure inputs and the weather. Soil Use Manag.
16, 244–250 (2000).
85. Cardoso, A. S. et al. Impact of the intensication of beef production in
Brazil on greenhouse gas emissions and land use. Agric. Syst. 143,
86–96 (2016).
86. de Figueiredo, E. B. et al. Greenhouse gas balance and carbon footprint of
beef cattle in three contrasting pasture-management systems in Brazil.
J. Clean. Prod. 142, 420–431 (2017).
87. Dick, M., Abreu Da Silva, M. & Dewes, H. Life cycle assessment of
beef cattle production in two typical grassland systems of southern Brazil.
J. Clean. Prod. 96, 426–434 (2015).
88. Florindo, T. J., de Medeiros Florindo, G. I. B., Talamini, E., da Costa, J. S. &
Ruviaro, C. F. Carbon footprint and life cycle costing of beef cattle in the
Brazilian midwest. J. Clean. Prod. 147, 119–129 (2017).
89. Mazzetto, A. M., Feigl, B. J., Schils, R. L. M., Cerri, C. E. P. & Cerri, C. C.
Improved pasture and herd management to reduce greenhouse gas
emissions from a Brazilian beef production system. Livest. Sci. 175,
101–112 (2015).
90. Pashaei Kamali, F. et al. Environmental and economic performance of beef
farming systems with dierent feeding strategies in southern Brazil. Agric.
Syst. 146, 70–79 (2016).
91. Ruviaro, C. F., De Léis, C. M., Lampert, V. D. N., Barcellos, J. O. J. &
Dewes, H. Carbon footprint in dierent beef production systems on a
southern Brazilian farm: a case study. J. Clean. Prod. 96, 435–443 (2015).
92. Ruviaro, C. F. et al. Economic and environmental feasibility of beef
production in dierent feed management systems in the Pampa biome,
southern Brazil. Ecol. Indic. 60, 930–939 (2016).
93. Dick, M., Da Silva, M. A. & Dewes, H. Mitigation of environmental impacts
of beef cattle production in southern Brazil - evaluation using farm-based
life cycle assessment. J. Clean. Prod. 87, 58–67 (2015).
94. Lesno, M. DynMod: a Tool for Demographic Projections of Tropical
Livestock Populations Under Microso Excel, User’s Manual - Version 1
(CIRAD, Montpelier, Cedex; ILRI, Nairobi, Kenya, 2008).
95. Broom, D. M., Galindo, F. A. & Murgueitio, E. Sustainable, ecient
livestock production with high biodiversity and good welfare for animals.
Proc. R. Soc. B 280, 20132025 (2013).
96. Junior, C. C. et al. Brazilian beef cattle feedlot manure management: a
country survey. J. Anim. Sci. 91, 1811–1818 (2013).
97. Garnsworthy, P. C. e environmental impact of fertility in dairy cows: a
modelling approach to predict methane and ammonia emissions. Anim.
Feed Sci. Technol. 112, 211–223 (2004).
98. Collins, A. L. & Zhang, Y. Exceedance of modern ‘background’ ne-grained
sediment delivery to rivers due to current agricultural land use and uptake
of water pollution mitigation options across England and Wales. Environ.
Sci. Policy 61, 61–73 (2016).
99. Chadwick, D. et al. Manure management: implications for greenhouse gas
emissions. Anim. Feed Sci. Technol. 166–167, 514–531 (2011).
100. Organic Dairy Cows: Milk Yield and Lactation Characteristics in irteen
Established Herds and Development of a Herd Simulation Model for Organic
Milk Production Project Report OF0170 (DEFRA, 2000); https://randd.
defra.gov.uk/Default.aspx?Menu= Menu&Module= More&Location=
None&Completed= 0&ProjectID= 8431
101. Wilkinson, J. M. Re-dening eciency of feed use by livestock. Animal 5,
1014–1022 (2011).
102. Webb, J., Audsley, E., Williams, A., Pearn, K. & Chatterton, J. Can UK
livestock production be congured to maintain production while meeting
targets to reduce emissions of greenhouse gases and ammonia? J. Clean.
Prod. 83, 204–211 (2014).
103. de Ponti, T., Rijk, B. & van Ittersum, M. K. e crop yield gap between
organic and conventional agriculture. Agric. Syst. 108, 1–9 (2012).
104. Gerber, P, Vellinga, T, Opio, C, Henderson, B. & Steinfeld, H. Greenhouse
Gas Emissions from the Dairy Sector: A Life Cycle Assessment (Food and
Agriculture Organization of the United Nations: 2010); http://www.fao.org/
docrep/012/k7930e/k7930e00.pdf
105. Brown, K. et al. UK Greenhouse Gas Inventory, 1990 to 2010: Annual Report
for Submission under the Framework Convention on Climate Change
(DEFRA, 2012); https://uk-air.defra.gov.uk/assets/documents/reports/
cat07/1204251149_ukghgi-90-10_main_chapters_issue2_print_v1.pdf
106. Misselbrook, T. H., Sutton, M. A. & Scholeeld, D. A simple process-based
model for estimating ammonia emissions from agricultural land aer
fertilizer applications. Soil Use Manag. 20, 365–372 (2006).
107. Misselbrook, T. H., Gilhespy, S. L., Cardenas, L. M., Williams, J. &
Dragosits, U. Inventory of Ammonia Emissions from UK Agriculture2015:
DEFRA Contract Report (SCF0102) (DEFRA, 2016); https://uk-air.defra.gov.
uk/library/reports?report_id= 928
108. Vellinga, T. V et al. Methodology Used in FeedPrint: a Tool Quantifying
Greenhouse Gas Emissions of Feed Production and Utilization Report 674
(Wageningen UR Livestock Research, 2013).
109. Anthony, S., Quinn, P. & Lord, E. Catchment scale modelling of nitrate
leaching. Asp. Appl. Biol. 46, 23–32 (1996).
110. Wang, L. et al. e changing trend in nitrate concentrations in major aquifers
due to historical nitrate loading from agricultural land across England and
Wales from 1925 to 2150. Sci. Total Environ. 542, 694–705 (2016).
111. Davison, P. S., Lord, E. I., Betson, M. J. & Strömqvist, J. PSYCHIC – A
process-based model of phosphorus and sediment mobilisation and delivery
within agricultural catchments. Part 1: Model description and
parameterisation. J. Hydrol. 350, 290–302 (2008).
112. Koponen, K. & Soimakallio, S. Foregone carbon sequestration due to land
occupation - the case of agro-bioenergy in Finland. Int. J. Life Cycle Assess.
20, 1544–1556 (2015).
113. Guo, L. B. & Giord, R. M. Soil carbon stocks and land use change: a meta
analysis. Glob. Change Biol. 8, 345–360 (2002).
Acknowledgements
We are grateful for funding from the Cambridge Conservation Initiative Collaborative
Fund and Arcadia, the Grantham Foundation for the Protection of the Environment,
the Kenneth Miller Trust, the UK-China Virtual Joint Centre for Agricultural Nitrogen
(CINAg, BB/N013468/1, financed by the Newton Fund via BBSRC and NERC), BBSRC
(BBS/E/C/000I0330), DEVIL (NE/M021327/1), U-GRASS (NE/M016900/1), Soils-R-
GRREAT (NE/P019455/1), N-Circle (BB/N013484/1), BBSRC Soil to Nutrition (S2N)
strategic programme (BBS/E/C/000I0330), UNAM-PAPIIT (IV200715), the Belmont
Forum/FACEE-JPI (NE/M021327/1 ‘DEVIL’) and the Cambridge Earth System Science
NERC DTP (NE/L002507/1); A.B. is supported by a Royal Society Wolfson Research
Merit award. We thank F. Brendrup, E. Caton, A. Dobermann, T. J. Florindo, E. Fonte,
O. Leyser, A. Mazzetto, J. Murthwaite, F. P. Kamali, R. Olea-Perez, S. Ramsden,
C. Ruviaro, J. Storkey, B. Strassburg, M. Topliff, J. N. V. da Silva, D. Williams, X. Yan and
Y. Zhang for advice, data or analysis, and K. Willott for much practical support.
Author contributions
A.B., T.A., H.B., D.C., D.E., R.F., P.G., R.G., P.S., H.W., A.W. and R.E. designed the study
and performed the research; D.M.B., A.C., J.C., T.F., E.G., A.G.-H., J.H.-M., M.H., F.H.,
A.L., T.M., B.P., B.I.S., T.T., J.V. and E.z.E. contributed and analysed data and results;
and all authors contributed substantially to the analysis and interpretation of results and
writing of the manuscript.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
s41893-018-0138-5.
Reprints and permissions information is available at www.nature.com/reprints.
Correspondence and requests for materials should be addressed to A.B.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
NATURE SUSTAINABILITY | VOL 1 | SEPTEMBER 2018 | 477–485 | www.nature.com/natsustain 485