ENVIRONMENTAL ECONOMICS

Do biofuel policies seek to cut
emissions by cutting food?
Major models should make trade-offs more transparent
By T. Searchinger1*, R. Edwards2*,
D. Mulligan2, R. Heimlich3, R. Plevin4

D

ebates about biofuels tend to focus
separately on estimates of adverse
effects on food security, poverty, and
greenhouse gas (GHG) emissions
driven by land-use change (LUC)
(1–4). These estimates often rely on
global agriculture and land-use models. Because models differ substantially in their
estimates of each of these adverse effects
(2, 3, 5), some argue that each individual
effect is too uncertain to influence policy
(6, 7). Yet these arguments fail to recognize
the trade-offs; much of the uncertainty is
only about which adverse effects predominate, not whether adverse effects occur at
all. Our analysis of the three major models used to set government policies in the
United States and Europe suggests that
ethanol policies in effect are relying on decreases in food consumption to generate
GHG savings (1).
When biofuels divert crops from food
and feed, three basic responses are possible.
First, farms may replace crops by expanding
cropland into forests and grasslands. This
LUC releases carbon. Second, farms may replace crops by increasing yields on existing
cropland more than they otherwise would.
This “yield response” is the most desirable,
but it can lead to greater use of fertilizer or
water and increased emissions
POLICY and may leave fewer options to
boost yields to meet rising food
demands. Third, some of the food may not
be replaced, meaning that someone will eat
less or less well. Although effects on different groups of the world’s poor will differ,
any reduction in global food consumption
is likely to disproportionately affect some
groups of the poor because they can less
afford higher prices (2, 8–10). In general,
models predict at least some of each basic
response. To the extent that a model pre1

Woodrow Wilson School of Public and International Afairs,
Princeton University, Princeton, NJ 08544, USA. 2Joint
Research Centre, European Commission, Ispra 21027, Italy.
3
Agricultural Conservation Economics, Laurel, MD 20723, USA.
4
Institute of Transportation Studies, University of California at
Davis, Davis, CA 95616, USA.
*E-mail: tsearchi@princeton.edu, robert.edwards@jrc.it

1420

dicts larger reductions in food consumption, it will predict less LUC, and vice versa.
The role of food reductions depends on
the percentage of crops diverted to biofuels
that are not replaced. Modelers generally
do not report this percentage—although
they sometimes report food effects in less
informative ways (1)—so we calculated the
percentage from model outputs. We considered the consumption and supply effects on
all crops, not just the biofuel feedstock, and
accounted for the return to the food supply
of biofuel feed by-products.

“…25 to 50% of net calories…
diverted to ethanol are not
replaced…”
We analyzed results of the 2009 and 2014
versions of the Purdue Global Trade Analysis Project (PURDUE-GTAP) model (11, 12)
directly incorporated by the California Air
Resources Board (CARB) into regulations,
and of the Food and Agricultural Policy
Research Institute Center for Agricultural
and Rural Development (FAPRI-CARD)
model similarly used by the U.S. Environmental Protection Agency (EPA) (13). We
also analyzed the International Food Policy
Research Institute Modeling International
Relationships in Applied General Equilibrium (IFPRI-MIRAGE) model, used by the
European Commission (EC) (14), to estimate
LUC in proposed European legislation.
These models estimate changes in land
use, crop production, and consumption,
but agencies use separate models to flesh
out remaining parts of life-cycle emissions
estimates. Our intent is not to endorse any
model but to illuminate trade-offs in what
models predict.
CUTTING CALORIES AND QUALITY. These

models estimate that roughly 25 to 50% of
the net calories in corn or wheat diverted to
ethanol are not replaced but instead come
out of food and feed consumption (see table
S3 in the supplementary materials). [Net calories account for return to the food and feed
supply of ethanol by-products (see table S2).]

Replacing fewer crops leads to lower GHG
emissions from LUC. It also reduces direct
emissions of carbon dioxide (CO2) by people
and livestock.
The table presents model estimates of
major GHG sources and sinks attributed to
the use of ethanol and compares them to
relevant estimates of total gasoline emissions. The first column shows CARB, EPA,
and the EC’s Joint Research Centre (JRC)
estimates of fossil carbon and trace gas
emissions from growing crops and refining
them into ethanol using methods separate
from the land-use models (column A). Onethird of net crop carbon is released during fermentation into ethanol (column B),
and two-thirds during combustion of the
ethanol (column C). Adding up all emission
sources (A + B + C), before any credits for
biogenic (plant-based) carbon offsets, GHG
emissions vary from 67 to 100% higher than
those of gasoline.
Most biofuel GHG analyses assume that
carbon absorbed by growth of crops diverted to ethanol automatically offsets the
carbon released by fermenting and burning them (columns B and C). But carbon
absorbed by crops that would grow anyway
cannot provide a valid offset because, by
definition, it is not caused by biofuels and is
not additional (15, 16) (fig. S4). Thus, a valid
crop growth offset comes only from growth
of additional crops to replace those diverted
to ethanol, whether from new cropland or
from boosting yields on existing cropland
(column D). However, conversion of forests
or grassland to produce some more crops
also releases carbon from LUC (column F),
reducing or negating the net offset from
producing more crops.
Significantly, crops that are not replaced
provide an offset in another way. If those
crops were not diverted to ethanol production, people or livestock would consume
them and emit their carbon through respiration and waste. Reducing crop consumption
therefore means that people or livestock emit
less carbon. This offset ranges from 23 to 53
g CO2/MJ for corn or wheat ethanol, equivalent to roughly 20 to 50% of the carbon in the
diverted food (column E).
Including the offset for reduced food consumption, all but CARB’s original GTAPbased results show modestly lower total
emissions for ethanol than for gasoline (column G). However, without crediting reduced
food consumption, none of these models
would project lower GHGs for ethanol than
for gasoline (column H). In that sense, the
lower GHGs for ethanol depend on reductions in food consumption.
Although the table shows the flow of carbon that the models predict, the role of crop
carbon emissions and offsets is normally
sciencemag.org SCIENCE

27 MARCH 2015 • VOL 347 ISSUE 6229

Published by AAAS

Downloaded from www.sciencemag.org on March 26, 2015

INSIGHTS   P E R S P E C T I V E S

 obscured because modelers present results
following a convention of ignoring crop carbon altogether (columns B to E). That works
mathematically because, by definition, the
net carbon in crops devoted to ethanol and
emitted (B + C) must equal and be canceled
out by the carbon in crops that are either
replaced or not replaced (D + E). However,
this practice hides the role played by reduced
food consumption. It also conceals that if all
diverted crops were replaced, LUC emissions
would be larger.
The MIRAGE model also projects reductions in the quality of food consumed (1, 14).
Much of the additional land used for corn
or wheat for ethanol previously produced
higher-value crops, such as oils and vegetables, which is not reflected by the loss of their
calories. Without food quantity and quality reductions—if farmers replaced both the
quantity and types of crops used for food—
farms would require ~5 times as much LUC,

resulting in 46% higher emissions for wheat
ethanol than for gasoline and 68% higher
emissions for corn ethanol.
MIRAGE results are informative because
the model optimistically projects little LUC
due to biofuels, in part because higher yields
in response to higher prices replace three
times as much crop area as does area expansion. Although the economic literature is
sparse and debated, it generally finds a much
larger area response than yield response (1,
17). The fact that global yields more than
doubled from 1961 to 2005, even as prices declined, also casts doubt on such a large yield
response to price (1). Yet, even with little LUC
for this reason, lower ethanol emissions than
gasoline still rely on food reductions.

is an important but separate subject. Many
model parameters and functions are uncertain, debatable, or assumed for mathematical ease. We generally consider likely yield
responses overstated and LUC understated.
Models also may not depict consequences of
increasing biofuels, for example, because one
MJ of ethanol may not fully avoid one MJ of
gasoline (18).
Yet, simpler economic methods also suggest that a substantial fraction of crops diverted to biofuels results in food reductions.
That fraction depends only on the ratio of
the demand response (reduced consumption) to the supply response (increased
production). Typical estimates of supplyand-demand elasticities for individual
crops (19, 20) alone suggest that reduced
consumption is substantial. One direct estimate of global supply-and-demand elasticities for aggregate calories from major
staple crops implies that roughly one-fifth

MAKE TRADE-OFFS TRANSPARENT. Al-

though we illuminate model outputs, whether
any of these models accurately project food or
GHG consequences of increasing ethanol use

Role of reduced food consumption in life-cycle greenhouse gas emissions
DIRECT PRODUCTION AND USE EMISSIONS
(CO2EQ/MJ)

TOTALS AND % CHANGE FROM
GASOLINE (CO2EQ/MJ)

NET OFFSETS (CO2EQ/MJ)

EMISSIONS AND OFFSETS OF CROP CARBON
SOURCE OF FUEL

A Production
and refining
emissions
from fossil
fuels and
trace gases

B
Fermentation
of grain

C
Vehicle
exhaust

D Additional
crop production from both
yield gains and
new cropland
(offset)

E Reduced
respiration and
waste due to
reduced crop
consumption
(offset)

F LUC
(emission
from new
cropland)

G Total
including
reduced food
consumption
(A+B+C+
D+E+F)

H Total
excluding
reduced food
consumption
(A+B+C+D+F)

GASOLINE = 99

CALIFORNIA AIR RESOURCES BOARD
GTAP US CORN
(2009)

69

36

71

–54

–53

42

111 (12%)

164 (65%)

GTAP NEW US
CORN (HIGHER
YIELD ELASTICITY

69

36

71

–75

–32

13

82 (-17%)

114 (15%)

GTAP NEW US
CORN (LOWER
YIELD ELASTICITY

69

36

71

–63

–44

25

94 (–5%)

138 (40%)

GTAP EU WHEAT
(ORIGINAL)

67

36

71

–63

–44

155

223 (125%)

267 (169%)

GASOLINE = 93

U.S. ENVIRONMENTAL PROTECTION AGENCY
FAPRI US CORN
(2022 ESTIMATE)

49

36

71

–86

–25

34

79 (–15%)

104 (12%)

GASOLINE = 87

EUROPEAN UNION
IFPRI-MIRAGE
WHEAT

67

36

71

–73

–34

17

84 (–4%)

118 (36%)

IFPRI-MIRAGE EC
CORN

69

36

71

–84

–23

11

80 (–8%)

103 (19%)

Benefits of biofuels relative to gasoline depend on reductions in food consumption. Ethanol emits fossil carbon and trace gases through production (A) and biogenic
carbon from the crops during fermentation (B) and ethanol combustion (C). Additional crop production can offset these emissions (D), but the net effect of producing more crops
must account for carbon released by LUC (F) (amortized 20 years for IFPRI and 30 years for EPA and CARB). An offset also occurs to the extent that crops are not replaced because
people or livestock eat fewer crops and therefore respire and waste less carbon (E). Total including reduced food production can show lower emissions for ethanol (G) but excluding
that offset shows higher emissions (H). MIRAGE results use updated JRC growing and refining emissions, rather than those in EU legislation, because of greater similarity with CARB
and EPA methods. EPA LUC estimate is based on FAPRI only (1).
SCIENCE sciencemag.org

27 MARCH 2015 • VOL 347 ISSUE 6229

Published by AAAS

1421

 INSIGHTS   P E R S P E C T I V E S

www.sciencemag.org/content/347/6229/1420/suppl/DC1
REFERENCES AND NOTES

1. See supplementary materials for details.
2. High Level Panel on Food Security, Biofuels and food security
(Food and Agricultural Organization, Rome, 2013).
3. National Research Council, Renewable Fuel Standard:
Potential Economic and Environmental Effects of U.S. Biofuel
Policy (National Academies Press, Washington, DC, 2011).
4. U.K. Renewable Fuels Agency, The Gallagher Review of the
Indirect Effects of Biofuels Production (Renewable Fuels
Agency, London, 2008).
5. S. A. Decara et al., Land-Use Change and Environmental
Consequences of Biofuels: A Quantitative Review of the
Literature (Institut National de la Recherche Agronomique,
Paris, 2012).
6. M. Finkbeiner, Biomass Bioenergy 62, 218 (2014).
7. D. Zilberman, G. Hochman, D. Rajagopal, AgBiol. Forum 13, 11
(2010).
8. High Level Panel on Food Security, Price volatility and food
security (Food and Agricultural Organization, Rome, 2011)
9. A. Dorward, Food Security 4, 633 (2012).
10. M. Filipski, K. Covarrubia, in Agricultural Policies for Poverty
Reduction, J. Brooks et al., Eds. (OECD, Paris, 2010).
11. CARB, Proposed regulation to implement the low carbon
fuel standard, Volume I, Staff Report: Initial Statement of
Reasons (California Air Resources Board, Sacramento,
CA 2009).
12. R. Edwards, D. Mulligan, L. Marelli, Indirect land use
change from increased biofuel demand: Comparison
of models and results for marginal biofuels production
from different feedstocks (European Commission Joint
Research Centre, Ispra, Italy 2010).
13. U.S. EPA, Renewable Fuel Standard Program (RFS2)
Regulatory Impact Analysis, (U.S. Environmental
Protection Agency, Washington, DC, 2010).
14. D. Laborde, Assessing the Land Use Change
Consequences of European Biofuel Policies (International
Food Policy Research Institute, Washington, DC, 2011)
15. T. D. Searchinger et al., Science 326, 527 (2009).
16. T. D. Searchinger, Environ. Res. Lttrs. 5, 024007 (2010).
17. S. Berry, Biofuel Policy and Empirical Inputs to GTAP
Models, Report to the California Air Resources Board
(CARB, Sacramento 2011).
18. R. Plevin, M. Delucchi, F. Creutzig, J. Ind. Ecol. 18, 73
(2014).
19. G. Hochman, D. Rajagopal, G. Timilsina, D. Zilberman,
Biomass Bioenergy 68, 106 (2014).
20. Food and Agricultural Policy Research Institute, Elasticity
Database (accessed August 12, 2014); www.fapri.iastate.
edu/tools/elasticity.aspx
21. M. Roberts, W. Schlenker, Am. Econ. Rev. 103, 2265
(2013).

For complex disease genetics,
collaboration drives progress
Exome sequencing identifies a gene that causes
amyotrophic lateral sclerosis
genome-wide association (examining genetic
variants in a genome-wide manner across
individuals for association with a trait), and
family-based next-generation sequencing
have each produced startling new phases of
genetic discovery. The study by Cirulli et al.
(1), reported on page 1436 of this issue, marks
an early success in another phase of gene discovery: the application of exome sequencing in large case-control cohorts to identify
genetic factors involved in complex disease.
This is one of the first successes in this area
and provides insights into amyotrophic lateral sclerosis (ALS), as well as broader lessons for disease gene discovery.
The design of Cirulli et al.’s effort to identify new genes that associate with ALS was
essentially that of a two-phase exome-wide
association study. Several approaches were
used in the initial stage, each centering on
the identification of rare variants within
the protein-coding regions of the genome

By Andrew B. Singleton
and Bryan J. Traynor

A

principal tool in the development of
etiology-based therapies is the identification of the genetic determinants
of disease. The logic of this approach
rests on the expectation that knowledge of the genetic variants underlying disease will enhance our understanding
of the molecular pathogenesis of disease and
reveal viable points for therapeutic intervention. Because of the general adoption of this
paradigm, genetics has been a dominant
force in disease investigation over the past 25
years. Success in this area has come in waves,
each driven by new methods and technology. Linkage (identifying segments of the genome that are associated with given traits),
positional sequencing (sequencing specific
candidate genes based on their location),

ALS genetics
Genetic discoveries in ALS, showing the percentage of cases each gene is involved in and the
type of genetic analysis by which the gene was discovered [adapted from (7)].
TBK1
TUBA4A

Linkage and Sanger sequencing
Second-generation sequencing in families
Linkage, genome-wide association,
and second-generation method
Second-generation sequencing in cohorts

15

CHCHD10
MATR3

PFN1
C90RF72

UBGLN2

10

VCP
OPTN
FUS

5

TDP43

SOD1

ACKNOWLEDGMENTS

The authors thank the David and Lucile Packard Foundation
for financial support.

0
1993

1995

1997

1999

1422

2001

2003

2005

2007

2009

2011

2013

2015

Year

10.1126/science.1261221

sciencemag.org SCIENCE

27 MARCH 2015 • VOL 347 ISSUE 6229

Published by AAAS

CREDIT: ADAPTED FROM (7)

SUPPLEMENTARY MATERIALS

GENETICS

Percent of cases

of calories diverted to biofuels are not replaced (21).
Regardless of model merits, biofuel policies are relying on estimates of reduction in
food consumption to claim GHG benefits.
Food reductions result not from a tailored
tax on overconsumption or high-carbon
foods but from broad global increases in crop
prices (8, 9). If models overstate the food reductions, then they understate the GHGs.
Policy-makers who do not wish to mitigate
climate change in this way could exclude the
GHG credit for these reductions from their
GHG calculations. Modelers need to make
the trade-offs transparent so that policymakers can consider whether to seek climate
mitigation through less food. ■