Little evidence of avoided yield loss in US corn when short-term forecasts correctly predict extreme heat

Successful agricultural adaptation to extreme heat has the potential to avert large crop losses and improve food security. Because adaptation is costly, accurate weather forecasts have the potential to improve targeting of adaptation efforts. To understand the role of short-term (1–7 day) forecasts in reducing heat-related yield loss, we analyze a novel dataset combining corn yields, short-term weather forecasts, and weather realizations in the United States from 2008 to 2021. We find no evidence that forecasts facilitate avoidance of heat-related yield losses on average, and only limited benefits when we allow for forecast benefits to vary with irrigation prevalence. While our results paint a pessimistic picture of in-season adaptation to heat, forecasts may be more valuable for other crops and regions, especially given continuing investment in adaptation technologies.


Introduction
Exposure to high temperatures is known to reduce crop yields (Deschênes and Greenstone 2007, Schlenker and Roberts 2009, Lobell et al 2013, Challinor et al 2014, Tack et al 2015, Lesk et al 2016) and may also depress farm labor productivity (Stevens 2017), alter labor supply and wages (Lee et al 2017, Colmer 2021), and imperil agricultural workers (Tigchelaar et al 2020), ultimately dragging down agricultural output (Dell et al 2012, Burke et al 2015, Miller et al 2021).The yield effects alone are economically significant: high temperatures accounted for $27.0 billion (19%) of crop losses in the United States between 1991and 2017(Diffenbaugh et al 2021), and climate sensitivity of U.S. agriculture is increasing in some areas (Lobell et al 2014, Ortiz-Bobea et al 2018).Adaptation can, of course, dampen these effects (Butler andHuybers 2013, Moore andLobell 2014).At the start of the season, farmers can plant heat-tolerant varietals, switch crops, or install or upgrade irrigation equipment.Mid-season, farmers may deploy shade, provide nutrients, use protective sprays, manage biotic co-stressors, or irrigate (Atkinson and Urwin 2012 Weather forecasts could play an important role in facilitating these adaptations to extreme heat.Because irrigation and other adaptation efforts are costly, forecasts of extreme heat-which can guide cost-effective deployment of adaptation strategiesmay be valuable.For example, center pivot irrigation systems, used in many agricultural areas in the US, can take several days to complete an irrigation cycle (Guerra et al 2005, Domínguez et al 2022), so that advance warning of heat is essential to effective adaptation.To this end, a number of farm technology providers bundle weather forecasts with their products.Others promise smart irrigation that uses forecasts, soil moisture readings, and other data to guide costeffective and resource-conserving water use.
Despite this potential, other factors may limit the value of heat forecasts.While forecasts of high heat have improved dramatically (Bauer et al 2015, de Perez et al 2018), farmers may still perceive forecasts as unreliable-especially at longer timescales (Crane et al 2010)-and choose not act upon them (Kusunose and Mahmood 2016).The availability of insurance could weaken economic incentives to adapt to extreme heat (Annan and Schlenker 2015).Alternatively, a given action that reduces heat damage could be so costly it is never employed, or so beneficial that it always is, so that a forecast does not alter behavior.For example, when crop water demands are highest during the summer and weather is typically hot, some farmers may already be irrigating at capacity regardless of what the short-term forecast says.On other farms not equipped for irrigation, the cost of installing a new system or adapting in other ways (e.g.sprays, dynamic shading) may simply be too high, especially for large-scale farms growing field crops.In short, for heat forecasts to have value, they must trigger a change in adaptive behavior, which is unlikely to occur if costs and benefits of adaptation are far out of balance.
These observations motivate our research question: do weather forecasts help reduce the yield losses associated with extreme heat, and if so, where and when?To this end, we statistically examine the role of short-term (one week and under) temperature forecasts in mitigating the effects of extreme heat on U.S. agriculture.We examine this question using a countylevel dataset spanning 2008-2021 combining yields from the U.S. Department of Agriculture (USDA) National Agricultural Statistical Service (NASS), daily weather data from the Parameter-elevation Regressions on Independent Slopes Model (PRISM) project, detailed land use data from USDA, and weather forecasts from the National Digital Forecast Database (NDFD).We focus our analysis on corn, which has the largest crop footprint in the U.S. and is vulnerable to extreme heat, especially during pollination (Butler andHuybers 2015, Luo et al 2023).
Both our question and approach differ from most prior work on adaptation to and impacts of extreme heat on agriculture.The numerous studies that quantify effects of heat on yield often implicitly embed adaptation efforts and, as a result, do not seek to quantify avoided losses (Schlenker and Roberts 2009) (Burke and Emerick 2016).We are similarly agnostic as to what form adaptation takes, but our interest is in the value of adaptation-enabling information, and not the adaptations themselves.There is, of course, a large literature on the utility of forecasts to farmers, but it primarily consists of ex-ante, theoretical assessments (Meza et al 2008) or ex-post studies based on smallscale surveys (Hu et al 2006, Frisvold andMurugesan 2013).Moreover, most investigations of forecast value to agriculture have focused on seasonal forecasts or precipitation (Asseng et al 2012, Gunda et al 2017, Ceglar and Toreti 2021, Toreti et al 2022); the value of short term heat forecasts is comparatively understudied.In sum, our construction and analysis of a novel, large-scale, observational dataset focused on shortterm heat forecasts provides a complementary line of evidence to these varied literatures.

Data
We constructed a panel dataset spanning 2008-2021 linking corn yield to temperature, precipitation, and weather forecasts at the county-year level in the United States.The panel combines data from NASS, PRISM, and NDFD.Construction of key variables is described next, followed by a summary of the final dataset.

Quantifying extreme heat exposure
We quantified extreme heat exposure using extreme degree days (EDDs) derived from PRISM data.As the PRISM data is reported on a 4 km grid, we first constructed daily county-level spatial averages of daily temperature minima and maxima, weighting by the fraction of each grid cell devoted to corn according to the USDA Cropland Data Layer for the relevant year.We then approximated intra-daily temperature exposure using sinusoidal interpolation between the maximum and minimum temperature for a county and day (Snyder 1985, Ortiz-Bobea et al 2019).From that interpolation, we computed EDD on each day as the area above 29 C but beneath the interpolated temperature curve (Schlenker and Roberts 2009); see SI figure S1.This approach generalizes simpler EDD measures based on maximum temperature alone: were minimum and maximum temperatures to coincide on a day, the sinusoidal, area-based approach would yield an identical contribution to EDD.Finally, we summarized county-level exposure for a year by summing EDD in a county from April through September.In other analyses, we separate EDD exposure in what we refer to as the primary silking month (when >50% of acres achieved silking over 2018-2022 according to 2023 USDA crop progress reports) to account for greater crop sensitivity during reproductive stages.See SI section 1 for details.

Quantifying correctly predicted extreme heat exposure
We quantified predicted extreme heat exposure using 1-7 day lead NDFD temperature forecasts.Specifically, we first used the same spatial aggregation, sinusoidal interpolation, and area calculations applied to realized temperature, but instead used maximum and minimum temperature forecasts.We call the resulting quantity predicted EDD, or PEDD.We then considered the minimum of EDD and PEDD to be correctly predicted extreme heat on a day, which we denote CPEDD.Formally, CPEDD id in county i on day d is defined as CPEDD id ≡ min{EDD id , PEDD id }.
This definition of CPEDD partitions EDD into two kinds of extreme heat: correctly predicted and unanticipated.Finally, we summed correctly predicted EDD in April-September and separately in the primary silking month by state, paralleling our summary of heat exposure.See SI section 1 for details.

Quantifying yield
We used yield for corn harvested for grain, measured in bushels per acre, extracted from USDA NASS.

Quantifying irrigation prevalence
We defined irrigation prevalence to be the fraction of operations in a county and year with any irrigated corn harvested for grain from USDA NASS.We used the fraction of operations employing irrigation rather than the fraction of irrigated acres because the area-based measure is only available for a small fraction (13.8%) of county-year observations, while the operations-based measure is available for the vast majority (87.4%).Further, among observations where both measures are available, the correlation is high (0.958).

Data filtering
We restricted our sample to counties with more than one year of corn yield available and included only county-years with nonzero corn yield.Inclusion of zero yield observations is precluded by use of log yields as an outcome of interest; this only removes 11 observations.Focusing on counties with more than one year of yield is immaterial; our inclusion of county-level fixed effects in our empirical specification (section 2.3) implies those observations would not contribute to estimates anyway.

Data summary
Combining these data sources produced an unbalanced panel containing 21 386 observations from 2107 unique counties across 14 years.The median extreme heat exposure was 34.1 EDD in a county and year between April and September.Extreme heat exposure was highest in the sample from 2011 to 2012, when the US also experienced major drought.With the exception of those drought years, corn yields generally trended upward, averaging 145 bushels/acre across the full sample.
Patterns of extreme heat forecasts in our data highlight several challenges for estimation (figure 1).While forecast quality is moderately high on average (66.1% correct prediction across counties and years), it varies across both space and time in systematic ways.First, counties in the South and West tend to have higher percentages of their extreme heat predicted (figure 1(a)).Because those counties may differ in other ways (including typical climate) relying on cross-sectional variation in forecast quality is unlikely to uncover a causal relationship between correct prediction of extreme heat and yield.Second, the fraction of correctly predicted EDD is far higher in June, July, and August (figure 1(b), left axis).Those months also contain the most extreme heat (figure 1(b), right axis) and are when corn is most sensitive to heat (during silking).Finally, the percent of EDD that are correctly predicted also varies within a county through time (figure 1(c)), allowing for comparison of yields within a county across years as a means to identify a potential role of heat forecasts in limiting yield loss.Our empirical strategy, which we describe next, was designed to account for all of these features.

Overall approach and estimation
We estimated a series of regression models relating log yield in a U.S. county, and year to variables measuring exposure to extreme heat, predicted extreme heat, average temperature and its square, precipitation and its square, and a suite of fixed effects.As is standard in related prior work, we used the natural log of yields as an outcome to model heat impacts in (approximate) percentage terms.Our preferred specification allowed for three types of heterogeneity in the effects of extreme heat, intended to deal with potential confounding of the effects of forecasts as motivated by figure 1.First, we interacted EDD with the long-run average temperature in a county (expressed as a difference from 26 C to render the main effect of EDD more interpretable), since selected crop varieties and practices in warmer counties may be better adapted to handle extreme heat.Second, we interacted EDD with the deviation of the current year annual average temperature from the long-run average temperature in the county.That interaction accounted for the potential that crop stress due to warmer (but not extreme) temperatures could modify the effects of extreme heat, and forecast quality could differ in those same years.Finally, we separately measured all weather-related variables (extreme heat, predicted extreme heat, average temperature, and precipitation) in two periods: (1) during the month when most silking occurs in each state, and (2) in all other months during the April-September growing season.This separate measurement was motivated by the sensitivity of corn to heat during silking and pollination, as well as the increased water demands of corn in late vegetative stages just prior to silking and pollination.
Following this logic, our baseline specification was as follows: (1) Here y it denotes log corn yield yield, CPEDD it the number of correctly predicted EDDs, and EDD it the number of EDDs in county i and year t.Superscripts of Silking and NotSilking indicate the month range over which the variable was calculated.Annual average temperature Tit and long-run average temperature Ti are interacted with EDD it to allow for heterogeneity as described above.Controls X it include average maximum temperature and its square and total precipitation and its square, each measured separately in the primary silking month in a county and other months, as well as county-specific linear and quadratic time trends.Average maximum temperature terms are included due to potential confounding: a hot day may influence both seasonal averages and EDD exposure-both of which may influence yield-and our interest is specifically in the effects of extreme heat.Fixed effects α i and ϕ t adjust for time-invariant county differences in yield and spatially invariant shocks in a given year.Our results reflect separate models for each forecast lead, with different values of CPEDD it defined using that forecast.Given the short timespan of our data, our use of county fixed effects, and the correlation of forecasts at different leads, it is unlikely we could separately identify forecast effects at different leads in the same model.We employed two-way (county and year) clustered standard errors for inference.
The parameters of interest in this specification are β S P1 and β NS P2 .Under a null hypothesis in which forecasts have no yield-related value for farmers, these parameters should be zero: either forecasts are not informative or there are no profitable yield-saving measures farmers can undertake in response.If forecasts are beneficial, we expect these coefficients to be positive.Note that any adaptations taken in response to realized weather rather than forecasts should be reflected in the coefficients associated with EDD terms.
In secondary analyses, we estimated this same specification separately on subsamples of our data.First, we defined those subsamples according to whether the county centroid is east or west of 100 W longitude to account for regional differences in farming practices.Second, we narrowed in on Kansas and Nebraska-two high production states split by that 100 W longitude line-and divided counties based on whether they fell in the interquartile range of irrigation prevalence in those states.We used a split-sample approach rather than interaction terms because we expected location and irrigation to influence not only the role of forecasts, but also other factors affecting yield.

Baseline results
Estimates from our baseline model provide no evidence that forecasts facilitate adaptation that reduces yield losses from extreme heat (figure 2).Point estimates of forecast benefits are near and not statistically different from zero, whether those forecasts occur during the primary silking month for a county (triangles) or not (circles).These general patterns are  1) an extreme degree day (EDD; gray) on corn yield in a county and year, and (2) the differential effect of an EDD if correctly predicted in advance (green) for models using different forecast leads (horizontal axis).Estimates are separated for EDD and predictions during the primary silking month (triangles) in a state vs. other (circles).Points represent mean estimates and bars 95% confidence intervals.
robust, holding when we seek to account for potential nonlinear effects of extreme heat that are not captured by EDD, drop per-county linear trends, controls for seasonal average temperature and its square, ignore intraseasonal heterogeneity in the effects of weather, or weight observations by corn acres planted.See the supplementary information (SI) for details (table

error
A potential concern noise in our measure of extreme heat is our estimates of forecast benefits, which would our baseline results misleading.In particular, examining county-wide temperatures-both forecast and realized-could miss important variation at smaller spatial scales.Specifically, localized extreme heat could get washed out via spatial averaging prior calculation of EDD.To address this concern, we reestimate our using of both EDD and correctly EDD by first computing those variper pixel in the raw temperature and forecast data, spatially averaging the pixel-level measures of extreme heat exposure.This approach yields qualitatively similar results to our baseline model (SI figure S1).

Heterogeneity
We next probe whether this absence of evidence for average forecast benefits is masking important heterogeneity, with some farms experiencing benefits, and others not.A common division used in analyses of US field crop yields is to separate counties east and west of the 100 W meridian (Schlenker and Roberts 2009, Burke andEmerick 2016, Massetti et al 2016).Farming practices are substantively different in the west, where a far higher fraction of cropland is irrigated.To this end, we estimate our model separately for counties with centroids falling east and west of the 100 W meridian.This exercise suggests there may be a limited benefit of advance warning of heat in western counties, in particular at longer lead times (figure 3).Moreover, the apparent benefits of forecasts accrue only for heat occurring outside of the silking window.We find no evidence of forecast benefits in the silking window in western counties, nor at any time of year in eastern counties.
To put these estimates in perspective, we calculated their implied avoided production loss in western counties.We asked what would have happened if all predicted EDD outside of the primary silking month in western counties had instead been unpredicted, focusing on the largest point estimate, which occurred for a model with 7-day-lead forecasts.Even under these favorable conditions, avoided losses amounted to only 0.48% of national production across the entire sample.Further, we consider this exercise to yield an optimistic take on forecast benefits.Even in the absence of federally provided forecasts, farmers still have other valuable knowledge of seasonal patterns, experience, and news of weather in other regions.As such, eliminating forecasts would be unlikely to make all extreme heat unanticipated, and the true benefit of forecasts is likely smaller.
We next probed whether these modest forecast benefits likely operated via irrigation.Economic intuition suggests that irrigation-mediated forecast benefits are largest when a farmer's beliefs about net benefits of irrigation are centered near zero but uncertain.If anticipated irrigation benefits are far larger (or smaller) than costs, a forecast is unlikely to change a farmer's choice to irrigate.While we do Figure 3. Effects of extreme heat (gray) and differential effects when that heat is correctly forecast (green).Estimates are presented for different forecast leads (horizontal axis), with separate effects for western (left panels) and eastern counties (right panels) and heat occurring during the silking period in a county (bottom panels) and not (top panels).Dots represent point estimates and bars 95% confidence intervals.
not observe farm-level beliefs about net benefits of irrigation, irrigation prevalence serves as a useful proxy.When irrigation net benefits are near zero but uncertain, we might expect moderate levels of irrigation prevalence; far higher (e.g. in California) or lower (e.g. in Minnesota) prevalence suggests either benefits or costs of irrigation dominate.Based on this logic, we reestimated our baseline specification in Nebraska and Kansas-two of the top ten producing states that exhibit a wide range of irrigation prevalence (figure 4(a)).We divided counties based on whether each county's irrigation prevalence fell between the 25th and 75th percentile among counties in those states.Based on the logic above, we expected greater forecast benefits in counties with interquartile irrigation prevalence to exhibit the greatest forecast benefits.
Estimates from this exercise suggest that any forecast benefits in those states were concentrated in counties within the interquartile range of irrigation prevalence (figure 4(b)).Further, those benefits continue to accrue outside of the primary silking month, at times of year when irrigation was not likely to be deemed as critical.While this evidence is suggestive only, it is consistent with the economic intuition described above.

Perils of naive estimates
We end by highlighting the importance of adjusting for heterogeneous impacts of extreme heat when quantifying forecast benefits, as is the case in equation ( 1).Forecasts induce heterogeneity in the effects of extreme heat which can easily be mistaken for other forms of heterogeneity if examined in isolation.To illustrate this point, a naive approach allowing for a single, average effect of EDD and a single, average benefit of forecasts yields far larger and statistically significant estimated forecast benefits (SI figure S2).For example, in that naive model, correct prediction of extreme heat at a 3 day lead enables a 28% reduction in yield loss from extreme heat.The upward bias in those naive estimates is consistent with patterns in figure 1 and heterogeneity in the effects of extreme heat.More extreme heat is correctly forecast in hotter areas (figure 1(a)) where farms and selected crop varieties are likely better adapted to heat.Smaller heat-related yield losses in those areas would otherwise be attributed to better forecasts if other forms of heterogeneity remain unmodeled.Further, we note that the loss of statistical significance when moving from this naive model to our primary specification appears to be driven by this bias rather than a simple loss of statistical power (SI section 4).We hope the misleading estimates from this naive model illustrate an important pitfall to other potential users of NDFD data.

Discussion
Our estimates offer a sobering view of the yieldrelated value of short-term weather forecasts for US corn farmers.However, while we find little evidence  b) Effects of extreme heat (gray) and differential effects when that heat is correctly forecast (green).Estimates are presented for different forecast leads (horizontal axis), with separate effects for counties with low (<25% of operations) or high (>75% of operations) levels of irrigation (left panels) and those with moderate (25%-75% of operations) levels of irrigation (right panels), and heat occurring during the silking period in a county (bottom panels) and not (top panels).Dots represent point estimates and bars 95% confidence intervals.
of avoided yield loss, that does not mean heat forecasts are of no use to farmers.Forecasts could enable cost savings rather than preventing yield loss, if farmers are able to more efficiently apply inputs (e.g.water).Because expected drops in harvest can drive up futures prices (Summer and Mueller 1989), farmers anticipating supply shortages due to impending heat may instead seek to lock in higher prices by selling futures or forward contracts.Finally, corn growers anticipating high heat could also begin making preparations to harvest their crop for silage rather than grain.All of these adaptations imply a potential value of extreme heat forecasts that will not be reflected in our focus on yield.
We also acknowledge that our investigation of irrigation as a mechanism for adaptive responses to heat forecasts is limited.Spatially resolved, high frequency data on within-season water applications is not available at the scale of our analysis, leading to our focus on how forecast benefits vary with the share of operators in a county who harvest irrigated corn.As a noisy proxy for irrigation use, this variable could attenuate the estimated value of forecasts.The prevalence of irrigation could also covary with other adaptive practices that are unmeasured, which could either attenuate or inflate our estimates of the role of irrigation.As such, we encourage interpretation of irrigation results as suggestive only; better irrigation data could help resolve these concerns in future analyses.
The value of extreme heat forecasts for other crops may also be higher.Growers of specialty crops rather than a field crop like corn may have a wider range of adaptation options.In particular, higher cost or more labor-intensive adaptations to forecasts of high heat may be warranted for higher value crops.For example, anecdotal evidence suggests farmers may preemptively alter plant canopy structure to shade crops (Nicholas and Durham 2012), and spray coatings on fruits can reduce heat damage (Glenn 2012).The value of weather and climate information may also be higher in other regions (Anuga and Gordon 2016), though that value may be constrained by the availability of high-quality climate services (Georgeson et al 2017), including short-term heat forecasts.Identifying where and when adaptations to heat are triggered by forecasts is a useful avenue for future work.
It is also possible that the value of short-term heat forecasts will change in the future.Both new and existing agrochemicals are increasingly being marketed as helping crops cope with heat stress, both for corn and other crops.If that marketing is justified, those products could give farmers additional adaptation responses to heat forecasts.For example, external application of plant bioregulators may directly limit oxidative damage that results from a primary stressor such as heat (Vardhini and Anjum 2015, Sarwar et al 2017, EL Sabagh et al 2020, Zulfiqar and Ashraf 2021).Foliar sprays may also help regulate levels and ratios of key plant hormones (e.g.ethylene) that normally respond to heat stress in ways that reduce yield (Wilkinson et al 2012).Finally, some fungicides are also marketed to help water-stressed crops cope with extreme heat via a reduction in stomatal conductance, though evidence of their efficacy and costeffectiveness in that role is mixed (Wise and Mueller 2011, Sulewska et al 2019).Despite this potential, our reduced-form estimates suggest that, even if these and other tools are being deployed in response to heat forecasts, they are not currently resulting in a detectable reduction in yield loss.
On the other hand, countervailing trends could dampen the limited forecast value we do see.To the extent that irrigation is a primary channel through which adaptive responses occur, shifting water availability may constrain forecast-responsive adaptation (Elliott et al 2014).Depletion of groundwater resources (Jasechko and Perrone 2021) may constrain adaptive irrigation (though surface water availability has declined in some areas but increased in others (Pekel et al 2016)).Trust in and use of forecasts to guide adaptation could also decline with climate change (Guido et al 2021).On a more positive note, the value of forecasts could shrink with continued progress in breeding heat-and drought-tolerant crop varietals, which require less in-season adaptation to retain yield.

Conclusion
We have quantified the value of short-term heat forecasts in helping corn growers in the US avoid yield losses from extreme heat.Using an underutilized dataset on historical short-term (1-7 day lead) daily weather forecasts, we separated extreme heat into that which was predicted and that which was not, and asked whether those types of exposure differed in their impacts on yield.While we found no evidence of an average benefit of forecasts, we did observe small potential benefits in areas where irrigation is a viable adaptation response.This narrow scope for forecast benefits underscores that information is valuable only when it triggers a change in behavior, which will not occur if preventive measures are too costly (eastern US) or already being deployed regardless of the forecast (irrigation during silking).On a more positive note, our analysis has offered a template for largescale, ex-post investigation of forecast value that we hope can enable a range of future research for other crops, regions, and channels via which forecasts could benefit farmers.

Figure 1 .
Figure 1.Spatial and temporal variation in the percentage of extreme degree days (EDDs) that are correctly predicted at a 3-day forecast lead.(a) Spatial variation in the long-run average percent EDD that are correctly predicted per county.(b) Monthly variation in the percentage of EDD that are correctly predicted (left axis, black) and that occur (right axis, dark red), with box plots showing distribution across counties.(c) Inter-annual variation in the percent EDD that are correctly predicted for all individual counties (light gray), two example counties (dark gray; Phelps County, Nebraska and Calhoun County, Georgia), and total EDD across all counties in the US (red).

Figure 2 .
Figure2.Estimated effects of (1) an extreme degree day (EDD; gray) on corn yield in a county and year, and (2) the differential effect of an EDD if correctly predicted in advance (green) for models using different forecast leads (horizontal axis).Estimates are separated for EDD and predictions during the primary silking month (triangles) in a state vs. other (circles).Points represent mean estimates and bars 95% confidence intervals.

Figure 4 .
Figure 4. Effects of extreme heat by irrigation prevalence in Nebraska and Kansas.(a) Percent of operations in each county harvesting irrigated corn acres (blue gradient), with interquartile range highlighted in pink.(b) Effects of extreme heat (gray) and differential effects when that heat is correctly forecast (green).Estimates are presented for different forecast leads (horizontal axis), with separate effects for counties with low (<25% of operations) or high (>75% of operations) levels of irrigation (left panels) and those with moderate (25%-75% of operations) levels of irrigation (right panels), and heat occurring during the silking period in a county (bottom panels) and not (top panels).Dots represent point estimates and bars 95% confidence intervals.
. Others explicitly estimate adaptation benefits, whether focusing on specific actions such as irrigation (Siebert et al 2017, Tack et al 2017), nutrient maintenance (Waraich et al 2012, Mengutay et al 2013), or development of more tolerant varietals (Fahad et al 2017), or estimating aggregate benefits across adaptation approaches Deschênes O and Greenstone M 2007 The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather Am.Econ.Rev. 97 354-85 Diffenbaugh N S, Davenport F V and Burke M 2021 Historical warming has increased US crop insurance losses Environ.Wang H, Carrillo C M and Ault T R 2019 Unpacking the climatic drivers of US agricultural yields Environ.Res.Lett.14 064003 Pekel J-F, Cottam A, Gorelick N and Belward A S 2016 High-resolution mapping of global surface water and its long-term changes Nature 540 418-22 Sarwar M, Saleem M, Najeeb U, Shakeel A, Ali S and Bilal M 2017 Hydrogen peroxide reduces heat-induced yield losses in cotton (Gossypium hirsutum L.) by protecting cellular membrane damage J. Agron.Crop Sci.203 429-41 Schlenker W and Roberts M J 2009 Nonlinear temperature effects indicate severe damages to US crop yields under climate change Proc.Natl Acad.Sci.106 15594-8 Siebert S, Webber H, Zhao G and Ewert F 2017 Heat stress is overestimated in climate impact studies for irrigated agriculture Environ.Res.Lett.12 054023 Snyder R L 1985 Hand calculating degree days Agric.Forest Meteorol.35 353-8 Stevens A 2017 Temperature, wages, and agricultural labor productivity Technical Report (UC Berkeley Working Paper, Accessible on UC Berkeley Website) (available at: http://are.berkeley.edu/sites/default/files/job-candidates/paper/stevens_jmp_jan16.pdf)Sulewska H, Ratajczak K, Panasiewicz K and Kalaji H M 2019 Can pyraclostrobin and epoxiconazole protect conventional and stay-green maize varieties grown under drought stress?PLoS One 14 e0221116 Summer D A and Mueller R A 1989 Are harvest forecasts news?USDA announcements and futures market reactions Am.J. Agric.Econ.71 1-8 Suzuki N, Rivero R M, Shulaev V, Blumwald E and Mittler R 2014 Abiotic and biotic stress combinations New Phytol.203 32-43 Tack J, Barkley A and Hendricks N 2017 Irrigation offsets wheat yield reductions from warming temperatures Environ.Res.Lett.12 114027 Tack J, Barkley A and Nalley L L 2015 Effect of warming temperatures on US wheat yields Proc.Natl Acad.Sci.112 6931-6 Tigchelaar M, Battisti D S and Spector J T 2020 Work adaptations insufficient to address growing heat risk for US agricultural workers Environ.Res.Lett.15 094035 Toreti A, Bassu S, Asseng S, Zampieri M, Ceglar A and Royo C 2022 Climate service driven adaptation may alleviate the impacts of climate change in agriculture Commun.Biol. 5 1235 Vardhini B V and Anjum N A 2015 Brassinosteroids make plant life easier under abiotic stresses mainly by modulating major components of antioxidant defense system Front.Environ.Sci. 2 67 Vogel E, Donat M G, Alexander L V, Meinshausen M, Ray D K, Karoly D, Meinshausen N and Frieler K 2019 The effects of climate extremes on global agricultural yields Environ.Res.Lett.14 054010 Waraich E, Ahmad R, Halim A and Aziz T 2012 Alleviation of temperature stress by nutrient management in crop plants: a review J. Soil Sci.Plant Nutrition 12 221-44 Wilkinson S, Kudoyarova G R, Veselov D S, Arkhipova T N and Davies W J 2012 Plant hormone interactions: innovative targets for crop breeding and management J. Exp.Bot.63 3499-509 Wise K and Mueller D 2011 Are fungicides no longer just for fungi?An analysis of foliar fungicide use in corn APSnet Features 10 Zaveri E and Lobell D 2019 The role of irrigation in changing wheat yields and heat sensitivity in India Nat.Commun.10 1-7 Zulfiqar F and Ashraf M 2021 Bioregulators: unlocking their potential role in regulation of the plant oxidative defense system Plant Mol.Biol.105 11-41