This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 71.192.160.26

This content was downloaded on 02/11/2015 at 05:06

Please note that terms and conditions apply.

US major crops’ uncertain climate change risks and greenhouse gas mitigation benefits

View the table of contents for this issue, or go to the journal homepage for more

2015 Environ. Res. Lett. 10 115002

(http://iopscience.iop.org/1748-9326/10/11/115002)

Home Search Collections Journals About Contact us My IOPscience

Environ. Res. Lett. 10 (2015) 115002 doi:10.1088/1748-9326/10/11/115002

LETTER

USmajor crops’ uncertain climate change risks and greenhouse gasmitigation benefits

Ian SueWing1, ErwanMonier2, Ari Stern3 andAnupriyaMundra4

1 Department of Earth and Environment, BostonUniversity, USA2 Joint Programon the Science&Policy of Global Change,MIT,USA3 NMRGroup, Somerville,MA,USA4 Joint Global Change Research Institute, University ofMaryland, USA

Keywords: agriculture, climate change impacts, error-correctionmodel, integrated assessment, econometricmodeling

Supplementarymaterial for this article is available online

AbstractWeestimate the costs of climate change toUS agriculture, and associated potential benefits of abatinggreenhouse gas emissions. Fivemajor crops’ yield responses to climatic variation aremodeledempirically, and the results combinedwith climate projections for a no-policy, high-warming future,as well asmoderate and stringentmitigation scenarios. Unabatedwarming reduces yields of wheat andsoybeans by 2050, and cotton by 2100, butmoderate warming increases yields of all crops exceptwheat. Yield changes aremonetized using the results of economic simulations within an integratedclimate-economymodeling framework.Uncontrolledwarming’s economic effects onmajor crops areslightly positive—annual benefits<$4 B. These are amplified by emission reductions, but subject todiminishing returns—by 2100 reaching $17Bundermoderatemitigation, but only $7Bwith stringentmitigation. Costs and benefits are sensitive to irreducible uncertainty about the fertilization effects ofelevated atmospheric carbon dioxide, withoutwhich unabatedwarming incurs net costs of up to $18B, generating benefits tomoderate (stringent)mitigation as large as $26 B ($20 B).

1. Introduction

What are the costs and benefits to US agriculture ofmitigating greenhouse gas (GHG) emissions? Agricul-ture has significant climate change exposure, butdespite being a sector that has long been studied (e.g.,Mendelsohn et al 1994), projections of future impactson crops, and the associated costs of damage, remaintoo uncertain to provide a definitive answer. The issueis highlighted by disagreements over the responses ofUS agricultural yields and profits inferred fromhistorical observations, and their implications for thesign andmagnitude of future climate impacts.

The empirical climate economics literature pro-vides ample evidence that yields of major US crops areadversely affected by exposure to cumulative growingseason degree day extremes (Schlenker et al 2006,Fisher et al 2012) and temperatures above a 86 °F(30 °C) threshold (Schlenker and Roberts 2009, Ortiz-Bobea 2013). But the economic consequences are con-tested. The robustness of accumulated heat’s adverseeffects on farm profits (Schlenker et al 2006, Fisher

et al 2012) has been questioned in light of the poten-tially confounding influence of spatially and tempo-rally varying non-climatic factors (Deschênes andGreenstone 2007, 2012). When combined with earthsystemmodel (ESM) simulations of future climate, thelatter responses suggest that climatic changes experi-enced by 2100 would have only small impacts ontoday’s agricultural system (annual losses of US(2002)-$4 B to -$16 B).

Additional uncertainty abounds in the future tra-jectory of production, and meteorological exposure,of US agriculture—even for a given warming scenario.Despite improved understanding of climate changefeedbacks on land use (Hurtt et al 2011), the futuregeographic distribution and output expansion of USfield crops remain indeterminate. With fixed croppingpatterns and warming trajectories, assessment usingmeteorological exposures from an ensemble of ESMscan increase the range of projected impacts and themagnitude of ‘worst-case’ losses (Deschênes andGreenstone 2012, and especially Burke et al 2015).Addressing the latter uncertainty, the Inter-Sectoral

OPEN ACCESS

RECEIVED

8April 2015

REVISED

28 September 2015

ACCEPTED FOR PUBLICATION

30 September 2015

PUBLISHED

28October 2015

Content from this workmay be used under theterms of theCreativeCommonsAttribution 3.0licence.

Any further distribution ofthis workmustmaintainattribution to theauthor(s) and the title ofthework, journal citationandDOI.

© 2015 IOPPublishing Ltd

Impact Model Intercomparison Project (ISI-MIP)uses global gridded crop models (GGCMs) forced byESM ensemble projections to quantify the range ofcrop shocks to crop yields (Rosenzweig et al 2014),which are in turn employed as input forcings to inte-grated assessment models (IAMs) that simulate con-comitant crop production, price and economicwelfare impacts (Nelson et al 2014).

The dollar value of damages depends critically onthe uncertain state of the economy in the future dec-ades when climate change affects crop yields. How-ever, the ISI-MIP impact modeling protocol’s baselinesocio-economic, technological and GHG mitigationtrajectories are not synchronized with the assump-tions used by the IAMs that simulate the representa-tive concentration pathway scenarios forcing ESMprojections. The key omission is the relative priceeffects of the GHGmitigationmeasures that are neces-sary to realize low-radiative forcing futures. Cost-ben-efit analysis requires a modeling framework that cansimulate the economic effects of mitigation, the cli-matic consequences of the resulting emissions, theconcomitant biophysical impacts and their effects onthe perturbed economy. A further limitation is that theresulting economic impacts understate the potentialbenefits of GHG mitigation by including adaptationthat arises out of IAMs’ price-driven substitutionresponses—among the inputs to crop production andthe outputs of agricultural sectors, and between othergoods and agriculture, and domestic and importedvarieties of each commodity (Nelson et al 2014)—whose cost-reducing effects are difficult tomonetize5.

This paper draws upon and extends aforemen-tioned approaches to estimate the costs of climatechange to US agriculture, and the potential benefits ofGHG mitigation, in a manner that is both economic-ally and climatically consistent, and exclusive of adap-tation. We first econometrically model the long-runyield response of five major crops (corn, soybeans,wheat, cotton, sorghum) to climatic variation, usingdata on weather, output and harvested area for∼3000counties in the coterminous US over the period1948–2010. We then combine the resulting yieldresponses with ESM simulations of climate changescenarios prepared for the US Environmental Protec-tion Agency’s climate change impacts and risk analysis(CIRA) project (Waldhoff et al 2014) to estimate yieldchanges under a no-policy high-warming future aswell as two lower-warming emission mitigation sce-narios. Finally, we use the resulting yield shocks inconjunction with the output of the computable gen-eral equilibrium (CGE) economic model that gener-ated the CIRA emission scenarios (MIT-EPPA—Paltsev et al 2013) to calculate aggregate economiccosts in terms of future revenue changes in each sce-nario. Cost differences between the no-policy and

mitigation scenarios indicate the benefits of reducingGHGemissions.

We find that unmitigated climate change has sub-stantial adverse effects on yields of soybeans and wheatby mid-century, and cotton by century’s end, butcompensating beneficial impacts on corn and sor-ghum yields. Climatic shocks exhibit substantial geo-graphic variation, with yield increases (reductions) inregions that currently have cooler (warmer) averageclimates, and, over time, increasingly severe impacts atlower latitudes. If climatic changes projected by 2100under the reference warming scenario were to occurtoday, annual major crop revenues would be largelyunaffected. But once the agriculture sector’s projectedfuture expansion is taken into account, the upshot isan annual net benefit of US (2010) $3 B by 2050, whichfalls to $1.3 B by 2100. But forgoing less vigorous cli-mate change is nonetheless costly. Both mitigationpolicy scenarios have net beneficial effects, up to $1.2B by 2050 and $2 B by 2100 if climate change were toimpact today’s agricultural system, and $3.3 B by 2050and almost $17 B by 2100 with the price and outputlevel that are projected in the future. However, theseresults are sensitive to the specification of the carbondioxide (CO2) fertilization effect (CFE), whose influ-ence on yields is thought to be positive but subject toconsiderable uncertainty. Omitting the CFE flips thesign of our impact estimates, giving rise to net agri-culture sector costs as high as $18 B, with attendantamplification ofmitigation benefits.

The rest of the paper is organized as follows.Section 2 summarizes our methodology for empiri-cally modeling climate-yield relationships and cou-pling these with ESM simulations. Section 3 presentsthe resulting yield responses to climate change, chan-ges in crop output at the county and regional levels,and monetized damages. In section 4 we offer a sum-mary of ourfindings and discussion of their caveats.

2.Methods

2.1. Empirical analysis: using historical observationsto infer climate impact on yieldsFollowing the recent climate-economics literature(Schlenker and Roberts 2006, 2009, Deschênes andGreenstone 2007, 2012, Lobell and Burke 2010, Ortiz-Bobea 2013, Burke and Emerick 2015)we quantify thepotentially nonlinear influence of climate on yieldsusing semi-parametric cross section-time seriesregressions. Previous studies exploit the historical co-variation between yields and weather shocks to inferthe effects of future climate. Motivated by Burke andEmerick’s (2015) finding that long-run adaptationsare limited in their ability to alleviate the short-runimpacts of extreme heat, we extend their approachusing a dynamic modeling framework that statisticallydistinguishes between the effects of short-run(weather) and long-run (climate) shocks. Since

5These are passive adaptations mediated by relative price changes

(see SueWing and Fisher-Vanden 2013).

2

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

farmers’ planting, management and harvesting deci-sions are based on land quality and expectations ofweather, yields and meteorological variables share along-run equilibrium relationship. In any given year,weather shocks cause yields to diverge from theirexpected long-run values, prompting farmers to revisetheir long-run expectations, and make managementdecisions that can have persistent effects6.

To statistically identify the former equilibrium andlatter disequilibrium responses we employ an error-correctionmodel (ECM)7. Our data are an unbalancedpanel of c counties over t years, recording yields, Y(calculated as the ratio of production, Q, to harvestedarea, H ,) as well as three-hourly observations ofgrowing season temperature, precipitation and soilmoisture, indexed by v T P S, ,{ }= , respectively.Interannual variation in log annual yield (y) is mod-eled as a function of a vector of county-specific effects( ,m which capture the influence of unobserved time-invariant local characteristics such as topography and

soils), a vector of climatic covariates (within each

annual growing season, the cumulative exposure overg crop growth stages to j temperature intervals, ,j g

T,x k

precipitation intervals, ,k gP,x and l soil moisture inter-

vals, ,l gS, )x and a vector of time-varying county-level

statistical controls X( ).Our model, which is derived and explained in the

supplementary information (SI), is written:

X

X

y y y

Q

H

Q

H

y

log log

1

c t c t c t

c t

c t

c t

c t

c g j j gT

j g c tT

k k gP

k g c tP

l l gS

l g c tS

c t c t

g j j gT

j g c tT

k k gP

k g c tP

l l gS

l g c tS

c t c t

, , , 1

,

,

, 1

, 1

, , , , , , , ,

, , , , , , 1

, , , , 1 , , , , 1

, , , , 1 , 1 ,

{}

{}

( )

⎛⎝⎜

⎞⎠⎟

⎛⎝⎜

⎞⎠⎟

⎡⎣

⎤⎦

g

l

m b x b x

b x q

h x h x

h x e

D = -

= -

= + S S D + S D

+ S D + D +

- S S + S

+ S - +

-

-

-

-

- -

- -

and is estimated via ordinary least squares for fivecrops (corn, soybeans, wheat, cotton, sorghum),indexed by i. Interannual difference terms (prefixedby )D model the yield impacts of transitory disequili-brium shocks, the expression in square braces capturesthe long-run equilibrium relationship between yieldsand the covariates, and e is a random disturbanceterm. Parameters to be estimated are the disequili-brium (weather) impacts (β v), equilibrium (climate)impacts (η v), short- and long-run effects of non-

climatic variables (γ and λ), and the error-correctionparameter ( )q measuring producers’ speed of adjust-ment to the long-run equilibrium. The parameters(η v) are vectors of semi-elasticities indicating thepercentage by which yields shift relative to theirconditional mean levels in response to additional timespent in a given interval. The vectors’ elements—theindividual coefficient estimates—each capture thedistinct marginal effect of exposure within the corre-sponding interval (e.g., the average impact of anadditional hour to 70–80 °F versus 80–90 °F tempera-tures). Collectively, the elements of vh flexibly capturev ’s overall long-run effect as a piecewise linear spline.The shape of the resulting function is identified fromthe covariation between observed yields andmeteorol-ogy within each interval, as well as the distribution ofobservations across intervals over the historical periodof the sample. With regard to temperature, theadvantage of this approach is that it more preciselyresolves the yield impacts of extreme heat relative tothe standard degree-day specification (see SchlenkerandRoberts 2009). Our dataset is described in the SI.

Omitted from equation (1) is the CFE. Rising CO2

concentrations are a time-varying shock that simulta-neously affects yields in all counties. However, there isnear-perfect collinearity between the CFE and long-run impacts of other strongly trending, spatiallyhomogeneous, beneficial influences such as total fac-tor productivity improvements or technological pro-gress. Data constraints preclude quantification of thelatter with accuracy sufficient to construct crediblestatistical controls8. Given the potential for the long-run coefficient on CO2 concentrations to erroneouslycapture these confounding secular effects, we eschewempirical estimation of the CFE and instead incorpo-rate its effect on our yield projections using relation-ships based on the literature.

2.2. Projecting yield impacts of future climatechangeClimate change impacts are quantified by combiningthe fitted values of the equilibrium meteorologicalparameters ( vˆ )h with meteorological exposuresderived from ESM simulations of different warmingscenarios. We spatially aggregate simulated 3-hourlyfields of temperature, precipitation and soil moistureto the county level (T ,c̃ Pc̃ and Sc˜ ) and bin theresults into the j, k and l intervals (respectively) overcrop growth stages in current and future growingseasons to generate exact analogs of the regressioncovariates,

vx̃ 9. Yields under irrigated and rainfed

management regimes (indexed by m I R,{ })= exhi-bit different responses to precipitation andmoisture as

6A key example is soil amendments. With agricultural profits,

analogous decisions involve inventory adjustments (Deschênes andGreenstone 2012).7See Nickell (1985). Prior research employing ECMs to understand

climate change impacts on agriculture (e.g., Blanc 2012) has not toour knowledge sought to explicitly partition yield variance into theeffects of weather versus climate. For a general application of ECMsto agricultural supply response, seeHallam andZanoli (1993).

8Absent specific indicators of technological advance (e.g. patent

stocks), productivity improvements are customarily modeled usinga time trend.9Comparing the future climate simulated by an ESM against

current climate simulated by the same model (as opposed toobservations) is a way of minimizing the impact of potential bias inESMprojections.

3

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

well as elevated CO2. Accordingly, we model themseparately (see SI), specifying rainfed impacts as afunction of temperature, precipitation, soil moistureand ambient CO2 concentrations ,( ˜ )C and irrigatedimpacts as a functions of temperature andCO2:

T P S

a

, , ,

log , 2

iR

c c c

gj

i j gT

j g c

T

ki k gP

k g c

P

li l gS

l g c

SiR

, , , , , , , ,

, , , ,

( )

( )

˜ ˜ ˜ ˜

ˆ ˜ ˆ ˜

ˆ ˜ ˜ ( )

C

C

⎪

⎪

⎧⎨⎩

⎫⎬⎭

å å

å

y

h x h x

h x

= S +

+ +

T

b

, log .

2

iI

c gj

i j gT

j g c

TiI

, , , ,( ) ( )˜ ˜ ˆ ˜ ˜

( )

C C⎪

⎪

⎪

⎪

⎧⎨⎩

⎫⎬⎭

åy h x= S +

Here, im is a concave function that captures

the differential benefits of CO2 fertilization underdifferent moisture stress conditions, based onHatfield et al (2011) and McGrath and Lobell (2013).Our calibration of the CFE index is documented inthe SI.

The terms i cm,y indicate the partial effects of cli-

mate on the logarithm of irrigated and rainfed yields.Our normalized decadal index of climate impact is theyield ratio:

T P S

T P

S

T

T

exp , ,

, ,

,

exp ,

,

3

i c cR

iR

c c c

iR

c c

c

cI

iI

c

iI

c

,

FutureClimate

FutureClimate

FutureClimate

FutureClimate

CurrentClimate

CurrentClimate

CurrentClimate

CurrentClimate

FutureClimate

FutureClimate

CurrentClimate

CurrentClimate

{

˜ ˜ ˜

˜ ˜

˜

˜

˜

¯

˜

˜

¯ ˜

˜

( )

C

C

C

C

⎪

⎪

⎪

⎪

⎪

⎪

⎡

⎣⎢⎢

⎧⎨⎩

⎛⎝⎜⎜

⎞⎠⎟

⎛⎝⎜⎜

⎞⎠⎟⎟

⎫⎬⎭

⎛⎝⎜⎜

⎞⎠⎟⎟

⎛⎝⎜⎜

⎞⎠⎟⎟

⎫⎬⎭

⎤

⎦⎥⎥

f y

y

f y

y

Y =

-

+

-

in which is the expectation operator and cvf̄ denotes

the average shares of irrigated and rainfed cultivationfrom theMIRCA dataset (Portmann et al 2010), whichwe treat as remaining fixed into the future. (Thisassumption is discussed in § 3.2.) i c,Y is interpretableas the climatically-attributable fractional change in acounty’s average yield relative to its own conditionalmean10. Accordingly, holding the current geographicdistribution of harvested area constant as well, thechange in production of each crop is simply the

quantity Yi c i c, ,

CurrentClimate¯Y ´ .

Our simulated meteorological fields and ambientCO2 concentrations are taken from the CIRA project(Waldhoff et al 2014), a 15-member ensemble ofsimulations using the MIT Integrated Global SystemModel-Community Atmosphere Model (IGSM-CAM) modeling framework (Monier et al 2013)11.CIRA is underlain by three consistent socioeconomicand emissions scenarios: a reference scenario withunconstrained emissions and two climate stabiliza-tion scenarios that impose uniform global taxes onGHGs to limit total radiative forcing to 4.5 Wm−2

and 3.7 Wm−2 by century’s end. Reductions in cli-mate damages to agriculture in moving from thereference to the policy scenarios are interpretable asthe benefits of GHG mitigation, and the associateddifferences in US agriculture sector output and rela-tive prices are crucial to our cost estimates (Paltsevet al 2013). For each emission scenario, IGSM-CAMwas run with different values of climate sensitivityand aerosol forcing, and different representations ofnatural variability, resulting in a 60-member ensem-ble (Monier et al 2015).We focus on simulations witha climate sensitivity of 3 °C, with each scenario run asa 5-member initial condition ensemble in an attemptto span the potential range of natural variability. Spa-tially disaggregating these projections to the countyscale and using equations (2) and (3) enables us tocalculate yield impacts at the middle and the end ofthe century (2036–2065 and 2086–2115) for eachcombination of scenario and ensemble member. Weanalyze 30-year time periods over 5 ensemble mem-bers with different representations of natural varia-bility, resulting in a total of 150 years definingchanges from the present day to the middle and endof the century, in order to obtain robust estimates ofclimate impacts on yield where the anthropogenicsignal is extracted from the noise associated with nat-ural variability.

3. Results

3.1. Yield responses to climate changeOur long-run estimates are for the most part broadlyconsistent with current agronomic understanding ofweather effects on field crop yields. Space constraintspreclude detailed description of these results for allfivecrops. We highlight our findings for corn and consignthe remaining results to the SI. Figure 1 shows corn’smeteorological yield response functions (panel A) andthe changes in exposure to weather conditions experi-enced by an average county in our three scenarios circa2050 and 2100 (panels B and C). Yields declineprecipitously with extreme temperature (Schlenker

10If 0,1i ( )Y Î the shift in themean climatic exposure reduces crop

productivity, and increases it otherwise.

11IGSM-CAM links the IGSM, an integrated assessment model

coupling an earth systemmodel of intermediate complexity (EMIC)to a global economic model (MIT EPPA—Paltsev et al 2013), withtheNational Center for Atmospheric Research (NCAR)CommunityAtmosphereModel (CAM—Collins et al 2006).

4

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

and Roberts 2009, Burke and Emerick 2015), butstratification of our responses by growth phase high-lights the large impact in the first half of the growingseason (Ortiz-Bobea 2013)—each additional 3 h per-iod below 15 °C increases yields by as much as 0.005%

relative to their conditional mean, but similar expo-sure above 40 °C triggers a reduction of more than0.01%. Over the second half of the growing season,temperatures cause slight yield declines below thelatter threshold but a marked increase above (as much

Figure 1.Corn empirical yield response functions (A) and the change in the distributions of average county temperature, precipitationand soilmoisture exposure circa year 2050 (B) and 2100 (C), over growing season sub-periods. Gaps in splines correspond toomittedmodal intervals. Histograms show the differences in the distributions of exposure between the no-policy reference scenarioand the current climate, and between the 4.5 W m−2 and 3.7 W m−2 GHGmitigation scenarios and the reference case. The verticalaxes of the differenced exposure distributions have non-linear scales to better illustrate the shifts inmeteorology due to climatechange.

5

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

as 0.01% per 3 h)12. A key point of divergence withprior results is our finding that yields increase stronglyand approximately linearly with precipitation, withtrace amounts associated with slight declines but 3 hextreme exposures (>15 mm) increasing yields by upto 0.01% (0.015%) in the early (late) sub-periods. Theinstantaneous soil moisture response exhibits a gen-eralized inverse U shape with an apex at the condi-tional mean exposure, a very slight negative influenceover most of its range and sharply negative impact atthe upper extreme (>35 kg m−2), with reduction of upto 0.004% (0.008%) in the early (late) sub-periods.Other crops’ responses share many of these character-istics (figures S3–S6)13.

In panels B and C, movements in the average dis-tribution of exposure due to unmitigated climatechange shifts the relative weights on different splinesegments. The larger the increases in exposure withinintervals associated with negative semi-elasticities (thepink bars), the greater the downward pressure onyield. Throughout the growing season temperatureincreases shift exposure out of low-temperature inter-vals into high-temperature intervals, negatively affect-ing yields, particularly in early stages of crop growth.Climate change is also associated with increases inboth rainfall and soil moisture that shift exposurefrom low to high ranges of these variables, which isgenerally beneficial. But where the latter intervals areassociated with negative marginal effects, the sign ofthe yield impacts is negative aswell.

GHG mitigation’s broad influence is to partiallyreverse these shifts in probability mass (the green andblue bars). Across meteorological variables, the mostcommon pattern is for the bars indicating the policyscenarios to be of smallermagnitude butmostly oppo-site sign to those corresponding to the reference case.Crucially, such reversals are not always beneficial. In

the reference scenario, exposure to low precipitationdeclines in both halves of the growing season circa2100, and, relative to this outcome, GHG emissionmitigation increases the average frequency of such dryepisodes, with adverse late-season yield impacts. Simi-larly, increases in large precipitation events under thereference scenario improve yields, but mitigationreduces these increases, curtailing this particular bene-fit from unmitigated climate change. Finally, as indi-cated in the SI (figure S2), a pervasive consequence ofmitigation is the reduction in the CFE and its atten-dant yield benefits.

3.2. Projected changes in crop yields and productionConditions within individual counties can divergemarkedly from the aforementioned average changes inexposure. Maps of projected yield changes in figure 2indicate the spatial patterns of the threat to the fivemajor crops posed by unmitigated climate change, aswell as the substantial threat reduction due to moder-ate mitigation. Panel A shows the counties where cropproduction is concentrated, while panels B and Cillustrate the percentage changes in yields calculatedusing equation (3). All crops experience both bene-ficial and adverse effects, depending on the region. Inthe reference scenario, wheat yields increase in thenorthwest and decline in the south central and south-west regions circa 2050, a pattern that intensifiesmarkedly toward century’s end. For the remainingcrops, the patterns of impact tend to follow the north-south temperature gradient. Soybean and corn yieldssuffer pronounced negative impacts in the South andthe Mississippi River valley that first lessen beforeturning positive with proximity to Canada. For cotton,the largest adverse effects are dispersed across thecrescent of southernmost counties, while for sorghumnegative impacts are concentrated in the southwest.The reductions in changes in climatic variables as aconsequence of mitigation policies attenuate theamplitude of beneficial as well as adverse yield shocks.Even a 4.5Wm−2 GHG stabilization policy limitsimpacts to ±10% from baseline levels in the majorityof cultivated counties. Results for the stringent3.7Wm−2 scenario (not shown) are similar butfurther accentuated.

However, the risk to agricultural supply arises outof the spatial intersection of yield shocks and patternsof crop production in future decades when climaticchanges occur. Although the latter will likely differfrom today, given the challenges that attend predictionof agriculture’s future geographic distribution (see,e.g., Ortiz-Bobea and Just 2012, Iizumi and Raman-kutty 2015), we follow the empirical literature inassuming that irrigated and rainfed crop cultivationwill continue to follow the current geographic patternin panel A.We quantify the implications at the scale ofUS climate regions in table 1.

12The positive late response to high temperature is not the result of

outlying observations. Historically, corn has regularly been exposedto this kind of heat, albeit in tiny amounts. In 23% of our 136 000county × year observations, corn was exposure to one or more 3 hperiods with T 43 C> in the second half of the growing season,covering 2307 out of 2842 counties and all the years of our sample.Using simpler empirical models, Blanc and Sultan (2015: figuresC1–C4) uncover similar beneficial yield responses to late extremeheat in the results of ISI-MIP GGCMs. Work remains to be done tounderstand themechanisms responsible for this phenomenon, bothinGGCMs and thefield.13

We find negative and strongly nonlinear temperature sensitivityof sorghum, soybeans, and, to a lesser extent wheat, especially in thefirst half of the growing season (see Tack et al 2015). For the mostpart, precipitation’s long-run effect is positive or statisticallyinsignificant over most of its range (especially in the second half ofthe growing period). Exceptions are the negative and significantimpacts of early extreme precipitation on soybeans and wheat.Long-run soil moisture responses peak apex at the modal exposure,suggesting detrimental yield impacts of soil waterlogging as well asdrying, with the exception of cotton early in the growing season.These results are generally in line with recent empirical findings. Ashortcoming of our model is its omission of freezing temperaturesand the associated negative response of wheat yields, whoseamelioration in a warming climate provides an offsetting beneficialeffect (Tack et al 2015).

6

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

Under reference warming, circa 2050, there areincreases in yields of corn and sorghum, declines inwheat and soybeans, and mixed impacts on cotton inthe regions where production of these crops is con-centrated. Climate impacts thatmanifest in one or tworegions at mid-century often spread geographically by2100, with production in regions that suffer earlyadverse impacts (e.g., the southeast, and, to a lesserextent, south central regions) often experiencing fur-ther declines.Mitigation often only softens the blow inregions with the largest percentage losses, and, para-doxically, changes in weather patterns associated withstringent emission reductionsmay have smaller ameli-orative impacts (see south, southeast and central sor-ghum and corn), likely due to the interplay betweenthe impact of changes in meteorological variables andthe CFE. Conversely, climate change improves yieldsin the cooler northeast, northwest, and, less reliably,north central areas aroundmid-century, with declinesin the pool of regions experiencing beneficial weatheras warming proceeds. Mitigation offsets outputdeclines from regions experiencing losses at the cost ofcurtailing gains to those benefiting from climatechange. However, rarely does mitigation transformgains under the reference into outright losses: morecommonly regions that gain experience smallerbenefits.

Our projected percentage changes in aggregateyield understate those of prior studies, although aclean comparison is elusive because of differences inthe scenarios of future warming and their meteor-ological consequences as elaborated by ESMs (see table

S3). Our inclusion of the CFE accounts for some ofthis divergence. Re-running our projections withoutthe CFE14 results in yield losses that are 20% larger forwheat and between 2 and 4 times as large for soybeansand cotton, and yield gains for corn and sorghum thatare 10%–15% smaller (table S2). More consequentialare our findings of countervailing effects of extremeheat on corn yields (adverse early, beneficial late), andthe importance of precipitation generally. The latter isparticularly important given that our estimatesassume no water stress, and therefore no impact ofprojected changes in precipitation or soil moisture, onthe irrigated fraction of the crop in each county. Rela-tive to the customarymethod of applying a single fittedyield response function everywhere, our approachreduces yield losses (gains) in regions experiencingprecipitation and soilmoisture declines (increases).

3.3. Economic costs of agricultural impacts andbenefits of GHGmitigationThe implications for aggregate climate damage costsand GHG mitigation benefits are summarized intable 2. Costs (negative entries)and benefits (positiveentries) are assessed by establishing two baselines fromwhich to compute the absolute changes in output thatcorrespond to US-wide percentage changes. Panel A,which collapses table 1, demonstrates that, underreference warming, adverse national average yieldimpacts are dominated by wheat, soybeans and,toward century’s end, cotton. Conversely, corn and

Figure 2.Geographic distributions of 1980–2010 crop production (A), and%change in yields offivemajorUS crops relative tocurrent climate circa years 2050 and 2100, under a no-policy reference scenario (B) and amoderate GHGmitigation scenario (C).

14This is achieved simply by setting 0i

mk = in equation (S9).

7

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

sorghumexperience large increases in national averageyield. Panel B shows the result of a comparative staticcalculation of the associated costs and benefits ofclimate change if crop production and prices remain attoday’s levels. Output losses (i) follow the referencepattern in Panel A, but negative impacts on wheat arelessened and on soybeans are reversed by mitigation’sreduction of warming to beneficial levels. The corre-sponding economic impacts (ii) are expressed as

changes in revenue, calculated by multiplying thequantity shocks by each crop’s 1981–2010 average realfarmgate price. Broadly echoing Deschênes andGreenstone’s (2007) findings, climate change has a netbeneficial impact which is modest at mid-century ($1B) but becomes negligibly small by 2100. Relative tothe reference, moderate mitigation generates addi-tional annual benefits of $1 B ($2 B) circa 2050 (2100),while stringentmitigation’s benefits are smaller: $0.6 B

Table 1.Changes in crop production (%) relative to current climate in the no-policy reference andGHGmitigation scenarios, circa years2050 and 2100: byUS climate regions.

Note: Square braces: changes in output from the reference scenario; shaded cells: losses relative to the current period (in cells with square

braces, relative to reference scenario); bold:major producing regions.

8

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

($0.8 B) per year by 2050 (2100). Panel B’s estimatesincorporate future increases in production, and asso-ciated price changes, as simulated by MIT-EPPA’sCIRA runs. Impacts are identical in sign, but expan-sions in crop output and revenue increase the magni-tude of costs and benefits15. Net annual benefits underreference warming are still small ($3 B circa 2050, $1Bby 2100), while moderate (stringent) mitigation gives

rise tomodest additional annual benefits of $3 B ($1 B)by 2050 and $17 B ($7 B) by 2100.

4.Discussion and conclusions

By combining empirical analysis with integratedeconomic and climate projections, we demonstratethat climate change effects on US crop yields are likelyto be slight around mid-century but substantiallycostly near century’s end, with regions where climatesare already warm suffering losses but cooler regions

Table 2.Aggregate annual changes in crop yields, production and associated gross costs and benefits relative tocurrent climate in the no-policy reference scenario, and aggregate avoided changes in crop yields and associatedcosts and benefits under GHGmitigation scenarios, circa years 2050 and 2100. (A) aggregate yield changes; (B)prices and quantities in current agricultural system; (C)prices and quantities scaled according to future growthsimulated by theMIT-EPPAmodel’s CIRA simulations.

2036–2055 2086–2115

Ref 4.5 W m−2 3.7 W m−2 Ref 4.5 W m−2 3.7 W m−2

(A)Average change in yield relative to current climate (%)

Wheat −3.5 −0.3 −0.8 −7.9 −1.3 −1.9

Soybeans −1.6 3.5 1.8 −7.8 1.4 0.3

Sorghum 6.7 3.0 2.6 18.1 5.3 1.7

Cotton 2.2 3.6 2.6 −3.5 3.2 1.5

Corn 6.4 5.3 4.8 10.4 6.6 3.8

(B)Current agricultural system

(i)Average change in production relative to current climate (106 tons)Wheat −2.1 −0.2 −0.5 −4.9 −0.8 −1.2

Soybeans −1.1 2.3 1.2 −5.2 1.0 0.2

Sorghum 1.1 0.5 0.4 2.9 0.9 0.3

Cotton 0.1 0.1 0.1 −0.1 0.1 0.1

Corn 14.9 12.3 11.0 24.1 15.2 8.7

(ii) Impact gross cost (negative) or benefit (positive) in reference scenario;mitigation net benefit

(positive) or cost (negative) in policy scenarios (2010$M)Wheat −388 360 297 −887 741 673

Soybeans −336 1050 698 −1608 1904 1664

Sorghum 106 −58 −65 288 −204 −261

Cotton 138 85 24 −222 424 313

Corn 1503 −258 −390 2434 −895 −1553

Total 1024 1179 563 5 1971 836

(C)Projected future agricultural system

(i)Average change in production (106 tons)Wheat −5.5 −0.4 −1.1 −32.4 −4.6 −5.8

Soybeans −2.8 5.5 2.6 −34.3 6.3 1.2

Sorghum 2.8 1.2 0.9 19.5 5.7 1.8

Cotton 0.2 0.3 0.2 −0.3 0.3 0.1

Corn 38.0 29.4 24.7 159.1 100.6 57.6

(ii) Impact gross cost (negative) or benefit (positive) in reference scenario;mitigation net benefit

(positive) or cost (negative) in policy scenarios (2010$M)Wheat −1232 1170 1011 −8395 7368 7475

Soybeans −1065 3251 2143 −15 230 18 442 17 125

Sorghum 336 −199 −231 2726 −1973 −2682

Cotton 439 229 12 −811 1539 1199

Corn 4769 −1093 −1730 23 048 −8666 −15 985

Total 3247 3358 1205 1338 16710 7132

15Relative to today, composite agricultural output is projected to

increase by a factor of 2.5 by 2050 and 6.5 by 2100, with realcomposite agricultural prices increasing by 25% and 44%,respectively.

9

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

enjoying gains, and declines in production concen-trated in soybeans, cotton and wheat that are partiallyoffset by increases in output of corn and sorghum.Reductions in radiative forcing from GHG mitigationgenerally offset output declines from regions and cropsthat experience losses, but at the cost of curtailinggains to those that benefit from climate change.

As summarized in figure 3, our results suggest thatthe overall effect of mitigation policies on agriculturalrevenues will be positive, but the magnitude is sensi-tive to the beneficial impacts of CO2 fertilization.Without the CFE, the impact of unmitigated climatechange flips sign, incurring annual net costs of $3 B($18 B) circa 2050 (2100). This amplifies the positiveeffect of emission reductions, increasing the benefitsof moderate mitigation by $1.4 B ($10 B) circa 2050(2100), and of stringent mitigation by $1.1 B ($13 B)circa 2050 (2100). These estimates, which should beconsidered upper bounds on the costs of climatechange impact and corresponding emission reductionbenefits, highlight the critical importance of assump-tions regarding the CFE. They are also somewhat lar-ger than, but in the same general range as, climatechange damages generated by prior studies (see tableS2), though simple comparisons of total dollar valuesare not appropriate given the use of different impactendpoints (land values, agricultural profits, non-mon-etized yields, or affected crops) and climate changeprojections, as well as the lack of accounting for theCFE. As a case in point, the study most closely relatedto ours—Beach et al (2015)—employs a crop model

forced by the CIRA IGSM-CAM simulations to con-struct gross-of-CFE changes in corn, soybean andwheat yields and which are then applied as exogenousshocks in a partial equilibrium simulation of the USagriculture and forestry sector. Yield impacts aremostly positive in the reference scenario and becomemore beneficial with stringent mitigation, over2015–2100 increasing cumulative agricultural surplusby (2005) $45 B—or an average annual mitigationbenefit of $0.5 B.

Our analysis represents an advance over currentapproaches to quantifying the costs and benefits of cli-mate change. We use projections of the future state ofthe agricultural economy using CGE model resultswhose simulated growth rates of agricultural outputand prices are consistent with the economic expan-sion, general equilibrium price and quantity effects ofmitigation policies, and concomitant GHG emissions,radiative forcing and meteorological changes thatdetermine the shocks to crop yields in the decades inwhich these impacts occur. By contrast, empirical stu-dies’ comparative statics valuation of impacts as chan-ges in agricultural revenues or profits under currentproduction and prices can understate costs (seetable 2). Modeling studies that simply impose GGCM-simulated yield changes onto economic models riskbeing inconsistent with the future economic condi-tions, GHG emissions and the climatic forcing of yieldshocks that we argue is essential to consistent estima-tion of costs and benefits. But the critical feature ofsuch model-based economic consequence analyses is

Figure 3.Annual costs (negative entries) and benefits (positive entries) of climate change impacts andmitigation onUS agriculturewith andwithoutCO2 fertilization, circa 2050 and 2100. (A) Impacts by crop (total shown at bottom). (B)Costs and benefits of4.5 W m−2 and 3.7 W m−2 stabilization policies by crop (total shown at bottom).

10

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

the additional uncertainty they introduce by simulat-ing the moderating effects of adaptation via market-mediated price and quantity adjustments (see Beachet al 2015). Although adaptation will almost surelyoccur, its associated indirect economic costs and bene-fits are poorly characterized and difficult to estimate,yet require accurate quantification to avoid potentialdouble-counting when estimating the net benefit ofmitigation. Our deliberately conservative approach istherefore to exclude the effects of future adaptationfrom our cost-benefit calculations. Instead we valuethe impact of climate change under the economic con-ditions likely to prevail at the instant such a shockoccurs, before producers and consumers have anopportunity to react (Fisher-Vanden et al 2013).

Nevertheless, several caveats to our analysisremain. Our narrow focus on well-documentedimpact pathways omits myriad indirect climate-rela-ted changes in crops’ growing environment (e.g.,ozone concentrations, diseases, pathogens and weeds)on which the literature provides less guidance regard-ing yield responses. Space constraints preclude a fulluncertainty analysis of the underlying economicassumptions in the MIT EPPA model and the climatesystem response, and particularly the CFE’s yield ben-efits, which were difficult to bound (see SI). Thedependence of yield changes on the assumption ofperfect water application in currently irrigated areashighlights the sensitivity of our cost-benefit projec-tions to the availability of water resources sufficient forirrigation as crop production expands out to century’send. The latter, while driven by shifting precipitationpatterns, requires hydrological analysis (e.g., futurewater infrastructure and efficiency assumptions, chan-ges in runoff and discharge, competition with growingmunicipal and industrial demands, groundwaterresource development and depletion) to determinehow, andwhere, it might influence our results. Finally,the CGE model that we use resolves future changes inaggregate agricultural activity, not individual crops.Research is ongoing to address these issues.

More broadly, our results illustrate the potential ofreduced-form empirical analysis as an alternative toGGCMs in evaluating climate change impacts on agri-culture (Nelson et al 2014, Rosenzweig et al 2014).Although crop models incorporate both detailed pro-cess-based understanding of crop physiology and theameliorating effects of a plethora of managementoptions, concerns regarding their accuracy in captur-ing crop yields’ responses to meteorological change(Hertel and Lobell 2014) have been slow to promptextensive testing, especially at the geographic scalesexamined here16. Our methodology can usefully beapplied to model the relationships between GGCMsimulated yields and their climatic drivers, and

thereby facilitate head-to-head comparisons that canlead to more robust estimates of impact response,future yield shocks, and associated economic costs andbenefits.

Acknowledgments

ISW, AS and AMgratefully acknowledge support fromNSF (grant nos. EAR-1038907 and GEO-1240507),and US Department of Energy Office of Science (BER)(grant no. DE-SC005171). EM gratefully acknowl-edges support from the US Environmental ProtectionAgency’s ClimateChangeDivision, underCooperativeAgreement #XA-83600001 and from the US Depart-ment of Energy, Office of Biological and Environmen-tal Research, under grantDEFG02-94ER61937.

References

BeachR et al 2015Climate change impacts onUS agriculture andforestry: benefits of global climate stabilization Environ. Res.Lett. 10 095004

Blanc E 2012The impact of climate change on crop yields in sub-saharanAfricaAm. J. Clim. Change 1 1–13

Blanc E and Sultan B 2015Emulatingmaize yields from globalgridded cropmodels using statistical estimatesAgric. ForestMeteorol. 214–215 134–47

BurkeM,Dykema J, Lobell D,Miguel T and Satyanath 2015Incorporating climate uncertainty into estimates of climatechange impactsRev. Econ. Stat. 97 461–71

BurkeMandEmerick K 2015Adaptation to climate change:evidence fromUS agricultureAm. Econ. J.—Econ. Policy atpress

CollinsW et al 2006The community climate systemmodel version 3(CCSM3) J. Clim. 19 2122–43

DeschênesO andGreenstoneM2007The economic impacts ofclimate change: evidence from agricultural output andrandomfluctuations inweatherAm. Econ. Rev. 97 354–85

DeschênesO andGreenstoneM2012The economic impacts ofclimate change: evidence from agricultural output andrandomfluctuations inweather: replyAm. Econ. Rev. 1023761–73

Estes LD, Bradley BA, BeukesH,HoleDG, LauM,OppenheimerMG, Schulze R, TadrossMA andTurnerWR2013aComparingmechanistic and empiricalmodelprojections of crop suitability and productivity: implicationsfor ecological forecastingGlob. Ecol.Biogeogr. 22 1007–18

Estes LD, BeukesH, Bradley BA,Debats S R,OppenheimerM,RuaneAC, Schulze R andTadrossM2013bProjected climateimpacts to SouthAfricanmaize andwheat production in2055: a comparison of empirical andmechanisticmodelingapproachesGlob. Change Biol. 19 3762–74

Fisher AC,HanemannWM,RobertsM J and SchlenkerW2012The economic impacts of climate change: evidence fromagricultural output and randomfluctuations inweather:commentAm. Econ. Rev. 102 3749–60

Fisher-VandenK, SueWing I, Lanzi E and PoppDC2013Modeling climate change feedbacks and adaptationresponses: recent approaches and shortcomingsClim.Change 117 481–95

HallamZ andZanoli R 1993 Error correctionmodels andagricultural supply response Eur. Rev. Agric. Econ. 20 151–66

Hansen ZK, Lowe S E andXuW2014 Long-term impacts ofmajorwater storage facilities on agriculture and the naturalenvironment: evidence from Idaho (US)Ecol. Econ. 100106–18

Hatfield J L, Boote K J, Kimball BA, Ziska LH, Izaurralde RC,OrtD, ThomsonAMandWolfeD 2011Climate impacts on

16Fortunately, this situation is beginning to change—see, e.g.,

Lobell and Burke (2010), Estes et al (2013a, 2013b), Blanc and Sultan(2015) andMontesino-SanMartin et al (2015).

11

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al

agriculture: implications for crop productionAgronomy J.103 351–70

Hertel TWand Lobell DB 2014Agricultural adaptation to climatechange in rich and poor countries: currentmodeling practiceand potential for empirical contributions Energy Econ. 46562–75

Hurtt GC et al 2011Harmonization of land-use scenarios for theperiod 1500–2100: 600 years of global gridded annual land-use transitions, wood harvest, and resulting secondary landsClim. Change 109 117–61

Iizumi T andRamankuttyN 2015Howdoweather and climateinfluence cropping area and intensity?Glob. Food Security 446–50

Lobell DB and Field CB 2008 Estimation of the carbon dioxide(CO2) fertilization effect using growth rate anomalies of CO2

and crop yields since 1961Glob. Change Biol. 14 39–45Lobell DB andBurkeMB2010On the use of statisticalmodels to

predict crop yield responses to climate changeAgric. ForestMeteorol. 150 1443–52

McGrath JM and Lobell DB 2013Regional disparities in theCO2

fertilization effect and implications for crop yieldsEnviron.Res. Lett. 8 014054

MendelsohnR,NordhausWDand ShawD1994The impact ofglobal warming on agriculture: a Ricardian analysisAm. Econ.Rev. 84 753–71

Monier E, GaoA, Scott J, Sokolov A and Schlosser A 2015Aframework formodeling uncertainty in regional climatechangeClim. Change 131 51–66

Monier E, Scott J, Sokolov A, Forest C and Schlosser A 2013Anintegrated assessmentmodeling framework for uncertaintystudies in global and regional climate change: theMIT IGSM-CAM (version 1.0)Geosci.Model Dev. 6 2063–85

Montesino-SanMartínM,Olesen J E andPorter J R 2015Can crop-climatemodels be accurate andprecise?A case study forwheatproduction inDenmarkAgric. ForestMeteorol.202 51–60

NelsonG et al 2014Climate change effects on agriculture: economicresponses to biophysical shocks Proc. Natl Acad. Sci. USA 1113274–9

Nickell S 1985 Error correction, partial adjustment and all that: anexpository noteOxford Bull. Econ. Stat. 47 119–29

Ortiz-Bobea A 2013Understanding temperature andmoistureinteractions in the economics of climate change impacts andadaptation on agricultureAgricultural andApplied EconomicsAssociation AnnualMeeting (Washington, DC, 4–6August)

Ortiz-Bobea A and Just RE 2012Modeling the structure ofadaptation in climate change impact assessmentAm. J. Agric.Econ. 95 244–51

Paltsev S, Reilly J, JacobyH, Eckaus R,McFarland J, SarofimM,AsadoorianMandBabikerM2005TheMIT emissionsprediction and policy analysis (EPPA)model: version 4MITJoint Program on the Science and Policy of Global ClimateChange Report No. 125

Paltsev S,Monier E, Scott J, Sokolov A andReilly J 2013 Integratedeconomic and climate projections for impact assessmentClim. Change 131 21–33

Portmann FT, Siebert S andDoll P 2010MIRCA2000–globalmonthly irrigated and rainfed crop areas around the year2000: a newhigh-resolution data set for agricultural andhydrologicalmodelingGlob. Biogeochem. Cycles 24GB1011

Rosenzweig C et al 2014Assessing agricultural risks of climatechange in the 21st century in a global gridded cropmodelintercomparison Proc. Natl Acad. Sci. USA 111 3268–73

SchlenkerWandRobertsM J 2009Nonlinear temperature effectsindicate severe damages toUS crop yields under climatechangeProc. Natl Acad. Sci. USA 106 15594–8

SchlenkerW,HanemannWMand Fisher AC 2006The impactof global warming onUS agriculture: an econometricanalysis of optimal growing conditionsRev. Econ. Stat. 88113–25

SueWing I and Fisher-VandenK 2013Confronting the challenge ofintegrated assessment of climate adaptation: a conceptualframeworkClim. Change 117 497–514

Tack J, Barkley A andNalley L L 2015 Effect of warmingtemperatures onUSwheat yieldsProc. Natl Acad. Sci. USA112 6931–6

Waldhoff S T,Martinich J, SarofimM,DeAngelo B,McFarland J,Jantarasami L, Shouse K, Crimmins A,Ohrel S and Li J 2014Overview of the special issue: amulti-model framework toachieve consistent evaluation of climate change impacts intheUnited StatesClim. Change 131 1–20

12

Environ. Res. Lett. 10 (2015) 115002 I SueWing et al