Lesk et al. (2016). Influence of extreme weather disasters on global crop production

8 4 | N A T U R E | V O L 5 2 9 | 7 J A N U A R y 2 0 1 6

LETTER

doi:10.1038/nature16467

Influence of extreme weather disasters on global

crop production

Corey Lesk

1

, Pedram Rowhani

2

& Navin Ramankutty

1,3

In recent years, several extreme weather disasters have partially or

completely damaged regional crop production

1–5

. While detailed

regional accounts of the effects of extreme weather disasters exist,

the global scale effects of droughts, floods and extreme temperature

on crop production are yet to be quantified. Here we estimate for the

first time, to our knowledge, national cereal production losses across

the globe resulting from reported extreme weather disasters during

1964–2007. We show that droughts and extreme heat significantly

reduced national cereal production by 9–10%, whereas our analysis

could not identify an effect from floods and extreme cold in the

national data. Analysing the underlying processes, we find that

production losses due to droughts were associated with a reduction

in both harvested area and yields, whereas extreme heat mainly

decreased cereal yields. Furthermore, the results highlight

~

7%

greater production damage from more recent droughts and 8–11%

more damage in developed countries than in developing ones. Our

findings may help to guide agricultural priorities in international

disaster risk reduction and adaptation efforts.

In many regions of the world, there have been considerable changes

in the nature of droughts, floods and extreme temperature events since

the middle of the twentieth century

6–8

. Over agricultural areas, disasters

arising from extreme weather can cause marked damage to crops and

food system infrastructure, with the potential to destabilize food sys-

tems and threaten local to global food security. In recent years, nearly

one-quarter of all damage and losses from climate-related disasters has

been in the agricultural sector in developing countries

9

. With such dis-

asters expected to become more common in the future

1,6,7

, policymak-

ers need robust scientific information to develop effective disaster risk

management and adaptation interventions (for example, infrastructure,

technology, management and insurance) to protect the most vulnerable

populations and to ensure global food security.

Whether an extreme weather event results in a disaster depends not

only on the severity of the event itself, but also on the vulnerability and

exposure of the human and natural systems that experience it

6

. Past

research has addressed agricultural effects of specific weather extremes

with fixed definitions, such as degree days above some threshold

10–15

.

This approach probably underestimates the crop effects of extreme

weather disasters (EWDs), because similar extreme weather events may

have differing effects depending on the vulnerability of the exposed

system.

In this study, we address this bias by using a disaster data set com-

piled based on human impact. In addition, we attend to two further

limitations of previous work on extreme weather and agriculture. First,

several regional empirical studies have highlighted the adverse effects

of extreme heat events on crop yields

10–13

, and global modelling efforts

have estimated future crop yield declines due to increasing extreme

heat stress

14,15

. But this emphasis on crop yields offers an incomplete

picture of crop production because of the potential for compensation

or compounding of yield impacts by changes in harvested area

16

;

and because crop production (and not yields)—together with access

and utilization—determines food security

2,4,7,17,18

. Second, we seek

to investigate the agricultural effects of often-overlooked extreme

weather events, namely floods and extreme cold

2,3

. Thus, our study is

the first, to our knowledge, that takes an empirical approach to estimat-

ing the influence of EWDs on crop area, yields and production at the

global scale.

We use a statistical method, superposed epoch analysis

19

(also known

as compositing, see Methods), to estimate average national per-disaster

cereal production losses (Food and Agriculture Organization of the

United Nations (FAO), http://faostat3.fao.org) across the globe due

to reported droughts, floods and extreme temperature disasters from

1964 to 2007. Furthermore, we estimate the effects on cereal yield and

harvested area separately to identify processes leading to production

losses. On the basis of

~

2,800 reported extreme hydro-meteorological

disasters collated by the Emergency Events Database (EM-DAT, http://

www.emdat.be/database), we find that national cereal production

during a drought was significantly reduced by 10.1% on average (95%

confidence interval 9.9–10.2%), while years with extreme heat led to

national production deficits of 9.1% (8.4–9.5%; Fig. 1a, b and Extended

Data Table 1). These production deficits were equivalent to roughly

6 years of production growth; however, no significant lasting effects

were noted in the years after the disasters. Estimated mean production

losses were driven mainly by a preponderance of disasters with moder-

ate effects on crops, as opposed to a few extreme cases (Extended Data

Fig. 1), and were not strongly influenced by sample size (see Extended

Data Fig. 2, Extended Data Table 2 and Supplementary Discussion).

During 1964–2007, these estimated EWD effects represent a loss of

1,820 million Mg due to droughts (approximately equal to the global

maize and wheat production in 2013), and 1,190 million Mg due to

extreme heat disasters (more than the global 2013 maize harvest). Over

2000–2007 (the period with the most complete disaster reporting com-

pared with earlier decades), 6.2% of total global cereal production was

lost due to EWDs relative to an estimated counterfactual global produc-

tion without EWD effects (3.0% to extreme heat and 3.2% to drought).

Cereal yield declines during EWDs were 5.1% (4.9–5.2%) and 7.6%

(7.0–8.1%) for drought and extreme heat, respectively (Fig. 2a). The

harvested area dropped 4.1% (4.0–4.3%) during droughts, but was not

significantly affected by extreme heat (Fig. 2b). This may be due to the

shorter duration of extreme heat relative to droughts—while approx-

imately one-third of droughts in this study spanned several years, all

extreme heat disasters took place within a single year. Droughts may

thus be more likely to last long enough to cause complete crop failure

and discourage planting, while extreme heat disasters, especially out-

side key crop developmental stages, may affect crop growth and reduce

yields without critically damaging harvests.

Our estimated yield deficits from EWDs cannot be directly com-

pared to previous studies of the impact of seasonal mean climate trends

over the same period

20

(see Supplementary Discussion). However, we

derived a comparable measure to that reported previously

21

, and esti-

mated a yield sensitivity of 6–7% per 1 °C increase in seasonal mean

1

Department of Geography, McGill University, Montreal H3A 0B9, Canada.

2

Department of Geography, University of Sussex, Brighton BN1 9QJ, UK.

3

Liu Institute for Global Issues and Institute for

Resources, Environment and Sustainability, University of British Columbia, Vancouver V6T 1Z2, Canada.

7 J A N U A R y 2 0 1 6 | V O L 5 2 9 | N A T U R E | 8 5

Letter

reSeArCH

weather associated with extreme heat disasters, which suggests that our

observed extreme heat effects are not necessarily independent from

those detected in studies examining changes in seasonal temperatures

(Extended Data Fig. 3). Methodological differences and uncertainties

prevent us from drawing strong conclusions based on this comparison.

Our drought impacts, however, seem to be independent of previous

estimates that used seasonal weather anomalies (see Supplementary

Discussion).

Our results do not show significant production effects from extreme

cold and floods (Fig. 1c, d). A potential explanation for this is that

floods tend to occur in the spring in temperate regions as a result of

snowmelt, and cold weather susceptibility in most agricultural regions

is highest outside the growing season, which may render a sizeable

portion of the flood and extreme cold disasters analysed in this study

agriculturally irrelevant. The estimated lack of response may also be

an artefact of the spatial dimension of these disasters. While drought

and extreme temperature affect broad regions, floods are a function

of both weather and topography, and can be highly localized within

a country

22

. Since this study uses country-level agricultural statistics,

one may speculate that a more noticeable flood effect on sub-national

production is masked at the national scale.

Several additional analyses offer more detailed insights into the

effects of these EWDs on cereal production. Cereals in the more tech-

nically developed agricultural systems of North America, Europe and

Australasia suffered most from droughts, facing on average a 19.9%

production deficit compared to 12.1% in Asia, 9.2% in Africa, and no

significant effect in Latin America and the Caribbean (overall differ-

ence in means

P

=

0.02; Fig. 3a and Extended Data Tables 3 and 4). This

more severe production impact in the developed nations was driven

by a substantial yield deficit of 15.9%, with no significant reduction in

harvested area (Fig. 3b, c). We see three possible explanations for this

pattern. First, it may arise from a tendency among lower-income coun-

tries to encompass diverse crops and management across many small

fields, which may allow for some fields to resist drought better than

others. This might reduce the national drought sensitivity compared

to higher-income countries, where large-scale monocultures are more

dominant. Second, lower-income countries may better resist drought

because smallholders tend to use risk-minimizing strategies compared

to the yield-maximizing ones prevalent in higher-income countries.

Third, the pattern may relate to generally lower fair-weather yields in

lower-income countries. In Asia, we found a significant reduction of

8.8% in harvested area during droughts with no corresponding yield

deficit, suggesting that this region has a greater tendency for total crop

failure in the event of a drought rather than harvesting with reduced

yields

16

. The production effects in Africa did not correspond to signif-

icant deficits in either yield or harvested area.

While the production of all three crops was similarly affected by

droughts (5–6% deficit each; Fig. 4a, Extended Data Tables 2 and 5),

only maize was significantly affected by extreme heat (11.7% deficit,

P

=

0.01) (Fig. 4d). Maize was also the only crop with significant yield

Drought (

n

= 222)

Extreme heat (

n

= 32)

Extreme cold (

n

= 51)

Flood (

n

= 756)

Normalized composite

a

b

c

d

1.2

1.1

1.0

0.9

0.8

1.2

1.1

1.0

0.9

0.8

1.2

1.1

1.0

0.9

0.8

10.1%

9.1%

Year from event

1.2

1.1

1.0

0.9

0.8

–3

–2

–1

0

1

2

3

–3

–2

–1

0

1

2

3

Figure 1

|

Influence of EWDs on national cereal production.

a

–

d

, Normalized production composites for drought (

n

=

222)

(

a

), extreme heat (

n

=

32) (

b

), flood (

n

=

756) (

c

) and extreme cold

disasters (

n

=

51) (

d

) over 7-year windows centred on the disaster year

(blue lines). Box plots depict the distributions of 1,000 false-disaster

control composites, with red crosses denoting extreme outliers, and red

dashes denoting medians. Production during drought and extreme heat

years was 10.1% and 9.1% below the control mean, respectively, whereas

no significant production signal was detected for floods or extreme cold.

Production resumed normal levels immediately after drought and extreme

heat. The increasing trend in production over the 7-year window reflects

the observed growth trend.

Year from event

a

b

Year from event

Drought (

n

= 222)

Extreme heat (

n

= 32)

Yield

Harvested area

1

2

3

0

–1

–2

–3

0.94

1.02

1.04

1.06

0.98
0.96

0.92

Normalized composite

1

2

3

0

–1

–2

–3

0.94

1.02

1.04

1.06

0.98
0.96

0.92

1.00

*

Figure 2

|

Influence of EWDs on national cereal yields and harvested

area. a

,

b

, Yield (blue) and harvested area (red) composites for drought

(

n

=

222) (

a

) and extreme heat (

n

=

32) (

b

), with significant points (those

lying beyond the control box plot whiskers) marked by asterisks (box plots

not shown for clarity). Drought was associated with significant deficits

in both yield and harvested area (5.1 and 4.1%), whereas extreme heat

revealed only significant yield impacts of 7.6% with no significant effect on

harvested area.

8 6 | N A T U R E | V O L 5 2 9 | 7 J A N U A R y 2 0 1 6

Letter

reSeArCH

effects (12.4%,

P

=

0.002) (Fig. 4b, e). We are hesitant to draw strong

conclusions based on this difference, as it may be due to differing var-

iance as well as mean (see Extended Data Table 6 and Supplementary

Discussion). Furthermore, it may reflect the fact that maize is generally

grown during summer months, which have the highest probabilities of

extreme heat as defined in EM-DAT, while wheat is grown during the

spring. Disaster data with monthly or daily resolution would enable us

to investigate whether this apparent susceptibility of maize is a result

of differing growing season.

Finally, more recent droughts (1985–2007) caused cereal produc-

tion losses averaging 13.7%, greater than the estimated 6.7% during

earlier droughts (1964–1984) (

P

=

0.008, Fig. 5), which may be due

to any combination of rising drought severity (although whether

drought severity has increased globally is presently debated)

23–26

,

increasing vulnerability

6,27

and exposure to drought

6

, and/or chang-

ing reporting dynamics (Extended Data Fig. 4). Sample size limita-

tions prevented us from repeating a regional and temporal analysis

for extreme heat.

Some limitations of our analyses are worth noting. First, we mainly

focus on four principal types of EWDs, but follow-up studies should

include tropical storms and extreme precipitation and wind events,

especially since they may have an increasingly important effect on

agriculture in the context of climate change

28

. Second, our estimates

are biased towards more recent disasters as they are more abundantly

reported in EM-DAT than older ones (see Extended Data Fig. 4 and

Supplementary Discussion). Third, we use EWDs from the EM-DAT

database, which collates disasters based on several criteria for sub-

stantial human impact (Methods). We may be underestimating the

true effects of EWDs if disasters are included mainly based on urban

impacts, or if extreme events occurring in sparsely populated areas are

less likely to qualify as disasters. Finally, since we observe agricultural

impacts at the national level, more notable local and regional effects of

disasters may be muted (but conversely, finding a signal at the national

level highlights the substantial influence of droughts and extreme heat).

Future studies may arrive at a more detailed estimate by using subna-

tional agricultural data, localizing the reported disasters within nations,

selecting disasters taking place during the growing season, and con-

trolling for severity of disasters. Linking the definitions of EWDs used

in this study with statistical meteorological definitions will also enable

a forecasting of future impacts.

Overall, there are four main conclusions from our study. First, over

the period 1964–2007, drought and extreme heat substantially damaged

national agricultural production across the globe. Within the frame-

work of this study, no effect on agriculture was identified from floods

and extreme cold. Second, drought reduced cereal yield and completely

damaged crops, whereas extreme heat only affected yield, reflecting

clear differences in the processes leading to overall production effects.

Third, this study highlights an important temporal dimension to these

impacts. While the damage to cereal production is considerable, this

effect is only short term, as agricultural output rebounds and continues

its growth trend after the disaster. Furthermore, we show that recent

droughts had a larger effect on cereal production than earlier ones.

Production

Yield

Harvested area

Drought

a

b

c

d

e

f

1

–3

2

3

0

–1

–2

Normalized composite

1.1

1.0

0.9

Extreme
heat

1.2

1.1

1.0

0.9

0.8

1.1

1.0

0.9

1.2

1.1

1.0

0.9

0.8

1.1

1.0

0.9

1.2

1.1

1.0

0.9

0.8

1

–3

2

3

0

–1

–2

1

–3

2

3

0

–1

–2

P

= 0.61

P

= 0.68

P

= 0.65

P

= 0.01

P

= 0.002

P

= 0.18

Year from event

Maize (

n

= 28)

Rice (

n

= 16)

Wheat (

n

= 32)

Maize (

n

= 208)

Rice (

n

= 171)

Wheat (

n

= 234)

*

Figure 4

|

The influence of drought and extreme heat on maize, rice and

wheat. a

–

f

, Drought and extreme heat composites of production (

a

,

d

),

yield (

b

,

e

) and harvested area (

c

,

f

) for maize (blue), rice (red) and wheat

(green), with significant points (those lying beyond the control box plot

whiskers) marked by asterisks (box plots not shown for clarity).

P

values

reflect significant differences between crops in disaster-year response

(Kruskal–Wallis test). Maize production (

n

=

28) responds more (

P

=

0.01)

to extreme heat than wheat (

n

=

32) and rice (

n

=

16), an effect driven by a

substantial yield deficit (

P

=

0.002). For drought data, maize (

n

=

208), rice

(

n

=

171) and wheat (

n

=

234).

Production

Yield

Harvested area

Year from event

1

–3

2

3

0

–1

–2

1

–3

2

3

0

–1

–2

P

= 0.02

P

= 0.002

P

= 0.13

a

b

c

Normalized composite

1

–3

2

3

0

–1

–2

1.10

1.05

1.00

0.95

0.90

0.85

0.80

1.10

1.05

1.00

0.95

0.90

0.85

0.80

1.10

1.05

1.00

0.95

0.90

0.85

0.80

Eur., N.Am., Aus. (

n

= 28)

Asia (

n

= 32)

Africa (

n

= 125)

L.Am. & Carib. (

n

= 37)

*

Figure 3

|

A regional analysis of the influence of drought. a

–

c

, Regional

composites of production (

a

), yield (

b

) and harvested area (

c

) for

drought, with significant points (those lying beyond the control box plot

whiskers) marked by asterisks (box plots not shown for clarity).

P

values

reflect significant differences between regions in drought-year response

(Kruskal–Wallis test). The drought-year normalized production is 7.8%

and 10.7% lower (

P

=

0.02) in developed Western countries (

n

=

28)

than in Asia (

n

=

32) and Africa (

n

=

125) (

a

), a difference driven by a

significantly greater yield deficit (

P

=

0.002) (

b

). Meanwhile, the Latin

America (L.Am) and Caribbean (Carib.) region (

n

=

37) exhibits no

significant response to drought. Aus., Australasia; Eur, Europe; N.Am.,

North America.

7 J A N U A R y 2 0 1 6 | V O L 5 2 9 | N A T U R E | 8 7

Letter

reSeArCH

Finally, our regional and crop-specific analysis finds that developed

nations suffer most from these EWDs.

Present climate projections suggest that extreme heat events will be

increasingly common and severe in the future

1

. Droughts are likely

to become more frequent in some regions, although considerable

uncertainty persists in the projections

6

. This study, by highlighting

the important historical effects of EWDs on agriculture, emphasizes

the urgency with which the global cereal production system must

adapt to extremes in a changing climate. Understanding the key pro-

cesses leading to such crop losses enables an informed prioritiza-

tion of disaster risk reduction and adaptation interventions to better

protect the most vulnerable farming systems and the populations

dependent on them.

Online Content

Methods, along with any additional Extended Data display items and

Source Data, are available in the online version of the paper; references unique to

these sections appear only in the online paper.

received 29 April; accepted 16 November 2015.

1. Battisti, D. S. & Naylor, R. L. Historical warnings of future food insecurity with

unprecedented seasonal heat.

Science

323,

240–244 (2009).

2. World Food Programme.

Pakistan – Flood Impact Assessment

http://documents.

wfp.org/stellent/groups/public/documents/ena/wfp225987.pdf (2010).

3. Gu, L.

et al.

The 2007 eastern US spring freeze: increased cold damage in a

warming world.

Bioscience

58,

253 (2008).

4. Barriopedro, D., Fischer, E. M., Luterbacher, J., Trigo, R. M. & García-Herrera, R.

The hot summer of 2010: redrawing the temperature record map of Europe.

Science

332,

220–224 (2011).

5. Coumou, D. & Rahmstorf, S. A decade of weather extremes.

Nature Clim.

Change

2,

491–496 (2012).

6. Intergovernmental Panel on Climate Change. in

Managing the Risks of Extreme

Events and Disasters to Advance Climate Change Adaptation

(eds Field, C. B.

et al.

)

1–19 (Cambridge Univ. Press, 2012).

7. United Nations Office for Disaster Risk Reduction.

The Pocket GAR 2013 –

From Shared Risk to Shared Value: the Business Case for Disaster Risk Reduction

(UNISDR, 2013).

8. World Meterological Organization.

Atlas of Mortality and Economic Losses from

Weather, Climate and Water Extremes (1970–2012).

(WMO, 2014).

9. Food and Agriculture Organization of the United Nations.

The Impact of Natural

Hazards and Disasters on Agriculture and Food and Nutrition Security – A Call for

Action to Build Resilient Livelihoods

(FAO, 2015).

10. Lobell, D. B.

et al.

The critical role of extreme heat for maize production in the

United States.

Nature Clim. Change

3,

497–501 (2013).

11. Lobell, D. B., Sibley, A. & Ivan Ortiz-Monasterio, J. Extreme heat effects on

wheat senescence in India.

Nature Clim. Change

2,

186–189 (2012).

12. Lobell, D. B., Bänziger, M., Magorokosho, C. & Vivek, B. Nonlinear heat effects

on African maize as evidenced by historical yield trials.

Nature Clim. Change

1,

42–45 (2011).

13. Moriondo, M., Giannakopoulos, C. & Bindi, M. Climate change impact

assessment: the role of climate extremes in crop yield simulation.

Clim. Change

104,

679–701 (2010).

14. Teixeira, E. I., Fischer, G., van Velthuizen, H., Walter, C. & Ewert, F. Global

hot-spots of heat stress on agricultural crops due to climate change.

Agric. For.

Meteorol.

170,

206–215 (2013).

15. Deryng, D., Conway, D., Ramankutty, N., Price, J. & Warren, R. Global crop yield

response to extreme heat stress under multiple climate change futures.

Environ. Res. Lett.

9,

034011 (2014).

16. Iizumi, T. & Ramankutty, N. How do weather and climate influence cropping

area and intensity?

Glob. Food Sec

.

4,

46–50 (2015).

17. Johnstone, S. & Mazo, J. Global warming and the Arab Spring.

Survival

53,

11–17 (2011).

18. Welton, G.

The Impact of Russia’s 2010 Grain Export Ban

(Oxfam International,

2011).

19. Brad Adams, J., Mann, M. E. & Ammann, C. M. Proxy evidence for an El

Niño-like response to volcanic forcing.

Nature

426,

274–278 (2003).

20. Lobell, D. B., Schlenker, W. & Costa-Roberts, J. Climate trends and global crop

production since 1980.

Science

333,

616–620 (2011).

21. Lobell, D. B. & Field, C. B. Global scale climate–crop yield relationships and the

impacts of recent warming.

Environ. Res. Lett.

2,

014002 (2007).

22. Thornton, P. K., Ericksen, P. J., Herrero, M. & Challinor, A. J. Climate variability

and vulnerability to climate change: a review.

Glob. Chang. Biol.

20,

3313–3328

(2014).

23. Dai, A. Increasing drought under global warming in observations and models.

Nature Clim. Change

3,

52–58 (2012).

24. Sheffield, J., Wood, E. F. & Roderick, M. L. Little change in global drought over

the past 60 years.

Nature

491,

435–438 (2012).

25. Trenberth, K. E.

et al.

Global warming and changes in drought.

Nature Clim.

Change

4,

17–22 (2014).

26. Greve, P.

et al.

Global assessment of trends in wetting and drying over land.

Nature Geosci

.

7,

716–721 (2014).

27. Lobell, D. B.

et al.

Greater sensitivity to drought accompanies maize yield

increase in the U.S. Midwest.

Science

344,

516–519 (2014).

28. Gornall, J.

et al.

Implications of climate change for agricultural productivity in

the early twenty-first century.

Phil. Trans. R. Soc. Lond. B

365,

2973–2989

(2010).

Supplementary Information

is available in the online version of the paper.

Acknowledgements

We thank R. Below, who is in charge of the EM-DAT project

at the Centre for Research on the Epidemiology of Disasters, for sharing the

data. We thank C. Champalle for testing the original idea using data over East

Africa in a class project. This research was supported by a Discovery Grant from

the Natural Science and Engineering Research Council of Canada to N.R.

Author Contributions

This research was designed and coordinated by N.R. All

authors performed analyses, discussed the results, and wrote the manuscript.

Author Information

Reprints and permissions information is available at

www.nature.com/reprints. The authors declare no competing financial

interests. Readers are welcome to comment on the online version of the paper.

Correspondence and requests for materials should be addressed to

N.R. (navin.ramankutty@ubc.ca).

a

b

Earlier (1964–1984)

n

= 126

Later (1985–2007)

n

= 121

Year from event

1.1

1.0

0.9

1.1

1.0

0.9

–3

–2

–1

0

1

2

3

–3

–2

–1

0

1

2

3

Normalized composite

Figure 5

|

A temporal analysis of the influence of drought.

a

,

b

, Production composites for earlier (1964–1984,

n

=

126) (

a

) and

later (1985–2007,

n

=

121) (

b

) droughts, with boxplots of 100 respective

control composites. In later instances, mean drought-year production

losses were greater (13.7%) than in earlier instances (6.7%;

P

=

0.008,

Kruskal–Wallis test).

Letter

reSeArCH

MethOdS

Superposed epoch analysis (SEA) is used to isolate an average EWD response

signal using time series of national agricultural production data and EWDs. SEA is

a statistical approach that has been used to enhance the signal (that is, influence of

particular events) in time-series data, while reducing noise due to extraneous vari-

ables

19

. The EWDs are compiled from the Emergency Events Database (EM-DAT;

http://www.emdat.be/database) and consist of 2,184 floods, 497 droughts, 138

extreme heat and 194 extreme cold disasters from 177 countries over the period

1964–2007. EM-DAT collects information on a reported disaster if at least ten peo-

ple died, a state of emergency was declared, international assistance was called, or

at least 100 people were injured, made homeless or required immediate assistance.

Disaster reports are gathered from various organizations including United Nations

agencies, governments, and the International Federation of Red Cross and Red

Crescent Societies

8

. The agricultural data consist of country-level total production,

average yield, and total harvested area data for 16 cereals (http://faostat3.fao.org)

covering the 177 countries in the set of EWDs from 1961 to 2010.

From the time series of agricultural data, we extracted shorter sets of time series

using a 7-year window centred on the year of occurrence of each EWD, with 3 years

of data preceding and following each EWD. The data were normalized to the aver-

age of the 3 years preceding and following the event to remove the absolute magni-

tude of national data from the signal. For multi-year droughts, we averaged across

all drought years to produce a single disaster year datum. For a 3-year drought,

for example, the 7-year window became a 9-year window with seven data points

(with the middle 3 years being averaged and assigned to year 0). The 7-year sets of

EWD time series were then centred on the disaster year and averaged year-wise to

yield single composited time-series of production, yield and harvested area for each

EWD type (a total of 12 composited time series). The averaging thus strengthens

the signal at the central year of EWD occurrence, while also cancelling the noise

in the non-disaster years preceding and following the event.

During compositing, points on individual time series co-occurring with another

disaster in the set were excluded from the mean. This procedure resulted in variable

sample size across the 7 years of the composites. For brevity, we have presented

mean sample sizes across all years; complete tabulated sample sizes are displayed in

Extended Data Tables 2 and 4. Our composited mean estimate does not seem to be

influenced by outliers (see Extended Data Fig. 1 and Supplementary Discussion).

The signal-to-noise strength will certainly depend on the sample size, and we

performed an analysis to estimate the influence of sample size (see Extended Data

Tables 2 and 4 , Extended Data Fig. 2 and Supplementary Discussion).

In addition to average per-disaster estimates, we also calculated aggregate pro-

duction losses over specific time periods. For each extreme heat or drought, we

first applied the average per-disaster percentage loss estimate (different values for

extreme heat or drought) to the average national production across the six adjacent

non-disaster years. We then computed the aggregate drought or heat-related global

production loss for each year by summing the production losses for each disaster

over the given time period. We estimated the percentage of global production lost

to the EWDs relative to an estimated counterfactual global production in a world

without EWDs (the latter being the sum of observed global production plus the

estimated production loss).

The significance-testing procedure involved setting up a ‘control’ estimate by

randomly resampling the agricultural data using sets of fictitious disasters with

randomly generated years and countries of occurrence. The fictitious EWD time

series were averaged as for the true ones to yield composited ‘control’ time series,

and the entire process was repeated 1,000 times. We quantified EWD-year deficits

in production, yield and harvested area by subtracting the true EWD time series

from the mean of the controls. Excluding randomly generated disasters that hap-

pened to be real disasters systematically raised the impact estimates by

~

1%; to

present a more conservative and rigorous detection of the disaster signal, we elected

not to exclude such pseudo-disasters. Note that we chose not to de-trend the time

series before compositing to remove technology-driven growth, but rather simply

estimate the disaster signal as difference from control (see Fig. 1). We estimated the

95% confidence intervals for our point estimates of impacts using an approach sim-

ilar to a delete-one jackknife resampling method (see Supplementary Discussion).

The percentage significance of each estimate of the EWD composites relative

to control was estimated as the percentage of 1,000 control points less than the

EWD composite estimate for each year. Points with estimated significance of

<

0.5% or

>

99.5% were considered significant deficits and surpluses, respectively,

corresponding to a two-tailed 99% confidence level. While we chose a two-tailed

approach for robustness, we found no significant surpluses. The significant points

appear as asterisks in Figs 2–4, while for Figs 1 and 5 we present the EWD compos-

ites with the distribution of controls represented as an array of box-and-whisker

plots for a visual representation of significance. The complete tabulated percentage

significance values are presented in Extended Data Tables 1 , 3 and 5.

The earlier-versus-later analysis for droughts was performed by applying the

SEA procedure to the set of droughts divided roughly equally into earlier and later

halves. Similarly, the regional analysis was conducted by repeating SEA for full set

of disasters divided into four regional groupings, and the by-crop composites were

obtained by repeating SEA on the full disaster sets using crop-specific agricultural

data from the FAO (http://faostat3.fao.org). Statistical significance of differences

between crop-specific, regional and earlier-versus-later composites was assessed

using the Kruskal–Wallis test. We applied a quadratic transformation to the data

for comparison to equalize variance between groups (verified using Levene’s test),

and used non-parametric tests when comparing groups as normal assumptions

were not met (see Supplementary Discussion).

Code availability.

All the core programs including codes to perform superposed

epoch analysis and the various statistics described in this paper are available on

Github (https://github.com/nramankutty/SEA-code).

Letter

reSeArCH

Extended Data Figure 1

|

Distributions of individual responses to

drought and extreme heat. a

–

f

, Histograms of disaster-year differences

from means of 1,000 resampled controls for drought (

n

=

222) (

a

–

c

) and

extreme heat (

n

=

32) (

d

–

f

). A preponderance of moderately negative

values (falling towards the right of the red shaded areas) underlies the

negative mean disaster year signals, with a limited influence of extreme

cases (those at the left of the red shaded areas).

Letter

reSeArCH

Extended Data Figure 2

|

The influence of sample size on estimated

disaster effects. a

,

b

, Estimated mean 16-cereal aggregated production

deficit for extreme heat (

a

) and drought (

b

) in 200 sub-samples with size

of (1, 2, …,

n

) (points). Dotted grey line shows the final estimated mean

production deficit (9.1% for extreme heat, 10.1% for drought). Most of the

initial variability at low sample sizes dissipates into the mean at well below

the actual sample size (

n

=

39 for extreme heat,

n

=

247 for drought).

Letter

reSeArCH

Extended Data Figure 3

|

Seasonal weather anomalies of drought and

extreme heat disasters in EM-DAT. a

–

c

, Normalized composite mean

growing season temperature for extreme heat (

n

=

32) (

a

) and drought

(

n

=

222) (

b

), and total precipitation for drought (

c

). Box plots depict the

distributions of 1,000 false-disaster control composites, with red crosses

denoting extreme outliers and red dashes denoting medians. Years with

extreme heat correspond to seasonal temperature anomalies of 1.2 °C,

while drought years have only 0.15 °C warmer temperatures, with no

significant precipitation anomaly.

Letter

reSeArCH

Extended Data Figure 4

|

Time series of the number of extreme heat and

drought disasters per year from the EM-DAT database.

The EM-DAT

database is based on a compilation of disaster reports gathered from various

organizations including United Nations agencies, governments and the

International Federation of Red Cross and Red Crescent Societies. The time

series of reported disasters per year exhibits an increasing trend, probably

the result of more complete disaster reporting in more recent decades with a

possible contribution from increasing disaster incidence. There is also large

inter-annual variability in the number of disasters.

Letter

reSeArCH

extended data table 1

|

Statistical significance of 16-cereal aggregate analysis

Percentage of points on control composites less than EWD composites for 16-cereal aggregate, 1,000 control replicates total.

Letter

reSeArCH

extended data table 2

|

Sample sizes for individual crop and 16-cereal aggregate analyses

Letter

reSeArCH

extended data table 3

|

Statistical significance of regional analysis

Percentage of points on control composites less than EWD composites for 16-cereal aggregate by region, 1,000 control replicates total.

Letter

reSeArCH

extended data table 4

|

Sample sizes for regional analysis

Letter

reSeArCH

extended data table 5

|

Statistical significance of individual crop analysis

Percentage of points on control composites less than EWD composites for individual crop analysis, 1,000 control replicates total.

Letter

reSeArCH

extended data table 6

|

Kruskal–Wallis assumptions test results for group comparison analyses

Lesk et al. (2016). Influence of extreme weather disasters on global crop production

Page 1

Page 2

Page 3

Page 4

Page 5

Page 6

Page 7

Page 8

Page 9

Page 10

Page 11

Page 12

Page 13

Page 14

Page 15

Document Outline