Global food demand and the sustainable intensification
of agriculture
David Tilmana,1, Christian Balzerb, Jason Hillc, and Belinda L. Beforta
a
Department of Ecology, Evolution, and Behavior, University of Minnesota, St. Paul, MN 55108; bDepartment of Ecology, Evolution, and Marine Biology,
University of California, Santa Barbara, CA 93106; and cDepartment of Bioproducts and Biosystems Engineering, University of Minnesota, St. Paul, MN 55108
Contributed by David Tilman, October 12, 2011 (sent for review August 24, 2011)
Global food demand is increasing rapidly, as are the environmental impacts of agricultural expansion. Here, we project global
demand for crop production in 2050 and evaluate the environmental impacts of alternative ways that this demand might be
met. We find that per capita demand for crops, when measured
as caloric or protein content of all crops combined, has been a
similarly increasing function of per capita real income since 1960.
This relationship forecasts a 100–110% increase in global crop demand from 2005 to 2050. Quantitative assessments show that the
environmental impacts of meeting this demand depend on how
global agriculture expands. If current trends of greater agricultural
intensification in richer nations and greater land clearing (extensification) in poorer nations were to continue, ∼1 billion ha of land
would be cleared globally by 2050, with CO2-C equivalent greenhouse gas emissions reaching ∼3 Gt y−1 and N use ∼250 Mt y−1 by
then. In contrast, if 2050 crop demand was met by moderate intensification focused on existing croplands of underyielding
nations, adaptation and transfer of high-yielding technologies to
these croplands, and global technological improvements, our analyses forecast land clearing of only ∼0.2 billion ha, greenhouse gas
emissions of ∼1 Gt y−1, and global N use of ∼225 Mt y−1. Efficient
management practices could substantially lower nitrogen use. Attainment of high yields on existing croplands of underyielding
nations is of great importance if global crop demand is to be
met with minimal environmental impacts.
food security
| land-use change | biodiversity | climate change | soil fertility
G
lobal demand for agricultural crops is increasing, and may
continue to do so for decades, propelled by a 2.3 billion
person increase in global population and greater per capita incomes anticipated through midcentury (1). Both land clearing and
more intensive use of existing croplands could contribute to the
increased crop production needed to meet such demand, but the
environmental impacts and tradeoffs of these alternative paths
of agricultural expansion are unclear (1, 2). Agriculture already
has major global environmental impacts: land clearing and habitat fragmentation threaten biodiversity (3), about one-quarter of
global greenhouse gas (GHG) emissions result from land clearing,
crop production, and fertilization (4), and fertilizer can harm
marine, freshwater, and terrestrial ecosystems (5). Understanding
the future environmental impacts of global crop production and
how to achieve greater yields with lower impacts requires quantitative assessments of future crop demand and how different
production practices affect yields and environmental variables.
Here, we forecast 2050 global crop demand and then quantitatively evaluate the global impacts on land clearing, nitrogen
fertilizer use, and GHG release of alternative approaches by
which this global crop demand might be achieved. To do these
analyses, we compiled annual agricultural and population data
for 1961–2007 obtained from the FAOSTAT database (Food
and Agriculture Organization of the United Nations; http://faostat.fao.org/) and other sources (SI Materials and Methods) for
each of 100 large nations that comprised 91% of the 2006 global
population (Table S1). We then calculated net national demand
for crop calories and crop protein for each nation for each year
20260–20264 | PNAS | December 13, 2011 | vol. 108 | no. 50
based on national annual yields, production, imports, and
exports of 275 major crops (those crops used as human foods or
livestock and fish feeds) (Table S2). The resultant per capita
demand for calories or protein from all food or feed crops
combined (SI Materials and Methods) encompasses annual human crop consumption, crop use for livestock and fish production, and all losses (waste and spoilage during food and crop
production, storage, transport, and manufacturing). To determine long-term global trends and better control for economic
differences among nations, nations were aggregated into seven
economic groups ranging from highest (Group A) to lowest
(Group G) national average per capita real (inflation-adjusted)
gross domestic product (GDP) (Table S1).
Results and Discussion
Global Crop Demand. Analyses reveal a simple and temporally
consistent global relationship between per capita GDP and per
capita demand for crop calories or protein. Across all years, per
capita crop use was similarly dependent on per capita GDP both
within and among the seven economic groups (Fig. 1). The
magnitude of this dependence is surprisingly large. In 2000, for
example, per capita use of calories and protein by the richest
nations (Group A) were 256% and 430% greater, respectively,
than use by the poorest nations (Groups F and G). These large
differences in crop demand partially result from greater dietary
meat consumption at higher income (6, 7) and the low efficiency
with which some types of livestock convert crop calories and
protein into edible foods (8).
We suggest that the observed relationships between per capita
crop use and per capita real GDP (Fig. 1) provide a means of
forecasting future crop demand. Specifically, using the fitted
curves in Fig. 1, we forecasted per capita crop caloric and protein
demand for 2050 for each economic group by its estimated 2050
per capita GDP (Fig. 2 B and C) (Table S3). The GDP estimate
(SI Materials and Methods and Fig. S1) assumes that per capita
real GDP would grow at ∼2.5% per year globally, with rates for
developing nations being greater than developed nations (Fig.
2A). Using United Nations (UN) projections of 2050 population
(9) (Fig. S2), we next calculated the total 2050 demand for crop
calories or crop protein for each economic group and then
summed these values to estimate 2050 global crop demand (SI
Materials and Methods).
These analyses forecast that global demand for crop calories
would increase by 100% ± 11% and global demand for crop
protein would increase by 110% ± 7% (mean ± SE) from 2005 to
2050 (Table S3). This projected doubling is lower than the 176%
Author contributions: D.T. designed research; D.T., C.B., J.H., and B.L.B. performed research; D.T., C.B., and B.L.B. analyzed data; and D.T., J.H., and B.L.B. wrote the paper.
The authors declare no conflict of interest.
Freely available online through the PNAS open access option.
See Commentary on page 19845.
1
To whom correspondence should be addressed. E-mail: [email protected].
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
1073/pnas.1116437108/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1116437108
1961
7,000
6,000
2003
5,000
4,000
1961
3,000
1961
2,000
Economic
group
2003
p < 0.0001
F = 7,860
n = 301
1,000
A
B
C
D
E
F
G
0
0
5,000
10,000 15,000 20,000 25,000
Per Capita GDP (1990$ yr -1)
300
250
35,000
30,000
2005-2050 Growth
101%
2005 Actual
25,000
246%
20,000
338%
15,000
379%
10,000
104%
0%
F
G
0
A
B
C
10%
34%
64%
D
E
B
10,000
107%
8,000
6,000
62%
4,000
20%
0%
2,000
0
200
A
B
13%
50%
C
D
E
F
G
C
150
p < 0.0001
F = 7,860
n = 301
100
50
0
0
5,000 10,000 15,000 20,000 25,000
Per Capita GDP (1990$ yr -1)
Fig. 1. Annual dependence of per capita demand for (A) crop calories and
(B) protein on per capita real GDP for each of economic Groups A–G (SI
Materials and Methods). Each color of points shows the trajectory for a
particular economic group (one point per year for each group). Curves are
fitted to the square root of per capita GDP.
(caloric) and 238% (protein) increases in global crop use that
would occur if per capita demands of all nations in 2050 reached
the 2005 levels of Group A nations.
Any projection of future global crop production entails many
elements of uncertainty and of necessity emphasizes some potentially causative factors over others. Our forecast of a 100–
110% increase in global crop production by 2050 is larger than
the 70% increase that has been projected for this same period
(10). Although our projection methods and the methods of the
earlier study differ in many ways, the different forecasts may
occur because of our use of quantitative global trends in per
capita crop demand that emphasize income-dependent dietary
choices (Fig. 1) vs. their use of expert opinion of national and
regional demand trends (10).
Quantification of Yield, Input, and Climate Relationships. The environmental impacts of doubling global crop production will depend on how increased production is achieved (11, 12). Production
could be increased by agricultural extensification (that is, clearing additional land for crop production) or intensification (that
is, achieving higher yields through increased inputs, improved
agronomic practices, improved crop varieties, and other innovations). Here, we quantify the global impacts on land clearing,
GHG emissions, and nitrogen fertilization of alternative pathways of agricultural development that meet the 2050 global crop
production that we forecast. In particular, we evaluate the comTilman et al.
29%
5,000
Per capita protein demand
(g day-1)
Per capita protein demand
(g day -1)
B
350
A
SEE COMMENTARY
2003
400
350
300
250
200
150
100
50
0
80%
116%
89%
28%
0%
A
High
B
C
D
E
Economic Group
Per capita GDP
F
G
Low
Fig. 2. (A) Per capita GDP, (B) per capita demand for crop calories, and (C)
per capita demand for crop protein in 2005 (black) and mean projected 2050
increases (white; percent increases above bars).
binations of current or improved agricultural technologies,
enhancements to soil fertility, and land clearing that could meet
our projected 2050 global caloric demand and what their environmental impacts would be. For brevity, results for protein are
not presented here but are similar. Because of data availability,
we use past N fertilization rates as quantitative measures of soil
fertility enhancement, but we emphasize that soil fertility can
also be enhanced by legumes, cover crops, and other means and
that yields could increase with less N fertilizer than in the past if
N use efficiency increases (1, 2, 13).
We used multiple regressions to quantify how nation to nation
and year to year differences in caloric yields have been related to
N fertilization intensity (N ha−1) and other variables that are
thought to impact yields (SI Materials and Methods). We found
that caloric yields were simultaneously related to N fertilization
intensity, precipitation, potential evapotranspiration, soil pH, elevation, time (year), and economic group (Table S4). A simpler
regression that included just N fertilization intensity, precipitation,
economic group, and time gave similar results (Table S4). Two
otherwise similar regressions used just 2005 data (Table S5).
These four regressions show that ∼80% of national-level variation in caloric yields was statistically explained by a few underlying variables. We use these fitted relationships to quantify
PNAS | December 13, 2011 | vol. 108 | no. 50 | 20261
SUSTAINABILITY
SCIENCE
Per capita caloric demand
(kcal day -1)
8,000
Per capita GDP
(1990$ yr-1)
9,000
Per capita caloric demand
(kcal day-1)
A
scenarios, exploring the potential effects of changes in these
variables on caloric yields and the environment. We do so with
the caveat that the fitted relationships need not be indicative of
causation, while noting that fits are consistent with other analyses
of controls of yields (12, 14, 15).
After controlling for N fertilization intensity, climate, soil, and
elevation in these regressions, we will, for brevity, refer to the
residual yield differences ascribed to economic groups as mainly
reflecting technological and infrastructure disparities among the
economic groups, and we will refer to the residual yield differences that are ascribed to time (year) as mainly reflecting technological improvements from 1965 to 2005.
Alternative Pathways of Agricultural Expansion. These regressions
can estimate the dependence of global yields on N use (soil
fertility enhancement) if future technological advances were to
continue along observed temporal trends to 2050 (technology
improvement), if underyielding nations were to overcome technological disparities by adapting and then adopting the highyielding technologies of Group A nations (technology transfer),
or if both technology improvement and technology transfer were
to occur. In particular, we used our regression results to quantify
curves defining the dependence of global caloric yields on global
N use for four cases that all meet our projected 2050 crop caloric
demand forecast (Fig. 3A and SI Materials and Methods). For all
cases, we assumed that the currently large disparities among
nations in agricultural intensities (measured here as N ha−1)
were eliminated by 2050. We call this equalization of N use
strategic N utilization, because it provides a larger increase in
global crop production per unit of N than would occur from
greater N use in nations already applying N at high rates.
The current technology curve in Fig. 3 retains each economic
group’s N-dependent yield at its 2005 relationship and thus
assumes no technological improvements or transfer from 2005 to
2050. This curve provides a potential lower bound for 2050
yields. It is defined by six data points calculated from each of the
two regressions that used just 2005 data (Table S5). These two
regressions, which differ in the number of variables included,
give results so similar to each other as to be almost indistinguishable in Fig. 3 A–C.
A potential upper bound is provided by the technology improvement and transfer curve for which complete technology
transfer is assumed to allow all nations to achieve (by 2050) the
technological improvements and soil- and climate-adjusted yields
projected for Group A nations by 2050. The two regressions on
which it is based also gave highly similar predictions (Fig. 3 A–C
and Table S4).
Two intermediate curves, each defined by two regressions,
provide benchmarks within the region defined by the upper and
lower bounds. The technology improvement curve assumes that
yields continue to increase until 2050 along the 1965–2005 time
trajectory (Table S4) but that all nations otherwise retain the
technology of their economic group. The technology transfer
curve has each nation, based on its climate and soils, achieve (in
2050) the climate- and soil-adjusted N-dependent yield of Group
A nations in 2005 (Table S5). All four curves in Fig. 3 explore
what might occur should lower-yielding nations achieve, by 2050,
significant soil fertility enhancements (here quantified by increased N use but potentially achievable by other means).
Any point in the shaded region of Fig. 3A represents different
combinations of technology improvement and technology transfer that, for the given global N use or its equivalent soil fertility
enhancement, would meet global caloric demand in 2050. The
increased global yields that could result from various degrees of
technology improvement, technology transfer, or N use would
meet 2050 crop demand with less cropland clearing (1, 2) (Fig.
3B). For instance, if global N use were held at 200 Mt, achieving
the technology transfer and improvement benchmark by 2050
20262 | www.pnas.org/cgi/doi/10.1073/pnas.1116437108
would decrease land clearing by ∼1.2 billion ha compared with
current yields (Fig. 3B).
Land clearing, soil cultivation, and N fertilizer manufacturing
and use all emit GHG. We quantified global emissions from
these sources for each curve using Intergovernmental Panel on
Climate Change (IPCC) Tier 1 methods (16, 17) (SI Materials
and Methods and Tables S6 and S7). Although estimates of N2O
emissions that result from N fertilizer are variable (18), such
variability is small compared with the other sources of emissions
that we quantified. Our analyses found that, when increased
global N is focused on croplands of underyielding nations, projected global 2050 net GHG emissions are reduced, as shown by
the negative slopes for each of the four curves of Fig. 3C. Reduced GHG emissions occurred because increased N use decreased land clearing. The resultant reduction in GHG emissions
from lower land clearing was approximately three times the
emissions increase from the N fertilizer.
Environmental Impacts of Meeting Increased Crop Demand. These
relationships among global N use, yield, land clearing, and GHG
emissions allow exploration of the environmental impacts of different pathways of global agricultural development. Four hypothetical pathways that start on the current technology curve at the
2005 global average N use intensity of 94 kg ha−1 (Fig. 3 D–F) illustrate that our forecast of 2050 global crop demand may be met
in ways that have markedly different environmental impacts. First,
consider a pathway that mimics past trends (black arrows), with
poorer, lower-yielding nations increasing crop production mainly
through land clearing and richer, higher-yielding nations doing
so mainly by yield increases from intensification and yield improvement. The environmental impacts of this past trend trajectory would, as illustrated, increase global land clearing to a total of
∼1 billion ha by 2050, global agricultural GHG emissions to ∼3 Gt
y−1 of CO2-carbon equivalents, and global N use to ∼250 Mt y−1.
These increases would have major environmental impacts through
resultant species extinctions, loss of ecosystem services, elevated
atmospheric GHG levels, and water pollution (3–5, 19).
Greater global investments in technology improvement and
technology adaptation and transfer could markedly reduce these
impacts, as illustrated by the other three trajectories, all of which
attain the technology improvement and technology adaptation
and transfer frontier by 2050. For instance, the N-minimizing
trajectory shown (brown arrows) (Fig. 3 D–F) could retain global
N use at its current 100 Mt y−1, have land clearing of ∼0.5 billion
ha, and have GHG emissions of 1.6 Gt y−1. Alternatively, a current N-intensity trajectory (yellow arrows), with global N intensity
staying at 94 kg ha−1 until 2050, would move global values to N use
of ∼125 Mt, land clearing of ∼0.4 billion ha, and GHG emissions
of ∼1.4 Gt y−1 in 2050.
A land sparing trajectory (white arrows) would minimize both
land clearing and GHG emissions. It could meet our 2050 projected global crop demand while clearing only ∼0.2 billion ha
land globally and producing global GHG emissions of just ∼1 Gt
y−1. Global N use would be ∼225 Mt y−1. This analysis suggests
that a land sparing trajectory of agricultural development might
be the best option for minimizing biodiversity loss and GHG
emissions, but it comes with the environmental cost associated
with greater global N use.
However, a variety of practices can greatly decrease this environmental cost by increasing the efficiency of agricultural nitrogen
utilization (1, 11–13, 20, 21). For instance, recent field trials of an
integrated soil–crop management system in China achieved a 90%
increase in maize yields with no increase in N use (13). Because N
inputs in excess of plant uptake increase nitrate loading into
surface and ground waters and contribute to marine anoxic zones
(20, 22), greater development and adoption of agronomic practices that increase nutrient efficiency (23, 24) could further decrease environmental impacts of increased yields (25–27).
Tilman et al.
Land cleared for crops
(109 ha)
nly
ro
fe
ans
tr
h
Tec
pr
Im
160
ech
140
nt t
e
r
r
120
Cu
1002050 Average
80
N intensity (kg ha-1)
60
250
150
200
Global N use in 2050 (106 tonnes)
80
0.5 Im
prove
me
sfer
nt & t
0
100
C
2050 Total annual GHG emissions
from land clearing and N fertilizer
(109 tonnes CO2e)
men
tran
2
1
ech t
120
ech 140
rre
rove
Tec
h
100
Cu
Imp
1.0
3
2050 Average N
intensity (kg ha-1)
60
1.5
4
300
nt t
t on
160
ly
only
150
200
250
Global N use in 2050 (106 tonnes)
SEE COMMENTARY
0.5
Land
sparing
Current N intensity
100
F
0
300
Past
trend
1.0
300
300
Current average
N intensity
(kg ha-1)
94
N minimizing
0
2050 Average N
-1
80 intensity (kg ha )
100
Cur
ren 120 140
Imp
t te
rove
ch
160
men
t on
ly
Tech
trans
fer on
ly
Improv
ement
& tech
transfe
r
Current
94 average N
intensity (kg ha-1)
150
200
250
Global N use in 2050 (106 tonnes)
1.5
er
60
100
N minimizing
2.0
ransf
150
200
250
Global N use in 2050 (106 tonnes)
Land
sparing
Past
trend
100
E
2.0
5
only
ent
m
ove
Current N
intensity
150
200
250
Global N use in 2050 (106 tonnes)
300
Current average
N intensity
94
(kg ha-1)
5
4
N minimizing
3
Past
trend
2
Land
sparing
1
Current N
intensity
0
100
150
200
250
Global N use in 2050 (106 tonnes)
300
Fig. 3. Projections of 2050 values for (A) global yields, (B) global land clearing, and (C) global agricultural GHG emissions and (D–F) the yields and environmental impacts of four alternative hypothetical trajectories along which agriculture might develop by 2050. Tonnes CO2e in (C) and (F) represents the
equivalent tonnes of C that would have been emitted had all GHG emissions in our analyses been in the form of CO2. All abscissas are global annual N use in
2050 calculated as the sum across all economic groups of N use intensity (N ha−1) times total cropland area (ha) needed to meet projected 2050 caloric
demand. To derive the curves shown, current N use intensities of lower-yielding nations were strategically increased to equal global mean intensity, which
was set at 60, 80, 100, 120, 140, or 160 kg ha−1 to calculate the 12 points shown for each curve (SI Materials and Methods). The four curves shown in each
graph (magenta, current technology; blue, improvement only; orange, technology adaptation and transfer only; green, improvement and technology
transfer) are regression-based estimates of yields (A and D) and associated land clearing (B and E) needed to meet 2050 global caloric demand and resultant
GHG emissions (C and F). Land clearing = (cropland needed to meet 2050 crop demand) − (2005 cropland).
Conclusions
Trajectories of global agricultural development that are directed
to greater achievement of the technology improvement and technology transfer frontier would meet 2050 crop demand with
much lower environmental impacts than trajectories that were
Tilman et al.
continuations of past trends (19). This difference occurs because
strategic intensification that elevates yields of existing croplands of
underyielding nations can meet the majority of 2050 global crop
demand, and in so doing can greatly reduce land clearing and
GHG emissions.
PNAS | December 13, 2011 | vol. 108 | no. 50 | 20263
SUSTAINABILITY
SCIENCE
vem
ro
Imp
tec
19
18
17
16
15
14
13
12
11
10
9
8
2050 Global average yield
(106 kcal ha-1)
&
ent
100
B
D
r
sfe
an
h tr
Land cleared for crops
(109 ha)
2050 Global average yield
(106 kcal ha-1)
19
18
17
16
15
14
13
12
11
10
9
8
2050 Total annual GHG emissions
from land clearing and N fertilizer
(109 tonnes CO2e)
A
Current yield differences among nations are large. In 2005, for
example, caloric yields for Group A nations were 308% greater
than for Groups F and G, 138% greater than for Group E, and
37% greater than for Groups B, C, and D. Our analyses, which
incorporate the effects of climate and soils on yields, suggest that
agricultural intensification through technology adaptation and
transfer and enhancement of soil fertility in poorer nations
would greatly reduce these yield gaps (14), provide a more equitable global food supply, and greatly decrease the GHG
emissions and species extinctions that otherwise would have
resulted from land clearing (4). Our analyses explore the implications of the 100% increase from 2005 to 2050 in global crop
production that we forecast. If global crop demand were lower
(10), less land clearing and/or N use would be needed, and environmental impacts would be smaller.
Our evaluations of the environmental benefits of alternative
pathways of global agricultural development are not meant to
imply that they might be similarly attainable or feasible. Global
yields will likely be impacted by climate change (15). Yield
increases in richer nations may be more difficult than in the past if
some major crops are approaching yield ceilings (21, 28). Yield
increases in poorer nations will require significant investments in
innovative adaptation of technologies to new soil types, climates,
and pests (29, 30) as well as new infrastructure (2). However,
yields have been increased in some nations in which they were
long stagnant, such as Malawi and Zimbabwe (31, 32). In Zimbabwe, for instance, field trials on >1,200 farms showed that
technology transfer (farmer education) and intensification (N
fertilizer use at ∼13% of use by Group A nations) increased maize
1. Godfray HCJ, et al. (2010) Food security: The challenge of feeding 9 billion people.
Science 327:812–818.
2. The Government Office for Science (2011) Foresight. The Future of Food and Farming.
Final Project Report (The Government Office for Science, London).
3. Dirzo R, Raven PH (2003) Global state of biodiversity and loss. Annu Rev Environ
Resour 28:137–167.
4. Burney JA, Davis SJ, Lobell DB (2010) Greenhouse gas mitigation by agricultural intensification. Proc Natl Acad Sci USA 107:12052–12057.
5. Vitousek PM, et al. (1997) Human alteration of the global nitrogen cycle. Ecol Appl 7:
737–750.
6. Poleman TT, Thomas LT (1995) Income and dietary change: International comparisons
using purchasing-power parity conversions. Food Policy 20:149–159.
7. Keyzer MA, Merbis MD, Pavel IFPW, van Wesenbeeck CFA (2005) Diet shifts towards
meat and the effects on cereal use: Can we feed the animals in 2030? Ecol Econ 55:
187–202.
8. Smil V (2002) Nitrogen and food production: Proteins for human diets. Ambio 31:126–131.
9. United Nations (2009) World Population Prospects: The 2008 Revision. Working Paper
No. ESA/P/WP.210 (United Nations Population Division, New York).
10. Alexandratos N (2009) How to feed the world in 2050. Proceedings of a Technical
Meeting of Experts (FAO, Rome), pp 1–32.
11. Tilman D, Cassman KG, Matson PA, Naylor R, Polasky S (2002) Agricultural sustainability and intensive production practices. Nature 418:671–677.
12. Foley JA, et al. (2011) Solutions for a cultivated planet. Nature 478:337–342.
13. Chen XP, et al. (2011) Integrated soil-crop system management for food security. Proc
Natl Acad Sci USA 108:6399–6404.
14. Lobell DB, Cassman KG, Field CB (2009) Crop yield gaps: Their importance, magnitudes and causes. Annu Rev Environ Resour 34:179–204.
15. Lobell DB, Field CB (2007) Global scale climate-crop yield relationships and the impacts of recent warming. Environ Res Lett, 2:014002.
16. Crutzen PJ, Mosier AR, Smith KA, Winiwarter W (2008) N2O release from agro-biofuel
production negates global warming deduction by replacing fossil fuels. Atmos Chem
Phys 8:389–395.
17. IPCC (2007) Climate Change 2007: The Physical Science Basis. Contribution of Working
Group I to the Fourth Assessment Report of IPCC, eds Solomon S, et al. (Cambridge
University Press, New York).
20264 | www.pnas.org/cgi/doi/10.1073/pnas.1116437108
yields 40%, giving market value returns ∼400% greater than the
cost of fertilization (32).
Global food demand is growing rapidly, much of the world’s
current cropland has yields well below their potential, and the
current global trajectory of agricultural expansion has serious longterm implications for the environment. The environmental impacts
of escalating crop demand will depend on the trajectory along
which global agriculture develops. The preservation of global
biodiversity and the minimization of the GHG impacts of agriculture may well hinge on this trajectory. A trajectory that adapts
and transfers technologies to underyielding nations, enhances
their soil fertility, employs more efficient nutrient use worldwide,
and minimizes land clearing provides a promising path to more
environmentally sustainable agricultural intensification and more
equitable global food supplies.
Materials and Methods
Detailed descriptions of our data sources and analyses can be found in the SI
Materials and Methods. These include data on political, economic, agricultural, and climatic variables; analyzed nations, economic aggregates, and
global estimates; crop demand; projections of per capita GDP; yield regressions; four 2050 yield curves and land cleared curves in (Fig. 3); calculation
and projections of GHG emissions from land use conversion; and estimation
of GHG emissions from N fertilizer manufacture and application.
ACKNOWLEDGMENTS. We thank Charles Godfray, Jon Foley, Gretchen
Daily, Stephen Polasky, Matt Burgess, and Jonathan Levine for comments.
We thank the National Science Foundation for Grant DEB-0620652 and the
University of Minnesota Institute on the Environment for support.
18. IPCC (2006) 2006 IPCC Guidelines for National Greenhouse Gas Inventories, eds
Eggleston S, et al. (IGES, Hayama, Japan).
19. Tilman D, et al. (2001) Forecasting agriculturally driven global environmental change.
Science 292:281–284.
20. Vitousek PM, et al. (2009) Agriculture. Nutrient imbalances in agricultural development. Science 324:1519–1520.
21. Cassman KG, Dobermann A, Walters DT, Yang H (2003) Meeting cereal demand while
protecting natural resources and improving environmental quality. Annu Rev Environ
Resour 28:315–358.
22. Galloway JN, et al. (2008) Transformation of the nitrogen cycle: Recent trends,
questions, and potential solutions. Science 320:889–892.
23. Dobermann A, Cassman KG (2005) Cereal area and nitrogen use efficiency are drivers
of future nitrogen fertilizer consumption. Sci China C Life Sci 48:745–758.
24. Snapp SS, Blackie MJ, Gilbert RA, Bezner-Kerr R, Kanyama-Phiri GY (2010) Biodiversity can support a greener revolution in Africa. Proc Natl Acad Sci USA 107:
20840–20845.
25. Drinkwater LE, Snapp SS (2007) Nutrients in agroecosystems: Rethinking the management paradigm. Adv Agron 92:163–186.
26. Beddington J (2010) Food security: Contributions from science to a new and greener
revolution. Philos Trans R Soc Lond B Biol Sci 365:61–71.
27. European Commission (2010) Report on Implementation of Council Directive 91/676/
EEC (European Union, Brussels).
28. Fischer RA, Edmeades GO (2010) Breeding and cereal yield progress. Crop Sci 50:
S85–S98.
29. Licker R, et al. (2010) Mind the gap: How do climate and agricultural management
explain the ‘yield gap’ of croplands around the world. Glob Ecol Biogeogr 19:
769–782.
30. Neumann K, Verburg PH, Stehfest E, Müller C (2010) The yield gap of global grain
production: A spatial analysis. Agric Syst 103:316–326.
31. Denning G, et al. (2009) Input subsidies to improve smallholder maize productivity in
Malawi: Toward an African green revolution. PLoS Biol 7:e23.
32. Twomlow S, et al. (2010) Micro-dosing as a pathway to Africa’s Green Revolution:
Evidence from broad-scale on-farm trials. Nutr Cycl Agroecosyst 88:3–15.
Tilman et al.
Supporting Information
Tilman et al. 10.1073/pnas.1116437108
SI Materials and Methods
Data on Political, Economic, Agricultural, and Climatic Variables. We
compiled annual national data (Table S1) for 1961–2007 on
population from Food and Agriculture Organization’s PopSTAT
database (Food and Agriculture Organization of the United
Nations, Rome, 2009; http://faostat.fao.org) (Fig. S2) and gross
domestic product (GDP; expressed in real 1990 international
dollars) from the Total Economy Database (Groningen Growth
and Development Centre, New York, 2008; www.conferenceboard.org/data/economydatabase/) on the production, area harvested, imports, exports, and stores of up to 275 nutritious crops
(Table S2) and N fertilizer use from the FAOSTAT database
(Food and Agriculture Organization of the United Nations,
Rome, 2009; http://faostat.fao.org). Annual N fertilizer use intensity was calculated as the annual N use of a nation divided by
the total area harvested for all crops. Using commodity- and
nation-specific conversion factors from FAOSTAT, we calculated total annual caloric and protein production for each crop,
summing across all crops to find national calorie and protein
production totals and yields for all nutritious crops combined.
We obtained global 5-min latitude/longitude raster data on
the within-nation geographic distribution of croplands as well as
climate, soil, and elevation variables from the SAGE 1992
Croplands Dataset (Center for Sustainability and the Global
Environment, University of Wisconsin–Madison, 1998; www.
sage.wisc.edu/mapsdatamodels.html). The national average
values of precipitation, elevation, potential evapotranspiration
(PET), soil organic carbon, and soil pH weighted across areas
cropped in each nation in 1992 were calculated using Esri
ArcGIS 9.3.
Analyzed Nations, Economic Aggregates, and Global Estimates. We
analyzed crop demand (utilization) and yield patterns and trends
using data from the 100 more populous nations, representing 91%
of total 2006 global population, for which sufficient annual data
on nutritious crop production and other variables were available
for 1961–2007. Some nations, however, had been formed during
this period when a former nation split into two nations. Their
data reported after the split had to be combined to make them
comparable with the earlier data. This combination reduced the
number of distinct entities, which for brevity, we call nations in
the text, to 95. A few large nations, including the former USSR
and its derivatives, could not be included in our trend analyses
because of years of missing data associated with dissolution of
state or periods of civil unrest or war. Each nation was assigned
to one of seven economic groups ranging from highest (Group
A) to lowest (Group G) national average 2000–2007 per capita
real (inflation-adjusted) GDP (Table S1).
For specific analyses, some otherwise included nations were
excluded from regressions. Malaysia (Group B) was excluded
from caloric analyses, but not protein analyses, because of poor
data on exports and food vs. biodiesel use of palm nut oil, an
energy dense commodity. Data from Ireland (Group A), The
Netherlands (Group A), and New Zealand (Group B) were excluded when performing regressions against N use, because these
nations have applied large proportions of their N fertilizer to
pasture rather than nutritious crops, with annual data on amounts
so applied being unavailable.
The global estimates of the environmental affects of future crop
demand that we present in this paper are scaled to encompass the
populations of all nations, including the remaining small nations
and the few larger nations not included in trend analyses for lack
Tilman et al. www.pnas.org/cgi/content/short/1116437108
of data. For all these nations, we obtained populations, GDPs,
and cropland areas, assigning each nation the mean per capita
crop demands and yields of the economic group to which it
belonged (according to its per capita GDP).
Crop Demand. We use the term crop demand interchangeably with
FAOSTAT’s annual domestic supply, both being defined as annual production + imports − exports + decrease in stocks. Because of low use of crops for biofuels before 2007 (1), we did not
exclude such use from demand. For 1961–2003, per capita nutritious crop demand of each economic group was dependent on
per capita inflation-adjusted GDP. Two fits were obtained for
each of the two nutritional units (tonnes protein and calories),
with per capita GDP square root transformed to capture the
nonlinear response observed. One was a universal fit, where data
from all nations were fit simultaneously to the square root of per
capita GDP (Fig. 1 A and B). The other was a specific fit, where
data from each economic group were fit separately to the square
root of per capita GDP by adding an interaction between economic
group and the square root of per capita GDP to the statistical
model. Square root was chosen over logarithmic transformation
because of its superior fit based on R2 and Akaike Information
Criterion (AIC) values.
We then used the four resulting fitted formulas to estimate
2050 per capita nutritious crop demand for each of the economic
groups based on their predicted 2050 per capita GDP (see below).
Group per capita 2050 demand is reported as the average of the
two forecasts for each nutritional unit (Table S3). Global crop
demand in 2050 was calculated by multiplying each group’s 2050
per capita demand by the total population (included + nonincluded nations) forecasted for 2050 by the United Nations
(UN) Population Division’s median variant projection (UN
Department of Economic and Social Affairs, New York, 2011;
http://esa.un.org/unpd/wpp/unpp/panel_population.htm) (Fig.
S2) and then summing demand across all seven economic groups
(Table S3).
Projections of per Capita GDP. Trajectories of the rate of change of
per capita GDP are a Kuznets (2) function of per capita GDP
(2–4). This empirical functional relationship is used to forecast
per capita GDP (4), especially when information on future
values of other terms in the Solow (5) income model are unavailable. To project the per capita GDP that each economic
group might achieve by 2050, we statistically fit by least squares
the dependence of the observed annual rate of change of per
capita GDP for each economic group on its per capita GDP,
obtaining the Kuznets (2) curve of Fig. S1. Then, using the
method of Seldon and Song (4), we numerically solved the resultant ordinary differential equation to forecast per capita GDP
to 2050 for each economic group. We did not make any forecasts
for individual nations because these forecasts are not needed
for our projections and might be more idiosyncratic than projections for groups of many nations. The model solutions (Fig.
S1) gave results similar to other estimates of 2050 per capita
GDP (6, 7).
Yield Regressions. We used past yield relationships and trends
to estimate the yields that might be achievable by 2050. All
analyses used 5-y mean data on a nation by nation basis (for
example, the mean for 1965 is the average for 1963–1967). In
particular, we used four regressions paired in two. One pair,
the time series regressions, used nine points in time comprised
1 of 8
of 1965, 1970, 1975, . . . , and 2005 averages and included year
as one of its variables (Table S4). The other pair, the 2005
regressions, only used average 2005 (average for 2003–2007)
data (Table S5).
Within each of these pairs, a four-variable regression determined the dependence of yield on economic group, the 1.333
root of national N use intensity per hectare (that is, N0.75),
precipitation, and year (if a time series regression). The other
kind of regression, a seven-variable regression, determined the
dependence of yield on those variables plus soil pH, elevation,
and potential evapotranspiration. Only N0.75 was used in all regressions, because the relationship between yield and N use intensity for the 95 analyzed nations was increasing but nonlinear
and N0.75 provided a better fit based on R2 and AIC than N
(linear), log[N], N0.5 (square root), or N0.667 (the 1.5 root of N
use intensity) for all regressions.
For the two time series regressions, each of the variables in both
the four-variable regression and the seven-variable regression was
significant at P < 0.01, and the overall regressions were significant at P < 0.0001. For the two 2005 regressions, all variables
were significant at P < 0.01 in the four-variable regression and
remained so in the seven-variable regression, but in the sevenvariable regression, none of the three added variables was significant (P > 0.05 for each).
Four 2050 Yield Curves and Land-Cleared Curves in Fig. 3. Each of the
two 2005 regression fits provides estimates on which the current
technology yield curve in Fig. 3A is based. In particular, each
2005 regression was used to calculate the caloric yield of each
nation (adjusted for all other regression variables) for each of six
different N use intensities (60, 80, 100, 120, 140, and 160 kg ha−1
N fertilizer). We then calculated the yield for each economic
group at each N use intensity by weighting national yields by
areas harvested in each nation. Global yields for each N use
intensity were similarly calculated as area-weighted means of
yields of all seven economic groups; 6 of the 12 2050 global yield
points to which the current technology yield curve in Fig. 3A is
fit come from one of the 2005 regression variations (one point
per N use intensity), and the other 6 come from the other variations. All 12 points are shown in Fig. 3A. The forecasts based
on the seven-variable regression and those forecasts based on
the four-variable regression are indistinguishable. Although
none of the three added variables in the seven-variable regression was significant, those six points predicted by this regression are shown precisely to illustrate how insensitive these
forecasts were to inclusion of added soil and climate variables.
The 2050 total global N use values associated with these 12
global yields were calculated from 2050 economic group yields
and projected 2050 economic group caloric demands as follows
(Eq. S1):
2050 Area Harvested ¼ ð2050 Projected DemandÞ=ð2050 YieldÞ
[S1]
and (Eq. S2)
2050 N Use ¼ ð2050 N Use IntensityÞ × ð2050 Area HarvestedÞ:
[S2]
Global N use is the sum of the 2050 N use across all seven
economic groups.
The 2005 regressions were also used to estimate the 2050
yields of the technological transfer curve. The calculations were
identical to those calculations just described except that, in using
the regressions to project 2050 values, all nations were assigned
to be members of economic Group A. The yields so calculated
thus represent the yields that could be achieved if each nation by
2050 was able to achieve a yield comparable with the yields of
Tilman et al. www.pnas.org/cgi/content/short/1116437108
Group A nations in 2005 but adjusted for its own climate and/or
soil and elevation.
The two time series regressions were similarly used to obtain
the technology improvement curve by using the observed time
dependence of yield on year in the multiple regressions to estimate the yields of all nations in 2050 under the assumption that
yields would continue to increase along the fitted trend line until
2050. As before, yields of each nation were adjusted for its own
climate and/or soil and elevation. Group yields and then global
yields were then calculated as above.
The technology improvement and transfer curve was also
similarly obtained from the two time series regressions under the
assumption that all nations would achieve by 2050 the yields that
the temporal trends projected for Group A nations in 2050. As
before, yields of each nation were adjusted for their own climate
and/or soil and elevation. Group yields and then global yields
were then calculated as above.
The four curves of cropland cleared in Fig. 3B were calculated
from the group yields calculated above and the groups’ 2050
projected caloric demands. Each group’s cropland cleared for
a given model at a given N use intensity was calculated using the
following equation (Eq. S3):
2050 Cropland Required ¼ ð2050 Area HarvestedÞ
× ð2005 CroplandÞ=ð2005 Area HarvestedÞ
[S3]
and (Eq. S4)
Cropland Cleared ¼ ð2050 Cropland RequiredÞ
− ð2005 Actual CroplandÞ:
[S4]
Global cropland cleared is the sum of cropland cleared across all
seven economic groups.
The greenhouse gas (GHG) emission curves in Fig. 3C were
calculated, as described below, using forecasts of global cropland
cleared and associated global N use values. GHG emissions were
calculated as the sum of GHG emissions from land conversions
to cropland and from N fertilizer manufacture and application in
accordance with Intergovernmental Panel on Climate Change
(IPCC) Tier 1 methodology (8).
Calculation and Projections of GHG Emissions from Land Use
Conversion. IPCC Tier 1 methodology (8) was used to quantify
the following three pathways of global GHG emissions after land
conversion to cropland.
i) Immediate C release from biomass in year of clearing
(c1 = 84 tC ha−1) (Table S6).
ii) Annual soil C loss (c20 = 1.195 tC ha−1) (Table S7) during
the first 20 y after clearing.
iii) Annual N2O release from soil N mineralization (n20 =
0.102 tC eq ha−1) associated with the annual soil C loss
during the first 20 y after clearing [using a 100-y global
warming potential (GWP) of 298 for N2O] (8).
Annual land clearing rates were calculated on a group by group
basis, assuming a constant clearing rate between 2006 (mean =
2005–2007) and 2050.
Total annual GHG release from land conversion to cropland
was calculated using the equation (cropland cleared past year)
(c1) + (cropland cleared past 20 y) (c20 + n20).
For annual soil C loss during the first 20 y after clearing, we
made the conservative assumption that all cleared croplands
were on mineral as opposed to organic soils. For each of the five
biomes, the appropriate average IPCC default organic soil C
stocks (IPCC table 2.3) (8) together with the appropriate average IPCC relative stock change factors for cropland use
(Table S7) (IPCC table 5.5) (8) were used to calculate annual
2 of 8
soil C loss for 20 y after clearing with the following formula
(IPCC equation 2.25) (8) (Eq. S5):
annual soil C loss
ðsoil org:C stockÞð1 − stock change factor over 20 yÞ
:
C20 ¼
20 y
[S5]
The average value for all biomes was weighted, as c1 above, by the
relative proportion of anticipated land use change, yielding the
value of c20 = 1.195 tC ha−1.
The annual N2O release from soil N mineralization during the
first 20 y after clearing (n20 = 0.102 tC ha−1) was calculated from
c20 using IPCC equation 11.8 (8), assuming the IPCC default C:
N ratio of 15:1 and default N2O emission factor of 0.01 t(N2O-N)
t(N)−1 and converting to tC equivalents using a GWP of 298
(IPCC equations 11.1 and 11.8) (8) (Eq. S6):
tC
1t N
0:01t N2 O − N
44t N2 O
×
×
×
n20 ¼ c20
ha
15t C
tN
28t N2 O − N
298t CO2
12t CO2 − C
tC
×
¼ 0:102
×
:
t N2 O
44t CO2
ha
[S6]
sions from N fertilizer manufacture and application were estimated using IPCC Tier 1 methodology (8):
i) CO2 emissions from ammonia production (IPCC table 3.1)
(8): good practice default value of 3.273 (tCO2) (t NH3)−1
= 1.08 (tC equiv.) (tNH3-N)−1.
ii) Direct N2O emissions from N addition to managed soils:
default value (IPCC table 11.1) (8) of 0.01 (tN2O-N emission) (tN input)−1 = 1.28 (tC equiv.) (tN input)−1.
iii) Indirect N2O emissions from N volatilization (IPCC equation 11.9) (8): 0.001 (tN2O-N emission) (tN input)−1 =
0.13 (tC equiv.) (tN input)−1.
The sum of GHG emissions from these three pathways equals
2.49 (tC equiv.) (tN fertilizer)−1. Multiplying this factor by annual
global N fertilizer consumption yields annual global GHG
emission values from N fertilizer manufacture and application.
We acknowledge the uncertainties with ii and iii above (9). In
our analyses, increased N use in underyielding nations gives a net
GHG benefit compared with the same crop production on newly
cleared land, even if N2O emissions were up to about four times
the values used by the IPCC.
7. Alexandratos N, et al., eds (2006) World Agriculture: Towards 2030/2050—Interim
Report (United Nations Food and Agriculture Organization, Rome).
8. IPCC (2006) IPCC Guidelines for National Greenhouse Gas Inventories, eds Eggleston S
et al. (IGES, Hayama, Japan).
9. Crutzen PJ, Mosier AR, Smith KA, Winiwarter W (2007) N2O release from agro-biofuel production negates global warming deduction by replacing fossil fuels. Atmos Chem Phys 7:
11191e111205.
2
y = -0.6284 + 0.1570*ln(x) - 0.0093*ln(x)
0.10
0.05
0.00
Economic group
Specific rate of change of per capita GDP
(yr -1)
1. Fargione JE, Plevin RJ, Hill JD (2010) The ecological impact of biofuels. Annu Rev Ecol
Evol Syst 41:351e377.
2. Kuznets S (1955) Economic growth and income inequality. Am Econ Rev 45:1e28.
3. Baumol WJ (1986) Productivity growth, convergence, and welfare: What the long-run
data show. Am Econ Rev 76:1072e1085.
4. Seldon TM, Song D (1994) Environmental quality and development: Is there a Kuznets
curve for air pollution? J Environ Econ Manage 27:147e162.
5. Solow RM (1956) A contribution to the theory of economic growth. Q J Econ 70:65e94.
6. Hawksworth J, Cookson G (2006) The World in 2050: How big will the major emerging
market economies get and how can the OECD compete? (PriceWaterhouseCoopers,
London).
Estimation of GHG Emissions from N Fertilizer Manufacture and
Application. The following three pathways of global GHG emis-
-0.05
-0.10
0
5,000
10,000
15,000
20,000
A
B
C
D
E
F
G
25,000
Per capita GDP (1990$ International)
Fig. S1. Fits of the annual specific rates of change of per capita gross domestic product (GDP) against per capita real GDP from 1961 to 2006 for economic
Groups A–G give a Kuznets (1) curve. Note the long-term decline in growth rates as economies become wealthier. In the fitted equation shown, y is dG/dt × 1/G,
where G is the per capita GDP, and x is per capita GDP or G. The differential equation is dG=dt ¼ Gð− 0:6284 þ 0:157 × ln½G − 0:0093 × ln½G2 Þ. Using the work by
Seldon and Song (2), we solved this differential equation to forecast 2050 values of per capita GDP for each economic group based on 2005 initial conditions.
Our numerical solutions used the NDSolve[ ] function of Mathematica.
1. Kuznets S (1955) Economic growth and income inequality. Am Econ Rev 45:1e28.
2. Seldon TM, Song D (1994) Environmental quality and development: Is there a Kuznets curve for air pollution? J Environ Econ Manage 27:147e162.
Tilman et al. www.pnas.org/cgi/content/short/1116437108
3 of 8
Population (Billions)
2005-2050 Growth
3.5
2005 Actual
3.0
57%
12%
2.5
2.0
1.5
1.0
143%
40%
15%
22%
0.5
177%
0.0
A
B
C
D
E
F
G
Fig. S2. Population in 2005 (black bars) and projected increase to 2050 (white bars; percent increase above) for economic Groups A–G (all nations) from the
UN Population Divisions 2009 Medium Variant forecasts (UN Department of Economic and Social Affairs, New York, 2011; http://esa.un.org/unpd/wpp/unpp/
panel_population.htm).
Table S1.
Analyzed nations in the seven economic groups (A–G)
Economic group
A
B
C
D
E
F
G
A–G
100 analyzed nations used as basis for projections
Australia, Austria, Canada, Denmark, Finland, France, Germany, Ireland, Japan, The Netherlands,
Norway, Sweden, Switzerland, United Kingdom, and United States
Argentina, Chile, Greece, Israel, Italy, Malaysia, Mauritius, New Zealand, Portugal, Saudi Arabia,
South Korea, Spain, Trinidad and Tobago, Uruguay, and Venezuela
Botswana, Brazil, China, Colombia, Costa Rica, Ecuador, Guatemala, Iran, Jordan, Mexico, South Africa,
Syria, Thailand, Tunisia, and Turkey
Algeria, Bolivia, Cuba, Dominican Republic, Egypt, El Salvador, Indonesia, Jamaica, Lebanon, Morocco,
Paraguay, Peru, Philippines, Sri Lanka, and Swaziland
Bangladesh and Pakistan, Benin, Cameroon, Cote d’Ivoire, Ghana, Honduras, India, Libya, Mozambique,
Myanmar (Burma), Nicaragua, Nigeria, North Korea, Senegal, and Vietnam
Burkina Faso, Eritrea and Ethiopia, Gambia, Guinea, Haiti, Kenya, Madagascar, Malawi, Zambia
and Zimbabwe, Mali, Nepal, Rwanda and Burundi, Sudan, Tanzania, Togo, and Uganda
Central African Republic, Chad, Democratic Republic of the Congo (former Zaire), Niger,
and Sierra Leone
Total of all Groups A–G
Analyzed percent of world
population 2006 (2050)
11.4 (9.4)
4.9 (4.3)
30.2 (24.9)
8.3 (8.6)
28.7 (31.8)
5.9 (9.7)
1.4 (2.8)
90.8 (91.5)
Nations were assigned to economic Groups A–G based on their ranking by per capita GDP (average for 2000–2007); Group A had the highest and Group G
had the lowest per capita GDP. Of the 100 current nations listed, eight nations have had their data reported by the FAO in composite with another of those
eight nations in the past, resulting in the 95 distinct nation/composite data units we used in our analyses. The average population of these 95 nation/
composites was 67 million in 2006. The 128 other nations had a much smaller average population of 4.9 million, and data sets that generally had too many
missing points to be usable.
Tilman et al. www.pnas.org/cgi/content/short/1116437108
4 of 8
Table S2. Crop categories that encompass the 275 nutritious crops included in our analyses and the percent that each
crop contributed to total 2005 kcal production within the 100 analyzed nations
Crops (in aggregate)
Maize (3)
Rice from paddies
Wheat (3)
Soybeans
Oil palm fruit
Sugar cane
Barley
Rapeseed
Cassava (4)
Sorghum (3)
Cottonseed
Potatoes
Groundnuts
Sweet potatoes
Sugar beet
Millet (7)
Coconuts
Beans (8)
Oats
Other fresh vegetables (21)
Yams
Sunflower seed
Bananas
Grapes
Dry peas (2)
Chick peas
Plantains
Apples (2)
Tomatoes
Onions
Sesame seed
Triticale
Oranges
Rye
Cow peas (2)
Broad beans (3)
Linseed
Other cereals (4)
Lentils (2)
Mangoes and guavas (3)
Olives
%
Crops (in aggregate)
%
Crops (in aggregate)
%
23.56
17.20
15.98
7.43
5.09
3.78
3.23
2.23
2.05
1.99
1.60
1.51
1.38
1.23
1.20
1.01
0.68
0.58
0.56
0.55
0.48
0.44
0.41
0.36
0.30
0.29
0.29
0.22
0.21
0.21
0.18
0.18
0.17
0.16
0.15
0.15
0.13
0.13
0.13
0.12
0.12
Watermelons
Pigeon peas
Other pulses (6)
Cabbages (4)
Taro
Other fresh fruit (15)
Other roots/tubers (10)
Dates
Green chillies and peppers (3)
Eggplants
Dry chillies/peppers (3)
Tangerines (2)
Pears
Other tropical fruit (15)
Peaches and nectarines (3)
Carrots and turnips (2)
Cashew nuts
Pineapples
Lupins
Other melons
Almonds (2)
Green maize
Karite nuts
Cucumbers
Squash and gourds (3)
Buckwheat
Plums and sloes (2)
Mixed grain
Green peas
Cauliflowers and broccoli (2)
Lettuce and chicory (4)
Persimmons
Green beans
Melon seed
Walnuts
Avocados
Spinach
Lemons and limes (3)
Other oilseeds (12)
Safflower seed
Other sugar crops (3)
0.11
0.11
0.10
0.10
0.09
0.09
0.09
0.08
0.07
0.07
0.07
0.07
0.07
0.06
0.06
0.06
0.06
0.05
0.04
0.04
0.04
0.04
0.04
0.04
0.04
0.03
0.03
0.03
0.03
0.03
0.03
0.03
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
Pistachios
Hazelnuts
Papayas
Other citrus fruits (4)
Okra (2)
Vetches
Kapok seed
Fonio (2)
Apricots
Canary seed
Strawberries
Asparagus
Cashew apple
Green onions (3)
Other nuts (6)
Grapefruit (3)
Cherries (3)
Figs
Kiwi fruit
String beans
Mushrooms (3)
Yautia
Chestnuts
Artichokes
Bambara beans
Hempseed
Other legumes
Other berries (5)
Quinoa
Cranberries
Poppy seed
Sour cherries
Carobs
Blueberries
Quinces
Other stone fruit
Raspberries
Currants (2)
Gooseberries
Brazil nuts
0.02
0.02
0.02
0.01
0.01
0.01
0.01
0.01
0.01
0.01
0.009
0.008
0.008
0.008
0.007
0.007
0.006
0.006
0.005
0.004
0.004
0.004
0.003
0.003
0.003
0.003
0.002
0.002
0.002
0.002
0.002
0.001
0.001
0.001
0.001
0.001
0.001
0.000
0.000
0.000
(Number) of distinct crops included in a category if the number is more than one.
Tilman et al. www.pnas.org/cgi/content/short/1116437108
5 of 8
Table S3. Caloric and protein demand for 2005 and projections for 2050 for the group of 100 analyzed nations and
extrapolations to all global nations
Per capita demand for analyzed
nations*
Calories
Universal fit
Specific fit
Mean
Protein
Universal fit
Specific fit
Mean
Total demand for analyzed
nations†
Total global demand
(all nations)†
2005
2050
Change (%)
2005
2050
Change (%)
2005
2050
Change (%)
1.74
1.74
1.74
2.45
2.50
2.48
41
44
42
10,200
10,200
10,200
20,400
20,800
20,600
99
103
101
11,300
11,400
11,300
22,400
22,900
22,700
98
102
100
0.054
0.054
0.054
0.086
0.077
0.082
58
41
50
323
323
323
722
645
683
123
100
112
355
355
355
790
706
748
122
98
110
The universal fit is a regression, using one data point per nation, of demand against the square root of per capita GDP. In contrast,
the specific fit also groups nations by economic group and separately determines the dependence of demand on the square root of per
capita GDP by adding an interaction term between economic group and the square root of per capita GDP to the statistical model. The
mean is the average of the universal and specific fits. The mean total global demands for calories and protein are the estimates of 2050
demand for the 275 nutritious crops that we use in our 2050 projections.
*Units of measure are million kilocalories per year for calories and tonnes protein per year for protein.
†
Units of measure are trillion kilocalories per year for calories and million tonnes protein per year for protein.
Table S4. Fit statistics for the two 1965–2005 regressions of yield against N intensity,
precipitation, year, and economic group (four-variable analysis) or against these variables
plus potential evapotranspiration, elevation, and soil pH (seven-variable analysis)
1965–2005 Four-variable analysis
Parameters
Overall
Intercept
N intensity (N0.75)
Precipitation
Year
Economic group
Fit
F9, 778 = 338
P < 0.0001
R2 = 0.796
F1, 778 = 77.4
P < 0.0001
F1, 778 = 547
P < 0.0001
F1, 778 = 213
P < 0.0001
F1, 778 = 81.7
P < 0.0001
F6, 778 = 39.6
P < 0.0001
PET
Elevation
Soil pH
Tilman et al. www.pnas.org/cgi/content/short/1116437108
6
Estimate (10 kcal/ha)
−103.3
25.7
0.68
0.05
Group A: 3.52
Group B: 0.16
Group C: −0.44
Group D: −0.12
Group E: −1.05
Group F: −0.98
1965–2005 Seven-variable analysis
Fit
F12, 755 = 259
P < 0.0001
R2 = 0.804
F1, 755 = 76.7
P < 0.0001
F1, 755 = 552
P < 0.0001
F1, 755 = 55.4
P < 0.0001
F1, 755 = 84.8
P < 0.0001
F6, 755 = 30.9
P < 0.0001
F1, 755 = 4.99
P = 0.0258
F1, 755 = 7.58
P = 0.0060
F1, 755 = 10.1
P = 0.0015
Estimate (106 kcal/ha)
−103.6
26.6
0.49
0.05
Group A: 3.72
Group B: 0.26
Group C: −0.27
Group D: −0.05
Group E: −1.05
Group F: −1.23
0.01
0.001
−0.50
6 of 8
Table S5. Fit statistics for the two 2005 regressions of yield against N intensity, precipitation,
year, and economic group (four-variable analysis) or against these variables plus potential
evapotranspiration, elevation, and soil pH (seven-variable analysis)
2005 Four-variable analysis
Parameters
Fit
Overall
Intercept
N intensity (N0.75)
Precipitation
Economic group
2005 Seven-variable analysis
Estimate (106 kcal/ha)
F8, 71 = 30.3
P < 0.0001
R2 = 0.773
F1, 71 = 15.0
P < 0.0001
F1, 71 = 48.5
P < 0.0001
F1, 71 = 26.6
P < 0.0001
F6, 71 = 4.14
P = 0.0013
3.06
30.1
1.03
Group A: 4.32
Group B: −0.24
Group C: 0.19
Group D: −0.12
Group E: −1.21
Group F: −1.47
PET
Estimate (106 kcal/ha)
Fit
F11, 66 = 22.2
P < 0.0001
R2 = 0.787
F1, 66 = 1.85
P = 0.1787
F1, 66 = 47.2
P < 0.0001
F1, 66 = 8.70
P = 0.0044
F6, 66 = 2.49
P = 0.0314
F1, 66 = 0.237
P = 0.6283
F1, 66 = 1.11
P = 0.2952
F1, 66 = 0.563
P = 0.4557
Elevation
Soil pH
6.76
30.7
0.87
Group A: 4.47
Group B: 0.39
Group C: 0.33
Group D: 0.14
Group E: −1.37
Group F: −1.93
0.01
0.001
−0.73
Table S6. Calculations of carbon emissions per hectare from land cleared for crops in the first year after clearing based on IPCC Tier 1
methods (1)
Biome
Cropland use
increase
weight*
Grassland
2
Savanna
3
Tropical forests
4
Mediterranean
1
Southern
temperate
forest
1
Biomass
estimation
Mean of four temperate
categories in IPCC table 6.4;
CF = 0.47 (above- +
below-ground biomass)
Subtropical steppe in
IPCC table 4.12
Mean of three tropical
forest categories in
IPCC table 4.12; CF = 0.5
Mean subtropical
dry forest and
subtropical steppe
in IPCC table 4.12; CF = 0.5
Temperate oceanic
forest in IPCC table 4.12,
(see also IPCC figure 4.1);
CF = 0.5
Ratio of total to
Live
Above-ground above-ground
biomass
biomass
biomass
loss, year 1
(tC ha−1)
(IPCC table 4.4) (tC ha−1)
4.66
N/A
4.66
35.00
1.32
46.20
101.67
1.33
135.22
50.00
1.37
68.50
90.00
1.30
117.36
Weighted
average
Dead organic
matter (IPCC
table 2.2)
N/A
Dead organic Live + dead
matter loss,
biomass
year 1
loss, year 1
(tC ha−1)
(tC ha−1)
0.00
4.66
Mean of
subtropical
value
Mean tropical
value
3.45
49.65
3.65
138.87
Mean
subtropical
value
3.45
71.95
20.88
138.24
Mean warm
temperate (dry
and moist) value
84.0
*4, most use; 1, least use.
1. IPCC (2006) IPCC Guidelines for National Greenhouse Gas Inventories, eds Eggleston S et al. (IGES, Hayama, Japan).
Tilman et al. www.pnas.org/cgi/content/short/1116437108
7 of 8
Table S7. Carbon emissions per hectare from land cleared for crops for the 2nd to 20th y after clearing based on IPCC Tier 1 methodology
and land use changes (1)
Biome
Cropland use
increase
weight*
Grassland
2
Savanna
3
Tropical forests
4
Mediterranean
1
Southern
temperate forests
Weighted average
1
Soil climate region
(IPCC table 2.3)
Mean four temperate
categories
Mean tropical dry and
tropical moist
Mean three tropical
categories (excluding
tropical montane)
Mean of warm temperate
dry, moist and tropical
dry, moist
Mean two warm
temperate categories
Initial soil
Stock change factor
Soil organic C stock
Annual soil C loss from
organic C stocks
climate region
change factor, croplands
cropland years 1–20
(tC ha−1)
(IPCC table 5.5)
(over 20 y)
(tC ha−1)
65.09
0.745
0.830
54.70
Mean temperate/
boreal; dry, moist
Mean tropical
0.530
1.285
62.20
Mean tropical
0.530
1.462
56.95
Mean tropical,
temperate/boral;
dry, moist
Mean temperate/
boreal; dry, moist
0.638
1.032
0.745
0.755
59.20
1.195
*4, most use; 1, least use.
1. IPCC (2006) IPCC Guidelines for National Greenhouse Gas Inventories, eds Eggleston S, et al. (IGES, Hayama, Japan).
Tilman et al. www.pnas.org/cgi/content/short/1116437108
8 of 8