Originally published as:
Qiao, L., Wang, X., Smith, P., Fan, J., Lu, Y., Emmett, B., Li, R., Dorling, S., Chen, H., Liu, S.,
Benton, T. G., Wang, Y., Ma, Y., Jiang, R., Zhang, F., Piao, S.,
, Yang, H., Hao, Y.,
Li, W., Fan, M. (2022): Soil quality both increases crop production and improves
resilience to climate change. - Nature Climate Change, 12, 6, 574-580.
DOI:
https://doi.org/10.1038/s41558-022-01376-8
1
Inventory
of
Supporting
Information
1
2
NCLIM-21122311A:
Soil quality both increases crop production and improves resilience to climate change.
3
4
Corresponding author name(s):
Mingsheng Fan.
5
6
1.
Extended Data
7
8
9
Figure #
Figure title
One sentence only
Filename
This should be the name
the file is saved as when it
is uploaded to our system.
Please include the file
extension. i.e.:
Smith_ED_Fig1.jpg
Figure Legend
If you are citing a reference for the first time in these legends, please include
all new references in the main text Methods References section, and carry on
the numbering from the main References section of the paper. If your paper
does not have a Methods section, include all new references at the end of the
main Reference list.
Extended Data Fig. 1
Yield
under
best
management
practices
Fan_ED_Fig
1.eps
a-c,
wheat,
d-f,
maize,
and
g-i
rice.
The
filled
orange
corresponds
to
a
bar
for
each
individual
Yield
BMPs
site-year
2
(Yield
BMPs
)
(ranked
from
high
to
low,
Numbers
shown
on
each
panel).
The
blue
dashed
lines
indicate
mean
Yield
BMPs
.
Yield
BMPs
(Mg/ha)
are
shown
as
mean
(±SD,
standard
deviation);
Coefficient
of
variation
(CV,
%)
calculated
by
dividing
mean
yield
by
SD;
N
refers
to
the
number
of
site-years
of
on-farm
trials
in
major
cropping
systems
in
China.
W-NCP,
winter
wheat
in
North
China
Plain;
W-YZB,
winter
wheat
in
Yangtze
River
Basin;
W-NWC,
winter
wheat
in
Northwest
China;
M-NEC,
rainfed
maize
in
Northeast
China;
M-NCP,
maize
in
North
China
Plain;
M-SWC,
rainfed
maize
in
Southwest
China;
SR-YZB,
single
rice
in
Yangtze
River
Basin;
ER-SC,
early
rice
in
South
China;
LR-SC,
late
rice
in
South
China
Extended Data Fig. 2
The
relative
contribution
(%)
of
explaining
variables
to
yields
under
best
management
practices.
Fan_ED_Fig
2.eps
Assessment
was
conducted
by
Gradient
Boosted
Regression
Tree
models
based
on
the
primary
data
comprising
all
on-farm
trials
in
major
cropping
systems
in
China.
Orange,
green
and
blue
bars
indicate
climate,
soil
and
management
variables
respectively.
Tmax
and
Tmin,
maximum
and
minimum
temperature;
PRE,
precipitation;
GDD,
growing
degree
days;
RAD,
solar
radiation;
SOM,
soil
organic
matter;
Olsen-P
and
Avail-K,
soil
available
phosphorus
and
potassium.
W-NCP,
winter
wheat
in
North
China
Plain;
W-YZB,
winter
wheat
in
Yangtze
River
Basin;
W-NWC,
winter
wheat
in
Northwest
China;
M-NEC,
rainfed
maize
in
Northeast
China;
M-NCP,
maize
in
North
China
Plain;
M-SWC,
rainfed
maize
in
Southwest
China;
SR-YZB,
single
rice
in
Yangtze
River
Basin;
ER-SC,
early
rice
in
South
China;
LR-SC,
late
rice
in
South
China.
3
Extended Data Fig. 3
Geographical
distribution
of
paired
on-
farm
trials
under
high-
and
low-quality
soils.
Fan_ED_Fig
3.eps
Symbols
of
blue
dot
and
red
triangle
indicate
on-farm
trials
conducted
in
high-
and
low-quality
soils
in
major
cropping
systems
in
China,
respectively.
Paired
on-farm
trials
were
conducted
in
high-(N=1665)
and
low
-
(N=1676)
quality
soils.
W-NCP,
winter
wheat
in
North
China
Plain;
W-YZB,
winter
wheat
in
Yangtze
River
Basin;
W-NWC,
winter
wheat
in
Northwest
China;
M-NEC,
rainfed
maize
in
Northeast
China;
M-
NCP,
maize
in
North
China
Plain;
M-SWC,
rainfed
maize
in
Southwest
China;
SR-YZB,
single
rice
in
Yangtze
River
Basin;
ER-
SC,
early
rice
in
South
China;
LR-SC,
late
rice
in
South
China.
Extended Data Fig. 4
Comparison
of
climate
variables
between
locations
of
paired
on-
farm
trials
conducted
in
high-
and
low-
quality
soils.
Fan_ED_Fig
4.eps
a-d,
average
maximum
temperature
(a),
average
minimum
temperature
(b),
cumulative
precipitation
(c)
and
cumulative
radiation
(d)
between
locations,
where
paired
on-farm
trials
were
conducted
in
high-(N=1665)
and
low
-
(N=1676)
quality
soils.
W-NCP,
winter
wheat
in
North
China
Plain;
W-YZB,
winter
wheat
in
Yangtze
River
Basin;
W-NWC,
winter
wheat
in
Northwest
China;
M-NEC,
rainfed
maize
in
Northeast
China;
M-
NCP,
maize
in
North
China
Plain;
M-SWC,
rainfed
maize
in
Southwest
China;
SR-YZB,
single
rice
in
Yangtze
River
Basin;
ER-
SC,
early
rice
in
South
China;
LR-SC,
late
rice
in
South
China.
***
refers
to
significance
at
p
=
0.001.
Extended Data Fig. 5
Projected
yield
change
in
future
climate
change.
Fan_ED_Fig
5.eps
a,b,
projected
yield
changes
under
RCP2.6
(a)
and
RCP8.5
(b)
pathways
up
to
2040-2059;
c,d,
projected
yield
changes
under
RCP2.6
(c)
and
RCP8.5
(d)
pathways
up
to
2080-2099.
Projections
were
based
on
Gradient
Boosted
Regression
Tree
(GBRT)
model
trained
on
the
primary
data
set
comprising
all
4
on-farm
trials
for
major
cropping
systems
in
China.
Boxes
represent
variability
across
each
cropping
system
over
the
2040-2099
periods.
Solid
lines
and
diamonds
in
this
figure
indicate
median
and
mean
yields,
respectively;
the
boundary
of
the
box
indicates
the
25th
and
75th
percentile;
whisker
caps
denote
the
90th
and
10th
percentiles.
W-NCP,
winter
wheat
in
North
China
Plain;
W-YZB,
winter
wheat
in
Yangtze
River
Basin;
W-NWC,
winter
wheat
in
Northwest
China;
M-NEC,
rainfed
maize
in
Northeast
China;
M-NCP,
maize
in
North
China
Plain;
M-SWC,
rainfed
maize
in
southwest
China;
SR-YZB,
single
rice
in
Yangtze
River
Basin;
ER-SC,
early
rice
in
South
China;
LR-SC,
later
rice
in
South
China.
Extended Data Fig. 6
Projected
yield
changes
and
their
difference
between
high-
and
low-
quality
soils.
Fan_ED_Fig
6.eps
a-d,
projected
yield
changes
and
their
difference
between
high-
and
low-quality
soils
under
climate
change
in
RCP
2.6
(a)
and
RCP
8.5
(b)
up
to
2040-2059,
and
RCP
2.6
(c)
and
RCP
8.5
(d)
up
to
2080-2099.
Projected
yield
changes
were
estimated
based
on
the
primary
dataset
comprising
all
on-farm
trials,
differences
of
yield
changes
were
estimated
based
on
sub-
datasets
comprising
on-farm
trials
with
paired
high-
and
low-
quality
soils
in
major
cropping
systems
in
China.
Horizontal
and
vertical
error
bars
(standard
deviation,
SD)
represent
inter-
annual
variations
of
yield
changes
and
difference
of
yield
change
between
high-
and
low-quality
soils,
respectively.
Solid
lines
represent
significant
difference
in
yield
change
between
high-
and
low-quality
soil
at
p
=
0.10,
while
dashed
lines
represent
no
significant
difference.
W-NCP,
winter
wheat
in
5
North
China
Plain;
W-YZB,
winter
wheat
in
Yangtze
River
Basin;
W-NWC,
winter
wheat
in
Northwest
China;
M-NEC,
rainfed
maize
in
Northeast
China;
M-NCP,
maize
in
North
China
Plain;
M-SWC,
rainfed
maize
in
Southwest
China;
SR-YZB,
single
rice
in
Yangtze
River
Basin;
ER-SC,
early
rice
in
South
China;
LR-SC,
late
rice
in
South
China.
2.
Supplementary Information:
10
11
A.
Flat Files
12
13
Item Present
Filename
This should be the name the
file is saved as when it is
uploaded to our system, and
should include the file
extension. The extension must
be .pdf
A brief, numerical description of file contents.
i.e.:
Supplementary Figures 1-4, Supplementary Discussion, and
Supplementary Tables 1-4.
Supplementary Information
Yes
SI-NCLIM-21122311A.
Supplementary Text S1 to S2, Supplementary Figures S1 to S7,
and Supplementary Table S1 to S6.
Reporting Summary
Yes
Reporting
summary-
NCLIM-21122311A
14
6
Soil quality both increases crop production and improves resilience to
15
climate change
16
Lei Qiao
1,2
, Xuhui Wang
3
, Pete Smith
4
, Jinlong Fan
5
，
Yuelai Lu
6
, Bridget Emmett
7
,
17
Rong Li
8
, Stephen Dorling
9
, Haiqing Chen
10
, Shaogui Liu
11
, Tim G. Benton
12
, Yaojun
18
Wang
13
, Yuqing Ma
1,2
, Rongfeng Jiang
1,2
, Fusuo Zhang
1,2
, Shilong Piao
3
, Christoph
19
M
ϋ
ller
14
, Huaqing Yang
1,2
, Yanan Hao
1,2
, Wangmei Li
1,2
, Mingsheng Fan
1,2*
20
1
College of Resources and Environmental Sciences, China Agricultural University,
21
Beijing 100193, China;
2
National Academy of Agriculture Green Development, China
22
Agricultural University, Beijing 100193, China;
3
Sino-French Institute for Earth
23
System Science, College of Urban and Environmental Science, Peking University,
24
Beijing 100871, China;
4
Institute of Biological and Environmental Sciences, University
25
of Aberdeen, Aberdeen AB24 3UU, United Kingdom;
5
National Satellite
26
Meteorological Center, China Meteorological Administration, Beijing 100081 China;
27
6
School of International Development, University of East Anglia, Norwich NR4 7TJ,
28
United Kingdom;
7
UK Centre for Ecology and Hydrology, Environment Centre Wales,
29
Bangor, Gwynedd, LL57 2UW, United Kingdom;
8
Cultivated Land Quality Monitoring
30
and Protection Center, Ministry of Agriculture and Rural Affairs of the People’s
31
Republic of China, Beijing 100125, China;
9
School of Environmental Sciences,
32
University of East Anglia, Norwich NR4 7TJ, United Kingdom
；
10
College of Land
33
Science and Technology, China Agricultural University, Beijing 100193, China;
34
11
Yangzhou Station of Farmland Quality Protection, Yangzhou 225101 China;
12
Royal
35
Institute of International Affairs, Chatham House, London SW1Y 4LE, United
36
Kingdom;
13
College of Information and Electrical Engineering, China Agricultural
37
University, Beijing 100193, China;
14
Potsdam Institute for Climate Impacts Research,
38
Member of the Leibniz Association, 14473 Potsdam, Germany.
39
*
Correspondence author (fanms@cau.edu.cn).
40
7
Interactions between soil quality and climate change may influence the
41
capacity of croplands to produce sufficient food. Here, we address this issue by
42
using a new dataset of soil, climate and associated yield observations for
12115
43
site-years representing 90% of total cereal production in China. Across crops and
44
environmental conditions, we show that high-quality soils reduced the sensitivity
45
of crop yield to climate variability leading to both higher mean crop yield
46
(10.3±6.7%) and higher yield stability (decreasing variability by
15.6±14.4%).
47
High-quality soils improve the outcome for yields under climate change by 1.7%
48
(0.5-4.0%), compared to low-quality soils. Climate-driven yield change could
49
result in reductions of national cereal production of 11.4Mt annually under
50
PCR8.5 by 2080-2099. While this production reduction was exacerbated by 14%
51
due to soil degradation; it can be reduced by 21% through soil improvement. This
52
study emphasises the vital role of soil quality in agriculture under climate change.
53
54
Food production may have to increase by as much as 60-100% by 2050 to meet
55
projected food demand
due to growing population
1,2
. Growth rates in crop productivity
56
are expected to be driven mainly by technological and agronomic improvements
2-5
, as
57
they were during the Green Revolution. However, agriculture now is facing greater
58
challenges than ever before, because increased global food production must be achieved
59
sustainably and under changing global biophysical stressors
6-11
.
60
Climate variability is known to impact crop production. For example, globally,
61
fluctuations in temperature, precipitation or their interaction were found to explain
62
8
roughly 32–39% of current crop yield variability
12
. Though uncertainties remains about
63
regional and local impacts of climate change, numerous studies have concluded that
64
continued warming will lead to substantial declines in global mean crop yields by the
65
mid-21
st
century, especially for tropical and sub-tropical agriculture
13-16
. At the same
66
time, there is active debate on where and how such warming will impact agriculture in
67
the temperate zone, such as in China or the United States
17-19
. China is the world’s most
68
populous and largest developing country, and agriculture is a fundamental component
69
of its national economy. Agriculture in China feeds ~20% of the global population with
70
only 7% of the world’s arable land, and 5% of water resources
5
, demonstrating its
71
global
importance.
72
Soil is one of the basic biophysical factors which together with climate determine
73
the major patterns of global agricultural land
20
. Soil quality improvement is recognized
74
increasingly as a fundamental mechanism to increase yield of crops and food insecurity
75
could be made even more acute through continuing soil degradation
10,11
. However, little
76
is known about how interactions between soil and climate change influence the capacity
77
of croplands to produce adequate food supply at regional to national and global scales.
78
Exploring the interactive effects of soil and climate on agricultural production at
79
regional and global scales is challenging since both are highly heterogeneous.
80
Assessments of the sensitivity of agricultural output to climate variability and change
81
have, to-date, relied either on process-based crop simulation models
21
or empirical and
82
statistical modelling of crop-climate relationships
22,23
. However, both approaches have
83
often neglected the heterogeneity of soil
24
, due to the quality and accessibility of
84
9
regional and/or global soil data in terms of accuracy and range of measured soil
85
characteristics.
86
Inadequate consideration of soil quality and interactions with climate change
87
impedes our understanding of the food security challenge in the face of rapidly
88
changing biophysical conditions and the implementation of appropriate risk
89
management strategies
24-26
. This is especially true in developing countries, where (a)
90
agriculture is a larger component of gross domestic product; (b) the majority of the
91
world’s food-insecure population resides with low-quality and/or severely degraded
92
soil; and (c) the worst effects of climate variability and change on food systems are
93
anticipated
14,21,27
. Similarly, agricultural production in China is also inherently fragile,
94
since it is also endangered by climate change and soil degradation
5,18
, and is among the
95
countries most affected by climate change
28
.
96
In this study, we focused on understanding the interaction of climate and soil
97
quality on yield and its variability, using a unique dataset of soil and associated yield
98
observations for 12115 site-years, complemented by multiple climate variables,
99
covering three major crops across major production regions which account for 90% of
100
total cereal production in China (Fig. 1, Table S1). We used a data-driven approach
101
based on a machine learning algorithm to quantify the potential benefits of enhanced
102
soil quality on crop yield and its variability under Best Management Practices
103
(Yield
BMPs
, see Methods section) for both current and future climates.
104
105
Yield variation and biophysical explanation
106
10
It is well known that yields under farmers’ practices are highly variable, especially
107
for smallholder systems, and management practices can be a major cause of this
108
variablity
29,30
. We find that Yield
BMPs
are also heterogeneous across and within major
109
cropping systems (Extended Data Fig. 1), though best management practices
110
sustainably increased yields by, on average, 10.6% compared to those under farmers’
111
actual practices
4
over the major cropping systems. The yield variations were measured
112
by both standard deviation (SD) and coefficient of variation (CV, SD/mean*100%),
113
with the former termed as absolute stability and the latter as relative stability
31
.The CV
114
of Yield
BMPs
for wheat, maize and rice were 18-22 %, 17-19 % and 13-16 % across
115
systems, which correspond to 1.2 to 1.5 Mg/ha, 1.4 to 1.8 Mg/ha and 1.1 Mg/ha in
116
absolute terms (SD), respectively. The degree of yield variability in this study was
117
higher than that estimated by Ray et al.
12
, in which average inter-annual yield variability
118
in China corresponded to 0.7, 0.9 and 0.7 Mg/ha for wheat, maize and rice, respectively.
119
This may be because Yield
BMPs
variability in the current study was derived from both
120
geographic and decade-scale temporal variation in climate and/or soil conditions
32
, in
121
contrast to the inter-annual and climate-induced yield variability considered in Ray et
122
al.
12
.
123
A Gradient Boosted Regression Tree statistical model (GBRT) was used to relate
124
biophysical factors to yield variations for each cropping system. The mean error (E)
125
values were relatively small, and were not significantly different from zero. The average
126
of normalized root mean square errors (nRMSE) ranged from 10.5 – 15.6 % across
127
crops and regions (Table S2), indicating good performance of the GBRT model in
128
11
modelling yield
33
.We also compared the GBRT approach with traditional stepwise
129
multiple linear regression (SMLR) for fitting data. In general, the descriptive statistics
130
indicate a higher level of prediction accuracy of the GBRT than the SMLR (Table S2).
131
In addition to prediction accuracy, GBRT also provides the relative importance of each
132
variable with their partial plots representing the marginal effect of single variables on
133
yields. For all cropping systems excluding winter wheat (W-YZB) and single rice (SR-
134
YZB) in the Yangtze River Basin and maize in northeast China (M-NEC), climatic and
135
soil variables were always ranked among the top four to seven explanatory factors
136
(Extended Data Fig. 2), providing evidence for joint climate-soil control in Yield
BMPs
.
137
However, the most influential bio-physical factors varied among cropping systems. For
138
W-YZB and SR-YZB, and M-NEC, nitrogen (N) rate remains the most important factor
139
in determining yield, showing potential for further improvement in N management
140
(Extended Data Fig. 2).
141
142
Buffer effect of high-quality soil to climate variability
143
To assess the buffering effects of high-quality soil to climate variables, we further
144
established a sub-set of data composed of local pairs of high- and low-quality soils
145
farmed using the same BMPs and under the same climate conditions (see Methods,
146
Extended Data Fig. 3 and 4). High-quality and low-quality soils were grouped
147
according to the two most important and sensitive soil factors and their partial plots in
148
explaining crop- and region-specific yield, based on the above GBRT models (Extended
149
Data Fig. 2, Fig. S1-S3). Dependent upon cropping systems, soil organic matter (SOM),
150
12
soil available Phosphorus (soil Olsen-P), and/or soil type and soil texture were
151
identified as the most important factors in explaining yield variations (Table S3). The
152
yield stability was compared between the two soil quality groups by measuring both
153
SD and CV. The mean Yield
BMPs
from high-quality soils were significantly higher, on
154
average by 0.69 Mg/ha across all cropping systems, than those from low-quality soils
155
(Table 1). The SD of yield produced on low-quality soils was either similar or
156
significantly higher than those in high-quality soils (Table 1). Accordingly, the CV in
157
all cropping systems was consistently lower in high-quality soils despite very small
158
differences being found in rice systems, suggesting higher yield stability under high-
159
quality soils. On average, high-quality soils increased relative yield stability compared
160
to low-quality soil by 8.8-51.0% for wheat, 8.8-22.0% for maize and 2.2–12.9% for rice
161
cropping systems (Table 1). Higher yield stability in high-quality soils in wheat and
162
maize cropping systems shows that wheat and maize productivity is more dependent on
163
soil conditions. The lower impacts of soil quality on yield of paddy rice is also expected
164
as the flooding over most of the growing period leads to smaller effects of soil and
165
climatic variables on crop growth.
166
We further explored how, and to what extent, the total Yield
BMPs
variations could
167
be explained by climate variables in both high- and low-quality soils. On average, 17.2%
168
(±4.3%) of Yield
BMPs
variation was explained by climate variability in high-quality soil
169
over all systems excluding late rice in the south of China (LR-SC) and maize in the
170
southwest of China (M-SWC), but the equivalent value was 26.4% (±10.5%) in low-
171
quality soils (Table 1), suggesting that high-quality soil generally reduces the sensitivity
172
13
of crop production to climate, lowering the climate-driven share of yield variability.
173
Overall, the climate-explained R
2
in those low-quality soils was 1.7 and 1.5 times
174
higher than in good quality soils for wheat and maize, compared with 1.2 times for rice.
175
176
Interactions of climate change and soil quality on yield
177
We derive the yield response to climate change based on the trained GBRT model
178
under future climate conditions (during both 2040-2059 and 2080–2099) following
179
RCP 2.6 and RCP 8.5, assuming no adaptation. The future climate was projected by
180
using the bias-corrected global gridded climate data at 0.5°× 0.5° horizontal resolution
181
from five Earth System Models
34
. Warming is simulated over China even under RCP
182
2.6 and accompanied by increased precipitation and solar radiation (Fig. S4 and S5).
183
However, depending on region and crop, the effects of climate change on yield in China
184
were diverse, ranging from a decrease by 6.9 % to an increase by 8.6 % over cropping
185
systems, RCPs and periods (Extended Data Fig. 5 and Fig. 6). Generally, cropping
186
systems such as winter wheat in the North China Plain (W-NCP) and late rice in the
187
south of China (LR-SC), benefit from climate change according to the GBRT model.
188
Winter wheat in Yangtze River Basin (W-YZB) and northwest of China (W-NWC),
189
maize in North China Plain (M-NCP), M-SWC and SR-YZB showed yield reductions
190
even in the most positive scenarios of RCP 2.6 (Extended Data Fig. 5). Maize in
191
Northeast of China (M-NEC) showed mixed impacts on yield trends, in contrast to other
192
combinations of RCPs and periods (Extended Data Fig. 5 a,b,c), RCP 8.5 could lead to
193
a decrease in yield during 2080-2099 (Extended Data Fig. 5d). Overall, the negative
194
14
effects of climate change on yield were more prominent under drastic climate change
195
scenarios at the end of century. Climate change impacts estimated in the current study
196
qualitatively support earlier findings projected using a range of approaches
16,35-37
.
197
China is located in the mid-latitudes and spans temperate, subtropical and tropical
198
climate zones, with very diverse biophysical conditions of arable cropping (Fig. 1, Text
199
S1). Cereal crops are grown either close to temperature thresholds or at suboptimal
200
temperatures, so that a mix of effects of climate change on crop yield over cropping
201
systems and regions was anticipated
18, 21, 38
.
202
Significant interactive effects of soil quality on yield in response to climate change
203
were found in almost all cropping systems across combinations of periods and RCPs,
204
except for M-SWC and SR-YZB (Fig 2). In regions projected to have a negative yield
205
response to climate change, high-quality soils led to smaller yield loss, whereas in
206
regions with positive yield response to climate change, the climate-induced yield
207
increment was larger (Fig. 2, Extended Data Fig. 6). Interestingly, in some cases,
208
especially for wheat, high-quality soil can shift climate-induced yield decreases in low-
209
quality soils to yield increases in high-quality soils (Fig. 2, Extended Data Fig. 6 a,b,c).
210
The significant differences in relative yield change response to climate change between
211
high- and low-quality soils were found in six and five out of nine major cropping
212
systems in middle and at end of century, respectively, with the mean amount of 1.68 %
213
ranging from 0.51 % to 4.02 % across cropping systems, RCPs and periods (Extended
214
Data Fig. 6).
215
Soil hydrology, soil temperature and evapotranspiration are driven by both
216
15
climatic and soil factors. High-quality soils (e.g. with medium-textured and high SOM)
217
may better moderate the impact of rainfall variability on soil moisture and crop
218
growth
26,39-41
. Ideally, nutrient additions should be managed to continuously satisfy
219
plant nutrient demand, which requires a thorough understanding of plant requirements
220
and soil nutrient availability
42
. This can be achieved in simulations by assuming that
221
nutrients match demand by setting optimal amount and daily crop demand
21
, thus, soil
222
nutrient-related yield variability estimated by a model can be largely underestimated
24
.
223
However, this has proved difficult to achieve in practice because applications must be
224
made before the demand exists
43
, and the impulse type management approach, -
225
applying nutrients (particularly N) at key growing stage even in BMPs, fails to match
226
perfectly and dynamically with crop demand in the whole crop growth cycle. Interactive
227
effects of soil P availability and climate in crop production can also be expected,
228
because soil temperature and moisture substantially affect P diffusion, and consequently
229
modulate P bio-availability to the crop
44
. Thus, the nutrient storage and supply capacity
230
provided by soils also enables them to either buffer or reinforce impacts of climate
231
variability and change on crop growth and yield. This could be the underlying
232
mechanism for what we observed in this study, in view of the facts that soil texture,
233
SOM, and/or soil Olsen-P were important factors in classification of soil quality levels.
234
However, the mechanisms by which soil modulates impacts of climate change and
235
variability on crop productivity are highly complex due to the many processes
236
involved
41
. They differed substantially between regions and cropping systems, but to
237
fully disentangle them is beyond the scope of this study.
238
16
239
Production fluctuation derived by climate-soil interactions
240
Finally, we assessed to what extent climate-derived yield change could be
241
translated into changes in national production fluctuations, and the relative importance
242
of climate-soil interactions. Here, the interactions of soil-climate were the difference in
243
production responses between either a scenario of soil improvement or soil degradation
244
and business as usual (BAU).
245
Under RCP 2.6, both climate-driven production fluctuations as the sum of total
246
wheat, maize and rice production were small (Fig. 3 a,c). However, high climate forcing
247
scenarios led to more prominent production fluctuations, with annual climate-driven
248
production loss was, on average, 11.4 Mt under RCP 8.5 during 2080-2099, accounting
249
for 3.3% of national total production (Fig. 3 d). This was mainly due to a climate
250
change-driven production loss in wheat in NWC and in wheat and rice in YZB, and in
251
all maize cropping systems, which exceeded the climate change-induced production
252
gain in other cropping systems. Further, under the scenario of all soils being degraded
253
to a low-quality level, the climate change derived annual production loss averaged 13.0
254
Mt, comprised of 3.8 Mt from wheat, 6.4 Mt from maize and 2.8 Mt for rice (Fig. 3d),
255
accounting for 4.2% of national total wheat, 5.4% of maize and 2.0% of rice production,
256
respectively
45
. These changes in average annual production are similar to the wheat
257
production of some European countries, and higher than the maize production of most
258
African countries
45
. The size of such loss could represent a substantial threat to
259
sustaining the production growth rates necessary to keep up with demand in China, in
260
17
view of an annual growth rate in cereal production of 3.7% during 1961-2009 in China
261
and 2% globally over the same period
5
. The climate change-derived production loss
262
and risk of short-term food price shocks could be larger, when considering inter-annual
263
variability (Fig. 3). In contrast, if all soils were improved to a high-quality level by
264
2080-2099, the climate change derived annual production loss could be reduced to 9.0
265
Mt, with 2.4 Mt for wheat, 3.9 Mt for maize and 2.7 Mt for rice (Fig. 3d). Overall, the
266
interactions of climate and soil accounted for 14% of the climate-driven production loss
267
under BAU under soil degradation and 21% under soil improvement scenarios,
268
respectively.
269
The soil-climate interaction may be underestimated in the current study, due to
270
other factors not considered here, such as topsoil depth, soil compaction and erosion,
271
and soil biota which could also be important in China
46
. We did not consider elevated
272
[CO
2
] and adaptation potential of improved technology, such as improved crop
273
germplasm and adjustment of agricultural structure and planting systems, in assessing
274
both climate-derived yield change and national future production fluctuations. However,
275
these effects could occur on both high- and low-quality soils. We assume that the
276
omission of these factors does not generally challenge conclusions that high-quality
277
soils are better suited to buffer adverse conditions under climate change. However, it
278
must be acknowledged that restoring and/or improving soil quality is a challenging task,
279
especially under warmer climates and more variable precipitation patterns in future,
280
which necessitates a national and international coordinated approach
10, 26
.
281
282
18
Increasing production and delivering stable food supplies in a changing and more
283
variable climate requires integrated solutions. We demonstrate here the value of
284
controlled management practice trials on working farms for revealing crop- and region-
285
specific soil and climatic controls on crop production. Our results show that high-
286
quality soils moderate the effects of climate change and climate variability on yield and
287
improve yield stability (Fig. 4). These findings show that improving soil quality could
288
be an effective strategy for increasing the resilience of regional, national and global
289
food production under a changing climate, as a vital component of “climate-smart
290
agriculture”.
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
19
Acknowledgements
308
We thank Dr Jie Pan in Chinese Academy of Agricultural Sciences for her help in
309
projecting future climate by using the global gridded climate data of 0.5°× 0.5°
310
horizontal resolution of five Earth System Models. We thank Prof. Jianchang Yang, Prof.
311
Mingrong He, and Dr. Peng Hou for their help in
categorizing
types of crop varieties.
312
We also thank Sustainable Agriculture Innovation Network (SAIN) in organizing
313
workshop on soil quality, climate change and food security and discussing early version
314
of manuscript. This work was financially supported by the National Key Research and
315
Development Program of China (2017YFD0200108) and the National Natural Science
316
Foundation of China (31972520) for M.F, L.Q, H.C, Y.M, H.Y, Y.H, W.L. The input of
317
P.S. contributes to the Newton Fund/UKRI-funded project N-Circle (BB/N013484/1).
318
The input of B.E. was supported by the Newton Fund/UKRI-funded project CINAg
319
project (BB/N013468/1). Supplementary information associated with this article can be
320
found in the online version.
321
322
Author Contributions:
323
M.F., designed the research. M.F., L.Q., J.F., R.L., H.C., S.L., F.Z., Y.M., Y.H. R.J., H.Y.
324
W.L., collected data. M.F., L.Q., X.W., P.S., H.C., Y.W., Y.M., contributed to data
325
analysis. M.F., L.Q., wrote the manuscript with
edits from
X.W., P.S., Y.L., B.E., S.D.,
326
T.B. S.P., C.M., All authors read and approved the final manuscript.
327
328
Competing interests:
The authors declare no competing interests.
329
330
331
332
333
20
Table 1. Observed Mean yield and yield variability (CV) under best management
334
practices (Yield
BMPs
) in high- and low-quality soils and yield variability explained by
335
climate variability for major cropping systems in China.
336
Crop
types
Production
regions
Soil quality
levels
N
Yield
BMPs
(Mg/ha)
Yield
BMPs
variation
explained by
climate variability
(%)
Mean
SD CV
(%)
Winter
Wheat
North China Plain
High
327
7.1 a*
1.0 b
14.5 b
12.8
Low
328
6.5 b
1.1 a
17.1 a
19.4
Yangtze River
Basin
High
152
7.0 a
0.9 a
12.5 b
20.1
Low
158
6.4 b
0.9 a
13.7 a
31.6
Northwest China
High
106
7.1 a
1.1 b
15.9 b
23.1
Low
71
5.6 b
1.8 a
32.4 a
42.6
Maize
Northeast China
High
102
10.0 a
1.4 b
14.6 b
20.3
Low
92
9.2 b
1.5 a
16.0 a
36.7
North China Plain
High
180
8.3 a
1.1 b
13.1 b
16.3
Low
175
7.8 b
1.3 a
16.8 a
20.0
Southwest China
High
130
8.1 a
1.3 b
16.6 b
15.6
Low
127
7.4 b
1.4 a
19.1 a
14.9
Single
rice
Yangtze River
Basin
High
241
8.7 a
1.1 a
13.1 b
16.7
Low
244
8.4 b
1.1 a
13.4 a
18.2
Early
rice
South China
High
188
7.1 a
1.0 b
14.2 b
11.2
Low
184
6.7 b
1.1 a
16.3 a
16.1
Late rice
South China
High
202
7.5 a
0.9 a
12.3 b
17.7
Low
253
6.6 b
0.9 a
13.1 a
7.4
High- and low-quality soils were grouped according to the two most important and
337
sensitive soil variables in explaining yield variations (See Method and Table S3). N
338
represents the number of paired on-farm trails with different soil quality but the same
339
management practices and climate conditions.
Yield
BMPs
(Mg/ha) are shown as mean,
340
SD (standard deviation), and CV (%, coefficient of variation calculated by dividing
341
mean yield by standard deviation).Climate impacts were assessed by explained
342
variability (R
2
) in climate-yield relationship assessed by Gradient Boosted Regression
343
Tree model for high and low soil quality groups. *Different lowercase showed
344
significant difference in mean Yield
BMPs
, SD and CV between high- and low-quality
345
soils for each cropping systems at p=0.05, respectively.
346
347
21
Figure Legends
348
Fig. 1. Geographical distribution of on-farm trials.
a-c, distributions on-farm trials
349
for winter wheat, maize, and rice, respectively. Symbols of purple dot represent on-
350
farm trials. Numbers in brackets indicate the number of on-farm trials for each region
351
of each crop. Map sections of different colours indicate the major wheat, maize, and
352
rice production agroecological regions in China.
Harvested area fractions represent the
353
proportion of harvested area of Gridcell (10 km
2
) for each crop (Data source:
354
http://www.earthstat.org/). The shade of colour section indicates the size of the
355
harvested area.
356
357
Fig. 2. Projected yield change in high- and low- quality soils in future climate
358
change.
Projections were conducted under RCP2.6 and RCP8.5 pathways up to 2040-
359
2059 and 2080-2099, and based on Gradient Boosted Regression Tree model trained on
360
sub-data set composed of on-farm trials with paired trials of high- and low-quality soil
361
in major cropping systems in China. Solid lines and diamonds in this figure indicate
362
median and mean yields, respectively; the boundary of the box indicates the 25th and
363
75th percentile; whisker caps denote the 90th and 10th percentiles. Paired data refer to
364
585 and 557 for wheat, 412 and 394 for maize, and 631 high- and 681 low-quality soils
365
for rice, respectively. Asterisks represent significant difference in yield change between
366
high- and low-soil quality at p = 0.10. W-NCP, winter wheat in North China Plain; W-
367
YZB, winter wheat in Yangtze River Basin; W-NWC, winter wheat in Northwest China;
368
M-NEC, rainfed maize in Northeast China; M-NCP, maize in North China Plain; M-
369
22
SWC, rainfed maize in Southwest China; SR-YZB, single rice in Yangtze River Basin;
370
ER-SC, early rice in South China;
LR-SC, later rice in South China.
371
372
373
Fig. 3. Climate-change driven change in cereal production.
a-d, Climate-change
374
driven change in cereal production of three soil quality scenarios
under RCP2.6 (a) and
375
RCP8.5 (b) pathways by 2040-2059, and RCP 2.6 (c) and RCP8.5 (d) by 2080-2099 for
376
major cropping systems in China. The bars (standard deviation, SD) show the average
377
plus inter-annual variability in total cereal production caused by climate change for the
378
three conditions: soil quality maintained at current quality level as business as usual
379
(BAU), soil quality uniformly improved to a high-quality level (SQ improvement), soil
380
quality uniformly degraded to a low-quality level (SQ degradation) for all farmlands of
381
major cropping systems. Green, dark green and light green, columns represent BAU,
382
SQ improvement, and SQ degradation scenarios, respectively. Asterisks refer to
383
cropping systems with significant difference in yield response to future climate changes
384
between high- and low-quality soil at p = 0.1. W-NCP, winter wheat in North China
385
Plain; W-YZB, winter wheat in Yangtze River Basin; W-NWC, winter wheat in
386
Northwest China; M-NEC, rainfed maize in Northeast China; M-NCP, maize in North
387
China Plain; M-SWC, rainfed maize in southwest China; SR-YZB, single rice in
388
Yangtze River Basin; ER-SC, early rice in South China;
LR-SC, later rice in South
389
China.
390
391
23
Fig. 4. Schematic representation of the pattern of soil quality (SQ) moderating
392
the yield resilience to climate variability and change.
High-quality soil leads to
393
higher attainable/mean yield and a less variable response to climate impacts than a
394
low-quality soil. Further, where climate change positively impacts crop yields, then a
395
good quality soil would enhance that positive effect. In contrast, if climate change
396
negatively affects yield, then high-quality soil would at least partially offset those
397
negative impacts.
398
399
References
400
1. Alexandratos, N. & Bruinsma, J.
World Agriculture Towards 2030/2050. The 2012
401
Revision
(FAO, 2012).
402
2. Tilman, D. Balzer, C., Hill, J. & Befort, B. L. Global food demand and the
403
sustainable intensification of agriculture.
Proc. Natl. Acad. Sci.
108
, 20260-20264
404
(2011).
405
3. Mueller, N. D.
et al.
Closing yield gaps through nutrient and water management.
406
Nature
490
, 254 (2012).
407
4. Chen, X.
et al.
Producing more grain with lower environmental costs.
Nature
514
,
408
486-489 (2014)
409
5-. Fan, M. S.
et al.
Improving crop productivity and resource use efficiency to
410
ensure food security and environmental quality in China.
J. Exp. Bot.
63
, 13-24
411
(2012).
412
6. Godfray, H. C. J.
et al.
Food security: the challenge of feeding 9 billion people.
413
Science
327
, 812-818 (2010).
414
7. Foley, J. A. et al. Solutions for a cultivated planet.
Nature
478
, 337 (2011).
415
8. Porter, J. R.
et al
.
Food security and food production systems
(Cambridge Univ
416
24
Press, Cambridge, 2014).
417
9. Ray, D.K. & Foley, J.A. Increasing global crop harvest frequency: recent trends
418
and future directions.
Environ. Res. Lett.
8
044041 (2013).
419
10. Lal, R. Restoring soil quality to mitigate soil degradation.
Sustainability
7
, 5875-
420
5895 (2015).
421
11.Wall, D. & Six, J.
Science
347
, 695 (2015).
422
12. Ray, D. K.
et al
. Climate variation explains a third of global crop yield variability.
423
Nat. Commun.
6
, 5989 (2015).
424
13. Battisti, D. S. & Naylor, R. L. Historical warnings of future food insecurity with
425
unprecedented seasonal heat.
Science
323
, 240-244 (2009).
426
14. Nelson, G. C.
et al
.
Climate change: Impact on agriculture and costs of
427
adaptation
(Intl Food Policy Res Inst, 2009).
428
15. Challinor, A. J., Koehler, A. K., Ramirez-Villegas, J., Whitfield, S. & Das, B.
429
Current warming will reduce yields unless maize breeding and seed systems
430
adapt immediately.
Nat. clim. change
6
, 954 (2016).
431
16. Zhao, C.
et al
. Temperature increase reduces global yields of major crops in four
432
independent estimates.
Proc. Natl. Acad. Sci.
114
, 9326-9331 (2017).
433
17. Schlenker, W., Hanemann, M. & Fisher, A. Will US agriculture really benefit
434
from global warming? Accounting for irrigation in the hedonic approach.
Am.
435
Econ. Rev.
95
, 395-406 (2005).
436
18. Piao, S. L.
et al
. The impacts of climate change on water resources and agriculture
437
in China.
Nature
467,
43
(2010).
438
19. Ray D. K
. et al.
Climate change has likely already affected global food production.
439
PloS one
14
: e0217148 (2019).
440
25
20. Ramankutty, N.
et al
. The global distribution of cultivable lands: current patterns
441
and sensitivity to possible climate change.
Global Ecol. Biogeogr.
11
, 377-392
442
(2002).
443
21. Rosenzweig, C.
et al
. Assessing agricultural risks of climate change in the 21st
444
century in a global gridded crop model intercomparison.
Proc. Natl. Acad. Sci.
445
111
, 3268-3273 (2014).
446
22. Lobell, D. B. & Burke, M. B. On the use of statistical models to predict crop yield
447
responses to climate change.
Agr. Forest. Meteorol.
150
,1443-1452 (2010).
448
23. Auffhammer, M. & Schlenker, W. Empirical studies on agricultural impacts and
449
adaptation.
Energ. Econ.
46
, 555-561 (2014).
450
24. Folberth, C.
et al.
Uncertainty in soil data can outweigh climate impact signals in
451
global crop yield simulations.
Nat. Commun.
7
, 11872 (2016).
452
25. Asseng S.
et al.
Uncertainty in simulating wheat yields under climate change.
Nat.
453
Clim. Change
3
, 827-832 (2013).
454
26. Basso, B.
et al.
Soil Organic Carbon and Nitrogen Feedbacks on Crop Yields
455
under Climate Change.
Agr. Environ. Letters
3
, (2018).
456
27. M
ϋ
ller, C. et al. Implication of climate mitigation for future agricultural
457
production.
Environ. Res. Lett.
10,
125004 (2015).
458
28. IPCC, Climate Change 2022:
Impacts, Adaptation, and Vulnerability.
459
Contribution of Working Group II to the Sixth Assessment Report of the
460
Intergovernmental Panel on Climate Change
(eds. Pörtner, et al.) (Cambridge
461
University Press. In Press, 2022).
462
29. Zhang, W.
et al
. Closing yield gaps in China by empowering smallholder farmers.
463
Nature
537
, 671–674 (2016).
464
26
30. Cui, Z. L.
et al.
Pursuing sustainable productivity with millions of smallholder
465
farmers.
Nature
555
, 363-368 (2018).
466
31. Knapp, S. & van der Heijden, M. G. A. A global meta-analysis of yield stability in
467
organic and conservation agriculture.
Nat. Commun.
9
, 3632 (2018)
468
32. Müller, C.
et al.
Global Gridded Crop Model evaluation: benchmarking, skills,
469
deficiencies and implications.
Geosci. Model Dev. Discuss
, 1-39 (2016).
470
33. Jamieson, P.D., Porter, J.R. & Wilson, D.R. A test of the computer simulation model
471
ARC-WHEAT on wheat crops grown in New Zealand.
Field Crops Res.
27
,337–
472
350 (1991).
473
34. Warszawski, L.
et al
. The inter-sectoral impact model intercomparison project
474
(ISI–MIP): project framework.
Proc. Natl. Acad. Sci.
111
, 3228-3232 (2014).
475
35. Xiong, W.
et al
. National level study: the impacts of climate change on cereal
476
production in China. In:
The Impacts of Climate Change on Chinese Agricultur
e-
477
Phase II Final Report
(AEA Group, 2008).
478
36. Liu, B.
et al.
Similar estimates of temperature impacts on global wheat yield by
479
three independent methods.
Nat. Clim. Change
6
, 1130-1136 (2016).
480
37. Tao, F.
et al
. Global warming, rice production, and water use in China:
481
Developing a probabilistic assessment.
Agr. Forest Meteorol.
148
, 94-110
482
(2008).
483
38. Xiong, W.
et al
. Different uncertainty distribution between high and low latitudes
484
in modelling warming impacts on wheat.
Nat. Food
(2019) doi:10.1038/s43016-
485
019-0004-2
486
39. Fernandez-Illescas, C. P. Porporato, A. Laio, F. & Rodriguez-Iturbe, I. The
487
ecohydrological role of soil texture in a water-limited ecosystem.
Water Resour.
488
Res.
37, 2863-2872 (2001).
489
27
40. Wang, E. L.
et al
. Capacity of soils to buffer impact of climate variability and
490
value of seasonal forecasts.
Agr. Forest Meteorol.
149
, 38-50 (2009).
491
41. Vereecken, H.
et al
. Modeling soil processes: Review, key challenges, and new
492
perspectives.
Vadose Zone J
15
(2016).
493
42. Myers, R. J. K.
et al
. The synchronization of nutrient mineralization and plant
494
nutrient demand. In:
The Biological Management of Tropical Soil Fertility
(eds.
495
Woomer, P.I. & Swift, M.J.) (Wiley, Chichester, New York, 1994).
496
43. Smith, P. & Gregory, P. J. Climate change and sustainable food production.
P.
497
Nutr. Soc.
72
, 21-28 (2013).
498
44. Khasawneh, F. E., Sample, E. C., & Kamprath, E. J. (1980).
The role of
499
phosphorus in agriculture
(American Society of Agronomy, Madison, 1980).
500
45. FAO. FAOSTAT. Statistics Division of the Food and Agriculture Organization of
501
the United Nations http://www.fao.org/faostat/en/#home (FAO, 2006).
502
46. Fan, M. S.
et al.
Plant-based assessment of inherent soil productivity and
503
contributions to China’s cereal crop yield increase since 1980.
Plos One
8
,
504
e74617 (2013)
505
28
Methods
506
The agroecological zones and major cereal cropping systems
507
Wheat (
Triticum aestivum L.
), maize (
Zea mays L.
) and rice (
Oryza sativa L.
) are the
508
principal staple foods in China, cultivated across China from cold to subtropical and
509
from arid to semi-arid and humid regions
47
. Nine major cropping systems, accounting
510
for more than 90% of the total production of rice, maize and wheat, were included in
511
the current study. They are defined according to their agroecological and geographical
512
location: 1) winter wheat in North China Plain (W-NCP), 2) winter wheat in Yangtze
513
River Basin (W-YZB), 3) winter wheat in Northwest China (W-NWC), 4) rainfed maize
514
in Northeast China (M-NEC), 5) maize in North China Plain (M-NCP), 6) rainfed maize
515
in Southwest China (M-SWC), 7) single rice in Yangtze River Basin (SR-YZB), 8) early
516
rice in South China (ER-SC), 9) late rice in South China (LR-SC). An overview of the
517
major cropping systems and the geographical distribution of on-farm trials is shown in
518
Fig. 1 and Text S1. A publicly released base map of China was obtained from the
519
Resource and Environmental Data Cloud Platform (http://www.resdc.cn).
All map-
520
related operations were performed using ArcGIS 10.8.1 software (www.esri.com/en-
521
us/arcgis).
522
523
524
On-farm trials and data set
525
A total of 12115 site-year on-farm trials (n=3883 for wheat, 3694 for maize and
526
4538 for rice) were obtained from the National Soil Test and Fertilizer
527
29
Recommendation projects (2005-2013), with sites spread across all the involved
528
agroecological zones (Fig. 1). On-farm trials were conducted to study optimized
529
fertilizer recommendation, in which the three nutrients of N, P and potassium (K) with
530
four rates, and a total of fourteen treatments were included
48
. These experiments were
531
designed and managed by local agricultural experts and/or trained extension officers,
532
and were implemented in on-farm fields. In the present study, only treatments with
533
optimal NPK rates were used, with an exception for LR-SC, for which yield in control
534
plots were used as one of the indicators in classification of soil quality. These optimal
535
NPK treatments were developed specifically to maximise both yield and nutrient use
536
efficiency for a given location based on integrated nutrient management strategies
49
,
537
also using locally available practices based on best science and understanding in
538
cultivar choice, sowing date and density, supplementary irrigation (in irrigated cropping
539
systems), weed, insect and disease control (hereafter referred to as BMPs treatments).
540
Based on these on-farm trials, paired agronomic, climate and soil data sets were
541
established (Table S1). Agronomic data collected according to a standard protocol
48
in
542
the current study included crop varieties, sowing
and harvest time, NPK rate, and grain
543
yield of BMPs treatments in each of the on-farm trials. Using yields under locally
544
defined BMPs allowed us to focus on the relative importance of soil quality and climate
545
variability in determining yield and yield variability, and avoiding the impacts of any
546
sub-optimal management impacting yield and its variability. Wheat varieties were
547
classified into small-, medium- and large-spike variety types; Maize and rice varieties
548
were classified into early-, medium- and late-maturity variety types. Soil data consists
549
30
of soil type, soil texture and SOM, soil Olsen-P and Available-K concentration and pH,
550
which are established indicators of soil quality
50
. Here, soil quality is defined as the
551
capacity of the soil to provide nutrients and water, and to support crop productivity
50
.
552
Soil type was represented as soil genetic classification in China
51
and soil texture was
553
in accordance with USDA texture class, both of which were used as natural genetic
554
attributes. SOM, soil Olsen-P, soil Available-K concentration and pH were measured
555
using standard methods
52
, are dynamic over time and represented as manageable soil
556
indicators. Weather data recorded during the crop growing period for each on-farm trial
557
comprised daily mean temperature (Tave), maximum (Tmax) and minimum
558
temperature (Tmin), precipitation (PRE) and sunshine duration (SSD) from the county
559
or municipality where the trial was conducted, and were obtained from the Chinese
560
Meteorological Administration (Table S1). Sunshine duration was converted into daily
561
solar radiation (RAD) using the Weather Aid module in the Hybrid-Maize model
562
(http://www.hybridmaize.unl.edu/). Growing degree days (GDD) was calculated as an
563
annual sum of daily mean temperatures based on
sowing and harvest time of BMPs
564
over a base temperature, 0 °C for wheat and 10 °C for maize and rice according to
565
Ramankutty, et al.
20
, representing the “growing season length” of crops and which is
566
sufficient to define the cold boundaries of agricultural land
53
. Generally, the present
567
study was built upon the most comprehensive dataset across a wide range of
568
agroecological zones in China. But, the effect of the other omitted variables could have
569
been important in some specific locations.
570
571
31
Explaining yield variation by GBRT
572
GBRT analysis was performed to assess the relative importance of explanatory
573
variables on Yield
BMPs
variation. The GBRT algorithm is an efficient machine learning
574
method, which combines regression trees and a boosting technique to optimize the
575
predictive performance of multiple single models
54
. The regression tree is a decision
576
tree model that can be used for regression. The specific formula of the decision
577
regression tree shown in Eq.1.
f
t
(x)
is the prediction function for the input variable.
I(x)
578
is an indicator function,
I(x)
=1 if
x
∈
R
m
, and
I(x)
=0 otherwise.
R
m
indicates partition
579
units of the input space. A regression tree corresponds to a partition of the input space
580
(i.e. feature space) and an output value on these partitioned units. In contrast to a
581
classification tree, the regression tree uses a heuristic method to divide the input space.
582
In the training process, the model traverses all the input variables, finds the optimal
583
segmentation variable
j
and the optimal segmentation point
s
to form a partition. In this
584
study,
j
indicate the elements of input explanatory variables, including 13 to 15 climatic,
585
soil and management variables (Table S1). Suppose that an input space is divided into
586
M units to form a partition of input space {
R
1
,R
2
,…,R
M
}. Each input variable of the
587
model falls on one unit
R
m
. There is a fixed output value
c
m
on each unit represents the
588
optimal output value on unit
R
m
, which is obtained by calculating the average of the
589
output values corresponding to all input instances on
R
m
.
y
i
represents the observed
590
Yield
BMPs
for
i
th on-farm trial.
591
min
,
min
∑
−
+ min
∑
−
∈
2
,
∈
1
,
(Eq.1)
592
,
=
≤
,
,
=
>
593
32
=
1
∈
,
,
∈
,
=
1,2
594
=
,
∈
595
GBRT model obtained by iterating multiple regression trees using stochastic
596
gradient boosting method. Stochastic gradient boosting is a forward stage-wise process,
597
in which a subset of the data is randomly selected to iteratively fit new tree models to
598
minimize the loss function
55
(Eq.2.).
f
0
(x)
is the initial regression tree with only one
599
terminal node, estimating a constant value that minimizes the loss function.
L()
is a loss
600
function fitted by least-squares to calculate the residual value between
c
(predicted yield)
601
and
y
i
(observed yield).
refers to residual estimate by negative gradient of the loss
602
function.
f
t
(x)
refer to the
t
th regression tree function for the prediction of dependent
603
variable y, which equal to the sum of the predicted residual value and the predicted
604
value by (
t
-1)th regression tree. Final model
f
T
(x)
is obtained by integrating the results
605
of total
T
regression trees. Boosting generates a final model by shrinking the
606
contribution of each tree and averaging across the final selected set, which is more
607
robust than a single regression tree model and enables fitting of curvilinear
608
functions
54,56
.
609
(1)
=
arg
∑
,
(Eq.2)
610
(2)
For
t
= 1 to
T
do:
611
=
−
,
,
=
1,2
⋯
612
=
arg
∑
,
,
613
R
,
=
614
33
For j=1 to J do:
615
,
=
arg
∑
,
+
616
=
+
∑
,
,
(
∈
R
,
)
617
(3)
Output:
( ) = ( )
+
∑
,
( )
(
∈
R
,
)
618
To run GBRT analysis, four main parameters are needed to define a GBRT
619
algorithm: learning rate (LR), the contribution of each tree to the final fitted model;
620
interaction depth (ID), tree depth and number of iterations; number of trees (NT),
621
integer specifying the total number of trees to fit; bag fraction (BF), the fraction of the
622
training set observations randomly selected to propose the next tree in the expansion.
623
In general, it is suggested that BF is set at around 0.5
55
. Then we set a series of
624
combinations of parameter values (LR and ID) to test GBRT models, thereafter
625
choosing the optimal parameter combination which provided the minimum predictive
626
deviation. These combinations can generate optimal NT using a 10-fold cross-
627
validation method. The relative importance of variables can be estimated based on the
628
number of times a variable is selected for modelling, weighted by the square
629
improvement to each split, and averaged across all trees
57
.
630
We selected climatic, soil and management variables as explanatory variables, and
631
Yield
BMPs
as the explained variable to include in the final model. Therefore, the final
632
regression model for each crop was:
633
y
i
= F(f
T
(X
i
),Q
i
) +
ε
i
(Eq. 3)
634
(1)
y
i
represents Yield
BMPs
for cropping systems
i
;
635
(2)
f
T
(X
i
) is the GBRT function, X
i
=
[
C
i
, S
i
, M
i
],
X
i
represents input explanatory
636
34
variables including climatic variables
C
i
(Tmax, Tmin, GDD, PRE and RAD),
637
soil variables
S
i
(Soil type, Soil texture, SOM, Olsen-P, Avail-K and pH) and
638
management variables
M
i
(application rates of N, P and K);
639
(3)
Q
i
=
[
LR
i
, ID
i
, NT
i
, BF
i
],
Q
i
represents the GBRT model parameters including
640
learning rate (
LR
i
), interaction depth (
ID
i
), number of trees (
NT
i
) and bag
641
fraction (
BF
i
);
642
(4)
ε
i
represents the error.
643
For each dataset of cropping systems, 10% of the total on-farm trials were
644
randomly excluded to act as independent test datasets. The remaining 90% of trials were
645
used to build GBRT models. To evaluate the robustness of the modelling, we randomly
646
sampled test datasets and run models for 50 times, and evaluated summary statistics of
647
modelling performances (Table S2). GBRT models are developed using the “caret” and
648
“gbm” packages of R software
58
, and R scripts are provided by Kuhn & Johnson
59
.
649
The degree of agreement between simulated and observed values was assessed by
650
mean error (E), root mean square error (RMSE), normalized
RMSE (nRMSE), which
651
are indices commonly used in both model calibration and validation processes
60
. E is
652
the bias between predicted value and observed value, an index to determine if the model
653
under-(negative) or over-estimates (positive) the observed data
61
. A paired t test was
654
also used to detect whether the E was significantly different from zero
62
. RMSE takes
655
on the same unit of deviation
61
, and nRMSE, as a metric of percentage deviation from
656
the average yield, gives a measure of the relative difference of simulated versus
657
observed data. The simulation is considered excellent, good, fair and poor, with nRMSE
658
35
< 10%, 10% < nRMSE < 20%, 20%< nRMSE < 30% and nRMSE > 30%, respectively
33
.
659
E, RMSE and nRMSE were calculated according to Eq 4-6:
660
E
=
∑
(
k
−
k
)
(Eq. 4)
661
RMSE
=
∑
(
k
−
k
)
(Eq. 5)
662
nRMSE
=
× 100
(Eq. 6)
663
Where,
P
k
and
O
k
are the predicted and observed yield values at site k, respectively;
664
is the mean of observed yield; n is the number of samples.
665
Summary statistics of modelling performances for each of the cropping systems
666
are shown in Table S2. The mean E values were relatively small. None of the E values
667
were significantly different from zero. Model evaluation produced average RMSE
668
value ranges of 818-1035 kg ha
−
1
for wheat, 1155-1494 kg ha
−
1
for maize, and 895-996
669
kg ha
-1
for rice, which was comparable with those of the latest simulation studies based
670
on multiple site-years dataset
63-66
. Average nRMSE ranged from 10.5 – 15.6 % across
671
three crops and regions, indicating good performance of GBRT model in modelling
672
yield. However, it should be noted that the empirical models are agnostic on the
673
underlying mechanisms. GBRT approach is not exception for this.
674
675
Yield response to climate variability in different quality soils
676
To assess yield resilience to both current climate variability and future change in
677
different quality soils, we developed a sub-set of data composed of locally paired on-
678
farm trials, for high- and low-quality soils in the same climatic conditions and with the
679
same BMPs.
680
36
All 6 soil indicators explained integrated yield variations (Extended Data Fig. 2).
681
Thus, we identified the two most important and sensitive soil variables as indicators in
682
grouping high- and low-quality soils in each cropping system. Both soil variables were
683
ranked in the top two of soil factors in explaining yield variation by GBRT (Extended
684
Data Fig. 2), and had strong partial dependence relationships with crop yield (Fig. S1-
685
S3). Then, we divided the entire on-farm trial database based on the two identified soil
686
indicators into “both high”, “both low”, and “low-high”, and “high-low” sub-databases.
687
Without a clear
threshold value between yield and manageable soil indicators
，
high and
688
low value groups were identified according to their mean; when soil type and soil
689
texture were selected as indicators, we divided them into two groups, with half of them
690
as “high” and the remaining half as the “low” group (Table S3). The “both high” and
691
“both low” groups were identified as “high” and “low” quality soil sub-databases,
692
which also were paired with the same management practices and sharing the same
693
climate observed station (Extended Data Fig. 3 and Fig. 4). The increase trends in mean
694
Yield
BMPs
along soil quality gradients (Fig.S6) suggested that defining soil quality based
695
on two major soil indicators was valid. Further, we grouped low- and high-quality soils
696
based on integrated soil quality index (SQI) and compared yield between two quality
697
levels (Text S2). Difference in yield and yield variation between low- and high-quality
698
soil using the SQI approach was similar to the trend based on sensitive soil variables
699
approach (Table 1 and Table S6). An overall soil quality index is often desired but is
700
actually not very meaningful
50
. However, sensitive soil variables approach allows us to
701
identify feasible soil management practices in diverse crop systems and regions and to
702
37
contribute to improved soil quality. A final sub-set of data comprised locally paired
703
n=585 high- and 557 low-quality soils for wheat, 412 high- and 394 low-quality soil
704
for maize, and 631 high- and 681 low-quality soil for rice cropping systems
(Extended
705
Data Fig. 3).
706
To assess the yield response to climate variability, we compared mean Yield
BMPs
,
707
SD, and CV between high- and low-quality soils for each cropping system. The SD is
708
termed the absolute yield stability
31
. The CV is termed relative yield stability and
709
captures both changes in the SD and mean of yield across site-years
31,67
. CV of
710
Yield
BMPs
is calculated using the following equation:
711
ij
(
%
) =
(
im
)
(
im
)
× 100
(Eq. 7)
712
Where, SD (Yield
im
)) and Mean (Yield
im
) are Yield variation and mean yield under
713
BMPs of high- and low-quality soil for each cropping system; i and m represent
714
cropping systems and soil quality groups, respectively.
715
We performed a bootstrapping exercise (1000 bootstrap samples) combined with
716
T-test to assess the statistical significance of differences at P=0.05 in mean yield, SD
717
and CV between high- and low-quality soils for each cropping system.
718
Furthermore, we used a variation partitioning method to differentiate the relative
719
contribution of climatic variables in explaining yield variation for two soil quality
720
datasets. For each cropping system, two GBRT models were respectively performed
721
with high- and low-quality soil datasets, using Yield-BMPs as the dependent predictor
722
and climatic variables as independent predictors. The relative contribution of climate
723
variability on yield variability was determined by coefficient of determination (R
2
),
724
38
which were estimated through a 10-fold cross validation procedure conducted using the
725
caret:train function
68
.
726
727
Projecting yield of different quality soil in climate change
728
For the future climate scenarios, four Representative Concentration Pathways
729
(RCPs), extending to the year 2100 with radiative forcing values from 2.6 to 8.5 Wm
-2
,
730
were proposed to represent different greenhouse gas emission scenarios
69,70
. In this
731
study, we considered RCP2.6 and RCP8.5. The former represented a very low forcing
732
level and a stringent pathway, peaking in radiative forcing at circa 3 W m
−
2
around the
733
year 2050 and then declining to 2.6 W m
−
2
by 2100; while the latter is a high-end forcing
734
pathway, a continuously increasing radiative forcing pathway to 8.5 W m
−
2
by 2100.
735
We do not explicitly consider RCP 4.5 and RCP 6 assuming results for these pathways
736
would lie between RCP 2.6 and RCP 8.5.
737
The future climate conditions under RCP 2.6 and RCP 8.5 were projected by using
738
the global gridded climate data of 0.5°× 0.5° horizontal resolution of five Earth System
739
Models (ESMs; GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC-ESM-
740
CHEM, NorESM1-M), which were taken from the ISI-MIP Fast Track input-data
741
catalogue
34
. The original data were retrieved from the CMIP5 archive and interpolated
742
and bias-corrected with respect to historical observations by Hempel et al.
71
to remove
743
systematic biases. The CMIP6 models exhibit an improvement in simulation of climate
744
extremes but the model spreads are still comparable between CMIP5 and CMIP6
72
, thus
745
we used climate data of CMIP5
.
746
39
The projected changes in mean Tmax, Tmin, accumulated PRE, and accumulated
747
SSD during the growing season of the major crop growing-areas in both 2040-2059 and
748
2080-2099 in comparison with 1986–2005 under two RCPs (2.6 and 8.5) are shown in
749
Fig. S4 and S5. RAD and GDD were calculated as described in a previous section. In
750
summary, warming occurs in all seasons even under RCP2.6. Projected Tmax and Tmin
751
on average increased by 1.6
℃
and 1.7
℃
over major production regions during 2040-
752
2059, then stabilized at a similar level up to 2100 for RCP 2.6 (Fig. S4 a,b); while both
753
Tmax and Tmin increased on average by 2.7
℃
and 2.5
℃
during 2040-2059, and by
754
5.6 and 5.1
℃
during 2080-2099 for RCP 8.5, respectively (Fig. S5 a,b). Both RCPs
755
show that increases in temperature will be accompanied by increased PRE and SSD
756
during both 2040-2059 and 2080-2099 (Fig.S4 c,d; Fig.S5 c,d), with an exception under
757
RCP2.6 during the 2040-2059 period (Fig. S4 d), when accumulated SSD could
758
decrease for the wheat growing area in NWC. However, PRE and SSD projection show
759
high spatial variability and greater differences between ESMs than temperature.
760
Yield change was estimated by comparing the yield differences predicted by
761
GBRT models, between future periods (2040-2059 and 2080-2099) and a baseline
762
period (1985-2005) for each cropping system. In running GBRT models, climatic
763
variables were derived from the above climate change scenarios, while soil and
764
management variables used were based either on the whole dataset or on high- and low-
765
soil quality groups. Management and soil variables were paired spatially with projected
766
climate data at 0.5°× 0.5° horizontal resolution. An unpaired t-test was conducted for
767
statistical comparison of yield changes to assess the significance of differences between
768
40
high and low soil quality groups. Factors tested were considered to be statistically
769
significant at p = 0.10. We also tested the sensitivity of yield change to soil quality by
770
comparing projected yield changes up to 2080-2099 by adjusting data distributions
771
based on the mean soil quality indicator threshold values by -20, -10, +10 and +20%,
772
finding no prominent difference between them (Fig. S7).
773
To assess further production response derived from interaction of soil and climate
774
for RCP2.6 and RCP8.5 during 2040-2059 and 2080-2099, we established three soil
775
quality scenarios: (1) where soil quality is maintained at the current level as business
776
as usual (BAU), (2) where soil was improved throughout to the high-quality level, and
777
(3) where soil was degraded throughout to the low-quality level for all farmlands of
778
major cropping systems. The definition of high-quality and low-quality soil (Table S1),
779
and projected yield changes per unit area under future climate scenarios (Fig. 2,
780
Extended Data Fig. 5 and 6) were described and shown in the above sections. The total
781
harvested area of each farming system (10
6
ha) was obtained from the China Agriculture
782
Yearbook
73
, which is assumed to be maintained the same as at present in the future;
783
thus, the total production response is the product of yield change and harvested area of
784
each of the cropping systems for each of the soil quality scenarios. The interactions of
785
climate-soil were the difference in production responses between either a scenario of
786
soil improvement or soil degradation and BAU.
787
788
Data availability
789
Data that support these findings are available via GitHub
790
41
(https://github.com/FMS321/soilquality_climatechange_paper.git).
791
Code availability
792
Codes for processing the data are available via GitHub
793
(https://github.com/FMS321/soilquality_climatechange_paper.git).
794
795
47. Liu, X. & Chen, F.
Farming system in China
(China Agr. Press, Beijing, 2005).
796
48. Chen, X. P. Using 3414 experiment to establish the index system of soil testing
797
and fertilizer recommendation. In:
Fertilization Technology Highlights
(eds.
798
Zhang, F. S) (Chinese Agricultural University Press, Beijing, China, 2006).
799
49. Zhang, F.
et al.
Integrated nutrient management for food security and
800
environmental quality in China.
Adv. Agron.
116
, 1-40 (2012).
801
50. Bünemann, E.K.
et al.
Soil quality – A critical review.
Soil Biol. Biochem.
120
,
802
105-125 (2018).
803
51. National Soil Survey Office.
Chinese Soil
(China Agriculture Press, Beijing,
804
1998).
805
52. Jiang, R. F. & Cui, J. Y. Soil test. In:
Fertilization Technology Highlights
(eds
806
Zhang, F. S) (China Agricultural University Press, Beijing, 2006).
807
53. Cramer, W.P. & Solomon, A.M. Climatic classification and future global
808
redistribution of agricultural land.
Clim. Res.
3
, 97-110 (1993).
809
54. Elith, J., Leathwick, J. R. & Hastie, T. A working guide to boosted regression
810
trees.
J. Anim. Ecol.
77
, 802-813 (2008).
811
55. Friedman, J. H. Stochastic gradient boosting.
Comput. Stat. Data An.
38
, 367-378
812
(2002).
813
42
56. Buston, P. M. & Elith, J. Determinants of reproductive success in dominant pairs
814
of clownfish: a boosted regression tree analysis.
J. Anim. Ecol.
80
, 528-538
815
(2011).
816
57. Friedman, J. H. & Meulman, J. J. Multiple additive regression trees with
817
application in epidemiology.
Stat. Med.
22
, 1365-1381 (2003).
818
58. R Development Core Team. R: A Language and Environment for Statistical
819
Computing. R Foundation for Statistical Computing: Vienna, Austria. ISBN 3-
820
900051-07-0, http://www.R-project.org (2018).
821
59. Kuhn, M. & Johnson, K.
Applied Predictive Modeling
(Springer, Now York,
822
2013).
823
60.
Yang, J.M., Yang, J.Y. & Liu, S., Hoogenboom, G. An evaluation of the statistical
824
methods for testing the performance of crop models with observed data.
Agric.
825
Syst
.
127
, 81-89 (2014).
826
61. Loague, K. & Green, R. E. Statistical and graphical methods for evaluating solute
827
transport models: overview and application.
J. Contamin. Hydro.
7
, 51–73 (1991).
828
62. Akinremi, O. O.
et al.
Evaluation of LEACHMN under Dryland conditions. I.
829
Simulation of water and solute transport.
Can. J. Soil Sci.
85
, 223–232 (2005).
830
63. Palosuo, T.
et al.
Simulation of winter wheat yield and its variability in different
831
climates of Europe: A comparison of eight crop growth models.
Europ. J.
832
Agronomy
35
,103–114 (2011).
833
64. Deng, N.
et al.
Closing yield gaps for rice self-sufficiency in China.
Nat. Commun.
834
10
,1725 (2019).
835
65. Correndo, A. A.
et al.
Assessing the uncertainty of maize yield without nitrogen
836
43
fertilization.
Field Crops Res.
260
, 107985 (2021).
837
66. Rattalino Edreira, J.I. et al., Spatial frameworks for robust estimation of yield gaps.
838
Nature Food.
2
, 773–779 (2021).
839
67. Tilman, D., Reich, P. B. & Knops, J. M. H. Biodiversity and ecosystem stability
840
in a decade-long grassland experiment.
Nature
441
, 629 (2006).
841
68. Kuhn, M. Building predictive models in R using the caret package.
J. Stat.
842
Softw.
28
, 1-26 (2008).69. IPCC. Climate Change 2014:
Climate Change:
843
Synthesis Report
(eds Working Group I, II and III) (Cambridge University Press,
844
Cambridge, 2014).
845
70. van Vuuren, D. P.
et al
. The representative concentration pathways: an overview.
846
Clim. Change
109
, 5-31 (2011).
847
71. Hempel, S., Frieler, K., Warszawski, L., Schewe, J. & Piontek, F. A trend-
848
preserving bias correction–the ISI-MIP approach.
Earth Syst. Dynam.
4
, 219-236
849
(2013).
850
72. Chen, H., Sun, J., Lin, W., Xu, H., Comparison of CMIP6 and CMIP5 models in
851
simulating climate extremes.
Sci. Bull.
65
, 1415–1418 (2020).
852
73. Editorial Board of China Agriculture Yearbook.
China Agriculture Yearbook
853
(China Agriculture Press, Beijing, 2005).
854
80°E
130°E
50°N
20°N
80°E
130°E
North China Plain
Yangtze River Basin
Northwest China
Winter wheat trials
0-0.05
0.05-0.10
0.10-0.15
0.15-0.20
0.20-0.25
0.25-0.30
>0.30
Wheat harvested
area fraction
North China Plain
(n=1722)
Northwest China
(n=539)
Yangtze River Basin
(n=1622)
North China Plain
Northwest China
Northeast China
Southwest China
Maize trials
0-0.05
0.05-0.10
0.10-0.15
0.15-0.20
0.20-0.25
0.25-0.30
>0.30
Maize harvested
area fraction
80°E
130°E
80°E
130°E
0-0.05
0.05-0.10
0.10-0.15
0.15-0.20
0.20-0.25
0.25-0.30
>0.30
Rice harvested
area fraction
Yangtze River Basin
Southwest China
South China
Northeast China
Rice trials
Northeast China
(n=1030)
Southwest China
(n=1056)
North China Plain
(n=1608)
Yangtze River Basin
(n=2104)
South China
(Early rice,n=1262
Late rice, n=1172)
80°E
130°E
80°E
130°E
a
b
c
50°N
20°N
NEC
NCP
NWC
YZB
SWC
SC
*
*
*
*
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
2040-2059
2080-2099
W-NCP
*
*
*
*
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
M-NCP
2040-2059
2080-2099
*
*
*
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
W-YZB
2040-2059
2080-2099
*
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
M-NEC
2040-2059
2080-2099
*
*
*
*
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
W-NWC
2040-2059
2080-2099
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
M-SWC
2040-2059
2080-2099
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
SR-YZB
2040-2059
2080-2099
*
*
*
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
ER-SC
2040-2059
2080-2099
*
*
*
−20
−15
−10
−5
0
5
10
15
20
RCP2.6 RCP8.5 RCP2.6 RCP8.5
Y
ield change (%)
high
low
LR-SC
2040-2059
2080-2099
W-NCPW
-YZ
B
W-NW
C
M-NE
C
M-NCPM-S
WC
SR-Y
ZB
ER-S
C
LR-S
C
Cere
al
pro
duct
io
n
chan
ges
(Mt
)
-8
-6
-4
-2
0
2
4
6
8
BAU
SQ improvement
SQ degradation
Cropping systems
W-NCPW
-YZ
B
W-NW
C
M-NE
C
M-NCPM-S
WC
SR-Y
ZB
ER-S
C
LR-S
C
Cere
al
pro
duct
io
n
chan
ges
(Mt
)
-8
-6
-4
-2
0
2
4
6
8
Cropping systems
W-NCPW
-YZ
B
W-NW
C
M-NE
C
M-NCPM-S
WC
SR-Y
ZB
ER-S
C
LR-S
C
-8
-6
-4
-2
0
2
4
6
8
W-NCPW
-YZ
B
W-NW
C
M-NE
C
M-NCPM-S
WC
SR-Y
ZB
ER-S
C
LR-S
C
-8
-6
-4
-2
0
2
4
6
8
2040-2059
RCP2.6
2080-2099
RCP2.6
2040-2059
RCP8.5
2080-2099
RCP8.5
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
a
b
c
d
*
*
*
yti
vit
c
u
d
or
p
p
or
C
Climate
High SQ
Low SQ
Yield variability under
current climate
Climate
suitability
Current climate variability
Future climate variability
Yield variability under
future climate
Yi
e
ld
BMPs
(Mg
/h
a
)
0
2
4
6
8
10
12
14
16
18
0
2
4
6
8
10
12
Yield
BMPs
Mean Yield
BMPs
0
2
4
6
8
10
12
14
Ranked site-year number
W-NCP
M-NCP
W-YZB
W-NWC
M-NEC
M-SWC
SR-YZB
ER-SC
LR-SC
a
b
c
d
e
f
g
h
i
Yield
BMPs
(Mg/ha): 6.7 (± 1.2)
CV (%): 18
N=1722
Yield
BMPs
(Mg/ha): 6.3 (± 1.2)
CV (%): 20
N=1622
Yield
BMPs
(Mg/ha): 6.6 (± 1.5)
CV (%): 22
N=539
Yield
BMPs
(Mg/ha): 9.6 (± 1.8)
CV (%): 19
N=1030
Yield
BMPs
(Mg/ha): 8.1 (± 1.6)
CV (%): 17
N=1608
Yield
BMPs
(Mg/ha): 7.8 (± 1.4)
CV (%): 18
N=1056
Yield
BMPs
(Mg/ha): 8.5 (± 1.1)
CV (%): 13
N=2104
Yield
BMPs
(Mg/ha): 6.8 (± 1.1)
CV (%): 16
N=1262
Yield
BMPs
(Mg/ha): 7.0 (± 1.1)
CV (%): 16
N=1172
NEC
NCP
NWC
YZB
SWC
SC
W−NCP
Cultivar
P rate
K rate
Soil type
Soil texture
pH
N rate
GDD
Avail−K
SOM
Tmin
RAD
PRE
Olsen−P
Tmax
0
2
4
6
8
10 12 14
Relative influence (%)
Climate
Management
Soil
W−YZB
Cultivar
pH
K rate
P rate
Avail−K
GDD
SOM
Tmax
Olsen−P
Tmin
PRE
Soil texture
Soil type
N rate
RAD
0 2 4 6 8 10 12 14 16 18 20
Relative influence (%)
Climate
Management
Soil
W−NWC
pH
P rate
K rate
Cultivar
Soil texture
Avail−K
GDD
SOM
Tmax
Tmin
N rate
PRE
RAD
Olsen−P
Soil type
0
2
4
6
8 10 12 14 16
Relative influence (%)
Climate
Management
Soil
M−NEC
Cultivar
P rate
K rate
GDD
Soil texture
pH
RAD
Soil type
Tmax
Avail−K
Olsen−P
Tmin
PRE
SOM
N rate
0
2
4
6
8
10 12 14
Relative influence (%)
Climate
Management
Soil
M−NCP
Cultivar
P rate
Soil type
K rate
Soil texture
pH
N rate
SOM
Avail−K
Olsen−P
GDD
RAD
Tmax
Tmin
PRE
0
2
4
6
8
10 12 14
Relative influence (%)
Climate
Management
Soil
M−SWC
Cultivar
P rate
pH
GDD
Olsen−P
SOM
Tmax
RAD
K rate
N rate
Tmin
Avail−K
PRE
Soil type
0
2
4
6
8
10 12 14
Relative influence (%)
Climate
Management
Soil
SR−YZB
Cultivar
Soil type
P rate
SOM
pH
K rate
Olsen−P
GDD
Tmin
Avail−K
PRE
Tmax
RAD
Soil texture
N rate
0
2
4
6
8
10 12 14
Relative influence (%)
Climate
Management
Soil
ER−SC
Cultivar
P rate
N rate
K rate
pH
RAD
Avail−K
Olsen−P
Tmax
PRE
SOM
GDD
Tmin
0
2
4
6
8
10 12 14
Relative influence (%)
Climate
Management
Soil
LR−SC
Cultivar
K rate
P rate
pH
Olsen−P
GDD
Soil texture
SOM
N rate
Tmax
Avail−K
PRE
Tmin
RAD
0
2
4
6
8
10 12 14
Relative influence (%)
Climate
Management
Soil
0
560
1,120
1,680
280
Kilometers
0
540
1,080
1,620
270
Kilometers
0
510
1,020
1,530
255
Kilometers
0
450
900
1,350
225
Kilometers
0
660
1,320
1,980
330
Kilometers
0
450
900
1,350
225
Kilometers
0
420
840
1,260
210
Kilometers
0
450
900
1,350
225
Kilometers
0
490
980
1,470
245
Kilometers
high
low
M-NEC
M-NCP
W-NCP
W-YZB
SR-YZB
W-NWC
M-SWC
ER-SC
LR-SC
Legend
a
R
2
= 0.99 ***
10
20
30
40
10
20
30
40
Average maximum temperature
in high quality soils (
℃
)
A
v
e
ra
g
e
ma
xi
m
u
m
te
mp
e
ra
tu
re
i
n
l
o
w
q
u
a
lit
y
so
ils
(
℃
)
W−NCP
W−YZB
W−NWC
M−NEC
M−NCP
M−SWC
SR−YZB
ER−SC
LR−SC
b
R
2
= 0.99 ***
0
10
20
30
0
10
20
30
Average minimum temperature
in high quality soils (
℃
)
A
v
e
ra
g
e
mi
n
im
u
m
te
mp
e
ra
tu
re
i
n
l
o
w
q
u
a
lit
y
so
ils
(
℃
)
c
R
2
= 0.92 ***
0
500
1000
1500
0
500
1000
1500
Cumulative precipitation
in high quality soils (mm)
C
u
m
u
la
ti
v
e
p
re
ci
p
it
a
ti
o
n
i
n
l
o
w
q
u
a
lit
y
so
ils
(mm)
d
R
2
= 0.96 ***
0
1000
2000
3000
4000
0
1000
2000
3000
4000
Cumulative radiation
in high quality soils (MJ/m²)
C
u
m
u
la
ti
v
e
r
a
d
ia
ti
o
n
i
n
l
o
w
q
u
a
lit
y
so
ils
(MJ/
m²)
2040−2059
RCP2.6
−20
−15
−10
−5
0
5
10
15
20
W
−N
C
P
W
−YZ
B
W
−N
W
C
M
−N
EC
M−N
C
P
M−SW
C
SR
−YZ
B
ER
−SC
LR
−SC
Y
ie
ld
ch
a
n
g
e
(%
)
2040−2059
RCP8.5
−20
−15
−10
−5
0
5
10
15
20
W
−N
C
P
W
−YZ
B
W
−N
W
C
M
−N
EC
M−N
C
P
M−SW
C
SR
−YZ
B
ER
−SC
LR
−SC
2080−2099
RCP2.6
−20
−15
−10
−5
0
5
10
15
20
W
−N
C
P
W
−YZ
B
W
−N
W
C
M−N
EC
M−N
C
P
M−SW
C
S
R
−YZ
B
ER
−SC
LR
−SC
Cropping systems
Y
ie
ld
ch
a
n
g
e
(%
)
2080−2099
RCP8.5
−20
−15
−10
−5
0
5
10
15
20
W
−N
C
P
W
−YZ
B
W
−N
W
C
M−N
EC
M−N
C
P
M
−SW
C
SR
−YZ
B
ER
−SC
LR
−SC
Cropping systems
a
b
c
d
2040−2059
RCP2.6
a
−1
0
1
2
3
4
5
6
7
−8
−6
−4
−2
0
2
4
6
8
10
D
if
fe
re
n
ce
o
f
yi
e
ld
ch
a
n
g
e
b
e
tw
e
e
n
h
ig
h
a
n
d
l
o
w
q
u
a
lit
y
so
ils
(%
)
W−NCP
W−YZB
W−NWC
M−NEC
M−NCP
M−SWC
SR−YZB
ER−SC
LR−SC
2040−2059
RCP8.5
b
−1
0
1
2
3
4
5
6
7
−8
−6
−4
−2
0
2
4
6
8
10
2080−2099
RCP2.6
c
−1
0
1
2
3
4
5
6
7
−8
−6
−4
−2
0
2
4
6
8
10
Projected yield change under climate
change scenarios (%)
D
if
fe
re
n
ce
o
f
yi
e
ld
ch
a
n
g
e
b
e
tw
e
e
n
h
ig
h
a
n
d
l
o
w
q
u
a
lit
y
so
ils
(%
)
2080−2099
RCP8.5
d
−1
0
1
2
3
4
5
6
7
−8
−6
−4
−2
0
2
4
6
8
10
Projected yield change under climate
change scenarios (%)