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

Download details:

IP Address: 140.247.105.215

This content was downloaded on 18/12/2015 at 20:04

Please note that terms and conditions apply.

Variations in the sensitivity of US maize yield to extreme temperatures by region and growth

phase

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

2015 Environ. Res. Lett. 10 034009

(http://iopscience.iop.org/1748-9326/10/3/034009)

Home Search Collections Journals About Contact us My IOPscience

Environ. Res. Lett. 10 (2015) 034009 doi:10.1088/1748-9326/10/3/034009

LETTER

Variations in the sensitivity of USmaize yield to extremetemperatures by region and growth phase

EthanEButler andPeterHuybersDepartment of Earth andPlanetary Science,HarvardUniversity, 20Oxford St. Cambridge,MA02138USA

E-mail: eebutler@fas.harvard.edu

Keywords: agriculture,maize, climate change, adaptation, phenology

Supplementarymaterial for this article is available online

AbstractMaize yield is sensitive to high temperatures, andmost large scale analyses have used a single,fixedsensitivity to represent this vulnerability over the course of a growing season. Field scale studies, incontrast, highlight how temperature sensitivity varies over the course of development. Here we coupleUnited StatesDepartment of Agriculture yield and development data from1981–2012withweatherstation data to resolve temperature sensitivity according to both region and growth interval. Onaverage, temperature sensitivity peaks during silking and grainfilling, but there aremajor regionalvariations. InNorthern states grainfilling phases are shorter when temperatures are higher, whereasSouthern states show little yield sensitivity and have longer grainfilling phases during hotter seasons.This pattern of grainfilling sensitivity and duration accords with thewhole-season temperaturesensitivity inUSmaize identified in recent studies. Further exploration of grainfilling duration and itsresponse to high temperaturesmay be useful in determining the degree towhichmaize agriculture canbe adapted to a hotter climate.

1. Introduction

Most large-scale empirical studies of the effects ofextreme temperatures upon maize yield haveemployed a single, fixed sensitivity [1–7]. A moreresolved analysis is useful, however, because sensitivityvaries substantially across development phases, andaccounting for these variations permits for betterdiscernment of physiological controls and quantifica-tion of the relationship betweenweather and yield.

Early field-scale work [8] demonstrated that thesensitivity of maize yield to temperature peaks at avalue approximately three times above the averagearound 90 days after planting, during silking. Sincethat time, many further field-scale studies have con-firmed that maize yield is particularly sensitive to ele-vated temperatures during silking as well as grainfilling [9]. Several large-scale empirical analyses havealso analyzed sensitivity during portions of maizedevelopment. For example, moderate sensitivity tohigh temperatures prior to silking, exceptional sensi-tivity during silking, and increased yields with elevatedtemperatures after the silking period were

demonstrated for sub-Saharan maize yields [10].Similarly, sensitivity to high temperatures during earlyreproductive stages has been demonstrated for USmaize [11–13].

Complementary to temporal variation in sensitiv-ity is to explore regional variations in sensitivity. Trialson various cultivars show that those planted in theSouth produce more heat shock proteins, lose lesswater, and have morphologies better adapted to hotenvironments relative to those typically planted in theNorth [14]. It has also been shown that temperate cul-tivars accelerate development in response to high tem-peratures more so than tropical varieties [15, 16].Consistent with these variations amongst cultivars,maize has been found to be more sensitive to hightemperatures in theUSNorth than South [7, 17].

Also suggested is that the observed spatial varia-tions in sensitivity may permit for inference of adapt-ability to future increases in temperature [7], thoughthe appropriateness of such an inference depends onthe basis for present spatial variations in sensitivity,and whether these qualities could be advantageouslyimported to new regions given warmer conditions

OPEN ACCESS

RECEIVED

9October 2014

REVISED

6 February 2015

ACCEPTED FOR PUBLICATION

10 February 2015

PUBLISHED

5March 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

[17–19]. Here we focus on identifying physiologicadaptations to high temperatures in rainfedmaize thatmay underlie variations in spatial sensitivity.

2.Methods

We first characterize spatial variation in temperaturesensitivity using a model with a single fixed sensitivityover the course of the entire growing season, similar tothe approach inmany previous studies [1–7], and thenintroduce a technique to incorporate developmentaldata into a regression model in order to explore howtemperature sensitivity varies through the growingseason.

As input to the regression model we use develop-ment data from the United States Department of Agri-culture/National Agriculture Statistics service1 andtemperature data from the United States HistoricalClimatology Network weather stations [20]. Develop-ment data are available for 17 states within the EasternUnited States, all extending from 1981–2012 exceptfor Georgia (1981–1999) and Texas (1985–2012). Vir-ginia, Tennessee, and North Dakota also have devel-opment data but are omitted because less than 15 yearsof data makes reliable parameter estimation difficult.Temperature data are maximum daily temperature,denoted T dmax, , and minimum daily temperature,T dmin, . These daily temperature values are interpolatedfrom a subset of 444 weather stations using a DelaunayTriangulation [21] to the center point associated witheach county contained within the 17 states having suf-ficient development data

Heavily irrigated counties are excluded from ouranalysis because irrigation significantly reduces tem-perature sensitivity [7] and to ensure a more homo-geneous sample. Specifically, irrigation data areaveraged across three available census years (1997,2002, 2007) and counties whose average irrigationexceeds 10% of its total harvested area are excluded.Colorado is removed entirely for having only a singleunirrigated county with a sufficiently long record ofmaize planting, bringing the total number of statesanalyzed to 16.

We use growing degree days (GDD) as an estimateof the beneficial effects of temperature. GDDs are typi-cally used as a measure of the thermal time requiredfor a specific cultivar to develop, but in this aggregateanalysis there are many maturity classes within anygiven state on any given year, and yearly GDDs helpdetermine which of those cultivars are most success-ful. This approach is in keeping with previous aggre-gate statistical studies [7, 10, 22]. The daily heat unit,GDDd, is defined on each day, d, using the representa-tion of [23]:

=+

−T T

TGDD2

, (1)dd dmin,

*max,*

low

where,

=< <

⩽⩾

T

T T T T

T T T

T T T

if ,

if ,

if .d

d d

d

d

max,*

max, low max, high

low max, low

high max, high

⎧⎨⎪

⎩⎪

T dmin,* is defined with analogous bounds. Similarly,

damaging heat units, killing degree days (KDDs), areused to quantify temperatures that may reduce yields,for example, through desiccation or accelerated devel-opment, and are defined as,

=− >

⩽T T T T

T TKDD

if ,

0 if .(2)d

d d

d

max, high max, high

max, high

⎪

⎪

⎧⎨⎩

In the above, Tlow is set to 9 °C and Thigh to 29 °C,similar to typical values for GDDs [23]. A Thigh valueof °29 C for damaging temperatures is consistent withprevious statistical studies of the influence of hightemperature on yield [2, 5–7, 10], though is notablycooler than thresholds established for protein dena-turing (≈ °45 C) [24] or photosynthetic inhibition(≈ °38 C) [25]. This discrepancy can be explained inthat KDDs represent many negative effects of hightemperature, which begin to accrue at temperaturesjust above the optimum. It is also noteworthy thatdiscrepancies can exist between ambient air tempera-ture—for which data are widely available—and cropcanopy temperature [26]. We also experimented withincluding freezing days, precipitation, and potentialsunshine hours in the model but these were ultimatelyrejected for adding little explanatory power relative tothe increased number of free parameters, particularlyin that our focus is on variations in the sensitivity totemperature.

Summing GDDd and KDDd across each year’sgrowing season and removing the sample mean acrossall years, y, gives anomalies in accumulated tempera-ture measures, GDD′y and KDD′y, with which we for-mulate a panel regressionmodel for the yield,

β β β

β ϵ

= + + ′

+ ′ +Y y GDD

KDD . (3)

y c c i i y c

i y c y c

, 0, 1, 2, ,

3, , ,

Yy c, represents the yield in county c and year yexpressed in metric tons per hectare (t/ha). The β c0,

term is a county dependent intercept, whereas other βterms are uniform across each state, i. The inclusion ofa state-wide linear time trend sensitivity, β i1, , accountsfor technological improvement over the study period,1981–2012. Overall positive yield trends are a result ofboth cultivar andmanagement improvement, with thegreatest increase generally attributed to greater standdensities, which have increased by approximately 1000plants per hectare per year [27–29]. In addition, kernelweight has increased over the course of plant breedingand contributes to the higher yields of modernvarieties [30]. Although the estimated β1, β2, and β3

1Of Agriculture/National Agriculture Statistics Service U S D

QuickstatsURL http://quickstats.nass.usda.gov/

2

Environ. Res. Lett. 10 (2015) 034009 EEButler and PHuybers

would not change if the magnitude of GDD and KDDwere used instead of anomalies, the value of β0 wouldthen not be interpretable as the mean county yield.This model is similar to that used in other recentstudies [6, 22], though here it is employed to estimatespatial variation in parameter estimates, as in [7].

Resolving temporal as well as spatial variations inyield sensitivity requires additional analysis. First, wedevelop and employ a technique to combine state-level development data with county-level yields andweather station temperatures. The United StatesDepartment of Agriculture/National Agriculture Sta-tistics Service development data indicates when cropspass through six distinct developmental stages: plant-ing, silking, doughing, dented, mature, and harvested.These stages are used to define four phases of maizedevelopment: (1) planting to silking is collectivelyreferred to as the vegetative phase; (2) silking todoughing is the early grain filling phase, (3) doughingtomature is the late grain filling phase; and (4) matureto harvested is the drydown phase. Note that phasethree encompasses passage through the dented devel-opmental stage, making the early and late grain fillingphases of similar duration. Although yield is largelybiologically insensitive to most environmental stressesduring phase 4, the model is ultimately constrained byactual yields as reported, and omitting possible influ-ences such as sufficiently high temperature for cropdrydown, which can influence harvesting, could intro-duce biases in the parameters inferred for the otherphases.

Developmental values are reported as percentagesof total acreage having attained a particular stage on aweekly basis.We linearly interpolate these data to dailyvalues and, if the data do not cover 0 or 100%, linearlyextrapolate to these end values in the adjacent weeks.Cumulative distributions of development stage areconverted into instantaneous daily fractions by sub-tracting the total percentile of acreage in the followingstage of development and dividing by a factor of 100,

= − +P C C( ) 100.p d s d s d, , 1, Pp d, represents the frac-tion of the planted area in each state within each phase,p, on day, d (figure 1). Note that crops across a statecan be in several phases on any given day, and inmanystates this is indeed the case during the middle of thegrowing season.

Incorporating this detailed temporal informationon growing phases is important because of the varia-bility in both the seasonal cycle of temperature andcropping calendars. Planting has shifted about twoweeks earlier across the US corn belt over the last threedecades [31], and Southern maize is typically plantedabout a month earlier than in the North. There is alsosubstantial interannual variability in the timing andamplitude of the seasonal cycle in temperature, as wellas evidence for a general shift toward earlier seasons bya few days over the last century [32]. The combinedeffects of variations in the seasonal cycle and planting

date permit for substantial changes in the tempera-tures that a cropmay experience during a given growthphase.

To incorporate development data into predictorvariables, daily GDD is weighted by the fraction ofcrop in a given phase, = ∑ PGDD GDDp d p d d, , wherethe sum is over the days comprising a given growthperiod as a function of state and year. KDDp is calcu-lated analogously. GDDp and KDDp are then used aspredictors for county yield in a multiple linear regres-sion

∑β β β

β ϵ

= + + ′

+ ′ +

=(

)

Y y GDD

KDD . (4)

y c c ip

i p y c p

i p y c p y c

, 0, 1,1

4

2, , , ,

3, , , , ,

Here, as in equation (3), the β terms are defined foreach state, i, as dictated by the scale at which thedevelopment data are reported, and values are esti-mated using ordinary least squares. As withequation (3), primes on GDD and KDD indicate thatthe mean has been removed to prevent interactionwith the β c0, term. These model results have a meancoefficient of determination of 0.69, with generallybetter fits in the North than the South and, of course,superior fits than the single season sensitivity (figureS1). We note that it is also possible to select a differentThigh for each growing phase [9] within each state, butthat such an approachwould introduce a large numberof adjustable parameters that would be partiallyredundant with the sensitivity parameters.

The use of specific crop phase dates for definingGDD and KDD is critical to the performance of ourmodel. The supplemental material, available at stacks.iop.org/erl/10/034009/mmedia, contains a counter-factual example where phase dates are fixed to theiraverage times across years, whereupon the inferred

Figure 1. The fraction of planted area in Iowa in each offour development phases. Solid lines show themean devel-opmental phase distributions across all reported years, dashedlines are for the earliest planting date (March 26th, 2012), anddot-dashed lines are for the latest planting date (April 21st,2008), nearly a fullmonth later. Iowa is the largest producerofmaize in theUS, with over a third of the state’s areadedicated tomaize cultivation.

3

Environ. Res. Lett. 10 (2015) 034009 EEButler and PHuybers

sensitivities are unphysical, and the explained varianceis reduced. Also see [11–13] for further discussionregarding the importance of correctly resolvinggrowth stages in considering yield sensitivity.

Generally, the sensitivity to GDDs are found tohave a positive coefficient across each phase and state,whereas KDDs have a negative coefficient across states(figure 2). As each state contains various cultivars hav-ing different GDDmaturity ratings, the generally posi-tive coefficient of GDD is interpreted as reflectinggreater yields for longer maturing varieties [33]. Thatan increase in GDD may further benefit yield is alsolikely related to the degree to which carbon uptake,chlorophyll production, and growth may be inhibitedat lower temperatures [9, 34, 35] as well as the fact thatGDDs and mean temperature are strongly correlated.Specifically, the correlation between daily GDDs andthe daily average temperature calculated for eachcounty included in our analysis gives an average Pear-son’s coefficient of r = 0.94. Also in keeping with theinferred sign of the response, KDDs are expected todamage yield through desiccation, accelerated devel-opment, and, at especially high temperatures, tissueand protein damage [15, 24, 25].

Uncertainties associated with βi p, sensitivities are

calculated using bootstrap resampling. Each county isresampled 1000 times using individual years as inde-pendent replicates. In addition, we perform a moreconservative block resampling in which all countieswithin a given state and year are resampled together asa unit. This reduces the effective degrees of freedom bya factor of nearly 100 and provides an upper estimatefor the uncertainty (figure S2), but in this case GDDand KDD sensitivity in many state and stage combina-tions are difficult to distinguish, which runs counter tothe the overall consistency observed between statesand physiological expectation. County based

bootstrap estimates are, therefore, used for citeduncertainties.

3. Results

Each phase of development represents a unique periodof the maize plant’s phenological cycle and entails adifferent response to environmental conditions. Theoverall life cycle is divided between vegetative andreproductive phases, with roughly half of each growingseason spent in each. During the vegetative phase aseedling expands its surface area and the leaf area indexof the plant is set, influencing the amount of radiationthat can be intercepted over the remainder of itslifecycle [23, 36]. Sensitivity to both KDDs and GDDsare low relative to other stages (see figure 2 and tableS1), though Texas shows pronounced sensitivity to

KDDs at −0.017 ((t/ha)/(°C day)) (95% c.i. −0.019 to−0.016), possibly because overall hotter temperaturesmake seedlings particularly vulnerable [37].

The early grain filling phase, as defined here,begins with the emergence of silks from the ends of theears and begins the reproductive portion of the plants’life cycle. As has long been recognized [8, 38], yieldsare highly susceptible to elevated temperatures duringsilking and early grain filling [9, 10, 12, 30, 39, 40].Indeed, we find that average sensitivity to KDDs acrossstates during this phase is a factor of four greater than

during the vegetative phase at −0.025 ((t/ha)/(°Cday)) (95% c.i. −0.024 to −0.026). Sensitivity to hightemperatures results from damage to silks and fewerkernels being fertilized or other factors influencingkernel viability [41–44]. This phase also accounts forabout 20%of final kernel drymass [23] and, therefore,entails responses similar to those found during lategrain filling, discussed below.

Figure 2. Yield sensitivity to growing and killing degree days.Phase sensitivity is indicated by shadingwithin thewheelscorresponding to: (1) vegetative, (2) early grainfilling, (3) late grainfilling, and (4) drydown. Thewhole season sensitivity is indicatedby the background shading of each state. (a) Growing degree day (GDD) sensitivity is largest in the northern states during early andlate grain filling. (b) Killing degree day (KDD) sensitivity is generally largest during early grain filling, but late grain filling is of acomparablemagnitude in theNorth.

4

Environ. Res. Lett. 10 (2015) 034009 EEButler and PHuybers

Late grain filling is the combination of the dough-ing and dented stages, and is when the majority ofphotosynthate and carbon reserves are transferredinto the growing kernels. At this point, yield damagesare generally a result of lower final kernel mass ratherthan lower total kernel number. High temperaturesare known to directly reduce kernel mass [45, 46],consistent with the high sensitivity we find. In addi-tion, high temperatures can lead to accelerationthrough the growth phase, reducing the total numberof days of grain filling and lowering kernel mass [47–53]. On average, yield sensitivity during late grain fill-ing is smaller than during early grain filling, withmean

losses from KDDs of −0.014 ((t/ha)/(°C day)) (95% c.i. −0.013 to −0.015). Wisconsin has the greatest sensi-

tivity with −0.05 ((t/ha)/(°C day)) (95% c.i. −0.04 to−0.06) losses and Texas is the least vulnerable to KDDs

at 0.006 ((t/ha)/(°Cday)) (95% c.i. 0.004 to 0.008).With regard to GDDs during late grain filling, the

greatest increase in yield occurs in Minnesota with

values of 0.015 ((t/ha)/(°C day)) (95% c.i. 0.014 to

0.016). The mean increase is 0.0055 ((t/ha)/(°C day))(95% c.i. 0.005 to 0.006), consistent with cultivarsthat are able to take advantage of greater accumula-tion of moderate temperatures and longer period ofradiation interception having a higher yield[45, 47, 54]. The fact that Texas has a positive lategrain filling KDD sensitivity estimate and a negativeGDD sensitivity may be because collinearity of thepredictors confounds some of the results, similar toIowa’s negative GDD sensitivity during early grainfilling. Note that whereas high KDDs usually shortenthe duration of grain filling, the total accumulatedGDDs are still higher in years with higher KDDs, con-sistent with previous work on development underhotter conditions [53].

The end of the plant’s life cycle is the drydownphase, when the plant ceases to photosynthesize andkernels lose water, generally drying from just over 30%moisture to about 20% moisture when ready for har-vest. Sensitivity parameters indicate a positive effectfrom KDDs in many northern states, possibly becauseinsufficient drydown ismost probable there, with con-sequences for vulnerability to pests, mold, or otherpre-harvest losses. Statistical studies including but notindependently resolving the drydown phase pre-sumably conflate the positive effects of KDDs duringdrydown with negative effects during grain develop-ment. Some states in the eastern corn belt still indicatestrong GDD sensitivity during the drydown phasepossibly because the late grain filling interval in theUSDA data extends into the drydown phase, or thatthe higher GDDs play a similar role to KDDs in pro-moting drying. Altogether, these results highlight thevariability of the temperature response according todevelopmental phase.

It is useful to make a direct comparison betweenwhole-season sensitivity and the phase-based

sensitivity estimates. Whole season sensitivity impli-citly sums over the product of the distribution ofKDDs (or GDDs) and sensitivity throughout thegrowing season. A development phase that was highlysensitive but experienced no KDDs, for example,would not contribute to whole-season KDD sensitiv-ity. Thus, in order to provide a direct comparison ofwhole-season sensitivity to phase-based sensitivity, weweight each phase according to the fraction of KDDsthat it accounts for and estimate the slope between thetwo measures. Slope is estimated using a York fit [55],which has the advantage of accounting for uncertain-ties in both the ordinate and abscissa.

The best-estimate slope between weighted dry-down-phase KDD sensitivity and whole-season KDDsensitivity is weak and negative at −0.1 (95% c.i., −0.2to 0.1), consistent with the foregoing discussion thatKDDs can be beneficial for drydown in the mostNorthern states, whereas they are typically detri-mental during other growth phases. The slope for thevegetative phase is also negative and weak at −0.2, aswell as uncertain (95% c.i., −0.5 to 1.1), consistentwith generally low KDD sensitivity during this phase.Of greater pertinence is that the slope between KDDsensitivity for whole season and early grain fill is 1.1(95% c.i., 0.5 to 2.8), indicating that the spatial varia-tion observed over the whole season are substantiallydetermined by variations in early grain filling (figureS3). Similarly, the slope with late grain fill is 0.6 (95%c.i., −1.2 to 1.7) indicating a further correspondence(figure S3). Together, the grain filling phases ofdevelopment account for the observed pattern of sen-sitivity to KDD, and are the focus of attentionhereafter.

The estimated spatial variations in grain fillingsensitivity are consistent with field trials where theresponses of temperate and tropical maize to heatexposure were examined prior to and through earlyreproductive development [15]. In these trials, well-watered temperate and tropical varieties gave similaryields under normal temperature conditions, butwhen exposed to higher temperatures during what waslikely the late blister, milking, and doughing stages, thetemperate varieties responded by accelerating throughgrain filling and having larger yield reductions. Speci-fically, temperate varieties showed a 21 day decrease inthe duration of the reproductive period as a con-sequence of heating, compared to afive day decrease intropical hybrids and no decrease in temperate–tropi-cal hybrids. Furthermore, temperate varieties had agreater decline in conversion of radiation to biomasswhen exposed to higher temperatures during earlygrain filling.

The USDA development data permits for anexamination of changes in the duration of grain fillingin response to high temperatures across the EasternUS.We find that the durational response to KDDs clo-sely follows the number of KDDs experienced in agiven state. States whose climatological average KDDs

5

Environ. Res. Lett. 10 (2015) 034009 EEButler and PHuybers

during early grain filling are below the median (<37KDDs forMI,WI,MN, PA, IA, OH, SD, and IN) showan average decrease in the duration of early grain fill-ing by 0.04 days per KDD. States clustered in the thirdquartile of exposure to KDDs ( < <37 KDD 66; IL,NE, KY, andMO) show a shortening of the early grainfilling phase by 0.01 days per KDD. And three out offour states in the upper quartile (>66 KDDs; NC, GA,TX, but not KS) show the opposite sign of response, onaverage across the four states, increasing the durationof early grain filling by 0.09 days per KDD(figure 3(a)). Similar to early grain filling, late grainfilling at the low end of the KDD climatologyshows an average decrease in duration of 0.12 daysper KDD(<50 KDDs); those at the mid-range( < <50 KDD 85) show a weaker decrease at 0.06days per KDD; and those at the highest range (>85KDDs) show an increase in duration of 0.04 days perKDD (figure 3(b)). Although the climatological valuesof KDDs differ between early and late grain filling, thesame states are kept in each climatological category.

Changes in the duration of grain filling are pri-marily associated with interannual variability, asopposed to long term trends. The detrended length ofearly grain filling has an average standard deviationacross states of 3 days, whereas the average trend istoward lengthening at 0.16 days/year. Similarly, lategrain filling varies by nearly 4 days and has a lengthen-ing trend of only 0.05 days/year. Standard deviationsare similarly larger than trends in just the three South-ern states with elongated grain filling phases duringhotter seasons. These results strongly suggest thatsteady technological changes are not the source of theKDD-duration relationship, and are consistent withvariations in the duration of grain filling as features ofgiven cultivars.

The variations of grain filling duration as aresponse to high temperatures found here generally

accord with the aforementioned field trial results [15],presuming that hybrids in the South are most similarto those of tropical or temperate–tropical origin. Thatboth early and late grain filling phases indicate similarpatterns with respect to climatological KDDs also sug-gests that these results are robust. Combining earlyand late grain filling into a single phase gives a similarresult but with smaller fractional changes in durationbecause high KDDs do not generally persist equallyacross the early and late phases.

4.Discussion and conclusion

Longer grain filling with hotter temperatures can beassociated with lower sensitivity to high temperatureson account of pattern correspondence between accel-eration and sensitivity across the US, similar responsesseen in field trials, and on the physiological basis oflonger duration leading to greater kernel mass, so longas crops remain adequately watered [15, 29]. Thephysiological pathway by which temperature variationinfluences the timing of the transition out of grainfilling is unclear, though evidence for the role ofreaching a critical moisture content [56] suggests thatincreased drying associated with higher temperaturesand higher vapor pressure deficits would lead to anearlier transition, as observed in the North. Refiningour understanding of how temperature influencesgrain filling duration, especially in the South and intropical varieties, may lead to a better understandingof how to reduce damages from a hotter climate.

Correspondence between KDD sensitivity and cli-matology over the whole season was identified in anearlier study and offered as a basis for inferring theadaptability of cultivars to higher temperatures [7].The association of yield sensitivity to KDDs and varia-tions in grain filling duration documented here (alsosee figure S4) provides a more physiologically based

Figure 3. Relationship between killing degree days and duration of grainfilling.Northern states show a decrease in the duration ofboth (a) early grain filling and (b) late grainfilling in seasonswith greater KDDs, whereas Southern states show an increase in duration.These regional variations in durationmay represent a physiological trait important in determining differential sensitivities toKDDs.For readability, the axes for early grainfilling truncate themost extreme values for Georgia, Texas, andNorthCarolina

6

Environ. Res. Lett. 10 (2015) 034009 EEButler and PHuybers

perspective on patterns of KDD sensitivity. An impor-tant question, however, is why has amodification suchas increasing the duration of grain filling in response tohigher temperature not yet been introduced intoNorthern states if it reduces yield losses? More generalversions of this question have been raised elsewhere[17], and one possibility is that such a modificationonly becomes physiologically feasible under warmerconditions, possibly because of a longer growing sea-son [12, 18]. In this speculative case, lengthening ofgrain fill duration in response to higher temperatureswould constitute a climate change adaptation in thesense articulated by [19]. The question remains, ofcourse, as to whether elongated duration of Southerngrain filling would be transferrable to Northern culti-vars under conditions of a warming climate and whe-ther it would prove effective in mitigating yield lossesthatwould otherwise accrue.

Building from previous controlled field-scale stu-dies of yield loss in the presence of altered temperature[15, 16] appears useful for furthering our under-standing of potential adaptation to climate change.Field trials could be conducted to determine whetherSouthern varieties could be advantageously grown atNorthern latitudes under higher temperatures, or,more elaborately, Northern cultivars could be simi-larly tested after breeding for longer duration of grainfill in response to higher temperatures.

The risks posed by climate change for agriculturalproduction remain uncertain, as highlighted by thefact that the Fourth Intergovernmental Panel on Cli-mate Change Assessment Report [57] suggested anincrease in maize production under conditions ofmoderate warming with adaptation, whereas the mostrecent Fifth Assessment Report [58] suggested losses.This variability appears to result from the ensemble ofavailable studies having substantial spread, and thatsubtle shifts in modeling frameworks can have sig-nificant influences on individual results. It is suggestedthat increasing the physiological interpretability ofsimple statistical models used to explore large-scalechanges in productionmay help to reduce some of thisambiguity through better bridging their results withmore complete agricultural models [59] and facilitat-ing testing of results againstfield-scale trials.

Acknowledgments

Comments that led to improvements in this manu-script were provided by Nathan Mueller, N MichelleHolbrook, an anonymous editor, and an anonymousreviewer. Funding was provided by the PackardFoundation andNSF grant 1304309.

References

[1] Thompson L 1975 Science 188 535–41[2] CarlsonR 1990 J. Prod. Agric. 3 293–7[3] SchlenkerWandRobertsM2006Rev. Agric. Econ. 28 391–8

[4] Lobell D and Field C 2007Environ. Res. Lett. 2 014002[5] SchlenkerWandRobertsM2009Proc. Natl Acad. Sci. USA

106 15594–8[6] Hawkins E, Fricker T, Challinor A, FerroC,HoC and

Osborne T 2013Glob. Change Biol. 19 937–47[7] Butler E andHuybers P 2013Nat. Clim. Change 3 68–72[8] de Bruyn L and de Jager J 1978Agric.Meteorol. 19 35–40[9] Sánchez B, Rasmussen A and Porter J 2014Glob. Change Biol.

20 408–17[10] Lobell DB, BänzigerM,MagorokoshoC andVivek B 2011

Nat. Clim. Change 1 42–45[11] Ortiz-Bobea A 2011 Improving agronomic structure in

econometricmodels of climate change impacts, presented attheAgricultural andApplied Economics Association’s 2011AAEA andNAREA Joint AnnualMeeting unpublished

[12] Ortiz-Bobea A and Just R 2012Am. J. Agric. Econ. 95 244–51[13] Berry S T, RobertsM J and SchlenkerW2013Corn Production

Shocks in 2012 and Beyond: Implications forHarvest Volatility(Chicage, IL: University of Chicago Press)

[14] Ristic Z,WilliamsG, YangG,Martin B and Fullerton S 1996J. Plant Physiol. 149 424–32

[15] Rattalino Edreira J andOteguiM2012 Field Crops Res. 13097–98

[16] Rattalino Edreira J andOteguiM2012 Field Crops Res. 14258–67

[17] SchlenkerW, RobertsM and Lobell D 2013Nat. Clim. Change3 690–1

[18] Butler E andHuybers P 2013Nat. Clim. Change 3 691–2[19] Lobell D 2014Glob. Food Secur. 3 72–6[20] MenneM J,WilliamsCN Jr andVose R S 2014United States

Historical ClimatologyNetworkDaily Temperature, Precipita-tion, and SnowData Tech. rep. (OakRidgeNational Labora-tory: CarbonDioxide InformationAnalysis Center)

[21] Boissonnat J D, DevillersO, Pion S, TeillaudMandYvinecM2002Computational Geometry pp 5–19 as implemented inMatlab 2010b

[22] DiffenbaughN,Hertel T, SchererM andVermaM2012Nat.Clim. Change 2 514–8

[23] Abendroth L, Elmore R, BoyerMandMarlay S 2011CornGrowth andDevelopment PMR1009 (Ames, IA: Iowa StateUniversity Extension)

[24] Taiz L andZeiger E 2002Plant Physiology 3rd edn (Sunderland,MA: Sinauer Associates)

[25] Crafts-Brandner S J and SalvucciM2002Plant Physiol. 1291773–80

[26] Gardner B, Blad B andWattsD 1981Agric.Meteorol. 25207–17

[27] Gifford R, Thorne J, HitzWandGiaquinta R 1984 Science 225801–8

[28] TollenaarM andWu J 1999Crop Sci. 39 1597–604[29] DuvickD2005Maydica 50 193–202[30] HayR and Porter J R 2006The Physiology of Crop Yield 2nd edn

(Oxford: Blackwell)[31] Kucharik C 2006Agron. J. 98 1544–50[32] Stine A,Huybers P and Fung I 2009Nature 457 435–40[33] TheCorn andClimate Report JordanB, KluckD,MockD,

Schulz P, TakleG andWash S (ed) (Ames, IA: Iowa StateUniversity)

[34] McWilliam J andNaylor A 1967Plant Physiol. 42 1711–5[35] Sage R andKubienD2007Plant Cell Environ. 30 1086–106[36] Boyer J 1970Plant Physiol. 46 233–5[37] Guimarães P d S, Bernini C S, Pedroso FK and PaternianiME

2014Maydica 59 72–79[38] SmithW1914Mon.Weather Rev. 42 78–94[39] Gourdji S, Sibley A and Lobell D 2013Environ. Res. Lett. 8 p10[40] Lobell D, RobertsM, SchlenkerW, BraunN, Little B,

Rejesus R andHammerG 2014 Science 344 516–9[41] Andrade F,OteguiM andVegaC 2000Agron. J. 92 92–97[42] Cárcova J andOteguiM2001Crop Sci. 41 1809–15[43] Craufurd P andWheeler T 2009 J. Exp. Bot. 60 2529–39[44] CicchinoM, Rattalino Edreira J andOteguiM2010Crop Sci.

50 1431–7[45] Jones R,Ouattar S andCrookstonRK1984Crop Sci. 2 133–7

7

Environ. Res. Lett. 10 (2015) 034009 EEButler and PHuybers

[46] Singletary G, Banisadr R andKeeling P 1994Aust. J. PlantPhysiol. 21 829–41

[47] Cavalieri A and SmithO 1984Crop Sci. 25 856–60[48] Borrás L, Slafer G andOteguiM2004 Field Crops Res. 86

131–46[49] Egli D 2004 Seed-fill duration and yield of grain cropsAdvances

in Agronomy vol 83 (Amsterdam: Elsevier)[50] Badu-Apraku B,Hunter R andTollenaarM1983Can. J. Plant

Sci. 63 357–63[51] TollenaarMandBruulesmaT1988Can. J. Plant Sci.68935–40[52] MuchowR, Sinclair T andBennett J 1990Agron. J. 82 338–43[53] WilhelmE,Mullen R, Keeling P and Singletary G 1999Crop

Sci. 39 1733–41[54] Maddonni G,OteguiM andBonhommeR1998 Field Crops

Res. 56 257–64[55] YorkD, EvensenNM,MartínezML andDelgado JDB 2004

Am. J. Phys. 72 367–75

[56] Gambín B, Borrás L andOteguiM2007 Field Crops Res. 1011–9

[57] EasterlingWE et al 2007 Food, fibre and forest productsClimate Change 2007: Impacts, Adaptation andVulnerability.Contribution ofWorking Group II to the Fourth AssessmentReport of the Intergovernmental Panel on Climate ChangeedMLParry et al (Cambridge: CambridgeUniversity Press)pp 273–313

[58] Porter J R, Xie L, Challinor A J, CochraneK,Howden SM,IqbalMM, Lobell DB andTravassoM I 2014 Food securityand food production systems.Climate Change 2014: Impacts,Adaptation andVulnerability Part A: Global and SectoralAspects. Contribution ofWorkingGroup II to the Fifth Assess-ment Report of the Intergovernmental Panel on Climate ChangeedCB Field et al (Cambridge: CambridgeUniversity Press)pp 485–533

[59] Rosenzweig C et al 2014Proc. Natl Acad. Sci. USA 111 3268–73

8

Environ. Res. Lett. 10 (2015) 034009 EEButler and PHuybers