Climate change signal in global agriculture emerges
earlier in new generation of climate and crop
models
Jonas Jaegermeyr (
jonas.jaegermeyr@columbia.edu )
https://orcid.org/0000-0002-8368-0018
Christoph Müller
Potsdam Institute for Climate Impact Research
https://orcid.org/0000-0002-9491-3550
Alex Ruane
NASA Goddard Institute for Space Studies
https://orcid.org/0000-0002-5582-9217
Joshua Elliott
Columbia University
Juraj Balkovic
International Institute for Applied Systems Analysis (IIASA)
Oscar Castillo
University of Florida
Babacar Faye
Institut de recherche pour le développement
Ian Foster
University of Chicago
Christian Folberth
International Institute for Applied Systems Analysis
https://orcid.org/0000-0002-6738-5238
James Franke
University of Chicago
Kathrin Fuchs
Karlsruhe Institute of Technology
Jose Guarin
https://orcid.org/0000-0002-3167-4329
Jens Heinke
Potsdam Institute for Climate Impacts Research
Gerrit Hoogenboom
https://orcid.org/0000-0002-1555-0537
Toshichika Iizumi
National Agriculture and Food Research Organization
https://orcid.org/0000-0002-0611-4637
Atul Jain
University of Illinois at Urbana-Champaign
https://orcid.org/0000-0002-4051-3228
David Kelly
University of Chicago
Nikolay Khabarov
International Institute for Applied Systems Analysis
https://orcid.org/0000-0001-5372-4668
Stefan Lange
Potsdam Institute for Climate Impact Research (PIK), Member of the Leibniz Association
https://orcid.org/0000-0003-2102-8873
Tzu-Shun Lin
University of Illinois at Urbana-Champaign
Wenfeng Liu
China Agricultural University
Oleksandr Mialyk
https://orcid.org/0000-0002-7495-2325
Sara Minoli
PIK
Elisabeth Moyer
University of Chicago
https://orcid.org/0000-0003-1829-5196
Masashi Okada
National Institute for Environmental Studies
Meridel Phillips
Columbia University
Cheryl Porter
https://orcid.org/0000-0001-7269-6543
Sam Rabin
Karlsruhe Institute of Technology
Clemens Scheer
Karlsruhe Institute of Technology
Julia Schneider
Ludwig-Maximilians-Universität München
Joep Schyns
https://orcid.org/0000-0001-5058-353X
Rastislav Skalský
International Institute for Applied Systems Analysis
https://orcid.org/0000-0002-0983-6897
Andrew Smerald
Karlsruhe Institute of Technology
Tommaso Stella
Leibniz Centre for Agricultural Landscape Research
https://orcid.org/0000-0002-3018-6585
Haynes Stephens
University of Chicago
https://orcid.org/0000-0002-2258-5244
Heidi Webber
Leibniz Centre for Agricultural Landscape Research
https://orcid.org/0000-0001-8301-5424
Florian Zabel
Ludwig-Maximilians-Universität München
https://orcid.org/0000-0002-2923-4412
Cynthia Rosenzweig
NASA Goddard Institute for Space Studies
https://orcid.org/0000-0002-8541-2201
Article
Keywords:
crop model, climate-related impacts, yield responses
Posted Date:
August 12th, 2021
DOI:
https://doi.org/10.21203/rs.3.rs-101657/v1
License:
This work is licensed under a Creative Commons Attribution 4.0 International License.
Version of Record:
A version of this preprint was published at Nature Food on November 1st, 2021. See
1
Climate change signal in global agriculture emerges
1
earlier in new generation of climate and crop models
2
3
Jonas Jägermeyr
1,2,3
, Christoph Müller
3
, Alex C. Ruane
1
, Joshua Elliott
4
, Juraj Balkovic
5,19
,
4
Oscar Castillo
6
, Babacar Faye
7
, Ian Foster
8
, Christian Folberth
5
, James A. Franke
9,4
, Kathrin
5
Fuchs
10
, Jose Guarin
1,2
, Jens Heinke
3
, Gerrit Hoogenboom
6,11
, Toshichika Iizumi
12
, Atul K.
6
Jain
13
, David Kelly
8
, Nikolay Khabarov
5
, Stefan Lange
3
, Tzu-Shun Lin
13
, Wenfeng Liu
14
,
7
Oleksandr Mialyk
15
, Sara Minoli
3
, Elisabeth J. Moyer
9,4
, Masashi Okada
16
, Meridel Phillips
1,2
,
8
Cheryl Porter
6
, Sam Rabin
10
, Clemens Scheer
10
, Julia M. Schneider
17
, Joep F. Schyns
15
,
9
Rastislav Skalsky
5,20
, Andrew Smerald
10
, Tommaso Stella
18
, Haynes Stephens
9
, Heidi
10
Webber
18
, Florian Zabel
17
, Cynthia Rosenzweig
1
11
12
1
NASA Goddard Institute for Space Studies, New York, USA
13
2
Columbia University, Center for Climate Systems Research, New York, USA
14
3
Potsdam Institute for Climate Impacts Research (PIK), Member of the Leibniz Association, Potsdam, Germany
15
4
Center for Robust Decision-making on Climate and Energy Policy (RDCEP), University of Chicago, Chicago, USA
16
5
International Institute for Applied Systems Analysis, Laxenburg, Austria
17
6
Agricultural & Biological Engineering Department, University of Florida, Gainesville, Florida, USA
18
7
Institut de recherche pour le développement (IRD) ESPACE-DEV, Montpellier, France
19
8
Department of Computer Science, University of Chicago, Chicago, Illinois, USA
20
9
Department of the Geophysical Sciences, University of Chicago, Chicago, Illinois, USA
21
10
Institute of Meteorology and Climate Research, Atmospheric Environmental Research, Karlsruhe Institute of
22
Technology, Garmisch-Partenkirchen, Germany
23
11
Institute for Sustainable Food Systems, University of Florida, Gainesville, Florida, USA
24
12
Institute for Agro-Environmental Sciences, National Agriculture and Food Research Organization, Tsukuba, Japan
25
13
Department of Atmospheric Sciences, University of Illinois, Urbana, Illinois, USA
26
14
Center for Agricultural Water Research in China, College of Water Resources and Civil Engineering, China
27
Agricultural University, Beijing, China
28
15
Multidisciplinary Water Management group, University of Twente, Enschede, The Netherlands
29
16
Center for Social and Environmental Systems Research, National Institute for Environmental Studies, Tsukuba,
30
Japan
31
17
Ludwig-Maximilians-Universität München (LMU), Munich, Germany
32
18
Leibniz Centre for Agricultural Landscape Research (ZALF), Müncheberg, Germany
33
19
Faculty of Natural Sciences, Comenius University in Bratislava, Bratislava, Slovak Republic
34
20
Soil Science and Conservation Research Institute, National Agricultural and Food Centre, Bratislava, Slovak
35
Republic
36
37
Potential climate-related impacts on future crop yield are a major societal concern first
38
surveyed in a harmonized multi-model effort in 2014. We report here on new 21
st
-century
39
projections using ensembles of latest-generation crop and climate models. Results
40
suggest markedly more pessimistic yield responses for maize, soybean, and rice
41
compared to the original ensemble. Mean end-of-century maize productivity is shifted
42
from +5 to -6% (SSP126) and +1 to -24% (SSP585) — explained by warmer climate
43
projections and improved crop model sensitivities. In contrast, wheat shows stronger
44
gains (+9 shifted to +18%, SSP585), linked to higher CO
2
concentrations and expanded
45
2
high-latitude gains. The ‘emergence’ of climate impacts — when the change signal
46
emerges from the noise — consistently occurs earlier in the new projections for several
47
main producing regions before 2040. While future yield estimates remain uncertain, these
48
results suggest that major breadbasket regions will face distinct anthropogenic climatic
49
risks sooner than previously anticipated.
50
51
52
Climate change already affects agricultural productivity worldwide via many mechanisms, driven
53
largely by warmer mean and extreme temperatures, altered precipitation regimes and drought
54
patterns, and elevated atmospheric CO
2
concentrations ([CO
2
])
1
. Uncertainties arising from
55
greenhouse gas emission scenarios, climate model projections, and the understanding and
56
representation of complex impact processes render estimates of future crop yield highly
57
uncertain
2
. A way towards improving yield projections is the development of benchmarked multi-
58
model ensemble simulations driven by harmonized simulation protocols
3
. Facilitated by the
59
Agricultural Model Intercomparison and Improvement Project (AgMIP)
4
and the Inter-Sectoral
60
Impact Model Intercomparison Project (ISIMIP)
5
, here we present a new systematic assessment
61
of agricultural yield projections, based on a protocol similar to the one used by Coupled Model
62
Intercomparison Project (CMIP) for climate models
6
. Previous projections of AgMIP’s Global
63
Gridded Crop Model Intercomparison (GGCMI) based on CMIP5 identified substantial climate
64
impacts on all major crops, with strong temperature and CO
2
responses and regional patterns of
65
losses and gains
7
. As the first systematic intercomparison, GGCMI-CMIP5 (hereafter ‘GC5’)
66
demonstrated in 2014 that crop models might indeed introduce larger uncertainty than current
67
climate models. CMIP6 now provides new reference climate model projections
8,9
, and recently
68
improved bias-adjustment and downscaling methods
10
benefit the impact modeling community
69
and support an advanced ensemble of process-based crop models. With improved and further
70
harmonized inputs and configuration of cropping systems, GGCMI is able to provide a new
71
standard in crop yield projections for the 21
st
century for several major crops using state-of-the-
72
art modeling approaches with CMIP6 scenarios (hereafter ‘GC6’).
73
74
3
Climate change impacts are often quantified in terms of differences over time, but especially in
75
view of adaptation measures, it is the amplitude of the change compared to the local
76
background variability and uncertainty of the recent past that is often more relevant
11
. Time of
77
climate impact emergence (TCIE) — the point in time by which the yield levels of exceptional
78
years (negative or positive) have become the new norm — is a critical measure for risk
79
assessment. Time of emergence
12
metrics have been applied to climate variables including
80
temperature
13,14
, precipitation
15
, and others
16,17
and demonstrate that major food producing
81
regions are increasingly facing changing climate profiles in the near term. Here we introduce the
82
TCIE concept with respect to future agricultural risks.
83
84
The analyses presented here shed new light on the projected effects of elevated [CO
2
], which
85
have been neglected in many previous studies that focused on direct temperature responses
18–
86
20
. CO
2
effects are among the largest sources of uncertainty inflating the range of crop model
87
projections by the end of the century
21–24
, but they
must be reflected in plausible future yield
88
projections
25
. The uncertainty in the mechanisms and overall size of the effects of CO
2
89
fertilization manifested in farmers’ fields are reflected in a wide range of CO
2
sensitivities among
90
the crop models contributing to the GGCMI archive
21,25
.
91
92
Here we present an ensemble of process-based projections of global productivity estimates for
93
the major crops for the 21
st
century. This work represents the first update since GC5 in 2014
7
94
and includes updated climate projections based on CMIP6 and latest-generation crop models
95
for maize, wheat, rice, and soybean. This study is based on constant management
96
assumptions, focusing on the isolated climate change effect on current crop production
97
systems. Opportunities associated with farming system adaptation and management trends will
98
be addressed in upcoming GGCMI simulations.
99
100
4
The simulation protocol is based on two Shared Socioeconomic Pathways related to
101
Representative Concentration Pathways (RCPs), RCP2.6 and RCP8.5 (hereafter ‘SSP126’ and
102
‘SSP585’; adaptation measures associated with the SSPs are not considered)
9
, chosen to
103
sample the range of available scenarios
26
and to make the results comparable with GC5.
104
Twelve GGCMs each simulated 5 GCM forcings, resulting in nearly 240 climate-crop model
105
realizations per crop (GGCMs x GCMs x RCPs x CO
2
settings). The climate projections from the
106
5 GCMs (Table S1), centrally bias-adjusted and downscaled for different research sectors, were
107
selected by ISIMIP based on benchmark performance, equilibrium climate sensitivity, and
108
output availability (see Methods). All simulations were carried out globally on a 0.5° grid,
109
covering the time period 1850 to 2100. We evaluate results based on the transient atmospheric
110
CO
2
concentration (i.e., ‘default’ [CO
2
]) and only refer to counterfactual simulations without
111
[CO
2
] increase after the year 2015 (‘constant’ [CO
2
]) to quantify the CO
2
fertilization effect for
112
further uncertainty evaluation and climate change factor attribution.
113
114
Recent literature has focused on capturing the temperature sensitivity of crops
18–20,27–29
in
115
isolation. To quantify climate change impacts more comprehensively, additional factors
116
including precipitation changes, temperature-moisture feedbacks, and [CO
2
] need to be
117
considered. The aims of this first GC6 study are to: i) provide new ensemble projections for the
118
productivity of major crops under climate change, ii) assess climate change impacts on crop
119
yields from a risk perspective, employing the TCIE concept, iii) improve understanding of
120
regional patterns of change, and iv) explore drivers of uncertainty related to climate models,
121
crop models, and responses to [CO
2
].
122
Global production response of major crops
123
The ensemble response across the new generation of climate and crop models to the SSP126
124
and SSP585 forcing is markedly more pronounced than in GC5
7
(Fig. 1). Wheat results are
125
5
more optimistic, while maize, soybean, and rice results are decisively more pessimistic. For
126
maize, the most important global crop in terms of total production and food security in many
127
regions, the mean end-of-century (2069-2099) global productivity response is ~10% (SSP126)
128
and ~20% (SSP585) lower than in GC5. This shifts the SSP585 estimate from +1%
129
(interquartile range of crop-climate model combinations: -10 to +8%) to -24% (-38 to -7%) and
130
for SSP126 from +5 to -6%. For wheat, the second largest global crop in terms of production,
131
the SSP585 ensemble estimate is shifted upwards from +10% (-1 to +15%) to +18% (-2 to
132
+39%), and under SSP126 from +5 to +9%. The SSP585 ensemble estimates for soybean are
133
revised downward from +15% (-8 to +36%) to -2% (-21 to +17%) and for rice from +23% (+1 to
134
+33%) to +2% (-15 to +12%). Overall, the new climate and crop model combinations narrow the
135
range of crop yield projections for soybean and rice, but disagreement among crop models
136
remains substantial and is largely indecisive about the sign of change at the global level (p-
137
value > 0.5 for both crops). The maize and wheat responses are robust and became more
138
distinct since GC5. While the range of crop projections somewhat increased, 85% of model
139
combinations indicate negative maize changes and 73% project positive wheat changes under
140
SSP585. Both responses are now statistically significant (p-value < 10
-5
); the maize response in
141
GC5 was not (p-value > 0.6). There is larger agreement on positive change for wheat under
142
SSP126 (89%) than under SSP585, indicating peak-and-decline trajectories for parts of the
143
ensemble under high-emissions scenarios (Fig. S1).
144
145
As a C
4
crop, maize has a smaller capacity to benefit from elevated [CO
2
]
30
, and it is also grown
146
across a wider range of low latitudes that are projected to experience the largest adverse
147
impacts due in large part to current proximity to crop-limiting temperature thresholds
31
. As a C
3
148
crop, the positive wheat response is explained by its relatively stronger CO
2
response and the
149
fact that global warming leads to wheat yield increases in high-latitude regions that are currently
150
temperature-limited
29
.
151
6
152
Three factors explain the more-pronounced crop yield response in GC6. First, CMIP6 has
153
markedly higher [CO
2
] than CMIP5 (Fig. 2), with year 2099 concentrations increased from 927
154
ppm (RCP8.5) to 1122 ppm (SSP585)
9
. Second, CMIP6 has a higher average end-of-century
155
warming level compared to CMIP5, adequately represented in the 5 GCMs sampled here (Table
156
S1, S2). While both RCP2.6 and RCP8.5 are on average ~0.3 °C warmer in CMIP6 than CMIP5
157
over land and oceans, the difference is even more pronounced (>0.5 °C) across main maize-
158
producing regions (Fig. 2). Third, the new crop model ensemble features advanced versions of
159
previous models, several new members, and improved input data, which resulted in more
160
realistic sensitivities to climate and [CO
2
] changes (see details below).
161
Emergence of the climate change signal in agriculture
162
The Time of Climate Impact Emergence (TCIE) describes the point in time when average
163
climate change impacts are projected to occur outside the envelope of historical variability and
164
uncertainty (‘noise’). We define TCIE as the year in which the multi-model 25yr moving-average
165
crop production change (‘signal’) emerges from the noise (i.e., standard deviation of simulated
166
variability across all GCM x GGCM combinations in 1983-2013).
167
168
Maize consistently shows emerging negative productivity changes (‘negative TCIE’) among
169
major producer regions. The ensemble median signal emerges from the noise at global level in
170
the year 2032 under SSP585 and the year 2051 under SSP126 (Fig. 3). Of all individual GCM x
171
GGCM realizations, 84% show a negative TCIE by 2099 under SSP585 (52% under SSP126)
172
and the inter-quartile range spans from 2014 to 2056, indicating sizeable agreement among
173
models. This is a substantial shift away from the GC5 simulations in which the ensemble
174
median shows no emergence by 2099 under any emission pathway, only seen in 46% of
175
individual GCM x GGCM combinations under RCP8.5 (inter-quartile range 2044-2080). Overall,
176
7
the TCIE signal at global level is shifted 30-40 years earlier and is more pronounced in the new
177
generation of climate and crop model projections (Fig. 4).
178
179
By the end of the century, 10% (SSP126) to 74% (SSP585) of current global maize cultivation
180
areas are projected to undergo negative TCIE (Fig. 5). Under SSP585 this trajectory is markedly
181
earlier, with higher late-century fractions of cropland area affected compared to the respective
182
47% in GC5 (RCP8.5). Crop models indicate early negative maize TCIE before 2040 even
183
under SSP126 in Central Asia, the Middle East, Southern Europe, Western USA, and tropical
184
South America. Projections referencing the 1983-2013 period suggest that the mean yield signal
185
is already starting to emerge in some of these regions (Fig. 3e and Fig. 5), patterns largely in
186
line with recent observations
15,32,33
. The tropical zone is the only climate zone in which the GC5
187
ensemble median also indicated a negative maize TCIE (Fig. 3e).
188
189
The standard deviation of grid-level TCIE estimates under SSP585 ranges between 25 and 35
190
years across most breadbasket regions, with slightly higher values under SSP126 (Fig. S2).
191
Such uncertainty ranges are in line with time of emergence estimates for climatological
192
variables, yet somewhat higher due to the additional layer of crop model uncertainties
12,13
.
193
Clearest emergence signals, i.e., largest signal-to-noise ratios with values < -2, are found
194
among lower latitudes in the tropics but also in Central Asia, the Middle East, and Western USA
195
(Fig. S3). As internal variability — and thus total noise — decreases with averaging, earlier
196
TCIE is generally found for larger spatial scales.
197
198
For wheat, ensemble projections indicate TCIE of positive productivity changes (‘positive TCIE’)
199
at the global level (Fig. 3b) and across large parts of currently cultivated areas (Fig. 5). While
200
also found in GC5 simulations, TCIE is shifted ~10 years earlier in GC6, suggesting that
201
climate-related increases might occur globally within the next few years (year 2023 under
202
8
SSP585, year 2025 under SSP126; inter-quartile ranges 2014-2029 and 2015-2029) and across
203
major breadbasket regions within the next two decades (Fig. 5). In some regions we already
204
detect a TCIE signal today, which is in line with the range of time of emergence estimates for
205
temperature and precipitation
13,15
. Such effects are difficult to distinguish from rapidly changing
206
management practices in observational data, but climate change impacts have been
207
documented for example in Central and South Asia, Northern China, and the USA
32,34
. The
208
TCIE estimates for wheat show high consistencies across the model ensemble — 76%
209
(SSP126) and 88% (SSP585) of individual model combinations show positive TCIE by 2099. As
210
for maize, the TCIE signal is shifted earlier and is more pronounced in GC6 than in GC5 (Fig.
211
4).
212
213
The share of wheat cultivation areas projected to see positive TCIE increased substantially in
214
GC6, from 8% (GC5, RCP8.5) to 37% (GC6, SSP585; Fig. 5f). This share levels off by mid-
215
century, a result of peak-and-decline trajectories seen in some crop models (Fig. 5f ; compare
216
Fig. 3d and Fig. S3 for regions that show TCIE early on but not by late century). Wheat also
217
exhibits negative TCIE among important growing regions in South Asia, Southern USA, Mexico,
218
and parts of South America around mid-century. The uncertainty among grid-level TCIE
219
estimates is generally higher for wheat than for maize (Fig. S2) and the extent of areas with very
220
high signal-to-noise ratios (i.e., >2) is smaller (Fig. S3).
221
222
Ensemble median soybean and rice productivity peak mid-century and decline towards the end
223
of the century at the global level (Fig. S4). The soybean response exhibits late-century negative
224
TCIE (year 2096) under SSP585; rice on the other hand shows early positive TCIE (year 2030,
225
SSP585) but late-century declines are not projected to reach the level of negative TCIE at the
226
global level. Rice is the only crop in this study that indicates positive TCIE in the tropics, which
227
9
drives early net global gains before productivity is simulated to decline again by about 2060
228
(Fig. S4c).
229
Regional patterns of yield change
230
Projections of crop yield changes include regions of losses and gains for all crops (Fig. 3, S4).
231
Global average responses can hide important regional changes, which are supported by strong
232
crop model agreement. Maize projections show spatially homogeneous losses especially
233
among main growing regions in North America, Mexico, West Africa, Central Asia, and China,
234
where crop model agreement is high (Fig. 3c). The high-latitude gains found in GC5 are not as
235
prevalent in GC6 and associated with high crop model uncertainty and low baseline yields.
236
Wheat shows distinct geographic gradients with losses in spring wheat regions in Mexico,
237
Southern USA, South America, and South Asia, supported by good model agreement. Sizable
238
wheat gains are projected by many models for the North China Plains, Australia, Central Asia,
239
Middle East, and for the winter wheat growing regions in the Northern USA and Canada (Fig.
240
3d). Soybean shows the greatest losses in the main-producer regions — the USA, Brazil, and
241
Southeast Asia — paired with large gains across parts of China and generally higher latitudes
242
(Fig. S4). Major declines in rice yields are simulated in Central Asia, and gains in South Asia,
243
NE China, and South America. Both soybean and rice yield changes must be interpreted in view
244
of the wide range in crop model ensemble results (Fig. 1, S4). A breakdown of yield responses
245
for the top-10 producer countries per crop highlights a wide range of CO
2
effects embedded in
246
the signal (Fig. S5, S6).
247
248
A latitudinal profile of yield changes under SSP585 — simulated in all grid cells irrespective of
249
the current cropland distribution — indicates that losses are most prevalent among low-latitude
250
tropical regions with highest gains found at higher latitudes beyond 50°N and 30°S for all crops
251
(Fig. 6). Maize exhibits widespread losses between 50°N and 30°S, while losses for the other
252
10
crops are more concentrated in the tropics with a less distinct signal for soybean and rice. Major
253
wheat breadbaskets are generally located at higher latitudes than maize, which further
254
contributes to overall wheat gains when aggregated across currently cultivated areas. Although
255
more than 90% of maize and wheat is currently produced in the temperate and subtropical
256
climate zones, major yield losses will affect the livelihoods and food security of many
257
smallholder farmers in the tropics. Overall, our results show that lower latitudes face the largest
258
losses for all crops, while higher latitudes see potential gains. These conclusions are in line with
259
the IPCC AR5
35
and recent studies
7,36,37
and such uneven distribution of impacts may further
260
increase regional disparities that are a ‘Reason for Concern’
38
regarding climate change risks.
261
Drivers of more pronounced ensemble response
262
It is difficult to determine to what degree the differences in crop yield projections between GC6
263
and GC5 can be explained by the new atmospheric forcing, the new crop model ensemble, or
264
new input data. A subset of GC6 and GC5 crop models that participated in both ensembles
265
(albeit in different versions) shows very similar responses compared with the respective full
266
ensemble, suggesting that the crop model selection does not explain the differences (Fig. 7).
267
Further, standardized comparisons of crop model responses to specific mean temperature
268
increases over cropland areas (‘warming sensitivity’; under constant [CO
2
] conditions, but
269
including changes in other climate variables) from 1-2°C and from 2-3°C, respectively, highlights
270
that the isolated warming sensitivity in GC6 has substantially increased for maize (from 2-3% in
271
GC5 to 8-9% in GC6) and decreased for wheat (from 7% to 3-6%; Fig. 7). With higher overall
272
warming levels in CMIP6, net warming-related maize losses by 2069-2099 thus increased from
273
12% (4.6°C maize cropland warming) to 30% (5°C maize cropland warming) in GC6. Further,
274
the CO
2
sensitivity at 500 and 700 ppm, but also net effects by the end of the century, have
275
decreased for both maize and wheat. In summary, the more pessimistic maize response in GC6
276
can largely be attributed to a higher sensitivity to warming and a lower compensating effect due
277
11
to CO
2
fertilization in the crop models, and to a smaller extent to the higher absolute warming
278
levels in CMIP6. For wheat on the other hand, the more optimistic response in GC6 can be
279
explained by lower losses per degree warming (with stronger temperature-related gains in high-
280
latitude regions), overcompensating for a lower CO
2
fertilization effect than in GC5 (despite
281
higher total [CO
2
] levels). For soybean and rice, in contrast, the more pessimistic response in
282
GC6 is largely attributed to higher warming levels in CMIP6 compounded by a higher crop
283
model sensitivity to warming, with similar sensitivities to changes in [CO
2
] (Fig. S7).
284
Crop and climate model uncertainty
285
The range of crop model responses under SSP585 (mean across climate models) is
286
substantially larger than the range introduced by the five climate models (mean across crop
287
models; Fig. 1). However, for all crops and RCPs, the uncertainty associated with the five
288
CMIP6 climate models has increased compared to the five climate models sampled in GC5. In
289
turn, the fraction of total variance induced by the crop models is substantially reduced for all
290
crops in GC6 (for maize from 97 to 69%; Fig. 8), which highlights that the crop response
291
became more consistent, even though the number of crop models increased. Absolute variance
292
induced by the climate models has increased for all crops (Fig. 8),
293
which is explained by a wider distribution of climate sensitivities tracked by the five CMIP6
294
GCMs (Table S1, S2), but also by higher [CO
2
] assumed in CMIP6 (Fig. 2). In this sample,
295
UKESM1 is the most pessimistic GCM for both RCPs and all crops, the global mean warming
296
level by 2099 is about 2.6°C higher than in GFDL-ESM4, and the Transient Climate Response
297
is 1.2°C higher (see Table S1 for more details)
6
. Generally, the least pessimistic crop impacts
298
are found with MRI-ESM2 (Fig. 1).
299
300
Higher emission scenarios inflate the crop model uncertainty (SSP585), while the overall
301
climate- and crop model-induced uncertainty range in GC6 is of comparable size under SSP126
302
12
(Fig. 1). Uncertainty in the CO
2
effect causes much of the crop model uncertainty for wheat,
303
soybean, and rice (Fig. S8), yet the range of maize responses is not fundamentally reduced
304
without the CO
2
effect. In line with physiological knowledge
30
, crop models mostly show the
305
smallest CO
2
effects for C
4
crops (maize) and much larger responses for C
3
crops (wheat,
306
soybean, rice). However, the CO
2
effects differ widely across crop models; the ensemble
307
median rainfed response is 19% for maize, 33% for wheat, 48% for soybean, and 37% for rice
308
by the year 2099 (Fig. S8), which is generally in line with field experiments given that model
309
simulations include nutrient limitations
25,30
. CYGMA and CROVER exhibit a strong peak-and-
310
decline CO
2
response for some crops, resulting in negative CO
2
effects for maize in CYGMA
311
after 2090 (Fig. S8). This is driven by increased water use efficiencies under elevated [CO
2
],
312
eventually leading to adverse excess moisture effects in humid regions — a new feedback
313
represented primarily in CYGMA and underexplored in previous studies
39
.
314
315
In addition to the CO
2
effect, climate change affects simulations of crop growth and
316
development in various ways. These include for example changed precipitation patterns,
317
extreme heat and drought events, and importantly, accelerated maturity. Higher temperatures
318
lead to faster phenological development and substantial reductions in the growing season
319
length in all crop models (Fig. S9), which in turn lead to complex processes affecting yield,
320
including shorter grain filling periods, smaller canopy, and reduction in photosynthesis. This
321
effect varies across models and additional work is needed to further narrow the range of crop
322
model responses
40
. After all, the standard deviation of simulated yield variability matches
323
observational data to a much higher degree in GC6 (R = 79%) than in GC5 (R = 44%), adding
324
to more realistic yield responses (Fig. S10).
325
13
Discussion
326
We introduce the concept of climate impact emergence to the field of agriculture impacts,
327
highlighting that major shifts in global crop productivity due to climate change are projected to
328
occur within the next twenty years, several decades sooner than estimates based on previous
329
model projections. The impact on crop productivity under SSP126 and SSP585 is largely similar
330
for the coming decade, which leaves little room for climate mitigation efforts. In light of the much
331
larger climate and crop model agreement for these short-term projections than for the late
332
century, the findings highlight challenges for food system adaptation faced with significantly
333
shorter lead times.
334
335
These CMIP6 multi-model crop yield projections suggest that climate change impacts on global
336
agriculture will be more pronounced than in GC5, with substantially larger losses for maize,
337
soybean, and rice and additional gains for wheat. This is supported by a generally more
338
consistent crop model ensemble. However, large uncertainties remain, particularly in TCIE
339
estimates — the standard deviation for global maize TCIE is 24 years (SSP585), which is
340
similar to estimates of temperature emergence
12
. Yet the signal is robust: More than 80% of the
341
GCM-GGCM combinations indicate TCIE for maize and wheat by late century across major
342
breadbaskets (SSP585). TCIE estimates based on different metrics qualitatively agree (e.g.,
343
multi-model ensemble mean TCIE for maize is found in the year 2032, the median of individual
344
GCM x GGCM estimates in the year 2027, and the mean in the year 2036). Leaving one crop
345
model out at a time introduces a TCIE standard deviation of only 1.5 years for both maize and
346
wheat (SSP585). That said, time of emergence estimates are sensitive to the underlying
347
definitions (e.g., noise, pre-industrial or recent climate, smoothing approach, threshold
348
selection) and can push the emergence date earlier or later in time
12,13,15,41
. Absolute TCIE
349
14
estimates are therefore more challenging to interpret than relative comparisons among regions,
350
crops, and especially the two ensemble projections GC5 and GC6.
351
352
Wheat yield increases are projected to level off by midcentury and part of the climate-crop
353
model ensemble indicates net losses under SSP585 by 2099 (Fig. 1, S1). Maize yield on the
354
other hand is projected to decline steadily, supported by higher model agreement than for
355
wheat. These general response differences are also in line with previous findings
42
. The more
356
pronounced response of the new projections can be explained primarily by higher equilibrium
357
climate sensitivities, higher [CO
2
], and different crop model sensitivities per degree warming and
358
[CO
2
] changes. With regard to CMIP6, higher and wider-ranging climate sensitivities are
359
critically discussed and associated with differing parameterizations of cloud feedback and cloud-
360
aerosol interactions
14,43–49
. While better simulations of cloud liquid water contents and their
361
radiative behavior render the climate models more realistic, it is unclear whether these
362
improvements translate into more accurate estimates of equilibrium climate sensitivity (ECS)
363
and overall warming levels. Additional improvements of the GCMs, and the bias-adjustment and
364
downscaling methods used, result in better representations of extreme events and internal
365
variability
10,47,50–52
, which are critical for crop modeling. Higher [CO
2
] in CMIP6 are due to a
366
revised tradeoff between [CO
2
] and [CH
4
] resulting from updated observations and assumptions
367
in the MAGICC7.0 model
53
.
368
369
The GGCMI crop model ensemble has substantially changed and consists of revised and new
370
members. For example, LPJmL contributed to GC5 and has since been fundamentally improved
371
with the addition of the nitrogen cycle
54
and heat unit parameterization
55
. In addition, input data
372
and model harmonization have been improved, including growing season harmonization based
373
on a new crop calendar developed for this study (see Methods). A comprehensive attribution of
374
crop response differences between GC5 and GC6 to changes in climate forcing, crop model
375
15
selection and sensitivities, and input data is not feasible. But standardized comparisons of
376
changes in cropland warming and [CO
2
] indicate that for maize and wheat changes in crop
377
model ensemble sensitivities dominate the response, and for soybean and rice higher warming
378
levels and warming sensitivity explain much of the differences (Fig. 7, S7).
379
380
The new GCM bias adjustment, crop model advancement, improved input data, and a new crop
381
yield bias correction serve to substantially reduce the amount of variance induced by the crop
382
models compared to the climate models, rendering the new GC6 ensemble more balanced and
383
consistent than GC5 despite a larger ensemble size (12 crop models in GC6, 7 in GC5; Fig. 8).
384
In a similar vein, Müller et al.
56
comprehensively compared crop yield uncertainties under all
385
CMIP5 and CMIP6 GCMs based on GGCMI crop model emulators
57
, confirming that CMIP6
386
introduces a wider range of yield responses with more pessimistic average impacts. In view of
387
improved model harmonization, inputs, and GGCM versions and performance, we consider
388
GC6 more reliable than GC5 – despite ongoing discussions on the temperature sensitivity in
389
CMIP6.
390
391
The wide range of CO
2
effects across GGCMI models is generally in line with field
392
experiments
25,58,59
, but the broad range of simulated CO
2
fertilization effects merits more
393
rigorous model testing at the process level, which in turn requires better reference data,
394
especially at high [CO
2
] levels. Moreover, elevated [CO
2
] boosts crop yield, but it may also affect
395
the nutritional content of the crops
60–62
. Impacts related to excess moisture, water resource
396
limitations, and new distributions of pests and diseases may lead to additional regional biotic
397
stresses requiring follow-on analysis.
398
399
Cropping system adaptation can substantially reduce and even outweigh adverse climate
400
change impacts, for example by switching to other crops
63
or better-adapted varieties
27,64
.
401
16
Integrated into ISIMIP’s wider cross-sector activities, GGCMI will systematically evaluate
402
farming system adaptation and changes in yield variability and extreme event impacts in
403
subsequent efforts.
404
405
In conclusion, the new generation of AgMIP’s GGCMI provides the most comprehensive
406
ensemble of process-based future crop yield projections under climate change to date. The
407
degree to which even high mitigation climate change scenarios are projected to push global
408
farming outside of its historical regimes suggests that current food production systems will soon
409
face fundamentally changed risk profiles. Despite prevailing uncertainties, these ensemble
410
projections spotlight the need for targeted food system adaptation and risk management across
411
the main producer regions in the coming decades.
412
413
17
Figures
414
415
Fig. 1: Ensemble end-of-century crop productivity response.
Global productivity changes (2069-2099
416
compared to 1983-2013) for SSP126 and SSP585 are shown as the mean across climate and crop models
417
for the four major crops (highlighted by bullets underneath the plot). Whiskers indicate the range of individual
418
climate model realizations (dashed line, as the mean across crop models), and the range across crop
419
models (solid line, as the mean across climate models). Individual model results are indicated by the bullets
420
along the whisker lines (for SSP585 only); violin shades additionally highlight the model distribution. For
421
context, gray bars and whiskers reference previous GGCMI simulations based on CMIP5 (GC5; Rosenzweig
422
et al. 2014)
7
in the same way, without specifying individual models. Data are shown for the default [CO
2
].
423
Not all crop models simulate all crops, see Table S3 for details.
424
425
−
6.4
+4.9
−
24.1
+1.3
●
e
●
b
●
c
●
a
●
d
●
3
●
9
●
4
●
6
●
12
●
2
●
1
●
10
●
5
●
8
●
11
●
7
Maize
−
60
−
40
−
20
0
20
40
60
Global productivity change [%]
+8.8
+5
+17.5
+9.9
●
b
●
e
●
c
●
a
●
d
●
2
●
7
●
4
●
5
●
6
●
10
●
3
●
8
●
9
●
11
●
12
●
1
Wheat
+2
+6.5
−
2.1
+15.3
●
e
●
b
●
c
●
a
●
d
●
3
●
2
●
6
●
7
●
10
●
5
●
9
●
12
●
1
●
8
●
11
Soybean
+3.4
+7.7
+1.7
+23.3
80
●
e
●
b
●
d
●
a
●
c
●
2
●
3
●
9
●
5
●
10
●
6
●
7
●
8
●
1
●
11
Rice
−
60
−
40
−
20
0
20
40
60
Rosenzweig et al. 2014 (CMIP5)
SSP126
SSP585
Climate model range
Crop model range
Climate models
a GFDL
−
ESM4
b IPSL
−
CM6A
−
LR
c MPI
−
ESM1
−
2
−
HR
d MRI
−
ESM2
−
0
e UKESM1
−
0
−
LL
Crop models
1 ACEA
2 CROVER
3 CYGMA1p74
4 DSSAT
−
Pythia
5 EPIC
−
IIASA
6 ISAM
7 LandscapeDNDC
8 LPJmL
9 pDSSAT
10 PEPIC
11 PROMET
12 SIMPLACE
−
LINTUL5
18
426
Fig. 2: Comparison of [CO
2
] and temperature changes between CMIP5 and CMIP6.
[CO
2
] pathways for
427
RCP26 and RCP85 in CMIP5 compared to SSP126 and SSP585 in CMIP6 (
a
). Box-and-whisker plots (
b
)
428
show the difference of the average maize growing season temperature changes [°C] (2069-2099 compared
429
to 1983-2013) between the CMIP6 and CMIP5 ensemble. Each ensemble is represented by the mean of 5
430
GCMs (Table S1 and S2) in each grid cell. CMIP6 and CMIP5 differences are separated for SSP126 (green)
431
and SSP585 (yellow) for all grid cells (maize production > 0; lighter shade) and for the highest-producing
432
grid cells that together account for 50% of global production (darker shade).
433
421
927
447
1122
2020
2040
2060
2080
2100
400
600
800
1000
1200
Atmospher
ic CO
2
concentr
ation [ppm]
a
SSP126 (CMIP6)
RCP26 (CMIP5)
SSP585 (CMIP6)
RCP85 (CMIP5)
0.32
n = 12141
0.59
n = 658
0.31
n = 12141
0.53
n = 658
−
0.5
0.0
0.5
1.0
1.5
2.0
T
emper
ature change diff
erence [°C]
b
SSP126
−
RCP26
SSP126
−
RCP26 (high
−
producing regions)
SSP585
−
RCP85
SSP585
−
RCP85 (high
−
producing regions)
19
434
Fig. 3: Projections of global crop productivity for the 21
st
century.
For maize (
a
) and wheat (
b
),
435
productivity time series are shown as relative changes to the 1983-2013 reference period under SSP126
436
(green) and SSP585 (yellow). Shaded ranges illustrate the interquartile range of all climate-crop model
437
combinations (5 GCMs x 12 GGCMs). The solid line shows the median response (and a 25yr moving
438
average). Horizontal dashed lines mark the standard deviation of historical yield variability and model
439
uncertainty (i.e., ‘noise’ from individual climate-crop model combinations) and open circles highlight the
440
‘Time of Climate Impact Emergence’ (TCIE), the year in which the smoothed climate change response
441
emerges from the noise. For context, the TCIE calculated from GC5
7
simulations is indicated in lighter
442
shades above the TCIE based on GC6 (>2099 if no TCIE occurs by 2099). The maps (
c, d
) show median
443
yield changes (2069-2099) under SSP585 across climate and crop models for current growing regions (>10
444
ha). Hatching indicates areas where less than 70% of the climate-crop model combinations agree on the
445
sign of impact. Regional productivity time series (
e, f
) are similar to (
a
), but stratified for the four major
446
Koeppen-Geiger climate zones (temperature limited, temperate/humid, subtropical, and tropical). The
447
percentage of the total global production contributed by each zone is indicated in the top right corner of the
448
insets. All data are shown for the default [CO
2
] (see Fig. S4 for all four crops).
449
20
450
Fig. 4: Shift towards earlier and more pronounced climate impact emergence.
Density plots of
451
individual TCIE estimates across the GCM x GGCM ensemble under SSP585 are shown for global maize
452
productivity (
a
;
negative TCIE) and wheat (
b
; positive TCIE). Histogram counts are smoothed with a loess fit
453
(span=0.5) and shown as the fraction of the respective ensemble size. The GGCMI-CMIP6 ensemble
454
includes 12 crop models, GGCMI-CMIP5 includes 7 crop models; both comprise 5 GCMs. The total
455
ensemble fraction that shows TCIE by 2099 is indicated in the top-right corner (‘Sum’). The ensemble
456
median TCIE is highlighted with vertical dashed lines.
457
Year
Sum = 84%
Sum = 46%
Maize
a
2000
2020
2040
2060
2080
2100
0
5
10
15
20
25
30
Fr
action of ensemb
le [%]
Year
Sum = 88%
Sum = 69%
Wheat
b
2000
2020
2040
2060
2080
2100
0
5
10
15
20
25
30
Fr
action of ensemb
le [%]
GGCMI
−
CMIP6
GGCMI
−
CMIP5
21
458
459
Fig. 5: Geographic patterns in TCIE.
The maps show TCIE estimates for maize (
a, b
) and wheat (
c, d
)
460
under SSP126 and SSP585 — calculated as the median of individual TCIE estimates from each climate-
461
crop model combination. Hatching indicates areas in which less than 70% of the crop models agree on the
462
emergence signal by 2099. See Figure S2 for the associated standard deviation of TCIE estimates, and
463
Figure S3 for the signal-to-noise ratio. Panel (
e)
and
(f)
illustrate the annual percentage of the respective
464
global cropland area affected by negative (maize) and positive (wheat) TCIE under SSP126 and SSP585,
465
separated for results from GC5
7
and GC6. Vertical bars indicate the inter-quartile range of all climate-crop
466
model combinations, with the median value in the circle. The maps show the first TCIE occurrence, even if
467
the signal is reversed by late century (e.g., parts of India for wheat; compare with Fig. S3); estimates of the
468
affected areas (
e
,
f
) account for signal changes.
469
470
22
471
472
Fig. 6: Latitudinal profile of crop yield changes.
Yield changes (SSP585, 2069-2099) are shown as
473
latitude averages for maize (
a
), wheat (
b
), soybean (
c
), and rice (
d
), based on crop simulations in all grid
474
cells, unconstrained by current cropland extent (bottom x-axis). For context, the current cropland extent is
475
shown across latitude bands as fractions of the crop-specific global extent (top x-axis; mirrored to allow
476
overlaps with both positive and negative yield changes). Yield data are shown as the climate and crop model
477
median (marginal areas with yield lower than the 20
th
percentile per crop are excluded).
478
Cropland extent
Yield decline
Yield increase
−
40
−
20
0
20
40
Yield change [%]
10
5
0
5
10
Fraction of global cropland [%]
−
60
−
30
0
30
60
90
Latitude
Maize
a
−
40
−
20
0
20
40
Yield change [%]
10
5
0
5
10
Fraction of global cropland [%]
Wheat
b
−
40
−
20
0
20
40
Yield change [%]
10
5
0
5
10
Fraction of global cropland [%]
Soybean
c
−
40
−
20
0
20
40
Yield change [%]
10
5
0
5
10
Fraction of global cropland [%]
−
60
−
30
0
30
60
90
Latitude
Rice
d
23
479
Fig. 7:
Driver attribution of crop model responses
. Projected end-of-century global productivity changes
480
for maize (
a
) and wheat (
b
) under RCP8.5 (climate model mean) are shown for all members of the crop
481
model ensemble GGCMI-CMIP5 (GC5) and GGCMI-CMIP6 (GC6), and for a subset of crop models that
482
participated in both rounds (note substantial differences between model versions). The sensitivity to global
483
mean warming (
c
,
d
) of the full ensembles is shown for temperature changes (over respective cropland
484
areas per crop) from 1 to 2°C, from 2 to 3°C, and for the total change between 1983-2013 and 2069-2099.
485
The warming sensitivity is based on [CO
2
] held constant at the 2015 level but includes changes in other
486
climate variables. The CO
2
sensitivity (
e
,
f
) in GC5 and GC6 is shown at specific [CO
2
] concentrations and
487
for the 2069-2099 mean concentrations. Warming and CO
2
sensitivities are calculated based on crop model
488
responses over a 21-year window centered on the year in which a certain temperature change or [CO
2
]
489
concentration occurs in each climate model. Filled circles indicate the median crop model response,
490
●
●
●
1
2
5
7
GC5
subset
●
−
1.2
●
●
●
5
8
9
10
GC6
subset
●
−
19.8
●
●
●
1
2
3
4
5
6
7
GC5
●
−
0.6
●
●
●
1
2
3
4
5
6
7
8
9
10
11
12
GC6
●
−
22.4
−
60
−
40
−
20
0
20
40
Global productivity change [%]
End
−
of
−
centur
y response
GGCMI
−
CMIP5
GGCMI
−
CMIP6
Maize
a
●
●
●
1
2
3
4
5
6
7
●
−
3.1
●
●
●
2
3
5
6
7
8
9
10
11
12
●
−
7.5
From 1 to 2 °C
cropland T change
−
20
−
15
−
10
−
5
0
5
Global productivity change [%]
W
arming sensitivity
(constant CO
2
)
c
●
●
●
1
2
3
4
5
6
7
●
−
1.9
●
●
●
2
3
5
6
7
8
9
10
11
12
●
−
7.4
From 2 to 3 °C
cropland T change
●
●
●
1
2
3
4
5
6
7
●
−
11.9
●
●
●
2
3
5
6
7
8
9
10
11
12
●
−
30.2
(2069
−
2099)
4.6 °C
5 °C
−
60
−
40
−
20
0
20
40
●
●
●
1
2
3
4
5
6
7
●
4.6
●
●
●
2
3
5
6
7
8
9
10
11
12
●
2.3
500
ppm
●
●
●
1
2
3
4
5
6
7
●
10.3
●
●
●
2
3
5
6
7
8
9
10
11
12
●
7.4
700
ppm
●
●
●
1
2
3
4
5
6
7
●
12.7
●
●
●
2
3
5
6
7
8
9
10
11
12
●
10.3
(2069
−
2099)
794
ppm
921
ppm
0
10
20
30
Global productivity change [%]
CO
2
sensitivity
e
●
●
●
1
2
5
7
GC5
subset
●
6.1
●
●
●
5
8
9
10
GC6
subset
●
11.3
●
●
●
1
2
3
4
5
6
7
GC5
●
2.6
●
●
●
1
2
3
4
5
6
7
8
9
10
11
12
GC6
●
9.8
−
20
0
20
40
Global productivity change [%]
Wheat
b
●
●
●
1
2
3
4
5
6
7
●
−
5.6
●
●
●
2
3
5
6
7
8
9
10
11
12
●
−
4.7
From 1 to 2 °C
cropland T change
−
10
−
5
0
Global productivity change [%]
d
●
●
●
1
2
3
4
5
6
7
●
−
5.7
●
●
●
2
3
5
6
7
8
9
10
11
12
●
−
4.4
From 2 to 3 °C
cropland T change
●
●
●
1
2
3
4
5
6
7
●
−
13.6
●
●
●
2
3
5
6
7
8
9
10
11
12
●
−
15.3
(2069
−
2099)
4.9 °C
5.2 °C
−
40
−
30
−
20
−
10
0
10
●
●
●
1
2
3
4
5
6
7
●
14.1
●
●
●
2
3
5
6
7
8
9
10
11
12
●
8.3
500
ppm
●
●
●
1
2
3
4
5
6
7
●
28.3
●
●
●
2
3
5
6
7
8
9
10
11
12
●
20.3
700
ppm
●
●
●
1
2
3
4
5
6
7
●
32.6
●
●
●
2
3
5
6
7
8
9
10
11
12
●
28.8
(2069
−
2099)
794
ppm
921
ppm
0
20
40
60
80
Global productivity change [%]
f
GC5 crop models
1 EPIC
2 GEPIC
3 IMAGE
4 LPJ
−
GUESS
5 LPJmL
6 PEGASUS
7 pDSSAT
GC6 crop models
1 ACEA
2 CROVER
3 CYGMA1p74
4 DSSAT
−
Pythia
5 EPIC
−
IIASA
6 ISAM
7 LandscapeDNDC
8 LPJmL
9 pDSSAT
10 PEPIC
11 PROMET
12 SIMPLACE
−
LINTUL5
24
additionally highlighted by circled numbers underneath each plot. Black bars show the inter-quartile range
491
and individual models are indicated by numbers. Note that both panel
c
and
d
include two different legends.
492
See Figure S7 for soybean and rice results. ACEA and DSSAT-Pythia have not submitted simulations for
493
the constant [CO
2
] setting and are excluded from panel
c
-
f
.
494
495
496
497
498
499
500
Fig. 8: Variance decomposition of ensemble projections.
Stacks show the fraction of total variance of
501
mid-century crop production changes
(2030-2070 mean) induced by the climate model ensemble (GCMs;
502
yellow) and by the crop model ensemble (GGCMs; pink), for GGCMI-CMIP6 (GC6) and GGCMI-CMIP5
503
(GC5), respectively. Variance fractions are normalized by the variance cross term to be additive. The
504
absolute variance introduced by GGCMs and GCMs is indicated at the base of each stack. The GCM
505
ensemble has 5 members in both cases, the GGCM ensemble has 12 members in GC6 and 7 members in
506
GC5, which further highlights that the crop model response became more consistent in GC6 compared to
507
the climate model uncertainty.
508
0.31
0.69
0.03
0.97
95
119
42
3
Absolute
variance:
GGCMs
GCMs
0.0
0.5
1.0
GC6 GC5
Maize
0.12
0.88
0.08
0.92
165
96
24
9
GC6 GC5
Wheat
0.3
0.7
0.08
0.92
175
422
75
38
GC6 GC5
Soybean
0.34
0.66
0.02
0.98
63
300
32
5
0.0
0.5
1.0
GC6 GC5
Rice
Variance induced by GCMs
Variance induced by GGCMs
25
Methods
509
Time of emergence metric
510
We define Time of Climate Impact Emergence (TCIE) as the year in which the smoothed climate
511
change signal (‘signal’) exceeds the underlying internal variability and model uncertainty (‘noise’). The
512
signal is the multi-model ensemble mean crop productivity change against the 1983-2013 reference
513
period (smoothed with a 25-yr moving window). Noise is defined as the standard deviation of simulated
514
historical variability of crop productivity across all individual GCM x GGCM combinations (1983-2013).
515
TCIE is the first year in which the signal emerges from the noise, i.e., when the signal-to-noise ratio
516
becomes greater than 1. Similar time of emergence definitions have been used in previous
517
studies
e.g.10,13,68,69
. Historical productivity time series are not detrended as we hold all management
518
factors constant throughout the simulations. To assess TCIE uncertainties, we calculate TCIE also for
519
each individual climate-crop model realization as suggested by Hawkins and Sutton 2012
12
, and we
520
analyze the distribution of the individual estimates (including mean, median, inter-quartile range, and
521
SD). We find that the multi-model ensemble mean TCIE usually occurs between the median and the
522
mean of individual TCIE estimates. For example, global-level maize production under RCP8.5 shows a
523
multi-model ensemble mean TCIE in year 2032, the median of individual estimates occurs in year 2027,
524
the mean in year 2036. Wheat shows the same pattern and results are qualitatively the same across
525
the different methods. To test the robustness of results in another way, we calculate the multi-model
526
ensemble mean TCIE iteratively while removing one crop model at a time. The SD of this distribution at
527
global level is marginal; 1.5 years for both maize and wheat under RCP8.5. As a final metric, we also
528
compare the number of climate and crop model combinations that show an emergence signal by the
529
end of the century. We calculate TCIE at global level, for different Koeppen-Geiger climate zones, and
530
for individual grid cells. Earlier TCIE is generally found for larger spatial scales as the variance of
531
internal variability decreases with averaging. For additional discussions see for example references
11–
532
13
.
533
26
ISIMIP climate input datasets
534
GGCMI simulation efforts for CMIP6 impact assessment are aligned with the ISIMIP
3
activity in which
535
GGCMI represents the agriculture sector. Key modeling inputs such as information on climate, land
536
use, fertilizer input, soils, among others, are harmonized across various research sectors. CMIP6
537
climate model outputs are centrally bias-adjusted and downscaled by the ISIMIP framework to provide
538
climate-input datasets on a daily regular 0.5°x0.5° global grid. The bias-adjustment method employs a
539
quantile mapping approach and uses the observational W5E5 v1.0 dataset
67,68
. This historical dataset
540
compares favorably with climatic forcing datasets that have been used previously by AgMIP GGCMI
69
.
541
The new quantile-mapping method adjusts biases and preserves trends in all quantiles of the
542
distribution of simulated daily climate model outputs; for more details see Lange (2019)
10
. To lower the
543
barrier for participation in this study we provide climate input data for five CMIP6 GCMs: GFDL-ESM4,
544
IPSL-CM6A-LR, MPI-ESM1-2-HR, MRI-ESM2-0, UKESM1-0-LL (see Table S1 for further details). The
545
GCM selection is based on data availability at the time of selection, performance in the historical period,
546
structural independence, process representation and equilibrium climate sensitivity (ECS). The five
547
GCMs are structurally independent in terms of their ocean and atmosphere model components and
548
overall they represent the range of ECS across the full CMIP6 ensemble, including three models with
549
below-average ECS (GFDL-ESM4, MPI-ESM1-2-HR, MRI-ESM2-0) and two models with above-
550
average ECS (IPSL-CM6A-LR, UKESM1-0-LL)
8
. ECS and transient climate response (TCR) for all
551
GCMs used are listed in Table S1. The mean and standard deviation (SD) of both ECS (mean = 3.7°C,
552
SD = 1.1) and TCR (mean = 2.0°C, SD = 0.5) across the five GCMs used here precisely match the
553
mean and SD across the full CMIP6 ensemble with 38 members (Table S1 and S2), much better than
554
in GC5, although the range of ECS in the CMIP6 ISIMIP models is larger than in the CMIP5 ISIMIP
555
models.
556
557
The daily weather variables at a 0.5° spatial resolution that are used as input for the crop models
558
include: daily mean, minimum, and maximum 2-m air temperature (T, Tmin, and Tmax, respectively
559
27
[°C]), daily total precipitation (P [mm]), and daily mean shortwave and longwave radiation (SR and LR
560
[W/m2]).
561
GGCMI Phase 3 crop modeling protocol
562
Bias-adjusted climate model projections are used to drive transient crop model simulations, i.e.,
563
uninterrupted runs for the historical (1850-2014), and future (2015-2100) time period. Potential future
564
trajectories are represented by SSP1 with RCP2.6 (here SSP126) and SSP5 with RCP8.5 (here
565
SSP585). Therefore, each crop model performs 20 future simulation runs for each crop (5 GCM x 2
566
RCP x 2 [CO
2
] settings). Note that in this study any socio-economic forcing or adaptation effort
567
associated with the SSP storylines is held constant at the year 2015 level to isolate the climate signal
568
(i.e., year 2015 land-use, fertilizer application, growing seasons, crop cultivars, but also NO
3
and NH
4
569
deposition rates, are used in years after 2015). To help isolate yield effects associated with the CO
2
570
fertilization effect, all crop model simulations are run for two separate assumptions: i) transient [CO
2
] in
571
line with the respective RCP (‘default [CO
2
]’), and ii) [CO
2
] concentration held constant at the 2015 level
572
at 399.95 ppmv (‘constant [CO
2
]’). Differences between the two [CO
2
] levels are not a measure of [CO
2
]
573
uncertainty, as there is no plausible climate change scenario without increasing [CO
2
]
22
. Instead, this
574
setup is used to quantify the size of the CO
2
fertilization effect. All simulations are carried out at the 0.5°
575
global grid. In addition to the GCM forcing, we include historical simulations based on the reanalysis
576
product GSWP3-W5E5 v1.0
67,68
for each crop model and crop to better evaluate crop model
577
performance against observational data.
578
579
We focus on the four major global grain crops, that is, maize (
Zea mays L.
), wheat (
Triticum sp. L.
), rice
580
(
Oryza sativa L.
), and soybean (
Glycine max L. Merr.
). Wheat is simulated as winter and spring wheat
581
individually; grain and silage maize are not distinguished. These four main crops contribute 90% of
582
today’s global caloric production of all cereals and soybean
70
.
583
584
28
All crops are simulated under both rainfed conditions and full irrigation (where soil moisture is set to
585
field capacity every day, without constraints to water availability) in all grid cells — independent of the
586
current cropland distribution. The physical cropland extent is applied in post-processing based on the
587
MIRCA2000 (Monthly Irrigated and Rainfed Crop Areas around the year 2000) reference dataset
71
and
588
irrigated fractions are adapted from Siebert et al. (2015)
72
; both are held constant over time.
589
590
Soil moisture and soil temperature for various soil layers are calculated by most crop models in a
591
transient way, that is, without reinitializing at the beginning of each year. All models use a classic
592
phenological heat sum approach to determine physiological stages between planting and maturity. Heat
593
unit accumulation can be modified by the sensitivity to day length (photoperiod) and for winter wheat it
594
is stalled until vernalization requirements are reached, that is, the exposure to cold temperatures before
595
anthesis. Planting dates (see section ‘Crop calendar and crop varieties’ below) are constant over time
596
but the heat sum approach leads to different growing season lengths depending on the daily
597
temperature distribution in each growing season. Except for rice, we simulate only one growing season
598
per calendar year. The first and last years of the transient runs are removed from crop model
599
simulations due to partially incomplete growing seasons. Simulations in grid cells with a growing
600
season length less than 50 days are removed, as are simulations resulting in premature harvest (i.e.,
601
accumulated heat units <80% of required heat units and applies only to those models that can provide
602
such outputs).
603
604
The harmonization of crop models includes the required use of a central crop calendar product (new
605
development for this study, see below), fertilizer inputs, and soil information. Additional protocol
606
characteristics are recommended but not required, as not all models can address all features (see
607
below).
608
609
Simulation protocols determine mineral and organic fertilizer [kg N/ha] inputs per crop and grid cell.
610
Mineral fertilizer (ammonium nitrate; NH
4
NO
3
) application is crop-specific and is derived from the LUH2
611
29
product
73,74
, harmonized by ISIMIP. Manure application inputs (C:N ratio of 14.5) are grid cell specific,
612
but constant across crops
75
. All other nutrients are considered non-limiting. Fertilizer scheduling follows
613
a simple assumption with 20% applied at sowing and 80% applied when 25% of the heat units required
614
to reach maturity are accumulated. As all other management aspects, fertilizer application is held
615
constant throughout the simulation period. Atmospheric N deposition is considered, separating NH
x
and
616
NO
y
, based on Tian et al. (2018)
76
and held constant at the year 2015 level.
617
618
Soil input is harmonized across crop models for the first time in GGCMI, derived from the Harmonized
619
World Soil Database (HWSD)
77
. While the same HWSD dataset is used across ISIMIP sectors, in this
620
study we employ a different algorithm to aggregate the data to 0.5° in order to be cropland specific. The
621
pDSSAT model uses the Global Soil Data set for Earth system modeling (GSDE)
78
and DSSAT-Pythia
622
uses the Global High-Resolution Soil Profile Database for Crop Modeling Applications
79
due to
623
difficulties in retrieving all soil parameters from HWSD.
624
625
Finally, the following management aspects are encouraged to be harmonized across crop models, but
626
are not accounted for by all teams: tillage (2 tillage events, planting day and harvest day, 200 mm
627
depth, full inversion), residues (70% of above-ground residues removed), no pest and disease damage,
628
no soil erosion, and no cover crops. Except for rice and wheat, which are simulated for two separate
629
growing seasons, we do not consider multi-cropping systems or crop rotations. Inputs are provided for
630
18 different crops globally, but most crop models can only simulate the major crops, which we focus on
631
in this study. All socio-economic and farm management input data are publicly available via
632
www.isimip.org.
633
Participating GGCMI crop models
634
Twelve process-based global crop models participate in this study: ACEA, CROVER, CYGMA1p74,
635
DSSAT-Pythia, EPIC-IIASA, ISAM, LandscapeDNDC, LPJmL, pDSSAT, PEPIC, PROMET,
636
SIMPLACE-LINTUL5 (see Table S3 for further details and references). The full ensemble, therefore,
637
30
consists of roughly 240 future crop model simulations per crop plus one historical reference run for
638
each crop and climate model and one historical reanalysis run per crop model. Due to computational
639
constraints, ACEA has only run GCMs UKESM1-0-LL and MRI-ESM2-0 so far, and DSSAT-Pythia has
640
not yet run UKESM1-0-LL. ACEA and DSSAT-Pythia have not yet finished simulations for the constant
641
[CO
2
] setting.
642
643
All crop models are considered independent. LPJmL, pDSSAT, EPIC-IIASA, PROMET, and PEPIC
644
have participated in previous GGCMI protocols
7,80–82
, and while the other models are new GGCMI
645
ensemble members, they have been thoroughly evaluated individually (see references in Table S3). In
646
order to participate in this study, each model was required to go through a benchmark performance
647
evaluation for the historical period based on GSWP3-W5E5 reanalysis data (results available upon
648
request). An overview of the degree to which the GC6 crop models explain observed inter-annual yield
649
variability is presented in Figure S11. For the top five producer countries per crop, the ensemble mean
650
generally shows higher performance in terms of correlation and root-mean-square error than the bulk of
651
individual models. Generally, explained variability in individual models is satisfactory for most maize,
652
wheat, and soybean main-producer countries. The metrics are lower for rice which also links to the fact
653
that the weather signal in (largely irrigated) rice is smaller than in other crops, and the overall observed
654
inter-annual variability in these rice producer countries is smaller than for the other crops. Since
655
management decisions (planting dates, crop rotations and areas, fertilizer application, irrigation, etc.)
656
are held constant over time, the crop models can only capture the interannual weather signal in
657
reported yields, which in general is much smaller in the tropics compared to mid- to high-latitude
658
regions. Additional in-depth GGCMI model comparison and evaluation is presented by Müller et al.
659
(2017)
81
. Overall, crop model performance evaluation based on historical yield variability provides
660
limited insight into the models’ capability to project future yield impacts
83
.
661
662
Since GCM-based crop model simulations are difficult to compare with observed inter-annual yield
663
levels (e.g., the 1988 drought does not necessarily occur in 1988 in the GCM), we compare the overall
664
31
range of simulated and observed yield variability across the historical reference period. The standard
665
deviation of observed national yield variability is matched to a substantially higher degree in GC6 (R =
666
79%, RMSE = 0.11) than in GC5 (R = 44%, RMSE = 0.17), which is indicative of more realistic yield
667
responses in GC6 (Fig. S10). These improvements are linked to a combination of factors, including
668
different internal variability in CMIP6, new GCM bias-adjustment method, improved crop model
669
ensemble, new crop yield bias-correction, and improved crop model inputs. The match with observed
670
yield variability using GC6 simulations based on GSWP3-W5E5 reanalysis data is only slightly better (R
671
= 87%, RMSE = 0.09) than with GCM-forced simulations, which highlights that the CMIP6 GCMs do not
672
introduce substantial errors in terms of historical variability (Fig. S10).
673
674
While the models generally reproduce yield declines in extreme years, adverse impacts of excess water
675
on crop growth due to lower aeration, waterlogging, and nitrogen leaching are generally
676
underrepresented in current global crop models
39
. As an exception, the crop model CYGMA accounts
677
for effects due to excess moisture stress
84
. ACEA, EPIC-based, and DSSAT-based crop models also
678
have processes related to waterlogging and root aeration but associated stresses occur rarely and
679
foremost on sensitive soils
85
. Many models do not handle direct effects of extreme heat (e.g., on leaf
680
senescence, pollen sterility; see Table S3)
3
. Individual model responses to elevated [CO
2
] are shown in
681
Figure 7 and S8 and discussed in the main text. The ISAM model requires sub-daily weather data and
682
therefore uses CRU–National Centers for Environmental Prediction (CRUNCEP) diurnal factors to
683
convert daily bias-adjusted climate model data to diurnal data. The PROMET model also requires sub-
684
daily weather data and uses ERA5-derived diurnal factors to convert climate model data to diurnal
685
inputs; it also uses WFDE5 instead of GSWP3-W5E5 for reanalysis simulations.
686
687
All models use spin-up simulations of various lengths to reach soil and carbon pool equilibrium. EPIC-
688
IIASA uses dynamic soil handling during spin-up to generate soil attributes. Subsequently these are
689
used as an input in the actual simulations with static soil handling, i.e. annual re-initialization of all soil
690
attributes (including soil organic matter fractions and soil texture among others) except mineral nutrient
691
32
pools, temperature, and soil moisture. The models do not account for human management intervention
692
other than fertilizer application, irrigation, seed selection, growing periods, and basic field management
693
such as tillage and residue removal.
694
695
All models follow a phenology calibration with respect to grid cell-specific cultivar parameterizations
696
(i.e., phenological heat units) based on the respective crop calendar and weather forcing (Table S3).
697
Yield calibration is not harmonized across crop models and each team follows their individual protocol,
698
including grid cell-specific calibration against SPAM
86
reference yields (e.g., pDSSAT), various site-
699
specific efforts based on field experiments (e.g., ISAM), and calibrations with national FAO
70
statistics
700
(e.g., PEPIC).
701
Crop yield bias correction
702
Crop production is calculated as yield times harvested area of the respective crop. We omit grid cells
703
with <10 ha cropland area for each crop. To compare results across crop models, but also to represent
704
realistic overall crop production estimates and spatial pattern, we calculate fractional yield changes
705
from each individual crop model simulation between the historical reference period (1983-2013) and the
706
respective future projection and multiply these with a spatially explicit (0.5°) observational yield
707
reference dataset (see Fig. S14 in ref.
87
). SPAM2005 (Spatial Production Allocation Model 2005)
88
is
708
used as the main reference yield data as it separates rainfed and irrigated systems. Grid cells with
709
missing SPAM2005 yield data but with >10 ha MIRCA2000 harvested area are gap-filled with Ray et al.
710
(2012)
89
yield data; both SPAM2005 and Ray et al. represent the time period 2003 to 2007.
711
Winter and spring wheat separation
712
While winter and spring wheat are simulated separately by the crop models covering all land areas, our
713
analyses distinguish winter and spring wheat harvested areas using a rule-based approach. We
714
assume that winter wheat is grown in a specific grid cell if: i) the average temperature of the coldest
715
month is between -10°C and +7°C, ii) the growing season length exceeds 150 days, and iii) the growing
716
33
season includes December (Northern Hemisphere) or July (Southern Hemisphere). These assumptions
717
are slightly modified from the rule set in MIRCA2000
71
; we use 7°C instead of 6°C as the upper
718
temperature threshold to allow for more winter wheat in Argentina, South Africa, and Australia, but also
719
to extend winter wheat in the US slightly towards the south (Fig. S12). This modification is done to
720
better represent the winter wheat mega environments used by CIMMYT
90
. The winter and spring wheat
721
rule set is also used to separate wheat crop calendars in case the two are not distinguished in the
722
original crop calendar data. In line with other cropland areas as well, winter and spring wheat areas are
723
held constant over time.
724
Crop calendar and crop varieties
725
We provide planting and maturity dates for each crop in each grid cell, separate for rainfed and irrigated
726
systems, based on a new observational crop calendar product. See section ‘GGCMI crop calendar’ and
727
Fig. S13-S15 in the Supplement for details. Growing season inputs are static over time throughout the
728
historical and future time period to avoid confounding trends. Each model calculated required reference
729
heat units to reach physiological maturity for each crop in each grid cell by averaging annual heat sums
730
over all growing seasons between 1979-2010.
731
Koeppen-Geiger climate class aggregation
732
Koeppen-Geiger climate zones
91
are aggregated to 0.5° spatial resolution and the 32 individual classes
733
are aggregated to the following four main climate types: temperature-limited
734
("Dfc","Dfd","Dsc","Dsd","Dwc","Dwd","ET","EF","H","BSk"), temperate/humid
735
("Csb","Cfa","Cfb","Cfc","Csc","Cwa","Cwb","Cwc","Dfa","Dfb","Dsa","Dsb","Dwa","Dwb"),
736
subtropical/Mediterranean ("Csa","BSh","Af","Am","As","Aw"), and tropical/other (all other classes).
737
Map projection and smoothing
738
Global maps are based on the Robinson projection
92
and grid-level data are smoothed to improve
739
clarity and visual appearance. Smoothing is done by first resampling the raw data to 5 times finer
740
34
resolution, followed by a 5x5 grid cell focal mean window aggregation. Map smoothing is done for
741
visualization purposes only and all analyses are based on the raw data.
742
Acknowledgements
743
J.J., A.C.R., C.R., and M.P.P. were supported by NASA GISS Climate Impacts Group and Indicators for
744
the National Climate Assessment funding from the NASA Earth Sciences Division. J.J. received support
745
from the Open Philanthropy Project and thanks the University of Chicago Research Computing Center
746
for supercomputer allocations to run the pDSSAT model. Ludwig-Maximilians-Universität München
747
thanks the Leibniz Supercomputing Center of the Bavarian Academy of Sciences and Humanities for
748
providing capacity on the Cloud computing infrastructure to run the PROMET model. J.M.S.
was
749
supported by the German Federal Ministry of Education and Research (grant-number 031B0230A:
750
BioNex—The Future of the Biomass Nexus). O.M. and J.F.S. were supported by funding from the
751
European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation
752
programme (Earth@lternatives project, grant agreement No 834716). J.A.F. and H.S. were supported
753
by the NSF NRT program (grant no. DGE-1735359). J.A.F was supported by the NSF Graduate
754
Research Fellowship Program (grant no. DGE-1746045). RDCEP is funded by NSF through the
755
Decision Making Under Uncertainty program (grant #SES-1463644). T.I. was partly supported by the
756
Environment Research and Technology Development Fund (2-2005) of the Environmental Restoration
757
and Conservation Agency and Grant-in-Aid for Scientific Research B (18H02317) of the Japan Society
758
for the Promotion of Science. M.O. was supported by the Climate Change Adaptation Research
759
Program of NIES, Japan. S.L. was supported by the German Federal Office for Agriculture and Food
760
(BLE) in the framework of OptAKlim (grant no. 281B203316). S.R. acknowledges funding from the
761
German Federal Ministry of Education and Research (BMBF) via the ISIpedia project.
762
Author contributions
763
J.J. and C.M. conceived the paper and coordinate GGCMI. J.J., C.M., and S.R. developed the
764
simulation protocol. A.R. and C.R. coordinate AgMIP integration. C.M., J.J., J.B., O.C., B.F., C.F., K.F.,
765
G.H. T.I., A.J. N.K, T.L., W.L., S.M., M.O., O.M., C.P. S.R., J.S., J.S. R.S., A.S., T. S., F.Z. conducted
766
crop model simulations, S.L. prepared climate data inputs, J.J. developed the manuscript and figures,
767
all coauthors supported writing and discussion of the results.
768
Data and materials availability
769
All data needed to evaluate the conclusions in the paper are present in the paper and/or the
770
Supplementary Materials. Model inputs are publicly available via
https://www.isimip.org/
or from the
771
corresponding author. Crop model simulations will be made public under the CC0 license pending
772
publication.
773
774
The authors declare no competing interest. This article contains supporting information online.
775
References
776
1.
Mbow, C.
et al.
Food security. in
Climate Change and Land: an IPCC special report on climate
777
change, desertification, land degradation, sustainable land management, food security, and
778
35
greenhouse gas fluxes in terrestrial ecosystems
1
, 270 (2019).
779
2.
Asseng, S.
et al.
Uncertainty in simulating wheat yields under climate change.
Nat. Clim. Chang.
780
3
, 827–832 (2013).
781
3.
Wang, E.
et al.
The uncertainty of crop yield projections is reduced by improved temperature
782
response functions.
Nat. Plants
3
, (2017).
783
4.
Rosenzweig, C.
et al.
The Agricultural Model Intercomparison and Improvement Project (AgMIP):
784
Protocols and pilot studies.
Agric. For. Meteorol.
170
, 166–182 (2013).
785
5.
ISIMIP. The Inter-Sectoral Impact Model Intercomparison Project.
2021
Available at:
786
https://www.isimip.org/.
787
6.
Eyring, V.
et al.
Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6)
788
experimental design and organization.
Geosci. Model Dev.
9
, 1937–1958 (2016).
789
7.
Rosenzweig, C.
et al.
Assessing agricultural risks of climate change in the 21st century in a
790
global gridded crop model intercomparison.
Proc. Natl. Acad. Sci.
111
, 3268–3273 (2014).
791
8.
Meehl, G. A.
et al.
Context for interpreting equilibrium climate sensitivity and transient climate
792
response from the CMIP6 Earth system models.
Sci. Adv.
6
, 1–11 (2020).
793
9.
O’Neill, B. C.
et al.
The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6.
794
Geosci. Model Dev.
9
, 3461–3482 (2016).
795
10.
Lange, S. Trend-preserving bias adjustment and statistical downscaling with ISIMIP3BASD
796
(v1.0).
Geosci. Model Dev.
12
, 3055–3070 (2019).
797
11.
Hawkins, E.
et al.
Observed Emergence of the Climate Change Signal: From the Familiar to the
798
Unknown.
Geophys. Res. Lett.
47
, (2020).
799
12.
Hawkins, E. & Sutton, R. Time of emergence of climate signals.
Geophys. Res. Lett.
39
, 1–6
800
(2012).
801
13.
Kirtman, B.
et al.
Near-term climate change: Projections and predictability.
Clim. Chang. 2013
802
Phys. Sci. Basis Work. Gr. I Contrib. to Fifth Assess. Rep. Intergov. Panel Clim. Chang.
803
9781107057
, 953–1028 (2013).
804
14.
Seneviratne, S. I. & Hauser, M. Regional Climate Sensitivity of Climate Extremes in CMIP6
805
Versus CMIP5 Multimodel Ensembles.
Earth’s Futur.
8
, 1–12 (2020).
806
15.
Rojas, M., Lambert, F., Ramirez-Villegas, J. & Challinor, A. J. Emergence of robust precipitation
807
changes across crop production areas in the 21st century.
Proc. Natl. Acad. Sci.
116
, 6673–
808
6678 (2019).
809
16.
Raymond, C., Matthews, T. & Horton, R. M. The emergence of heat and humidity too severe for
810
human tolerance.
Sci. Adv.
6
, (2020).
811
17.
Park, C. E.
et al.
Keeping global warming within 1.5 °c constrains emergence of aridification.
Nat.
812
Clim. Chang.
(2018). doi:10.1038/s41558-017-0034-4
813
18.
Liu, B.
et al.
Similar estimates of temperature impacts on global wheat yield by three
814
36
independent methods.
Nat. Clim. Chang.
6
, 1130–1136 (2016).
815
19.
Zhao, C.
et al.
Plausible rice yield losses under future climate warming.
Nat. Plants
3
, 1–5
816
(2016).
817
20.
Zhao, C.
et al.
Temperature increase reduces global yields of major crops in four independent
818
estimates.
Proc. Natl. Acad. Sci.
201701762 (2017). doi:10.1073/pnas.1701762114
819
21.
Deryng, D.
et al.
Regional disparities in the beneficial effects of rising CO2 concentrations on
820
crop water productivity.
Nat. Clim. Chang.
(2016). doi:10.1038/nclimate2995
821
22.
Ruane, A. C.
et al.
Biophysical and economic implications for agriculture of +1.5° and +2.0°C
822
global warming using AgMIP Coordinated Global and Regional Assessments.
Clim. Res.
76
, 17–
823
39 (2018).
824
23.
Ahmed, M.
et al.
Novel multimodel ensemble approach to evaluate the sole effect of elevated
825
CO2 on winter wheat productivity.
Sci. Rep.
9
, 1–15 (2019).
826
24.
Leakey, A. D. B., Bishop, K. A. & Ainsworth, E. A. A multi-biome gap in understanding of crop
827
and ecosystem responses to elevated CO 2.
Current Opinion in Plant Biology
(2012).
828
doi:10.1016/j.pbi.2012.01.009
829
25.
Toreti, A.
et al.
Narrowing uncertainties in the effects of elevated CO2 on crops.
Nat. Food
1
,
830
775–782 (2020).
831
26.
Hausfather, Z. & Peters, G. P. Emissions – the ‘business as usual’ story is misleading.
Nature
832
577
, 618–620 (2020).
833
27.
Minoli, S.
et al.
Global Response Patterns of Major Rainfed Crops to Adaptation by Maintaining
834
Current Growing Periods and Irrigation.
Earth’s Futur.
7
, 1464–1480 (2019).
835
28.
Wang, X.
et al.
Emergent constraint on crop yield response to warmer temperature from field
836
experiments.
Nat. Sustain.
3
, 908–916 (2020).
837
29.
Asseng, S.
et al.
Rising temperatures reduce global wheat production.
Nat. Clim. Chang.
5
, 143–
838
147 (2014).
839
30.
Kimball, B. A. Crop responses to elevated CO2 and interactions with H2O, N, and temperature.
840
Current Opinion in Plant Biology
(2016). doi:10.1016/j.pbi.2016.03.006
841
31.
Zabel, F.
et al.
Large potential for crop production adaptation depends on available future
842
varieties.
Glob. Chang. Biol.
gcb.15649 (2021). doi:10.1111/gcb.15649
843
32.
Ray, D. K.
et al.
Climate change has likely already affected global food production.
PLoS One
844
14
, 1–18 (2019).
845
33.
Lobell, D. B., Schlenker, W. & Costa-Roberts, J. Climate trends and global crop production since
846
1980.
Science
333
, 616–20 (2011).
847
34.
Ahmad, S.
et al.
Climate warming and management impact on the change of phenology of the
848
rice-wheat cropping system in Punjab, Pakistan.
F. Crop. Res.
230
, 46–61 (2019).
849
35.
Porter, J. R.
et al.
Food security and food production systems. in
Climate Change 2014: Impacts,
850
37
Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working
851
Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change
852
(eds. Field, C. B. et al.) 485–533 (Cambridge University Press, 2014).
853
36.
Levis, S., Badger, A., Drewniak, B., Nevison, C. & Ren, X. CLMcrop yields and water
854
requirements: avoided impacts by choosing RCP 4.5 over 8.5.
Clim. Change
146
, 501–515
855
(2018).
856
37.
Falconnier, G. N.
et al.
Modelling climate change impacts on maize yields under low nitrogen
857
input conditions in sub
-
Saharan Africa.
Glob. Chang. Biol.
26
, 5942–5964 (2020).
858
38.
O’Neill, B. C.
et al.
IPCC reasons for concern regarding climate change risks.
Nat. Clim. Chang.
859
7
, 28–37 (2017).
860
39.
Li, Y., Guan, K., Schnitkey, G. D., DeLucia, E. & Peng, B. Excessive rainfall leads to maize yield
861
loss of a comparable magnitude to extreme drought in the United States.
Glob. Chang. Biol.
25
,
862
2325–2337 (2019).
863
40.
Zhu, P., Zhuang, Q., Archontoulis, S. V., Bernacchi, C. & Müller, C. Dissecting the nonlinear
864
response of maize yield to high temperature stress with model-data integration.
Glob. Chang.
865
Biol.
25
, 2470–2484 (2019).
866
41.
Christensen, J. H.
et al.
Regional Climate Projections. in
Climate Change 2007: The Physical
867
Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the
868
Intergovernmental Panel on Climate Change
(ed. Solomon, S., D. Qin, M. Manning, Z. Chen, M.
869
Marquis, K.B. Averyt, M. T. and H. L. M.) 11–15 (Cambridge University Press, 2007).
870
doi:10.1007/978-81-322-1967-5_4
871
42.
Iizumi, T.
et al.
Responses of crop yield growth to global temperature and socioeconomic
872
changes.
Sci. Rep.
7
, 1–10 (2017).
873
43.
Sherwood, S. C.
et al.
An Assessment of Earth’s Climate Sensitivity Using Multiple Lines of
874
Evidence.
Rev. Geophys.
58
, 1–92 (2020).
875
44.
Flynn, C. M. & Mauritsen, T. On the climate sensitivity and historical warming evolution in recent
876
coupled model ensembles.
Atmos. Chem. Phys.
20
, 7829–7842 (2020).
877
45.
Zelinka, M. D.
et al.
Causes of Higher Climate Sensitivity in CMIP6 Models.
Geophys. Res. Lett.
878
47
, 1–12 (2020).
879
46.
Tokarska, K. B.
et al.
Past warming trend constrains future warming in CMIP6 models.
Sci. Adv.
880
6
, 1–14 (2020).
881
47.
Williams, K. D., Hewitt, A. J. & Bodas-Salcedo, A. Use of Short-Range Forecasts to Evaluate
882
Fast Physics Processes Relevant for Climate Sensitivity.
J. Adv. Model. Earth Syst.
12
, 1–9
883
(2020).
884
48.
Brunner, L.
et al.
Reduced global warming from CMIP6 projections when weighting models by
885
performance and independence.
Earth Syst. Dyn.
11
, 995–1012 (2020).
886
38
49.
Nijsse, F. J. M. M., Cox, P. M. & Williamson, M. S. Emergent constraints on transient climate
887
response (TCR) and equilibrium climate sensitivity (ECS) from historical warming in CMIP5 and
888
CMIP6 models.
Earth Syst. Dyn.
11
, 737–750 (2020).
889
50.
Ridder, N. N., Pitman, A. J. & Ukkola, A. M. Do CMIP6 Climate Models simulate Global or
890
Regional Compound Events skilfully?
Geophys. Res. Lett.
1–11 (2020).
891
doi:10.1029/2020gl091152
892
51.
Fan, X., Miao, C., Duan, Q., Shen, C. & Wu, Y. The Performance of CMIP6 Versus CMIP5 in
893
Simulating Temperature Extremes Over the Global Land Surface.
J. Geophys. Res. Atmos.
125
,
894
1–16 (2020).
895
52.
Xin, X., Wu, T., Zhang, J., Yao, J. & Fang, Y. Comparison of <scp>CMIP6</scp> and
896
<scp>CMIP5</scp> simulations of precipitation in China and the East Asian summer monsoon.
897
Int. J. Climatol.
40
, 6423–6440 (2020).
898
53.
Meinshausen, M.
et al.
The shared socio-economic pathway (SSP) greenhouse gas
899
concentrations and their extensions to 2500.
Geosci. Model Dev.
13
, 3571–3605 (2020).
900
54.
Von Bloh, W.
et al.
Implementing the nitrogen cycle into the dynamic global vegetation,
901
hydrology, and crop growth model LPJmL (version 5.0).
Geosci. Model Dev.
11
, 2789–2812
902
(2018).
903
55.
Jägermeyr, J. & Frieler, K. Spatial variations in crop growing seasons pivotal to reproduce global
904
fluctuations in maize and wheat yields.
Sci. Adv.
4
, eaat4517 (2018).
905
56.
Müller, C.
et al.
Exploring uncertainties in global crop yield projections in a large ensemble of
906
crop models and CMIP5 and CMIP6 climate scenarios.
Environ. Res. Lett.
16
, 034040 (2021).
907
57.
Franke, J. A.
et al.
The GGCMI Phase 2 emulators: global gridded crop model responses to
908
changes in CO2, temperature, water, and nitrogen (version 1.0).
Geosci. Model Dev.
13
, 2315–
909
2336 (2020).
910
58.
Allen, L. H.
et al.
Fluctuations of CO2 in Free-Air CO2 Enrichment (FACE) depress plant
911
photosynthesis, growth, and yield.
Agric. For. Meteorol.
284
, (2020).
912
59.
Durand, J. L.
et al.
How accurately do maize crop models simulate the interactions of
913
atmospheric CO2 concentration levels with limited water supply on water use and yield?
Eur. J.
914
Agron.
(2018). doi:10.1016/j.eja.2017.01.002
915
60.
Myers, S. S.
et al.
Increasing CO2 threatens human nutrition.
Nature
510
, 139–42 (2014).
916
61.
Zhu, C.
et al.
Carbon dioxide (CO2) levels this century will alter the protein, micronutrients, and
917
vitamin content of rice grains with potential health consequences for the poorest rice-dependent
918
countries.
Sci. Adv.
4
, eaaq1012 (2018).
919
62.
Köhler, I. H., Huber, S. C., Bernacchi, C. J. & Baxter, I. R. Increased temperatures may
920
safeguard the nutritional quality of crops under future elevated CO2 concentrations.
Plant J.
97
,
921
872–886 (2019).
922
39
63.
Rising, J. & Devineni, N. Crop switching reduces agricultural losses from climate change in the
923
United States by half under RCP 8.5.
Nat. Commun.
11
, 4991 (2020).
924
64.
Asseng, S.
et al.
Climate change impact and adaptation for wheat protein.
Glob. Chang. Biol.
25
,
925
155–173 (2019).
926
65.
Hawkins, E. & Sutton, R. The potential to narrow uncertainty in regional climate predictions.
Bull.
927
Am. Meteorol. Soc.
90
, 1095–1107 (2009).
928
66.
Giorgi, F. & Bi, X. Time of emergence (TOE) of GHG-forced precipitation change hot-spots.
929
Geophys. Res. Lett.
36
, L06709 (2009).
930
67.
Lange, S. WFDE5 over land merged with ERA5 over the ocean (W5E5).
V. 1.0. GFZ Data
931
Services
(2019). doi:https://doi.org/10.5880/pik.2019.023
932
68.
Cucchi, M.
et al.
WFDE5: Bias-adjusted ERA5 reanalysis data for impact studies.
Earth Syst.
933
Sci. Data
12
, 2097–2120 (2020).
934
69.
Ruane, A. C.
et al.
Strong regional influence of climatic forcing datasets on global crop model
935
ensembles.
Agric. For. Meteorol.
300
, 108313 (2021).
936
70.
FAO. FAOSTAT, United Nation’s Food and Agricultural Organization, Rome. (2019). Available
937
at: http://www.fao.org/faostat/. (Accessed: 10th July 2019)
938
71.
Portmann, F. T., Siebert, S. & Döll, P. MIRCA2000 - Global monthly irrigated and rainfed crop
939
areas around the year 2000: A new high-resolution data set for agricultural and hydrological
940
modeling.
Global Biogeochem. Cycles
24
, 1–24 (2010).
941
72.
Siebert, S.
et al.
A global data set of the extent of irrigated land from 1900 to 2005.
Hydrol. Earth
942
Syst. Sci.
19
, 1521–1545 (2015).
943
73.
Mueller, N. D.
et al.
Closing yield gaps through nutrient and water management.
Nature
490
,
944
254–7 (2012).
945
74.
Hurtt, G. C.
et al.
Harmonization of global land use change and management for the period 850–
946
2100 (LUH2) for CMIP6.
Geosci. Model Dev.
13
, 5425–5464 (2020).
947
75.
Zhang, B.
et al.
Global manure nitrogen production and application in cropland during 1860-
948
2014: A 5 arcmin gridded global dataset for Earth system modeling.
Earth Syst. Sci. Data
9
,
949
667–678 (2017).
950
76.
Tian, H.
et al.
The Global N2O Model Intercomparison Project.
Bull. Am. Meteorol. Soc.
99
,
951
1231–1251 (2018).
952
77.
Nachtergaele, F.
et al.
Harmonized World Soil Database (version 1.2).
FAO, Rome, Italy IIASA,
953
Laxenburg, Austria
1–50 (2012). doi:3123
954
78.
Shangguan, W., Dai, Y., Duan, Q., Liu, B. & Yuan, H. A global soil data set for earth system
955
modeling.
J. Adv. Model. Earth Syst.
6
, 249–263 (2014).
956
79.
Hengl, T.
et al.
SoilGrids1km - Global soil information based on automated mapping.
PLoS One
957
9
, (2014).
958
40
80.
Elliott, J.
et al.
The Global Gridded Crop Model Intercomparison: data and modeling protocols for
959
Phase 1 (v1.0).
Geosci. Model Dev.
8
, 261–277 (2015).
960
81.
Müller, C.
et al.
Global gridded crop model evaluation: benchmarking, skills, deficiencies and
961
implications.
Geosci. Model Dev.
10
, 1403–1422 (2017).
962
82.
Franke, J. A.
et al.
The GGCMI Phase 2 experiment: global gridded crop model simulations
963
under uniform changes in CO2, temperature, water, and nitrogen levels (protocol version 1.0).
964
Geosci. Model Dev.
13
, 2315–2336 (2020).
965
83.
Ruane, A. C.
et al.
Multi-wheat-model ensemble responses to interannual climate variability.
966
Environ. Model. Softw.
81
, 86–101 (2016).
967
84.
Wang, R., Bowling, L. C. & Cherkauer, K. A. Estimation of the effects of climate variability on
968
crop yield in the Midwest USA.
Agric. For. Meteorol.
216
, 141–156 (2016).
969
85.
Folberth, C., Gaiser, T., Abbaspour, K. C., Schulin, R. & Yang, H. Regionalization of a large-
970
scale crop growth model for sub-Saharan Africa: Model setup, evaluation, and estimation of
971
maize yields.
Agric. Ecosyst. Environ.
151
, 21–33 (2012).
972
86.
International Food Policy Research Institute. Global Spatially-Disaggregated Crop Production
973
Statistics Data for 2010 Version 1.0.
Harvard Dataverse
V1 (2019). Available at:
974
https://doi.org/10.7910/DVN/PRFF8V. (Accessed: 15th February 2019)
975
87.
Jägermeyr, J.
et al.
A regional nuclear conflict would compromise global food security.
Proc.
976
Natl. Acad. Sci.
117
, 7071–7081 (2020).
977
88.
International Food Policy Research Institute (IFPRI) & International Institute for Applied Systems
978
Analysis (IIASA). Global Spatially-Disaggregated Crop Production Statistics Data for 2005
979
Version 3.2.
Harvard Dataverse
V9 (2016). Available at: https://doi.org/10.7910/DVN/DHXBJX.
980
(Accessed: 15th February 2019)
981
89.
Ray, D. K., Ramankutty, N., Mueller, N. D., West, P. C. & Foley, J. a. Recent patterns of crop
982
yield growth and stagnation.
Nat. Commun.
3
, 1293 (2012).
983
90.
Reynolds, M. & Braun, H. Benefits to low-input agriculture.
Nat. Plants
5
, 652–653 (2019).
984
91.
Kottek, M., Grieser, J., Beck, C., Rudolf, B. & Rubel, F. World Map of the Köppen-Geiger climate
985
classification updated.
Meteorol. Zeitschrift
15
, 259–263 (2006).
986
92.
John P. Snyder & Voxland, P. M.
An album of map projections
. (1989). doi:10.3133/pp1453
987
988
Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.