Kornhuber et al. (2023). Risks of synchronized low yields are underestimated in climate and crop model projections

Article

https://doi.org/10.1038/s41467-023-38906-7

Risks of synchronized low yields are

underestimated in climate and crop

model projections

Kai Kornhuber

1,2,3

, Corey Lesk

1,4

, Carl F. Schleussner

2,5

,

Jonas Jägermeyr

6,7,8

, Peter P

ﬂ

eiderer

2,5

& Radley M. Horton

1

Simultaneous harvest failures across major crop-producing regions are a
threat to global food security. Concurrent weather extremes driven by a
strongly meandering jet stream could trigger such events, but so far this has
not been quanti

ﬁ

ed. Speci

ﬁ

cally, the ability of state-of-the art crop and climate

models to adequately reproduce such high impact events is a crucial com-
ponent for estimating risks to global food security. Here we

ﬁ

nd an increased

likelihood of concurrent low yields during summers featuring meandering jets
in observations and models. While climate models accurately simulate atmo-
spheric patterns, associated surface weather anomalies and negative effects on
crop responses are mostly underestimated in bias-adjusted simulations. Given
the identi

ﬁ

ed model biases, future assessments of regional and concurrent

crop losses from meandering jet states remain highly uncertain. Our results
suggest that model-blind spots for such high-impact but deeply-uncertain
hazards have to be anticipated and accounted for in meaningful climate risk
assessments.

Extreme weather events like heatwaves, droughts and extreme pre-
cipitation can adversely impact crop production

and food security

,

Global warming is increasing the frequency and intensity

4

–

7

of weather

extremes and the likelihood of their simultaneous occurrence
globally

8

–

10

. Extremes occurring in close temporal vicinity

11

–

13

can lead to

outsized societal impacts, often beyond the sum of each extreme
occurring in isolation

14

. In particular, synchronized crop failures due to

simultaneous weather extremes across multiple breadbasket regions
pose a risk to global food security and food system supply chains

15

,

16

,

with potential disproportional impacts for import-dependent regions

2

,

3

.

Risks of synchronized breadbasket failures have been assessed on

a purely statistical basis

17

,

18

and in relation to dominant modes of

climate variability

16

that act on annual and seasonal timescales. In

the mid-latitudes, however, concurrent weather extremes are to a

large degree driven by the jet stream, the fast

ﬂ

owing winds in the

upper tropospheric mid-latitude circulation

19

–

21

. Speci

ﬁ

c summertime

circulation regimes in the jet

stream act as circumglobal

teleconnections

22

–

24

, promoting simultaneous heat

,

and rainfall

,

extremes with adverse effects on agricultural production across the
mid-latitudes

19

. In Northern Hemisphere (NH) summer, recurrent pat-

terns have been identi

ﬁ

ed as quasi-stationary Rossby waves with

wavenumbers 5 and 7 (wave-5 and wave-7 from hereon), where the
wavenumber refers to the number of ridges and troughs observed
within the mid-latitudes, whenever their amplitude is high. Such high
amplitude waves have been observed during major NH summer
weather extremes recently

25

,

27

,

28

.

While earlier analyses have shown the importance of atmospheric

wave patterns for local

29

or coinciding

19

,

30

extreme weather events,

Received: 29 April 2022

Accepted: 17 May 2023

Check for updates

1

Lamont-Doherty Earth Observatory, Columbia University, New York, USA.

2

Climate Analytics, Berlin, Germany.

3

German Council on Foreign Relations,

Berlin, Germany.

4

Department of Geography and Neukom Institute, Dartmouth College, Hanover, USA.

5

Integrative Research Institute on Transformations of

Human-Environment Systems (IRI THESys) and the Geography Department, Humboldt-Universität zu Berlin, Berlin, Germany.

6

Center for Climate Systems

Research, Columbia University, Climate School, NY, USA.

7

NASA GISS, Columbia University, New York, USA.

8

Potsdam Institute for Climate Impact Research,

Potsdam, Germany.

e-mail:

kk3397@columbia.edu

Nature Communications

| (2023) 14:3528

1

12

3

4

5

6

78

12

3

4

5

6

7

8

their impact on yield anomalies has to date only been quanti

ﬁ

ed on a

regional basis. The effect of such wave patterns on yield co-variability
between pairs of regions remains unquanti

ﬁ

ed in observations as well

as in model experiments. Purely model based risk assessments require
both crop models that show skill in reproducing observed extreme
weather yield responses with suf

ﬁ

cient accuracy and climate models

that reproduce observed relationships between wave patterns and
extreme weather. A recent analysis by Luo et al.

31

shows that the sur-

face imprint of wave patterns is largely underestimated in three cli-
mate models, while their phase position is reasonably well depicted.
Recent CMIP6 models and bias-adjusted model experiments, however,
have not been investigated. Thus, models on which recent risk
assessments rely still lack a thorough review. Further, the performance
of climate-crop models in simulating historical wave pattern-induced
concurrent low yields across regions remains unassessed. In addition,
potential future changes in wave patterns and associated surface
anomalies and yield co-variability impacts have not been quanti-

ﬁ

ed so far.

Here, we map the observed concurrence patterns of low yields in

major breadbasket regions associated with wave-5 and wave-7 patterns
and probe the skill of models to reproduce the observed relationships.
To do so, we use the most recent experiments from the Global Gridded
Crop Model Intercomparison (GGCMI

32

) over

ﬁ

ve key crop regions that

account for a large part of global maize and wheat production (maize:
~66%; wheat: ~70%

33

), and compare results based on historical model

experiments to observed crop data

33

. To evaluate if current climate and

crop models are suitable for credible multiple breadbasket risk assess-
ments, we provide results from crop models driven by observations and
crop models driven by bias-adjusted CMIP6 models (see data and
methods section for details) over the historical time period (1960

–

2014).

In doing so we aim at answering the following questions:

Estimating biases in climate models: How well are upper tropo-

spheric jet patterns and associated surface weather anomalies repro-
duced in bias-adjusted CMIP6 simulations?

Regional impacts on crop yields in observation and models: Can a

crop model driven by reanalysis data and bias-adjusted CMIP6 climate
simulations reliably reproduce observed regional crop yield anomalies
related to speci

ﬁ

c wave patterns?

Cross-regional impacts in observations and models: How do wave

events modulate the concurrence of low yields in major crop produ-
cing regions, and how do results from the crop model driven by rea-
nalysis and climate models compare to the observed signals of low
yield concurrence?

The Results section is organized around answering these three

categories of questions, primarily by analyzing observations and his-
torical climate and crop simulations. However, our methodology also
allows us to brie

ﬂ

y offer projections towards the end of each of the

three Results subsections, informed and in some cases tempered by
important caveats identi

ﬁ

ed in the comparison of historical simula-

tions to observations.

Results

Circulation patterns and associated surface anomalies

While upper tropospheric wave patterns are well reproduced in the
bias-adjusted CMIP6 multi-model mean, associated multi-model mean
temperature (Fig.

1

) and precipitation anomalies (Fig. S3, Table S1) are

largely underestimated. To determine these composite maps wave
events and anomaly

ﬁ

elds are calculated for each model separately

following the methodology of Kornhuber et al.

19

(see methods: wave

events) before averaging.

To quantify model agreement with reanalysis-data, we calculate the

Pearson correlation

r

and coef

ﬁ

cient of determination

R

2

of model-

based composite

ﬁ

elds (Fig.

1

b, c, f, g) with those based on ERA-5 (Fig.

1

a,

b) over the mid-latitudinal belt (38°

−

58°N). Note that

R

2

is calculated

between the values of the two respective

ﬁ

elds and not between values

and a

ﬁ

t as is usually done. We

ﬁ

nd high values for meridional winds

(

cor

w

7

v

,

hist

= 0

:

89,

R

2

= 0

:

89;

cor

w

7

v

,

ssp

585

= 0

:

92,

R

2

= 0

:

92

;

cor

w

5

v

,

hist

= 0.82,

R

2

= 0

:

76;

cor

w

5

v

,

ssp

585

= 0.84,

R

2

= 0

:

8), but low values for temperature

a

180° W

135°

W

90° W

45°

W

0°

45°

E

90° E

135°

E

b

180° W

13

5°

W

90° W

45°

W

0°

45°

E

90° E

135°

E

e

180° W

135

°W

90° W

45°

W

0°

45°

E

90° E

135°

E

c

180° W

135°

W

90° W

45°

W

0°

45°

E

90° E

13

5°

E

d

180° W

13

5°

W

90° W

45°

W

0°

45

°E

90° E

135°

E

h

180° W

13

5°

W

90° W

45°

W

0°

45°

E

90° E

13

5°E

f

180° W

13

5°

W

90° W

45°

W

0°

45

°E

90° E

135°

E

g

180° W

13

5°

W

90° W

45°

W

0°

45

°E

90° E

135°

E

2m T

emperature

Anomaly (K)

Fig. 1 | Circumglobal wave-7 and 5 patterns and associated 2 m air temperature
anomalies in ERA-5 reanalysis data and bias-adjusted CMIP6 models.

Mer-

idional winds in m/s (contours; purple: southerly, orange: northerly winds, in
(

a

–

c

,

e

–

g

) contours start at an absolute value of 3 m/s and increase/decrease by 3

respectively, in (

d

,

h

) contours start an absolute value of 0.5 and increase/decrease

by steps of one) and near surface temperature anomalies

ﬁ

lled contours during

(

a

–

c

) wave-7 and (

e

–

g

) wave- 5 events relative to the respective climatology in the

northern hemisphere summer (JJA) based on (

a

,

e

) ERA5 reanalysis (1960

–

2014),

(

b

,

f

) historical (1960

–

2014) and (

c

,

g

) future (SSP5-8.5, 2045

–

2099) bias-adjusted

output from CMIP6 simulations (four models).

d

,

h

) Difference in meridional winds

and temperature response during wave events comparing historical and future
patterns in four bias-adjusted CMIP6 models (for twelve non adjusted models see
Fig. S6). Hatching shows statistical signi

ﬁ

cance on a 95% con

ﬁ

dence level (

a

,

d

,

e

,

h

)

or 100% model agreement in sign (4 out of 4 models,

b

,

c

,

f

,

g

) While the phase

positions and intensity of the wave patterns (line contour) are well represented in
the models their surface imprint are considerably underestimated in historical
simulations. Changes in the temperature response are identi

ﬁ

ed over North

America, Eurasia and East Asia (

d

,

h

).

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

2

anomalies in the multi-model mean (

cor

w

7

t

2

m

,

hist

= 0

:

35,

R

2

= 0

:

14;

cor

w

7

t

2

m

,

ssp

585

= 0

:

38,

R

2

= 0

:

15

;

cor

w

5

t

2

m

,

hist

= 0.03,

R

2

= 0

:

02;

cor

w

5

t

2

m

,

ssp

585

=

0

:

51,

R

2

= 0

:

28) and even lower values for precipitation anomalies

(Table S1, Fig. S3). Luo et al.

31

found similar results for three models

partly based on CMIP5 experiments, suggesting that models might have
not considerably improved in the newer CMIP6 experiments. Notably,
bias-adjusted model outputs do not exhibit considerably improved
spatial correlation values and anomaly

ﬁ

elds compared to the original

CMIP6 simulations (Table S2, Figs. S4, S5), possibly because bias
adjustment optimizes

ﬁ

elds for different subsamples rather than high

amplitude wave events.

Circulation related changes in temperature during wave events

comparing historical and future experiments for four bias-adjusted
models are shown in Fig.

1

, d, h. Future extreme warming simulations

do not project an increase in mean wave amplitudes (Fig. S1), which
have a tendency to be underestimated in historical experiments
(Fig.

1

). This is in line with no detectable trend in wave events over the

historical time-period (Fig. S9).

However, ridges over certain regions do amplify in the future

projections and contribute to larger positive temperature anomalies in
the four bias-adjusted models (Fig.

1

d, h, see Fig. S6 for anomalies

based on all CMIP6 models). For wave-7, temperature anomalies
increase signi

ﬁ

cantly over Western North America (NA) and over East

Asia (EAS), while for wave-5 signi

ﬁ

cant increases are detected for

Western NA and most of Eurasia. The correlations with observed
temperature anomalies are higher for future compared to historical
simulations in part due to increased temperature anomalies beyond
mean warming alone.

Concerning crop producing regions, the majority of the climate

models analyzed here underestimate temperature and precipitation

anomalies during wave events (Fig.

2

). Underestimation of surface

anomalies in models during wave events may translate into particular
underestimated risk to crop productivity in key breadbasket regions.
To assess this, we calculate the spatially averaged surface response
in temperature and precipitation in important NH mid-latitude
crop regions (Fig.

2

a, see methods) during wave-7 and wave-5 events

(Fig.

2

b

–

k). We compare two different reanalysis datasets (ERA5,

1960

–

2014; W5E5, 1979

–

2014) to the bias-adjusted output from four

CMIP6 models and their multi-model mean under historic and future
simulations.

Strongest discrepancies between surface anomalies in reanalysis

and models are found where the crop-producing regions spatially align
with a wave-induced temperature anomaly, such as WEU for wave-7
and NA, WEU, and EEU for wave-5. For reanalysis, we

ﬁ

nd that NA shows

above average temperatures during wave-5 events, and dry anomalies
during wave-7 events. Strongest bivariate differences between wave
patterns occur over western European (WEU) croplands which are
wetter and colder during wave-5 events and hotter and drier during
wave-7 events. Eastern Europe (EEU) exhibits warmer than average
temperatures during wave-5 events and wetter than average condi-
tions during wave-7 events. India (IND) and Eastern Asia (EAS) are drier
than average during wave-7. Here, the reanalysis products covering the
recent years (W5E5, both 1979-2014) exhibit wetter and drier condi-
tions respectively, whereas ERA5 shows precipitation anomalies of
opposite sign to the other reanalyses. The bias-adjusted model output
does not exhibit a notable improvement relative to the other
CMIP6 models over the historical period (Fig.

2

, Fig. S7). Projected

relative anomalies however do move towards the observed values in
some regions, e.g. temperature anomalies for wave-5 in EEU and pre-
cipitation in IND for both waves (Fig.

2

, Fig. S5), which aligns with the

Fig. 2 | Mean response in precipitation and 2 m temperature anomalies over
major crop-producing regions during wave events in reanalysis data and
CMIP6 climate models. a

Major crop producing regions in the Northern Hemi-

sphere mid-latitudes de

ﬁ

ned by a threshold of 25% harvested area per grid-point.

Weekly mean temperature and aggregated precipitation anomalies averaged over
the regions outlined in (

a

) for (

b

–

f

) wave-7 and (

g

–

k

) for wave-5. We compare two

different reanalysis datasets ERA-5 (dark red,1960

–

2014) and W5E5 (red,

1979

–

2014) with bias-adjusted output from four CMIP6 models under historical

(green, 1960

–

2014) and future (2045

–

2099, SSP5-8.5, yellow) conditions, whereas

their mean values are shown as dashed lines. Note the different

y

-axis range for (

k

)

and (

f

) compared to the other panels. Temperature anomalies are dominantly

underestimated in the bias-adjusted output in WEU (wave-7, wave-5), EEU and NA
(wave-5). Precipitation anomalies are underestimated in NA, WEU (wave-7).

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

3

increased pattern correlation of future temperature anomaly

ﬁ

elds

identi

ﬁ

ed earlier (Table S1).

Observed and simulated regional yield losses

The occurrence of multiple wave events in summer negatively affects
combined maize and wheat yield at the regional and global level in
observational yield statistics (FAOSTAT

33

), but these impacts are not

accurately reproduced by the crop-model experiments in most
regions, especially when driven by bias-adjusted climate models
(Fig.

3

). We show this by comparing estimates of the waves

’

composite

impact on combined wheat and maize yield in observations (ERA5 x
FAO-data: Obs/Obs), a crop model driven by a reanalysis product
(W5E5 x LPJmL: Obs./Model) and the mean crop-model response from
four climate models (bias-adjusted CMIP6 x LPJmL; Model/Model)
under historical (Fig.

3

a, c) and future conditions (Fig.

3

b, d, see

methods for further details).

We

ﬁ

nd that years with more than one wave event are associated

with regional crop yield anomalies of up to

−

7% in EAS for,

−

6% to in NA

and

−

3% in EEU for wave-7 with an average response across the selected

regions of

−

2% to

−

3%. For wave-5 we observe values of

−

1% to

−

3% for

respectively with highest regional anomalies observed in EEU (

−

3.7%)

and NA (

−

2%) (Fig.

3

a, c Obs/Obs). These results are in agreement

with Kornhuber et al. 2020

19

, whose analysis relied on NCEP.NCAR

reanalysis over a shorter period (1979

–

2018) and different crop data.

Averaged across all regions, the response for wave-7 in CMIP6-driven
experiments is essentially zero, which represents an underestimation
of 3% globally relative to observations. While the yield anomalies
averaged across all regions are in good agreement for wave-5, regional
anomalies in models deviate from observed values: a 3% over-
estimation in NA compensates for the underestimation in all other

regions. Historic crop simulations underestimate the response in EAS
by 9% and by 3% in EEU (Fig.

3

c).

In some regions, disagreements between purely observation and

model-based assessments are substantially reduced when the crop
model is driven with bias-adjusted ERA5 reanalysis data (W5E5, Fig.

1

a

Obs/Model). While disagreements with observed values remain over
EAS and WEU, yield anomalies in NA, EEU and IND are found to be
well within the error-margins of the observed impact. This provides
evidence that surface anomaly biases in CMIP6 models are to a large
degree responsible for ensuing crop model disagreement with obser-
vations in some regions. However, we note that this reduction in dis-
agreement for the reanalysis-model hybrid is not consistent for all
regions, suggesting regional variation in the accuracy of the crop
model response to wave events.

Although linked climate and crop model simulations tend to

underestimate observed crop impacts, a comparison of historical and
projected future impacts might be still instructive given the large
societal risks associated with even a small percentage change in yield.
Future impacts on crop yield increase for wave-7 globally (Fig.

3

b, d),

and averaged over all regions, driven mainly by increased negative
impacts in NA where yield is reduced by 6% compared to historical
simulations, in agreement with the projected ampli

ﬁ

ed heat response

in that region in bias-adjusted model experiments (Fig.

1

d, Fig.

2

b).

Meanwhile, wave-5 projected impacts dampen at aggregated geo-
graphic scales, primarily driven by much-reduced impact over NA that
is only slightly offset by an increased impact over EEU. This increase
over EEU is in agreement with the ampli

ﬁ

ed temperature response

observed in the multi-model mean for that region (Figs.

1

h,

2

i). We note

that since the temperature and yield anomalies are normalized to
respective time periods, these comparisons isolate the in

ﬂ

uence of

Wave 7

World All Regs NA EEU WEU EAS IND

Wave 5

0 (n = 167) Hist.

0 (n = 130) Fut.

>1 (n = 55) Hist.

>1 (n = 50) Fut.

World All Regs NA EEU WEU EAS IND

)

%(

yl

a

m

o

n

a

dl

ei

y

eti

s

o

p

m

o

C

Wave 5

World All Regs NA EEU WEU EAS IND

Historical Experiments and Observations

Modelled Historical and Future Experiments

)

%(

yl

a

m

o

n

a

dl

ei

y

eti

s

o

p

m

o

C

Wave 7

World All Regs NA EEU WEU EAS IND

a

b

c

d

0 (n = 164) Hist.

0 (n = 136) Fut.

>1 (n = 39) Hist.

>1 (n = 50) Fut.

Fig. 3 | Combined Wheat and Maize yield anomalies during wave event years in
observations and models.

Composite yield anomalies based on wave events from

ERA-5 reanalysis and annual reported national yield statistics from FAO (Obs/Obs),
wave events from ERA-5 reanalysis and yield anomalies from a crop model (LPJmL)
driven by reanalysis data (W5E5, Obs/Model) and crop model (LPJmL) driven by
four bias-adjusted CMIP6 simulations (Model/Model) for (

a

) wave-7 and (

c

) wave-5

over the historical time period (1960

–

2014). Composites compare years in which

two or more wave events are detected in JJA (red to purple bars) with the control
case of years without such events (gray, light yellow bars). Bars and whiskers depict

the distribution of 500 resampled replicate composite yield effects, where each
replicated preserves the sample size of the underlying observations (wave events).
Differences in detected wave events across datasets cause the difference in dis-
tribution variance. Differences in modelled crop impacts (both Obs/Model and
Model/Model) are large compared to observations, but are smaller in some regions
when driving the crop model with bias-adjusted reanalysis weather data (Obs/
Model) instead of GCM simulations (Model/Model). (

b

,

d

) as in (

a

,

b

) but showing

crop yield anomalies simulated by LPJmL based on historical (1960

–

2014) and

future (SSP5-8.5, 2045

–

2099) simulations.

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

4

wave patterns relative to contemporaneous climate, masking large
mean warming and yield changes

32

along with their own attendant

uncertainties. Given the large biases in regional crop response to wave
events, these results have to be considered with caution.

When quantifying the crop response to extreme weather events,

their timing relative to the sowing and harvesting calendars is
crucial

34

,

35

. For instance, the lack of wave event impact for wave events

on WEU crop yield in observations might be related to dominance of
winter wheat in this region, with a growing season extending far out-
side of JJA. Furthermore, we

ﬁ

nd that the timing of wave events within

JJA differs between observations and CMIP6 (Fig. S8) presenting an
additional potential explanation (beyond surface anomalies) for yield
discrepancies between observations and CMIP6 models. While wave-5
events are evenly distributed over the summer season with a slight
preference for the later weeks in ERA-5, these events skew towards the
beginning of the season in the models, a tendency that is reinforced
under a high emission scenario (Fig. S8). While wave-7 events pre-
dominantly occur in June and early July in ERA-5, CMIP6 models sug-
gest a more evenly distributed occurrence of wave-7 events that reach
into August.

Observed and simulated concurrent yield losses

The occurrence of one or more wave event in summer elevates the
likelihood of poor harvests in pairs of two important crop producing
regions in observations (Fig.

4

a, e). To quantify the effect of wave

events on concurrent poor yields we introduce the

likelihood multi-

plication factor

(LMF) in analogue to e.g. Zscheischler & Seneviratne

36

(see Eq. (

3

) in Methods). The LMF is larger than one for a pair of two

regions when wave events amplify the likelihood of concurrent low
yields and lower than one when wave events have a lessening effect.

Here we de

ﬁ

ne a poor yield year as a year in which the combined wheat

and maize yields are below the multiyear trend. In analogue we test the
effect on concurrent yields above the long-term trends to test a
potential bene

ﬁ

cial effect on yield in two regions.

Results investigating concurrent negative anomalies are shown in

upper right tiles of heatmaps in Fig.

4

, while results on concurrent

positive anomalies are presented in lower left tiles. In experiments
based on observations (Obs./Obs., Fig.

4

a, e) wave events increase the

likelihood of concurrent negative yield anomalies in particular in
pairings that include NA while they mostly decrease the likelihood of
concurrent positive yield years overall. Both aspects are well repre-
sented in experiments based on LPJmL driven by reanalysis data
(Fig.

4

b, f), however values are higher for parings that include NA and

EAS for wave-7. Further discrepancies between observation-based and
modelled yields are found for pairings with WEU and EAS exhibiting
overestimated LMF values for wave-7, while being mostly under-
estimated for wave-5. We

ﬁ

nd that LMF values for concurrent low

yields and high yields differ statistically signi

ﬁ

cant for both experi-

ments (Obs./Obs., Obs./Model) and both waves throughout regions
(Figs. S12

–

S15). Historical CMIP-6 models (Fig.

4

c, g) exhibit a good

agreement with observations, in particular for wave-5 where three out
of four models agree in sign for most of the regions that have been
identi

ﬁ

ed as teleconnected. For wave-7 model agreement is strongest

for the link between for NA x EEU, while the co-occurrence of positive
yield anomalies is overestimated in some regions. This tendency
increases for wave-7 in future projections, where concurrent positive
yields dominate (Fig.

4

d). The only pairings that see an increase in

multi-model mean LMF for co-occurring negative yield anomalies are
NA x EAS for wave-7 and NA x EEU and IND x WEU for wave-5. Notably,
for future projections the four models exhibit poor agreement with

EAS

EEU

IND

NA

WEU

U

E

ES

A

E

A

ND

NI

WEU

1.1

0.57

1.1

1.6

1.7

0.89

1.7

0.67

1.8

0.89

1.1

0.67

1

1.6

1.1

0.33

0.57

1.3

a

Obs. / Obs.

EAS

EEU

IND

NA

WEU

2.1

1.4

6.5

2.7

0.8

0.83

8.7

1.1

1.2

1.4

1.2

0.62

0.5

0.55

0.68

3.7

0.85

0.9

1.4

0.78

b

Obs. / Model

EAS

EEU

IND

NA

WEU

1.1

0.62

0.82

1.7

0.98

2.5

1.6

1.3

0.88

1.5

0.86

0.81

1.2

1.8

0.96

1.5

0.96

1.2

0.88

c

Model / Model (hist.)

EAS

EEU

IND

NA

WEU

0.99

0.82

1.4

0.81

1

0.89

1.1

0.96

1.4

1.6

0.86

0.83

2.5

1

1.4

0.94

0.85

1.3

1.5

0.78

d

Model / Model (fut.)

0.00

0.25

0.50

0.75

1.00

1.25

1.50

1.75

2.00

EAS

EEU

IND

NA

WEU

U

E

S

A

E

A

ND

NI

WEU

1.5

1.2

3

2.3

0.5

1.8

2

1.2

1

2

2.5

1.5

0.75

0.5

1.5

0.75

0.2

1

0.33

e

Obs. / Obs.

EAS

EEU

IND

NA

WEU

0.75

1

10

0.92

0.37

1.8

7.8

0.92

0.83

0.99

7.4

1.1

0.39

0.14

0.2

2.8

0.83

0.38

0.68

0.36

f

Obs. / Model

EAS

EEU

IND

NA

WEU

1.4

1.9

2.7

1.9

0.52

6.2

1.3

6.4

0.75

1.3

2.3

1.1

0.45

0.71

1.1

1.3

0.74

1.6

1.1

0.74

g

Model / Model (hist.)

EAS

EEU

IND

NA

WEU

1.3

1.1

2

1.3

0.76

1.2

2.2

1.7

1.4

0.74

1.9

1.6

1.1

0.96

1

1.2

0.88

0.86

0.95

0.73

h

Model / Model (fut.)

0.00

0.25

0.50

0.75

1.00

1.25

1.50

1.75

2.00

Region 1

r

ot

c

a

F

n

oit

a

cil

pit

l

u

M

d

o

hil

e

ki

L

Region 2

r

ot

c

a

F

n

oit

a

cil

pit

l

u

M

d

o

hil

e

ki

L

Wave 7

Wave 5

Fig. 4 | Likelihood multiplication factors (LMF) of concurrent yield losses in
observations and models.

LMF of concurrent negative yield anomalies of com-

bined wheat and maize yield in two regions for wave-7 (

a

–

d

) and wave-5 events pear

year (

e

–

h

) (upper right corner). LMF for concurrent positive yield anomalies are

provided in the lower left corner of each heatmap.

a

,

e

Values based on ERA-5 and

FAO data (1960

–

2014, Obs./Obs.).

b

,

f

Values based on wave events from ERA-5 and

yield anomalies from LPJmL driven by W5E5 (Obs/Model.). LMF values for

concurrent low and high yields differ signi

ﬁ

cantly for (

a

,

b

,

e

,

f

) (Figs. S12

–

S15).

Averaged LMF values based on LPJmL driven by four bias-adjusted CMIP6 models
separately are shown for (

c

,

g

) historical experiments (1960

–

2014) and (

d

,

h

) future

projections (2045

–

2099, SSP5-8.5). Model agreement is provided by dots where

one dot indicates an agreement (above or below a LMF value of one) among three
out of four models while two dots indicate an agreement among all four models.

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

5

respect to changes in low yield concurrence (compare Figs. S10, S11).
While for wave-5 LMF values increase for MPI-ESM1-2-HR, values
decrease in all regions besides NA for UKESM1-0-LL. Such disagree-
ment in simulated concurrent yield loss across varying climate model
inputs complicates drawing clear conclusions on on how LMF values
will change in the future.

Discussion

Concurrent crop failures in major crop-producing regions constitute a
systemic risk as associated spikes in food prices can lead to con

ﬂ

ict

and undernutrition in countries that rely on imports

1

,

3

. Thus, under-

standing the likelihood of concurrent crop failures and the degree to
which models are able to reproduce observed relationships is impor-
tant for increasing the resilience of the global food system

15

and miti-

gating climate risks.

While circulation patterns associated with high amplitude Rossby

waves are accurately reproduced in climate models (Fig.

1

, Table S1),

the magnitude of surface temperature and precipitation anomalies is
largely underestimated, including in important crop producing
regions (Figs.

1

,

2

). Similar results have been reported for other climate

models mostly following the CMIP5 protocol

31

, however it is notable,

that bias-adjusted output from CMIP6 experiments do not exhibit
considerably improved spatial correlation and an accurate magnitude
in surface anomaly

ﬁ

elds (Fig.

1

, Tables S1, S2), possibly because bias

adjustment optimizes

ﬁ

elds for different subsamples rather than high

amplitude wave events.

Investigating future projections under a high emission scenario,

we

ﬁ

nd no consistent global increase in wave amplitudes in models

(Figs. S1, S2). This might be due to the fact that our wave diagnostic
is applied to the mid-latitudes (37.5

–

57.5°N) and is therefore not

sensitive to suggested

37

changes in sinuosity in higher latitudes where

the increased temperature gradient from increased land warming
increases zonal winds and improves waveguidability

37

,

38

. Instead, lower

wave amplitudes might be a consequence of the projected weakening
of summertime stormtracks, associated with an increase in weather
persistence

in the mid-latitudes

, which might lower the magni-

tude of meridional winds on which the wave diagnostic is based on
(see methods). We identify a regional ampli

ﬁ

cation of troughs and

ridges, in particular over the NA West coast and Eurasia (Fig.

1

d, h,

Fig.

2

b, i, f) for wave-7 and wave-5 respectively. Ampli

ﬁ

ed land-

atmosphere feedbacks which are acting on top of regional circulation
changes in a warmer climate

42

–

44

are other potential factors for

an increased regional temperature response. With the impacts of
recent extreme heat events

45

,

46

and associated wild

ﬁ

res

47

,

48

, such as the

severe Paci

ﬁ

c northwestern heatwave of 2021

45

and the extraordinary

Siberian heatwave of 2020

49

, these regions potentially emerge as high

risk areas.

We

ﬁ

nd that simultaneous extremes linked to a meandering jet

stream from ampli

ﬁ

ed Rossby waves

19

lead to regional yield losses

(Fig.

3

) and to concurrent low harvests across the mid-latitudes (Fig.

4

).

This increased likelihood of concurrent low yields in major breadbas-
kets, is mostly reproduced by historical model experiments, whether
driven by reanalysis data or climate models in particular for wave-5
(Fig.

4

). Regionally, however, we

ﬁ

nd that yield losses are mostly

underestimated in crop models driven by climate model output
(Fig.

3

), while crop models driven by reanalysis data show more

accurate responses.

The mostly adequate model representation of concurrent low

yields combined with predominant underestimation of local impacts
on yields, parallels the reasonable representation of modeled wave
patterns and an underestimation of associated surface anomalies in
bias-adjusted model output.

The biases in modelled crop yield response to wave event draw

the reliability of future projections into question, but the societal
importance of yield projections and the absence of a better approach

argue for discussion under consideration of identi

ﬁ

ed caveats.

Increased local impacts on yields are identi

ﬁ

ed in regions where future

surface anomalies are projected to increase (e.g. NA for wave-7 and
EAS for wave-5 (Fig.

2

b, I, Fig.

3

). However, the projected concurrence

of poor yields, is found to be less conclusive (Fig.

4

d, h), as models

show divergent responses to future warming scenarios (Figs. S15, S16).

With positive trends observed in magnitude and frequency of

extreme weather events, in particular for extreme heat, concurrent
weather extremes causing interconnected and potentially disruptive
impacts have become more common and will increase further if
greenhouse gas emissions remain unmitigated

. Informed adapta-

tion measures depend on models that simulate not just mean changes
but also the changes in complex low probability but high-risk scenarios
such as the concurrent and persistent extreme heat and rainfall
extremes as observed e.g. in the extreme summers of 2018

25

,

51

in Eur-

ope and Russia in 2010

28

,

52

, both with severe agricultural impacts

53

,

54

.

Assessing these complex risks depends upon an adequate repre-

sentation of the location, magnitude, frequency and sub-seasonal
distribution of extreme weather events, which may change under
future emission scenarios. Our results highlight how the evolution
of risks of multiple breadbasket failures under climate change are
characterized by deep uncertainty in part due to the insuf

ﬁ

cient

representation of the underlying climate impact drivers in models

55

.

Other major observational and modelling uncertainties regarding cli-
mate change risks to global crop production include the magnitude of
the CO2 fertilization effect and general changes in the hydrologic
cycle. Our results point to an additional modelling uncertainty with
import speci

ﬁ

cally for inter-regional crop yield covariability, which

have unique consequences for the global food system. Future work
should examine potential interactions among these key uncertainties,
particularly the potential modulation of jet-related hazards by mean
hydrologic and thermodynamic change.

While climate models have been excellent in projecting the mean

response to continued anthropogenic greenhouse gas emissions

56

, our

analysis suggests that they might provide a conservative estimate of
how concurrent extreme weather events driven by speci

ﬁ

c circulation

regimes might evolve in future and how they might affect regional
crop yield and covariability across regions. Further we highlight that
the underestimation of surface extremes identi

ﬁ

ed in CMIP5 models

31

and impacts on yields from their bias-adjusted output still persist in
most recent climate and crop simulations. Physically constrained
machine learning methods designed to maintain patterns and coher-
ence across variables might offer an effective tool for an improved bias
adjustment for more accurate impact assessments

57

.

Our study points towards potential high-impact blind spots in

current climate risk assessments, highlighting the urgent need for
more empirical and process-based research to support model
improvements in the climate and agriculture domains, supple-
mented by expert elicitation, qualitative storylines

58

, and decision-

centric approaches

59

. Evidence for high-risk blind spots such as an

underestimation of synchronized harvest failures as identi

ﬁ

ed here,

manifests the urgency of rapid emission reductions, lest climate
extremes and their complex interactions might increasingly become
unmanageable.

Methods

Data

ERA5

60

reanalysis data for years (1960

–

2014) data using the recently

published back-extension was downloaded from the Copernicus Data
Store (

https://cds.climate.copernicus.eu/#!/home

). Daily mean

ﬁ

elds

of 2 m temperature, precipitation and meridional winds are based on
hourly data at timesteps 0:00, 6:00, 12:00 and 18:00. Downloaded at a
0.25 × 0.25 resolution, the data was re-gridded to 1 × 1 resolution.
Temperature and precipitation data was detrended by subtracting
linear trends on a grid-point basis over the investigated period and a

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

6

subsequent subtraction of the day-in-year climatological value over
the same period in order to remove the seasonal cycle. For tempera-
ture and meridional winds weekly anomalies are calculated by aver-
aging daily mean data centered around the day for which a high
amplitude is detected. For precipitation, weekly anomalies are calcu-
lated by a 7-day aggregation of daily anomalies centered around the
day for which an event is detected. Meridional winds are analyzed a
250 mb level.

W5E5

61

is the primary input data for the ISIMIP3-project, is based

on bias-adjusted ERA5 data for the period 1979-2014 and GSWP3
homogenized to W5E5 for years 1960-1978 and was acquired from the
internal repository of the Potsdam Institute for Climate Impact
research. For a more detailed description about the bias adjustment
process see

62

and

https://www.isimip.org/gettingstarted/input-data-

bias-correction/details/80/

. Wave events for Fig.

2

and Figs. S4, S5 are

based on ERA5 wind-

ﬁ

elds. As years prior to 1979 are not based on

ERA5 we limit the analysis to years 1979

–

2014.

CMIP6

63

daily 2-meter temperature, precipitation and meridional

wind

ﬁ

elds were retrieved from the

google data store

and re-gridded to

1 × 1° resolution for years 1960

–

2014 (historical runs) and 2045-2099

(SSP5-8.5). Temperature and precipitation

ﬁ

elds were detrended fol-

lowing the approach described above. Precipitation

ﬁ

elds from CMIP6

were converted from precipitation

ﬂ

ux (kg m

−

2

s

−

1

) to meter per unit

cell per hour by multiplying each value with a factor of 3.6. Models
included in this analysis were chosen based on the availability of
meridional wind

ﬁ

elds on daily time-scale for both time-periods

investigated. The following twelve models were included: BCC-CSM2-
MR, CESM2-WACCM, CNRM-CM6-1, CNRM-ESM21, CanESM5, EC-
Earth3-Veg, GFDL-CM4, IPSL-CM6A-LR, MIROC6, MPI-ESM1-2-HR, MRI-
ESM2-0, UKESM1-0-LL.

Bias-adjusted models and yield-data

We identify maize and wheat responses to ampli

ﬁ

ed wave patterns

based on a 3-tier protocol: (i) FAO

33

crop yield statistics compared with

reanalysis weather data, (ii) crop model-based yield estimates driven
by reanalysis weather data, (iii) crop model-based yield estimates
based on 4 downscaled and bias-adjusted CMIP6 climate models. Our
analyses leverage crop model simulations facilitated by AgMIP

’

s Global

Gridded Crop Model Intercomparison (GGCMI)

32

. Here we use global

LPJmL simulations at 0.5° for maize and wheat driven by W5E5

61

rea-

nalysis data (1960-2014) and climate model simulations from 4
CMIP6 GCMs (2045-2099), bias-adjusted and downscaled by ISIMIP

64

(IPSL-CM6A-LR, MPI-ESM1-2-HR, MRI-ESM2-0, UKESM1-0-LL) to esti-
mate the effect of high amplitude waves on crop yield anomalies. For
more details on the crop modelling protocol, see Jägermeyr et al.

32

.

Crop producing regions and regional aggregation: Crop-

producing regions in the northern hemisphere mid-latitudes are
de

ﬁ

ned similar to prior studies: North America (United States,

Canada), Western Europe (France, Switzerland, Spain, Portugal, United
Kingdom, Belgium, Netherlands, Germany), Eastern Europe (Greece,
Bulgaria, Moldova, North Macedonia, Ukraine, Romania, Serbia, Alba-
nia, Russia), India and Eastern Asia (China, Mongolia). Grid-points
relevant for crop production for the analysis shown in Fig.

2

are

de

ﬁ

ned by a grid-point based threshold of 25 % harvested area

fraction based on 2005

ﬁ

elds from Ray et al.

65

. We further include an

‘

all regions

’

grouping including all wave event-affected countries, as

well as global total yield levels as a

‘

world region

’

which refers to the

global yield level in a given year. We aggregate crop yield data to the
regional scale by summing national FAO production and divide it by
total harvested area in each year, and summing gridded modeled crop
production data, which are based on static harvested area over time.
Because the crop model cannot simulate annual variations in har-
vested area, we strictly compare simulated yield (i.e., productivity)
with observational yield levels from FAO. We then simply detrend
modeled and observed crop yields using singular spectrum analysis

(SSA), a non-parametric method that avoids assumptions about the
functional form of the climate and yield trends

66

. This isolates inter-

annual yield variability from longer-term trends, and thus quanti

ﬁ

es

yield variations relative to ~10-year moving windows. Thus, the com-
posites re

ﬂ

ect yield variations relative to contemporaneous yields,

enabling us to composite wave event impacts relative larger long-term
mean changes in yield.

Wave analysis and wave event de

ﬁ

nition

We analyze wave patterns following the approach from Kornhuber
et al.

19

, decomposing weekly averages of the weighted mean of mid-

latitude (37.5°N

–

57.5°N), meridional wind

ﬁ

elds (250 mb) with a fast

Fourier decomposition in June-August (JJA). The 92 days in the JJA yield
13 weeks. The weekly averaging assures some degree of anti-aliasing,
and serves as a low-pass

ﬁ

lter. High amplitude wave events are

detected when the weekly amplitude exceeds a value of 1.5 standard
deviation above the mean relative to the respective wave

’

s JJA clima-

tology. As events are based on a threshold that varies with the
respective climatology, we

ﬁ

nd similar event numbers for each dataset.

For ERA5 that is 54 wave-7 events and 62 wave-5 for JJA 1960

–

2014

(see Fig. S9 for number of events per year for each wave).

Composite analysis

For each tier in the protocol described above, we use a compositing
method similar to Kornhuber et al.

19

to estimate the regional compo-

site yield anomalies for both years with >1 wave event, and years with
no wave events (the

‘

control

’

). The method isolates the signal of spe-

ci

ﬁ

c events in noisy time series by averaging anomalies across the

sample of either >1 or 0 wave event years within each tier of the pro-
tocol (i.e., observed or modeled future/historical wave events). Prior to
averaging regional yield values for the sample of >1 or 0 event years,
the values are divided by the average yield values for the three years
preceding and following the event, generating standardized compo-
sites (in terms of % of regional production). To estimate uncertainty in
this point estimate of the standardized composite yield anomaly, we
re-estimate 500 replicates of the composites using ~90% of the full
sample. This bootstrap approach yields a distribution of resampled
composites, which we depict as box-and-whisker plots. Values men-
tioned in the results re

ﬂ

ect median composite estimates across the

500 replicates. Further details on the compositing method can be
found in Kornhuber

19

.

Likelihood multiplication factor and de

ﬁ

nition of poor

yield years

The LMF for a speci

ﬁ

c wave and a pair of regions is determined by

ﬁ

rst

calculating the number of years that feature concurrent below average
yields

y1

and

y2

in those speci

ﬁ

c regions conditioned on the occur-

rence of more than one wave event of the speci

ﬁ

c wavenumber

N

y

1<0,

y

2<0

_

w

>1

ð

Þ

divided by the total number of years that feature more

than one wave event per year

N

w

>1

:

p

w

>1

=

N

y

1<0

\

y

2<0

_

w

>1

ð

Þ

=

N

w

>1

ð

1

Þ

To determine the LMF we calculate the ratio of

p

w

>1

and

p

0

=

N

y

1<0

\

y

2<0

_

w

= 0

ð

Þ

=

N

w

= 0,

ð

2

Þ

the

‘

hit rate

’

of concurrent low yields conditioned on the occurrence of

no wave event:

LMF =

p

w

>1

=

p

0

:

ð

3

Þ

Thus, LMF provides a factor by which the occurrence of wave

events increases the occurrence of concurrent low yields in two
regions, relative to years in which no events occur. In Fig.

4

we provide

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

7

results on concurrent high yields in analogue to this approach. Poor or
good yields years are de

ﬁ

ned as those years in which a region exhibits

combined wheat and maize yields below or above the multiyear trend
respectively. This moderate threshold ensures a more reliable statistic
through a higher sample size. The following example is based on ERA-5
and FAO data and regions WEU and NA and wave 7: we

ﬁ

nd

N

y

1<0,

y

2<0

_

w

>1

ð

Þ

= 6 years where yields are below the long-term trend in

both regions that also feature more than one wave event. Throughout
the dataset

N

w

>1

= 15 years are identi

ﬁ

ed in which more than one wave

event occurs, thus:

p

w

>1

=

6

15

. Of those, only one event shows con-

current positive yield anomalies. In contrast, we

ﬁ

nd

N

w

= 0

= 20 zero

event years of which

N

y

1<0

\

y

2<0

_

w

= 0

ð

Þ

= 5 feature concurrent low yields.

Thus

p

0

=5/20. For wave-7 and NA x WEU we

ﬁ

nd:

LMF

=

6

15

20

5

= 1

:

6

:

During years with more than one wave-7 event the probability of

concurrent low yields is increased by a factor of 1.6 or 60% compared
to years with no wave-7 event (see Fig.

4

a).

Data availability

Reanalysis and CMIP6 datasets used in this study are publicly available
at

https://cds.climate.copernicus.eu/#!/home

and

https://esgf-node.

llnl.gov/search/cmip6/

respectively, the GGCMI datasets are available

at

https://agmip.org/aggrid-ggcmi/

.

Code availability

All scripts relevant for the presented analysis will be made available at
GitHub:

https://github.com/KaiKornhuber/BreadBaskets_code.git

.

References

1.

Lesk, C., Rowhani, P. & Ramankutty, N. In

ﬂ

uence of extreme

weather disasters on global crop production.

Nature

529

,

84

–

87 (2016).

2.

Puma, M. J., Bose, S., Chon, S. Y. & Cook, B. I. Assessing the evolving
fragility of the global food system.

Environ. Res Lett.

10

,

024007 (2015).

3.

Bren d

’

Amour, C., Wenz, L., Kalkuhl, M., Christoph Steckel, J. &

Creutzig, F. Teleconnected food supply shocks.

Environ. Res Lett.

11

, 035007 (2016).

4.

Horton, R. M., Mankin, J. S., Lesk, C., Coffel, E. & Raymond, C. A
review of recent advances in research on extreme heat events.

Curr.

Clim. Chang Rep.

2

, 242

–

259 (2016).

5.

Lehmann, J., Coumou, D. & Frieler, K. Increased record-breaking
precipitation events under global warming.

Clim. Change

132

,

501

–

515 (2015).

6.

IPCC.

Managing the Risks of Extreme Events and Disasters to

Advance Climate Change Adaptation

. (eds. Field, C. B., Barros, V.,

Stocker, T. F. & Dah

ś

e, Q., eds.) (Cambridge University Press, 2012).

https://doi.org/10.1017/CBO9781139177245

.

7.

Meehl, G. A. & Tebaldi, C. More intense, more frequent, and longer
lasting heat waves in the 21st century.

Science

305

,

994

–

997 (2004).

8.

Sarhadi, A., Ausín, M. C., Wiper, M. P., Touma, D. & Diffenbaugh N. S.
Multidimensional risk in a nonstationary climate: joint probability
of increasingly severe warm and dry conditions.

Sci Adv

.

4

.

https://

doi.org/10.1126/sciadv.aau3487

(2018).

9.

Rogers, C. D. W., Kornhuber, K., Perkins-Kirkpatrick, S. E., Loikith, P.
C. & Singh, D. Six-fold increase in historical Northern Hemisphere
concurrent large heatwaves driven by warming and changing
atmospheric circulations.

J Clim

. 1

–

39.

https://doi.org/10.1175/jcli-

d-21-0200.1

(2021).

10. Singh, J., Ashfaq, M., Skinner, C. B., Anderson, W. B. & Singh, D.

Ampli

ﬁ

ed risk of spatially compounding droughts during

co-occurrences of modes of natural ocean variability.

npj Clim.

Atmos. Sci.

4

, 1

–

14 (2021).

11.

Ortiz-Bobea, A., Wang, H., Carrillo, C. M. & Ault, T. R. Unpacking the
climatic drivers of US agricultural yields.

Environ. Res. Lett

.

14

.

https://doi.org/10.1088/1748-9326/ab1e75

(2019).

12. Rigden, A. J., Mueller, N. D., Holbrook, N. M., Pillai, N. & Huybers, P.

Combined in

ﬂ

uence of soil moisture and atmospheric evaporative

demand is important for accurately predicting US maize yields.

Nat.

Food

1

, 127

–

133 (2020).

13. Lesk, C. et al. Stronger temperature

–

moisture couplings exacer-

bate the impact of climate warming on global crop yields.

Nat. Food

2

, 683

–

691 (2021).

14. Raymond, C. et al. Understanding and managing connected

extreme events.

Nat. Clim. Chang

10

, 611

–

621 (2020).

15. Mehrabi, Z. Food system collapse.

Nat. Clim. Chang

10

,

16

–

17 (2020).

16. Anderson, W. B., Seager, R., Baethgen, W., Cane, M. & You, L.

Synchronous crop failures and climate-forced production varia-
bility.

Sci. Adv.

5

, 1

–

10 (2019).

17. Gaupp F., P

ﬂ

ug, G., Hochrainer-stigler, S., Hall, J. & Dadson, S.

Dependency of crop production between global breadbaskets: a
copula approach for the assessment of global and regional risk
pools.

Risk Anal.

37

, 2212

–

2228 (2017).

18. Tigchelaar, M., Battisti, D. S., Naylor, R. L. & Ray, D. K. Future

warming increases probability of globally synchronized maize
production shocks.

Proc. Natl Acad. Sci. USA

115

,

6644

–

6649 (2018).

19. Kornhuber, K. et al. Ampli

ﬁ

ed Rossby waves enhance risk of con-

current heatwaves in major breadbasket regions.

Nat. Clim. Chang.

10

, 48

–

53 (2020).

20. Saeed, S., Van Lipzig, N., Müller, W. A., Saeed, F. & Zanchettin, D.

In

ﬂ

uence of the circumglobal wave-train on European summer

precipitation.

Clim. Dyn.

43

, 503

–

515 (2014).

21. White, R. H., Kornhuber, K., Martius, O. & Wirth, V. From atmo-

spheric waves to heatwaves: a waveguide perspective for under-
standing and predicting concurrent, persistent and extreme
extratropical weather.

Bull. Am. Meteorol. Soc

. 2021:1-35.

https://

doi.org/10.1175/bams-d-21-0170.1

22. Teng, H., Branstator, G., Wang, H., Meehl, G. A. & Washington, W. M.

Probability of US heat waves affected by a subseasonal planetary
wave pattern.

Nat. Geosci.

6

, 1056

–

1061 (2013).

23. Branstator, G. Circumglobal teleconnections, the jet stream wave-

guide, and the North Atlantic Oscillation.

J. Clim.

15

,

1893

–

1910 (2002).

24. Ding, Q. & Wang, B. Circumglobal teleconnection in the Northern

Hemisphere summer.

J. Clim.

18

, 3483

–

3505 (2005).

25. Kornhuber, K. et al. Extreme weather events in early summer 2018

connected by a recurrent hemispheric wave-7 pattern.

Environ. Res.

Lett

.

14

, 054002 (2019).

26. Wang, S.-Y., Davies, R. E. & Gillies, R. R. Identi

ﬁ

cation of extreme

precipitation threat across midlatitude regions based on short-wave
circulations.

J. Geophys. Res. Atmos.

118

, 11059

–

11074 (2013).

27. Mitchell, D., Kornhuber, K., Huntingford, C. & Uhe, P. The day the

2003 European heatwave record was broken.

Lancet Planet Heal

3

,

e290

–

e292 (2019).

28. Di Capua, G. et al. Drivers behind the summer 2010 wave train

leading to Russian heatwave and Pakistan

ﬂ

ooding.

npj Clim. Atmos

Sci

.

4

.

https://doi.org/10.1038/s41612-021-00211-9

(2021).

29. Screen, J. A. & Simmonds, I. Ampli

ﬁ

ed mid-latitude planetary waves

favour particular regional weather extremes.

Nat. Clim. Chang.

4

,

704

–

709 (2014).

30. Coumou, D., Petoukhov, V., Rahmstorf, S., Petri, S. & Schellnhuber,

H. J. Quasi-resonant circulation regimes and hemispheric syn-
chronization of extreme weather in boreal summer.

Proc. Natl Acad.

Sci.

111

, 12331

–

12336 (2014).

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

8

31. Luo, F. et al. Summertime circumglobal Rossby waves in climate

models: small biases in upper-level circulation create substantial
biases in surface imprint.

Weather Clim. Dyn.

3

, 1

–

30 (2021).

32. Jägermeyr, J. et al. Climate impacts on global agriculture emerge

earlier in new generation of climate and crop models.

Nat. Food

2

.

https://doi.org/10.1038/s43016-021-00400-y

(2021).

33. Food and Agriculture Organization of the United Nations.

FAOSTAT

Statistical Database

(Food and Agriculture Organization of the

United Nations, 1997).

34. Jägermeyr, J. & Frieler, K. Spatial variations in crop growing seasons

pivotal to reproduce global

ﬂ

uctuations in maize and wheat yields.

Sci Adv

.

4

.

https://doi.org/10.1126/sciadv.aat4517

(2018).

35. Butler, E. E. & Huybers, P. Variations in the sensitivity of US maize

yield to extreme temperatures by region and growth phase.

Environ

Res Lett

.

10

.

https://doi.org/10.1088/1748-9326/10/3/

034009

(2015).

36. Zscheischler, J. & Seneviratne, S. I. Dependence of drivers affects

risks associated with compound events.

Sci. Adv.

3

, e1700263 (2017).

37. Xu, P. et al. Ampli

ﬁ

ed waveguide teleconnections along the polar

front jet favor summer temperature extremes over Northern Eurasia.

Geophys. Res. Lett

.

48

.

https://doi.org/10.1029/2021GL093735

(2021).

38. Mann, M. E., et al. Projected changes in persistent extreme summer

weather events: the role of quasi-resonant ampli

ﬁ

cation.

Sci. Adv.

4

, 1

–

10 (2018).

39. Kornhuber, K. & Tamarin-Brodsky, T. Future changes in Northern

Hemisphere summer weather persistence linked to projected Arc-
tic warming.

Geophys Res Lett.

48

, 1

–

12 (2021).

40. P

ﬂ

eiderer, P., Schleussner, C-F, Kornhuber, K. & Coumou, D. Sum-

mer weather becomes more persistent in a 2C world.

Nat. Clim.

Chang

.

9

, 666

–

671 (2019).

41. Harvey, B. J., Cook, P., Shaffrey, L. C. & Schiemann, R. The response

of the Northern Hemisphere storm tracks and jet streams to climate
change in the CMIP3, CMIP5, and CMIP6 climate models.

J. Geo-

phys Res Atmos.

125

, 1

–

10 (2020).

42. Berg, A. et al. Impact of soil moisture-atmosphere interactions on

surface temperature distribution.

J. Clim.

27

, 7976

–

7993 (2014).

43. Miralles, D. G., Teuling, A. J., van Heerwaarden, C. C. & Vilà-Guerau

de Arellano, J. Mega-heatwave temperatures due to combined soil
desiccation and atmospheric heat accumulation.

Nat. Geosci.

7

,

345

–

349 (2014).

44. Mueller, N. D. et al. Cooling of US Midwest summer temperature

extremes from cropland intensi

ﬁ

cation.

Nat. Clim. Chang.

6

,

317

–

322 (2016).

45. Philip, S. Y., Kew, S. F., Oldenborgh & Van, G. J. et al. Rapid attri-

bution analysis of the extraordinary heatwave on the Paci

ﬁ

c Coast

of the US and Canada June 2021.

World Weather Attrib

. 119

–

123

(2021).

https://www.worldweatherattribution.org/wp-content/

uploads/NW-US-extreme-heat-2021-scienti

ﬁ

c-report-WWA.pdf

.

46. Li, J. & Thompson, D. W. J. Widespread changes in surface tem-

perature persistence under climate change.

Nature

599

,

425

–

430 (2021).

47. Yasunari T. J., et al. Relationship between circum-Arctic atmo-

spheric wave patterns and large-scale wild

ﬁ

res in boreal summer.

Environ Res Lett

.

16

.

https://doi.org/10.1088/1748-9326/

abf7ef

(2021).

48. Petoukhov, V. et al. Alberta wild

ﬁ

re 2016: Apt contribution from

anomalous planetary wave dynamics.

Sci. Rep.

8

, 12375 (2018).

49. Ciavarella, A. et al. Prolonged Siberian heat of 2020 almost

impossible without human in

ﬂ

uence.

Clim. Change

166

, 1

–

18 (2021).

50. Raymond, C. et al. Increasing spatiotemporal proximity of heat and

precipitation extremes in a warming world quanti

ﬁ

ed by a large

model ensemble.

Environ. Res Lett.

17

, 035005 (2022).

51. Vogel, M. M., Zscheischler, J., Wartenburger, R., Dee, D. &

Seneviratne, S. I. Concurrent 2018 hot extremes across Northern

Hemisphere due to human-induced climate change.

Earth

’

s Futur.

7

, 692

–

703 (2019).

52. Lau, W. K. M. & Kim, K.-M. The 2010 Pakistan

ﬂ

ood and Russian heat

wave: teleconnection of hydrometeorological extremes.

J. Hydro-

meteorol.

13

, 392

–

403 (2012).

53. Welton, G.

The Impact of Russia

’

s 2010 Grain Export Ban

(2011).

54. Beillouin D., Schauberger, B., Bastos, A., Ciais, P. & Makowski, D.

Impact of extreme weather conditions on European crop pro-
duction in 2018: Random forest

—

yield anomalies.

Philos. Trans. R.

Soc. B: Biol. Sci

.

375

.

https://doi.org/10.1098/rstb.2019.0510

(2020).

55. Masson-Delmotte V., et al.

IPCC, 2021: Climate Change 2021: The

Physical Science Basis. Contribution of Working Group I to the Sixth
Assessment Report of the Intergovernmental Panel on Climate
Change

(2021).

56. Hausfather, Z., Drake, H. F., Abbott, T. & Schmidt, G. A. Evaluating

the performance of past climate model projections.

Geophys Res

Lett.

47

, 1

–

10 (2020).

57. Hess, P., Drüke, M., Petri, S., Strnad, F. M. & Boers, N. Physically

constrained generative adversarial networks for improving pre-
cipitation

ﬁ

elds from Earth system models.

Nat. Mach. Intell.

4

,

828

–

839 (2022).

58. Zappa, G. & Shepherd, T. G. Storylines of atmospheric circulation

change for European regional climate impact assessment.

J. Clim.

30

, 6561

–

6577 (2017).

59. Weaver, C. P. et al. Improving the contribution of climate model

information to decision making: the value and demands of robust
decision frameworks.

WIREs Clim. Chang.

4

, 39

–

60 (2013).

60. Hersbach, H. et al. The ERA5 global reanalysis.

Q J. R. Meteorol. Soc.

146

, 1999

–

2049 (2020).

61. Lange S. WFDE5 over land merged with ERA5 over the ocean

(W5E5).

V 10 GFZ Data Serv

https://doi.org/10.5880/pik.2019.

023

(2019).

62. Cucchi, M. et al. WFDE5: bias adjusted ERA5 reanalysis data for

impact studies.

Earth Syst. Sci. Data

12

, 1

–

32 (2020).

63. Eyring, V. et al. Overview of the Coupled Model Intercomparison

Project Phase 6 (CMIP6) experimental design and organization.

Geosci. Model Dev.

9

, 1937

–

1958 (2016).

64. Lange S. Trend-preserving bias adjustment and statistical down-

scaling with ISIMIP3BASD (v1.0).

Geosci. Model Dev. Discuss

. 1

–

24.

https://doi.org/10.5194/gmd-2019-36

(2019).

65. Ray, D. K., Gerber, J. S., MacDonald, G. K. & West, P. C. Climate

variation explains a third of global crop yield variability.

Nat. Com-

mun.

6

, 5989 (2015).

66. Vautard, R., Yiou, P. & Ghil, M. Singular-spectrum analysis: a toolkit

for short, noisy chaotic signals.

Phys. D: Nonlinear Phenom.

58

,

95

–

126 (1992).

Acknowledgements

The authors want to thank the ECMWF and the Copernicus Climate
Change Service for providing the ERA5 reanalysis and World Cli-
mate Research Programme, which, through its Working Group on
Coupled Modelling, coordinated and promoted CMIP6. We thank
the climate modeling groups for producing and making available
their model output, the Earth System Grid Federation (ESGF) for
archiving the data and providing access, and the multiple funding
agencies who support CMIP6 and ESGF. K.K. was partially sup-
ported by the NSF project NSF AGS-1934358. K.K. and R.H. were
both in part supported by the NOAA RISA Program award
NA20OAR4310147A. C.L. was supported by National Science
Foundation Graduate Research Fellowship under Grant No. DGE

—

1644869. PP and CFS acknowledge support by the German Federal
Ministry of Education and Research (01LN1711A) and European
Union

’

s Horizon 2020 Research and Innovation Programme under

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

9

grant no. 101003687 (PROVIDE). J.J. was supported by the NASA
GISS Climate Impacts Group and the Open Philanthropy Project.

Author contributions

K.K. conceived and designed the research, analyzed the data and wrote
the

ﬁ

rst draft of the manuscript. C.L. analyzed the data. J.J. and P.P.

provided the data. C.F.S., R.H. and all authors helped with the design of
the research and writing of the manuscript.

Competing interests

The authors declare no competing interests.

Additional information

Supplementary information

The online version contains

supplementary material available at

https://doi.org/10.1038/s41467-023-38906-7

.

Correspondence

and requests for materials should be addressed to Kai

Kornhuber.

Peer review information

Nature Communications

thanks the anon-

ymous reviewer(s) for their contribution to the peer review of this work.

Reprints and permissions information

is available at

http://www.nature.com/reprints

Publisher

’

s note

Springer Nature remains neutral with regard to jur-

isdictional claims in published maps and institutional af

ﬁ

liations.

Open Access

This article is licensed under a Creative Commons

Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as
long as you give appropriate credit to the original author(s) and the
source, provide a link to the Creative Commons license, and indicate if
changes were made. The images or other third party material in this
article are included in the article

’

s Creative Commons license, unless

indicated otherwise in a credit line to the material. If material is not
included in the article

’

s Creative Commons license and your intended

use is not permitted by statutory regulation or exceeds the permitted
use, you will need to obtain permission directly from the copyright
holder. To view a copy of this license, visit

http://creativecommons.org/

licenses/by/4.0/

.

Article

https://doi.org/10.1038/s41467-023-38906-7

Nature Communications

| (2023) 14:3528

10

Kornhuber et al. (2023). Risks of synchronized low yields are underestimated in climate and crop model projections

Page 1

Page 2

Page 3

Page 4

Page 5

Page 6

Page 7

Page 8

Page 9

Page 10

Document Outline