Page 1
RESEARCH ARTICLE
Possible factors determining global-scale
patterns of crop yield sensitivity to drought
Vempi Satriya Adi Hendrawan
ID
1,2
*
, Daisuke Komori
2,3
*
, Wonsik Kim
4
1 Department of Civil and Environmental Engineering, Universitas Gadjah Mada, Yogyakarta, Indonesia,
2 Graduate School of Environmental Studies, Tohoku University, Sendai, Japan, 3 Graduate School of
Engineering, Tohoku University, Sendai, Japan, 4 Institute for Agro-Environmental Sciences, National
Agriculture and Food Research Organization, Tsukuba, Japan
*
vempi@ugm.ac.id
(VSAH);
daisuke.komori.e8@tohoku.ac.jp
(DK)
Abstract
In recent decades, droughts have critically limited crop production, inducing food system
shocks regionally and globally. It was estimated that crop yield variability in around one-third
to three-fourths of global harvested areas is explained significantly by drought, revealing the
notable vulnerability of crop systems to such climate-related stressors. However, under-
standing the key factors determining the global pattern of crop yield sensitivity to drought is
limited. Here, we investigate a wide range of physical and socioeconomic factors that may
determine crop-drought vulnerability in terms of yield sensitivity to drought based on the
Standardized Precipitation Index at 0.5˚ resolution from 1981 to 2016 using machine learn-
ing approaches. The results indicate that the spatial variations of the crop-drought sensitivity
were mainly explained by environmental factors (i.e., annual precipitation, soil water-holding
capacity, soil acidity, annual potential evapotranspiration) and crop management factors
(i.e., fertilizer rate, growing season). Several factors might have a positive effect in mitigating
crop-drought vulnerability, such as annual precipitation, soil water holding capacity, and fer-
tilizer rate. This study quantitatively assesses the possible effect of various determinants
which might control crop vulnerability to drought. This understanding may provide insights
for further studies addressing better crop vulnerability measures under future drought
stress.
Introduction
The impact of global climate variabilities on crops has been well investigated. Empirically,
around one-third of the variability of crop measures (i.e., crop yield, production) [
1
–
3
] and
harvested area globally [
4
] are explained by climate variabilities during the last several decades.
As one climate extreme, drought has been widely recognized as a notable limiting factor for
crop yield and production [
5
,
6
]. One-third to three-fourths of the global harvested areas expe-
rienced yield losses due to drought [
7
,
8
], corresponding to approximately 166 billion US dol-
lars from 1983 to 09 [
7
]. The serious drought impact on crops may unveil the global crop
system’s vulnerability to climate-related stressors, which can later threaten local or global food
security in the current and future climate [
3
,
9
,
10
].
PLOS
ONE
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
1 / 20
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Hendrawan VSA, Komori D, Kim W
(2023) Possible factors determining global-scale
patterns of crop yield sensitivity to drought. PLoS
ONE 18(2): e0281287.
https://doi.org/10.1371/
journal.pone.0281287
Editor: Arun Kumar Shanker, Central Research
Institute for Dryland Agriculture, INDIA
Received: September 2, 2022
Accepted: January 19, 2023
Published: February 2, 2023
Peer Review History: PLOS recognizes the
benefits of transparency in the peer review
process; therefore, we enable the publication of
all of the content of peer review and author
responses alongside final, published articles. The
editorial history of this article is available here:
https://doi.org/10.1371/journal.pone.0281287
Copyright:
©
2023 Hendrawan et al. This is an
open access article distributed under the terms of
the
Creative Commons Attribution License
, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: Datasets generated
or analyzed during the current study are available at
http://doi.org/10.6084/m9.figshare.21298785
.
Funding: D.K. was partly supported by the
International Joint Graduate Program in Resilience
Page 2
Approaches for understanding the drought risk on the crop system may include assessing
its potential impact and crop vulnerability under specific dry condition definitions (e.g., pre-
cipitation deficit, soil moisture stress) [
11
–
13
] during the growing period as a hazard and
exposure measures [
9
]. The drought impact on crops can be measured by damages or losses
on crop production measures (i.e., crop yield, crop production, or harvested area) due to cer-
tain drought conditions, whereas the vulnerability can be presented by the sensitivity or sus-
ceptibility of the crop production responding a drought severity [
9
,
14
]. While the drought
impact on crops and their vulnerability is growingly evident (through field-scale experiments
to larger scale based on global physical modeling and statistics [
1
,
15
–
18
]), quantitative identi-
fication of multiple factors that may determine such crop-drought relationship patterns across
global agricultural systems is limited in previous works.
Agricultural risk and vulnerability to drought in a region may be determined by several fac-
tors [
14
,
17
], which can be generally grouped as physical (e.g., climate, soil, topography, crop
production factors) and socio-economic (e.g., crop management, agricultural input, field size,
demography, farmer income, Gross Domestic Product (GDP) per capita, technology) [
19
–
21
].
These determinants govern the adaptive capacity of a system to climate disruptions over time.
For instance, studies [
7
,
22
] reported that technological improvements (i.e., high-yield breed-
ing, adoption of drought-tolerant seed) associated with per capita GDP growth could further
increase the resilience of crop production systems to droughts in a region. Current reports also
revealed that more resilient agricultural systems, due to social, institutional, and agroecological
factors [
23
] could adapt to stress and recover better than those less resilient, even with a com-
parable or higher climate extreme exposure [
20
,
22
]. Therefore, quantifying the adverse impact
on crops due to extreme events such as a drought is imperative to understanding crop vulnera-
bility to climate disruptions since drought impact is known to be one of the most disastrous
climate extremes causing crop damage.
Furthermore, how socio-economic factors drive the global crop-drought relationship is
often overlooked. Several studies [
12
,
19
,
21
] attempted to quantitatively explore how various
physical and socio-economic factors might control crop productivity measures (i.e., crop yield,
production). Even though numerous existing local studies with a specific context and underly-
ing physical and socio-economic conditions existed [
24
,
25
], these studies might not represent
the understanding to a larger extent. On a global scale, previous studies often use aggregated
country or sub-country level data to assess socio-economic factors’ influence to crop vulnera-
bility [
19
,
20
,
26
,
27
]. This approach might not represent heterogeneous agricultural systems,
given that adaptive capacity may differ within local or smaller levels [
28
–
30
].
For instance, existing global scale studies by Simelton et al. [
20
] and currently Kinnunen
et al. [
31
] have attempted to understand the socioeconomic factors which determine crop vul-
nerability to climatic stress (i.e., drought and heat). They employed several socioeconomic
proxies at subnational and national levels, such as governance, GDP per capita, human devel-
opment index, fertilizer, water stress index, and irrigation infrastructure indicator. They
emphasized the importance of socioeconomic factors (e.g., income level, governance, fertilizer
application) in defining global crop vulnerability to climatic stress. These existing studies also
suggested crop production in countries with higher agricultural investments (i.e., higher fertil-
izer application) is generally less vulnerable to drought, despite a variation across crops and
the type of regions. However, some complex and nonlinear relationship has been reported,
particularly for economic indicator (e.g., GDP per capita, human development index) [
20
].
Moreover, these previous studies have not discussed the effect of importance and the direction
of how a wide range of factors combining physical (e.g., climate, soil, topography, irrigation,
crop production factors) and socioeconomic variables affect crop-drought vulnerability in a
finer spatial scale (i.e., grid-scale).
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
2 / 20
and Safety Studies, Tohoku University. The funder
had no role in study design, data collection and
analysis, decision to publish, or preparation of the
manuscript.
Competing interests: The authors have declared
that no competing interests exist.
Page 3
Therefore, this study addresses the above gaps of lacking quantitative evidence of the effect
of multiple factors (i.e., physical, socioeconomic) in determining global crop-drought vulnera-
bility based on gridded historical data. Particularly, this present study explores 1) how much
global harvested area and crop yield anomaly are significantly affected by drought and 2) how
the key factors mitigate or exacerbate the crop-drought vulnerability in terms of their crop
yield sensitivity to drought. Here, we examine the sensitivity of maize, rice, soybean, and
wheat yield to drought based on a meteorological drought indicator as drought proxy (Stan-
dardized Precipitation Index, SPI) with a spatial resolution of 0.5˚ grid-cell from 1981 to 2016.
Finally, we explore how the key factors determine global crop-drought sensitivity using
machine learning approaches. These study results are necessary to indicate which factors are
important to determine the global crop sensitivity to drought on a global scale. This may pro-
vide light for further attempts to understand crop system vulnerability and its determinants
for future adaptation and mitigation efforts in the global context.
Materials
The dataset used in this study mainly consists of global gridded crop yield data, the Standard-
ized Precipitation Index (SPI), and several indicators that possibly control crop sensitivity to
drought. We consider all the data for 1981–2016 associated with this study analysis period.
Since the spatial resolution of the datasets varies, we re-gridded all the datasets to 0.5˚ using
bilinear interpolation.
Drought indicators
Here, the drought indicator is based on one of the meteorological drought indices, the Stan-
dardized Precipitation Index (SPI) [
32
]. While SPI is solely based on precipitation, this index
has been widely used and well-recognized in previous drought studies and applications [
33
–
36
]. Other drought indexes such as the Palmer Drought Severity Index (PDSI), the Standard-
ized Precipitation Evapotranspiration Index (SPEI), and other indices (e.g., soil moisture
anomaly) have also been used but require more parameterization, which may introduce addi-
tional uncertainties [
37
,
38
]. Therefore, the simple but robust drought definition based on SPI
is employed in this study to reveal the global drought pattern from 1981 to 2016.
SPI is obtained based on the standardization of monthly precipitation accumulation within
various timescales (e.g., 3, 6, 9, 12 months) over the window period of 1981–2016; thus, the
standardized value finally has a mean of zero and a standard deviation of one. Before standard-
ization based on the normal Gaussian distribution, the data is fitted based on the Gamma dis-
tribution [
39
]. Here, SPI is developed based on several historical global gridded precipitations
to obtain the best agreement among datasets according to their ensemble mean (GPCC, CRU,
PRECL, UDEL, CPC, MSWEP, MERRA-2, and ERA-5 dataset) [
8
]. SPI is produced for each
dataset, and then the mean of the eight selected datasets is obtained to represent ensemble SPI
as the drought index.
The drought index (
DI
) based on negative SPI is used to represent the inter-annual drought
condition in this study (
Eq 1
) [
8
]. We obtain yearly
DI
based on monthly SPI on harvest
month each year. The harvest month data for each grid cell is obtained from the crop calendar
dataset during the year 2000 [
40
] and is constantly applied every year from 1981 to 2016. The
drought index (
DI
) is calculated as:
DI
t
¼
SPI
m
;
t
;
SPI
m
;
t
<
0
null
;
SPI
m
;
t
�
0
ð
1
Þ
(
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
3 / 20
Page 4
where
SPI
m
,
t
is the 9-month SPI (SPI-9) in harvest month
m
and year
t
. SPI with a 9-month
timescale is used to indicate a long-term precipitation anomaly which may indicate a more
prominent drought [
32
,
33
]. The spatial variation of the global crop-drought relationship is
broadly similar when different SPI timescales are used [
8
].
Crop yield anomaly
The percentage of yield anomaly (
Δ
Y
) is estimated based on the gridded dataset of historical
yield (GDHY) [
41
] for the major crops: maize, rice, soybean, and wheat (
Eq 2
). We calculate
long-term crop yield trends as normal or expected yields (
�
Y
) using the local polynomial
regression method [
8
]. This method is selected since it can account for possible nonlinear
trends and works well for limited data series, particularly compared to the moving average
method [
42
]. This long-term trend generally represents the improvement of technological
advances in producing higher yields in a region, while the yield anomaly subtracted from the
expected yield is assumed to be caused by climate-related factors, although other affecting fac-
tors, such as socioeconomic factors, may also influence this short-term anomaly.
D
Y
t
¼
Y
t
�
Y
t
�
Y
t
�
100
%
ð
2
Þ
where
Y
is crop yield (t ha
-1
) and
�
Y
is the trend obtained by local polynomial regression (t
ha
-1
) in a given year
t
.
Crop-drought determinants
We obtain several possible factors determining crop yield and drought relationship. We define
the possible determinants as (1) climate factors [
1
], (2) terrain factors [
43
,
44
], (3) soil factors
[
45
,
46
], (4) irrigation factors [
26
,
27
], (5) crop production (hereafter production) factors [
47
],
(6) fertilizer factors [
20
,
48
], and (7) socio-economic factors [
12
] (see
Table 1
). All variables
shown in
Table 1
are then used as independent variables to train the Random Forest model
(see
Methods
). Moreover, it is noteworthy that we use an average value or a subset of the data-
set in a specific year between 1981–2016, despite a possible change during the study period.
Previous studies [
46
] have also used this approach, and year-specific datasets are widely
employed, especially in the case of crop calendar and harvest area datasets [
49
,
50
], which were
only available around the year 2000.
Methods
Crop-drought sensitivity
We estimate crop-drought sensitivity as the slope coefficient (
β
) of a linear regression between
yield anomaly (
Δ
Y
) and
DI
in the given year as time series (
Eq 3
). Statistical significance con-
sidered in this study is based on a two-tailed t-test with a threshold of
P
<
0.05.
D
Y
¼
a
þ
b
�
DI
ð
3
Þ
where
a
and
β
are the model intercept and slope, respectively. These linear regression param-
eters are fitted based on the ordinary least-squares method using the “statsmodels” library in
Python [
60
] (see
S1 Fig
for an example of the regression model in a grid cell). This calculation is
applied for each grid cell and crop. Here, crop-drought sensitivity refers to the simple linear
relationship, despite possible nonlinear responses and more complex relationships occurring
across regions. Nevertheless, this simple model allows us to understand the general pattern of
how crop yield responds to drought, as demonstrated by previous studies [
1
,
61
].
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
4 / 20
Page 5
Random forest
Here we use one of the machine learning algorithms, Random Forest [
62
], to reveal the key
factors defining the crop-drought relationship. We use Random Forest algorithm due to its rel-
atively better accuracy, interpretability, and efficiency with large training datasets than other
machine learning methods (e.g., decision trees, nearest neighbors) [
62
–
64
]. In addition, we
test the sensitivity of our results by comparing the model performances estimating crop-
drought sensitivity with the outputs of two other machine learning algorithms: support vector
regression (SVR) [
65
] and gradient boosting (XGBoost) [
66
], implemented in the R library,
"caret" [
67
].
We set the crop-drought sensitivity (
β
) as the response variable and various possible deter-
minants (
Table 1
) as independent variables or predictors. The Random Forest model has been
used to detect a nonlinear climate-crop system relationship [
2
,
46
]. Here, the Random Forest
algorithm is applied using the “randomForest” library in R [
68
]. We use parameter input as
default settings (e.g., number of trees = 500) since the model skill with basic setting is adequate
to reveal the variable importance and their influence on global crop sensitivity [
46
]. The model
is fitted with the data from all available grid cells (representing spatial dimension) for each
crop. During the calculation process, explained variance is calculated based on OOB (out-of-
bag) error estimation by subsampling with replacement in the training samples [
68
].
Further, we obtain relative importance [
66
] for each variable based on the fitted models.
Here we use the metric measuring number of tree nodes employing the variable to reduce the
model’s impurity averaged across all trees [
68
]. The “randomForest” calculates and scales it as
Table 1. Determinants used to describe the crop-drought sensitivity analyzed in the study. The dataset sources and format are listed in
S1 Table
.
Class
Variable
Unit
Period
Source
Climate
Mean annual precipitation
mm year
-1
1981–2016
[
51
]
Mean annual PET
mm year
-1
Mean temperature
˚C
Terrain
Elevation
m
2000
[
52
]
Slope
-
Soil
Topsoil saturated hydraulic conductivity
a
cm day
-1
2013
[
53
]
Topsoil clay amount
% weight
2000
[
54
]
Topsoil organic carbon
kg C m
-2
Topsoil acidity
pH
Water-holding capacity
mm
1993
[
55
]
Irrigation
Area equipped for irrigation (AEI)
% of land area
2005
[
50
,
56
]
Area actually irrigated (AAI)
% of AEI
Area irrigated with groundwater (AEIGW)
% of AEI
Area irrigated with water from non-conventional sources (AEINC)
% of AEI
Production
Growing season length
days
The 1990s –early 2000s
[
40
]
Harvested area
km
2
year
-1
1998–2002
[
49
]
Fertilizer
Nitrogen rate application
kg ha
-1
2000
[
57
]
Phosphorus rate application
kg ha
-1
Potassium rate application
kg ha
-1
Socioeconomic
Gross Domestic Product (GDP)
billion United States Dollar (USD)
2015
[
58
]
GDP per capita
USD
Population density
people km
-2
2015
[
59
]
a
30cm depth
https://doi.org/10.1371/journal.pone.0281287.t001
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
5 / 20
Page 6
the total importance for all variables equals one. Then, for ease of comparison, we normalize
them into the 0–1 range, representing the minimum and maximum values, respectively.
Finally, we plot all models based on different settings and rank them from the most important
parameter to the least.
We use a partial dependence plot from the fitted models to understand each variable’s rela-
tionship with crop-drought sensitivity as the response variable [
66
]. The partial dependence
allows selecting small subsets or single variables to reveal a relationship between predictor(s)
and a response variable, similar to sensitivity analysis [
2
]. Here, we pick up the six most influen-
tial variables based on their overall relative importance (from the average of all experiment set-
ups) to demonstrate their relationship with crop-drought sensitivity. This enables us to estimate
the effect of the important variable to determine global crop sensitivity to drought. This study
also calculates the variable importance and partial dependence plots based on SVR and XGBoost
algorithms using “caret” and "pdp" R library [
69
] to investigate the robustness of our results.
Results
Crop-drought sensitivity pattern
We obtain crop-drought sensitivity based on the slope of the linear regression (
β
) between crop
yield anomalies and the meteorological Drought Index (
DI
) based on SPI with a 9-month time-
scale (
Fig 1
). Significant relationships are evident in around a quarter of croplands or over around
Fig 1. Crop yield sensitivity to drought based on 9-month SPI across global croplands based on GDHY for maize, rice, soybean, and wheat. Sensitivity is
calculated by the linear slope coefficient (
β
) as crop yield anomaly (%) per drought magnitude based on SPI, while white cells show no cropland or crop yield
data. We show a map over entire global croplands for a spatially broader interpretation. The map showing only areas with significant crop yield sensitivity to
drought is provided in
S2 Fig
(
P
<
0.05). The map is made with the “cartopy” library in Python (version 0.17) and the Natural Earth free vector and raster map
data (
naturalearthdata.com
), which are freely available through the public domain [
70
].
https://doi.org/10.1371/journal.pone.0281287.g001
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
6 / 20
Page 7
one-fourth of global maize, wheat, and soybean croplands, respectively (
P
<
0.05). Almost entire
significant croplands show a negative correlation (~96%), confirming the damaging impact of
drought on major crops. In contrast, the negative impact of drought on rice seems offset by yield
benefits and remains low globally, with only 7% significantly correlated croplands [
7
].
Globally, drought reduces yield by 4%, 1%, 3%, and 6% for maize, rice, soybean, and wheat,
respectively, with increasing drought index
DI
(i.e., precipitation deviations from normal condi-
tion per precipitation standard deviations). Over a quarter of the total croplands with significant
drought impact, yields drop by around 14% by one drought unit (maize, soybean, and wheat).
Furthermore, we point out primary regions with a substantial drought impact, comprising gen-
erally dry croplands such as The Great Plains, Africa (southern Africa, The Sahel, and the Horn
of Africa), Australia, northeast China, the Mediterranean basin, and Central Asia (
Fig 1
) [
7
,
8
].
Droughts also slightly damage maize and wheat yield in wetter regions such as South America,
The Pampas, and the African and Southeast Asian tropics. While droughts induce rice yield loss
in parts of The Sahel region and Northeast China and slightly in central Asia, rice in the tropics
broadly shows weak sensitivity to drought. The higher crop yield sensitivity over drylands may
indicate their higher vulnerability along with higher drought hazard exposures.
Possible driving factors
We investigate the importance of several possible determinants as independent variables to
control crop-drought sensitivity. We train the Random Forest models for regression with pre-
dictors of 22 variables. Results show that the variable predictors can explain overall 43%, 36%,
66%, and 37% of crop-drought sensitivity variations (R
2
) with root mean square error (RMSE)
of 8.61%, 5.89%, 4.63%, 10% (crop yield anomaly per drought magnitude,
β
) for maize, rice,
soybean, and wheat, respectively. The explanatory skills of the 22 variables are robust across
different types of machine learning algorithms (SVR and XGBoost), showing similar model
performances with a small standard deviation of R
2
and RMSE by 1% and 0.2%, respectively,
across crop types and machine learning algorithms (
S2 Table
). Furthermore, in rice, where the
models less explain the crop-drought sensitivity variation, there are likely other more complex
effects of factors contributing to rice vulnerability to drought [
1
,
7
].
Here, we derive each variable’s relative contribution to the models. While the important
parameters vary across crops, we average their relative importance to reveal overall important
variables (
Fig 2
). It shows that mean annual precipitation, water-holding capacity, soil acidity,
Potential Evapotranspiration (PET), elevation, and potassium application rate are among the
most driving factors, revealing that climate, soil, terrain, and agricultural input factors play an
important role in crop-drought sensitivity variations. In particular, for each crop, the result
suggests that the annual precipitation pattern becomes important for all crops, while water-
holding capacity, elevation, PET, acidity, and potassium rate are relevant for some crops
(
Table 2
). Regarding management factors, groundwater irrigation potential and population
density are slightly important for maize, soybean, and rice.
The importance of variables is generally consistent across the machine learning models,
particularly the order of the most important variables (see
S3 Fig
). Note that the importance of
variables from the three machine learning algorithms (
Fig 2
and
S3 Fig
) may slightly vary in
absolute value due to different methods implemented [
67
]. The variable importance and par-
tial dependence plot of SVR and XGBoost are obtained using “caret” library.
Impact behavior of each factor
We obtain a partial dependence plot for each variable to describe their functional relationship
with crop-drought sensitivity patterns (
Fig 3
and
S4 Fig
). The subsequent sections discuss the
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
7 / 20
Page 8
variable relationship for each parameter among different factors (climate, terrain, soil, produc-
tion, irrigation, fertilizer, and socioeconomic factors) based on Random Forest model. We also
obtain the partial dependence plots based on SVR and XGBoost (
S5
and
S6
Figs). The results
confirm that most estimated relationships between various determinants and crop yield sensi-
tivity are generally robust across the models.
Climate factors
The most important factor, mean annual precipitation, demonstrates a clear relationship with
crop-drought sensitivity; a higher precipitation rate alleviates crop yield reduction due to
Fig 2. The relative importance of the determinant factors in explaining crop-drought sensitivity based on random forest model. The importance values
are normalized to the 0–1 range computed separately across different crops. The colored points represent the relative importance of each variable. Label order
of the y-axis indicates the order from the most important variables to the least based on the overall average across crops.
https://doi.org/10.1371/journal.pone.0281287.g002
Table 2. Top six important parameters for each crop based on the average relative importance. The overall average represents the average from combinations across
crops corresponding to the order of the y-axis labels in
Fig 2
.
Rank
Maize
Rice
Soybean
Wheat
Overall average
1
Precipitation
Precipitation
Water-holding capacity
Precipitation
Precipitation
2
Water-holding capacity
PET
Potassium rate
Acidity
Water-holding capacity
3
Elevation
Population density
Irrigation (groundwater)
Water-holding capacity
Acidity
4
Growing season length
Acidity
Precipitation
Nitrogen rate
PET
5
Irrigation (groundwater)
Elevation
Elevation
Hydraulic conductivity
Elevation
6
Soil organic carbon
Irrigation (groundwater)
PET
Temperature
Potassium rate
https://doi.org/10.1371/journal.pone.0281287.t002
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
8 / 20
Page 9
drought. For instance, based on functional relationships in maize and wheat in which the yield
loss sensitivity is high, the higher annual precipitation (e.g., ~500–1000 mm year
-1
) increase
may mitigate yield loss (
Fig 3
). This result indicates that crop yield in semi-arid and arid
regions with less precipitation is more sensitive to drought impact, as reported elsewhere [
7
,
71
–
73
]. Furthermore, this confirms that crop loss sensitivity increases in most drylands, where
severe hazards and higher vulnerability occur concurrently. The other climate factor, PET,
shows an inverse response to precipitation—higher PET tends to aggravate yield loss sensitiv-
ity due to drought, while temperature slightly determines the response variables.
Terrain factors
The higher elevation is found to be an important parameter with a diminishing effect on yield
loss sensitivity. The response is particularly evident for maize, soybean, and wheat. This
implies that more serious crop-yield loss sensitivity due to drought is generally located in low-
lying cropland; while in higher elevations, where the precipitation rate is generally higher,
drought impact may be lessened [
74
]. Moreover, atmospheric vapor demand would generally
be lower at higher altitudes, which may alleviate moisture stress due to drought [
75
]. On the
other hand, the other terrain factor, the slope, does not show a clear relationship with the
drought-induced yield sensitivity.
Fig 3. Relationship between the response variable of crop-drought sensitivity (
β
) as the response variable and the six selected determinant factors as
independent variables based on each partial dependence using random forest models. The order shows higher to lower overall relative importance across
crops (see
Table 2
). The lower negative y-axis indicates higher yield loss due to drought. The partial dependence plot for all variables (22 variables) is provided
in
S4 Fig
.
https://doi.org/10.1371/journal.pone.0281287.g003
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
9 / 20
Page 10
Soil factors
The most important soil factor, water-holding capacity, generally shows a consistent pattern for
all crops. The higher value may slightly moderate drought-induced yield loss, as indicated in the
previous studies [
76
,
77
]. For instance, this parameter’s functional relationship shows a decrease
in drought-induced yield loss in all crop regions by around 330 mm water-holding capacity.
The other soil factor, soil acidity, tends to increase the drought-induced yield loss by pH values
ranging from 7 to 8. In this range, soil condition is generally more saline, while soil may become
alkaline with a pH of more than 8.5 [
72
]. With these conditions, salt and alkali stress may simul-
taneously restrict crop development in semi-arid and arid regions, further exacerbating crop
yield loss to drought [
78
,
79
]. Moreover, hydraulic conductivity may also lessen drought-
induced yield loss, especially for maize and wheat within hydraulic conductivity values (i.e., up
to 25 cm d
-1
). This dependence generally may be related to more effective water transport and
increasing soil water content potential as soil hydraulic conductivity increases [
80
,
81
], despite
more complex mechanisms that may be involved [
15
,
82
]. The other parameters: soil organic
carbon and clay soil amount, however, generally remain less influence given lower relative
importance in the model, despite their importance in some cases (e.g., higher clay soil amount
in wheat seems to slightly aggravate negative crop sensitivity to drought).
Production factors
The crop production factor, growing season length, is revealed as one of the important factors
that may worsen wheat yield loss due to longer season length, particularly for maize and
wheat. The longer season length is generally attributable to winter wheat being dominated by a
global rainfed system, and the crop may be highly reliant on precipitation and sensitive to pre-
cipitation variability. For instance, the longer maize growing season (
>
~130 days) may exacer-
bate the yield loss. This growing season covers most of the major maize belts in higher
latitudes, including most semi-arid and arid lands, while the shorter season spans around the
tropics. The other parameter, the harvested area, may exhibit lesser effects.
Irrigation factors
Overall, irrigation parameters (i.e., percentage of area equipped for irrigation, AEI; actual irri-
gation, groundwater; and non-conventional water sources) show an unclear relationship with
the yield sensitivity, despite a slightly higher relative importance, especially for the groundwa-
ter irrigation parameter. The remaining irrigation parameters tend to have less control over
the drought-induced yield sensitivity. We acknowledge that these results contrast with previ-
ous studies arguing that irrigation may partly mitigate drought impact to crop yield.
Fertilizer factors
The fertilizer factors, potassium and phosphorus application rates, are revealed among the
important variables. In general, their higher application rate may slightly alleviate yield loss
due to drought. These fertilizer applications may effectively relieve drought stress on crops and
enhance drought tolerance [
83
–
85
]. This mechanism may be generally explained by reducing
the uptake of toxic nutrients and strengthening the physiological efficiency [
83
], which may
improve overall resilience to drought, while in the particular further mechanism of the role of
nutrients to mitigate drought stress on plants has been discussed in previous studies [
86
].
Moreover, the other fertilizer factor, nitrogen application rate, does not show an apparent rela-
tionship with the yield sensitivity indicator, despite a notable exception for the wheat with an
easing effect on the drought-induced yield loss.
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
10 / 20
Page 11
Socio-economic factors
GDP per capita relatively shows a stronger relationship with the yield sensitivity than the GDP
parameter, especially in the case of wheat and soybean. However, these relationships remain
weak, preventing us from drawing a conclusion. The other factor, the population indicator,
does not determine yield sensitivity to drought.
Discussion
The present study assesses crop sensitivity to drought from 1981 to 2016 based on a meteoro-
logical drought index using a 9-month Standardized Precipitation Index (SPI) and gridded
crop yield datasets for maize, rice, soybean, and wheat. Our estimates reveal that over one-
fourth of global cropland is significantly sensitive to inter-annual drought, which generally
agrees with previous studies [
2
,
7
,
73
], despite slightly lower estimates than the previous study
using a similar approach and dataset [
7
]. A negligible drought impact on rice is also confirmed
in this study, which may be attributable to its low exposure to drought with general moisture
surplus as reported elsewhere [
61
,
87
]. Here, we highlight a corresponding result pattern
among related studies based on specific drought indicators and crop statistics during historical
periods [
7
,
88
,
89
]. The major global drylands are the hotspots of the drought-sensitive regions,
particularly for the three most affected crops: maize, soybean, and wheat. The significance of
the regions is also widely highlighted by previous studies [
7
,
11
,
73
].
This study explores several possible factors determining crop sensitivity to drought. We
indicate that environmental factors (i.e., climate, soil, terrain) have an important role in miti-
gating crop yield loss due to drought, while management factors such as fertilizer, growing sea-
son, and GDP per capita show moderate influence, while the irrigation factor seems to have
lesser control, in contrast to their typical role in mitigating the impact of drought [
2
,
27
,
61
,
87
,
90
]. This result may suggest that drought events in our scale are too large to prevent crop losses
by irrigation; water deficits occurred extensively in the case of large-scale droughts. Previous
study by Kinnunen et al. [
31
] also indicated the importance of fertilizer application to alleviate
crop yield variation during dry and hot years, while the irrigation factor was also relatively less
important due to possible differences between actual conditions and data representation.
We further discuss these findings regarding the irrigation effect as follows. First, we may
not detect the significant importance of irrigation factors as we cannot independently disen-
tangle the impact of drought on irrigated and rainfed crop yields. Here, we use mean annual
yield involving multiple crop growing systems (irrigated and rainfed) and seasons (major and
minor) [
41
]. Therefore, our estimates may offset the mitigating effect of irrigation due to a
comparable or even larger negative impact of drought on rainfed crops within the same grid
location. Previous studies primarily conducted on a local or regional scale employed irrigated
and rainfed yield data separately to obtain drought impact on each system [
90
], while further
challenges related to data availability may arise when dealing with a global scale analysis.
On the other hand, another study [
2
] detected response of inter-annual yield using sub-
country-level data (combined rainfed and irrigated yields) on each classified irrigated and
rainfed cropland based on a specific definition (e.g., irrigated cropland is defined as
>
80% of
harvested area irrigated and otherwise is rainfed). Similarly, we classify irrigated and rainfed
cropland based on the similar global harvested area dataset [
49
] and the classification used in
the previous study [
2
]. We then calculate the share of drought-affected areas for each crop-
drought sensitivity bin (
Fig 4
). We find that the rainfed system considerably dominates the
drought-affected regions resulted in this study, especially for soybean (99% of the total calcu-
lated harvested area), maize (89%), and wheat (83%), while in the case of rice, the proportion
of irrigated cropland is relatively higher (63%). Irrigation tends to mitigate rice yield loss as the
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
11 / 20
Page 12
centroid of the bins tends to be positive (yield gain), while rainfed rice tends to have yield loss.
For the major rainfed crops (maize, soybean, wheat), there seems to be no considerable differ-
ence between irrigated and rainfed system responses; both exhibit mainly yield loss despite a
large difference in cropland share. Nevertheless, the vast rainfed system is negatively impacted
by drought, shown by an extensive area proportion attributable to yield loss, which indicates
the substantial drought impact on this cropping system globally. Particularly, maize, soybean,
and wheat are damaged by drought, equal to 65%, 69%, and 62% of the calculated harvested
area. Therefore, we may argue that variability of irrigation fraction may not be adequate to
explain crop yield sensitivity to drought owing to the dominance of a rainfed system over
global croplands calculated in this study.
Further study may consider other important parameters for drought parameterization,
such as temperature or evapotranspiration (ET), as they play a significant role in exacerbating
drought impact through increasing evaporative demand [
61
,
91
,
92
]. The drought indicator
can also be represented by other drought parameters, such as growing season soil moisture
conditions based on physical modeling or remote sensing estimates on a global scale [
37
,
93
].
Previous studies reported that the impact of drought on agricultural production due to lacking
soil moisture is often associated with extreme heat, posing a more severe risk to crops [
61
,
94
,
95
]. Joint drought and heat events indicators may provide insight into these relationships [
88
,
96
,
97
]. We suggest considering these possible mechanisms accounting for hydrological fea-
tures on drought estimates (rather than only meteorological factors, e.g., precipitation), espe-
cially in some regions agricultural systems impose more complex atmospheric soil interaction
Fig 4. Crop-drought sensitivity in each rainfed and irrigated defined by
>
80% of irrigated harvested area for maize, rice, soybean, and wheat. The x-axis
represents the bins of crop yield sensitivity, and the y-axis represents the percentage of the harvested area where data is available. The dashed line indicates the
center value (zero), separating yield loss (negative) and yield gain (positive).
https://doi.org/10.1371/journal.pone.0281287.g004
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
12 / 20
Page 13
and are heavily reliant on the other common water sources such as irrigation and groundwater
[
61
,
98
].
Moreover, we indicate that the overall effect of socio-economic factors, i.e., GDP per capita,
is relatively weak in determining crop-drought sensitivity—which may be underestimated in
this study. Specifically for soybean, higher GDP may slightly induce higher yield loss sensitivity
to drought, while maize slightly responds differently. Previous studies [
20
,
99
] reported that
higher GDP per capita generally reflects larger average farm sizes [
100
], which are typically
market-oriented. Furthermore, a previous study [
99
] suggested that this type of system may be
more sensitive to drought when they tend to achieve maximum yield potential associated with
higher risk instead of adjusting low-risk management practices during climate stress. Possible
reasons are highlighted: when GDP per capita increases, conventional drought coping strate-
gies may be reduced, while in poorer regions, traditional and well-settled adaptation strategies
may be more preserved, possibly reducing the harvested yield volatility [
20
,
99
]. However, this
socio-economic indicator effect in defining the crop sensitivity to drought may remain com-
plex in actual conditions. In contrast, numerous pieces of literature have indicated that the
importance of GDP or capital indicators may decrease vulnerability (i.e., increasing adaptive
capacity) due to more financial reserves for investing in, e.g., drought-tolerant breeding, irriga-
tion facility, mechanization, or weather forecasting [
13
,
20
,
25
,
26
]. Other studies also
highlighted that crop resilience to drought is higher in rich and developing countries than in
middle-income countries [
20
]. It is worth noting that this present study only considers a spe-
cific vulnerability measure (i.e., crop yield sensitivity, yield loss) based on the relationship
between crop yield and meteorological drought. In contrast, a vulnerability in its comprehen-
sive definition may consider more factors (i.e., drought-induced production and economic
loss, food trade) [
7
,
9
].
Finally, we acknowledge the limitation of this study related to the datasets used, particularly
the reliance on the gridded dataset of historical yield (GDHY) and SPI-based drought indica-
tor as the two main input parameters [
8
]. The crop-drought sensitivity patterns are also subject
to change if other datasets (i.e., sub-country crop yield data [
3
]) and other drought indicators
are used [
37
]. The drought definition used in this study can be further improved by consider-
ing other parameters (i.e., temperature, ET, soil moisture). Therefore, we acknowledge that
crop yield sensitivity here is exclusively determined by the specific datasets used here, mainly
obtained from the SPI-based drought indicator and yield dataset from GDHY [
41
]. It is note-
worthy that the result of key determinants of the global pattern of crop sensitivity to drought
may be sensitive to the data and model selection in this study.
At this point, challenges and questions remain for empirically understanding the complete
drought characterization on global agriculture. Nevertheless, this study is worth undertaking
since this provides the first attempt to broadly link several possible determinants with the
crop-drought relationship to reveal the most influential parameter in defining the relationship
patterns over the global scale. Overall, this study’s results may provide a basis for measures to
achieve food security against the pressures such as increasing severity and frequency of
extreme events under climate changes, food demand due to the growing global population,
and land availability [
5
,
13
,
101
].
Conclusion
This paper explores key factors determining crop-drought sensitivity for maize, rice, soybean,
and wheat for 1981–2016. Results suggest that maize, soybean, and wheat yield are significantly
affected by drought in major dry croplands where they are mostly cultivated (e.g., The Great
Plains, Africa, Australia, the Mediterranean basin, and Central Asia), while rice is generally
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
13 / 20
Page 14
less affected. We link the crop-drought sensitivity with various determinants using the Ran-
dom Forest model. Results reveal that environmental factors (i.e., annual precipitation, soil
water-holding capacity, soil acidity, annual potential evapotranspiration) and management
factors (i.e., fertilizers application rate, growing season) are among the key factors in control-
ling crop sensitivity to drought. Our study also reveals how such important factors affect crop-
drought vulnerability: 1) crop yield in semi-arid and arid regions with low annual precipitation
and higher PET is more sensitive to drought impact; 2) other important factors such as higher
fertilizer rate, elevation, soil water holding capacity and shorter growing season have a moder-
ate association with lower vulnerability, 3) the effect of other socioeconomic indicators (i.e.,
GDP per capita, irrigation) may be underestimated in our models generally due to limitation
in modeling the actual mechanism by the selected proxies based the currently available global
dataset. These results may improve our understanding of global crop-drought sensitivity pat-
terns and their key determinants. Future studies are expected to address the remaining gaps
demonstrated in this study to understand crop system vulnerability to drought toward future
challenges to achieve food security.
Supporting information
S1 Table. List of dataset sources. All datasets are freely available via websites.
(PDF)
S2 Table. Output performance (R
2
and RMSE) of each machine learning algorithm trained
in this study. The support vector machine model is built using ´´´svmRadial´´´ based on the
"caret" R library with the parameter sigma value (
σ
) of 0.1 and cost (C) of 1. Extreme gradient
boosting is developed using "xgbTree" method in "caret" using its default fitting parameters
(e.g., number of rounds of boosting = 500, maximum depth of a tree = 6, learning rate = 0.3).
The values of the performance skills are based on the mean value from 10-fold cross-valida-
tion, except in the case of random Forest, which is solely based on OOB (out-of-bag) validation
in this study. The unit percentage for RMSE denotes crop yield anomaly per drought magni-
tude (
β
).
(PDF)
S1 Fig. Example of the linear regression between yield anomaly (
Δ
Y
) and
DI
in a grid cell.
Crop drought sensitivity (
β
) is obtained based on the slope coefficient.
(TIF)
S2 Fig. Crop yield sensitivity to drought based on 9-month SPI across global croplands
based on GDHY for maize, rice, soybean, and wheat. Sensitivity is calculated by the linear
slope coefficient (
β
) with crop yield anomaly (%) per drought magnitude. White grid cells
show no cropland or crop yield data, and grey shows the non-significant grid cells (
P
�
0.05).
(TIF)
S3 Fig. The relative importance of the determinant factors based on Support Vector
Regression (SVR) and Extreme Gradient Boosting models (XGBoost). The importance val-
ues are normalized to the 0–1 range computed separately across different models. The colored
bars represent the average relative importance across different algorithms. Label order of the
y-axis indicates the order from the most important variables to the least based on the overall
average across crops and models.
(TIF)
S4 Fig. Relationship between crop-drought sensitivity (
β
) as the response variable and the
determinants factors as independent variables based on each partial dependence using
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
14 / 20
Page 15
Random Forest model. The order shows high to lower average relative importance across
crops based on this specific model (not overall). The lower negative y-axis indicates higher
yield loss due to drought.
(TIF)
S5 Fig. Same as
S4 Fig
but for SVR model.
(TIF)
S6 Fig. Same as
S4 Fig
but for XGBoost model.
(TIF)
Author Contributions
Conceptualization: Daisuke Komori, Wonsik Kim.
Data curation: Vempi Satriya Adi Hendrawan.
Formal analysis: Vempi Satriya Adi Hendrawan.
Funding acquisition: Daisuke Komori.
Investigation: Vempi Satriya Adi Hendrawan.
Methodology: Vempi Satriya Adi Hendrawan, Daisuke Komori.
Project administration: Daisuke Komori.
Resources: Daisuke Komori.
Software: Vempi Satriya Adi Hendrawan.
Supervision: Daisuke Komori, Wonsik Kim.
Validation: Daisuke Komori, Wonsik Kim.
Visualization: Vempi Satriya Adi Hendrawan.
Writing – original draft: Vempi Satriya Adi Hendrawan.
Writing – review & editing: Daisuke Komori, Wonsik Kim.
References
1.
Lobell DB, Field CB. Global scale climate-crop yield relationships and the impacts of recent warming.
Environmental Research Letters. 2007; 2.
https://doi.org/10.1088/1748-9326/2/1/014002
2.
Vogel E, Donat MG, Alexander L v., Meinshausen M, Ray DK, Karoly D, et al. The effects of climate
extremes on global agricultural yields. Environmental Research Letters. 2019; 14.
https://doi.org/10.
1088/1748-9326/ab154b
3.
Ray DK, West PC, Clark M, Gerber JS, Prishchepov A V., Chatterjee S Climate change has likely
already affected global food production. PLoS One. 2019; 14: 1–18.
https://doi.org/10.1371/journal.
pone.0217148
PMID:
31150427
4.
Iizumi T, Sakuma H, Yokozawa M, Luo JJ, Challinor AJ, Brown ME, et al. Prediction of seasonal cli-
mate-induced variations in global food production. Nat Clim Chang. 2013; 3: 904–908.
https://doi.org/
10.1038/nclimate1945
5.
Seneviratne SI, Nicholls N, Easterling D, Goodess CM, Kanae S, Kossin J, et al. Changes in Climate
Extremes and their Impacts on the Natural Physical Environment. In: Field CB, Barros V, Stocker TF,
Dahe Q, editors. Managing the Risks of Extreme Events and Disasters to Advance Climate Change
Adaptation. Cambridge: Cambridge University Press; 2012. pp. 109–230.
https://doi.org/10.1017/
CBO9781139177245.006
6.
Eckholm E. Losing ground. Environment: Science and Policy for Sustainable Development. 1976; 18:
6–11.
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
15 / 20
Page 16
7.
Kim W, Iizumi T, Nishimori M. Global patterns of crop production losses associated with droughts from
1983 to 2009. J Appl Meteorol Climatol. 2019; 58: 1233–1244.
https://doi.org/10.1175/JAMC-D-18-
0174.1
8.
Hendrawan VSA, Kim W, Touge Y, Ke S, Komori D. A global-scale relationship between crop yield
anomaly and multiscalar drought index based on multiple precipitation data. Environmental Research
Letters. 2022; 17: 014037.
https://doi.org/10.1088/1748-9326/ac45b4
9.
IPCC. Climate Change 2007: Impacts, Adaptation and Vulnerability. Working Group II Contribution to
the Intergovernmental Panel on Climate Change Fourth Assessment Report. Brussels, Belgium:
IPCC; 2007.
10.
Hasegawa T, Sakurai G, Fujimori S, Takahashi K, Hijioka Y, Masui T. Extreme climate events increase
risk of global food insecurity and adaptation needs. Nat Food. 2021; 2: 587–595.
https://doi.org/10.
1038/s43016-021-00335-4
11.
Liu D, Mishra AK, Ray DK. Sensitivity of global major crop yields to climate variables: A non-parametric
elasticity analysis. Science of the Total Environment. 2020; 748: 141431.
https://doi.org/10.1016/j.
scitotenv.2020.141431
PMID:
32805570
12.
Simelton E, Fraser EDG, Termansen M, Forster PM, Dougill AJ. Typologies of crop-drought vulnera-
bility: an empirical analysis of the socio-economic factors that influence the sensitivity and resilience to
drought of three major food crops in China (1961–2001). Environ Sci Policy. 2009; 12: 438–452.
https://doi.org/10.1016/j.envsci.2008.11.005
13.
Hagenlocher M, Meza I, Anderson CC, Min A, Renaud FG, Walz Y, et al. Drought vulnerability and risk
assessments: State of the art, persistent gaps, and research agenda. Environmental Research Let-
ters. 2019;14.
https://doi.org/10.1088/1748-9326/ab225d
14.
Wilhelmi O V, Wilhite DA. Assessing Vulnerability to Agricultural Drought: A Nebraska Case Study.
Natural Hazards. 2002; 25: 37–58.
https://doi.org/10.1023/A:1013388814894
15.
Fahad S, Bajwa AA, Nazir U, Anjum SA, Farooq A, Zohaib A, et al. Crop production under drought and
heat stress: Plant responses and management options. Front Plant Sci. 2017; 8: 1–16.
https://doi.org/
10.3389/fpls.2017.01147
PMID:
28706531
16.
Rosenzweig C, Elliott J, Deryng D, Ruane AC, Mu¨ller C, Arneth A, et al. Assessing agricultural risks of
climate change in the 21st century in a global gridded crop model intercomparison. Proc Natl Acad Sci
U S A. 2014; 111: 3268–3273.
https://doi.org/10.1073/pnas.1222463110
PMID:
24344314
17.
Daryanto S, Wang L, Jacinthe PA. Global synthesis of drought effects on maize and wheat production.
PLoS One. 2016; 11: 1–15.
https://doi.org/10.1371/journal.pone.0156362
PMID:
27223810
18.
Wu JJ, Geng GP, Zhou HK, Liu JH, Wang QF, Yang JH. Global vulnerability to agricultural drought
and its spatial characteristics. Sci China Earth Sci. 2017; 60: 910–920.
https://doi.org/10.1007/
s11430-016-9018-2
19.
Fraser EDG, Simelton E, Termansen M, Gosling SN, South A. “Vulnerability hotspots”: Integrating
socio-economic and hydrological models to identify where cereal production may decline in the future
due to climate change induced drought. Agric For Meteorol. 2013; 170: 195–205.
https://doi.org/10.
1016/j.agrformet.2012.04.008
20.
Simelton E, Fraser EDG, Termansen M, Benton TG, Gosling SN, South A, et al. The socioeconomics
of food crop production and climate change vulnerability: a global scale quantitative analysis of how
grain crops are sensitive to drought. Food Secur. 2012; 4: 163–179.
https://doi.org/10.1007/s12571-
012-0173-4
21.
Antwi-Agyei P, Fraser EDG, Dougill AJ, Stringer LC, Simelton E. Mapping the vulnerability of crop pro-
duction to drought in Ghana using rainfall, yield and socioeconomic data. Applied Geography. 2012;
32: 324–334.
https://doi.org/10.1016/j.apgeog.2011.06.010
22.
Lobell DB, Deines JM, Tommaso S Di. Changes in the drought sensitivity of US maize yields. Nat
Food. 2020; 1: 729–735.
https://doi.org/10.1038/s43016-020-00165-w
23.
Fraser EDG. Food system vulnerability: Using past famines to help understand how food systems
may adapt to climate change. Ecological Complexity. 2006; 3: 328–335.
https://doi.org/10.1016/j.
ecocom.2007.02.006
24.
Kamali B, Abbaspour KC, Wehrli B, Yang H. Drought vulnerability assessment of maize in Sub-Saha-
ran Africa: Insights from physical and social perspectives. Glob Planet Change. 2018; 162: 266–274.
https://doi.org/10.1016/j.gloplacha.2018.01.011
25.
Murthy CS, Laxman B, Sesha Sai MVR. Geospatial analysis of agricultural drought vulnerability using
a composite index based on exposure, sensitivity and adaptive capacity. International Journal of
Disaster Risk Reduction. 2015; 12: 163–171.
https://doi.org/10.1016/j.ijdrr.2015.01.004
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
16 / 20
Page 17
26.
Carr
ã
o H, Naumann G, Barbosa P. Mapping global patterns of drought risk: An empirical framework
based on sub-national estimates of hazard, exposure and vulnerability. Global Environmental Change.
2016; 39: 108–124.
https://doi.org/10.1016/j.gloenvcha.2016.04.012
27.
Meza I, Siebert S, Do¨ll P, Kusche J, Herbert C, Rezaei EE, et al. Global-scale drought risk assessment
for agricultural systems. Natural Hazards and Earth System Sciences. 2020; 20: 695–712.
https://doi.
org/10.5194/nhess-20-695-2020
28.
Williams PA, Crespo O, Abu M, Simpson NP. A systematic review of how vulnerability of smallholder
agricultural systems to changing climate is assessed in Africa. Environmental Research Letters.
2018;13.
https://doi.org/10.1088/1748-9326/aae026
29.
Wu J, He B, Lu¨ A, Zhou L, Liu M, Zhao L. Quantitative assessment and spatial characteristics analysis
of agricultural drought vulnerability in China. Natural Hazards. 2011; 56: 785–801.
https://doi.org/10.
1007/s11069-010-9591-9
30.
Frischen J, Meza I, Rupp D, Wietler K, Hagenlocher M. Drought risk to agricultural systems in Zimba-
bwe: A spatial analysis of hazard, exposure, and vulnerability. Sustainability (Switzerland). 2020; 12.
https://doi.org/10.3390/su12030752
31.
Kinnunen P, Heino M, Sandstro
¨ m V, Taka M, Ray DK, Kummu M. Crop Yield Loss Risk Is Modulated
by Anthropogenic Factors. Earths Future. 2022; 10: 1–16.
https://doi.org/10.1029/2021EF002420
PMID:
36583138
32.
McKee T, Doesken N, Kleist J. The relationship of drought frequency and duration to time scales. Proc
8th Conf on Applied Climatology. 1993. pp. 179–183.
33.
Wilhite DA, Glantz MH. Understanding: the Drought Phenomenon: The Role of Definitions. Water Int.
1985; 10: 111–120.
https://doi.org/10.1080/02508068508686328
34.
Turco M, Jerez S, Donat MG, Toreti A, Vicente-Serrano SM, Doblas-Reyes FJ. A global probabilistic
dataset for monitoring meteorological droughts. Bull Am Meteorol Soc. 2020; 101: E1628–E1644.
https://doi.org/10.1175/BAMS-D-19-0192.1
35.
Mishra AK, Singh VP. A review of drought concepts. J Hydrol (Amst). 2010; 391: 202–216.
https://doi.
org/10.1016/j.jhydrol.2010.07.012
36.
Spinoni J, Naumann G, Carrao H, Barbosa P, Vogt J. World drought frequency, duration, and severity
for 1951–2010. Int J Climatol. 2014; 34: 2792–2804.
https://doi.org/10.1002/joc.3875
37.
Hoffmann D, Gallant AJE, Arblaster JM. Uncertainties in Drought From Index and Data Selection.
Journal of Geophysical Research: Atmospheres. 2020; 125: 1–21.
https://doi.org/10.1029/
2019JD031946
38.
Zargar A, Sadiq R, Naser B, Khan FI. A review of drought indices. Environmental Reviews. 2011; 19:
333–349.
https://doi.org/10.1139/a11-013
39.
Lloyd-Hughes B, Saunders MA. A drought climatology for Europe. International Journal of Climatol-
ogy. 2002; 22: 1571–1592.
https://doi.org/10.1002/joc.846
40.
Sacks WJ, Deryng D, Foley JA, Ramankutty N. Crop planting dates: An analysis of global patterns.
Global Ecology and Biogeography. 2010; 19: 607–620.
https://doi.org/10.1111/j.1466-8238.2010.
00551.x
41.
Iizumi T, Sakai T. The global dataset of historical yields for major crops 1981–2016. Sci Data. 2020; 7:
1–7.
https://doi.org/10.1038/s41597-020-0433-7
PMID:
32198349
42.
Lu J, Carbone GJ, Gao P. Detrending crop yield data for spatial visualization of drought impacts in the
United States, 1895–2014. Agric For Meteorol. 2017; 237–238: 196–208.
https://doi.org/10.1016/j.
agrformet.2017.02.001
43.
Liu H, Jin Y, Roche LM, O’Geen AT, Dahlgren RA. Understanding spatial variability of forage produc-
tion in California grasslands: Delineating climate, topography and soil controls. Environmental
Research Letters. 2021; 16.
https://doi.org/10.1088/1748-9326/abc64d
44.
Lychuk TE, Moulin AP, Lemke RL, Gossen BD, Leeson JY, Kirk A, et al. Effects of crop inputs, diver-
sity, environment, and terrain on yield in an 18-yr study in the semi-arid Canadian prairies. Canadian
Journal of Plant Science. 2017; 97: 715–730.
https://doi.org/10.1139/cjps-2016-0228
45.
Davis KF, Chhatre A, Rao ND, Singh D, Defries R. Sensitivity of grain yields to historical climate vari-
ability in India. Environmental Research Letters. 2019; 14.
https://doi.org/10.1088/1748-9326/ab22db
46.
Iizumi T, Hosokawa N, Wagai R. Soil carbon-food synergy: sizable contributions of small-scale farm-
ers. CABI Agriculture and Bioscience. 2021; 2: 1–15.
https://doi.org/10.1186/s43170-021-00063-6
47.
Williams A, Hunter MC, Kammerer M, Kane DA, Jordan NR, Mortensen DA, et al. Soil water holding
capacity mitigates downside risk and volatility in US rainfed maize: Time to invest in soil organic mat-
ter? PLoS One. 2016; 11: 1–11.
https://doi.org/10.1371/journal.pone.0160974
PMID:
27560666
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
17 / 20
Page 18
48.
Yamoah CF, Walters DT, Shapiro CA, Francis CA, Hayes MJ. Standardized precipitation index and
nitrogen rate effects on crop yields and risk distribution in maize. Agric Ecosyst Environ. 2000; 80:
113–120.
https://doi.org/10.1016/S0167-8809(00)00140-7
49.
Portmann FT, Siebert S, Do¨ll P. MIRCA2000-Global monthly irrigated and rainfed crop areas around
the year 2000: A new high-resolution data set for agricultural and hydrological modeling. Global Bio-
geochem Cycles. 2010; 24: n/a-n/a.
https://doi.org/10.1029/2008gb003435
50.
Siebert S, Do¨ll P, Hoogeveen J, Faures J-M, Frenken K, Feick S. Development and validation of the
global map of irrigation areas. Hydrol Earth Syst Sci. 2005; 9: 535–547.
https://doi.org/10.5194/hess-
9-535-2005
51.
Harris I, Osborn TJ, Jones P, Lister D. Version 4 of the CRU TS monthly high-resolution gridded multi-
variate climate dataset. Sci Data. 2020; 7: 1–18.
https://doi.org/10.1038/s41597-020-0453-3
PMID:
32246091
52.
IIASA/FAO. Global Agro-ecological Zones (GAEZ v3.0). Laxenburg, Austria and Rome, Italy.: IIASA
and FAO; 2010.
53.
Montzka C, Herbst M, Weihermu¨ller L, Verhoef A, Vereecken H. A global data set of soil hydraulic
properties and sub-grid variability of soil water retention and hydraulic conductivity curves. Earth Syst
Sci Data. 2017; 9: 529–543.
https://doi.org/10.5194/essd-9-529-2017
54.
Wieder W. Regridded Harmonized World Soil Database v1.2. ORNL Distributed Active Archive Cen-
ter; 2014.
https://doi.org/10.3334/ORNLDAAC/1247
55.
Webb RS, Rosenzweig CE, Levine ER. Specifying land surface characteristics in general circulation
models: Soil profile data set and derived water-holding capacities. Global Biogeochem Cycles. 1993;
7: 97–108.
56.
Siebert S, Burke J, Faures JM, Frenken K, Hoogeveen J, Do¨ll P, et al. Groundwater use for irrigation–
a global inventory. Hydrol Earth Syst Sci. 2010; 14: 1863–1880.
https://doi.org/10.5194/hess-14-
1863-2010
57.
Mueller ND, Gerber JS, Johnston M, Ray DK, Ramankutty N, Foley JA. Closing yield gaps through
nutrient and water management. Nature. 2012; 490: 254–257.
https://doi.org/10.1038/nature11420
PMID:
22932270
58.
Kummu M, Taka M, Guillaume JHA. Gridded global datasets for Gross Domestic Product and Human
Development Index over 1990–2015. Sci Data. 2018; 5: 1–15.
https://doi.org/10.1038/sdata.2018.4
PMID:
29406518
59.
CIESIN. Gridded Population of the World, Version 4 (GPWv4): Population Density, Revision 11. Pali-
sades, NY: NASA Socioeconomic Data and Applications Center (SEDAC); 2018.
60.
Seabold S, Perktold J. statsmodels: Econometric and statistical modeling with python. 9th Python in
Science Conference. 2010.
61.
Lesk C, Coffel E, Winter J, Ray D, Zscheischler J, Seneviratne SI, et al. Stronger temperature–mois-
ture couplings exacerbate the impact of climate warming on global crop yields. Nat Food. 2021; 2:
683–691.
https://doi.org/10.1038/s43016-021-00341-6
62.
Breiman L. Random Forests. Mach Learn. 2001; 45: 5–32.
https://doi.org/10.1023/A:1010933404324
63.
Park S, Im J, Jang E, Rhee J. Drought assessment and monitoring through blending of multi-sensor
indices using machine learning approaches for different climate regions. Agric For Meteorol. 2016;
216: 157–169.
https://doi.org/10.1016/j.agrformet.2015.10.011
64.
Jo T, Cheng J. Improving protein fold recognition by random forest. BMC Bioinformatics. 2014; 15: 1–
7.
https://doi.org/10.1186/1471-2105-15-S11-S14
PMID:
25350499
65.
Boser BE, Vapnik VN, Guyon IM. Training Algorithm Margin for Optimal Classifiers. Perception. 1992;
144–152.
66.
Friedman JH. Greedy function approximation: A gradient boosting machine. The Annals of Statistics.
2001; 29: 1189–1232.
https://doi.org/10.1214/aos/1013203451
67.
Kuhn M. Building predictive models in R using the caret package. J Stat Softw. 2008; 28: 1–26.
https://
doi.org/10.18637/jss.v028.i05
68.
Liaw A, Wiener M. Classification and Regression by randomForest. R News. 2002; 2: 18–22.
69.
Greenwell BM. pdp: An R package for constructing partial dependence plots. R Journal. 2017; 9: 421–
436.
https://doi.org/10.32614/rj-2017-016
70.
Met Office. Cartopy: a cartographic python library with a Matplotlib interface. Exeter, Devon; Available:
https://scitools.org.uk/cartopy
71.
Nielsen DC, Vigil MF, Benjamin JG. The variable response of dryland corn yield to soil water content
at planting. Agric Water Manag. 2009; 96: 330–336.
https://doi.org/10.1016/j.agwat.2008.08.011
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
18 / 20
Page 19
72.
Vicente-Serrano SM. Evaluating the impact of drought using remote sensing in a Mediterranean,
Semi-arid Region. Natural Hazards. 2007; 40: 173–208.
https://doi.org/10.1007/s11069-006-0009-7
73.
Ray DK, Gerber JS, Macdonald GK, West PC. Climate variation explains a third of global crop yield
variability. Nat Commun. 2015; 6: 1–9.
https://doi.org/10.1038/ncomms6989
PMID:
25609225
74.
Strauss F, Moltchanova E, Schmid E. Spatially Explicit Modeling of Long-Term Drought Impacts on
Crop Production in Austria. Am J Clim Change. 2013; 02: 1–11.
https://doi.org/10.4236/ajcc.2013.
23a001
75.
Lambert L, Chitrakar BD. Variation of potential evapotranspiration with elevation in Nepal. Mt Res Dev.
1989; 9: 145–152.
https://doi.org/10.2307/3673477
76.
He D, Wang E. On the relation between soil water holding capacity and dryland crop productivity. Geo-
derma. 2019; 353: 11–24.
https://doi.org/10.1016/j.geoderma.2019.06.022
77.
Sweet SK, Wolfe DW, DeGaetano A, Benner R. Anatomy of the 2016 drought in the Northeastern
United States: Implications for agriculture and water resources in humid climates. Agric For Meteorol.
2017; 247: 571–581.
https://doi.org/10.1016/j.agrformet.2017.08.024
78.
Wang J, Wang X, Wang J. Profile distribution of CO2 in an arid saline-alkali soil with gypsum and
wheat straw amendments: a two-year incubation experiment. Sci Rep. 2018; 8: 1–10.
https://doi.org/
10.1038/s41598-018-30312-0
PMID:
30093635
79.
Fang S, Hou X, Liang X. Response Mechanisms of Plants Under Saline-Alkali Stress. Front Plant Sci.
2021; 12: 1–20.
https://doi.org/10.3389/fpls.2021.667458
PMID:
34149764
80.
Shao Y, Irannejad P. On the choice of soil hydraulic models in land-surface schemes. Boundary Layer
Meteorol. 1999; 90: 83–115.
https://doi.org/10.1023/A:1001786023282
81.
Solone R, Bittelli M, Tomei F, Morari F. Errors in water retention curves determined with pressure
plates: Effects on the soil water balance. J Hydrol (Amst). 2012; 470–471: 65–74.
https://doi.org/10.
1016/j.jhydrol.2012.08.017
82.
Bodner G, Nakhforoosh A, Kaul HP. Management of crop water under drought: a review. Agron Sus-
tain Dev. 2015; 35: 401–442.
https://doi.org/10.1007/s13593-015-0283-4
83.
Samar Raza MA, Farrukh Saleem M, Mustafa Shah G, Jamil M, Haider Khan I. Potassium applied
under drought improves physiological and nutrient uptake performances of wheat (Triticum aestivun
L.). J Soil Sci Plant Nutr. 2013; 13: 175–185.
https://doi.org/10.4067/s0718-95162013005000016
84.
Ul-Allah S, Ijaz M, Nawaz A, Sattar A, Sher A, Naeem M, et al. Potassium Application Improves Grain
Yield and Maize Hybrids. Plants. 2020; 9: 1–11.
85.
Jin J, Wang G, Liu X, Pan X, Herbert SJ, Tang C. Interaction Between Phosphorus Nutrition and
Drought on Grain Yield, and Assimilation of Phosphorus and Nitrogen in Two Soybean Cultivars Differ-
ing in Protein Concentration in Grains. J Plant Nutr. 2006; 29: 1433–1449.
https://doi.org/10.1080/
01904160600837089
86.
Wang M, Zheng Q, Shen Q, Guo S. The critical role of potassium in plant stress response. Int J Mol
Sci. 2013; 14: 7370–7390.
https://doi.org/10.3390/ijms14047370
PMID:
23549270
87.
Lu J, Carbone GJ, Huang X, Lackstrom K, Gao P. Mapping the sensitivity of agriculture to drought and
estimating the effect of irrigation in the United States, 1950–2016. Agric For Meteorol. 2020; 292–293:
108124.
https://doi.org/10.1016/j.agrformet.2020.108124
88.
Matiu M, Ankerst DP, Menzel A. Interactions between temperature and drought in global and regional
crop yield variability during 1961–2014. PLoS One. 2017; 12: 1–23.
https://doi.org/10.1371/journal.
pone.0178339
PMID:
28552938
89.
Lesk C, Rowhani P, Ramankutty N. Influence of extreme weather disasters on global crop production.
Nature. 2016; 529: 84–87.
https://doi.org/10.1038/nature16467
PMID:
26738594
90.
Troy TJ, Kipgen C, Pal I. The impact of climate extremes and irrigation on US crop yields. Environmen-
tal Research Letters. 2015;10.
https://doi.org/10.1088/1748-9326/10/5/054013
91.
Rigden AJ, Mueller ND, Holbrook NM, Pillai N, Huybers P. Combined influence of soil moisture and
atmospheric evaporative demand is important for accurately predicting US maize yields. Nat Food.
2020; 1: 127–133.
https://doi.org/10.1038/s43016-020-0028-7
92.
Vicente-Serrano SM, McVicar TR, Miralles DG, Yang Y, Tomas-Burguera M. Unraveling the influence
of atmospheric evaporative demand on drought and its response to climate change. Wiley Interdiscip
Rev Clim Change. 2020; 11: 1–31.
https://doi.org/10.1002/wcc.632
93.
Carr
ã
o H, Russo S, Sepulcre-Canto G, Barbosa P. An empirical standardized soil moisture index for
agricultural drought assessment from remotely sensed data. International Journal of Applied Earth
Observation and Geoinformation. 2016; 48: 74–84.
https://doi.org/10.1016/j.jag.2015.06.011
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
19 / 20
Page 20
94.
Berg A, Lintner BR, Findell K, Seneviratne SI, Denhurk B Van, Ducharne A, et al. Interannual coupling
between summertime surface temperature and precipitation over land: Processes and implications for
climate change. J Clim. 2015; 28: 1308–1328.
https://doi.org/10.1175/JCLI-D-14-00324.1
95.
Seneviratne SI, Corti T, Davin EL, Hirschi M, Jaeger EB, Lehner I, et al. Investigating soil moisture-cli-
mate interactions in a changing climate: A review. Earth Sci Rev. 2010; 99: 125–161.
https://doi.org/
10.1016/j.earscirev.2010.02.004
96.
Lesk C, Anderson W. Decadal variability modulates trends in concurrent heat and drought over global
croplands. Environmental Research Letters. 2021; 16: 055024.
https://doi.org/10.1088/1748-9326/
abeb35
97.
Lipiec J, Doussan C, Nosalewicz A, Kondracka K. Effect of drought and heat stresses on plant growth
and yield: A review. Int Agrophys. 2013; 27: 463–477.
https://doi.org/10.2478/intag-2013-0017
98.
Zipper SC, Qiu J, Kucharik CJ. Drought effects on US maize and soybean production: Spatiotemporal
patterns and historical changes. Environmental Research Letters. 2016;11.
https://doi.org/10.1088/
1748-9326/11/9/094021
99.
Stringer LC, Fraser EDG, Harris D, Lyon C, Pereira L, Ward CFM, et al. Adaptation and development
pathways for different types of farmers. Environ Sci Policy. 2020; 104: 174–189.
https://doi.org/10.
1016/j.envsci.2019.10.007
100.
Lowder SK, Skoet J, Raney T. The Number, Size, and Distribution of Farms, Smallholder Farms, and
Family Farms Worldwide. World Dev. 2016; 87: 16–29.
https://doi.org/10.1016/j.worlddev.2015.10.
041
101.
United Nations. Transforming our world: the 2030 Agenda for Sustainable Development. 2015 [cited 7
Apr 2022] pp. 1–35. Available:
https://www.refworld.org/docid/57b6e3e44.html
PLOS ONE
Possible factors determining global-scale patterns of crop yield sensitivity to drought
PLOS ONE |
https://doi.org/10.1371/journal.pone.0281287
February 2, 2023
20 / 20